?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ipv4: Fix PMTU update.
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / ipv4 / route.c
blob209989cf7d1bf1e68fffb2f09c2a852701e50c1c
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(oldflp4) \
113 ((u32)(oldflp4->flowi4_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 /*
135 * Interface to generic destination cache.
136 */
137
138 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
139 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
140 static unsigned int ipv4_default_mtu(const struct dst_entry *dst);
141 static void ipv4_dst_destroy(struct dst_entry *dst);
142 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
143 static void ipv4_link_failure(struct sk_buff *skb);
144 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
145 static int rt_garbage_collect(struct dst_ops *ops);
146
147 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
148 int how)
149 {
150 }
151
152 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
153 {
154 struct rtable *rt = (struct rtable *) dst;
155 struct inet_peer *peer;
156 u32 *p = NULL;
157
158 if (!rt->peer)
159 rt_bind_peer(rt, 1);
160
161 peer = rt->peer;
162 if (peer) {
163 u32 *old_p = __DST_METRICS_PTR(old);
164 unsigned long prev, new;
165
166 p = peer->metrics;
167 if (inet_metrics_new(peer))
168 memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
169
170 new = (unsigned long) p;
171 prev = cmpxchg(&dst->_metrics, old, new);
172
173 if (prev != old) {
174 p = __DST_METRICS_PTR(prev);
175 if (prev & DST_METRICS_READ_ONLY)
176 p = NULL;
177 } else {
178 if (rt->fi) {
179 fib_info_put(rt->fi);
180 rt->fi = NULL;
181 }
182 }
183 }
184 return p;
185 }
186
187 static struct dst_ops ipv4_dst_ops = {
188 .family = AF_INET,
189 .protocol = cpu_to_be16(ETH_P_IP),
190 .gc = rt_garbage_collect,
191 .check = ipv4_dst_check,
192 .default_advmss = ipv4_default_advmss,
193 .default_mtu = ipv4_default_mtu,
194 .cow_metrics = ipv4_cow_metrics,
195 .destroy = ipv4_dst_destroy,
196 .ifdown = ipv4_dst_ifdown,
197 .negative_advice = ipv4_negative_advice,
198 .link_failure = ipv4_link_failure,
199 .update_pmtu = ip_rt_update_pmtu,
200 .local_out = __ip_local_out,
201 };
202
203 #define ECN_OR_COST(class) TC_PRIO_##class
204
205 const __u8 ip_tos2prio[16] = {
206 TC_PRIO_BESTEFFORT,
207 ECN_OR_COST(FILLER),
208 TC_PRIO_BESTEFFORT,
209 ECN_OR_COST(BESTEFFORT),
210 TC_PRIO_BULK,
211 ECN_OR_COST(BULK),
212 TC_PRIO_BULK,
213 ECN_OR_COST(BULK),
214 TC_PRIO_INTERACTIVE,
215 ECN_OR_COST(INTERACTIVE),
216 TC_PRIO_INTERACTIVE,
217 ECN_OR_COST(INTERACTIVE),
218 TC_PRIO_INTERACTIVE_BULK,
219 ECN_OR_COST(INTERACTIVE_BULK),
220 TC_PRIO_INTERACTIVE_BULK,
221 ECN_OR_COST(INTERACTIVE_BULK)
222 };
223
224
225 /*
226 * Route cache.
227 */
228
229 /* The locking scheme is rather straight forward:
230 *
231 * 1) Read-Copy Update protects the buckets of the central route hash.
232 * 2) Only writers remove entries, and they hold the lock
233 * as they look at rtable reference counts.
234 * 3) Only readers acquire references to rtable entries,
235 * they do so with atomic increments and with the
236 * lock held.
237 */
238
239 struct rt_hash_bucket {
240 struct rtable __rcu *chain;
241 };
242
243 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
244 defined(CONFIG_PROVE_LOCKING)
245 /*
246 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
247 * The size of this table is a power of two and depends on the number of CPUS.
248 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
249 */
250 #ifdef CONFIG_LOCKDEP
251 # define RT_HASH_LOCK_SZ 256
252 #else
253 # if NR_CPUS >= 32
254 # define RT_HASH_LOCK_SZ 4096
255 # elif NR_CPUS >= 16
256 # define RT_HASH_LOCK_SZ 2048
257 # elif NR_CPUS >= 8
258 # define RT_HASH_LOCK_SZ 1024
259 # elif NR_CPUS >= 4
260 # define RT_HASH_LOCK_SZ 512
261 # else
262 # define RT_HASH_LOCK_SZ 256
263 # endif
264 #endif
265
266 static spinlock_t *rt_hash_locks;
267 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
268
269 static __init void rt_hash_lock_init(void)
270 {
271 int i;
272
273 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
274 GFP_KERNEL);
275 if (!rt_hash_locks)
276 panic("IP: failed to allocate rt_hash_locks\n");
277
278 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
279 spin_lock_init(&rt_hash_locks[i]);
280 }
281 #else
282 # define rt_hash_lock_addr(slot) NULL
283
284 static inline void rt_hash_lock_init(void)
285 {
286 }
287 #endif
288
289 static struct rt_hash_bucket *rt_hash_table __read_mostly;
290 static unsigned rt_hash_mask __read_mostly;
291 static unsigned int rt_hash_log __read_mostly;
292
293 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
294 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
295
296 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
297 int genid)
298 {
299 return jhash_3words((__force u32)daddr, (__force u32)saddr,
300 idx, genid)
301 & rt_hash_mask;
302 }
303
304 static inline int rt_genid(struct net *net)
305 {
306 return atomic_read(&net->ipv4.rt_genid);
307 }
308
309 #ifdef CONFIG_PROC_FS
310 struct rt_cache_iter_state {
311 struct seq_net_private p;
312 int bucket;
313 int genid;
314 };
315
316 static struct rtable *rt_cache_get_first(struct seq_file *seq)
317 {
318 struct rt_cache_iter_state *st = seq->private;
319 struct rtable *r = NULL;
320
321 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
322 if (!rcu_dereference_raw(rt_hash_table[st->bucket].chain))
323 continue;
324 rcu_read_lock_bh();
325 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
326 while (r) {
327 if (dev_net(r->dst.dev) == seq_file_net(seq) &&
328 r->rt_genid == st->genid)
329 return r;
330 r = rcu_dereference_bh(r->dst.rt_next);
331 }
332 rcu_read_unlock_bh();
333 }
334 return r;
335 }
336
337 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
338 struct rtable *r)
339 {
340 struct rt_cache_iter_state *st = seq->private;
341
342 r = rcu_dereference_bh(r->dst.rt_next);
343 while (!r) {
344 rcu_read_unlock_bh();
345 do {
346 if (--st->bucket < 0)
347 return NULL;
348 } while (!rcu_dereference_raw(rt_hash_table[st->bucket].chain));
349 rcu_read_lock_bh();
350 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
351 }
352 return r;
353 }
354
355 static struct rtable *rt_cache_get_next(struct seq_file *seq,
356 struct rtable *r)
357 {
358 struct rt_cache_iter_state *st = seq->private;
359 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
360 if (dev_net(r->dst.dev) != seq_file_net(seq))
361 continue;
362 if (r->rt_genid == st->genid)
363 break;
364 }
365 return r;
366 }
367
368 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
369 {
370 struct rtable *r = rt_cache_get_first(seq);
371
372 if (r)
373 while (pos && (r = rt_cache_get_next(seq, r)))
374 --pos;
375 return pos ? NULL : r;
376 }
377
378 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
379 {
380 struct rt_cache_iter_state *st = seq->private;
381 if (*pos)
382 return rt_cache_get_idx(seq, *pos - 1);
383 st->genid = rt_genid(seq_file_net(seq));
384 return SEQ_START_TOKEN;
385 }
386
387 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
388 {
389 struct rtable *r;
390
391 if (v == SEQ_START_TOKEN)
392 r = rt_cache_get_first(seq);
393 else
394 r = rt_cache_get_next(seq, v);
395 ++*pos;
396 return r;
397 }
398
399 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
400 {
401 if (v && v != SEQ_START_TOKEN)
402 rcu_read_unlock_bh();
403 }
404
405 static int rt_cache_seq_show(struct seq_file *seq, void *v)
406 {
407 if (v == SEQ_START_TOKEN)
408 seq_printf(seq, "%-127s\n",
409 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
410 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
411 "HHUptod\tSpecDst");
412 else {
413 struct rtable *r = v;
414 int len;
415
416 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
417 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
418 r->dst.dev ? r->dst.dev->name : "*",
419 (__force u32)r->rt_dst,
420 (__force u32)r->rt_gateway,
421 r->rt_flags, atomic_read(&r->dst.__refcnt),
422 r->dst.__use, 0, (__force u32)r->rt_src,
423 dst_metric_advmss(&r->dst) + 40,
424 dst_metric(&r->dst, RTAX_WINDOW),
425 (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
426 dst_metric(&r->dst, RTAX_RTTVAR)),
427 r->rt_tos,
428 r->dst.hh ? atomic_read(&r->dst.hh->hh_refcnt) : -1,
429 r->dst.hh ? (r->dst.hh->hh_output ==
430 dev_queue_xmit) : 0,
431 r->rt_spec_dst, &len);
432
433 seq_printf(seq, "%*s\n", 127 - len, "");
434 }
435 return 0;
436 }
437
438 static const struct seq_operations rt_cache_seq_ops = {
439 .start = rt_cache_seq_start,
440 .next = rt_cache_seq_next,
441 .stop = rt_cache_seq_stop,
442 .show = rt_cache_seq_show,
443 };
444
445 static int rt_cache_seq_open(struct inode *inode, struct file *file)
446 {
447 return seq_open_net(inode, file, &rt_cache_seq_ops,
448 sizeof(struct rt_cache_iter_state));
449 }
450
451 static const struct file_operations rt_cache_seq_fops = {
452 .owner = THIS_MODULE,
453 .open = rt_cache_seq_open,
454 .read = seq_read,
455 .llseek = seq_lseek,
456 .release = seq_release_net,
457 };
458
459
460 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
461 {
462 int cpu;
463
464 if (*pos == 0)
465 return SEQ_START_TOKEN;
466
467 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
468 if (!cpu_possible(cpu))
469 continue;
470 *pos = cpu+1;
471 return &per_cpu(rt_cache_stat, cpu);
472 }
473 return NULL;
474 }
475
476 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
477 {
478 int cpu;
479
480 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
481 if (!cpu_possible(cpu))
482 continue;
483 *pos = cpu+1;
484 return &per_cpu(rt_cache_stat, cpu);
485 }
486 return NULL;
487
488 }
489
490 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
491 {
492
493 }
494
495 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
496 {
497 struct rt_cache_stat *st = v;
498
499 if (v == SEQ_START_TOKEN) {
500 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");
501 return 0;
502 }
503
504 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
505 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
506 dst_entries_get_slow(&ipv4_dst_ops),
507 st->in_hit,
508 st->in_slow_tot,
509 st->in_slow_mc,
510 st->in_no_route,
511 st->in_brd,
512 st->in_martian_dst,
513 st->in_martian_src,
514
515 st->out_hit,
516 st->out_slow_tot,
517 st->out_slow_mc,
518
519 st->gc_total,
520 st->gc_ignored,
521 st->gc_goal_miss,
522 st->gc_dst_overflow,
523 st->in_hlist_search,
524 st->out_hlist_search
525 );
526 return 0;
527 }
528
529 static const struct seq_operations rt_cpu_seq_ops = {
530 .start = rt_cpu_seq_start,
531 .next = rt_cpu_seq_next,
532 .stop = rt_cpu_seq_stop,
533 .show = rt_cpu_seq_show,
534 };
535
536
537 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
538 {
539 return seq_open(file, &rt_cpu_seq_ops);
540 }
541
542 static const struct file_operations rt_cpu_seq_fops = {
543 .owner = THIS_MODULE,
544 .open = rt_cpu_seq_open,
545 .read = seq_read,
546 .llseek = seq_lseek,
547 .release = seq_release,
548 };
549
550 #ifdef CONFIG_IP_ROUTE_CLASSID
551 static int rt_acct_proc_show(struct seq_file *m, void *v)
552 {
553 struct ip_rt_acct *dst, *src;
554 unsigned int i, j;
555
556 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
557 if (!dst)
558 return -ENOMEM;
559
560 for_each_possible_cpu(i) {
561 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
562 for (j = 0; j < 256; j++) {
563 dst[j].o_bytes += src[j].o_bytes;
564 dst[j].o_packets += src[j].o_packets;
565 dst[j].i_bytes += src[j].i_bytes;
566 dst[j].i_packets += src[j].i_packets;
567 }
568 }
569
570 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
571 kfree(dst);
572 return 0;
573 }
574
575 static int rt_acct_proc_open(struct inode *inode, struct file *file)
576 {
577 return single_open(file, rt_acct_proc_show, NULL);
578 }
579
580 static const struct file_operations rt_acct_proc_fops = {
581 .owner = THIS_MODULE,
582 .open = rt_acct_proc_open,
583 .read = seq_read,
584 .llseek = seq_lseek,
585 .release = single_release,
586 };
587 #endif
588
589 static int __net_init ip_rt_do_proc_init(struct net *net)
590 {
591 struct proc_dir_entry *pde;
592
593 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
594 &rt_cache_seq_fops);
595 if (!pde)
596 goto err1;
597
598 pde = proc_create("rt_cache", S_IRUGO,
599 net->proc_net_stat, &rt_cpu_seq_fops);
600 if (!pde)
601 goto err2;
602
603 #ifdef CONFIG_IP_ROUTE_CLASSID
604 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
605 if (!pde)
606 goto err3;
607 #endif
608 return 0;
609
610 #ifdef CONFIG_IP_ROUTE_CLASSID
611 err3:
612 remove_proc_entry("rt_cache", net->proc_net_stat);
613 #endif
614 err2:
615 remove_proc_entry("rt_cache", net->proc_net);
616 err1:
617 return -ENOMEM;
618 }
619
620 static void __net_exit ip_rt_do_proc_exit(struct net *net)
621 {
622 remove_proc_entry("rt_cache", net->proc_net_stat);
623 remove_proc_entry("rt_cache", net->proc_net);
624 #ifdef CONFIG_IP_ROUTE_CLASSID
625 remove_proc_entry("rt_acct", net->proc_net);
626 #endif
627 }
628
629 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
630 .init = ip_rt_do_proc_init,
631 .exit = ip_rt_do_proc_exit,
632 };
633
634 static int __init ip_rt_proc_init(void)
635 {
636 return register_pernet_subsys(&ip_rt_proc_ops);
637 }
638
639 #else
640 static inline int ip_rt_proc_init(void)
641 {
642 return 0;
643 }
644 #endif /* CONFIG_PROC_FS */
645
646 static inline void rt_free(struct rtable *rt)
647 {
648 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
649 }
650
651 static inline void rt_drop(struct rtable *rt)
652 {
653 ip_rt_put(rt);
654 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
655 }
656
657 static inline int rt_fast_clean(struct rtable *rth)
658 {
659 /* Kill broadcast/multicast entries very aggresively, if they
660 collide in hash table with more useful entries */
661 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
662 rt_is_input_route(rth) && rth->dst.rt_next;
663 }
664
665 static inline int rt_valuable(struct rtable *rth)
666 {
667 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
668 (rth->peer && rth->peer->pmtu_expires);
669 }
670
671 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
672 {
673 unsigned long age;
674 int ret = 0;
675
676 if (atomic_read(&rth->dst.__refcnt))
677 goto out;
678
679 age = jiffies - rth->dst.lastuse;
680 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
681 (age <= tmo2 && rt_valuable(rth)))
682 goto out;
683 ret = 1;
684 out: return ret;
685 }
686
687 /* Bits of score are:
688 * 31: very valuable
689 * 30: not quite useless
690 * 29..0: usage counter
691 */
692 static inline u32 rt_score(struct rtable *rt)
693 {
694 u32 score = jiffies - rt->dst.lastuse;
695
696 score = ~score & ~(3<<30);
697
698 if (rt_valuable(rt))
699 score |= (1<<31);
700
701 if (rt_is_output_route(rt) ||
702 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
703 score |= (1<<30);
704
705 return score;
706 }
707
708 static inline bool rt_caching(const struct net *net)
709 {
710 return net->ipv4.current_rt_cache_rebuild_count <=
711 net->ipv4.sysctl_rt_cache_rebuild_count;
712 }
713
714 static inline bool compare_hash_inputs(const struct rtable *rt1,
715 const struct rtable *rt2)
716 {
717 return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
718 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
719 (rt1->rt_iif ^ rt2->rt_iif)) == 0);
720 }
721
722 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
723 {
724 return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
725 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
726 (rt1->rt_mark ^ rt2->rt_mark) |
727 (rt1->rt_tos ^ rt2->rt_tos) |
728 (rt1->rt_oif ^ rt2->rt_oif) |
729 (rt1->rt_iif ^ rt2->rt_iif)) == 0;
730 }
731
732 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
733 {
734 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
735 }
736
737 static inline int rt_is_expired(struct rtable *rth)
738 {
739 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
740 }
741
742 /*
743 * Perform a full scan of hash table and free all entries.
744 * Can be called by a softirq or a process.
745 * In the later case, we want to be reschedule if necessary
746 */
747 static void rt_do_flush(struct net *net, int process_context)
748 {
749 unsigned int i;
750 struct rtable *rth, *next;
751
752 for (i = 0; i <= rt_hash_mask; i++) {
753 struct rtable __rcu **pprev;
754 struct rtable *list;
755
756 if (process_context && need_resched())
757 cond_resched();
758 rth = rcu_dereference_raw(rt_hash_table[i].chain);
759 if (!rth)
760 continue;
761
762 spin_lock_bh(rt_hash_lock_addr(i));
763
764 list = NULL;
765 pprev = &rt_hash_table[i].chain;
766 rth = rcu_dereference_protected(*pprev,
767 lockdep_is_held(rt_hash_lock_addr(i)));
768
769 while (rth) {
770 next = rcu_dereference_protected(rth->dst.rt_next,
771 lockdep_is_held(rt_hash_lock_addr(i)));
772
773 if (!net ||
774 net_eq(dev_net(rth->dst.dev), net)) {
775 rcu_assign_pointer(*pprev, next);
776 rcu_assign_pointer(rth->dst.rt_next, list);
777 list = rth;
778 } else {
779 pprev = &rth->dst.rt_next;
780 }
781 rth = next;
782 }
783
784 spin_unlock_bh(rt_hash_lock_addr(i));
785
786 for (; list; list = next) {
787 next = rcu_dereference_protected(list->dst.rt_next, 1);
788 rt_free(list);
789 }
790 }
791 }
792
793 /*
794 * While freeing expired entries, we compute average chain length
795 * and standard deviation, using fixed-point arithmetic.
796 * This to have an estimation of rt_chain_length_max
797 * rt_chain_length_max = max(elasticity, AVG + 4*SD)
798 * We use 3 bits for frational part, and 29 (or 61) for magnitude.
799 */
800
801 #define FRACT_BITS 3
802 #define ONE (1UL << FRACT_BITS)
803
804 /*
805 * Given a hash chain and an item in this hash chain,
806 * find if a previous entry has the same hash_inputs
807 * (but differs on tos, mark or oif)
808 * Returns 0 if an alias is found.
809 * Returns ONE if rth has no alias before itself.
810 */
811 static int has_noalias(const struct rtable *head, const struct rtable *rth)
812 {
813 const struct rtable *aux = head;
814
815 while (aux != rth) {
816 if (compare_hash_inputs(aux, rth))
817 return 0;
818 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
819 }
820 return ONE;
821 }
822
823 /*
824 * Pertubation of rt_genid by a small quantity [1..256]
825 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
826 * many times (2^24) without giving recent rt_genid.
827 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
828 */
829 static void rt_cache_invalidate(struct net *net)
830 {
831 unsigned char shuffle;
832
833 get_random_bytes(&shuffle, sizeof(shuffle));
834 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
835 }
836
837 /*
838 * delay < 0 : invalidate cache (fast : entries will be deleted later)
839 * delay >= 0 : invalidate & flush cache (can be long)
840 */
841 void rt_cache_flush(struct net *net, int delay)
842 {
843 rt_cache_invalidate(net);
844 if (delay >= 0)
845 rt_do_flush(net, !in_softirq());
846 }
847
848 /* Flush previous cache invalidated entries from the cache */
849 void rt_cache_flush_batch(struct net *net)
850 {
851 rt_do_flush(net, !in_softirq());
852 }
853
854 static void rt_emergency_hash_rebuild(struct net *net)
855 {
856 if (net_ratelimit())
857 printk(KERN_WARNING "Route hash chain too long!\n");
858 rt_cache_invalidate(net);
859 }
860
861 /*
862 Short description of GC goals.
863
864 We want to build algorithm, which will keep routing cache
865 at some equilibrium point, when number of aged off entries
866 is kept approximately equal to newly generated ones.
867
868 Current expiration strength is variable "expire".
869 We try to adjust it dynamically, so that if networking
870 is idle expires is large enough to keep enough of warm entries,
871 and when load increases it reduces to limit cache size.
872 */
873
874 static int rt_garbage_collect(struct dst_ops *ops)
875 {
876 static unsigned long expire = RT_GC_TIMEOUT;
877 static unsigned long last_gc;
878 static int rover;
879 static int equilibrium;
880 struct rtable *rth;
881 struct rtable __rcu **rthp;
882 unsigned long now = jiffies;
883 int goal;
884 int entries = dst_entries_get_fast(&ipv4_dst_ops);
885
886 /*
887 * Garbage collection is pretty expensive,
888 * do not make it too frequently.
889 */
890
891 RT_CACHE_STAT_INC(gc_total);
892
893 if (now - last_gc < ip_rt_gc_min_interval &&
894 entries < ip_rt_max_size) {
895 RT_CACHE_STAT_INC(gc_ignored);
896 goto out;
897 }
898
899 entries = dst_entries_get_slow(&ipv4_dst_ops);
900 /* Calculate number of entries, which we want to expire now. */
901 goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
902 if (goal <= 0) {
903 if (equilibrium < ipv4_dst_ops.gc_thresh)
904 equilibrium = ipv4_dst_ops.gc_thresh;
905 goal = entries - equilibrium;
906 if (goal > 0) {
907 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
908 goal = entries - equilibrium;
909 }
910 } else {
911 /* We are in dangerous area. Try to reduce cache really
912 * aggressively.
913 */
914 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
915 equilibrium = entries - goal;
916 }
917
918 if (now - last_gc >= ip_rt_gc_min_interval)
919 last_gc = now;
920
921 if (goal <= 0) {
922 equilibrium += goal;
923 goto work_done;
924 }
925
926 do {
927 int i, k;
928
929 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
930 unsigned long tmo = expire;
931
932 k = (k + 1) & rt_hash_mask;
933 rthp = &rt_hash_table[k].chain;
934 spin_lock_bh(rt_hash_lock_addr(k));
935 while ((rth = rcu_dereference_protected(*rthp,
936 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
937 if (!rt_is_expired(rth) &&
938 !rt_may_expire(rth, tmo, expire)) {
939 tmo >>= 1;
940 rthp = &rth->dst.rt_next;
941 continue;
942 }
943 *rthp = rth->dst.rt_next;
944 rt_free(rth);
945 goal--;
946 }
947 spin_unlock_bh(rt_hash_lock_addr(k));
948 if (goal <= 0)
949 break;
950 }
951 rover = k;
952
953 if (goal <= 0)
954 goto work_done;
955
956 /* Goal is not achieved. We stop process if:
957
958 - if expire reduced to zero. Otherwise, expire is halfed.
959 - if table is not full.
960 - if we are called from interrupt.
961 - jiffies check is just fallback/debug loop breaker.
962 We will not spin here for long time in any case.
963 */
964
965 RT_CACHE_STAT_INC(gc_goal_miss);
966
967 if (expire == 0)
968 break;
969
970 expire >>= 1;
971 #if RT_CACHE_DEBUG >= 2
972 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
973 dst_entries_get_fast(&ipv4_dst_ops), goal, i);
974 #endif
975
976 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
977 goto out;
978 } while (!in_softirq() && time_before_eq(jiffies, now));
979
980 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
981 goto out;
982 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
983 goto out;
984 if (net_ratelimit())
985 printk(KERN_WARNING "dst cache overflow\n");
986 RT_CACHE_STAT_INC(gc_dst_overflow);
987 return 1;
988
989 work_done:
990 expire += ip_rt_gc_min_interval;
991 if (expire > ip_rt_gc_timeout ||
992 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
993 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
994 expire = ip_rt_gc_timeout;
995 #if RT_CACHE_DEBUG >= 2
996 printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
997 dst_entries_get_fast(&ipv4_dst_ops), goal, rover);
998 #endif
999 out: return 0;
1000 }
1001
1002 /*
1003 * Returns number of entries in a hash chain that have different hash_inputs
1004 */
1005 static int slow_chain_length(const struct rtable *head)
1006 {
1007 int length = 0;
1008 const struct rtable *rth = head;
1009
1010 while (rth) {
1011 length += has_noalias(head, rth);
1012 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1013 }
1014 return length >> FRACT_BITS;
1015 }
1016
1017 static struct rtable *rt_intern_hash(unsigned hash, struct rtable *rt,
1018 struct sk_buff *skb, int ifindex)
1019 {
1020 struct rtable *rth, *cand;
1021 struct rtable __rcu **rthp, **candp;
1022 unsigned long now;
1023 u32 min_score;
1024 int chain_length;
1025 int attempts = !in_softirq();
1026
1027 restart:
1028 chain_length = 0;
1029 min_score = ~(u32)0;
1030 cand = NULL;
1031 candp = NULL;
1032 now = jiffies;
1033
1034 if (!rt_caching(dev_net(rt->dst.dev))) {
1035 /*
1036 * If we're not caching, just tell the caller we
1037 * were successful and don't touch the route. The
1038 * caller hold the sole reference to the cache entry, and
1039 * it will be released when the caller is done with it.
1040 * If we drop it here, the callers have no way to resolve routes
1041 * when we're not caching. Instead, just point *rp at rt, so
1042 * the caller gets a single use out of the route
1043 * Note that we do rt_free on this new route entry, so that
1044 * once its refcount hits zero, we are still able to reap it
1045 * (Thanks Alexey)
1046 * Note: To avoid expensive rcu stuff for this uncached dst,
1047 * we set DST_NOCACHE so that dst_release() can free dst without
1048 * waiting a grace period.
1049 */
1050
1051 rt->dst.flags |= DST_NOCACHE;
1052 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1053 int err = arp_bind_neighbour(&rt->dst);
1054 if (err) {
1055 if (net_ratelimit())
1056 printk(KERN_WARNING
1057 "Neighbour table failure & not caching routes.\n");
1058 ip_rt_put(rt);
1059 return ERR_PTR(err);
1060 }
1061 }
1062
1063 goto skip_hashing;
1064 }
1065
1066 rthp = &rt_hash_table[hash].chain;
1067
1068 spin_lock_bh(rt_hash_lock_addr(hash));
1069 while ((rth = rcu_dereference_protected(*rthp,
1070 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1071 if (rt_is_expired(rth)) {
1072 *rthp = rth->dst.rt_next;
1073 rt_free(rth);
1074 continue;
1075 }
1076 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1077 /* Put it first */
1078 *rthp = rth->dst.rt_next;
1079 /*
1080 * Since lookup is lockfree, the deletion
1081 * must be visible to another weakly ordered CPU before
1082 * the insertion at the start of the hash chain.
1083 */
1084 rcu_assign_pointer(rth->dst.rt_next,
1085 rt_hash_table[hash].chain);
1086 /*
1087 * Since lookup is lockfree, the update writes
1088 * must be ordered for consistency on SMP.
1089 */
1090 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1091
1092 dst_use(&rth->dst, now);
1093 spin_unlock_bh(rt_hash_lock_addr(hash));
1094
1095 rt_drop(rt);
1096 if (skb)
1097 skb_dst_set(skb, &rth->dst);
1098 return rth;
1099 }
1100
1101 if (!atomic_read(&rth->dst.__refcnt)) {
1102 u32 score = rt_score(rth);
1103
1104 if (score <= min_score) {
1105 cand = rth;
1106 candp = rthp;
1107 min_score = score;
1108 }
1109 }
1110
1111 chain_length++;
1112
1113 rthp = &rth->dst.rt_next;
1114 }
1115
1116 if (cand) {
1117 /* ip_rt_gc_elasticity used to be average length of chain
1118 * length, when exceeded gc becomes really aggressive.
1119 *
1120 * The second limit is less certain. At the moment it allows
1121 * only 2 entries per bucket. We will see.
1122 */
1123 if (chain_length > ip_rt_gc_elasticity) {
1124 *candp = cand->dst.rt_next;
1125 rt_free(cand);
1126 }
1127 } else {
1128 if (chain_length > rt_chain_length_max &&
1129 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1130 struct net *net = dev_net(rt->dst.dev);
1131 int num = ++net->ipv4.current_rt_cache_rebuild_count;
1132 if (!rt_caching(net)) {
1133 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1134 rt->dst.dev->name, num);
1135 }
1136 rt_emergency_hash_rebuild(net);
1137 spin_unlock_bh(rt_hash_lock_addr(hash));
1138
1139 hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1140 ifindex, rt_genid(net));
1141 goto restart;
1142 }
1143 }
1144
1145 /* Try to bind route to arp only if it is output
1146 route or unicast forwarding path.
1147 */
1148 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1149 int err = arp_bind_neighbour(&rt->dst);
1150 if (err) {
1151 spin_unlock_bh(rt_hash_lock_addr(hash));
1152
1153 if (err != -ENOBUFS) {
1154 rt_drop(rt);
1155 return ERR_PTR(err);
1156 }
1157
1158 /* Neighbour tables are full and nothing
1159 can be released. Try to shrink route cache,
1160 it is most likely it holds some neighbour records.
1161 */
1162 if (attempts-- > 0) {
1163 int saved_elasticity = ip_rt_gc_elasticity;
1164 int saved_int = ip_rt_gc_min_interval;
1165 ip_rt_gc_elasticity = 1;
1166 ip_rt_gc_min_interval = 0;
1167 rt_garbage_collect(&ipv4_dst_ops);
1168 ip_rt_gc_min_interval = saved_int;
1169 ip_rt_gc_elasticity = saved_elasticity;
1170 goto restart;
1171 }
1172
1173 if (net_ratelimit())
1174 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1175 rt_drop(rt);
1176 return ERR_PTR(-ENOBUFS);
1177 }
1178 }
1179
1180 rt->dst.rt_next = rt_hash_table[hash].chain;
1181
1182 #if RT_CACHE_DEBUG >= 2
1183 if (rt->dst.rt_next) {
1184 struct rtable *trt;
1185 printk(KERN_DEBUG "rt_cache @%02x: %pI4",
1186 hash, &rt->rt_dst);
1187 for (trt = rt->dst.rt_next; trt; trt = trt->dst.rt_next)
1188 printk(" . %pI4", &trt->rt_dst);
1189 printk("\n");
1190 }
1191 #endif
1192 /*
1193 * Since lookup is lockfree, we must make sure
1194 * previous writes to rt are comitted to memory
1195 * before making rt visible to other CPUS.
1196 */
1197 rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1198
1199 spin_unlock_bh(rt_hash_lock_addr(hash));
1200
1201 skip_hashing:
1202 if (skb)
1203 skb_dst_set(skb, &rt->dst);
1204 return rt;
1205 }
1206
1207 static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1208
1209 static u32 rt_peer_genid(void)
1210 {
1211 return atomic_read(&__rt_peer_genid);
1212 }
1213
1214 void rt_bind_peer(struct rtable *rt, int create)
1215 {
1216 struct inet_peer *peer;
1217
1218 peer = inet_getpeer_v4(rt->rt_dst, create);
1219
1220 if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1221 inet_putpeer(peer);
1222 else
1223 rt->rt_peer_genid = rt_peer_genid();
1224 }
1225
1226 /*
1227 * Peer allocation may fail only in serious out-of-memory conditions. However
1228 * we still can generate some output.
1229 * Random ID selection looks a bit dangerous because we have no chances to
1230 * select ID being unique in a reasonable period of time.
1231 * But broken packet identifier may be better than no packet at all.
1232 */
1233 static void ip_select_fb_ident(struct iphdr *iph)
1234 {
1235 static DEFINE_SPINLOCK(ip_fb_id_lock);
1236 static u32 ip_fallback_id;
1237 u32 salt;
1238
1239 spin_lock_bh(&ip_fb_id_lock);
1240 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1241 iph->id = htons(salt & 0xFFFF);
1242 ip_fallback_id = salt;
1243 spin_unlock_bh(&ip_fb_id_lock);
1244 }
1245
1246 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1247 {
1248 struct rtable *rt = (struct rtable *) dst;
1249
1250 if (rt) {
1251 if (rt->peer == NULL)
1252 rt_bind_peer(rt, 1);
1253
1254 /* If peer is attached to destination, it is never detached,
1255 so that we need not to grab a lock to dereference it.
1256 */
1257 if (rt->peer) {
1258 iph->id = htons(inet_getid(rt->peer, more));
1259 return;
1260 }
1261 } else
1262 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1263 __builtin_return_address(0));
1264
1265 ip_select_fb_ident(iph);
1266 }
1267 EXPORT_SYMBOL(__ip_select_ident);
1268
1269 static void rt_del(unsigned hash, struct rtable *rt)
1270 {
1271 struct rtable __rcu **rthp;
1272 struct rtable *aux;
1273
1274 rthp = &rt_hash_table[hash].chain;
1275 spin_lock_bh(rt_hash_lock_addr(hash));
1276 ip_rt_put(rt);
1277 while ((aux = rcu_dereference_protected(*rthp,
1278 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1279 if (aux == rt || rt_is_expired(aux)) {
1280 *rthp = aux->dst.rt_next;
1281 rt_free(aux);
1282 continue;
1283 }
1284 rthp = &aux->dst.rt_next;
1285 }
1286 spin_unlock_bh(rt_hash_lock_addr(hash));
1287 }
1288
1289 /* called in rcu_read_lock() section */
1290 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1291 __be32 saddr, struct net_device *dev)
1292 {
1293 struct in_device *in_dev = __in_dev_get_rcu(dev);
1294 struct inet_peer *peer;
1295 struct net *net;
1296
1297 if (!in_dev)
1298 return;
1299
1300 net = dev_net(dev);
1301 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1302 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1303 ipv4_is_zeronet(new_gw))
1304 goto reject_redirect;
1305
1306 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1307 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1308 goto reject_redirect;
1309 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1310 goto reject_redirect;
1311 } else {
1312 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1313 goto reject_redirect;
1314 }
1315
1316 peer = inet_getpeer_v4(daddr, 1);
1317 if (peer) {
1318 peer->redirect_learned.a4 = new_gw;
1319
1320 inet_putpeer(peer);
1321
1322 atomic_inc(&__rt_peer_genid);
1323 }
1324 return;
1325
1326 reject_redirect:
1327 #ifdef CONFIG_IP_ROUTE_VERBOSE
1328 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1329 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1330 " Advised path = %pI4 -> %pI4\n",
1331 &old_gw, dev->name, &new_gw,
1332 &saddr, &daddr);
1333 #endif
1334 ;
1335 }
1336
1337 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1338 {
1339 struct rtable *rt = (struct rtable *)dst;
1340 struct dst_entry *ret = dst;
1341
1342 if (rt) {
1343 if (dst->obsolete > 0) {
1344 ip_rt_put(rt);
1345 ret = NULL;
1346 } else if (rt->rt_flags & RTCF_REDIRECTED) {
1347 unsigned hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1348 rt->rt_oif,
1349 rt_genid(dev_net(dst->dev)));
1350 #if RT_CACHE_DEBUG >= 1
1351 printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1352 &rt->rt_dst, rt->rt_tos);
1353 #endif
1354 rt_del(hash, rt);
1355 ret = NULL;
1356 } else if (rt->peer &&
1357 rt->peer->pmtu_expires &&
1358 time_after_eq(jiffies, rt->peer->pmtu_expires)) {
1359 unsigned long orig = rt->peer->pmtu_expires;
1360
1361 if (cmpxchg(&rt->peer->pmtu_expires, orig, 0) == orig)
1362 dst_metric_set(dst, RTAX_MTU,
1363 rt->peer->pmtu_orig);
1364 }
1365 }
1366 return ret;
1367 }
1368
1369 /*
1370 * Algorithm:
1371 * 1. The first ip_rt_redirect_number redirects are sent
1372 * with exponential backoff, then we stop sending them at all,
1373 * assuming that the host ignores our redirects.
1374 * 2. If we did not see packets requiring redirects
1375 * during ip_rt_redirect_silence, we assume that the host
1376 * forgot redirected route and start to send redirects again.
1377 *
1378 * This algorithm is much cheaper and more intelligent than dumb load limiting
1379 * in icmp.c.
1380 *
1381 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1382 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1383 */
1384
1385 void ip_rt_send_redirect(struct sk_buff *skb)
1386 {
1387 struct rtable *rt = skb_rtable(skb);
1388 struct in_device *in_dev;
1389 struct inet_peer *peer;
1390 int log_martians;
1391
1392 rcu_read_lock();
1393 in_dev = __in_dev_get_rcu(rt->dst.dev);
1394 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1395 rcu_read_unlock();
1396 return;
1397 }
1398 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1399 rcu_read_unlock();
1400
1401 if (!rt->peer)
1402 rt_bind_peer(rt, 1);
1403 peer = rt->peer;
1404 if (!peer) {
1405 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1406 return;
1407 }
1408
1409 /* No redirected packets during ip_rt_redirect_silence;
1410 * reset the algorithm.
1411 */
1412 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1413 peer->rate_tokens = 0;
1414
1415 /* Too many ignored redirects; do not send anything
1416 * set dst.rate_last to the last seen redirected packet.
1417 */
1418 if (peer->rate_tokens >= ip_rt_redirect_number) {
1419 peer->rate_last = jiffies;
1420 return;
1421 }
1422
1423 /* Check for load limit; set rate_last to the latest sent
1424 * redirect.
1425 */
1426 if (peer->rate_tokens == 0 ||
1427 time_after(jiffies,
1428 (peer->rate_last +
1429 (ip_rt_redirect_load << peer->rate_tokens)))) {
1430 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1431 peer->rate_last = jiffies;
1432 ++peer->rate_tokens;
1433 #ifdef CONFIG_IP_ROUTE_VERBOSE
1434 if (log_martians &&
1435 peer->rate_tokens == ip_rt_redirect_number &&
1436 net_ratelimit())
1437 printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1438 &rt->rt_src, rt->rt_iif,
1439 &rt->rt_dst, &rt->rt_gateway);
1440 #endif
1441 }
1442 }
1443
1444 static int ip_error(struct sk_buff *skb)
1445 {
1446 struct rtable *rt = skb_rtable(skb);
1447 struct inet_peer *peer;
1448 unsigned long now;
1449 bool send;
1450 int code;
1451
1452 switch (rt->dst.error) {
1453 case EINVAL:
1454 default:
1455 goto out;
1456 case EHOSTUNREACH:
1457 code = ICMP_HOST_UNREACH;
1458 break;
1459 case ENETUNREACH:
1460 code = ICMP_NET_UNREACH;
1461 IP_INC_STATS_BH(dev_net(rt->dst.dev),
1462 IPSTATS_MIB_INNOROUTES);
1463 break;
1464 case EACCES:
1465 code = ICMP_PKT_FILTERED;
1466 break;
1467 }
1468
1469 if (!rt->peer)
1470 rt_bind_peer(rt, 1);
1471 peer = rt->peer;
1472
1473 send = true;
1474 if (peer) {
1475 now = jiffies;
1476 peer->rate_tokens += now - peer->rate_last;
1477 if (peer->rate_tokens > ip_rt_error_burst)
1478 peer->rate_tokens = ip_rt_error_burst;
1479 peer->rate_last = now;
1480 if (peer->rate_tokens >= ip_rt_error_cost)
1481 peer->rate_tokens -= ip_rt_error_cost;
1482 else
1483 send = false;
1484 }
1485 if (send)
1486 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1487
1488 out: kfree_skb(skb);
1489 return 0;
1490 }
1491
1492 /*
1493 * The last two values are not from the RFC but
1494 * are needed for AMPRnet AX.25 paths.
1495 */
1496
1497 static const unsigned short mtu_plateau[] =
1498 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1499
1500 static inline unsigned short guess_mtu(unsigned short old_mtu)
1501 {
1502 int i;
1503
1504 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1505 if (old_mtu > mtu_plateau[i])
1506 return mtu_plateau[i];
1507 return 68;
1508 }
1509
1510 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1511 unsigned short new_mtu,
1512 struct net_device *dev)
1513 {
1514 unsigned short old_mtu = ntohs(iph->tot_len);
1515 unsigned short est_mtu = 0;
1516 struct inet_peer *peer;
1517
1518 peer = inet_getpeer_v4(iph->daddr, 1);
1519 if (peer) {
1520 unsigned short mtu = new_mtu;
1521
1522 if (new_mtu < 68 || new_mtu >= old_mtu) {
1523 /* BSD 4.2 derived systems incorrectly adjust
1524 * tot_len by the IP header length, and report
1525 * a zero MTU in the ICMP message.
1526 */
1527 if (mtu == 0 &&
1528 old_mtu >= 68 + (iph->ihl << 2))
1529 old_mtu -= iph->ihl << 2;
1530 mtu = guess_mtu(old_mtu);
1531 }
1532
1533 if (mtu < ip_rt_min_pmtu)
1534 mtu = ip_rt_min_pmtu;
1535 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1536 unsigned long pmtu_expires;
1537
1538 pmtu_expires = jiffies + ip_rt_mtu_expires;
1539 if (!pmtu_expires)
1540 pmtu_expires = 1UL;
1541
1542 est_mtu = mtu;
1543 peer->pmtu_learned = mtu;
1544 peer->pmtu_expires = pmtu_expires;
1545 }
1546
1547 inet_putpeer(peer);
1548
1549 atomic_inc(&__rt_peer_genid);
1550 }
1551 return est_mtu ? : new_mtu;
1552 }
1553
1554 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1555 {
1556 unsigned long expires = peer->pmtu_expires;
1557
1558 if (time_before(jiffies, expires)) {
1559 u32 orig_dst_mtu = dst_mtu(dst);
1560 if (peer->pmtu_learned < orig_dst_mtu) {
1561 if (!peer->pmtu_orig)
1562 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1563 dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1564 }
1565 } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1566 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1567 }
1568
1569 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1570 {
1571 struct rtable *rt = (struct rtable *) dst;
1572 struct inet_peer *peer;
1573
1574 dst_confirm(dst);
1575
1576 if (!rt->peer)
1577 rt_bind_peer(rt, 1);
1578 peer = rt->peer;
1579 if (peer) {
1580 if (mtu < ip_rt_min_pmtu)
1581 mtu = ip_rt_min_pmtu;
1582 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1583 unsigned long pmtu_expires;
1584
1585 pmtu_expires = jiffies + ip_rt_mtu_expires;
1586 if (!pmtu_expires)
1587 pmtu_expires = 1UL;
1588
1589 peer->pmtu_learned = mtu;
1590 peer->pmtu_expires = pmtu_expires;
1591
1592 atomic_inc(&__rt_peer_genid);
1593 rt->rt_peer_genid = rt_peer_genid();
1594 }
1595 check_peer_pmtu(dst, peer);
1596
1597 inet_putpeer(peer);
1598 }
1599 }
1600
1601 static int check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1602 {
1603 struct rtable *rt = (struct rtable *) dst;
1604 __be32 orig_gw = rt->rt_gateway;
1605
1606 dst_confirm(&rt->dst);
1607
1608 neigh_release(rt->dst.neighbour);
1609 rt->dst.neighbour = NULL;
1610
1611 rt->rt_gateway = peer->redirect_learned.a4;
1612 if (arp_bind_neighbour(&rt->dst) ||
1613 !(rt->dst.neighbour->nud_state & NUD_VALID)) {
1614 if (rt->dst.neighbour)
1615 neigh_event_send(rt->dst.neighbour, NULL);
1616 rt->rt_gateway = orig_gw;
1617 return -EAGAIN;
1618 } else {
1619 rt->rt_flags |= RTCF_REDIRECTED;
1620 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE,
1621 rt->dst.neighbour);
1622 }
1623 return 0;
1624 }
1625
1626 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1627 {
1628 struct rtable *rt = (struct rtable *) dst;
1629
1630 if (rt_is_expired(rt))
1631 return NULL;
1632 if (rt->rt_peer_genid != rt_peer_genid()) {
1633 struct inet_peer *peer;
1634
1635 if (!rt->peer)
1636 rt_bind_peer(rt, 0);
1637
1638 peer = rt->peer;
1639 if (peer && peer->pmtu_expires)
1640 check_peer_pmtu(dst, peer);
1641
1642 if (peer && peer->redirect_learned.a4 &&
1643 peer->redirect_learned.a4 != rt->rt_gateway) {
1644 if (check_peer_redir(dst, peer))
1645 return NULL;
1646 }
1647
1648 rt->rt_peer_genid = rt_peer_genid();
1649 }
1650 return dst;
1651 }
1652
1653 static void ipv4_dst_destroy(struct dst_entry *dst)
1654 {
1655 struct rtable *rt = (struct rtable *) dst;
1656 struct inet_peer *peer = rt->peer;
1657
1658 if (rt->fi) {
1659 fib_info_put(rt->fi);
1660 rt->fi = NULL;
1661 }
1662 if (peer) {
1663 rt->peer = NULL;
1664 inet_putpeer(peer);
1665 }
1666 }
1667
1668
1669 static void ipv4_link_failure(struct sk_buff *skb)
1670 {
1671 struct rtable *rt;
1672
1673 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1674
1675 rt = skb_rtable(skb);
1676 if (rt &&
1677 rt->peer &&
1678 rt->peer->pmtu_expires) {
1679 unsigned long orig = rt->peer->pmtu_expires;
1680
1681 if (cmpxchg(&rt->peer->pmtu_expires, orig, 0) == orig)
1682 dst_metric_set(&rt->dst, RTAX_MTU, rt->peer->pmtu_orig);
1683 }
1684 }
1685
1686 static int ip_rt_bug(struct sk_buff *skb)
1687 {
1688 printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1689 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1690 skb->dev ? skb->dev->name : "?");
1691 kfree_skb(skb);
1692 return 0;
1693 }
1694
1695 /*
1696 We do not cache source address of outgoing interface,
1697 because it is used only by IP RR, TS and SRR options,
1698 so that it out of fast path.
1699
1700 BTW remember: "addr" is allowed to be not aligned
1701 in IP options!
1702 */
1703
1704 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1705 {
1706 __be32 src;
1707 struct fib_result res;
1708
1709 if (rt_is_output_route(rt))
1710 src = rt->rt_src;
1711 else {
1712 struct flowi4 fl4 = {
1713 .daddr = rt->rt_key_dst,
1714 .saddr = rt->rt_key_src,
1715 .flowi4_tos = rt->rt_tos,
1716 .flowi4_oif = rt->rt_oif,
1717 .flowi4_iif = rt->rt_iif,
1718 .flowi4_mark = rt->rt_mark,
1719 };
1720
1721 rcu_read_lock();
1722 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1723 src = FIB_RES_PREFSRC(res);
1724 else
1725 src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1726 RT_SCOPE_UNIVERSE);
1727 rcu_read_unlock();
1728 }
1729 memcpy(addr, &src, 4);
1730 }
1731
1732 #ifdef CONFIG_IP_ROUTE_CLASSID
1733 static void set_class_tag(struct rtable *rt, u32 tag)
1734 {
1735 if (!(rt->dst.tclassid & 0xFFFF))
1736 rt->dst.tclassid |= tag & 0xFFFF;
1737 if (!(rt->dst.tclassid & 0xFFFF0000))
1738 rt->dst.tclassid |= tag & 0xFFFF0000;
1739 }
1740 #endif
1741
1742 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1743 {
1744 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1745
1746 if (advmss == 0) {
1747 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1748 ip_rt_min_advmss);
1749 if (advmss > 65535 - 40)
1750 advmss = 65535 - 40;
1751 }
1752 return advmss;
1753 }
1754
1755 static unsigned int ipv4_default_mtu(const struct dst_entry *dst)
1756 {
1757 unsigned int mtu = dst->dev->mtu;
1758
1759 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1760 const struct rtable *rt = (const struct rtable *) dst;
1761
1762 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1763 mtu = 576;
1764 }
1765
1766 if (mtu > IP_MAX_MTU)
1767 mtu = IP_MAX_MTU;
1768
1769 return mtu;
1770 }
1771
1772 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *oldflp4,
1773 struct fib_info *fi)
1774 {
1775 struct inet_peer *peer;
1776 int create = 0;
1777
1778 /* If a peer entry exists for this destination, we must hook
1779 * it up in order to get at cached metrics.
1780 */
1781 if (oldflp4 && (oldflp4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS))
1782 create = 1;
1783
1784 rt->peer = peer = inet_getpeer_v4(rt->rt_dst, create);
1785 if (peer) {
1786 rt->rt_peer_genid = rt_peer_genid();
1787 if (inet_metrics_new(peer))
1788 memcpy(peer->metrics, fi->fib_metrics,
1789 sizeof(u32) * RTAX_MAX);
1790 dst_init_metrics(&rt->dst, peer->metrics, false);
1791
1792 if (peer->pmtu_expires)
1793 check_peer_pmtu(&rt->dst, peer);
1794 if (peer->redirect_learned.a4 &&
1795 peer->redirect_learned.a4 != rt->rt_gateway) {
1796 rt->rt_gateway = peer->redirect_learned.a4;
1797 rt->rt_flags |= RTCF_REDIRECTED;
1798 }
1799 } else {
1800 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1801 rt->fi = fi;
1802 atomic_inc(&fi->fib_clntref);
1803 }
1804 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1805 }
1806 }
1807
1808 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *oldflp4,
1809 const struct fib_result *res,
1810 struct fib_info *fi, u16 type, u32 itag)
1811 {
1812 struct dst_entry *dst = &rt->dst;
1813
1814 if (fi) {
1815 if (FIB_RES_GW(*res) &&
1816 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1817 rt->rt_gateway = FIB_RES_GW(*res);
1818 rt_init_metrics(rt, oldflp4, fi);
1819 #ifdef CONFIG_IP_ROUTE_CLASSID
1820 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1821 #endif
1822 }
1823
1824 if (dst_mtu(dst) > IP_MAX_MTU)
1825 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1826 if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40)
1827 dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40);
1828
1829 #ifdef CONFIG_IP_ROUTE_CLASSID
1830 #ifdef CONFIG_IP_MULTIPLE_TABLES
1831 set_class_tag(rt, fib_rules_tclass(res));
1832 #endif
1833 set_class_tag(rt, itag);
1834 #endif
1835 rt->rt_type = type;
1836 }
1837
1838 static struct rtable *rt_dst_alloc(bool nopolicy, bool noxfrm)
1839 {
1840 struct rtable *rt = dst_alloc(&ipv4_dst_ops, 1);
1841 if (rt) {
1842 rt->dst.obsolete = -1;
1843
1844 rt->dst.flags = DST_HOST |
1845 (nopolicy ? DST_NOPOLICY : 0) |
1846 (noxfrm ? DST_NOXFRM : 0);
1847 }
1848 return rt;
1849 }
1850
1851 /* called in rcu_read_lock() section */
1852 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1853 u8 tos, struct net_device *dev, int our)
1854 {
1855 unsigned int hash;
1856 struct rtable *rth;
1857 __be32 spec_dst;
1858 struct in_device *in_dev = __in_dev_get_rcu(dev);
1859 u32 itag = 0;
1860 int err;
1861
1862 /* Primary sanity checks. */
1863
1864 if (in_dev == NULL)
1865 return -EINVAL;
1866
1867 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1868 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1869 goto e_inval;
1870
1871 if (ipv4_is_zeronet(saddr)) {
1872 if (!ipv4_is_local_multicast(daddr))
1873 goto e_inval;
1874 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1875 } else {
1876 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1877 &itag, 0);
1878 if (err < 0)
1879 goto e_err;
1880 }
1881 rth = rt_dst_alloc(IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1882 if (!rth)
1883 goto e_nobufs;
1884
1885 rth->dst.output = ip_rt_bug;
1886
1887 rth->rt_key_dst = daddr;
1888 rth->rt_dst = daddr;
1889 rth->rt_tos = tos;
1890 rth->rt_mark = skb->mark;
1891 rth->rt_key_src = saddr;
1892 rth->rt_src = saddr;
1893 #ifdef CONFIG_IP_ROUTE_CLASSID
1894 rth->dst.tclassid = itag;
1895 #endif
1896 rth->rt_iif = dev->ifindex;
1897 rth->dst.dev = init_net.loopback_dev;
1898 dev_hold(rth->dst.dev);
1899 rth->rt_oif = 0;
1900 rth->rt_gateway = daddr;
1901 rth->rt_spec_dst= spec_dst;
1902 rth->rt_genid = rt_genid(dev_net(dev));
1903 rth->rt_flags = RTCF_MULTICAST;
1904 rth->rt_type = RTN_MULTICAST;
1905 if (our) {
1906 rth->dst.input= ip_local_deliver;
1907 rth->rt_flags |= RTCF_LOCAL;
1908 }
1909
1910 #ifdef CONFIG_IP_MROUTE
1911 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1912 rth->dst.input = ip_mr_input;
1913 #endif
1914 RT_CACHE_STAT_INC(in_slow_mc);
1915
1916 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1917 rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
1918 err = 0;
1919 if (IS_ERR(rth))
1920 err = PTR_ERR(rth);
1921
1922 e_nobufs:
1923 return -ENOBUFS;
1924 e_inval:
1925 return -EINVAL;
1926 e_err:
1927 return err;
1928 }
1929
1930
1931 static void ip_handle_martian_source(struct net_device *dev,
1932 struct in_device *in_dev,
1933 struct sk_buff *skb,
1934 __be32 daddr,
1935 __be32 saddr)
1936 {
1937 RT_CACHE_STAT_INC(in_martian_src);
1938 #ifdef CONFIG_IP_ROUTE_VERBOSE
1939 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1940 /*
1941 * RFC1812 recommendation, if source is martian,
1942 * the only hint is MAC header.
1943 */
1944 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1945 &daddr, &saddr, dev->name);
1946 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1947 int i;
1948 const unsigned char *p = skb_mac_header(skb);
1949 printk(KERN_WARNING "ll header: ");
1950 for (i = 0; i < dev->hard_header_len; i++, p++) {
1951 printk("%02x", *p);
1952 if (i < (dev->hard_header_len - 1))
1953 printk(":");
1954 }
1955 printk("\n");
1956 }
1957 }
1958 #endif
1959 }
1960
1961 /* called in rcu_read_lock() section */
1962 static int __mkroute_input(struct sk_buff *skb,
1963 const struct fib_result *res,
1964 struct in_device *in_dev,
1965 __be32 daddr, __be32 saddr, u32 tos,
1966 struct rtable **result)
1967 {
1968 struct rtable *rth;
1969 int err;
1970 struct in_device *out_dev;
1971 unsigned int flags = 0;
1972 __be32 spec_dst;
1973 u32 itag;
1974
1975 /* get a working reference to the output device */
1976 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1977 if (out_dev == NULL) {
1978 if (net_ratelimit())
1979 printk(KERN_CRIT "Bug in ip_route_input" \
1980 "_slow(). Please, report\n");
1981 return -EINVAL;
1982 }
1983
1984
1985 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1986 in_dev->dev, &spec_dst, &itag, skb->mark);
1987 if (err < 0) {
1988 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1989 saddr);
1990
1991 goto cleanup;
1992 }
1993
1994 if (err)
1995 flags |= RTCF_DIRECTSRC;
1996
1997 if (out_dev == in_dev && err &&
1998 (IN_DEV_SHARED_MEDIA(out_dev) ||
1999 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2000 flags |= RTCF_DOREDIRECT;
2001
2002 if (skb->protocol != htons(ETH_P_IP)) {
2003 /* Not IP (i.e. ARP). Do not create route, if it is
2004 * invalid for proxy arp. DNAT routes are always valid.
2005 *
2006 * Proxy arp feature have been extended to allow, ARP
2007 * replies back to the same interface, to support
2008 * Private VLAN switch technologies. See arp.c.
2009 */
2010 if (out_dev == in_dev &&
2011 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2012 err = -EINVAL;
2013 goto cleanup;
2014 }
2015 }
2016
2017 rth = rt_dst_alloc(IN_DEV_CONF_GET(in_dev, NOPOLICY),
2018 IN_DEV_CONF_GET(out_dev, NOXFRM));
2019 if (!rth) {
2020 err = -ENOBUFS;
2021 goto cleanup;
2022 }
2023
2024 rth->rt_key_dst = daddr;
2025 rth->rt_dst = daddr;
2026 rth->rt_tos = tos;
2027 rth->rt_mark = skb->mark;
2028 rth->rt_key_src = saddr;
2029 rth->rt_src = saddr;
2030 rth->rt_gateway = daddr;
2031 rth->rt_iif = in_dev->dev->ifindex;
2032 rth->dst.dev = (out_dev)->dev;
2033 dev_hold(rth->dst.dev);
2034 rth->rt_oif = 0;
2035 rth->rt_spec_dst= spec_dst;
2036
2037 rth->dst.input = ip_forward;
2038 rth->dst.output = ip_output;
2039 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2040
2041 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2042
2043 rth->rt_flags = flags;
2044
2045 *result = rth;
2046 err = 0;
2047 cleanup:
2048 return err;
2049 }
2050
2051 static int ip_mkroute_input(struct sk_buff *skb,
2052 struct fib_result *res,
2053 const struct flowi4 *fl4,
2054 struct in_device *in_dev,
2055 __be32 daddr, __be32 saddr, u32 tos)
2056 {
2057 struct rtable* rth = NULL;
2058 int err;
2059 unsigned hash;
2060
2061 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2062 if (res->fi && res->fi->fib_nhs > 1)
2063 fib_select_multipath(res);
2064 #endif
2065
2066 /* create a routing cache entry */
2067 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2068 if (err)
2069 return err;
2070
2071 /* put it into the cache */
2072 hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2073 rt_genid(dev_net(rth->dst.dev)));
2074 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2075 if (IS_ERR(rth))
2076 return PTR_ERR(rth);
2077 return 0;
2078 }
2079
2080 /*
2081 * NOTE. We drop all the packets that has local source
2082 * addresses, because every properly looped back packet
2083 * must have correct destination already attached by output routine.
2084 *
2085 * Such approach solves two big problems:
2086 * 1. Not simplex devices are handled properly.
2087 * 2. IP spoofing attempts are filtered with 100% of guarantee.
2088 * called with rcu_read_lock()
2089 */
2090
2091 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2092 u8 tos, struct net_device *dev)
2093 {
2094 struct fib_result res;
2095 struct in_device *in_dev = __in_dev_get_rcu(dev);
2096 struct flowi4 fl4;
2097 unsigned flags = 0;
2098 u32 itag = 0;
2099 struct rtable * rth;
2100 unsigned hash;
2101 __be32 spec_dst;
2102 int err = -EINVAL;
2103 struct net * net = dev_net(dev);
2104
2105 /* IP on this device is disabled. */
2106
2107 if (!in_dev)
2108 goto out;
2109
2110 /* Check for the most weird martians, which can be not detected
2111 by fib_lookup.
2112 */
2113
2114 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2115 ipv4_is_loopback(saddr))
2116 goto martian_source;
2117
2118 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2119 goto brd_input;
2120
2121 /* Accept zero addresses only to limited broadcast;
2122 * I even do not know to fix it or not. Waiting for complains :-)
2123 */
2124 if (ipv4_is_zeronet(saddr))
2125 goto martian_source;
2126
2127 if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2128 goto martian_destination;
2129
2130 /*
2131 * Now we are ready to route packet.
2132 */
2133 fl4.flowi4_oif = 0;
2134 fl4.flowi4_iif = dev->ifindex;
2135 fl4.flowi4_mark = skb->mark;
2136 fl4.flowi4_tos = tos;
2137 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2138 fl4.daddr = daddr;
2139 fl4.saddr = saddr;
2140 err = fib_lookup(net, &fl4, &res);
2141 if (err != 0) {
2142 if (!IN_DEV_FORWARD(in_dev))
2143 goto e_hostunreach;
2144 goto no_route;
2145 }
2146
2147 RT_CACHE_STAT_INC(in_slow_tot);
2148
2149 if (res.type == RTN_BROADCAST)
2150 goto brd_input;
2151
2152 if (res.type == RTN_LOCAL) {
2153 err = fib_validate_source(saddr, daddr, tos,
2154 net->loopback_dev->ifindex,
2155 dev, &spec_dst, &itag, skb->mark);
2156 if (err < 0)
2157 goto martian_source_keep_err;
2158 if (err)
2159 flags |= RTCF_DIRECTSRC;
2160 spec_dst = daddr;
2161 goto local_input;
2162 }
2163
2164 if (!IN_DEV_FORWARD(in_dev))
2165 goto e_hostunreach;
2166 if (res.type != RTN_UNICAST)
2167 goto martian_destination;
2168
2169 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2170 out: return err;
2171
2172 brd_input:
2173 if (skb->protocol != htons(ETH_P_IP))
2174 goto e_inval;
2175
2176 if (ipv4_is_zeronet(saddr))
2177 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2178 else {
2179 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2180 &itag, skb->mark);
2181 if (err < 0)
2182 goto martian_source_keep_err;
2183 if (err)
2184 flags |= RTCF_DIRECTSRC;
2185 }
2186 flags |= RTCF_BROADCAST;
2187 res.type = RTN_BROADCAST;
2188 RT_CACHE_STAT_INC(in_brd);
2189
2190 local_input:
2191 rth = rt_dst_alloc(IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2192 if (!rth)
2193 goto e_nobufs;
2194
2195 rth->dst.output= ip_rt_bug;
2196 rth->rt_genid = rt_genid(net);
2197
2198 rth->rt_key_dst = daddr;
2199 rth->rt_dst = daddr;
2200 rth->rt_tos = tos;
2201 rth->rt_mark = skb->mark;
2202 rth->rt_key_src = saddr;
2203 rth->rt_src = saddr;
2204 #ifdef CONFIG_IP_ROUTE_CLASSID
2205 rth->dst.tclassid = itag;
2206 #endif
2207 rth->rt_iif = dev->ifindex;
2208 rth->dst.dev = net->loopback_dev;
2209 dev_hold(rth->dst.dev);
2210 rth->rt_gateway = daddr;
2211 rth->rt_spec_dst= spec_dst;
2212 rth->dst.input= ip_local_deliver;
2213 rth->rt_flags = flags|RTCF_LOCAL;
2214 if (res.type == RTN_UNREACHABLE) {
2215 rth->dst.input= ip_error;
2216 rth->dst.error= -err;
2217 rth->rt_flags &= ~RTCF_LOCAL;
2218 }
2219 rth->rt_type = res.type;
2220 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2221 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2222 err = 0;
2223 if (IS_ERR(rth))
2224 err = PTR_ERR(rth);
2225 goto out;
2226
2227 no_route:
2228 RT_CACHE_STAT_INC(in_no_route);
2229 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2230 res.type = RTN_UNREACHABLE;
2231 if (err == -ESRCH)
2232 err = -ENETUNREACH;
2233 goto local_input;
2234
2235 /*
2236 * Do not cache martian addresses: they should be logged (RFC1812)
2237 */
2238 martian_destination:
2239 RT_CACHE_STAT_INC(in_martian_dst);
2240 #ifdef CONFIG_IP_ROUTE_VERBOSE
2241 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2242 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2243 &daddr, &saddr, dev->name);
2244 #endif
2245
2246 e_hostunreach:
2247 err = -EHOSTUNREACH;
2248 goto out;
2249
2250 e_inval:
2251 err = -EINVAL;
2252 goto out;
2253
2254 e_nobufs:
2255 err = -ENOBUFS;
2256 goto out;
2257
2258 martian_source:
2259 err = -EINVAL;
2260 martian_source_keep_err:
2261 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2262 goto out;
2263 }
2264
2265 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2266 u8 tos, struct net_device *dev, bool noref)
2267 {
2268 struct rtable * rth;
2269 unsigned hash;
2270 int iif = dev->ifindex;
2271 struct net *net;
2272 int res;
2273
2274 net = dev_net(dev);
2275
2276 rcu_read_lock();
2277
2278 if (!rt_caching(net))
2279 goto skip_cache;
2280
2281 tos &= IPTOS_RT_MASK;
2282 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2283
2284 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2285 rth = rcu_dereference(rth->dst.rt_next)) {
2286 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2287 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2288 (rth->rt_iif ^ iif) |
2289 rth->rt_oif |
2290 (rth->rt_tos ^ tos)) == 0 &&
2291 rth->rt_mark == skb->mark &&
2292 net_eq(dev_net(rth->dst.dev), net) &&
2293 !rt_is_expired(rth)) {
2294 if (noref) {
2295 dst_use_noref(&rth->dst, jiffies);
2296 skb_dst_set_noref(skb, &rth->dst);
2297 } else {
2298 dst_use(&rth->dst, jiffies);
2299 skb_dst_set(skb, &rth->dst);
2300 }
2301 RT_CACHE_STAT_INC(in_hit);
2302 rcu_read_unlock();
2303 return 0;
2304 }
2305 RT_CACHE_STAT_INC(in_hlist_search);
2306 }
2307
2308 skip_cache:
2309 /* Multicast recognition logic is moved from route cache to here.
2310 The problem was that too many Ethernet cards have broken/missing
2311 hardware multicast filters :-( As result the host on multicasting
2312 network acquires a lot of useless route cache entries, sort of
2313 SDR messages from all the world. Now we try to get rid of them.
2314 Really, provided software IP multicast filter is organized
2315 reasonably (at least, hashed), it does not result in a slowdown
2316 comparing with route cache reject entries.
2317 Note, that multicast routers are not affected, because
2318 route cache entry is created eventually.
2319 */
2320 if (ipv4_is_multicast(daddr)) {
2321 struct in_device *in_dev = __in_dev_get_rcu(dev);
2322
2323 if (in_dev) {
2324 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2325 ip_hdr(skb)->protocol);
2326 if (our
2327 #ifdef CONFIG_IP_MROUTE
2328 ||
2329 (!ipv4_is_local_multicast(daddr) &&
2330 IN_DEV_MFORWARD(in_dev))
2331 #endif
2332 ) {
2333 int res = ip_route_input_mc(skb, daddr, saddr,
2334 tos, dev, our);
2335 rcu_read_unlock();
2336 return res;
2337 }
2338 }
2339 rcu_read_unlock();
2340 return -EINVAL;
2341 }
2342 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2343 rcu_read_unlock();
2344 return res;
2345 }
2346 EXPORT_SYMBOL(ip_route_input_common);
2347
2348 /* called with rcu_read_lock() */
2349 static struct rtable *__mkroute_output(const struct fib_result *res,
2350 const struct flowi4 *fl4,
2351 const struct flowi4 *oldflp4,
2352 struct net_device *dev_out,
2353 unsigned int flags)
2354 {
2355 struct fib_info *fi = res->fi;
2356 u32 tos = RT_FL_TOS(oldflp4);
2357 struct in_device *in_dev;
2358 u16 type = res->type;
2359 struct rtable *rth;
2360
2361 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2362 return ERR_PTR(-EINVAL);
2363
2364 if (ipv4_is_lbcast(fl4->daddr))
2365 type = RTN_BROADCAST;
2366 else if (ipv4_is_multicast(fl4->daddr))
2367 type = RTN_MULTICAST;
2368 else if (ipv4_is_zeronet(fl4->daddr))
2369 return ERR_PTR(-EINVAL);
2370
2371 if (dev_out->flags & IFF_LOOPBACK)
2372 flags |= RTCF_LOCAL;
2373
2374 in_dev = __in_dev_get_rcu(dev_out);
2375 if (!in_dev)
2376 return ERR_PTR(-EINVAL);
2377
2378 if (type == RTN_BROADCAST) {
2379 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2380 fi = NULL;
2381 } else if (type == RTN_MULTICAST) {
2382 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2383 if (!ip_check_mc_rcu(in_dev, oldflp4->daddr, oldflp4->saddr,
2384 oldflp4->flowi4_proto))
2385 flags &= ~RTCF_LOCAL;
2386 /* If multicast route do not exist use
2387 * default one, but do not gateway in this case.
2388 * Yes, it is hack.
2389 */
2390 if (fi && res->prefixlen < 4)
2391 fi = NULL;
2392 }
2393
2394 rth = rt_dst_alloc(IN_DEV_CONF_GET(in_dev, NOPOLICY),
2395 IN_DEV_CONF_GET(in_dev, NOXFRM));
2396 if (!rth)
2397 return ERR_PTR(-ENOBUFS);
2398
2399 rth->rt_key_dst = oldflp4->daddr;
2400 rth->rt_tos = tos;
2401 rth->rt_key_src = oldflp4->saddr;
2402 rth->rt_oif = oldflp4->flowi4_oif;
2403 rth->rt_mark = oldflp4->flowi4_mark;
2404 rth->rt_dst = fl4->daddr;
2405 rth->rt_src = fl4->saddr;
2406 rth->rt_iif = 0;
2407 /* get references to the devices that are to be hold by the routing
2408 cache entry */
2409 rth->dst.dev = dev_out;
2410 dev_hold(dev_out);
2411 rth->rt_gateway = fl4->daddr;
2412 rth->rt_spec_dst= fl4->saddr;
2413
2414 rth->dst.output=ip_output;
2415 rth->rt_genid = rt_genid(dev_net(dev_out));
2416
2417 RT_CACHE_STAT_INC(out_slow_tot);
2418
2419 if (flags & RTCF_LOCAL) {
2420 rth->dst.input = ip_local_deliver;
2421 rth->rt_spec_dst = fl4->daddr;
2422 }
2423 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2424 rth->rt_spec_dst = fl4->saddr;
2425 if (flags & RTCF_LOCAL &&
2426 !(dev_out->flags & IFF_LOOPBACK)) {
2427 rth->dst.output = ip_mc_output;
2428 RT_CACHE_STAT_INC(out_slow_mc);
2429 }
2430 #ifdef CONFIG_IP_MROUTE
2431 if (type == RTN_MULTICAST) {
2432 if (IN_DEV_MFORWARD(in_dev) &&
2433 !ipv4_is_local_multicast(oldflp4->daddr)) {
2434 rth->dst.input = ip_mr_input;
2435 rth->dst.output = ip_mc_output;
2436 }
2437 }
2438 #endif
2439 }
2440
2441 rt_set_nexthop(rth, oldflp4, res, fi, type, 0);
2442
2443 rth->rt_flags = flags;
2444 return rth;
2445 }
2446
2447 /*
2448 * Major route resolver routine.
2449 * called with rcu_read_lock();
2450 */
2451
2452 static struct rtable *ip_route_output_slow(struct net *net,
2453 const struct flowi4 *oldflp4)
2454 {
2455 u32 tos = RT_FL_TOS(oldflp4);
2456 struct flowi4 fl4;
2457 struct fib_result res;
2458 unsigned int flags = 0;
2459 struct net_device *dev_out = NULL;
2460 struct rtable *rth;
2461
2462 res.fi = NULL;
2463 #ifdef CONFIG_IP_MULTIPLE_TABLES
2464 res.r = NULL;
2465 #endif
2466
2467 fl4.flowi4_oif = oldflp4->flowi4_oif;
2468 fl4.flowi4_iif = net->loopback_dev->ifindex;
2469 fl4.flowi4_mark = oldflp4->flowi4_mark;
2470 fl4.daddr = oldflp4->daddr;
2471 fl4.saddr = oldflp4->saddr;
2472 fl4.flowi4_tos = tos & IPTOS_RT_MASK;
2473 fl4.flowi4_scope = ((tos & RTO_ONLINK) ?
2474 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2475
2476 rcu_read_lock();
2477 if (oldflp4->saddr) {
2478 rth = ERR_PTR(-EINVAL);
2479 if (ipv4_is_multicast(oldflp4->saddr) ||
2480 ipv4_is_lbcast(oldflp4->saddr) ||
2481 ipv4_is_zeronet(oldflp4->saddr))
2482 goto out;
2483
2484 /* I removed check for oif == dev_out->oif here.
2485 It was wrong for two reasons:
2486 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2487 is assigned to multiple interfaces.
2488 2. Moreover, we are allowed to send packets with saddr
2489 of another iface. --ANK
2490 */
2491
2492 if (oldflp4->flowi4_oif == 0 &&
2493 (ipv4_is_multicast(oldflp4->daddr) ||
2494 ipv4_is_lbcast(oldflp4->daddr))) {
2495 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2496 dev_out = __ip_dev_find(net, oldflp4->saddr, false);
2497 if (dev_out == NULL)
2498 goto out;
2499
2500 /* Special hack: user can direct multicasts
2501 and limited broadcast via necessary interface
2502 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2503 This hack is not just for fun, it allows
2504 vic,vat and friends to work.
2505 They bind socket to loopback, set ttl to zero
2506 and expect that it will work.
2507 From the viewpoint of routing cache they are broken,
2508 because we are not allowed to build multicast path
2509 with loopback source addr (look, routing cache
2510 cannot know, that ttl is zero, so that packet
2511 will not leave this host and route is valid).
2512 Luckily, this hack is good workaround.
2513 */
2514
2515 fl4.flowi4_oif = dev_out->ifindex;
2516 goto make_route;
2517 }
2518
2519 if (!(oldflp4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2520 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2521 if (!__ip_dev_find(net, oldflp4->saddr, false))
2522 goto out;
2523 }
2524 }
2525
2526
2527 if (oldflp4->flowi4_oif) {
2528 dev_out = dev_get_by_index_rcu(net, oldflp4->flowi4_oif);
2529 rth = ERR_PTR(-ENODEV);
2530 if (dev_out == NULL)
2531 goto out;
2532
2533 /* RACE: Check return value of inet_select_addr instead. */
2534 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2535 rth = ERR_PTR(-ENETUNREACH);
2536 goto out;
2537 }
2538 if (ipv4_is_local_multicast(oldflp4->daddr) ||
2539 ipv4_is_lbcast(oldflp4->daddr)) {
2540 if (!fl4.saddr)
2541 fl4.saddr = inet_select_addr(dev_out, 0,
2542 RT_SCOPE_LINK);
2543 goto make_route;
2544 }
2545 if (!fl4.saddr) {
2546 if (ipv4_is_multicast(oldflp4->daddr))
2547 fl4.saddr = inet_select_addr(dev_out, 0,
2548 fl4.flowi4_scope);
2549 else if (!oldflp4->daddr)
2550 fl4.saddr = inet_select_addr(dev_out, 0,
2551 RT_SCOPE_HOST);
2552 }
2553 }
2554
2555 if (!fl4.daddr) {
2556 fl4.daddr = fl4.saddr;
2557 if (!fl4.daddr)
2558 fl4.daddr = fl4.saddr = htonl(INADDR_LOOPBACK);
2559 dev_out = net->loopback_dev;
2560 fl4.flowi4_oif = net->loopback_dev->ifindex;
2561 res.type = RTN_LOCAL;
2562 flags |= RTCF_LOCAL;
2563 goto make_route;
2564 }
2565
2566 if (fib_lookup(net, &fl4, &res)) {
2567 res.fi = NULL;
2568 if (oldflp4->flowi4_oif) {
2569 /* Apparently, routing tables are wrong. Assume,
2570 that the destination is on link.
2571
2572 WHY? DW.
2573 Because we are allowed to send to iface
2574 even if it has NO routes and NO assigned
2575 addresses. When oif is specified, routing
2576 tables are looked up with only one purpose:
2577 to catch if destination is gatewayed, rather than
2578 direct. Moreover, if MSG_DONTROUTE is set,
2579 we send packet, ignoring both routing tables
2580 and ifaddr state. --ANK
2581
2582
2583 We could make it even if oif is unknown,
2584 likely IPv6, but we do not.
2585 */
2586
2587 if (fl4.saddr == 0)
2588 fl4.saddr = inet_select_addr(dev_out, 0,
2589 RT_SCOPE_LINK);
2590 res.type = RTN_UNICAST;
2591 goto make_route;
2592 }
2593 rth = ERR_PTR(-ENETUNREACH);
2594 goto out;
2595 }
2596
2597 if (res.type == RTN_LOCAL) {
2598 if (!fl4.saddr) {
2599 if (res.fi->fib_prefsrc)
2600 fl4.saddr = res.fi->fib_prefsrc;
2601 else
2602 fl4.saddr = fl4.daddr;
2603 }
2604 dev_out = net->loopback_dev;
2605 fl4.flowi4_oif = dev_out->ifindex;
2606 res.fi = NULL;
2607 flags |= RTCF_LOCAL;
2608 goto make_route;
2609 }
2610
2611 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2612 if (res.fi->fib_nhs > 1 && fl4.flowi4_oif == 0)
2613 fib_select_multipath(&res);
2614 else
2615 #endif
2616 if (!res.prefixlen && res.type == RTN_UNICAST && !fl4.flowi4_oif)
2617 fib_select_default(&res);
2618
2619 if (!fl4.saddr)
2620 fl4.saddr = FIB_RES_PREFSRC(res);
2621
2622 dev_out = FIB_RES_DEV(res);
2623 fl4.flowi4_oif = dev_out->ifindex;
2624
2625
2626 make_route:
2627 rth = __mkroute_output(&res, &fl4, oldflp4, dev_out, flags);
2628 if (!IS_ERR(rth)) {
2629 unsigned int hash;
2630
2631 hash = rt_hash(oldflp4->daddr, oldflp4->saddr, oldflp4->flowi4_oif,
2632 rt_genid(dev_net(dev_out)));
2633 rth = rt_intern_hash(hash, rth, NULL, oldflp4->flowi4_oif);
2634 }
2635
2636 out:
2637 rcu_read_unlock();
2638 return rth;
2639 }
2640
2641 struct rtable *__ip_route_output_key(struct net *net, const struct flowi4 *flp4)
2642 {
2643 struct rtable *rth;
2644 unsigned int hash;
2645
2646 if (!rt_caching(net))
2647 goto slow_output;
2648
2649 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2650
2651 rcu_read_lock_bh();
2652 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2653 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2654 if (rth->rt_key_dst == flp4->daddr &&
2655 rth->rt_key_src == flp4->saddr &&
2656 rt_is_output_route(rth) &&
2657 rth->rt_oif == flp4->flowi4_oif &&
2658 rth->rt_mark == flp4->flowi4_mark &&
2659 !((rth->rt_tos ^ flp4->flowi4_tos) &
2660 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2661 net_eq(dev_net(rth->dst.dev), net) &&
2662 !rt_is_expired(rth)) {
2663 dst_use(&rth->dst, jiffies);
2664 RT_CACHE_STAT_INC(out_hit);
2665 rcu_read_unlock_bh();
2666 return rth;
2667 }
2668 RT_CACHE_STAT_INC(out_hlist_search);
2669 }
2670 rcu_read_unlock_bh();
2671
2672 slow_output:
2673 return ip_route_output_slow(net, flp4);
2674 }
2675 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2676
2677 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2678 {
2679 return NULL;
2680 }
2681
2682 static unsigned int ipv4_blackhole_default_mtu(const struct dst_entry *dst)
2683 {
2684 return 0;
2685 }
2686
2687 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2688 {
2689 }
2690
2691 static struct dst_ops ipv4_dst_blackhole_ops = {
2692 .family = AF_INET,
2693 .protocol = cpu_to_be16(ETH_P_IP),
2694 .destroy = ipv4_dst_destroy,
2695 .check = ipv4_blackhole_dst_check,
2696 .default_mtu = ipv4_blackhole_default_mtu,
2697 .default_advmss = ipv4_default_advmss,
2698 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2699 };
2700
2701 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2702 {
2703 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, 1);
2704 struct rtable *ort = (struct rtable *) dst_orig;
2705
2706 if (rt) {
2707 struct dst_entry *new = &rt->dst;
2708
2709 new->__use = 1;
2710 new->input = dst_discard;
2711 new->output = dst_discard;
2712 dst_copy_metrics(new, &ort->dst);
2713
2714 new->dev = ort->dst.dev;
2715 if (new->dev)
2716 dev_hold(new->dev);
2717
2718 rt->rt_key_dst = ort->rt_key_dst;
2719 rt->rt_key_src = ort->rt_key_src;
2720 rt->rt_tos = ort->rt_tos;
2721 rt->rt_iif = ort->rt_iif;
2722 rt->rt_oif = ort->rt_oif;
2723 rt->rt_mark = ort->rt_mark;
2724
2725 rt->rt_genid = rt_genid(net);
2726 rt->rt_flags = ort->rt_flags;
2727 rt->rt_type = ort->rt_type;
2728 rt->rt_dst = ort->rt_dst;
2729 rt->rt_src = ort->rt_src;
2730 rt->rt_iif = ort->rt_iif;
2731 rt->rt_gateway = ort->rt_gateway;
2732 rt->rt_spec_dst = ort->rt_spec_dst;
2733 rt->peer = ort->peer;
2734 if (rt->peer)
2735 atomic_inc(&rt->peer->refcnt);
2736 rt->fi = ort->fi;
2737 if (rt->fi)
2738 atomic_inc(&rt->fi->fib_clntref);
2739
2740 dst_free(new);
2741 }
2742
2743 dst_release(dst_orig);
2744
2745 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2746 }
2747
2748 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2749 struct sock *sk)
2750 {
2751 struct rtable *rt = __ip_route_output_key(net, flp4);
2752
2753 if (IS_ERR(rt))
2754 return rt;
2755
2756 if (flp4->flowi4_proto) {
2757 if (!flp4->saddr)
2758 flp4->saddr = rt->rt_src;
2759 if (!flp4->daddr)
2760 flp4->daddr = rt->rt_dst;
2761 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2762 flowi4_to_flowi(flp4),
2763 sk, 0);
2764 }
2765
2766 return rt;
2767 }
2768 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2769
2770 static int rt_fill_info(struct net *net,
2771 struct sk_buff *skb, u32 pid, u32 seq, int event,
2772 int nowait, unsigned int flags)
2773 {
2774 struct rtable *rt = skb_rtable(skb);
2775 struct rtmsg *r;
2776 struct nlmsghdr *nlh;
2777 long expires;
2778 u32 id = 0, ts = 0, tsage = 0, error;
2779
2780 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2781 if (nlh == NULL)
2782 return -EMSGSIZE;
2783
2784 r = nlmsg_data(nlh);
2785 r->rtm_family = AF_INET;
2786 r->rtm_dst_len = 32;
2787 r->rtm_src_len = 0;
2788 r->rtm_tos = rt->rt_tos;
2789 r->rtm_table = RT_TABLE_MAIN;
2790 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2791 r->rtm_type = rt->rt_type;
2792 r->rtm_scope = RT_SCOPE_UNIVERSE;
2793 r->rtm_protocol = RTPROT_UNSPEC;
2794 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2795 if (rt->rt_flags & RTCF_NOTIFY)
2796 r->rtm_flags |= RTM_F_NOTIFY;
2797
2798 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2799
2800 if (rt->rt_key_src) {
2801 r->rtm_src_len = 32;
2802 NLA_PUT_BE32(skb, RTA_SRC, rt->rt_key_src);
2803 }
2804 if (rt->dst.dev)
2805 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2806 #ifdef CONFIG_IP_ROUTE_CLASSID
2807 if (rt->dst.tclassid)
2808 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2809 #endif
2810 if (rt_is_input_route(rt))
2811 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2812 else if (rt->rt_src != rt->rt_key_src)
2813 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2814
2815 if (rt->rt_dst != rt->rt_gateway)
2816 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2817
2818 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2819 goto nla_put_failure;
2820
2821 if (rt->rt_mark)
2822 NLA_PUT_BE32(skb, RTA_MARK, rt->rt_mark);
2823
2824 error = rt->dst.error;
2825 expires = (rt->peer && rt->peer->pmtu_expires) ?
2826 rt->peer->pmtu_expires - jiffies : 0;
2827 if (rt->peer) {
2828 inet_peer_refcheck(rt->peer);
2829 id = atomic_read(&rt->peer->ip_id_count) & 0xffff;
2830 if (rt->peer->tcp_ts_stamp) {
2831 ts = rt->peer->tcp_ts;
2832 tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2833 }
2834 }
2835
2836 if (rt_is_input_route(rt)) {
2837 #ifdef CONFIG_IP_MROUTE
2838 __be32 dst = rt->rt_dst;
2839
2840 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2841 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2842 int err = ipmr_get_route(net, skb, r, nowait);
2843 if (err <= 0) {
2844 if (!nowait) {
2845 if (err == 0)
2846 return 0;
2847 goto nla_put_failure;
2848 } else {
2849 if (err == -EMSGSIZE)
2850 goto nla_put_failure;
2851 error = err;
2852 }
2853 }
2854 } else
2855 #endif
2856 NLA_PUT_U32(skb, RTA_IIF, rt->rt_iif);
2857 }
2858
2859 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2860 expires, error) < 0)
2861 goto nla_put_failure;
2862
2863 return nlmsg_end(skb, nlh);
2864
2865 nla_put_failure:
2866 nlmsg_cancel(skb, nlh);
2867 return -EMSGSIZE;
2868 }
2869
2870 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2871 {
2872 struct net *net = sock_net(in_skb->sk);
2873 struct rtmsg *rtm;
2874 struct nlattr *tb[RTA_MAX+1];
2875 struct rtable *rt = NULL;
2876 __be32 dst = 0;
2877 __be32 src = 0;
2878 u32 iif;
2879 int err;
2880 int mark;
2881 struct sk_buff *skb;
2882
2883 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2884 if (err < 0)
2885 goto errout;
2886
2887 rtm = nlmsg_data(nlh);
2888
2889 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2890 if (skb == NULL) {
2891 err = -ENOBUFS;
2892 goto errout;
2893 }
2894
2895 /* Reserve room for dummy headers, this skb can pass
2896 through good chunk of routing engine.
2897 */
2898 skb_reset_mac_header(skb);
2899 skb_reset_network_header(skb);
2900
2901 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2902 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2903 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2904
2905 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2906 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2907 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2908 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2909
2910 if (iif) {
2911 struct net_device *dev;
2912
2913 dev = __dev_get_by_index(net, iif);
2914 if (dev == NULL) {
2915 err = -ENODEV;
2916 goto errout_free;
2917 }
2918
2919 skb->protocol = htons(ETH_P_IP);
2920 skb->dev = dev;
2921 skb->mark = mark;
2922 local_bh_disable();
2923 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2924 local_bh_enable();
2925
2926 rt = skb_rtable(skb);
2927 if (err == 0 && rt->dst.error)
2928 err = -rt->dst.error;
2929 } else {
2930 struct flowi4 fl4 = {
2931 .daddr = dst,
2932 .saddr = src,
2933 .flowi4_tos = rtm->rtm_tos,
2934 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2935 .flowi4_mark = mark,
2936 };
2937 rt = ip_route_output_key(net, &fl4);
2938
2939 err = 0;
2940 if (IS_ERR(rt))
2941 err = PTR_ERR(rt);
2942 }
2943
2944 if (err)
2945 goto errout_free;
2946
2947 skb_dst_set(skb, &rt->dst);
2948 if (rtm->rtm_flags & RTM_F_NOTIFY)
2949 rt->rt_flags |= RTCF_NOTIFY;
2950
2951 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2952 RTM_NEWROUTE, 0, 0);
2953 if (err <= 0)
2954 goto errout_free;
2955
2956 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2957 errout:
2958 return err;
2959
2960 errout_free:
2961 kfree_skb(skb);
2962 goto errout;
2963 }
2964
2965 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2966 {
2967 struct rtable *rt;
2968 int h, s_h;
2969 int idx, s_idx;
2970 struct net *net;
2971
2972 net = sock_net(skb->sk);
2973
2974 s_h = cb->args[0];
2975 if (s_h < 0)
2976 s_h = 0;
2977 s_idx = idx = cb->args[1];
2978 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
2979 if (!rt_hash_table[h].chain)
2980 continue;
2981 rcu_read_lock_bh();
2982 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
2983 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
2984 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
2985 continue;
2986 if (rt_is_expired(rt))
2987 continue;
2988 skb_dst_set_noref(skb, &rt->dst);
2989 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
2990 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2991 1, NLM_F_MULTI) <= 0) {
2992 skb_dst_drop(skb);
2993 rcu_read_unlock_bh();
2994 goto done;
2995 }
2996 skb_dst_drop(skb);
2997 }
2998 rcu_read_unlock_bh();
2999 }
3000
3001 done:
3002 cb->args[0] = h;
3003 cb->args[1] = idx;
3004 return skb->len;
3005 }
3006
3007 void ip_rt_multicast_event(struct in_device *in_dev)
3008 {
3009 rt_cache_flush(dev_net(in_dev->dev), 0);
3010 }
3011
3012 #ifdef CONFIG_SYSCTL
3013 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3014 void __user *buffer,
3015 size_t *lenp, loff_t *ppos)
3016 {
3017 if (write) {
3018 int flush_delay;
3019 ctl_table ctl;
3020 struct net *net;
3021
3022 memcpy(&ctl, __ctl, sizeof(ctl));
3023 ctl.data = &flush_delay;
3024 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3025
3026 net = (struct net *)__ctl->extra1;
3027 rt_cache_flush(net, flush_delay);
3028 return 0;
3029 }
3030
3031 return -EINVAL;
3032 }
3033
3034 static ctl_table ipv4_route_table[] = {
3035 {
3036 .procname = "gc_thresh",
3037 .data = &ipv4_dst_ops.gc_thresh,
3038 .maxlen = sizeof(int),
3039 .mode = 0644,
3040 .proc_handler = proc_dointvec,
3041 },
3042 {
3043 .procname = "max_size",
3044 .data = &ip_rt_max_size,
3045 .maxlen = sizeof(int),
3046 .mode = 0644,
3047 .proc_handler = proc_dointvec,
3048 },
3049 {
3050 /* Deprecated. Use gc_min_interval_ms */
3051
3052 .procname = "gc_min_interval",
3053 .data = &ip_rt_gc_min_interval,
3054 .maxlen = sizeof(int),
3055 .mode = 0644,
3056 .proc_handler = proc_dointvec_jiffies,
3057 },
3058 {
3059 .procname = "gc_min_interval_ms",
3060 .data = &ip_rt_gc_min_interval,
3061 .maxlen = sizeof(int),
3062 .mode = 0644,
3063 .proc_handler = proc_dointvec_ms_jiffies,
3064 },
3065 {
3066 .procname = "gc_timeout",
3067 .data = &ip_rt_gc_timeout,
3068 .maxlen = sizeof(int),
3069 .mode = 0644,
3070 .proc_handler = proc_dointvec_jiffies,
3071 },
3072 {
3073 .procname = "gc_interval",
3074 .data = &ip_rt_gc_interval,
3075 .maxlen = sizeof(int),
3076 .mode = 0644,
3077 .proc_handler = proc_dointvec_jiffies,
3078 },
3079 {
3080 .procname = "redirect_load",
3081 .data = &ip_rt_redirect_load,
3082 .maxlen = sizeof(int),
3083 .mode = 0644,
3084 .proc_handler = proc_dointvec,
3085 },
3086 {
3087 .procname = "redirect_number",
3088 .data = &ip_rt_redirect_number,
3089 .maxlen = sizeof(int),
3090 .mode = 0644,
3091 .proc_handler = proc_dointvec,
3092 },
3093 {
3094 .procname = "redirect_silence",
3095 .data = &ip_rt_redirect_silence,
3096 .maxlen = sizeof(int),
3097 .mode = 0644,
3098 .proc_handler = proc_dointvec,
3099 },
3100 {
3101 .procname = "error_cost",
3102 .data = &ip_rt_error_cost,
3103 .maxlen = sizeof(int),
3104 .mode = 0644,
3105 .proc_handler = proc_dointvec,
3106 },
3107 {
3108 .procname = "error_burst",
3109 .data = &ip_rt_error_burst,
3110 .maxlen = sizeof(int),
3111 .mode = 0644,
3112 .proc_handler = proc_dointvec,
3113 },
3114 {
3115 .procname = "gc_elasticity",
3116 .data = &ip_rt_gc_elasticity,
3117 .maxlen = sizeof(int),
3118 .mode = 0644,
3119 .proc_handler = proc_dointvec,
3120 },
3121 {
3122 .procname = "mtu_expires",
3123 .data = &ip_rt_mtu_expires,
3124 .maxlen = sizeof(int),
3125 .mode = 0644,
3126 .proc_handler = proc_dointvec_jiffies,
3127 },
3128 {
3129 .procname = "min_pmtu",
3130 .data = &ip_rt_min_pmtu,
3131 .maxlen = sizeof(int),
3132 .mode = 0644,
3133 .proc_handler = proc_dointvec,
3134 },
3135 {
3136 .procname = "min_adv_mss",
3137 .data = &ip_rt_min_advmss,
3138 .maxlen = sizeof(int),
3139 .mode = 0644,
3140 .proc_handler = proc_dointvec,
3141 },
3142 { }
3143 };
3144
3145 static struct ctl_table empty[1];
3146
3147 static struct ctl_table ipv4_skeleton[] =
3148 {
3149 { .procname = "route",
3150 .mode = 0555, .child = ipv4_route_table},
3151 { .procname = "neigh",
3152 .mode = 0555, .child = empty},
3153 { }
3154 };
3155
3156 static __net_initdata struct ctl_path ipv4_path[] = {
3157 { .procname = "net", },
3158 { .procname = "ipv4", },
3159 { },
3160 };
3161
3162 static struct ctl_table ipv4_route_flush_table[] = {
3163 {
3164 .procname = "flush",
3165 .maxlen = sizeof(int),
3166 .mode = 0200,
3167 .proc_handler = ipv4_sysctl_rtcache_flush,
3168 },
3169 { },
3170 };
3171
3172 static __net_initdata struct ctl_path ipv4_route_path[] = {
3173 { .procname = "net", },
3174 { .procname = "ipv4", },
3175 { .procname = "route", },
3176 { },
3177 };
3178
3179 static __net_init int sysctl_route_net_init(struct net *net)
3180 {
3181 struct ctl_table *tbl;
3182
3183 tbl = ipv4_route_flush_table;
3184 if (!net_eq(net, &init_net)) {
3185 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3186 if (tbl == NULL)
3187 goto err_dup;
3188 }
3189 tbl[0].extra1 = net;
3190
3191 net->ipv4.route_hdr =
3192 register_net_sysctl_table(net, ipv4_route_path, tbl);
3193 if (net->ipv4.route_hdr == NULL)
3194 goto err_reg;
3195 return 0;
3196
3197 err_reg:
3198 if (tbl != ipv4_route_flush_table)
3199 kfree(tbl);
3200 err_dup:
3201 return -ENOMEM;
3202 }
3203
3204 static __net_exit void sysctl_route_net_exit(struct net *net)
3205 {
3206 struct ctl_table *tbl;
3207
3208 tbl = net->ipv4.route_hdr->ctl_table_arg;
3209 unregister_net_sysctl_table(net->ipv4.route_hdr);
3210 BUG_ON(tbl == ipv4_route_flush_table);
3211 kfree(tbl);
3212 }
3213
3214 static __net_initdata struct pernet_operations sysctl_route_ops = {
3215 .init = sysctl_route_net_init,
3216 .exit = sysctl_route_net_exit,
3217 };
3218 #endif
3219
3220 static __net_init int rt_genid_init(struct net *net)
3221 {
3222 get_random_bytes(&net->ipv4.rt_genid,
3223 sizeof(net->ipv4.rt_genid));
3224 return 0;
3225 }
3226
3227 static __net_initdata struct pernet_operations rt_genid_ops = {
3228 .init = rt_genid_init,
3229 };
3230
3231
3232 #ifdef CONFIG_IP_ROUTE_CLASSID
3233 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3234 #endif /* CONFIG_IP_ROUTE_CLASSID */
3235
3236 static __initdata unsigned long rhash_entries;
3237 static int __init set_rhash_entries(char *str)
3238 {
3239 if (!str)
3240 return 0;
3241 rhash_entries = simple_strtoul(str, &str, 0);
3242 return 1;
3243 }
3244 __setup("rhash_entries=", set_rhash_entries);
3245
3246 int __init ip_rt_init(void)
3247 {
3248 int rc = 0;
3249
3250 #ifdef CONFIG_IP_ROUTE_CLASSID
3251 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3252 if (!ip_rt_acct)
3253 panic("IP: failed to allocate ip_rt_acct\n");
3254 #endif
3255
3256 ipv4_dst_ops.kmem_cachep =
3257 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3258 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3259
3260 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3261
3262 if (dst_entries_init(&ipv4_dst_ops) < 0)
3263 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3264
3265 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3266 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3267
3268 rt_hash_table = (struct rt_hash_bucket *)
3269 alloc_large_system_hash("IP route cache",
3270 sizeof(struct rt_hash_bucket),
3271 rhash_entries,
3272 (totalram_pages >= 128 * 1024) ?
3273 15 : 17,
3274 0,
3275 &rt_hash_log,
3276 &rt_hash_mask,
3277 rhash_entries ? 0 : 512 * 1024);
3278 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3279 rt_hash_lock_init();
3280
3281 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3282 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3283
3284 devinet_init();
3285 ip_fib_init();
3286
3287 if (ip_rt_proc_init())
3288 printk(KERN_ERR "Unable to create route proc files\n");
3289 #ifdef CONFIG_XFRM
3290 xfrm_init();
3291 xfrm4_init(ip_rt_max_size);
3292 #endif
3293 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3294
3295 #ifdef CONFIG_SYSCTL
3296 register_pernet_subsys(&sysctl_route_ops);
3297 #endif
3298 register_pernet_subsys(&rt_genid_ops);
3299 return rc;
3300 }
3301
3302 #ifdef CONFIG_SYSCTL
3303 /*
3304 * We really need to sanitize the damn ipv4 init order, then all
3305 * this nonsense will go away.
3306 */
3307 void __init ip_static_sysctl_init(void)
3308 {
3309 register_sysctl_paths(ipv4_path, ipv4_skeleton);
3310 }
3311 #endif
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree