?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86: cpa use the new set_clr function
[linux-2.6/linux-loongson.git] / net / ipv4 / route.c
blob896c768e41a2d90baef4b4a3783be0164678ff9e
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 * Version: $Id: route.c,v 1.103 2002/01/12 07:44:09 davem Exp $
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
14 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
15 *
16 * Fixes:
17 * Alan Cox : Verify area fixes.
18 * Alan Cox : cli() protects routing changes
19 * Rui Oliveira : ICMP routing table updates
20 * (rco@di.uminho.pt) Routing table insertion and update
21 * Linus Torvalds : Rewrote bits to be sensible
22 * Alan Cox : Added BSD route gw semantics
23 * Alan Cox : Super /proc >4K
24 * Alan Cox : MTU in route table
25 * Alan Cox : MSS actually. Also added the window
26 * clamper.
27 * Sam Lantinga : Fixed route matching in rt_del()
28 * Alan Cox : Routing cache support.
29 * Alan Cox : Removed compatibility cruft.
30 * Alan Cox : RTF_REJECT support.
31 * Alan Cox : TCP irtt support.
32 * Jonathan Naylor : Added Metric support.
33 * Miquel van Smoorenburg : BSD API fixes.
34 * Miquel van Smoorenburg : Metrics.
35 * Alan Cox : Use __u32 properly
36 * Alan Cox : Aligned routing errors more closely with BSD
37 * our system is still very different.
38 * Alan Cox : Faster /proc handling
39 * Alexey Kuznetsov : Massive rework to support tree based routing,
40 * routing caches and better behaviour.
41 *
42 * Olaf Erb : irtt wasn't being copied right.
43 * Bjorn Ekwall : Kerneld route support.
44 * Alan Cox : Multicast fixed (I hope)
45 * Pavel Krauz : Limited broadcast fixed
46 * Mike McLagan : Routing by source
47 * Alexey Kuznetsov : End of old history. Split to fib.c and
48 * route.c and rewritten from scratch.
49 * Andi Kleen : Load-limit warning messages.
50 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
51 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
52 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
53 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
54 * Marc Boucher : routing by fwmark
55 * Robert Olsson : Added rt_cache statistics
56 * Arnaldo C. Melo : Convert proc stuff to seq_file
57 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
58 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
59 * Ilia Sotnikov : Removed TOS from hash calculations
60 *
61 * This program is free software; you can redistribute it and/or
62 * modify it under the terms of the GNU General Public License
63 * as published by the Free Software Foundation; either version
64 * 2 of the License, or (at your option) any later version.
65 */
66
67 #include <linux/module.h>
68 #include <asm/uaccess.h>
69 #include <asm/system.h>
70 #include <linux/bitops.h>
71 #include <linux/types.h>
72 #include <linux/kernel.h>
73 #include <linux/mm.h>
74 #include <linux/bootmem.h>
75 #include <linux/string.h>
76 #include <linux/socket.h>
77 #include <linux/sockios.h>
78 #include <linux/errno.h>
79 #include <linux/in.h>
80 #include <linux/inet.h>
81 #include <linux/netdevice.h>
82 #include <linux/proc_fs.h>
83 #include <linux/init.h>
84 #include <linux/workqueue.h>
85 #include <linux/skbuff.h>
86 #include <linux/inetdevice.h>
87 #include <linux/igmp.h>
88 #include <linux/pkt_sched.h>
89 #include <linux/mroute.h>
90 #include <linux/netfilter_ipv4.h>
91 #include <linux/random.h>
92 #include <linux/jhash.h>
93 #include <linux/rcupdate.h>
94 #include <linux/times.h>
95 #include <net/dst.h>
96 #include <net/net_namespace.h>
97 #include <net/protocol.h>
98 #include <net/ip.h>
99 #include <net/route.h>
100 #include <net/inetpeer.h>
101 #include <net/sock.h>
102 #include <net/ip_fib.h>
103 #include <net/arp.h>
104 #include <net/tcp.h>
105 #include <net/icmp.h>
106 #include <net/xfrm.h>
107 #include <net/netevent.h>
108 #include <net/rtnetlink.h>
109 #ifdef CONFIG_SYSCTL
110 #include <linux/sysctl.h>
111 #endif
112
113 #define RT_FL_TOS(oldflp) \
114 ((u32)(oldflp->fl4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
115
116 #define IP_MAX_MTU 0xFFF0
117
118 #define RT_GC_TIMEOUT (300*HZ)
119
120 static int ip_rt_min_delay = 2 * HZ;
121 static int ip_rt_max_delay = 10 * HZ;
122 static int ip_rt_max_size;
123 static int ip_rt_gc_timeout = RT_GC_TIMEOUT;
124 static int ip_rt_gc_interval = 60 * HZ;
125 static int ip_rt_gc_min_interval = HZ / 2;
126 static int ip_rt_redirect_number = 9;
127 static int ip_rt_redirect_load = HZ / 50;
128 static int ip_rt_redirect_silence = ((HZ / 50) << (9 + 1));
129 static int ip_rt_error_cost = HZ;
130 static int ip_rt_error_burst = 5 * HZ;
131 static int ip_rt_gc_elasticity = 8;
132 static int ip_rt_mtu_expires = 10 * 60 * HZ;
133 static int ip_rt_min_pmtu = 512 + 20 + 20;
134 static int ip_rt_min_advmss = 256;
135 static int ip_rt_secret_interval = 10 * 60 * HZ;
136 static int ip_rt_flush_expected;
137 static unsigned long rt_deadline;
138
139 #define RTprint(a...) printk(KERN_DEBUG a)
140
141 static struct timer_list rt_flush_timer;
142 static void rt_worker_func(struct work_struct *work);
143 static DECLARE_DELAYED_WORK(expires_work, rt_worker_func);
144 static struct timer_list rt_secret_timer;
145
146 /*
147 * Interface to generic destination cache.
148 */
149
150 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
151 static void ipv4_dst_destroy(struct dst_entry *dst);
152 static void ipv4_dst_ifdown(struct dst_entry *dst,
153 struct net_device *dev, int how);
154 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
155 static void ipv4_link_failure(struct sk_buff *skb);
156 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
157 static int rt_garbage_collect(struct dst_ops *ops);
158
159
160 static struct dst_ops ipv4_dst_ops = {
161 .family = AF_INET,
162 .protocol = __constant_htons(ETH_P_IP),
163 .gc = rt_garbage_collect,
164 .check = ipv4_dst_check,
165 .destroy = ipv4_dst_destroy,
166 .ifdown = ipv4_dst_ifdown,
167 .negative_advice = ipv4_negative_advice,
168 .link_failure = ipv4_link_failure,
169 .update_pmtu = ip_rt_update_pmtu,
170 .local_out = ip_local_out,
171 .entry_size = sizeof(struct rtable),
172 };
173
174 #define ECN_OR_COST(class) TC_PRIO_##class
175
176 const __u8 ip_tos2prio[16] = {
177 TC_PRIO_BESTEFFORT,
178 ECN_OR_COST(FILLER),
179 TC_PRIO_BESTEFFORT,
180 ECN_OR_COST(BESTEFFORT),
181 TC_PRIO_BULK,
182 ECN_OR_COST(BULK),
183 TC_PRIO_BULK,
184 ECN_OR_COST(BULK),
185 TC_PRIO_INTERACTIVE,
186 ECN_OR_COST(INTERACTIVE),
187 TC_PRIO_INTERACTIVE,
188 ECN_OR_COST(INTERACTIVE),
189 TC_PRIO_INTERACTIVE_BULK,
190 ECN_OR_COST(INTERACTIVE_BULK),
191 TC_PRIO_INTERACTIVE_BULK,
192 ECN_OR_COST(INTERACTIVE_BULK)
193 };
194
195
196 /*
197 * Route cache.
198 */
199
200 /* The locking scheme is rather straight forward:
201 *
202 * 1) Read-Copy Update protects the buckets of the central route hash.
203 * 2) Only writers remove entries, and they hold the lock
204 * as they look at rtable reference counts.
205 * 3) Only readers acquire references to rtable entries,
206 * they do so with atomic increments and with the
207 * lock held.
208 */
209
210 struct rt_hash_bucket {
211 struct rtable *chain;
212 };
213 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
214 defined(CONFIG_PROVE_LOCKING)
215 /*
216 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
217 * The size of this table is a power of two and depends on the number of CPUS.
218 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
219 */
220 #ifdef CONFIG_LOCKDEP
221 # define RT_HASH_LOCK_SZ 256
222 #else
223 # if NR_CPUS >= 32
224 # define RT_HASH_LOCK_SZ 4096
225 # elif NR_CPUS >= 16
226 # define RT_HASH_LOCK_SZ 2048
227 # elif NR_CPUS >= 8
228 # define RT_HASH_LOCK_SZ 1024
229 # elif NR_CPUS >= 4
230 # define RT_HASH_LOCK_SZ 512
231 # else
232 # define RT_HASH_LOCK_SZ 256
233 # endif
234 #endif
235
236 static spinlock_t *rt_hash_locks;
237 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
238
239 static __init void rt_hash_lock_init(void)
240 {
241 int i;
242
243 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
244 GFP_KERNEL);
245 if (!rt_hash_locks)
246 panic("IP: failed to allocate rt_hash_locks\n");
247
248 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
249 spin_lock_init(&rt_hash_locks[i]);
250 }
251 #else
252 # define rt_hash_lock_addr(slot) NULL
253
254 static inline void rt_hash_lock_init(void)
255 {
256 }
257 #endif
258
259 static struct rt_hash_bucket *rt_hash_table;
260 static unsigned rt_hash_mask;
261 static unsigned int rt_hash_log;
262 static unsigned int rt_hash_rnd;
263
264 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
265 #define RT_CACHE_STAT_INC(field) \
266 (__raw_get_cpu_var(rt_cache_stat).field++)
267
268 static int rt_intern_hash(unsigned hash, struct rtable *rth,
269 struct rtable **res);
270
271 static unsigned int rt_hash_code(u32 daddr, u32 saddr)
272 {
273 return (jhash_2words(daddr, saddr, rt_hash_rnd)
274 & rt_hash_mask);
275 }
276
277 #define rt_hash(daddr, saddr, idx) \
278 rt_hash_code((__force u32)(__be32)(daddr),\
279 (__force u32)(__be32)(saddr) ^ ((idx) << 5))
280
281 #ifdef CONFIG_PROC_FS
282 struct rt_cache_iter_state {
283 int bucket;
284 };
285
286 static struct rtable *rt_cache_get_first(struct seq_file *seq)
287 {
288 struct rtable *r = NULL;
289 struct rt_cache_iter_state *st = seq->private;
290
291 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
292 rcu_read_lock_bh();
293 r = rt_hash_table[st->bucket].chain;
294 if (r)
295 break;
296 rcu_read_unlock_bh();
297 }
298 return rcu_dereference(r);
299 }
300
301 static struct rtable *rt_cache_get_next(struct seq_file *seq, struct rtable *r)
302 {
303 struct rt_cache_iter_state *st = seq->private;
304
305 r = r->u.dst.rt_next;
306 while (!r) {
307 rcu_read_unlock_bh();
308 if (--st->bucket < 0)
309 break;
310 rcu_read_lock_bh();
311 r = rt_hash_table[st->bucket].chain;
312 }
313 return rcu_dereference(r);
314 }
315
316 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
317 {
318 struct rtable *r = rt_cache_get_first(seq);
319
320 if (r)
321 while (pos && (r = rt_cache_get_next(seq, r)))
322 --pos;
323 return pos ? NULL : r;
324 }
325
326 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
327 {
328 return *pos ? rt_cache_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
329 }
330
331 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
332 {
333 struct rtable *r = NULL;
334
335 if (v == SEQ_START_TOKEN)
336 r = rt_cache_get_first(seq);
337 else
338 r = rt_cache_get_next(seq, v);
339 ++*pos;
340 return r;
341 }
342
343 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
344 {
345 if (v && v != SEQ_START_TOKEN)
346 rcu_read_unlock_bh();
347 }
348
349 static int rt_cache_seq_show(struct seq_file *seq, void *v)
350 {
351 if (v == SEQ_START_TOKEN)
352 seq_printf(seq, "%-127s\n",
353 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
354 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
355 "HHUptod\tSpecDst");
356 else {
357 struct rtable *r = v;
358 char temp[256];
359
360 sprintf(temp, "%s\t%08lX\t%08lX\t%8X\t%d\t%u\t%d\t"
361 "%08lX\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X",
362 r->u.dst.dev ? r->u.dst.dev->name : "*",
363 (unsigned long)r->rt_dst, (unsigned long)r->rt_gateway,
364 r->rt_flags, atomic_read(&r->u.dst.__refcnt),
365 r->u.dst.__use, 0, (unsigned long)r->rt_src,
366 (dst_metric(&r->u.dst, RTAX_ADVMSS) ?
367 (int)dst_metric(&r->u.dst, RTAX_ADVMSS) + 40 : 0),
368 dst_metric(&r->u.dst, RTAX_WINDOW),
369 (int)((dst_metric(&r->u.dst, RTAX_RTT) >> 3) +
370 dst_metric(&r->u.dst, RTAX_RTTVAR)),
371 r->fl.fl4_tos,
372 r->u.dst.hh ? atomic_read(&r->u.dst.hh->hh_refcnt) : -1,
373 r->u.dst.hh ? (r->u.dst.hh->hh_output ==
374 dev_queue_xmit) : 0,
375 r->rt_spec_dst);
376 seq_printf(seq, "%-127s\n", temp);
377 }
378 return 0;
379 }
380
381 static const struct seq_operations rt_cache_seq_ops = {
382 .start = rt_cache_seq_start,
383 .next = rt_cache_seq_next,
384 .stop = rt_cache_seq_stop,
385 .show = rt_cache_seq_show,
386 };
387
388 static int rt_cache_seq_open(struct inode *inode, struct file *file)
389 {
390 return seq_open_private(file, &rt_cache_seq_ops,
391 sizeof(struct rt_cache_iter_state));
392 }
393
394 static const struct file_operations rt_cache_seq_fops = {
395 .owner = THIS_MODULE,
396 .open = rt_cache_seq_open,
397 .read = seq_read,
398 .llseek = seq_lseek,
399 .release = seq_release_private,
400 };
401
402
403 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
404 {
405 int cpu;
406
407 if (*pos == 0)
408 return SEQ_START_TOKEN;
409
410 for (cpu = *pos-1; cpu < NR_CPUS; ++cpu) {
411 if (!cpu_possible(cpu))
412 continue;
413 *pos = cpu+1;
414 return &per_cpu(rt_cache_stat, cpu);
415 }
416 return NULL;
417 }
418
419 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
420 {
421 int cpu;
422
423 for (cpu = *pos; cpu < NR_CPUS; ++cpu) {
424 if (!cpu_possible(cpu))
425 continue;
426 *pos = cpu+1;
427 return &per_cpu(rt_cache_stat, cpu);
428 }
429 return NULL;
430
431 }
432
433 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
434 {
435
436 }
437
438 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
439 {
440 struct rt_cache_stat *st = v;
441
442 if (v == SEQ_START_TOKEN) {
443 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");
444 return 0;
445 }
446
447 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
448 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
449 atomic_read(&ipv4_dst_ops.entries),
450 st->in_hit,
451 st->in_slow_tot,
452 st->in_slow_mc,
453 st->in_no_route,
454 st->in_brd,
455 st->in_martian_dst,
456 st->in_martian_src,
457
458 st->out_hit,
459 st->out_slow_tot,
460 st->out_slow_mc,
461
462 st->gc_total,
463 st->gc_ignored,
464 st->gc_goal_miss,
465 st->gc_dst_overflow,
466 st->in_hlist_search,
467 st->out_hlist_search
468 );
469 return 0;
470 }
471
472 static const struct seq_operations rt_cpu_seq_ops = {
473 .start = rt_cpu_seq_start,
474 .next = rt_cpu_seq_next,
475 .stop = rt_cpu_seq_stop,
476 .show = rt_cpu_seq_show,
477 };
478
479
480 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
481 {
482 return seq_open(file, &rt_cpu_seq_ops);
483 }
484
485 static const struct file_operations rt_cpu_seq_fops = {
486 .owner = THIS_MODULE,
487 .open = rt_cpu_seq_open,
488 .read = seq_read,
489 .llseek = seq_lseek,
490 .release = seq_release,
491 };
492
493 #ifdef CONFIG_NET_CLS_ROUTE
494 static int ip_rt_acct_read(char *buffer, char **start, off_t offset,
495 int length, int *eof, void *data)
496 {
497 unsigned int i;
498
499 if ((offset & 3) || (length & 3))
500 return -EIO;
501
502 if (offset >= sizeof(struct ip_rt_acct) * 256) {
503 *eof = 1;
504 return 0;
505 }
506
507 if (offset + length >= sizeof(struct ip_rt_acct) * 256) {
508 length = sizeof(struct ip_rt_acct) * 256 - offset;
509 *eof = 1;
510 }
511
512 offset /= sizeof(u32);
513
514 if (length > 0) {
515 u32 *dst = (u32 *) buffer;
516
517 *start = buffer;
518 memset(dst, 0, length);
519
520 for_each_possible_cpu(i) {
521 unsigned int j;
522 u32 *src;
523
524 src = ((u32 *) per_cpu_ptr(ip_rt_acct, i)) + offset;
525 for (j = 0; j < length/4; j++)
526 dst[j] += src[j];
527 }
528 }
529 return length;
530 }
531 #endif
532
533 static __init int ip_rt_proc_init(struct net *net)
534 {
535 struct proc_dir_entry *pde;
536
537 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
538 &rt_cache_seq_fops);
539 if (!pde)
540 goto err1;
541
542 pde = create_proc_entry("rt_cache", S_IRUGO, net->proc_net_stat);
543 if (!pde)
544 goto err2;
545
546 pde->proc_fops = &rt_cpu_seq_fops;
547
548 #ifdef CONFIG_NET_CLS_ROUTE
549 pde = create_proc_read_entry("rt_acct", 0, net->proc_net,
550 ip_rt_acct_read, NULL);
551 if (!pde)
552 goto err3;
553 #endif
554 return 0;
555
556 #ifdef CONFIG_NET_CLS_ROUTE
557 err3:
558 remove_proc_entry("rt_cache", net->proc_net_stat);
559 #endif
560 err2:
561 remove_proc_entry("rt_cache", net->proc_net);
562 err1:
563 return -ENOMEM;
564 }
565 #else
566 static inline int ip_rt_proc_init(struct net *net)
567 {
568 return 0;
569 }
570 #endif /* CONFIG_PROC_FS */
571
572 static __inline__ void rt_free(struct rtable *rt)
573 {
574 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
575 }
576
577 static __inline__ void rt_drop(struct rtable *rt)
578 {
579 ip_rt_put(rt);
580 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
581 }
582
583 static __inline__ int rt_fast_clean(struct rtable *rth)
584 {
585 /* Kill broadcast/multicast entries very aggresively, if they
586 collide in hash table with more useful entries */
587 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
588 rth->fl.iif && rth->u.dst.rt_next;
589 }
590
591 static __inline__ int rt_valuable(struct rtable *rth)
592 {
593 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
594 rth->u.dst.expires;
595 }
596
597 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
598 {
599 unsigned long age;
600 int ret = 0;
601
602 if (atomic_read(&rth->u.dst.__refcnt))
603 goto out;
604
605 ret = 1;
606 if (rth->u.dst.expires &&
607 time_after_eq(jiffies, rth->u.dst.expires))
608 goto out;
609
610 age = jiffies - rth->u.dst.lastuse;
611 ret = 0;
612 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
613 (age <= tmo2 && rt_valuable(rth)))
614 goto out;
615 ret = 1;
616 out: return ret;
617 }
618
619 /* Bits of score are:
620 * 31: very valuable
621 * 30: not quite useless
622 * 29..0: usage counter
623 */
624 static inline u32 rt_score(struct rtable *rt)
625 {
626 u32 score = jiffies - rt->u.dst.lastuse;
627
628 score = ~score & ~(3<<30);
629
630 if (rt_valuable(rt))
631 score |= (1<<31);
632
633 if (!rt->fl.iif ||
634 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
635 score |= (1<<30);
636
637 return score;
638 }
639
640 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
641 {
642 return ((__force u32)((fl1->nl_u.ip4_u.daddr ^ fl2->nl_u.ip4_u.daddr) |
643 (fl1->nl_u.ip4_u.saddr ^ fl2->nl_u.ip4_u.saddr)) |
644 (fl1->mark ^ fl2->mark) |
645 (*(u16 *)&fl1->nl_u.ip4_u.tos ^
646 *(u16 *)&fl2->nl_u.ip4_u.tos) |
647 (fl1->oif ^ fl2->oif) |
648 (fl1->iif ^ fl2->iif)) == 0;
649 }
650
651 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
652 {
653 return rt1->u.dst.dev->nd_net == rt2->u.dst.dev->nd_net;
654 }
655
656 /*
657 * Perform a full scan of hash table and free all entries.
658 * Can be called by a softirq or a process.
659 * In the later case, we want to be reschedule if necessary
660 */
661 static void rt_do_flush(int process_context)
662 {
663 unsigned int i;
664 struct rtable *rth, *next;
665
666 for (i = 0; i <= rt_hash_mask; i++) {
667 if (process_context && need_resched())
668 cond_resched();
669 rth = rt_hash_table[i].chain;
670 if (!rth)
671 continue;
672
673 spin_lock_bh(rt_hash_lock_addr(i));
674 rth = rt_hash_table[i].chain;
675 rt_hash_table[i].chain = NULL;
676 spin_unlock_bh(rt_hash_lock_addr(i));
677
678 for (; rth; rth = next) {
679 next = rth->u.dst.rt_next;
680 rt_free(rth);
681 }
682 }
683 }
684
685 static void rt_check_expire(void)
686 {
687 static unsigned int rover;
688 unsigned int i = rover, goal;
689 struct rtable *rth, **rthp;
690 u64 mult;
691
692 mult = ((u64)ip_rt_gc_interval) << rt_hash_log;
693 if (ip_rt_gc_timeout > 1)
694 do_div(mult, ip_rt_gc_timeout);
695 goal = (unsigned int)mult;
696 if (goal > rt_hash_mask)
697 goal = rt_hash_mask + 1;
698 for (; goal > 0; goal--) {
699 unsigned long tmo = ip_rt_gc_timeout;
700
701 i = (i + 1) & rt_hash_mask;
702 rthp = &rt_hash_table[i].chain;
703
704 if (need_resched())
705 cond_resched();
706
707 if (*rthp == NULL)
708 continue;
709 spin_lock_bh(rt_hash_lock_addr(i));
710 while ((rth = *rthp) != NULL) {
711 if (rth->u.dst.expires) {
712 /* Entry is expired even if it is in use */
713 if (time_before_eq(jiffies, rth->u.dst.expires)) {
714 tmo >>= 1;
715 rthp = &rth->u.dst.rt_next;
716 continue;
717 }
718 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout)) {
719 tmo >>= 1;
720 rthp = &rth->u.dst.rt_next;
721 continue;
722 }
723
724 /* Cleanup aged off entries. */
725 *rthp = rth->u.dst.rt_next;
726 rt_free(rth);
727 }
728 spin_unlock_bh(rt_hash_lock_addr(i));
729 }
730 rover = i;
731 }
732
733 /*
734 * rt_worker_func() is run in process context.
735 * If a whole flush was scheduled, it is done.
736 * Else, we call rt_check_expire() to scan part of the hash table
737 */
738 static void rt_worker_func(struct work_struct *work)
739 {
740 if (ip_rt_flush_expected) {
741 ip_rt_flush_expected = 0;
742 rt_do_flush(1);
743 } else
744 rt_check_expire();
745 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
746 }
747
748 /* This can run from both BH and non-BH contexts, the latter
749 * in the case of a forced flush event.
750 */
751 static void rt_run_flush(unsigned long process_context)
752 {
753 rt_deadline = 0;
754
755 get_random_bytes(&rt_hash_rnd, 4);
756
757 rt_do_flush(process_context);
758 }
759
760 static DEFINE_SPINLOCK(rt_flush_lock);
761
762 void rt_cache_flush(int delay)
763 {
764 unsigned long now = jiffies;
765 int user_mode = !in_softirq();
766
767 if (delay < 0)
768 delay = ip_rt_min_delay;
769
770 spin_lock_bh(&rt_flush_lock);
771
772 if (del_timer(&rt_flush_timer) && delay > 0 && rt_deadline) {
773 long tmo = (long)(rt_deadline - now);
774
775 /* If flush timer is already running
776 and flush request is not immediate (delay > 0):
777
778 if deadline is not achieved, prolongate timer to "delay",
779 otherwise fire it at deadline time.
780 */
781
782 if (user_mode && tmo < ip_rt_max_delay-ip_rt_min_delay)
783 tmo = 0;
784
785 if (delay > tmo)
786 delay = tmo;
787 }
788
789 if (delay <= 0) {
790 spin_unlock_bh(&rt_flush_lock);
791 rt_run_flush(user_mode);
792 return;
793 }
794
795 if (rt_deadline == 0)
796 rt_deadline = now + ip_rt_max_delay;
797
798 mod_timer(&rt_flush_timer, now+delay);
799 spin_unlock_bh(&rt_flush_lock);
800 }
801
802 /*
803 * We change rt_hash_rnd and ask next rt_worker_func() invocation
804 * to perform a flush in process context
805 */
806 static void rt_secret_rebuild(unsigned long dummy)
807 {
808 get_random_bytes(&rt_hash_rnd, 4);
809 ip_rt_flush_expected = 1;
810 cancel_delayed_work(&expires_work);
811 schedule_delayed_work(&expires_work, HZ/10);
812 mod_timer(&rt_secret_timer, jiffies + ip_rt_secret_interval);
813 }
814
815 /*
816 Short description of GC goals.
817
818 We want to build algorithm, which will keep routing cache
819 at some equilibrium point, when number of aged off entries
820 is kept approximately equal to newly generated ones.
821
822 Current expiration strength is variable "expire".
823 We try to adjust it dynamically, so that if networking
824 is idle expires is large enough to keep enough of warm entries,
825 and when load increases it reduces to limit cache size.
826 */
827
828 static int rt_garbage_collect(struct dst_ops *ops)
829 {
830 static unsigned long expire = RT_GC_TIMEOUT;
831 static unsigned long last_gc;
832 static int rover;
833 static int equilibrium;
834 struct rtable *rth, **rthp;
835 unsigned long now = jiffies;
836 int goal;
837
838 /*
839 * Garbage collection is pretty expensive,
840 * do not make it too frequently.
841 */
842
843 RT_CACHE_STAT_INC(gc_total);
844
845 if (now - last_gc < ip_rt_gc_min_interval &&
846 atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) {
847 RT_CACHE_STAT_INC(gc_ignored);
848 goto out;
849 }
850
851 /* Calculate number of entries, which we want to expire now. */
852 goal = atomic_read(&ipv4_dst_ops.entries) -
853 (ip_rt_gc_elasticity << rt_hash_log);
854 if (goal <= 0) {
855 if (equilibrium < ipv4_dst_ops.gc_thresh)
856 equilibrium = ipv4_dst_ops.gc_thresh;
857 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
858 if (goal > 0) {
859 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
860 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
861 }
862 } else {
863 /* We are in dangerous area. Try to reduce cache really
864 * aggressively.
865 */
866 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
867 equilibrium = atomic_read(&ipv4_dst_ops.entries) - goal;
868 }
869
870 if (now - last_gc >= ip_rt_gc_min_interval)
871 last_gc = now;
872
873 if (goal <= 0) {
874 equilibrium += goal;
875 goto work_done;
876 }
877
878 do {
879 int i, k;
880
881 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
882 unsigned long tmo = expire;
883
884 k = (k + 1) & rt_hash_mask;
885 rthp = &rt_hash_table[k].chain;
886 spin_lock_bh(rt_hash_lock_addr(k));
887 while ((rth = *rthp) != NULL) {
888 if (!rt_may_expire(rth, tmo, expire)) {
889 tmo >>= 1;
890 rthp = &rth->u.dst.rt_next;
891 continue;
892 }
893 *rthp = rth->u.dst.rt_next;
894 rt_free(rth);
895 goal--;
896 }
897 spin_unlock_bh(rt_hash_lock_addr(k));
898 if (goal <= 0)
899 break;
900 }
901 rover = k;
902
903 if (goal <= 0)
904 goto work_done;
905
906 /* Goal is not achieved. We stop process if:
907
908 - if expire reduced to zero. Otherwise, expire is halfed.
909 - if table is not full.
910 - if we are called from interrupt.
911 - jiffies check is just fallback/debug loop breaker.
912 We will not spin here for long time in any case.
913 */
914
915 RT_CACHE_STAT_INC(gc_goal_miss);
916
917 if (expire == 0)
918 break;
919
920 expire >>= 1;
921 #if RT_CACHE_DEBUG >= 2
922 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
923 atomic_read(&ipv4_dst_ops.entries), goal, i);
924 #endif
925
926 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
927 goto out;
928 } while (!in_softirq() && time_before_eq(jiffies, now));
929
930 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
931 goto out;
932 if (net_ratelimit())
933 printk(KERN_WARNING "dst cache overflow\n");
934 RT_CACHE_STAT_INC(gc_dst_overflow);
935 return 1;
936
937 work_done:
938 expire += ip_rt_gc_min_interval;
939 if (expire > ip_rt_gc_timeout ||
940 atomic_read(&ipv4_dst_ops.entries) < ipv4_dst_ops.gc_thresh)
941 expire = ip_rt_gc_timeout;
942 #if RT_CACHE_DEBUG >= 2
943 printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
944 atomic_read(&ipv4_dst_ops.entries), goal, rover);
945 #endif
946 out: return 0;
947 }
948
949 static int rt_intern_hash(unsigned hash, struct rtable *rt, struct rtable **rp)
950 {
951 struct rtable *rth, **rthp;
952 unsigned long now;
953 struct rtable *cand, **candp;
954 u32 min_score;
955 int chain_length;
956 int attempts = !in_softirq();
957
958 restart:
959 chain_length = 0;
960 min_score = ~(u32)0;
961 cand = NULL;
962 candp = NULL;
963 now = jiffies;
964
965 rthp = &rt_hash_table[hash].chain;
966
967 spin_lock_bh(rt_hash_lock_addr(hash));
968 while ((rth = *rthp) != NULL) {
969 if (compare_keys(&rth->fl, &rt->fl) && compare_netns(rth, rt)) {
970 /* Put it first */
971 *rthp = rth->u.dst.rt_next;
972 /*
973 * Since lookup is lockfree, the deletion
974 * must be visible to another weakly ordered CPU before
975 * the insertion at the start of the hash chain.
976 */
977 rcu_assign_pointer(rth->u.dst.rt_next,
978 rt_hash_table[hash].chain);
979 /*
980 * Since lookup is lockfree, the update writes
981 * must be ordered for consistency on SMP.
982 */
983 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
984
985 dst_use(&rth->u.dst, now);
986 spin_unlock_bh(rt_hash_lock_addr(hash));
987
988 rt_drop(rt);
989 *rp = rth;
990 return 0;
991 }
992
993 if (!atomic_read(&rth->u.dst.__refcnt)) {
994 u32 score = rt_score(rth);
995
996 if (score <= min_score) {
997 cand = rth;
998 candp = rthp;
999 min_score = score;
1000 }
1001 }
1002
1003 chain_length++;
1004
1005 rthp = &rth->u.dst.rt_next;
1006 }
1007
1008 if (cand) {
1009 /* ip_rt_gc_elasticity used to be average length of chain
1010 * length, when exceeded gc becomes really aggressive.
1011 *
1012 * The second limit is less certain. At the moment it allows
1013 * only 2 entries per bucket. We will see.
1014 */
1015 if (chain_length > ip_rt_gc_elasticity) {
1016 *candp = cand->u.dst.rt_next;
1017 rt_free(cand);
1018 }
1019 }
1020
1021 /* Try to bind route to arp only if it is output
1022 route or unicast forwarding path.
1023 */
1024 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1025 int err = arp_bind_neighbour(&rt->u.dst);
1026 if (err) {
1027 spin_unlock_bh(rt_hash_lock_addr(hash));
1028
1029 if (err != -ENOBUFS) {
1030 rt_drop(rt);
1031 return err;
1032 }
1033
1034 /* Neighbour tables are full and nothing
1035 can be released. Try to shrink route cache,
1036 it is most likely it holds some neighbour records.
1037 */
1038 if (attempts-- > 0) {
1039 int saved_elasticity = ip_rt_gc_elasticity;
1040 int saved_int = ip_rt_gc_min_interval;
1041 ip_rt_gc_elasticity = 1;
1042 ip_rt_gc_min_interval = 0;
1043 rt_garbage_collect(&ipv4_dst_ops);
1044 ip_rt_gc_min_interval = saved_int;
1045 ip_rt_gc_elasticity = saved_elasticity;
1046 goto restart;
1047 }
1048
1049 if (net_ratelimit())
1050 printk(KERN_WARNING "Neighbour table overflow.\n");
1051 rt_drop(rt);
1052 return -ENOBUFS;
1053 }
1054 }
1055
1056 rt->u.dst.rt_next = rt_hash_table[hash].chain;
1057 #if RT_CACHE_DEBUG >= 2
1058 if (rt->u.dst.rt_next) {
1059 struct rtable *trt;
1060 printk(KERN_DEBUG "rt_cache @%02x: %u.%u.%u.%u", hash,
1061 NIPQUAD(rt->rt_dst));
1062 for (trt = rt->u.dst.rt_next; trt; trt = trt->u.dst.rt_next)
1063 printk(" . %u.%u.%u.%u", NIPQUAD(trt->rt_dst));
1064 printk("\n");
1065 }
1066 #endif
1067 rt_hash_table[hash].chain = rt;
1068 spin_unlock_bh(rt_hash_lock_addr(hash));
1069 *rp = rt;
1070 return 0;
1071 }
1072
1073 void rt_bind_peer(struct rtable *rt, int create)
1074 {
1075 static DEFINE_SPINLOCK(rt_peer_lock);
1076 struct inet_peer *peer;
1077
1078 peer = inet_getpeer(rt->rt_dst, create);
1079
1080 spin_lock_bh(&rt_peer_lock);
1081 if (rt->peer == NULL) {
1082 rt->peer = peer;
1083 peer = NULL;
1084 }
1085 spin_unlock_bh(&rt_peer_lock);
1086 if (peer)
1087 inet_putpeer(peer);
1088 }
1089
1090 /*
1091 * Peer allocation may fail only in serious out-of-memory conditions. However
1092 * we still can generate some output.
1093 * Random ID selection looks a bit dangerous because we have no chances to
1094 * select ID being unique in a reasonable period of time.
1095 * But broken packet identifier may be better than no packet at all.
1096 */
1097 static void ip_select_fb_ident(struct iphdr *iph)
1098 {
1099 static DEFINE_SPINLOCK(ip_fb_id_lock);
1100 static u32 ip_fallback_id;
1101 u32 salt;
1102
1103 spin_lock_bh(&ip_fb_id_lock);
1104 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1105 iph->id = htons(salt & 0xFFFF);
1106 ip_fallback_id = salt;
1107 spin_unlock_bh(&ip_fb_id_lock);
1108 }
1109
1110 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1111 {
1112 struct rtable *rt = (struct rtable *) dst;
1113
1114 if (rt) {
1115 if (rt->peer == NULL)
1116 rt_bind_peer(rt, 1);
1117
1118 /* If peer is attached to destination, it is never detached,
1119 so that we need not to grab a lock to dereference it.
1120 */
1121 if (rt->peer) {
1122 iph->id = htons(inet_getid(rt->peer, more));
1123 return;
1124 }
1125 } else
1126 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1127 __builtin_return_address(0));
1128
1129 ip_select_fb_ident(iph);
1130 }
1131
1132 static void rt_del(unsigned hash, struct rtable *rt)
1133 {
1134 struct rtable **rthp;
1135
1136 spin_lock_bh(rt_hash_lock_addr(hash));
1137 ip_rt_put(rt);
1138 for (rthp = &rt_hash_table[hash].chain; *rthp;
1139 rthp = &(*rthp)->u.dst.rt_next)
1140 if (*rthp == rt) {
1141 *rthp = rt->u.dst.rt_next;
1142 rt_free(rt);
1143 break;
1144 }
1145 spin_unlock_bh(rt_hash_lock_addr(hash));
1146 }
1147
1148 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1149 __be32 saddr, struct net_device *dev)
1150 {
1151 int i, k;
1152 struct in_device *in_dev = in_dev_get(dev);
1153 struct rtable *rth, **rthp;
1154 __be32 skeys[2] = { saddr, 0 };
1155 int ikeys[2] = { dev->ifindex, 0 };
1156 struct netevent_redirect netevent;
1157
1158 if (!in_dev)
1159 return;
1160
1161 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev)
1162 || ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw)
1163 || ipv4_is_zeronet(new_gw))
1164 goto reject_redirect;
1165
1166 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1167 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1168 goto reject_redirect;
1169 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1170 goto reject_redirect;
1171 } else {
1172 if (inet_addr_type(&init_net, new_gw) != RTN_UNICAST)
1173 goto reject_redirect;
1174 }
1175
1176 for (i = 0; i < 2; i++) {
1177 for (k = 0; k < 2; k++) {
1178 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k]);
1179
1180 rthp=&rt_hash_table[hash].chain;
1181
1182 rcu_read_lock();
1183 while ((rth = rcu_dereference(*rthp)) != NULL) {
1184 struct rtable *rt;
1185
1186 if (rth->fl.fl4_dst != daddr ||
1187 rth->fl.fl4_src != skeys[i] ||
1188 rth->fl.oif != ikeys[k] ||
1189 rth->fl.iif != 0) {
1190 rthp = &rth->u.dst.rt_next;
1191 continue;
1192 }
1193
1194 if (rth->rt_dst != daddr ||
1195 rth->rt_src != saddr ||
1196 rth->u.dst.error ||
1197 rth->rt_gateway != old_gw ||
1198 rth->u.dst.dev != dev)
1199 break;
1200
1201 dst_hold(&rth->u.dst);
1202 rcu_read_unlock();
1203
1204 rt = dst_alloc(&ipv4_dst_ops);
1205 if (rt == NULL) {
1206 ip_rt_put(rth);
1207 in_dev_put(in_dev);
1208 return;
1209 }
1210
1211 /* Copy all the information. */
1212 *rt = *rth;
1213 INIT_RCU_HEAD(&rt->u.dst.rcu_head);
1214 rt->u.dst.__use = 1;
1215 atomic_set(&rt->u.dst.__refcnt, 1);
1216 rt->u.dst.child = NULL;
1217 if (rt->u.dst.dev)
1218 dev_hold(rt->u.dst.dev);
1219 if (rt->idev)
1220 in_dev_hold(rt->idev);
1221 rt->u.dst.obsolete = 0;
1222 rt->u.dst.lastuse = jiffies;
1223 rt->u.dst.path = &rt->u.dst;
1224 rt->u.dst.neighbour = NULL;
1225 rt->u.dst.hh = NULL;
1226 rt->u.dst.xfrm = NULL;
1227
1228 rt->rt_flags |= RTCF_REDIRECTED;
1229
1230 /* Gateway is different ... */
1231 rt->rt_gateway = new_gw;
1232
1233 /* Redirect received -> path was valid */
1234 dst_confirm(&rth->u.dst);
1235
1236 if (rt->peer)
1237 atomic_inc(&rt->peer->refcnt);
1238
1239 if (arp_bind_neighbour(&rt->u.dst) ||
1240 !(rt->u.dst.neighbour->nud_state &
1241 NUD_VALID)) {
1242 if (rt->u.dst.neighbour)
1243 neigh_event_send(rt->u.dst.neighbour, NULL);
1244 ip_rt_put(rth);
1245 rt_drop(rt);
1246 goto do_next;
1247 }
1248
1249 netevent.old = &rth->u.dst;
1250 netevent.new = &rt->u.dst;
1251 call_netevent_notifiers(NETEVENT_REDIRECT,
1252 &netevent);
1253
1254 rt_del(hash, rth);
1255 if (!rt_intern_hash(hash, rt, &rt))
1256 ip_rt_put(rt);
1257 goto do_next;
1258 }
1259 rcu_read_unlock();
1260 do_next:
1261 ;
1262 }
1263 }
1264 in_dev_put(in_dev);
1265 return;
1266
1267 reject_redirect:
1268 #ifdef CONFIG_IP_ROUTE_VERBOSE
1269 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1270 printk(KERN_INFO "Redirect from %u.%u.%u.%u on %s about "
1271 "%u.%u.%u.%u ignored.\n"
1272 " Advised path = %u.%u.%u.%u -> %u.%u.%u.%u\n",
1273 NIPQUAD(old_gw), dev->name, NIPQUAD(new_gw),
1274 NIPQUAD(saddr), NIPQUAD(daddr));
1275 #endif
1276 in_dev_put(in_dev);
1277 }
1278
1279 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1280 {
1281 struct rtable *rt = (struct rtable*)dst;
1282 struct dst_entry *ret = dst;
1283
1284 if (rt) {
1285 if (dst->obsolete) {
1286 ip_rt_put(rt);
1287 ret = NULL;
1288 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1289 rt->u.dst.expires) {
1290 unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1291 rt->fl.oif);
1292 #if RT_CACHE_DEBUG >= 1
1293 printk(KERN_DEBUG "ipv4_negative_advice: redirect to "
1294 "%u.%u.%u.%u/%02x dropped\n",
1295 NIPQUAD(rt->rt_dst), rt->fl.fl4_tos);
1296 #endif
1297 rt_del(hash, rt);
1298 ret = NULL;
1299 }
1300 }
1301 return ret;
1302 }
1303
1304 /*
1305 * Algorithm:
1306 * 1. The first ip_rt_redirect_number redirects are sent
1307 * with exponential backoff, then we stop sending them at all,
1308 * assuming that the host ignores our redirects.
1309 * 2. If we did not see packets requiring redirects
1310 * during ip_rt_redirect_silence, we assume that the host
1311 * forgot redirected route and start to send redirects again.
1312 *
1313 * This algorithm is much cheaper and more intelligent than dumb load limiting
1314 * in icmp.c.
1315 *
1316 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1317 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1318 */
1319
1320 void ip_rt_send_redirect(struct sk_buff *skb)
1321 {
1322 struct rtable *rt = (struct rtable*)skb->dst;
1323 struct in_device *in_dev = in_dev_get(rt->u.dst.dev);
1324
1325 if (!in_dev)
1326 return;
1327
1328 if (!IN_DEV_TX_REDIRECTS(in_dev))
1329 goto out;
1330
1331 /* No redirected packets during ip_rt_redirect_silence;
1332 * reset the algorithm.
1333 */
1334 if (time_after(jiffies, rt->u.dst.rate_last + ip_rt_redirect_silence))
1335 rt->u.dst.rate_tokens = 0;
1336
1337 /* Too many ignored redirects; do not send anything
1338 * set u.dst.rate_last to the last seen redirected packet.
1339 */
1340 if (rt->u.dst.rate_tokens >= ip_rt_redirect_number) {
1341 rt->u.dst.rate_last = jiffies;
1342 goto out;
1343 }
1344
1345 /* Check for load limit; set rate_last to the latest sent
1346 * redirect.
1347 */
1348 if (rt->u.dst.rate_tokens == 0 ||
1349 time_after(jiffies,
1350 (rt->u.dst.rate_last +
1351 (ip_rt_redirect_load << rt->u.dst.rate_tokens)))) {
1352 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1353 rt->u.dst.rate_last = jiffies;
1354 ++rt->u.dst.rate_tokens;
1355 #ifdef CONFIG_IP_ROUTE_VERBOSE
1356 if (IN_DEV_LOG_MARTIANS(in_dev) &&
1357 rt->u.dst.rate_tokens == ip_rt_redirect_number &&
1358 net_ratelimit())
1359 printk(KERN_WARNING "host %u.%u.%u.%u/if%d ignores "
1360 "redirects for %u.%u.%u.%u to %u.%u.%u.%u.\n",
1361 NIPQUAD(rt->rt_src), rt->rt_iif,
1362 NIPQUAD(rt->rt_dst), NIPQUAD(rt->rt_gateway));
1363 #endif
1364 }
1365 out:
1366 in_dev_put(in_dev);
1367 }
1368
1369 static int ip_error(struct sk_buff *skb)
1370 {
1371 struct rtable *rt = (struct rtable*)skb->dst;
1372 unsigned long now;
1373 int code;
1374
1375 switch (rt->u.dst.error) {
1376 case EINVAL:
1377 default:
1378 goto out;
1379 case EHOSTUNREACH:
1380 code = ICMP_HOST_UNREACH;
1381 break;
1382 case ENETUNREACH:
1383 code = ICMP_NET_UNREACH;
1384 IP_INC_STATS_BH(IPSTATS_MIB_INNOROUTES);
1385 break;
1386 case EACCES:
1387 code = ICMP_PKT_FILTERED;
1388 break;
1389 }
1390
1391 now = jiffies;
1392 rt->u.dst.rate_tokens += now - rt->u.dst.rate_last;
1393 if (rt->u.dst.rate_tokens > ip_rt_error_burst)
1394 rt->u.dst.rate_tokens = ip_rt_error_burst;
1395 rt->u.dst.rate_last = now;
1396 if (rt->u.dst.rate_tokens >= ip_rt_error_cost) {
1397 rt->u.dst.rate_tokens -= ip_rt_error_cost;
1398 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1399 }
1400
1401 out: kfree_skb(skb);
1402 return 0;
1403 }
1404
1405 /*
1406 * The last two values are not from the RFC but
1407 * are needed for AMPRnet AX.25 paths.
1408 */
1409
1410 static const unsigned short mtu_plateau[] =
1411 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1412
1413 static __inline__ unsigned short guess_mtu(unsigned short old_mtu)
1414 {
1415 int i;
1416
1417 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1418 if (old_mtu > mtu_plateau[i])
1419 return mtu_plateau[i];
1420 return 68;
1421 }
1422
1423 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1424 unsigned short new_mtu)
1425 {
1426 int i;
1427 unsigned short old_mtu = ntohs(iph->tot_len);
1428 struct rtable *rth;
1429 __be32 skeys[2] = { iph->saddr, 0, };
1430 __be32 daddr = iph->daddr;
1431 unsigned short est_mtu = 0;
1432
1433 if (ipv4_config.no_pmtu_disc)
1434 return 0;
1435
1436 for (i = 0; i < 2; i++) {
1437 unsigned hash = rt_hash(daddr, skeys[i], 0);
1438
1439 rcu_read_lock();
1440 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1441 rth = rcu_dereference(rth->u.dst.rt_next)) {
1442 if (rth->fl.fl4_dst == daddr &&
1443 rth->fl.fl4_src == skeys[i] &&
1444 rth->rt_dst == daddr &&
1445 rth->rt_src == iph->saddr &&
1446 rth->fl.iif == 0 &&
1447 !(dst_metric_locked(&rth->u.dst, RTAX_MTU)) &&
1448 rth->u.dst.dev->nd_net == net) {
1449 unsigned short mtu = new_mtu;
1450
1451 if (new_mtu < 68 || new_mtu >= old_mtu) {
1452
1453 /* BSD 4.2 compatibility hack :-( */
1454 if (mtu == 0 &&
1455 old_mtu >= rth->u.dst.metrics[RTAX_MTU-1] &&
1456 old_mtu >= 68 + (iph->ihl << 2))
1457 old_mtu -= iph->ihl << 2;
1458
1459 mtu = guess_mtu(old_mtu);
1460 }
1461 if (mtu <= rth->u.dst.metrics[RTAX_MTU-1]) {
1462 if (mtu < rth->u.dst.metrics[RTAX_MTU-1]) {
1463 dst_confirm(&rth->u.dst);
1464 if (mtu < ip_rt_min_pmtu) {
1465 mtu = ip_rt_min_pmtu;
1466 rth->u.dst.metrics[RTAX_LOCK-1] |=
1467 (1 << RTAX_MTU);
1468 }
1469 rth->u.dst.metrics[RTAX_MTU-1] = mtu;
1470 dst_set_expires(&rth->u.dst,
1471 ip_rt_mtu_expires);
1472 }
1473 est_mtu = mtu;
1474 }
1475 }
1476 }
1477 rcu_read_unlock();
1478 }
1479 return est_mtu ? : new_mtu;
1480 }
1481
1482 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1483 {
1484 if (dst->metrics[RTAX_MTU-1] > mtu && mtu >= 68 &&
1485 !(dst_metric_locked(dst, RTAX_MTU))) {
1486 if (mtu < ip_rt_min_pmtu) {
1487 mtu = ip_rt_min_pmtu;
1488 dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1489 }
1490 dst->metrics[RTAX_MTU-1] = mtu;
1491 dst_set_expires(dst, ip_rt_mtu_expires);
1492 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1493 }
1494 }
1495
1496 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1497 {
1498 return NULL;
1499 }
1500
1501 static void ipv4_dst_destroy(struct dst_entry *dst)
1502 {
1503 struct rtable *rt = (struct rtable *) dst;
1504 struct inet_peer *peer = rt->peer;
1505 struct in_device *idev = rt->idev;
1506
1507 if (peer) {
1508 rt->peer = NULL;
1509 inet_putpeer(peer);
1510 }
1511
1512 if (idev) {
1513 rt->idev = NULL;
1514 in_dev_put(idev);
1515 }
1516 }
1517
1518 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1519 int how)
1520 {
1521 struct rtable *rt = (struct rtable *) dst;
1522 struct in_device *idev = rt->idev;
1523 if (dev != dev->nd_net->loopback_dev && idev && idev->dev == dev) {
1524 struct in_device *loopback_idev =
1525 in_dev_get(dev->nd_net->loopback_dev);
1526 if (loopback_idev) {
1527 rt->idev = loopback_idev;
1528 in_dev_put(idev);
1529 }
1530 }
1531 }
1532
1533 static void ipv4_link_failure(struct sk_buff *skb)
1534 {
1535 struct rtable *rt;
1536
1537 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1538
1539 rt = (struct rtable *) skb->dst;
1540 if (rt)
1541 dst_set_expires(&rt->u.dst, 0);
1542 }
1543
1544 static int ip_rt_bug(struct sk_buff *skb)
1545 {
1546 printk(KERN_DEBUG "ip_rt_bug: %u.%u.%u.%u -> %u.%u.%u.%u, %s\n",
1547 NIPQUAD(ip_hdr(skb)->saddr), NIPQUAD(ip_hdr(skb)->daddr),
1548 skb->dev ? skb->dev->name : "?");
1549 kfree_skb(skb);
1550 return 0;
1551 }
1552
1553 /*
1554 We do not cache source address of outgoing interface,
1555 because it is used only by IP RR, TS and SRR options,
1556 so that it out of fast path.
1557
1558 BTW remember: "addr" is allowed to be not aligned
1559 in IP options!
1560 */
1561
1562 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1563 {
1564 __be32 src;
1565 struct fib_result res;
1566
1567 if (rt->fl.iif == 0)
1568 src = rt->rt_src;
1569 else if (fib_lookup(rt->u.dst.dev->nd_net, &rt->fl, &res) == 0) {
1570 src = FIB_RES_PREFSRC(res);
1571 fib_res_put(&res);
1572 } else
1573 src = inet_select_addr(rt->u.dst.dev, rt->rt_gateway,
1574 RT_SCOPE_UNIVERSE);
1575 memcpy(addr, &src, 4);
1576 }
1577
1578 #ifdef CONFIG_NET_CLS_ROUTE
1579 static void set_class_tag(struct rtable *rt, u32 tag)
1580 {
1581 if (!(rt->u.dst.tclassid & 0xFFFF))
1582 rt->u.dst.tclassid |= tag & 0xFFFF;
1583 if (!(rt->u.dst.tclassid & 0xFFFF0000))
1584 rt->u.dst.tclassid |= tag & 0xFFFF0000;
1585 }
1586 #endif
1587
1588 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1589 {
1590 struct fib_info *fi = res->fi;
1591
1592 if (fi) {
1593 if (FIB_RES_GW(*res) &&
1594 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1595 rt->rt_gateway = FIB_RES_GW(*res);
1596 memcpy(rt->u.dst.metrics, fi->fib_metrics,
1597 sizeof(rt->u.dst.metrics));
1598 if (fi->fib_mtu == 0) {
1599 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
1600 if (rt->u.dst.metrics[RTAX_LOCK-1] & (1 << RTAX_MTU) &&
1601 rt->rt_gateway != rt->rt_dst &&
1602 rt->u.dst.dev->mtu > 576)
1603 rt->u.dst.metrics[RTAX_MTU-1] = 576;
1604 }
1605 #ifdef CONFIG_NET_CLS_ROUTE
1606 rt->u.dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1607 #endif
1608 } else
1609 rt->u.dst.metrics[RTAX_MTU-1]= rt->u.dst.dev->mtu;
1610
1611 if (rt->u.dst.metrics[RTAX_HOPLIMIT-1] == 0)
1612 rt->u.dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1613 if (rt->u.dst.metrics[RTAX_MTU-1] > IP_MAX_MTU)
1614 rt->u.dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1615 if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0)
1616 rt->u.dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->u.dst.dev->mtu - 40,
1617 ip_rt_min_advmss);
1618 if (rt->u.dst.metrics[RTAX_ADVMSS-1] > 65535 - 40)
1619 rt->u.dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1620
1621 #ifdef CONFIG_NET_CLS_ROUTE
1622 #ifdef CONFIG_IP_MULTIPLE_TABLES
1623 set_class_tag(rt, fib_rules_tclass(res));
1624 #endif
1625 set_class_tag(rt, itag);
1626 #endif
1627 rt->rt_type = res->type;
1628 }
1629
1630 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1631 u8 tos, struct net_device *dev, int our)
1632 {
1633 unsigned hash;
1634 struct rtable *rth;
1635 __be32 spec_dst;
1636 struct in_device *in_dev = in_dev_get(dev);
1637 u32 itag = 0;
1638
1639 /* Primary sanity checks. */
1640
1641 if (in_dev == NULL)
1642 return -EINVAL;
1643
1644 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1645 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1646 goto e_inval;
1647
1648 if (ipv4_is_zeronet(saddr)) {
1649 if (!ipv4_is_local_multicast(daddr))
1650 goto e_inval;
1651 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1652 } else if (fib_validate_source(saddr, 0, tos, 0,
1653 dev, &spec_dst, &itag) < 0)
1654 goto e_inval;
1655
1656 rth = dst_alloc(&ipv4_dst_ops);
1657 if (!rth)
1658 goto e_nobufs;
1659
1660 rth->u.dst.output= ip_rt_bug;
1661
1662 atomic_set(&rth->u.dst.__refcnt, 1);
1663 rth->u.dst.flags= DST_HOST;
1664 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1665 rth->u.dst.flags |= DST_NOPOLICY;
1666 rth->fl.fl4_dst = daddr;
1667 rth->rt_dst = daddr;
1668 rth->fl.fl4_tos = tos;
1669 rth->fl.mark = skb->mark;
1670 rth->fl.fl4_src = saddr;
1671 rth->rt_src = saddr;
1672 #ifdef CONFIG_NET_CLS_ROUTE
1673 rth->u.dst.tclassid = itag;
1674 #endif
1675 rth->rt_iif =
1676 rth->fl.iif = dev->ifindex;
1677 rth->u.dst.dev = init_net.loopback_dev;
1678 dev_hold(rth->u.dst.dev);
1679 rth->idev = in_dev_get(rth->u.dst.dev);
1680 rth->fl.oif = 0;
1681 rth->rt_gateway = daddr;
1682 rth->rt_spec_dst= spec_dst;
1683 rth->rt_type = RTN_MULTICAST;
1684 rth->rt_flags = RTCF_MULTICAST;
1685 if (our) {
1686 rth->u.dst.input= ip_local_deliver;
1687 rth->rt_flags |= RTCF_LOCAL;
1688 }
1689
1690 #ifdef CONFIG_IP_MROUTE
1691 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1692 rth->u.dst.input = ip_mr_input;
1693 #endif
1694 RT_CACHE_STAT_INC(in_slow_mc);
1695
1696 in_dev_put(in_dev);
1697 hash = rt_hash(daddr, saddr, dev->ifindex);
1698 return rt_intern_hash(hash, rth, (struct rtable**) &skb->dst);
1699
1700 e_nobufs:
1701 in_dev_put(in_dev);
1702 return -ENOBUFS;
1703
1704 e_inval:
1705 in_dev_put(in_dev);
1706 return -EINVAL;
1707 }
1708
1709
1710 static void ip_handle_martian_source(struct net_device *dev,
1711 struct in_device *in_dev,
1712 struct sk_buff *skb,
1713 __be32 daddr,
1714 __be32 saddr)
1715 {
1716 RT_CACHE_STAT_INC(in_martian_src);
1717 #ifdef CONFIG_IP_ROUTE_VERBOSE
1718 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1719 /*
1720 * RFC1812 recommendation, if source is martian,
1721 * the only hint is MAC header.
1722 */
1723 printk(KERN_WARNING "martian source %u.%u.%u.%u from "
1724 "%u.%u.%u.%u, on dev %s\n",
1725 NIPQUAD(daddr), NIPQUAD(saddr), dev->name);
1726 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1727 int i;
1728 const unsigned char *p = skb_mac_header(skb);
1729 printk(KERN_WARNING "ll header: ");
1730 for (i = 0; i < dev->hard_header_len; i++, p++) {
1731 printk("%02x", *p);
1732 if (i < (dev->hard_header_len - 1))
1733 printk(":");
1734 }
1735 printk("\n");
1736 }
1737 }
1738 #endif
1739 }
1740
1741 static inline int __mkroute_input(struct sk_buff *skb,
1742 struct fib_result* res,
1743 struct in_device *in_dev,
1744 __be32 daddr, __be32 saddr, u32 tos,
1745 struct rtable **result)
1746 {
1747
1748 struct rtable *rth;
1749 int err;
1750 struct in_device *out_dev;
1751 unsigned flags = 0;
1752 __be32 spec_dst;
1753 u32 itag;
1754
1755 /* get a working reference to the output device */
1756 out_dev = in_dev_get(FIB_RES_DEV(*res));
1757 if (out_dev == NULL) {
1758 if (net_ratelimit())
1759 printk(KERN_CRIT "Bug in ip_route_input" \
1760 "_slow(). Please, report\n");
1761 return -EINVAL;
1762 }
1763
1764
1765 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1766 in_dev->dev, &spec_dst, &itag);
1767 if (err < 0) {
1768 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1769 saddr);
1770
1771 err = -EINVAL;
1772 goto cleanup;
1773 }
1774
1775 if (err)
1776 flags |= RTCF_DIRECTSRC;
1777
1778 if (out_dev == in_dev && err && !(flags & RTCF_MASQ) &&
1779 (IN_DEV_SHARED_MEDIA(out_dev) ||
1780 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1781 flags |= RTCF_DOREDIRECT;
1782
1783 if (skb->protocol != htons(ETH_P_IP)) {
1784 /* Not IP (i.e. ARP). Do not create route, if it is
1785 * invalid for proxy arp. DNAT routes are always valid.
1786 */
1787 if (out_dev == in_dev) {
1788 err = -EINVAL;
1789 goto cleanup;
1790 }
1791 }
1792
1793
1794 rth = dst_alloc(&ipv4_dst_ops);
1795 if (!rth) {
1796 err = -ENOBUFS;
1797 goto cleanup;
1798 }
1799
1800 atomic_set(&rth->u.dst.__refcnt, 1);
1801 rth->u.dst.flags= DST_HOST;
1802 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1803 rth->u.dst.flags |= DST_NOPOLICY;
1804 if (IN_DEV_CONF_GET(out_dev, NOXFRM))
1805 rth->u.dst.flags |= DST_NOXFRM;
1806 rth->fl.fl4_dst = daddr;
1807 rth->rt_dst = daddr;
1808 rth->fl.fl4_tos = tos;
1809 rth->fl.mark = skb->mark;
1810 rth->fl.fl4_src = saddr;
1811 rth->rt_src = saddr;
1812 rth->rt_gateway = daddr;
1813 rth->rt_iif =
1814 rth->fl.iif = in_dev->dev->ifindex;
1815 rth->u.dst.dev = (out_dev)->dev;
1816 dev_hold(rth->u.dst.dev);
1817 rth->idev = in_dev_get(rth->u.dst.dev);
1818 rth->fl.oif = 0;
1819 rth->rt_spec_dst= spec_dst;
1820
1821 rth->u.dst.input = ip_forward;
1822 rth->u.dst.output = ip_output;
1823
1824 rt_set_nexthop(rth, res, itag);
1825
1826 rth->rt_flags = flags;
1827
1828 *result = rth;
1829 err = 0;
1830 cleanup:
1831 /* release the working reference to the output device */
1832 in_dev_put(out_dev);
1833 return err;
1834 }
1835
1836 static inline int ip_mkroute_input(struct sk_buff *skb,
1837 struct fib_result* res,
1838 const struct flowi *fl,
1839 struct in_device *in_dev,
1840 __be32 daddr, __be32 saddr, u32 tos)
1841 {
1842 struct rtable* rth = NULL;
1843 int err;
1844 unsigned hash;
1845
1846 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1847 if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
1848 fib_select_multipath(fl, res);
1849 #endif
1850
1851 /* create a routing cache entry */
1852 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
1853 if (err)
1854 return err;
1855
1856 /* put it into the cache */
1857 hash = rt_hash(daddr, saddr, fl->iif);
1858 return rt_intern_hash(hash, rth, (struct rtable**)&skb->dst);
1859 }
1860
1861 /*
1862 * NOTE. We drop all the packets that has local source
1863 * addresses, because every properly looped back packet
1864 * must have correct destination already attached by output routine.
1865 *
1866 * Such approach solves two big problems:
1867 * 1. Not simplex devices are handled properly.
1868 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1869 */
1870
1871 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1872 u8 tos, struct net_device *dev)
1873 {
1874 struct fib_result res;
1875 struct in_device *in_dev = in_dev_get(dev);
1876 struct flowi fl = { .nl_u = { .ip4_u =
1877 { .daddr = daddr,
1878 .saddr = saddr,
1879 .tos = tos,
1880 .scope = RT_SCOPE_UNIVERSE,
1881 } },
1882 .mark = skb->mark,
1883 .iif = dev->ifindex };
1884 unsigned flags = 0;
1885 u32 itag = 0;
1886 struct rtable * rth;
1887 unsigned hash;
1888 __be32 spec_dst;
1889 int err = -EINVAL;
1890 int free_res = 0;
1891 struct net * net = dev->nd_net;
1892
1893 /* IP on this device is disabled. */
1894
1895 if (!in_dev)
1896 goto out;
1897
1898 /* Check for the most weird martians, which can be not detected
1899 by fib_lookup.
1900 */
1901
1902 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1903 ipv4_is_loopback(saddr))
1904 goto martian_source;
1905
1906 if (daddr == htonl(0xFFFFFFFF) || (saddr == 0 && daddr == 0))
1907 goto brd_input;
1908
1909 /* Accept zero addresses only to limited broadcast;
1910 * I even do not know to fix it or not. Waiting for complains :-)
1911 */
1912 if (ipv4_is_zeronet(saddr))
1913 goto martian_source;
1914
1915 if (ipv4_is_lbcast(daddr) || ipv4_is_zeronet(daddr) ||
1916 ipv4_is_loopback(daddr))
1917 goto martian_destination;
1918
1919 /*
1920 * Now we are ready to route packet.
1921 */
1922 if ((err = fib_lookup(net, &fl, &res)) != 0) {
1923 if (!IN_DEV_FORWARD(in_dev))
1924 goto e_hostunreach;
1925 goto no_route;
1926 }
1927 free_res = 1;
1928
1929 RT_CACHE_STAT_INC(in_slow_tot);
1930
1931 if (res.type == RTN_BROADCAST)
1932 goto brd_input;
1933
1934 if (res.type == RTN_LOCAL) {
1935 int result;
1936 result = fib_validate_source(saddr, daddr, tos,
1937 net->loopback_dev->ifindex,
1938 dev, &spec_dst, &itag);
1939 if (result < 0)
1940 goto martian_source;
1941 if (result)
1942 flags |= RTCF_DIRECTSRC;
1943 spec_dst = daddr;
1944 goto local_input;
1945 }
1946
1947 if (!IN_DEV_FORWARD(in_dev))
1948 goto e_hostunreach;
1949 if (res.type != RTN_UNICAST)
1950 goto martian_destination;
1951
1952 err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
1953 done:
1954 in_dev_put(in_dev);
1955 if (free_res)
1956 fib_res_put(&res);
1957 out: return err;
1958
1959 brd_input:
1960 if (skb->protocol != htons(ETH_P_IP))
1961 goto e_inval;
1962
1963 if (ipv4_is_zeronet(saddr))
1964 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1965 else {
1966 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1967 &itag);
1968 if (err < 0)
1969 goto martian_source;
1970 if (err)
1971 flags |= RTCF_DIRECTSRC;
1972 }
1973 flags |= RTCF_BROADCAST;
1974 res.type = RTN_BROADCAST;
1975 RT_CACHE_STAT_INC(in_brd);
1976
1977 local_input:
1978 rth = dst_alloc(&ipv4_dst_ops);
1979 if (!rth)
1980 goto e_nobufs;
1981
1982 rth->u.dst.output= ip_rt_bug;
1983
1984 atomic_set(&rth->u.dst.__refcnt, 1);
1985 rth->u.dst.flags= DST_HOST;
1986 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1987 rth->u.dst.flags |= DST_NOPOLICY;
1988 rth->fl.fl4_dst = daddr;
1989 rth->rt_dst = daddr;
1990 rth->fl.fl4_tos = tos;
1991 rth->fl.mark = skb->mark;
1992 rth->fl.fl4_src = saddr;
1993 rth->rt_src = saddr;
1994 #ifdef CONFIG_NET_CLS_ROUTE
1995 rth->u.dst.tclassid = itag;
1996 #endif
1997 rth->rt_iif =
1998 rth->fl.iif = dev->ifindex;
1999 rth->u.dst.dev = net->loopback_dev;
2000 dev_hold(rth->u.dst.dev);
2001 rth->idev = in_dev_get(rth->u.dst.dev);
2002 rth->rt_gateway = daddr;
2003 rth->rt_spec_dst= spec_dst;
2004 rth->u.dst.input= ip_local_deliver;
2005 rth->rt_flags = flags|RTCF_LOCAL;
2006 if (res.type == RTN_UNREACHABLE) {
2007 rth->u.dst.input= ip_error;
2008 rth->u.dst.error= -err;
2009 rth->rt_flags &= ~RTCF_LOCAL;
2010 }
2011 rth->rt_type = res.type;
2012 hash = rt_hash(daddr, saddr, fl.iif);
2013 err = rt_intern_hash(hash, rth, (struct rtable**)&skb->dst);
2014 goto done;
2015
2016 no_route:
2017 RT_CACHE_STAT_INC(in_no_route);
2018 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2019 res.type = RTN_UNREACHABLE;
2020 if (err == -ESRCH)
2021 err = -ENETUNREACH;
2022 goto local_input;
2023
2024 /*
2025 * Do not cache martian addresses: they should be logged (RFC1812)
2026 */
2027 martian_destination:
2028 RT_CACHE_STAT_INC(in_martian_dst);
2029 #ifdef CONFIG_IP_ROUTE_VERBOSE
2030 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2031 printk(KERN_WARNING "martian destination %u.%u.%u.%u from "
2032 "%u.%u.%u.%u, dev %s\n",
2033 NIPQUAD(daddr), NIPQUAD(saddr), dev->name);
2034 #endif
2035
2036 e_hostunreach:
2037 err = -EHOSTUNREACH;
2038 goto done;
2039
2040 e_inval:
2041 err = -EINVAL;
2042 goto done;
2043
2044 e_nobufs:
2045 err = -ENOBUFS;
2046 goto done;
2047
2048 martian_source:
2049 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2050 goto e_inval;
2051 }
2052
2053 int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2054 u8 tos, struct net_device *dev)
2055 {
2056 struct rtable * rth;
2057 unsigned hash;
2058 int iif = dev->ifindex;
2059 struct net *net;
2060
2061 net = skb->dev->nd_net;
2062 tos &= IPTOS_RT_MASK;
2063 hash = rt_hash(daddr, saddr, iif);
2064
2065 rcu_read_lock();
2066 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2067 rth = rcu_dereference(rth->u.dst.rt_next)) {
2068 if (rth->fl.fl4_dst == daddr &&
2069 rth->fl.fl4_src == saddr &&
2070 rth->fl.iif == iif &&
2071 rth->fl.oif == 0 &&
2072 rth->fl.mark == skb->mark &&
2073 rth->fl.fl4_tos == tos &&
2074 rth->u.dst.dev->nd_net == net) {
2075 dst_use(&rth->u.dst, jiffies);
2076 RT_CACHE_STAT_INC(in_hit);
2077 rcu_read_unlock();
2078 skb->dst = (struct dst_entry*)rth;
2079 return 0;
2080 }
2081 RT_CACHE_STAT_INC(in_hlist_search);
2082 }
2083 rcu_read_unlock();
2084
2085 /* Multicast recognition logic is moved from route cache to here.
2086 The problem was that too many Ethernet cards have broken/missing
2087 hardware multicast filters :-( As result the host on multicasting
2088 network acquires a lot of useless route cache entries, sort of
2089 SDR messages from all the world. Now we try to get rid of them.
2090 Really, provided software IP multicast filter is organized
2091 reasonably (at least, hashed), it does not result in a slowdown
2092 comparing with route cache reject entries.
2093 Note, that multicast routers are not affected, because
2094 route cache entry is created eventually.
2095 */
2096 if (ipv4_is_multicast(daddr)) {
2097 struct in_device *in_dev;
2098
2099 rcu_read_lock();
2100 if ((in_dev = __in_dev_get_rcu(dev)) != NULL) {
2101 int our = ip_check_mc(in_dev, daddr, saddr,
2102 ip_hdr(skb)->protocol);
2103 if (our
2104 #ifdef CONFIG_IP_MROUTE
2105 || (!ipv4_is_local_multicast(daddr) &&
2106 IN_DEV_MFORWARD(in_dev))
2107 #endif
2108 ) {
2109 rcu_read_unlock();
2110 return ip_route_input_mc(skb, daddr, saddr,
2111 tos, dev, our);
2112 }
2113 }
2114 rcu_read_unlock();
2115 return -EINVAL;
2116 }
2117 return ip_route_input_slow(skb, daddr, saddr, tos, dev);
2118 }
2119
2120 static inline int __mkroute_output(struct rtable **result,
2121 struct fib_result* res,
2122 const struct flowi *fl,
2123 const struct flowi *oldflp,
2124 struct net_device *dev_out,
2125 unsigned flags)
2126 {
2127 struct rtable *rth;
2128 struct in_device *in_dev;
2129 u32 tos = RT_FL_TOS(oldflp);
2130 int err = 0;
2131
2132 if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags&IFF_LOOPBACK))
2133 return -EINVAL;
2134
2135 if (fl->fl4_dst == htonl(0xFFFFFFFF))
2136 res->type = RTN_BROADCAST;
2137 else if (ipv4_is_multicast(fl->fl4_dst))
2138 res->type = RTN_MULTICAST;
2139 else if (ipv4_is_lbcast(fl->fl4_dst) || ipv4_is_zeronet(fl->fl4_dst))
2140 return -EINVAL;
2141
2142 if (dev_out->flags & IFF_LOOPBACK)
2143 flags |= RTCF_LOCAL;
2144
2145 /* get work reference to inet device */
2146 in_dev = in_dev_get(dev_out);
2147 if (!in_dev)
2148 return -EINVAL;
2149
2150 if (res->type == RTN_BROADCAST) {
2151 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2152 if (res->fi) {
2153 fib_info_put(res->fi);
2154 res->fi = NULL;
2155 }
2156 } else if (res->type == RTN_MULTICAST) {
2157 flags |= RTCF_MULTICAST|RTCF_LOCAL;
2158 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2159 oldflp->proto))
2160 flags &= ~RTCF_LOCAL;
2161 /* If multicast route do not exist use
2162 default one, but do not gateway in this case.
2163 Yes, it is hack.
2164 */
2165 if (res->fi && res->prefixlen < 4) {
2166 fib_info_put(res->fi);
2167 res->fi = NULL;
2168 }
2169 }
2170
2171
2172 rth = dst_alloc(&ipv4_dst_ops);
2173 if (!rth) {
2174 err = -ENOBUFS;
2175 goto cleanup;
2176 }
2177
2178 atomic_set(&rth->u.dst.__refcnt, 1);
2179 rth->u.dst.flags= DST_HOST;
2180 if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2181 rth->u.dst.flags |= DST_NOXFRM;
2182 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2183 rth->u.dst.flags |= DST_NOPOLICY;
2184
2185 rth->fl.fl4_dst = oldflp->fl4_dst;
2186 rth->fl.fl4_tos = tos;
2187 rth->fl.fl4_src = oldflp->fl4_src;
2188 rth->fl.oif = oldflp->oif;
2189 rth->fl.mark = oldflp->mark;
2190 rth->rt_dst = fl->fl4_dst;
2191 rth->rt_src = fl->fl4_src;
2192 rth->rt_iif = oldflp->oif ? : dev_out->ifindex;
2193 /* get references to the devices that are to be hold by the routing
2194 cache entry */
2195 rth->u.dst.dev = dev_out;
2196 dev_hold(dev_out);
2197 rth->idev = in_dev_get(dev_out);
2198 rth->rt_gateway = fl->fl4_dst;
2199 rth->rt_spec_dst= fl->fl4_src;
2200
2201 rth->u.dst.output=ip_output;
2202
2203 RT_CACHE_STAT_INC(out_slow_tot);
2204
2205 if (flags & RTCF_LOCAL) {
2206 rth->u.dst.input = ip_local_deliver;
2207 rth->rt_spec_dst = fl->fl4_dst;
2208 }
2209 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2210 rth->rt_spec_dst = fl->fl4_src;
2211 if (flags & RTCF_LOCAL &&
2212 !(dev_out->flags & IFF_LOOPBACK)) {
2213 rth->u.dst.output = ip_mc_output;
2214 RT_CACHE_STAT_INC(out_slow_mc);
2215 }
2216 #ifdef CONFIG_IP_MROUTE
2217 if (res->type == RTN_MULTICAST) {
2218 if (IN_DEV_MFORWARD(in_dev) &&
2219 !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2220 rth->u.dst.input = ip_mr_input;
2221 rth->u.dst.output = ip_mc_output;
2222 }
2223 }
2224 #endif
2225 }
2226
2227 rt_set_nexthop(rth, res, 0);
2228
2229 rth->rt_flags = flags;
2230
2231 *result = rth;
2232 cleanup:
2233 /* release work reference to inet device */
2234 in_dev_put(in_dev);
2235
2236 return err;
2237 }
2238
2239 static inline int ip_mkroute_output(struct rtable **rp,
2240 struct fib_result* res,
2241 const struct flowi *fl,
2242 const struct flowi *oldflp,
2243 struct net_device *dev_out,
2244 unsigned flags)
2245 {
2246 struct rtable *rth = NULL;
2247 int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2248 unsigned hash;
2249 if (err == 0) {
2250 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif);
2251 err = rt_intern_hash(hash, rth, rp);
2252 }
2253
2254 return err;
2255 }
2256
2257 /*
2258 * Major route resolver routine.
2259 */
2260
2261 static int ip_route_output_slow(struct net *net, struct rtable **rp,
2262 const struct flowi *oldflp)
2263 {
2264 u32 tos = RT_FL_TOS(oldflp);
2265 struct flowi fl = { .nl_u = { .ip4_u =
2266 { .daddr = oldflp->fl4_dst,
2267 .saddr = oldflp->fl4_src,
2268 .tos = tos & IPTOS_RT_MASK,
2269 .scope = ((tos & RTO_ONLINK) ?
2270 RT_SCOPE_LINK :
2271 RT_SCOPE_UNIVERSE),
2272 } },
2273 .mark = oldflp->mark,
2274 .iif = net->loopback_dev->ifindex,
2275 .oif = oldflp->oif };
2276 struct fib_result res;
2277 unsigned flags = 0;
2278 struct net_device *dev_out = NULL;
2279 int free_res = 0;
2280 int err;
2281
2282
2283 res.fi = NULL;
2284 #ifdef CONFIG_IP_MULTIPLE_TABLES
2285 res.r = NULL;
2286 #endif
2287
2288 if (oldflp->fl4_src) {
2289 err = -EINVAL;
2290 if (ipv4_is_multicast(oldflp->fl4_src) ||
2291 ipv4_is_lbcast(oldflp->fl4_src) ||
2292 ipv4_is_zeronet(oldflp->fl4_src))
2293 goto out;
2294
2295 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2296 dev_out = ip_dev_find(net, oldflp->fl4_src);
2297 if (dev_out == NULL)
2298 goto out;
2299
2300 /* I removed check for oif == dev_out->oif here.
2301 It was wrong for two reasons:
2302 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2303 is assigned to multiple interfaces.
2304 2. Moreover, we are allowed to send packets with saddr
2305 of another iface. --ANK
2306 */
2307
2308 if (oldflp->oif == 0
2309 && (ipv4_is_multicast(oldflp->fl4_dst) ||
2310 oldflp->fl4_dst == htonl(0xFFFFFFFF))) {
2311 /* Special hack: user can direct multicasts
2312 and limited broadcast via necessary interface
2313 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2314 This hack is not just for fun, it allows
2315 vic,vat and friends to work.
2316 They bind socket to loopback, set ttl to zero
2317 and expect that it will work.
2318 From the viewpoint of routing cache they are broken,
2319 because we are not allowed to build multicast path
2320 with loopback source addr (look, routing cache
2321 cannot know, that ttl is zero, so that packet
2322 will not leave this host and route is valid).
2323 Luckily, this hack is good workaround.
2324 */
2325
2326 fl.oif = dev_out->ifindex;
2327 goto make_route;
2328 }
2329 if (dev_out)
2330 dev_put(dev_out);
2331 dev_out = NULL;
2332 }
2333
2334
2335 if (oldflp->oif) {
2336 dev_out = dev_get_by_index(net, oldflp->oif);
2337 err = -ENODEV;
2338 if (dev_out == NULL)
2339 goto out;
2340
2341 /* RACE: Check return value of inet_select_addr instead. */
2342 if (__in_dev_get_rtnl(dev_out) == NULL) {
2343 dev_put(dev_out);
2344 goto out; /* Wrong error code */
2345 }
2346
2347 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2348 oldflp->fl4_dst == htonl(0xFFFFFFFF)) {
2349 if (!fl.fl4_src)
2350 fl.fl4_src = inet_select_addr(dev_out, 0,
2351 RT_SCOPE_LINK);
2352 goto make_route;
2353 }
2354 if (!fl.fl4_src) {
2355 if (ipv4_is_multicast(oldflp->fl4_dst))
2356 fl.fl4_src = inet_select_addr(dev_out, 0,
2357 fl.fl4_scope);
2358 else if (!oldflp->fl4_dst)
2359 fl.fl4_src = inet_select_addr(dev_out, 0,
2360 RT_SCOPE_HOST);
2361 }
2362 }
2363
2364 if (!fl.fl4_dst) {
2365 fl.fl4_dst = fl.fl4_src;
2366 if (!fl.fl4_dst)
2367 fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2368 if (dev_out)
2369 dev_put(dev_out);
2370 dev_out = net->loopback_dev;
2371 dev_hold(dev_out);
2372 fl.oif = net->loopback_dev->ifindex;
2373 res.type = RTN_LOCAL;
2374 flags |= RTCF_LOCAL;
2375 goto make_route;
2376 }
2377
2378 if (fib_lookup(net, &fl, &res)) {
2379 res.fi = NULL;
2380 if (oldflp->oif) {
2381 /* Apparently, routing tables are wrong. Assume,
2382 that the destination is on link.
2383
2384 WHY? DW.
2385 Because we are allowed to send to iface
2386 even if it has NO routes and NO assigned
2387 addresses. When oif is specified, routing
2388 tables are looked up with only one purpose:
2389 to catch if destination is gatewayed, rather than
2390 direct. Moreover, if MSG_DONTROUTE is set,
2391 we send packet, ignoring both routing tables
2392 and ifaddr state. --ANK
2393
2394
2395 We could make it even if oif is unknown,
2396 likely IPv6, but we do not.
2397 */
2398
2399 if (fl.fl4_src == 0)
2400 fl.fl4_src = inet_select_addr(dev_out, 0,
2401 RT_SCOPE_LINK);
2402 res.type = RTN_UNICAST;
2403 goto make_route;
2404 }
2405 if (dev_out)
2406 dev_put(dev_out);
2407 err = -ENETUNREACH;
2408 goto out;
2409 }
2410 free_res = 1;
2411
2412 if (res.type == RTN_LOCAL) {
2413 if (!fl.fl4_src)
2414 fl.fl4_src = fl.fl4_dst;
2415 if (dev_out)
2416 dev_put(dev_out);
2417 dev_out = net->loopback_dev;
2418 dev_hold(dev_out);
2419 fl.oif = dev_out->ifindex;
2420 if (res.fi)
2421 fib_info_put(res.fi);
2422 res.fi = NULL;
2423 flags |= RTCF_LOCAL;
2424 goto make_route;
2425 }
2426
2427 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2428 if (res.fi->fib_nhs > 1 && fl.oif == 0)
2429 fib_select_multipath(&fl, &res);
2430 else
2431 #endif
2432 if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2433 fib_select_default(net, &fl, &res);
2434
2435 if (!fl.fl4_src)
2436 fl.fl4_src = FIB_RES_PREFSRC(res);
2437
2438 if (dev_out)
2439 dev_put(dev_out);
2440 dev_out = FIB_RES_DEV(res);
2441 dev_hold(dev_out);
2442 fl.oif = dev_out->ifindex;
2443
2444
2445 make_route:
2446 err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2447
2448
2449 if (free_res)
2450 fib_res_put(&res);
2451 if (dev_out)
2452 dev_put(dev_out);
2453 out: return err;
2454 }
2455
2456 int __ip_route_output_key(struct net *net, struct rtable **rp,
2457 const struct flowi *flp)
2458 {
2459 unsigned hash;
2460 struct rtable *rth;
2461
2462 hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif);
2463
2464 rcu_read_lock_bh();
2465 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2466 rth = rcu_dereference(rth->u.dst.rt_next)) {
2467 if (rth->fl.fl4_dst == flp->fl4_dst &&
2468 rth->fl.fl4_src == flp->fl4_src &&
2469 rth->fl.iif == 0 &&
2470 rth->fl.oif == flp->oif &&
2471 rth->fl.mark == flp->mark &&
2472 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2473 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2474 rth->u.dst.dev->nd_net == net) {
2475 dst_use(&rth->u.dst, jiffies);
2476 RT_CACHE_STAT_INC(out_hit);
2477 rcu_read_unlock_bh();
2478 *rp = rth;
2479 return 0;
2480 }
2481 RT_CACHE_STAT_INC(out_hlist_search);
2482 }
2483 rcu_read_unlock_bh();
2484
2485 return ip_route_output_slow(net, rp, flp);
2486 }
2487
2488 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2489
2490 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2491 {
2492 }
2493
2494 static struct dst_ops ipv4_dst_blackhole_ops = {
2495 .family = AF_INET,
2496 .protocol = __constant_htons(ETH_P_IP),
2497 .destroy = ipv4_dst_destroy,
2498 .check = ipv4_dst_check,
2499 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2500 .entry_size = sizeof(struct rtable),
2501 };
2502
2503
2504 static int ipv4_dst_blackhole(struct rtable **rp, struct flowi *flp, struct sock *sk)
2505 {
2506 struct rtable *ort = *rp;
2507 struct rtable *rt = (struct rtable *)
2508 dst_alloc(&ipv4_dst_blackhole_ops);
2509
2510 if (rt) {
2511 struct dst_entry *new = &rt->u.dst;
2512
2513 atomic_set(&new->__refcnt, 1);
2514 new->__use = 1;
2515 new->input = dst_discard;
2516 new->output = dst_discard;
2517 memcpy(new->metrics, ort->u.dst.metrics, RTAX_MAX*sizeof(u32));
2518
2519 new->dev = ort->u.dst.dev;
2520 if (new->dev)
2521 dev_hold(new->dev);
2522
2523 rt->fl = ort->fl;
2524
2525 rt->idev = ort->idev;
2526 if (rt->idev)
2527 in_dev_hold(rt->idev);
2528 rt->rt_flags = ort->rt_flags;
2529 rt->rt_type = ort->rt_type;
2530 rt->rt_dst = ort->rt_dst;
2531 rt->rt_src = ort->rt_src;
2532 rt->rt_iif = ort->rt_iif;
2533 rt->rt_gateway = ort->rt_gateway;
2534 rt->rt_spec_dst = ort->rt_spec_dst;
2535 rt->peer = ort->peer;
2536 if (rt->peer)
2537 atomic_inc(&rt->peer->refcnt);
2538
2539 dst_free(new);
2540 }
2541
2542 dst_release(&(*rp)->u.dst);
2543 *rp = rt;
2544 return (rt ? 0 : -ENOMEM);
2545 }
2546
2547 int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2548 struct sock *sk, int flags)
2549 {
2550 int err;
2551
2552 if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2553 return err;
2554
2555 if (flp->proto) {
2556 if (!flp->fl4_src)
2557 flp->fl4_src = (*rp)->rt_src;
2558 if (!flp->fl4_dst)
2559 flp->fl4_dst = (*rp)->rt_dst;
2560 err = __xfrm_lookup((struct dst_entry **)rp, flp, sk,
2561 flags ? XFRM_LOOKUP_WAIT : 0);
2562 if (err == -EREMOTE)
2563 err = ipv4_dst_blackhole(rp, flp, sk);
2564
2565 return err;
2566 }
2567
2568 return 0;
2569 }
2570
2571 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2572
2573 int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2574 {
2575 return ip_route_output_flow(net, rp, flp, NULL, 0);
2576 }
2577
2578 static int rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
2579 int nowait, unsigned int flags)
2580 {
2581 struct rtable *rt = (struct rtable*)skb->dst;
2582 struct rtmsg *r;
2583 struct nlmsghdr *nlh;
2584 long expires;
2585 u32 id = 0, ts = 0, tsage = 0, error;
2586
2587 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2588 if (nlh == NULL)
2589 return -EMSGSIZE;
2590
2591 r = nlmsg_data(nlh);
2592 r->rtm_family = AF_INET;
2593 r->rtm_dst_len = 32;
2594 r->rtm_src_len = 0;
2595 r->rtm_tos = rt->fl.fl4_tos;
2596 r->rtm_table = RT_TABLE_MAIN;
2597 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2598 r->rtm_type = rt->rt_type;
2599 r->rtm_scope = RT_SCOPE_UNIVERSE;
2600 r->rtm_protocol = RTPROT_UNSPEC;
2601 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2602 if (rt->rt_flags & RTCF_NOTIFY)
2603 r->rtm_flags |= RTM_F_NOTIFY;
2604
2605 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2606
2607 if (rt->fl.fl4_src) {
2608 r->rtm_src_len = 32;
2609 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2610 }
2611 if (rt->u.dst.dev)
2612 NLA_PUT_U32(skb, RTA_OIF, rt->u.dst.dev->ifindex);
2613 #ifdef CONFIG_NET_CLS_ROUTE
2614 if (rt->u.dst.tclassid)
2615 NLA_PUT_U32(skb, RTA_FLOW, rt->u.dst.tclassid);
2616 #endif
2617 if (rt->fl.iif)
2618 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2619 else if (rt->rt_src != rt->fl.fl4_src)
2620 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2621
2622 if (rt->rt_dst != rt->rt_gateway)
2623 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2624
2625 if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
2626 goto nla_put_failure;
2627
2628 error = rt->u.dst.error;
2629 expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
2630 if (rt->peer) {
2631 id = rt->peer->ip_id_count;
2632 if (rt->peer->tcp_ts_stamp) {
2633 ts = rt->peer->tcp_ts;
2634 tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2635 }
2636 }
2637
2638 if (rt->fl.iif) {
2639 #ifdef CONFIG_IP_MROUTE
2640 __be32 dst = rt->rt_dst;
2641
2642 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2643 IPV4_DEVCONF_ALL(&init_net, MC_FORWARDING)) {
2644 int err = ipmr_get_route(skb, r, nowait);
2645 if (err <= 0) {
2646 if (!nowait) {
2647 if (err == 0)
2648 return 0;
2649 goto nla_put_failure;
2650 } else {
2651 if (err == -EMSGSIZE)
2652 goto nla_put_failure;
2653 error = err;
2654 }
2655 }
2656 } else
2657 #endif
2658 NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2659 }
2660
2661 if (rtnl_put_cacheinfo(skb, &rt->u.dst, id, ts, tsage,
2662 expires, error) < 0)
2663 goto nla_put_failure;
2664
2665 return nlmsg_end(skb, nlh);
2666
2667 nla_put_failure:
2668 nlmsg_cancel(skb, nlh);
2669 return -EMSGSIZE;
2670 }
2671
2672 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2673 {
2674 struct net *net = in_skb->sk->sk_net;
2675 struct rtmsg *rtm;
2676 struct nlattr *tb[RTA_MAX+1];
2677 struct rtable *rt = NULL;
2678 __be32 dst = 0;
2679 __be32 src = 0;
2680 u32 iif;
2681 int err;
2682 struct sk_buff *skb;
2683
2684 if (net != &init_net)
2685 return -EINVAL;
2686
2687 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2688 if (err < 0)
2689 goto errout;
2690
2691 rtm = nlmsg_data(nlh);
2692
2693 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2694 if (skb == NULL) {
2695 err = -ENOBUFS;
2696 goto errout;
2697 }
2698
2699 /* Reserve room for dummy headers, this skb can pass
2700 through good chunk of routing engine.
2701 */
2702 skb_reset_mac_header(skb);
2703 skb_reset_network_header(skb);
2704
2705 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2706 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2707 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2708
2709 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2710 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2711 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2712
2713 if (iif) {
2714 struct net_device *dev;
2715
2716 dev = __dev_get_by_index(&init_net, iif);
2717 if (dev == NULL) {
2718 err = -ENODEV;
2719 goto errout_free;
2720 }
2721
2722 skb->protocol = htons(ETH_P_IP);
2723 skb->dev = dev;
2724 local_bh_disable();
2725 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2726 local_bh_enable();
2727
2728 rt = (struct rtable*) skb->dst;
2729 if (err == 0 && rt->u.dst.error)
2730 err = -rt->u.dst.error;
2731 } else {
2732 struct flowi fl = {
2733 .nl_u = {
2734 .ip4_u = {
2735 .daddr = dst,
2736 .saddr = src,
2737 .tos = rtm->rtm_tos,
2738 },
2739 },
2740 .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2741 };
2742 err = ip_route_output_key(&init_net, &rt, &fl);
2743 }
2744
2745 if (err)
2746 goto errout_free;
2747
2748 skb->dst = &rt->u.dst;
2749 if (rtm->rtm_flags & RTM_F_NOTIFY)
2750 rt->rt_flags |= RTCF_NOTIFY;
2751
2752 err = rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2753 RTM_NEWROUTE, 0, 0);
2754 if (err <= 0)
2755 goto errout_free;
2756
2757 err = rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid);
2758 errout:
2759 return err;
2760
2761 errout_free:
2762 kfree_skb(skb);
2763 goto errout;
2764 }
2765
2766 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2767 {
2768 struct rtable *rt;
2769 int h, s_h;
2770 int idx, s_idx;
2771
2772 s_h = cb->args[0];
2773 if (s_h < 0)
2774 s_h = 0;
2775 s_idx = idx = cb->args[1];
2776 for (h = s_h; h <= rt_hash_mask; h++) {
2777 rcu_read_lock_bh();
2778 for (rt = rcu_dereference(rt_hash_table[h].chain), idx = 0; rt;
2779 rt = rcu_dereference(rt->u.dst.rt_next), idx++) {
2780 if (idx < s_idx)
2781 continue;
2782 skb->dst = dst_clone(&rt->u.dst);
2783 if (rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
2784 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2785 1, NLM_F_MULTI) <= 0) {
2786 dst_release(xchg(&skb->dst, NULL));
2787 rcu_read_unlock_bh();
2788 goto done;
2789 }
2790 dst_release(xchg(&skb->dst, NULL));
2791 }
2792 rcu_read_unlock_bh();
2793 s_idx = 0;
2794 }
2795
2796 done:
2797 cb->args[0] = h;
2798 cb->args[1] = idx;
2799 return skb->len;
2800 }
2801
2802 void ip_rt_multicast_event(struct in_device *in_dev)
2803 {
2804 rt_cache_flush(0);
2805 }
2806
2807 #ifdef CONFIG_SYSCTL
2808 static int flush_delay;
2809
2810 static int ipv4_sysctl_rtcache_flush(ctl_table *ctl, int write,
2811 struct file *filp, void __user *buffer,
2812 size_t *lenp, loff_t *ppos)
2813 {
2814 if (write) {
2815 proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
2816 rt_cache_flush(flush_delay);
2817 return 0;
2818 }
2819
2820 return -EINVAL;
2821 }
2822
2823 static int ipv4_sysctl_rtcache_flush_strategy(ctl_table *table,
2824 int __user *name,
2825 int nlen,
2826 void __user *oldval,
2827 size_t __user *oldlenp,
2828 void __user *newval,
2829 size_t newlen)
2830 {
2831 int delay;
2832 if (newlen != sizeof(int))
2833 return -EINVAL;
2834 if (get_user(delay, (int __user *)newval))
2835 return -EFAULT;
2836 rt_cache_flush(delay);
2837 return 0;
2838 }
2839
2840 ctl_table ipv4_route_table[] = {
2841 {
2842 .ctl_name = NET_IPV4_ROUTE_FLUSH,
2843 .procname = "flush",
2844 .data = &flush_delay,
2845 .maxlen = sizeof(int),
2846 .mode = 0200,
2847 .proc_handler = &ipv4_sysctl_rtcache_flush,
2848 .strategy = &ipv4_sysctl_rtcache_flush_strategy,
2849 },
2850 {
2851 .ctl_name = NET_IPV4_ROUTE_MIN_DELAY,
2852 .procname = "min_delay",
2853 .data = &ip_rt_min_delay,
2854 .maxlen = sizeof(int),
2855 .mode = 0644,
2856 .proc_handler = &proc_dointvec_jiffies,
2857 .strategy = &sysctl_jiffies,
2858 },
2859 {
2860 .ctl_name = NET_IPV4_ROUTE_MAX_DELAY,
2861 .procname = "max_delay",
2862 .data = &ip_rt_max_delay,
2863 .maxlen = sizeof(int),
2864 .mode = 0644,
2865 .proc_handler = &proc_dointvec_jiffies,
2866 .strategy = &sysctl_jiffies,
2867 },
2868 {
2869 .ctl_name = NET_IPV4_ROUTE_GC_THRESH,
2870 .procname = "gc_thresh",
2871 .data = &ipv4_dst_ops.gc_thresh,
2872 .maxlen = sizeof(int),
2873 .mode = 0644,
2874 .proc_handler = &proc_dointvec,
2875 },
2876 {
2877 .ctl_name = NET_IPV4_ROUTE_MAX_SIZE,
2878 .procname = "max_size",
2879 .data = &ip_rt_max_size,
2880 .maxlen = sizeof(int),
2881 .mode = 0644,
2882 .proc_handler = &proc_dointvec,
2883 },
2884 {
2885 /* Deprecated. Use gc_min_interval_ms */
2886
2887 .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL,
2888 .procname = "gc_min_interval",
2889 .data = &ip_rt_gc_min_interval,
2890 .maxlen = sizeof(int),
2891 .mode = 0644,
2892 .proc_handler = &proc_dointvec_jiffies,
2893 .strategy = &sysctl_jiffies,
2894 },
2895 {
2896 .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL_MS,
2897 .procname = "gc_min_interval_ms",
2898 .data = &ip_rt_gc_min_interval,
2899 .maxlen = sizeof(int),
2900 .mode = 0644,
2901 .proc_handler = &proc_dointvec_ms_jiffies,
2902 .strategy = &sysctl_ms_jiffies,
2903 },
2904 {
2905 .ctl_name = NET_IPV4_ROUTE_GC_TIMEOUT,
2906 .procname = "gc_timeout",
2907 .data = &ip_rt_gc_timeout,
2908 .maxlen = sizeof(int),
2909 .mode = 0644,
2910 .proc_handler = &proc_dointvec_jiffies,
2911 .strategy = &sysctl_jiffies,
2912 },
2913 {
2914 .ctl_name = NET_IPV4_ROUTE_GC_INTERVAL,
2915 .procname = "gc_interval",
2916 .data = &ip_rt_gc_interval,
2917 .maxlen = sizeof(int),
2918 .mode = 0644,
2919 .proc_handler = &proc_dointvec_jiffies,
2920 .strategy = &sysctl_jiffies,
2921 },
2922 {
2923 .ctl_name = NET_IPV4_ROUTE_REDIRECT_LOAD,
2924 .procname = "redirect_load",
2925 .data = &ip_rt_redirect_load,
2926 .maxlen = sizeof(int),
2927 .mode = 0644,
2928 .proc_handler = &proc_dointvec,
2929 },
2930 {
2931 .ctl_name = NET_IPV4_ROUTE_REDIRECT_NUMBER,
2932 .procname = "redirect_number",
2933 .data = &ip_rt_redirect_number,
2934 .maxlen = sizeof(int),
2935 .mode = 0644,
2936 .proc_handler = &proc_dointvec,
2937 },
2938 {
2939 .ctl_name = NET_IPV4_ROUTE_REDIRECT_SILENCE,
2940 .procname = "redirect_silence",
2941 .data = &ip_rt_redirect_silence,
2942 .maxlen = sizeof(int),
2943 .mode = 0644,
2944 .proc_handler = &proc_dointvec,
2945 },
2946 {
2947 .ctl_name = NET_IPV4_ROUTE_ERROR_COST,
2948 .procname = "error_cost",
2949 .data = &ip_rt_error_cost,
2950 .maxlen = sizeof(int),
2951 .mode = 0644,
2952 .proc_handler = &proc_dointvec,
2953 },
2954 {
2955 .ctl_name = NET_IPV4_ROUTE_ERROR_BURST,
2956 .procname = "error_burst",
2957 .data = &ip_rt_error_burst,
2958 .maxlen = sizeof(int),
2959 .mode = 0644,
2960 .proc_handler = &proc_dointvec,
2961 },
2962 {
2963 .ctl_name = NET_IPV4_ROUTE_GC_ELASTICITY,
2964 .procname = "gc_elasticity",
2965 .data = &ip_rt_gc_elasticity,
2966 .maxlen = sizeof(int),
2967 .mode = 0644,
2968 .proc_handler = &proc_dointvec,
2969 },
2970 {
2971 .ctl_name = NET_IPV4_ROUTE_MTU_EXPIRES,
2972 .procname = "mtu_expires",
2973 .data = &ip_rt_mtu_expires,
2974 .maxlen = sizeof(int),
2975 .mode = 0644,
2976 .proc_handler = &proc_dointvec_jiffies,
2977 .strategy = &sysctl_jiffies,
2978 },
2979 {
2980 .ctl_name = NET_IPV4_ROUTE_MIN_PMTU,
2981 .procname = "min_pmtu",
2982 .data = &ip_rt_min_pmtu,
2983 .maxlen = sizeof(int),
2984 .mode = 0644,
2985 .proc_handler = &proc_dointvec,
2986 },
2987 {
2988 .ctl_name = NET_IPV4_ROUTE_MIN_ADVMSS,
2989 .procname = "min_adv_mss",
2990 .data = &ip_rt_min_advmss,
2991 .maxlen = sizeof(int),
2992 .mode = 0644,
2993 .proc_handler = &proc_dointvec,
2994 },
2995 {
2996 .ctl_name = NET_IPV4_ROUTE_SECRET_INTERVAL,
2997 .procname = "secret_interval",
2998 .data = &ip_rt_secret_interval,
2999 .maxlen = sizeof(int),
3000 .mode = 0644,
3001 .proc_handler = &proc_dointvec_jiffies,
3002 .strategy = &sysctl_jiffies,
3003 },
3004 { .ctl_name = 0 }
3005 };
3006 #endif
3007
3008 #ifdef CONFIG_NET_CLS_ROUTE
3009 struct ip_rt_acct *ip_rt_acct __read_mostly;
3010 #endif /* CONFIG_NET_CLS_ROUTE */
3011
3012 static __initdata unsigned long rhash_entries;
3013 static int __init set_rhash_entries(char *str)
3014 {
3015 if (!str)
3016 return 0;
3017 rhash_entries = simple_strtoul(str, &str, 0);
3018 return 1;
3019 }
3020 __setup("rhash_entries=", set_rhash_entries);
3021
3022 int __init ip_rt_init(void)
3023 {
3024 int rc = 0;
3025
3026 rt_hash_rnd = (int) ((num_physpages ^ (num_physpages>>8)) ^
3027 (jiffies ^ (jiffies >> 7)));
3028
3029 #ifdef CONFIG_NET_CLS_ROUTE
3030 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct));
3031 if (!ip_rt_acct)
3032 panic("IP: failed to allocate ip_rt_acct\n");
3033 #endif
3034
3035 ipv4_dst_ops.kmem_cachep =
3036 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3037 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3038
3039 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3040
3041 rt_hash_table = (struct rt_hash_bucket *)
3042 alloc_large_system_hash("IP route cache",
3043 sizeof(struct rt_hash_bucket),
3044 rhash_entries,
3045 (num_physpages >= 128 * 1024) ?
3046 15 : 17,
3047 0,
3048 &rt_hash_log,
3049 &rt_hash_mask,
3050 0);
3051 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3052 rt_hash_lock_init();
3053
3054 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3055 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3056
3057 devinet_init();
3058 ip_fib_init();
3059
3060 setup_timer(&rt_flush_timer, rt_run_flush, 0);
3061 setup_timer(&rt_secret_timer, rt_secret_rebuild, 0);
3062
3063 /* All the timers, started at system startup tend
3064 to synchronize. Perturb it a bit.
3065 */
3066 schedule_delayed_work(&expires_work,
3067 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3068
3069 rt_secret_timer.expires = jiffies + net_random() % ip_rt_secret_interval +
3070 ip_rt_secret_interval;
3071 add_timer(&rt_secret_timer);
3072
3073 if (ip_rt_proc_init(&init_net))
3074 printk(KERN_ERR "Unable to create route proc files\n");
3075 #ifdef CONFIG_XFRM
3076 xfrm_init();
3077 xfrm4_init();
3078 #endif
3079 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3080
3081 return rc;
3082 }
3083
3084 EXPORT_SYMBOL(__ip_select_ident);
3085 EXPORT_SYMBOL(ip_route_input);
3086 EXPORT_SYMBOL(ip_route_output_key);
linux kernel for loongson