blob: 60b11aece5c3971a321411b9247723b367ea4538 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 *
Jesper Juhl02c30a82005-05-05 16:16:16 -070010 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * 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
Eric Dumazetbb1d23b2005-07-05 15:00:32 -070057 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
Ilia Sotnikovcef26852006-03-25 01:38:55 -080058 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
59 * Ilia Sotnikov : Removed TOS from hash calculations
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 *
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/config.h>
68#include <linux/module.h>
69#include <asm/uaccess.h>
70#include <asm/system.h>
71#include <linux/bitops.h>
72#include <linux/types.h>
73#include <linux/kernel.h>
74#include <linux/sched.h>
75#include <linux/mm.h>
Eric Dumazet424c4b72005-07-05 14:58:19 -070076#include <linux/bootmem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070077#include <linux/string.h>
78#include <linux/socket.h>
79#include <linux/sockios.h>
80#include <linux/errno.h>
81#include <linux/in.h>
82#include <linux/inet.h>
83#include <linux/netdevice.h>
84#include <linux/proc_fs.h>
85#include <linux/init.h>
86#include <linux/skbuff.h>
87#include <linux/rtnetlink.h>
88#include <linux/inetdevice.h>
89#include <linux/igmp.h>
90#include <linux/pkt_sched.h>
91#include <linux/mroute.h>
92#include <linux/netfilter_ipv4.h>
93#include <linux/random.h>
94#include <linux/jhash.h>
95#include <linux/rcupdate.h>
96#include <linux/times.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/ip_mp_alg.h>
108#ifdef CONFIG_SYSCTL
109#include <linux/sysctl.h>
110#endif
111
112#define RT_FL_TOS(oldflp) \
113 ((u32)(oldflp->fl4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
114
115#define IP_MAX_MTU 0xFFF0
116
117#define RT_GC_TIMEOUT (300*HZ)
118
119static int ip_rt_min_delay = 2 * HZ;
120static int ip_rt_max_delay = 10 * HZ;
121static int ip_rt_max_size;
122static int ip_rt_gc_timeout = RT_GC_TIMEOUT;
123static int ip_rt_gc_interval = 60 * HZ;
124static int ip_rt_gc_min_interval = HZ / 2;
125static int ip_rt_redirect_number = 9;
126static int ip_rt_redirect_load = HZ / 50;
127static int ip_rt_redirect_silence = ((HZ / 50) << (9 + 1));
128static int ip_rt_error_cost = HZ;
129static int ip_rt_error_burst = 5 * HZ;
130static int ip_rt_gc_elasticity = 8;
131static int ip_rt_mtu_expires = 10 * 60 * HZ;
132static int ip_rt_min_pmtu = 512 + 20 + 20;
133static int ip_rt_min_advmss = 256;
134static int ip_rt_secret_interval = 10 * 60 * HZ;
135static unsigned long rt_deadline;
136
137#define RTprint(a...) printk(KERN_DEBUG a)
138
139static struct timer_list rt_flush_timer;
140static struct timer_list rt_periodic_timer;
141static struct timer_list rt_secret_timer;
142
143/*
144 * Interface to generic destination cache.
145 */
146
147static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
148static void ipv4_dst_destroy(struct dst_entry *dst);
149static void ipv4_dst_ifdown(struct dst_entry *dst,
150 struct net_device *dev, int how);
151static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
152static void ipv4_link_failure(struct sk_buff *skb);
153static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
154static int rt_garbage_collect(void);
155
156
157static struct dst_ops ipv4_dst_ops = {
158 .family = AF_INET,
159 .protocol = __constant_htons(ETH_P_IP),
160 .gc = rt_garbage_collect,
161 .check = ipv4_dst_check,
162 .destroy = ipv4_dst_destroy,
163 .ifdown = ipv4_dst_ifdown,
164 .negative_advice = ipv4_negative_advice,
165 .link_failure = ipv4_link_failure,
166 .update_pmtu = ip_rt_update_pmtu,
167 .entry_size = sizeof(struct rtable),
168};
169
170#define ECN_OR_COST(class) TC_PRIO_##class
171
172__u8 ip_tos2prio[16] = {
173 TC_PRIO_BESTEFFORT,
174 ECN_OR_COST(FILLER),
175 TC_PRIO_BESTEFFORT,
176 ECN_OR_COST(BESTEFFORT),
177 TC_PRIO_BULK,
178 ECN_OR_COST(BULK),
179 TC_PRIO_BULK,
180 ECN_OR_COST(BULK),
181 TC_PRIO_INTERACTIVE,
182 ECN_OR_COST(INTERACTIVE),
183 TC_PRIO_INTERACTIVE,
184 ECN_OR_COST(INTERACTIVE),
185 TC_PRIO_INTERACTIVE_BULK,
186 ECN_OR_COST(INTERACTIVE_BULK),
187 TC_PRIO_INTERACTIVE_BULK,
188 ECN_OR_COST(INTERACTIVE_BULK)
189};
190
191
192/*
193 * Route cache.
194 */
195
196/* The locking scheme is rather straight forward:
197 *
198 * 1) Read-Copy Update protects the buckets of the central route hash.
199 * 2) Only writers remove entries, and they hold the lock
200 * as they look at rtable reference counts.
201 * 3) Only readers acquire references to rtable entries,
202 * they do so with atomic increments and with the
203 * lock held.
204 */
205
206struct rt_hash_bucket {
207 struct rtable *chain;
Eric Dumazet22c047c2005-07-05 14:55:24 -0700208};
209#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
210/*
211 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
212 * The size of this table is a power of two and depends on the number of CPUS.
213 */
214#if NR_CPUS >= 32
215#define RT_HASH_LOCK_SZ 4096
216#elif NR_CPUS >= 16
217#define RT_HASH_LOCK_SZ 2048
218#elif NR_CPUS >= 8
219#define RT_HASH_LOCK_SZ 1024
220#elif NR_CPUS >= 4
221#define RT_HASH_LOCK_SZ 512
222#else
223#define RT_HASH_LOCK_SZ 256
224#endif
225
226static spinlock_t *rt_hash_locks;
227# define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
228# define rt_hash_lock_init() { \
229 int i; \
230 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ, GFP_KERNEL); \
231 if (!rt_hash_locks) panic("IP: failed to allocate rt_hash_locks\n"); \
232 for (i = 0; i < RT_HASH_LOCK_SZ; i++) \
233 spin_lock_init(&rt_hash_locks[i]); \
234 }
235#else
236# define rt_hash_lock_addr(slot) NULL
237# define rt_hash_lock_init()
238#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
240static struct rt_hash_bucket *rt_hash_table;
241static unsigned rt_hash_mask;
242static int rt_hash_log;
243static unsigned int rt_hash_rnd;
244
Eric Dumazet2f970d82006-01-17 02:54:36 -0800245static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
Andrew Mortondbd29152006-01-17 21:58:01 -0800246#define RT_CACHE_STAT_INC(field) \
Paul Mackerrasbfe5d832006-06-25 05:47:14 -0700247 (__raw_get_cpu_var(rt_cache_stat).field++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
249static int rt_intern_hash(unsigned hash, struct rtable *rth,
250 struct rtable **res);
251
Ilia Sotnikovcef26852006-03-25 01:38:55 -0800252static unsigned int rt_hash_code(u32 daddr, u32 saddr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253{
Ilia Sotnikovcef26852006-03-25 01:38:55 -0800254 return (jhash_2words(daddr, saddr, rt_hash_rnd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 & rt_hash_mask);
256}
257
258#ifdef CONFIG_PROC_FS
259struct rt_cache_iter_state {
260 int bucket;
261};
262
263static struct rtable *rt_cache_get_first(struct seq_file *seq)
264{
265 struct rtable *r = NULL;
266 struct rt_cache_iter_state *st = seq->private;
267
268 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
269 rcu_read_lock_bh();
270 r = rt_hash_table[st->bucket].chain;
271 if (r)
272 break;
273 rcu_read_unlock_bh();
274 }
275 return r;
276}
277
278static struct rtable *rt_cache_get_next(struct seq_file *seq, struct rtable *r)
279{
280 struct rt_cache_iter_state *st = rcu_dereference(seq->private);
281
282 r = r->u.rt_next;
283 while (!r) {
284 rcu_read_unlock_bh();
285 if (--st->bucket < 0)
286 break;
287 rcu_read_lock_bh();
288 r = rt_hash_table[st->bucket].chain;
289 }
290 return r;
291}
292
293static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
294{
295 struct rtable *r = rt_cache_get_first(seq);
296
297 if (r)
298 while (pos && (r = rt_cache_get_next(seq, r)))
299 --pos;
300 return pos ? NULL : r;
301}
302
303static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
304{
305 return *pos ? rt_cache_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
306}
307
308static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
309{
310 struct rtable *r = NULL;
311
312 if (v == SEQ_START_TOKEN)
313 r = rt_cache_get_first(seq);
314 else
315 r = rt_cache_get_next(seq, v);
316 ++*pos;
317 return r;
318}
319
320static void rt_cache_seq_stop(struct seq_file *seq, void *v)
321{
322 if (v && v != SEQ_START_TOKEN)
323 rcu_read_unlock_bh();
324}
325
326static int rt_cache_seq_show(struct seq_file *seq, void *v)
327{
328 if (v == SEQ_START_TOKEN)
329 seq_printf(seq, "%-127s\n",
330 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
331 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
332 "HHUptod\tSpecDst");
333 else {
334 struct rtable *r = v;
335 char temp[256];
336
337 sprintf(temp, "%s\t%08lX\t%08lX\t%8X\t%d\t%u\t%d\t"
338 "%08lX\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X",
339 r->u.dst.dev ? r->u.dst.dev->name : "*",
340 (unsigned long)r->rt_dst, (unsigned long)r->rt_gateway,
341 r->rt_flags, atomic_read(&r->u.dst.__refcnt),
342 r->u.dst.__use, 0, (unsigned long)r->rt_src,
343 (dst_metric(&r->u.dst, RTAX_ADVMSS) ?
344 (int)dst_metric(&r->u.dst, RTAX_ADVMSS) + 40 : 0),
345 dst_metric(&r->u.dst, RTAX_WINDOW),
346 (int)((dst_metric(&r->u.dst, RTAX_RTT) >> 3) +
347 dst_metric(&r->u.dst, RTAX_RTTVAR)),
348 r->fl.fl4_tos,
349 r->u.dst.hh ? atomic_read(&r->u.dst.hh->hh_refcnt) : -1,
350 r->u.dst.hh ? (r->u.dst.hh->hh_output ==
351 dev_queue_xmit) : 0,
352 r->rt_spec_dst);
353 seq_printf(seq, "%-127s\n", temp);
354 }
355 return 0;
356}
357
358static struct seq_operations rt_cache_seq_ops = {
359 .start = rt_cache_seq_start,
360 .next = rt_cache_seq_next,
361 .stop = rt_cache_seq_stop,
362 .show = rt_cache_seq_show,
363};
364
365static int rt_cache_seq_open(struct inode *inode, struct file *file)
366{
367 struct seq_file *seq;
368 int rc = -ENOMEM;
369 struct rt_cache_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
370
371 if (!s)
372 goto out;
373 rc = seq_open(file, &rt_cache_seq_ops);
374 if (rc)
375 goto out_kfree;
376 seq = file->private_data;
377 seq->private = s;
378 memset(s, 0, sizeof(*s));
379out:
380 return rc;
381out_kfree:
382 kfree(s);
383 goto out;
384}
385
386static struct file_operations rt_cache_seq_fops = {
387 .owner = THIS_MODULE,
388 .open = rt_cache_seq_open,
389 .read = seq_read,
390 .llseek = seq_lseek,
391 .release = seq_release_private,
392};
393
394
395static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
396{
397 int cpu;
398
399 if (*pos == 0)
400 return SEQ_START_TOKEN;
401
402 for (cpu = *pos-1; cpu < NR_CPUS; ++cpu) {
403 if (!cpu_possible(cpu))
404 continue;
405 *pos = cpu+1;
Eric Dumazet2f970d82006-01-17 02:54:36 -0800406 return &per_cpu(rt_cache_stat, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 }
408 return NULL;
409}
410
411static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
412{
413 int cpu;
414
415 for (cpu = *pos; cpu < NR_CPUS; ++cpu) {
416 if (!cpu_possible(cpu))
417 continue;
418 *pos = cpu+1;
Eric Dumazet2f970d82006-01-17 02:54:36 -0800419 return &per_cpu(rt_cache_stat, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 }
421 return NULL;
422
423}
424
425static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
426{
427
428}
429
430static int rt_cpu_seq_show(struct seq_file *seq, void *v)
431{
432 struct rt_cache_stat *st = v;
433
434 if (v == SEQ_START_TOKEN) {
Olaf Rempel5bec0032005-04-28 12:16:08 -0700435 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");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 return 0;
437 }
438
439 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
440 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
441 atomic_read(&ipv4_dst_ops.entries),
442 st->in_hit,
443 st->in_slow_tot,
444 st->in_slow_mc,
445 st->in_no_route,
446 st->in_brd,
447 st->in_martian_dst,
448 st->in_martian_src,
449
450 st->out_hit,
451 st->out_slow_tot,
452 st->out_slow_mc,
453
454 st->gc_total,
455 st->gc_ignored,
456 st->gc_goal_miss,
457 st->gc_dst_overflow,
458 st->in_hlist_search,
459 st->out_hlist_search
460 );
461 return 0;
462}
463
464static struct seq_operations rt_cpu_seq_ops = {
465 .start = rt_cpu_seq_start,
466 .next = rt_cpu_seq_next,
467 .stop = rt_cpu_seq_stop,
468 .show = rt_cpu_seq_show,
469};
470
471
472static int rt_cpu_seq_open(struct inode *inode, struct file *file)
473{
474 return seq_open(file, &rt_cpu_seq_ops);
475}
476
477static struct file_operations rt_cpu_seq_fops = {
478 .owner = THIS_MODULE,
479 .open = rt_cpu_seq_open,
480 .read = seq_read,
481 .llseek = seq_lseek,
482 .release = seq_release,
483};
484
485#endif /* CONFIG_PROC_FS */
486
487static __inline__ void rt_free(struct rtable *rt)
488{
489 multipath_remove(rt);
490 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
491}
492
493static __inline__ void rt_drop(struct rtable *rt)
494{
495 multipath_remove(rt);
496 ip_rt_put(rt);
497 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
498}
499
500static __inline__ int rt_fast_clean(struct rtable *rth)
501{
502 /* Kill broadcast/multicast entries very aggresively, if they
503 collide in hash table with more useful entries */
504 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
505 rth->fl.iif && rth->u.rt_next;
506}
507
508static __inline__ int rt_valuable(struct rtable *rth)
509{
510 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
511 rth->u.dst.expires;
512}
513
514static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
515{
516 unsigned long age;
517 int ret = 0;
518
519 if (atomic_read(&rth->u.dst.__refcnt))
520 goto out;
521
522 ret = 1;
523 if (rth->u.dst.expires &&
524 time_after_eq(jiffies, rth->u.dst.expires))
525 goto out;
526
527 age = jiffies - rth->u.dst.lastuse;
528 ret = 0;
529 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
530 (age <= tmo2 && rt_valuable(rth)))
531 goto out;
532 ret = 1;
533out: return ret;
534}
535
536/* Bits of score are:
537 * 31: very valuable
538 * 30: not quite useless
539 * 29..0: usage counter
540 */
541static inline u32 rt_score(struct rtable *rt)
542{
543 u32 score = jiffies - rt->u.dst.lastuse;
544
545 score = ~score & ~(3<<30);
546
547 if (rt_valuable(rt))
548 score |= (1<<31);
549
550 if (!rt->fl.iif ||
551 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
552 score |= (1<<30);
553
554 return score;
555}
556
557static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
558{
559 return memcmp(&fl1->nl_u.ip4_u, &fl2->nl_u.ip4_u, sizeof(fl1->nl_u.ip4_u)) == 0 &&
560 fl1->oif == fl2->oif &&
561 fl1->iif == fl2->iif;
562}
563
564#ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
565static struct rtable **rt_remove_balanced_route(struct rtable **chain_head,
566 struct rtable *expentry,
567 int *removed_count)
568{
569 int passedexpired = 0;
570 struct rtable **nextstep = NULL;
571 struct rtable **rthp = chain_head;
572 struct rtable *rth;
573
574 if (removed_count)
575 *removed_count = 0;
576
577 while ((rth = *rthp) != NULL) {
578 if (rth == expentry)
579 passedexpired = 1;
580
581 if (((*rthp)->u.dst.flags & DST_BALANCED) != 0 &&
582 compare_keys(&(*rthp)->fl, &expentry->fl)) {
583 if (*rthp == expentry) {
584 *rthp = rth->u.rt_next;
585 continue;
586 } else {
587 *rthp = rth->u.rt_next;
588 rt_free(rth);
589 if (removed_count)
590 ++(*removed_count);
591 }
592 } else {
593 if (!((*rthp)->u.dst.flags & DST_BALANCED) &&
594 passedexpired && !nextstep)
595 nextstep = &rth->u.rt_next;
596
597 rthp = &rth->u.rt_next;
598 }
599 }
600
601 rt_free(expentry);
602 if (removed_count)
603 ++(*removed_count);
604
605 return nextstep;
606}
607#endif /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
608
609
610/* This runs via a timer and thus is always in BH context. */
611static void rt_check_expire(unsigned long dummy)
612{
Eric Dumazetbb1d23b2005-07-05 15:00:32 -0700613 static unsigned int rover;
614 unsigned int i = rover, goal;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 struct rtable *rth, **rthp;
616 unsigned long now = jiffies;
Eric Dumazetbb1d23b2005-07-05 15:00:32 -0700617 u64 mult;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
Eric Dumazetbb1d23b2005-07-05 15:00:32 -0700619 mult = ((u64)ip_rt_gc_interval) << rt_hash_log;
620 if (ip_rt_gc_timeout > 1)
621 do_div(mult, ip_rt_gc_timeout);
622 goal = (unsigned int)mult;
623 if (goal > rt_hash_mask) goal = rt_hash_mask + 1;
624 for (; goal > 0; goal--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 unsigned long tmo = ip_rt_gc_timeout;
626
627 i = (i + 1) & rt_hash_mask;
628 rthp = &rt_hash_table[i].chain;
629
Eric Dumazetbb1d23b2005-07-05 15:00:32 -0700630 if (*rthp == 0)
631 continue;
Eric Dumazet22c047c2005-07-05 14:55:24 -0700632 spin_lock(rt_hash_lock_addr(i));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 while ((rth = *rthp) != NULL) {
634 if (rth->u.dst.expires) {
635 /* Entry is expired even if it is in use */
636 if (time_before_eq(now, rth->u.dst.expires)) {
637 tmo >>= 1;
638 rthp = &rth->u.rt_next;
639 continue;
640 }
641 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout)) {
642 tmo >>= 1;
643 rthp = &rth->u.rt_next;
644 continue;
645 }
646
647 /* Cleanup aged off entries. */
648#ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
649 /* remove all related balanced entries if necessary */
650 if (rth->u.dst.flags & DST_BALANCED) {
651 rthp = rt_remove_balanced_route(
652 &rt_hash_table[i].chain,
653 rth, NULL);
654 if (!rthp)
655 break;
656 } else {
657 *rthp = rth->u.rt_next;
658 rt_free(rth);
659 }
660#else /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
661 *rthp = rth->u.rt_next;
662 rt_free(rth);
663#endif /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
664 }
Eric Dumazet22c047c2005-07-05 14:55:24 -0700665 spin_unlock(rt_hash_lock_addr(i));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666
667 /* Fallback loop breaker. */
668 if (time_after(jiffies, now))
669 break;
670 }
671 rover = i;
Eric Dumazetbb1d23b2005-07-05 15:00:32 -0700672 mod_timer(&rt_periodic_timer, jiffies + ip_rt_gc_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673}
674
675/* This can run from both BH and non-BH contexts, the latter
676 * in the case of a forced flush event.
677 */
678static void rt_run_flush(unsigned long dummy)
679{
680 int i;
681 struct rtable *rth, *next;
682
683 rt_deadline = 0;
684
685 get_random_bytes(&rt_hash_rnd, 4);
686
687 for (i = rt_hash_mask; i >= 0; i--) {
Eric Dumazet22c047c2005-07-05 14:55:24 -0700688 spin_lock_bh(rt_hash_lock_addr(i));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 rth = rt_hash_table[i].chain;
690 if (rth)
691 rt_hash_table[i].chain = NULL;
Eric Dumazet22c047c2005-07-05 14:55:24 -0700692 spin_unlock_bh(rt_hash_lock_addr(i));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
694 for (; rth; rth = next) {
695 next = rth->u.rt_next;
696 rt_free(rth);
697 }
698 }
699}
700
701static DEFINE_SPINLOCK(rt_flush_lock);
702
703void rt_cache_flush(int delay)
704{
705 unsigned long now = jiffies;
706 int user_mode = !in_softirq();
707
708 if (delay < 0)
709 delay = ip_rt_min_delay;
710
711 /* flush existing multipath state*/
712 multipath_flush();
713
714 spin_lock_bh(&rt_flush_lock);
715
716 if (del_timer(&rt_flush_timer) && delay > 0 && rt_deadline) {
717 long tmo = (long)(rt_deadline - now);
718
719 /* If flush timer is already running
720 and flush request is not immediate (delay > 0):
721
722 if deadline is not achieved, prolongate timer to "delay",
723 otherwise fire it at deadline time.
724 */
725
726 if (user_mode && tmo < ip_rt_max_delay-ip_rt_min_delay)
727 tmo = 0;
728
729 if (delay > tmo)
730 delay = tmo;
731 }
732
733 if (delay <= 0) {
734 spin_unlock_bh(&rt_flush_lock);
735 rt_run_flush(0);
736 return;
737 }
738
739 if (rt_deadline == 0)
740 rt_deadline = now + ip_rt_max_delay;
741
742 mod_timer(&rt_flush_timer, now+delay);
743 spin_unlock_bh(&rt_flush_lock);
744}
745
746static void rt_secret_rebuild(unsigned long dummy)
747{
748 unsigned long now = jiffies;
749
750 rt_cache_flush(0);
751 mod_timer(&rt_secret_timer, now + ip_rt_secret_interval);
752}
753
754/*
755 Short description of GC goals.
756
757 We want to build algorithm, which will keep routing cache
758 at some equilibrium point, when number of aged off entries
759 is kept approximately equal to newly generated ones.
760
761 Current expiration strength is variable "expire".
762 We try to adjust it dynamically, so that if networking
763 is idle expires is large enough to keep enough of warm entries,
764 and when load increases it reduces to limit cache size.
765 */
766
767static int rt_garbage_collect(void)
768{
769 static unsigned long expire = RT_GC_TIMEOUT;
770 static unsigned long last_gc;
771 static int rover;
772 static int equilibrium;
773 struct rtable *rth, **rthp;
774 unsigned long now = jiffies;
775 int goal;
776
777 /*
778 * Garbage collection is pretty expensive,
779 * do not make it too frequently.
780 */
781
782 RT_CACHE_STAT_INC(gc_total);
783
784 if (now - last_gc < ip_rt_gc_min_interval &&
785 atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) {
786 RT_CACHE_STAT_INC(gc_ignored);
787 goto out;
788 }
789
790 /* Calculate number of entries, which we want to expire now. */
791 goal = atomic_read(&ipv4_dst_ops.entries) -
792 (ip_rt_gc_elasticity << rt_hash_log);
793 if (goal <= 0) {
794 if (equilibrium < ipv4_dst_ops.gc_thresh)
795 equilibrium = ipv4_dst_ops.gc_thresh;
796 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
797 if (goal > 0) {
798 equilibrium += min_t(unsigned int, goal / 2, rt_hash_mask + 1);
799 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
800 }
801 } else {
802 /* We are in dangerous area. Try to reduce cache really
803 * aggressively.
804 */
805 goal = max_t(unsigned int, goal / 2, rt_hash_mask + 1);
806 equilibrium = atomic_read(&ipv4_dst_ops.entries) - goal;
807 }
808
809 if (now - last_gc >= ip_rt_gc_min_interval)
810 last_gc = now;
811
812 if (goal <= 0) {
813 equilibrium += goal;
814 goto work_done;
815 }
816
817 do {
818 int i, k;
819
820 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
821 unsigned long tmo = expire;
822
823 k = (k + 1) & rt_hash_mask;
824 rthp = &rt_hash_table[k].chain;
Eric Dumazet22c047c2005-07-05 14:55:24 -0700825 spin_lock_bh(rt_hash_lock_addr(k));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 while ((rth = *rthp) != NULL) {
827 if (!rt_may_expire(rth, tmo, expire)) {
828 tmo >>= 1;
829 rthp = &rth->u.rt_next;
830 continue;
831 }
832#ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
833 /* remove all related balanced entries
834 * if necessary
835 */
836 if (rth->u.dst.flags & DST_BALANCED) {
837 int r;
838
839 rthp = rt_remove_balanced_route(
Suresh Bhogavilli85259872006-02-21 13:42:22 -0800840 &rt_hash_table[k].chain,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 rth,
842 &r);
843 goal -= r;
844 if (!rthp)
845 break;
846 } else {
847 *rthp = rth->u.rt_next;
848 rt_free(rth);
849 goal--;
850 }
851#else /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
852 *rthp = rth->u.rt_next;
853 rt_free(rth);
854 goal--;
855#endif /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
856 }
Eric Dumazet22c047c2005-07-05 14:55:24 -0700857 spin_unlock_bh(rt_hash_lock_addr(k));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 if (goal <= 0)
859 break;
860 }
861 rover = k;
862
863 if (goal <= 0)
864 goto work_done;
865
866 /* Goal is not achieved. We stop process if:
867
868 - if expire reduced to zero. Otherwise, expire is halfed.
869 - if table is not full.
870 - if we are called from interrupt.
871 - jiffies check is just fallback/debug loop breaker.
872 We will not spin here for long time in any case.
873 */
874
875 RT_CACHE_STAT_INC(gc_goal_miss);
876
877 if (expire == 0)
878 break;
879
880 expire >>= 1;
881#if RT_CACHE_DEBUG >= 2
882 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
883 atomic_read(&ipv4_dst_ops.entries), goal, i);
884#endif
885
886 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
887 goto out;
888 } while (!in_softirq() && time_before_eq(jiffies, now));
889
890 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
891 goto out;
892 if (net_ratelimit())
893 printk(KERN_WARNING "dst cache overflow\n");
894 RT_CACHE_STAT_INC(gc_dst_overflow);
895 return 1;
896
897work_done:
898 expire += ip_rt_gc_min_interval;
899 if (expire > ip_rt_gc_timeout ||
900 atomic_read(&ipv4_dst_ops.entries) < ipv4_dst_ops.gc_thresh)
901 expire = ip_rt_gc_timeout;
902#if RT_CACHE_DEBUG >= 2
903 printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
904 atomic_read(&ipv4_dst_ops.entries), goal, rover);
905#endif
906out: return 0;
907}
908
909static int rt_intern_hash(unsigned hash, struct rtable *rt, struct rtable **rp)
910{
911 struct rtable *rth, **rthp;
912 unsigned long now;
913 struct rtable *cand, **candp;
914 u32 min_score;
915 int chain_length;
916 int attempts = !in_softirq();
917
918restart:
919 chain_length = 0;
920 min_score = ~(u32)0;
921 cand = NULL;
922 candp = NULL;
923 now = jiffies;
924
925 rthp = &rt_hash_table[hash].chain;
926
Eric Dumazet22c047c2005-07-05 14:55:24 -0700927 spin_lock_bh(rt_hash_lock_addr(hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 while ((rth = *rthp) != NULL) {
929#ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
930 if (!(rth->u.dst.flags & DST_BALANCED) &&
931 compare_keys(&rth->fl, &rt->fl)) {
932#else
933 if (compare_keys(&rth->fl, &rt->fl)) {
934#endif
935 /* Put it first */
936 *rthp = rth->u.rt_next;
937 /*
938 * Since lookup is lockfree, the deletion
939 * must be visible to another weakly ordered CPU before
940 * the insertion at the start of the hash chain.
941 */
942 rcu_assign_pointer(rth->u.rt_next,
943 rt_hash_table[hash].chain);
944 /*
945 * Since lookup is lockfree, the update writes
946 * must be ordered for consistency on SMP.
947 */
948 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
949
950 rth->u.dst.__use++;
951 dst_hold(&rth->u.dst);
952 rth->u.dst.lastuse = now;
Eric Dumazet22c047c2005-07-05 14:55:24 -0700953 spin_unlock_bh(rt_hash_lock_addr(hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
955 rt_drop(rt);
956 *rp = rth;
957 return 0;
958 }
959
960 if (!atomic_read(&rth->u.dst.__refcnt)) {
961 u32 score = rt_score(rth);
962
963 if (score <= min_score) {
964 cand = rth;
965 candp = rthp;
966 min_score = score;
967 }
968 }
969
970 chain_length++;
971
972 rthp = &rth->u.rt_next;
973 }
974
975 if (cand) {
976 /* ip_rt_gc_elasticity used to be average length of chain
977 * length, when exceeded gc becomes really aggressive.
978 *
979 * The second limit is less certain. At the moment it allows
980 * only 2 entries per bucket. We will see.
981 */
982 if (chain_length > ip_rt_gc_elasticity) {
983 *candp = cand->u.rt_next;
984 rt_free(cand);
985 }
986 }
987
988 /* Try to bind route to arp only if it is output
989 route or unicast forwarding path.
990 */
991 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
992 int err = arp_bind_neighbour(&rt->u.dst);
993 if (err) {
Eric Dumazet22c047c2005-07-05 14:55:24 -0700994 spin_unlock_bh(rt_hash_lock_addr(hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
996 if (err != -ENOBUFS) {
997 rt_drop(rt);
998 return err;
999 }
1000
1001 /* Neighbour tables are full and nothing
1002 can be released. Try to shrink route cache,
1003 it is most likely it holds some neighbour records.
1004 */
1005 if (attempts-- > 0) {
1006 int saved_elasticity = ip_rt_gc_elasticity;
1007 int saved_int = ip_rt_gc_min_interval;
1008 ip_rt_gc_elasticity = 1;
1009 ip_rt_gc_min_interval = 0;
1010 rt_garbage_collect();
1011 ip_rt_gc_min_interval = saved_int;
1012 ip_rt_gc_elasticity = saved_elasticity;
1013 goto restart;
1014 }
1015
1016 if (net_ratelimit())
1017 printk(KERN_WARNING "Neighbour table overflow.\n");
1018 rt_drop(rt);
1019 return -ENOBUFS;
1020 }
1021 }
1022
1023 rt->u.rt_next = rt_hash_table[hash].chain;
1024#if RT_CACHE_DEBUG >= 2
1025 if (rt->u.rt_next) {
1026 struct rtable *trt;
1027 printk(KERN_DEBUG "rt_cache @%02x: %u.%u.%u.%u", hash,
1028 NIPQUAD(rt->rt_dst));
1029 for (trt = rt->u.rt_next; trt; trt = trt->u.rt_next)
1030 printk(" . %u.%u.%u.%u", NIPQUAD(trt->rt_dst));
1031 printk("\n");
1032 }
1033#endif
1034 rt_hash_table[hash].chain = rt;
Eric Dumazet22c047c2005-07-05 14:55:24 -07001035 spin_unlock_bh(rt_hash_lock_addr(hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 *rp = rt;
1037 return 0;
1038}
1039
1040void rt_bind_peer(struct rtable *rt, int create)
1041{
1042 static DEFINE_SPINLOCK(rt_peer_lock);
1043 struct inet_peer *peer;
1044
1045 peer = inet_getpeer(rt->rt_dst, create);
1046
1047 spin_lock_bh(&rt_peer_lock);
1048 if (rt->peer == NULL) {
1049 rt->peer = peer;
1050 peer = NULL;
1051 }
1052 spin_unlock_bh(&rt_peer_lock);
1053 if (peer)
1054 inet_putpeer(peer);
1055}
1056
1057/*
1058 * Peer allocation may fail only in serious out-of-memory conditions. However
1059 * we still can generate some output.
1060 * Random ID selection looks a bit dangerous because we have no chances to
1061 * select ID being unique in a reasonable period of time.
1062 * But broken packet identifier may be better than no packet at all.
1063 */
1064static void ip_select_fb_ident(struct iphdr *iph)
1065{
1066 static DEFINE_SPINLOCK(ip_fb_id_lock);
1067 static u32 ip_fallback_id;
1068 u32 salt;
1069
1070 spin_lock_bh(&ip_fb_id_lock);
1071 salt = secure_ip_id(ip_fallback_id ^ iph->daddr);
1072 iph->id = htons(salt & 0xFFFF);
1073 ip_fallback_id = salt;
1074 spin_unlock_bh(&ip_fb_id_lock);
1075}
1076
1077void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1078{
1079 struct rtable *rt = (struct rtable *) dst;
1080
1081 if (rt) {
1082 if (rt->peer == NULL)
1083 rt_bind_peer(rt, 1);
1084
1085 /* If peer is attached to destination, it is never detached,
1086 so that we need not to grab a lock to dereference it.
1087 */
1088 if (rt->peer) {
1089 iph->id = htons(inet_getid(rt->peer, more));
1090 return;
1091 }
1092 } else
Stephen Hemminger9c2b3322005-04-19 22:39:42 -07001093 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1094 __builtin_return_address(0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
1096 ip_select_fb_ident(iph);
1097}
1098
1099static void rt_del(unsigned hash, struct rtable *rt)
1100{
1101 struct rtable **rthp;
1102
Eric Dumazet22c047c2005-07-05 14:55:24 -07001103 spin_lock_bh(rt_hash_lock_addr(hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 ip_rt_put(rt);
1105 for (rthp = &rt_hash_table[hash].chain; *rthp;
1106 rthp = &(*rthp)->u.rt_next)
1107 if (*rthp == rt) {
1108 *rthp = rt->u.rt_next;
1109 rt_free(rt);
1110 break;
1111 }
Eric Dumazet22c047c2005-07-05 14:55:24 -07001112 spin_unlock_bh(rt_hash_lock_addr(hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113}
1114
1115void ip_rt_redirect(u32 old_gw, u32 daddr, u32 new_gw,
Ilia Sotnikovcef26852006-03-25 01:38:55 -08001116 u32 saddr, struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117{
1118 int i, k;
1119 struct in_device *in_dev = in_dev_get(dev);
1120 struct rtable *rth, **rthp;
1121 u32 skeys[2] = { saddr, 0 };
1122 int ikeys[2] = { dev->ifindex, 0 };
1123
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 if (!in_dev)
1125 return;
1126
1127 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev)
1128 || MULTICAST(new_gw) || BADCLASS(new_gw) || ZERONET(new_gw))
1129 goto reject_redirect;
1130
1131 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1132 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1133 goto reject_redirect;
1134 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1135 goto reject_redirect;
1136 } else {
1137 if (inet_addr_type(new_gw) != RTN_UNICAST)
1138 goto reject_redirect;
1139 }
1140
1141 for (i = 0; i < 2; i++) {
1142 for (k = 0; k < 2; k++) {
1143 unsigned hash = rt_hash_code(daddr,
Ilia Sotnikovcef26852006-03-25 01:38:55 -08001144 skeys[i] ^ (ikeys[k] << 5));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
1146 rthp=&rt_hash_table[hash].chain;
1147
1148 rcu_read_lock();
1149 while ((rth = rcu_dereference(*rthp)) != NULL) {
1150 struct rtable *rt;
1151
1152 if (rth->fl.fl4_dst != daddr ||
1153 rth->fl.fl4_src != skeys[i] ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 rth->fl.oif != ikeys[k] ||
1155 rth->fl.iif != 0) {
1156 rthp = &rth->u.rt_next;
1157 continue;
1158 }
1159
1160 if (rth->rt_dst != daddr ||
1161 rth->rt_src != saddr ||
1162 rth->u.dst.error ||
1163 rth->rt_gateway != old_gw ||
1164 rth->u.dst.dev != dev)
1165 break;
1166
1167 dst_hold(&rth->u.dst);
1168 rcu_read_unlock();
1169
1170 rt = dst_alloc(&ipv4_dst_ops);
1171 if (rt == NULL) {
1172 ip_rt_put(rth);
1173 in_dev_put(in_dev);
1174 return;
1175 }
1176
1177 /* Copy all the information. */
1178 *rt = *rth;
1179 INIT_RCU_HEAD(&rt->u.dst.rcu_head);
1180 rt->u.dst.__use = 1;
1181 atomic_set(&rt->u.dst.__refcnt, 1);
1182 rt->u.dst.child = NULL;
1183 if (rt->u.dst.dev)
1184 dev_hold(rt->u.dst.dev);
1185 if (rt->idev)
1186 in_dev_hold(rt->idev);
1187 rt->u.dst.obsolete = 0;
1188 rt->u.dst.lastuse = jiffies;
1189 rt->u.dst.path = &rt->u.dst;
1190 rt->u.dst.neighbour = NULL;
1191 rt->u.dst.hh = NULL;
1192 rt->u.dst.xfrm = NULL;
1193
1194 rt->rt_flags |= RTCF_REDIRECTED;
1195
1196 /* Gateway is different ... */
1197 rt->rt_gateway = new_gw;
1198
1199 /* Redirect received -> path was valid */
1200 dst_confirm(&rth->u.dst);
1201
1202 if (rt->peer)
1203 atomic_inc(&rt->peer->refcnt);
1204
1205 if (arp_bind_neighbour(&rt->u.dst) ||
1206 !(rt->u.dst.neighbour->nud_state &
1207 NUD_VALID)) {
1208 if (rt->u.dst.neighbour)
1209 neigh_event_send(rt->u.dst.neighbour, NULL);
1210 ip_rt_put(rth);
1211 rt_drop(rt);
1212 goto do_next;
1213 }
1214
1215 rt_del(hash, rth);
1216 if (!rt_intern_hash(hash, rt, &rt))
1217 ip_rt_put(rt);
1218 goto do_next;
1219 }
1220 rcu_read_unlock();
1221 do_next:
1222 ;
1223 }
1224 }
1225 in_dev_put(in_dev);
1226 return;
1227
1228reject_redirect:
1229#ifdef CONFIG_IP_ROUTE_VERBOSE
1230 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1231 printk(KERN_INFO "Redirect from %u.%u.%u.%u on %s about "
1232 "%u.%u.%u.%u ignored.\n"
Ilia Sotnikovcef26852006-03-25 01:38:55 -08001233 " Advised path = %u.%u.%u.%u -> %u.%u.%u.%u\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 NIPQUAD(old_gw), dev->name, NIPQUAD(new_gw),
Ilia Sotnikovcef26852006-03-25 01:38:55 -08001235 NIPQUAD(saddr), NIPQUAD(daddr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236#endif
1237 in_dev_put(in_dev);
1238}
1239
1240static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1241{
1242 struct rtable *rt = (struct rtable*)dst;
1243 struct dst_entry *ret = dst;
1244
1245 if (rt) {
1246 if (dst->obsolete) {
1247 ip_rt_put(rt);
1248 ret = NULL;
1249 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1250 rt->u.dst.expires) {
1251 unsigned hash = rt_hash_code(rt->fl.fl4_dst,
1252 rt->fl.fl4_src ^
Ilia Sotnikovcef26852006-03-25 01:38:55 -08001253 (rt->fl.oif << 5));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254#if RT_CACHE_DEBUG >= 1
1255 printk(KERN_DEBUG "ip_rt_advice: redirect to "
1256 "%u.%u.%u.%u/%02x dropped\n",
1257 NIPQUAD(rt->rt_dst), rt->fl.fl4_tos);
1258#endif
1259 rt_del(hash, rt);
1260 ret = NULL;
1261 }
1262 }
1263 return ret;
1264}
1265
1266/*
1267 * Algorithm:
1268 * 1. The first ip_rt_redirect_number redirects are sent
1269 * with exponential backoff, then we stop sending them at all,
1270 * assuming that the host ignores our redirects.
1271 * 2. If we did not see packets requiring redirects
1272 * during ip_rt_redirect_silence, we assume that the host
1273 * forgot redirected route and start to send redirects again.
1274 *
1275 * This algorithm is much cheaper and more intelligent than dumb load limiting
1276 * in icmp.c.
1277 *
1278 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1279 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1280 */
1281
1282void ip_rt_send_redirect(struct sk_buff *skb)
1283{
1284 struct rtable *rt = (struct rtable*)skb->dst;
1285 struct in_device *in_dev = in_dev_get(rt->u.dst.dev);
1286
1287 if (!in_dev)
1288 return;
1289
1290 if (!IN_DEV_TX_REDIRECTS(in_dev))
1291 goto out;
1292
1293 /* No redirected packets during ip_rt_redirect_silence;
1294 * reset the algorithm.
1295 */
1296 if (time_after(jiffies, rt->u.dst.rate_last + ip_rt_redirect_silence))
1297 rt->u.dst.rate_tokens = 0;
1298
1299 /* Too many ignored redirects; do not send anything
1300 * set u.dst.rate_last to the last seen redirected packet.
1301 */
1302 if (rt->u.dst.rate_tokens >= ip_rt_redirect_number) {
1303 rt->u.dst.rate_last = jiffies;
1304 goto out;
1305 }
1306
1307 /* Check for load limit; set rate_last to the latest sent
1308 * redirect.
1309 */
1310 if (time_after(jiffies,
1311 (rt->u.dst.rate_last +
1312 (ip_rt_redirect_load << rt->u.dst.rate_tokens)))) {
1313 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1314 rt->u.dst.rate_last = jiffies;
1315 ++rt->u.dst.rate_tokens;
1316#ifdef CONFIG_IP_ROUTE_VERBOSE
1317 if (IN_DEV_LOG_MARTIANS(in_dev) &&
1318 rt->u.dst.rate_tokens == ip_rt_redirect_number &&
1319 net_ratelimit())
1320 printk(KERN_WARNING "host %u.%u.%u.%u/if%d ignores "
1321 "redirects for %u.%u.%u.%u to %u.%u.%u.%u.\n",
1322 NIPQUAD(rt->rt_src), rt->rt_iif,
1323 NIPQUAD(rt->rt_dst), NIPQUAD(rt->rt_gateway));
1324#endif
1325 }
1326out:
1327 in_dev_put(in_dev);
1328}
1329
1330static int ip_error(struct sk_buff *skb)
1331{
1332 struct rtable *rt = (struct rtable*)skb->dst;
1333 unsigned long now;
1334 int code;
1335
1336 switch (rt->u.dst.error) {
1337 case EINVAL:
1338 default:
1339 goto out;
1340 case EHOSTUNREACH:
1341 code = ICMP_HOST_UNREACH;
1342 break;
1343 case ENETUNREACH:
1344 code = ICMP_NET_UNREACH;
1345 break;
1346 case EACCES:
1347 code = ICMP_PKT_FILTERED;
1348 break;
1349 }
1350
1351 now = jiffies;
1352 rt->u.dst.rate_tokens += now - rt->u.dst.rate_last;
1353 if (rt->u.dst.rate_tokens > ip_rt_error_burst)
1354 rt->u.dst.rate_tokens = ip_rt_error_burst;
1355 rt->u.dst.rate_last = now;
1356 if (rt->u.dst.rate_tokens >= ip_rt_error_cost) {
1357 rt->u.dst.rate_tokens -= ip_rt_error_cost;
1358 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1359 }
1360
1361out: kfree_skb(skb);
1362 return 0;
1363}
1364
1365/*
1366 * The last two values are not from the RFC but
1367 * are needed for AMPRnet AX.25 paths.
1368 */
1369
Arjan van de Ven9b5b5cf2005-11-29 16:21:38 -08001370static const unsigned short mtu_plateau[] =
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371{32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1372
1373static __inline__ unsigned short guess_mtu(unsigned short old_mtu)
1374{
1375 int i;
1376
1377 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1378 if (old_mtu > mtu_plateau[i])
1379 return mtu_plateau[i];
1380 return 68;
1381}
1382
1383unsigned short ip_rt_frag_needed(struct iphdr *iph, unsigned short new_mtu)
1384{
1385 int i;
1386 unsigned short old_mtu = ntohs(iph->tot_len);
1387 struct rtable *rth;
1388 u32 skeys[2] = { iph->saddr, 0, };
1389 u32 daddr = iph->daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 unsigned short est_mtu = 0;
1391
1392 if (ipv4_config.no_pmtu_disc)
1393 return 0;
1394
1395 for (i = 0; i < 2; i++) {
Ilia Sotnikovcef26852006-03-25 01:38:55 -08001396 unsigned hash = rt_hash_code(daddr, skeys[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397
1398 rcu_read_lock();
1399 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1400 rth = rcu_dereference(rth->u.rt_next)) {
1401 if (rth->fl.fl4_dst == daddr &&
1402 rth->fl.fl4_src == skeys[i] &&
1403 rth->rt_dst == daddr &&
1404 rth->rt_src == iph->saddr &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 rth->fl.iif == 0 &&
1406 !(dst_metric_locked(&rth->u.dst, RTAX_MTU))) {
1407 unsigned short mtu = new_mtu;
1408
1409 if (new_mtu < 68 || new_mtu >= old_mtu) {
1410
1411 /* BSD 4.2 compatibility hack :-( */
1412 if (mtu == 0 &&
1413 old_mtu >= rth->u.dst.metrics[RTAX_MTU-1] &&
1414 old_mtu >= 68 + (iph->ihl << 2))
1415 old_mtu -= iph->ihl << 2;
1416
1417 mtu = guess_mtu(old_mtu);
1418 }
1419 if (mtu <= rth->u.dst.metrics[RTAX_MTU-1]) {
1420 if (mtu < rth->u.dst.metrics[RTAX_MTU-1]) {
1421 dst_confirm(&rth->u.dst);
1422 if (mtu < ip_rt_min_pmtu) {
1423 mtu = ip_rt_min_pmtu;
1424 rth->u.dst.metrics[RTAX_LOCK-1] |=
1425 (1 << RTAX_MTU);
1426 }
1427 rth->u.dst.metrics[RTAX_MTU-1] = mtu;
1428 dst_set_expires(&rth->u.dst,
1429 ip_rt_mtu_expires);
1430 }
1431 est_mtu = mtu;
1432 }
1433 }
1434 }
1435 rcu_read_unlock();
1436 }
1437 return est_mtu ? : new_mtu;
1438}
1439
1440static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1441{
1442 if (dst->metrics[RTAX_MTU-1] > mtu && mtu >= 68 &&
1443 !(dst_metric_locked(dst, RTAX_MTU))) {
1444 if (mtu < ip_rt_min_pmtu) {
1445 mtu = ip_rt_min_pmtu;
1446 dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1447 }
1448 dst->metrics[RTAX_MTU-1] = mtu;
1449 dst_set_expires(dst, ip_rt_mtu_expires);
1450 }
1451}
1452
1453static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1454{
1455 return NULL;
1456}
1457
1458static void ipv4_dst_destroy(struct dst_entry *dst)
1459{
1460 struct rtable *rt = (struct rtable *) dst;
1461 struct inet_peer *peer = rt->peer;
1462 struct in_device *idev = rt->idev;
1463
1464 if (peer) {
1465 rt->peer = NULL;
1466 inet_putpeer(peer);
1467 }
1468
1469 if (idev) {
1470 rt->idev = NULL;
1471 in_dev_put(idev);
1472 }
1473}
1474
1475static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1476 int how)
1477{
1478 struct rtable *rt = (struct rtable *) dst;
1479 struct in_device *idev = rt->idev;
1480 if (dev != &loopback_dev && idev && idev->dev == dev) {
1481 struct in_device *loopback_idev = in_dev_get(&loopback_dev);
1482 if (loopback_idev) {
1483 rt->idev = loopback_idev;
1484 in_dev_put(idev);
1485 }
1486 }
1487}
1488
1489static void ipv4_link_failure(struct sk_buff *skb)
1490{
1491 struct rtable *rt;
1492
1493 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1494
1495 rt = (struct rtable *) skb->dst;
1496 if (rt)
1497 dst_set_expires(&rt->u.dst, 0);
1498}
1499
1500static int ip_rt_bug(struct sk_buff *skb)
1501{
1502 printk(KERN_DEBUG "ip_rt_bug: %u.%u.%u.%u -> %u.%u.%u.%u, %s\n",
1503 NIPQUAD(skb->nh.iph->saddr), NIPQUAD(skb->nh.iph->daddr),
1504 skb->dev ? skb->dev->name : "?");
1505 kfree_skb(skb);
1506 return 0;
1507}
1508
1509/*
1510 We do not cache source address of outgoing interface,
1511 because it is used only by IP RR, TS and SRR options,
1512 so that it out of fast path.
1513
1514 BTW remember: "addr" is allowed to be not aligned
1515 in IP options!
1516 */
1517
1518void ip_rt_get_source(u8 *addr, struct rtable *rt)
1519{
1520 u32 src;
1521 struct fib_result res;
1522
1523 if (rt->fl.iif == 0)
1524 src = rt->rt_src;
1525 else if (fib_lookup(&rt->fl, &res) == 0) {
1526 src = FIB_RES_PREFSRC(res);
1527 fib_res_put(&res);
1528 } else
1529 src = inet_select_addr(rt->u.dst.dev, rt->rt_gateway,
1530 RT_SCOPE_UNIVERSE);
1531 memcpy(addr, &src, 4);
1532}
1533
1534#ifdef CONFIG_NET_CLS_ROUTE
1535static void set_class_tag(struct rtable *rt, u32 tag)
1536{
1537 if (!(rt->u.dst.tclassid & 0xFFFF))
1538 rt->u.dst.tclassid |= tag & 0xFFFF;
1539 if (!(rt->u.dst.tclassid & 0xFFFF0000))
1540 rt->u.dst.tclassid |= tag & 0xFFFF0000;
1541}
1542#endif
1543
1544static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1545{
1546 struct fib_info *fi = res->fi;
1547
1548 if (fi) {
1549 if (FIB_RES_GW(*res) &&
1550 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1551 rt->rt_gateway = FIB_RES_GW(*res);
1552 memcpy(rt->u.dst.metrics, fi->fib_metrics,
1553 sizeof(rt->u.dst.metrics));
1554 if (fi->fib_mtu == 0) {
1555 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
1556 if (rt->u.dst.metrics[RTAX_LOCK-1] & (1 << RTAX_MTU) &&
1557 rt->rt_gateway != rt->rt_dst &&
1558 rt->u.dst.dev->mtu > 576)
1559 rt->u.dst.metrics[RTAX_MTU-1] = 576;
1560 }
1561#ifdef CONFIG_NET_CLS_ROUTE
1562 rt->u.dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1563#endif
1564 } else
1565 rt->u.dst.metrics[RTAX_MTU-1]= rt->u.dst.dev->mtu;
1566
1567 if (rt->u.dst.metrics[RTAX_HOPLIMIT-1] == 0)
1568 rt->u.dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1569 if (rt->u.dst.metrics[RTAX_MTU-1] > IP_MAX_MTU)
1570 rt->u.dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1571 if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0)
1572 rt->u.dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->u.dst.dev->mtu - 40,
1573 ip_rt_min_advmss);
1574 if (rt->u.dst.metrics[RTAX_ADVMSS-1] > 65535 - 40)
1575 rt->u.dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1576
1577#ifdef CONFIG_NET_CLS_ROUTE
1578#ifdef CONFIG_IP_MULTIPLE_TABLES
1579 set_class_tag(rt, fib_rules_tclass(res));
1580#endif
1581 set_class_tag(rt, itag);
1582#endif
1583 rt->rt_type = res->type;
1584}
1585
1586static int ip_route_input_mc(struct sk_buff *skb, u32 daddr, u32 saddr,
1587 u8 tos, struct net_device *dev, int our)
1588{
1589 unsigned hash;
1590 struct rtable *rth;
1591 u32 spec_dst;
1592 struct in_device *in_dev = in_dev_get(dev);
1593 u32 itag = 0;
1594
1595 /* Primary sanity checks. */
1596
1597 if (in_dev == NULL)
1598 return -EINVAL;
1599
1600 if (MULTICAST(saddr) || BADCLASS(saddr) || LOOPBACK(saddr) ||
1601 skb->protocol != htons(ETH_P_IP))
1602 goto e_inval;
1603
1604 if (ZERONET(saddr)) {
1605 if (!LOCAL_MCAST(daddr))
1606 goto e_inval;
1607 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1608 } else if (fib_validate_source(saddr, 0, tos, 0,
1609 dev, &spec_dst, &itag) < 0)
1610 goto e_inval;
1611
1612 rth = dst_alloc(&ipv4_dst_ops);
1613 if (!rth)
1614 goto e_nobufs;
1615
1616 rth->u.dst.output= ip_rt_bug;
1617
1618 atomic_set(&rth->u.dst.__refcnt, 1);
1619 rth->u.dst.flags= DST_HOST;
1620 if (in_dev->cnf.no_policy)
1621 rth->u.dst.flags |= DST_NOPOLICY;
1622 rth->fl.fl4_dst = daddr;
1623 rth->rt_dst = daddr;
1624 rth->fl.fl4_tos = tos;
1625#ifdef CONFIG_IP_ROUTE_FWMARK
1626 rth->fl.fl4_fwmark= skb->nfmark;
1627#endif
1628 rth->fl.fl4_src = saddr;
1629 rth->rt_src = saddr;
1630#ifdef CONFIG_NET_CLS_ROUTE
1631 rth->u.dst.tclassid = itag;
1632#endif
1633 rth->rt_iif =
1634 rth->fl.iif = dev->ifindex;
1635 rth->u.dst.dev = &loopback_dev;
1636 dev_hold(rth->u.dst.dev);
1637 rth->idev = in_dev_get(rth->u.dst.dev);
1638 rth->fl.oif = 0;
1639 rth->rt_gateway = daddr;
1640 rth->rt_spec_dst= spec_dst;
1641 rth->rt_type = RTN_MULTICAST;
1642 rth->rt_flags = RTCF_MULTICAST;
1643 if (our) {
1644 rth->u.dst.input= ip_local_deliver;
1645 rth->rt_flags |= RTCF_LOCAL;
1646 }
1647
1648#ifdef CONFIG_IP_MROUTE
1649 if (!LOCAL_MCAST(daddr) && IN_DEV_MFORWARD(in_dev))
1650 rth->u.dst.input = ip_mr_input;
1651#endif
1652 RT_CACHE_STAT_INC(in_slow_mc);
1653
1654 in_dev_put(in_dev);
Ilia Sotnikovcef26852006-03-25 01:38:55 -08001655 hash = rt_hash_code(daddr, saddr ^ (dev->ifindex << 5));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 return rt_intern_hash(hash, rth, (struct rtable**) &skb->dst);
1657
1658e_nobufs:
1659 in_dev_put(in_dev);
1660 return -ENOBUFS;
1661
1662e_inval:
1663 in_dev_put(in_dev);
1664 return -EINVAL;
1665}
1666
1667
1668static void ip_handle_martian_source(struct net_device *dev,
1669 struct in_device *in_dev,
1670 struct sk_buff *skb,
1671 u32 daddr,
1672 u32 saddr)
1673{
1674 RT_CACHE_STAT_INC(in_martian_src);
1675#ifdef CONFIG_IP_ROUTE_VERBOSE
1676 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1677 /*
1678 * RFC1812 recommendation, if source is martian,
1679 * the only hint is MAC header.
1680 */
1681 printk(KERN_WARNING "martian source %u.%u.%u.%u from "
1682 "%u.%u.%u.%u, on dev %s\n",
1683 NIPQUAD(daddr), NIPQUAD(saddr), dev->name);
Olaf Kirch0b7f22a2005-07-11 21:01:42 -07001684 if (dev->hard_header_len && skb->mac.raw) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 int i;
1686 unsigned char *p = skb->mac.raw;
1687 printk(KERN_WARNING "ll header: ");
1688 for (i = 0; i < dev->hard_header_len; i++, p++) {
1689 printk("%02x", *p);
1690 if (i < (dev->hard_header_len - 1))
1691 printk(":");
1692 }
1693 printk("\n");
1694 }
1695 }
1696#endif
1697}
1698
1699static inline int __mkroute_input(struct sk_buff *skb,
1700 struct fib_result* res,
1701 struct in_device *in_dev,
1702 u32 daddr, u32 saddr, u32 tos,
1703 struct rtable **result)
1704{
1705
1706 struct rtable *rth;
1707 int err;
1708 struct in_device *out_dev;
1709 unsigned flags = 0;
1710 u32 spec_dst, itag;
1711
1712 /* get a working reference to the output device */
1713 out_dev = in_dev_get(FIB_RES_DEV(*res));
1714 if (out_dev == NULL) {
1715 if (net_ratelimit())
1716 printk(KERN_CRIT "Bug in ip_route_input" \
1717 "_slow(). Please, report\n");
1718 return -EINVAL;
1719 }
1720
1721
1722 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1723 in_dev->dev, &spec_dst, &itag);
1724 if (err < 0) {
1725 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1726 saddr);
1727
1728 err = -EINVAL;
1729 goto cleanup;
1730 }
1731
1732 if (err)
1733 flags |= RTCF_DIRECTSRC;
1734
1735 if (out_dev == in_dev && err && !(flags & (RTCF_NAT | RTCF_MASQ)) &&
1736 (IN_DEV_SHARED_MEDIA(out_dev) ||
1737 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1738 flags |= RTCF_DOREDIRECT;
1739
1740 if (skb->protocol != htons(ETH_P_IP)) {
1741 /* Not IP (i.e. ARP). Do not create route, if it is
1742 * invalid for proxy arp. DNAT routes are always valid.
1743 */
1744 if (out_dev == in_dev && !(flags & RTCF_DNAT)) {
1745 err = -EINVAL;
1746 goto cleanup;
1747 }
1748 }
1749
1750
1751 rth = dst_alloc(&ipv4_dst_ops);
1752 if (!rth) {
1753 err = -ENOBUFS;
1754 goto cleanup;
1755 }
1756
Julian Anastasovce723d82005-09-08 13:34:47 -07001757 atomic_set(&rth->u.dst.__refcnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 rth->u.dst.flags= DST_HOST;
1759#ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
1760 if (res->fi->fib_nhs > 1)
1761 rth->u.dst.flags |= DST_BALANCED;
1762#endif
1763 if (in_dev->cnf.no_policy)
1764 rth->u.dst.flags |= DST_NOPOLICY;
1765 if (in_dev->cnf.no_xfrm)
1766 rth->u.dst.flags |= DST_NOXFRM;
1767 rth->fl.fl4_dst = daddr;
1768 rth->rt_dst = daddr;
1769 rth->fl.fl4_tos = tos;
1770#ifdef CONFIG_IP_ROUTE_FWMARK
1771 rth->fl.fl4_fwmark= skb->nfmark;
1772#endif
1773 rth->fl.fl4_src = saddr;
1774 rth->rt_src = saddr;
1775 rth->rt_gateway = daddr;
1776 rth->rt_iif =
1777 rth->fl.iif = in_dev->dev->ifindex;
1778 rth->u.dst.dev = (out_dev)->dev;
1779 dev_hold(rth->u.dst.dev);
1780 rth->idev = in_dev_get(rth->u.dst.dev);
1781 rth->fl.oif = 0;
1782 rth->rt_spec_dst= spec_dst;
1783
1784 rth->u.dst.input = ip_forward;
1785 rth->u.dst.output = ip_output;
1786
1787 rt_set_nexthop(rth, res, itag);
1788
1789 rth->rt_flags = flags;
1790
1791 *result = rth;
1792 err = 0;
1793 cleanup:
1794 /* release the working reference to the output device */
1795 in_dev_put(out_dev);
1796 return err;
1797}
1798
1799static inline int ip_mkroute_input_def(struct sk_buff *skb,
1800 struct fib_result* res,
1801 const struct flowi *fl,
1802 struct in_device *in_dev,
1803 u32 daddr, u32 saddr, u32 tos)
1804{
Chuck Short7abaa272005-06-22 22:10:23 -07001805 struct rtable* rth = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 int err;
1807 unsigned hash;
1808
1809#ifdef CONFIG_IP_ROUTE_MULTIPATH
1810 if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
1811 fib_select_multipath(fl, res);
1812#endif
1813
1814 /* create a routing cache entry */
1815 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
1816 if (err)
1817 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
1819 /* put it into the cache */
Ilia Sotnikovcef26852006-03-25 01:38:55 -08001820 hash = rt_hash_code(daddr, saddr ^ (fl->iif << 5));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 return rt_intern_hash(hash, rth, (struct rtable**)&skb->dst);
1822}
1823
1824static inline int ip_mkroute_input(struct sk_buff *skb,
1825 struct fib_result* res,
1826 const struct flowi *fl,
1827 struct in_device *in_dev,
1828 u32 daddr, u32 saddr, u32 tos)
1829{
1830#ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
Julian Anastasovce723d82005-09-08 13:34:47 -07001831 struct rtable* rth = NULL, *rtres;
1832 unsigned char hop, hopcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 int err = -EINVAL;
1834 unsigned int hash;
1835
1836 if (res->fi)
1837 hopcount = res->fi->fib_nhs;
1838 else
1839 hopcount = 1;
1840
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 /* distinguish between multipath and singlepath */
1842 if (hopcount < 2)
1843 return ip_mkroute_input_def(skb, res, fl, in_dev, daddr,
1844 saddr, tos);
1845
1846 /* add all alternatives to the routing cache */
1847 for (hop = 0; hop < hopcount; hop++) {
1848 res->nh_sel = hop;
1849
Julian Anastasovce723d82005-09-08 13:34:47 -07001850 /* put reference to previous result */
1851 if (hop)
1852 ip_rt_put(rtres);
1853
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 /* create a routing cache entry */
1855 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos,
1856 &rth);
1857 if (err)
1858 return err;
1859
1860 /* put it into the cache */
Ilia Sotnikovcef26852006-03-25 01:38:55 -08001861 hash = rt_hash_code(daddr, saddr ^ (fl->iif << 5));
Julian Anastasovce723d82005-09-08 13:34:47 -07001862 err = rt_intern_hash(hash, rth, &rtres);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 if (err)
1864 return err;
1865
1866 /* forward hop information to multipath impl. */
1867 multipath_set_nhinfo(rth,
1868 FIB_RES_NETWORK(*res),
1869 FIB_RES_NETMASK(*res),
1870 res->prefixlen,
1871 &FIB_RES_NH(*res));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 }
Julian Anastasovce723d82005-09-08 13:34:47 -07001873 skb->dst = &rtres->u.dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 return err;
1875#else /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
1876 return ip_mkroute_input_def(skb, res, fl, in_dev, daddr, saddr, tos);
1877#endif /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
1878}
1879
1880
1881/*
1882 * NOTE. We drop all the packets that has local source
1883 * addresses, because every properly looped back packet
1884 * must have correct destination already attached by output routine.
1885 *
1886 * Such approach solves two big problems:
1887 * 1. Not simplex devices are handled properly.
1888 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1889 */
1890
1891static int ip_route_input_slow(struct sk_buff *skb, u32 daddr, u32 saddr,
1892 u8 tos, struct net_device *dev)
1893{
1894 struct fib_result res;
1895 struct in_device *in_dev = in_dev_get(dev);
1896 struct flowi fl = { .nl_u = { .ip4_u =
1897 { .daddr = daddr,
1898 .saddr = saddr,
1899 .tos = tos,
1900 .scope = RT_SCOPE_UNIVERSE,
1901#ifdef CONFIG_IP_ROUTE_FWMARK
1902 .fwmark = skb->nfmark
1903#endif
1904 } },
1905 .iif = dev->ifindex };
1906 unsigned flags = 0;
1907 u32 itag = 0;
1908 struct rtable * rth;
1909 unsigned hash;
1910 u32 spec_dst;
1911 int err = -EINVAL;
1912 int free_res = 0;
1913
1914 /* IP on this device is disabled. */
1915
1916 if (!in_dev)
1917 goto out;
1918
1919 /* Check for the most weird martians, which can be not detected
1920 by fib_lookup.
1921 */
1922
1923 if (MULTICAST(saddr) || BADCLASS(saddr) || LOOPBACK(saddr))
1924 goto martian_source;
1925
1926 if (daddr == 0xFFFFFFFF || (saddr == 0 && daddr == 0))
1927 goto brd_input;
1928
1929 /* Accept zero addresses only to limited broadcast;
1930 * I even do not know to fix it or not. Waiting for complains :-)
1931 */
1932 if (ZERONET(saddr))
1933 goto martian_source;
1934
1935 if (BADCLASS(daddr) || ZERONET(daddr) || LOOPBACK(daddr))
1936 goto martian_destination;
1937
1938 /*
1939 * Now we are ready to route packet.
1940 */
1941 if ((err = fib_lookup(&fl, &res)) != 0) {
1942 if (!IN_DEV_FORWARD(in_dev))
Dietmar Eggemann2c2910a2005-06-28 13:06:23 -07001943 goto e_hostunreach;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 goto no_route;
1945 }
1946 free_res = 1;
1947
1948 RT_CACHE_STAT_INC(in_slow_tot);
1949
1950 if (res.type == RTN_BROADCAST)
1951 goto brd_input;
1952
1953 if (res.type == RTN_LOCAL) {
1954 int result;
1955 result = fib_validate_source(saddr, daddr, tos,
1956 loopback_dev.ifindex,
1957 dev, &spec_dst, &itag);
1958 if (result < 0)
1959 goto martian_source;
1960 if (result)
1961 flags |= RTCF_DIRECTSRC;
1962 spec_dst = daddr;
1963 goto local_input;
1964 }
1965
1966 if (!IN_DEV_FORWARD(in_dev))
Dietmar Eggemann2c2910a2005-06-28 13:06:23 -07001967 goto e_hostunreach;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 if (res.type != RTN_UNICAST)
1969 goto martian_destination;
1970
1971 err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
1972 if (err == -ENOBUFS)
1973 goto e_nobufs;
1974 if (err == -EINVAL)
1975 goto e_inval;
1976
1977done:
1978 in_dev_put(in_dev);
1979 if (free_res)
1980 fib_res_put(&res);
1981out: return err;
1982
1983brd_input:
1984 if (skb->protocol != htons(ETH_P_IP))
1985 goto e_inval;
1986
1987 if (ZERONET(saddr))
1988 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1989 else {
1990 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1991 &itag);
1992 if (err < 0)
1993 goto martian_source;
1994 if (err)
1995 flags |= RTCF_DIRECTSRC;
1996 }
1997 flags |= RTCF_BROADCAST;
1998 res.type = RTN_BROADCAST;
1999 RT_CACHE_STAT_INC(in_brd);
2000
2001local_input:
2002 rth = dst_alloc(&ipv4_dst_ops);
2003 if (!rth)
2004 goto e_nobufs;
2005
2006 rth->u.dst.output= ip_rt_bug;
2007
2008 atomic_set(&rth->u.dst.__refcnt, 1);
2009 rth->u.dst.flags= DST_HOST;
2010 if (in_dev->cnf.no_policy)
2011 rth->u.dst.flags |= DST_NOPOLICY;
2012 rth->fl.fl4_dst = daddr;
2013 rth->rt_dst = daddr;
2014 rth->fl.fl4_tos = tos;
2015#ifdef CONFIG_IP_ROUTE_FWMARK
2016 rth->fl.fl4_fwmark= skb->nfmark;
2017#endif
2018 rth->fl.fl4_src = saddr;
2019 rth->rt_src = saddr;
2020#ifdef CONFIG_NET_CLS_ROUTE
2021 rth->u.dst.tclassid = itag;
2022#endif
2023 rth->rt_iif =
2024 rth->fl.iif = dev->ifindex;
2025 rth->u.dst.dev = &loopback_dev;
2026 dev_hold(rth->u.dst.dev);
2027 rth->idev = in_dev_get(rth->u.dst.dev);
2028 rth->rt_gateway = daddr;
2029 rth->rt_spec_dst= spec_dst;
2030 rth->u.dst.input= ip_local_deliver;
2031 rth->rt_flags = flags|RTCF_LOCAL;
2032 if (res.type == RTN_UNREACHABLE) {
2033 rth->u.dst.input= ip_error;
2034 rth->u.dst.error= -err;
2035 rth->rt_flags &= ~RTCF_LOCAL;
2036 }
2037 rth->rt_type = res.type;
Ilia Sotnikovcef26852006-03-25 01:38:55 -08002038 hash = rt_hash_code(daddr, saddr ^ (fl.iif << 5));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 err = rt_intern_hash(hash, rth, (struct rtable**)&skb->dst);
2040 goto done;
2041
2042no_route:
2043 RT_CACHE_STAT_INC(in_no_route);
2044 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2045 res.type = RTN_UNREACHABLE;
2046 goto local_input;
2047
2048 /*
2049 * Do not cache martian addresses: they should be logged (RFC1812)
2050 */
2051martian_destination:
2052 RT_CACHE_STAT_INC(in_martian_dst);
2053#ifdef CONFIG_IP_ROUTE_VERBOSE
2054 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2055 printk(KERN_WARNING "martian destination %u.%u.%u.%u from "
2056 "%u.%u.%u.%u, dev %s\n",
2057 NIPQUAD(daddr), NIPQUAD(saddr), dev->name);
2058#endif
Dietmar Eggemann2c2910a2005-06-28 13:06:23 -07002059
2060e_hostunreach:
2061 err = -EHOSTUNREACH;
2062 goto done;
2063
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064e_inval:
2065 err = -EINVAL;
2066 goto done;
2067
2068e_nobufs:
2069 err = -ENOBUFS;
2070 goto done;
2071
2072martian_source:
2073 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2074 goto e_inval;
2075}
2076
2077int ip_route_input(struct sk_buff *skb, u32 daddr, u32 saddr,
2078 u8 tos, struct net_device *dev)
2079{
2080 struct rtable * rth;
2081 unsigned hash;
2082 int iif = dev->ifindex;
2083
2084 tos &= IPTOS_RT_MASK;
Ilia Sotnikovcef26852006-03-25 01:38:55 -08002085 hash = rt_hash_code(daddr, saddr ^ (iif << 5));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086
2087 rcu_read_lock();
2088 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2089 rth = rcu_dereference(rth->u.rt_next)) {
2090 if (rth->fl.fl4_dst == daddr &&
2091 rth->fl.fl4_src == saddr &&
2092 rth->fl.iif == iif &&
2093 rth->fl.oif == 0 &&
2094#ifdef CONFIG_IP_ROUTE_FWMARK
2095 rth->fl.fl4_fwmark == skb->nfmark &&
2096#endif
2097 rth->fl.fl4_tos == tos) {
2098 rth->u.dst.lastuse = jiffies;
2099 dst_hold(&rth->u.dst);
2100 rth->u.dst.__use++;
2101 RT_CACHE_STAT_INC(in_hit);
2102 rcu_read_unlock();
2103 skb->dst = (struct dst_entry*)rth;
2104 return 0;
2105 }
2106 RT_CACHE_STAT_INC(in_hlist_search);
2107 }
2108 rcu_read_unlock();
2109
2110 /* Multicast recognition logic is moved from route cache to here.
2111 The problem was that too many Ethernet cards have broken/missing
2112 hardware multicast filters :-( As result the host on multicasting
2113 network acquires a lot of useless route cache entries, sort of
2114 SDR messages from all the world. Now we try to get rid of them.
2115 Really, provided software IP multicast filter is organized
2116 reasonably (at least, hashed), it does not result in a slowdown
2117 comparing with route cache reject entries.
2118 Note, that multicast routers are not affected, because
2119 route cache entry is created eventually.
2120 */
2121 if (MULTICAST(daddr)) {
2122 struct in_device *in_dev;
2123
2124 rcu_read_lock();
Herbert Xue5ed6392005-10-03 14:35:55 -07002125 if ((in_dev = __in_dev_get_rcu(dev)) != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 int our = ip_check_mc(in_dev, daddr, saddr,
2127 skb->nh.iph->protocol);
2128 if (our
2129#ifdef CONFIG_IP_MROUTE
2130 || (!LOCAL_MCAST(daddr) && IN_DEV_MFORWARD(in_dev))
2131#endif
2132 ) {
2133 rcu_read_unlock();
2134 return ip_route_input_mc(skb, daddr, saddr,
2135 tos, dev, our);
2136 }
2137 }
2138 rcu_read_unlock();
2139 return -EINVAL;
2140 }
2141 return ip_route_input_slow(skb, daddr, saddr, tos, dev);
2142}
2143
2144static inline int __mkroute_output(struct rtable **result,
2145 struct fib_result* res,
2146 const struct flowi *fl,
2147 const struct flowi *oldflp,
2148 struct net_device *dev_out,
2149 unsigned flags)
2150{
2151 struct rtable *rth;
2152 struct in_device *in_dev;
2153 u32 tos = RT_FL_TOS(oldflp);
2154 int err = 0;
2155
2156 if (LOOPBACK(fl->fl4_src) && !(dev_out->flags&IFF_LOOPBACK))
2157 return -EINVAL;
2158
2159 if (fl->fl4_dst == 0xFFFFFFFF)
2160 res->type = RTN_BROADCAST;
2161 else if (MULTICAST(fl->fl4_dst))
2162 res->type = RTN_MULTICAST;
2163 else if (BADCLASS(fl->fl4_dst) || ZERONET(fl->fl4_dst))
2164 return -EINVAL;
2165
2166 if (dev_out->flags & IFF_LOOPBACK)
2167 flags |= RTCF_LOCAL;
2168
2169 /* get work reference to inet device */
2170 in_dev = in_dev_get(dev_out);
2171 if (!in_dev)
2172 return -EINVAL;
2173
2174 if (res->type == RTN_BROADCAST) {
2175 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2176 if (res->fi) {
2177 fib_info_put(res->fi);
2178 res->fi = NULL;
2179 }
2180 } else if (res->type == RTN_MULTICAST) {
2181 flags |= RTCF_MULTICAST|RTCF_LOCAL;
2182 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2183 oldflp->proto))
2184 flags &= ~RTCF_LOCAL;
2185 /* If multicast route do not exist use
2186 default one, but do not gateway in this case.
2187 Yes, it is hack.
2188 */
2189 if (res->fi && res->prefixlen < 4) {
2190 fib_info_put(res->fi);
2191 res->fi = NULL;
2192 }
2193 }
2194
2195
2196 rth = dst_alloc(&ipv4_dst_ops);
2197 if (!rth) {
2198 err = -ENOBUFS;
2199 goto cleanup;
2200 }
2201
Julian Anastasovce723d82005-09-08 13:34:47 -07002202 atomic_set(&rth->u.dst.__refcnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 rth->u.dst.flags= DST_HOST;
2204#ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
2205 if (res->fi) {
2206 rth->rt_multipath_alg = res->fi->fib_mp_alg;
2207 if (res->fi->fib_nhs > 1)
2208 rth->u.dst.flags |= DST_BALANCED;
2209 }
2210#endif
2211 if (in_dev->cnf.no_xfrm)
2212 rth->u.dst.flags |= DST_NOXFRM;
2213 if (in_dev->cnf.no_policy)
2214 rth->u.dst.flags |= DST_NOPOLICY;
2215
2216 rth->fl.fl4_dst = oldflp->fl4_dst;
2217 rth->fl.fl4_tos = tos;
2218 rth->fl.fl4_src = oldflp->fl4_src;
2219 rth->fl.oif = oldflp->oif;
2220#ifdef CONFIG_IP_ROUTE_FWMARK
2221 rth->fl.fl4_fwmark= oldflp->fl4_fwmark;
2222#endif
2223 rth->rt_dst = fl->fl4_dst;
2224 rth->rt_src = fl->fl4_src;
2225 rth->rt_iif = oldflp->oif ? : dev_out->ifindex;
2226 /* get references to the devices that are to be hold by the routing
2227 cache entry */
2228 rth->u.dst.dev = dev_out;
2229 dev_hold(dev_out);
2230 rth->idev = in_dev_get(dev_out);
2231 rth->rt_gateway = fl->fl4_dst;
2232 rth->rt_spec_dst= fl->fl4_src;
2233
2234 rth->u.dst.output=ip_output;
2235
2236 RT_CACHE_STAT_INC(out_slow_tot);
2237
2238 if (flags & RTCF_LOCAL) {
2239 rth->u.dst.input = ip_local_deliver;
2240 rth->rt_spec_dst = fl->fl4_dst;
2241 }
2242 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2243 rth->rt_spec_dst = fl->fl4_src;
2244 if (flags & RTCF_LOCAL &&
2245 !(dev_out->flags & IFF_LOOPBACK)) {
2246 rth->u.dst.output = ip_mc_output;
2247 RT_CACHE_STAT_INC(out_slow_mc);
2248 }
2249#ifdef CONFIG_IP_MROUTE
2250 if (res->type == RTN_MULTICAST) {
2251 if (IN_DEV_MFORWARD(in_dev) &&
2252 !LOCAL_MCAST(oldflp->fl4_dst)) {
2253 rth->u.dst.input = ip_mr_input;
2254 rth->u.dst.output = ip_mc_output;
2255 }
2256 }
2257#endif
2258 }
2259
2260 rt_set_nexthop(rth, res, 0);
2261
2262 rth->rt_flags = flags;
2263
2264 *result = rth;
2265 cleanup:
2266 /* release work reference to inet device */
2267 in_dev_put(in_dev);
2268
2269 return err;
2270}
2271
2272static inline int ip_mkroute_output_def(struct rtable **rp,
2273 struct fib_result* res,
2274 const struct flowi *fl,
2275 const struct flowi *oldflp,
2276 struct net_device *dev_out,
2277 unsigned flags)
2278{
Chuck Short7abaa272005-06-22 22:10:23 -07002279 struct rtable *rth = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2281 unsigned hash;
2282 if (err == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 hash = rt_hash_code(oldflp->fl4_dst,
Ilia Sotnikovcef26852006-03-25 01:38:55 -08002284 oldflp->fl4_src ^ (oldflp->oif << 5));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 err = rt_intern_hash(hash, rth, rp);
2286 }
2287
2288 return err;
2289}
2290
2291static inline int ip_mkroute_output(struct rtable** rp,
2292 struct fib_result* res,
2293 const struct flowi *fl,
2294 const struct flowi *oldflp,
2295 struct net_device *dev_out,
2296 unsigned flags)
2297{
2298#ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 unsigned char hop;
2300 unsigned hash;
2301 int err = -EINVAL;
Chuck Short7abaa272005-06-22 22:10:23 -07002302 struct rtable *rth = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303
2304 if (res->fi && res->fi->fib_nhs > 1) {
2305 unsigned char hopcount = res->fi->fib_nhs;
2306
2307 for (hop = 0; hop < hopcount; hop++) {
2308 struct net_device *dev2nexthop;
2309
2310 res->nh_sel = hop;
2311
2312 /* hold a work reference to the output device */
2313 dev2nexthop = FIB_RES_DEV(*res);
2314 dev_hold(dev2nexthop);
2315
Julian Anastasovce723d82005-09-08 13:34:47 -07002316 /* put reference to previous result */
2317 if (hop)
2318 ip_rt_put(*rp);
2319
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 err = __mkroute_output(&rth, res, fl, oldflp,
2321 dev2nexthop, flags);
2322
2323 if (err != 0)
2324 goto cleanup;
2325
2326 hash = rt_hash_code(oldflp->fl4_dst,
2327 oldflp->fl4_src ^
Ilia Sotnikovcef26852006-03-25 01:38:55 -08002328 (oldflp->oif << 5));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 err = rt_intern_hash(hash, rth, rp);
2330
2331 /* forward hop information to multipath impl. */
2332 multipath_set_nhinfo(rth,
2333 FIB_RES_NETWORK(*res),
2334 FIB_RES_NETMASK(*res),
2335 res->prefixlen,
2336 &FIB_RES_NH(*res));
2337 cleanup:
2338 /* release work reference to output device */
2339 dev_put(dev2nexthop);
2340
2341 if (err != 0)
2342 return err;
2343 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 return err;
2345 } else {
2346 return ip_mkroute_output_def(rp, res, fl, oldflp, dev_out,
2347 flags);
2348 }
2349#else /* CONFIG_IP_ROUTE_MULTIPATH_CACHED */
2350 return ip_mkroute_output_def(rp, res, fl, oldflp, dev_out, flags);
2351#endif
2352}
2353
2354/*
2355 * Major route resolver routine.
2356 */
2357
2358static int ip_route_output_slow(struct rtable **rp, const struct flowi *oldflp)
2359{
2360 u32 tos = RT_FL_TOS(oldflp);
2361 struct flowi fl = { .nl_u = { .ip4_u =
2362 { .daddr = oldflp->fl4_dst,
2363 .saddr = oldflp->fl4_src,
2364 .tos = tos & IPTOS_RT_MASK,
2365 .scope = ((tos & RTO_ONLINK) ?
2366 RT_SCOPE_LINK :
2367 RT_SCOPE_UNIVERSE),
2368#ifdef CONFIG_IP_ROUTE_FWMARK
2369 .fwmark = oldflp->fl4_fwmark
2370#endif
2371 } },
2372 .iif = loopback_dev.ifindex,
2373 .oif = oldflp->oif };
2374 struct fib_result res;
2375 unsigned flags = 0;
2376 struct net_device *dev_out = NULL;
2377 int free_res = 0;
2378 int err;
2379
2380
2381 res.fi = NULL;
2382#ifdef CONFIG_IP_MULTIPLE_TABLES
2383 res.r = NULL;
2384#endif
2385
2386 if (oldflp->fl4_src) {
2387 err = -EINVAL;
2388 if (MULTICAST(oldflp->fl4_src) ||
2389 BADCLASS(oldflp->fl4_src) ||
2390 ZERONET(oldflp->fl4_src))
2391 goto out;
2392
2393 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2394 dev_out = ip_dev_find(oldflp->fl4_src);
2395 if (dev_out == NULL)
2396 goto out;
2397
2398 /* I removed check for oif == dev_out->oif here.
2399 It was wrong for two reasons:
2400 1. ip_dev_find(saddr) can return wrong iface, if saddr is
2401 assigned to multiple interfaces.
2402 2. Moreover, we are allowed to send packets with saddr
2403 of another iface. --ANK
2404 */
2405
2406 if (oldflp->oif == 0
2407 && (MULTICAST(oldflp->fl4_dst) || oldflp->fl4_dst == 0xFFFFFFFF)) {
2408 /* Special hack: user can direct multicasts
2409 and limited broadcast via necessary interface
2410 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2411 This hack is not just for fun, it allows
2412 vic,vat and friends to work.
2413 They bind socket to loopback, set ttl to zero
2414 and expect that it will work.
2415 From the viewpoint of routing cache they are broken,
2416 because we are not allowed to build multicast path
2417 with loopback source addr (look, routing cache
2418 cannot know, that ttl is zero, so that packet
2419 will not leave this host and route is valid).
2420 Luckily, this hack is good workaround.
2421 */
2422
2423 fl.oif = dev_out->ifindex;
2424 goto make_route;
2425 }
2426 if (dev_out)
2427 dev_put(dev_out);
2428 dev_out = NULL;
2429 }
2430
2431
2432 if (oldflp->oif) {
2433 dev_out = dev_get_by_index(oldflp->oif);
2434 err = -ENODEV;
2435 if (dev_out == NULL)
2436 goto out;
Herbert Xue5ed6392005-10-03 14:35:55 -07002437
2438 /* RACE: Check return value of inet_select_addr instead. */
2439 if (__in_dev_get_rtnl(dev_out) == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 dev_put(dev_out);
2441 goto out; /* Wrong error code */
2442 }
2443
2444 if (LOCAL_MCAST(oldflp->fl4_dst) || oldflp->fl4_dst == 0xFFFFFFFF) {
2445 if (!fl.fl4_src)
2446 fl.fl4_src = inet_select_addr(dev_out, 0,
2447 RT_SCOPE_LINK);
2448 goto make_route;
2449 }
2450 if (!fl.fl4_src) {
2451 if (MULTICAST(oldflp->fl4_dst))
2452 fl.fl4_src = inet_select_addr(dev_out, 0,
2453 fl.fl4_scope);
2454 else if (!oldflp->fl4_dst)
2455 fl.fl4_src = inet_select_addr(dev_out, 0,
2456 RT_SCOPE_HOST);
2457 }
2458 }
2459
2460 if (!fl.fl4_dst) {
2461 fl.fl4_dst = fl.fl4_src;
2462 if (!fl.fl4_dst)
2463 fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2464 if (dev_out)
2465 dev_put(dev_out);
2466 dev_out = &loopback_dev;
2467 dev_hold(dev_out);
2468 fl.oif = loopback_dev.ifindex;
2469 res.type = RTN_LOCAL;
2470 flags |= RTCF_LOCAL;
2471 goto make_route;
2472 }
2473
2474 if (fib_lookup(&fl, &res)) {
2475 res.fi = NULL;
2476 if (oldflp->oif) {
2477 /* Apparently, routing tables are wrong. Assume,
2478 that the destination is on link.
2479
2480 WHY? DW.
2481 Because we are allowed to send to iface
2482 even if it has NO routes and NO assigned
2483 addresses. When oif is specified, routing
2484 tables are looked up with only one purpose:
2485 to catch if destination is gatewayed, rather than
2486 direct. Moreover, if MSG_DONTROUTE is set,
2487 we send packet, ignoring both routing tables
2488 and ifaddr state. --ANK
2489
2490
2491 We could make it even if oif is unknown,
2492 likely IPv6, but we do not.
2493 */
2494
2495 if (fl.fl4_src == 0)
2496 fl.fl4_src = inet_select_addr(dev_out, 0,
2497 RT_SCOPE_LINK);
2498 res.type = RTN_UNICAST;
2499 goto make_route;
2500 }
2501 if (dev_out)
2502 dev_put(dev_out);
2503 err = -ENETUNREACH;
2504 goto out;
2505 }
2506 free_res = 1;
2507
2508 if (res.type == RTN_LOCAL) {
2509 if (!fl.fl4_src)
2510 fl.fl4_src = fl.fl4_dst;
2511 if (dev_out)
2512 dev_put(dev_out);
2513 dev_out = &loopback_dev;
2514 dev_hold(dev_out);
2515 fl.oif = dev_out->ifindex;
2516 if (res.fi)
2517 fib_info_put(res.fi);
2518 res.fi = NULL;
2519 flags |= RTCF_LOCAL;
2520 goto make_route;
2521 }
2522
2523#ifdef CONFIG_IP_ROUTE_MULTIPATH
2524 if (res.fi->fib_nhs > 1 && fl.oif == 0)
2525 fib_select_multipath(&fl, &res);
2526 else
2527#endif
2528 if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2529 fib_select_default(&fl, &res);
2530
2531 if (!fl.fl4_src)
2532 fl.fl4_src = FIB_RES_PREFSRC(res);
2533
2534 if (dev_out)
2535 dev_put(dev_out);
2536 dev_out = FIB_RES_DEV(res);
2537 dev_hold(dev_out);
2538 fl.oif = dev_out->ifindex;
2539
2540
2541make_route:
2542 err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2543
2544
2545 if (free_res)
2546 fib_res_put(&res);
2547 if (dev_out)
2548 dev_put(dev_out);
2549out: return err;
2550}
2551
2552int __ip_route_output_key(struct rtable **rp, const struct flowi *flp)
2553{
2554 unsigned hash;
2555 struct rtable *rth;
2556
Ilia Sotnikovcef26852006-03-25 01:38:55 -08002557 hash = rt_hash_code(flp->fl4_dst, flp->fl4_src ^ (flp->oif << 5));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558
2559 rcu_read_lock_bh();
2560 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2561 rth = rcu_dereference(rth->u.rt_next)) {
2562 if (rth->fl.fl4_dst == flp->fl4_dst &&
2563 rth->fl.fl4_src == flp->fl4_src &&
2564 rth->fl.iif == 0 &&
2565 rth->fl.oif == flp->oif &&
2566#ifdef CONFIG_IP_ROUTE_FWMARK
2567 rth->fl.fl4_fwmark == flp->fl4_fwmark &&
2568#endif
2569 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2570 (IPTOS_RT_MASK | RTO_ONLINK))) {
2571
2572 /* check for multipath routes and choose one if
2573 * necessary
2574 */
2575 if (multipath_select_route(flp, rth, rp)) {
2576 dst_hold(&(*rp)->u.dst);
2577 RT_CACHE_STAT_INC(out_hit);
2578 rcu_read_unlock_bh();
2579 return 0;
2580 }
2581
2582 rth->u.dst.lastuse = jiffies;
2583 dst_hold(&rth->u.dst);
2584 rth->u.dst.__use++;
2585 RT_CACHE_STAT_INC(out_hit);
2586 rcu_read_unlock_bh();
2587 *rp = rth;
2588 return 0;
2589 }
2590 RT_CACHE_STAT_INC(out_hlist_search);
2591 }
2592 rcu_read_unlock_bh();
2593
2594 return ip_route_output_slow(rp, flp);
2595}
2596
Arnaldo Carvalho de Melod8c97a92005-08-09 20:12:12 -07002597EXPORT_SYMBOL_GPL(__ip_route_output_key);
2598
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599int ip_route_output_flow(struct rtable **rp, struct flowi *flp, struct sock *sk, int flags)
2600{
2601 int err;
2602
2603 if ((err = __ip_route_output_key(rp, flp)) != 0)
2604 return err;
2605
2606 if (flp->proto) {
2607 if (!flp->fl4_src)
2608 flp->fl4_src = (*rp)->rt_src;
2609 if (!flp->fl4_dst)
2610 flp->fl4_dst = (*rp)->rt_dst;
2611 return xfrm_lookup((struct dst_entry **)rp, flp, sk, flags);
2612 }
2613
2614 return 0;
2615}
2616
Arnaldo Carvalho de Melod8c97a92005-08-09 20:12:12 -07002617EXPORT_SYMBOL_GPL(ip_route_output_flow);
2618
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619int ip_route_output_key(struct rtable **rp, struct flowi *flp)
2620{
2621 return ip_route_output_flow(rp, flp, NULL, 0);
2622}
2623
2624static int rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
Jamal Hadi Salimb6544c02005-06-18 22:54:12 -07002625 int nowait, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626{
2627 struct rtable *rt = (struct rtable*)skb->dst;
2628 struct rtmsg *r;
2629 struct nlmsghdr *nlh;
2630 unsigned char *b = skb->tail;
2631 struct rta_cacheinfo ci;
2632#ifdef CONFIG_IP_MROUTE
2633 struct rtattr *eptr;
2634#endif
Jamal Hadi Salimb6544c02005-06-18 22:54:12 -07002635 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 r = NLMSG_DATA(nlh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 r->rtm_family = AF_INET;
2638 r->rtm_dst_len = 32;
2639 r->rtm_src_len = 0;
2640 r->rtm_tos = rt->fl.fl4_tos;
2641 r->rtm_table = RT_TABLE_MAIN;
2642 r->rtm_type = rt->rt_type;
2643 r->rtm_scope = RT_SCOPE_UNIVERSE;
2644 r->rtm_protocol = RTPROT_UNSPEC;
2645 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2646 if (rt->rt_flags & RTCF_NOTIFY)
2647 r->rtm_flags |= RTM_F_NOTIFY;
2648 RTA_PUT(skb, RTA_DST, 4, &rt->rt_dst);
2649 if (rt->fl.fl4_src) {
2650 r->rtm_src_len = 32;
2651 RTA_PUT(skb, RTA_SRC, 4, &rt->fl.fl4_src);
2652 }
2653 if (rt->u.dst.dev)
2654 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
2655#ifdef CONFIG_NET_CLS_ROUTE
2656 if (rt->u.dst.tclassid)
2657 RTA_PUT(skb, RTA_FLOW, 4, &rt->u.dst.tclassid);
2658#endif
2659#ifdef CONFIG_IP_ROUTE_MULTIPATH_CACHED
2660 if (rt->rt_multipath_alg != IP_MP_ALG_NONE) {
2661 __u32 alg = rt->rt_multipath_alg;
2662
2663 RTA_PUT(skb, RTA_MP_ALGO, 4, &alg);
2664 }
2665#endif
2666 if (rt->fl.iif)
2667 RTA_PUT(skb, RTA_PREFSRC, 4, &rt->rt_spec_dst);
2668 else if (rt->rt_src != rt->fl.fl4_src)
2669 RTA_PUT(skb, RTA_PREFSRC, 4, &rt->rt_src);
2670 if (rt->rt_dst != rt->rt_gateway)
2671 RTA_PUT(skb, RTA_GATEWAY, 4, &rt->rt_gateway);
2672 if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
2673 goto rtattr_failure;
2674 ci.rta_lastuse = jiffies_to_clock_t(jiffies - rt->u.dst.lastuse);
2675 ci.rta_used = rt->u.dst.__use;
2676 ci.rta_clntref = atomic_read(&rt->u.dst.__refcnt);
2677 if (rt->u.dst.expires)
2678 ci.rta_expires = jiffies_to_clock_t(rt->u.dst.expires - jiffies);
2679 else
2680 ci.rta_expires = 0;
2681 ci.rta_error = rt->u.dst.error;
2682 ci.rta_id = ci.rta_ts = ci.rta_tsage = 0;
2683 if (rt->peer) {
2684 ci.rta_id = rt->peer->ip_id_count;
2685 if (rt->peer->tcp_ts_stamp) {
2686 ci.rta_ts = rt->peer->tcp_ts;
2687 ci.rta_tsage = xtime.tv_sec - rt->peer->tcp_ts_stamp;
2688 }
2689 }
2690#ifdef CONFIG_IP_MROUTE
2691 eptr = (struct rtattr*)skb->tail;
2692#endif
2693 RTA_PUT(skb, RTA_CACHEINFO, sizeof(ci), &ci);
2694 if (rt->fl.iif) {
2695#ifdef CONFIG_IP_MROUTE
2696 u32 dst = rt->rt_dst;
2697
2698 if (MULTICAST(dst) && !LOCAL_MCAST(dst) &&
2699 ipv4_devconf.mc_forwarding) {
2700 int err = ipmr_get_route(skb, r, nowait);
2701 if (err <= 0) {
2702 if (!nowait) {
2703 if (err == 0)
2704 return 0;
2705 goto nlmsg_failure;
2706 } else {
2707 if (err == -EMSGSIZE)
2708 goto nlmsg_failure;
2709 ((struct rta_cacheinfo*)RTA_DATA(eptr))->rta_error = err;
2710 }
2711 }
2712 } else
2713#endif
2714 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
2715 }
2716
2717 nlh->nlmsg_len = skb->tail - b;
2718 return skb->len;
2719
2720nlmsg_failure:
2721rtattr_failure:
2722 skb_trim(skb, b - skb->data);
2723 return -1;
2724}
2725
2726int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2727{
2728 struct rtattr **rta = arg;
2729 struct rtmsg *rtm = NLMSG_DATA(nlh);
2730 struct rtable *rt = NULL;
2731 u32 dst = 0;
2732 u32 src = 0;
2733 int iif = 0;
2734 int err = -ENOBUFS;
2735 struct sk_buff *skb;
2736
2737 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2738 if (!skb)
2739 goto out;
2740
2741 /* Reserve room for dummy headers, this skb can pass
2742 through good chunk of routing engine.
2743 */
Stephen Hemmingerd2c962b2006-04-17 17:27:11 -07002744 skb->mac.raw = skb->nh.raw = skb->data;
2745
2746 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2747 skb->nh.iph->protocol = IPPROTO_ICMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2749
2750 if (rta[RTA_SRC - 1])
2751 memcpy(&src, RTA_DATA(rta[RTA_SRC - 1]), 4);
2752 if (rta[RTA_DST - 1])
2753 memcpy(&dst, RTA_DATA(rta[RTA_DST - 1]), 4);
2754 if (rta[RTA_IIF - 1])
2755 memcpy(&iif, RTA_DATA(rta[RTA_IIF - 1]), sizeof(int));
2756
2757 if (iif) {
2758 struct net_device *dev = __dev_get_by_index(iif);
2759 err = -ENODEV;
2760 if (!dev)
2761 goto out_free;
2762 skb->protocol = htons(ETH_P_IP);
2763 skb->dev = dev;
2764 local_bh_disable();
2765 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2766 local_bh_enable();
2767 rt = (struct rtable*)skb->dst;
2768 if (!err && rt->u.dst.error)
2769 err = -rt->u.dst.error;
2770 } else {
2771 struct flowi fl = { .nl_u = { .ip4_u = { .daddr = dst,
2772 .saddr = src,
2773 .tos = rtm->rtm_tos } } };
2774 int oif = 0;
2775 if (rta[RTA_OIF - 1])
2776 memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
2777 fl.oif = oif;
2778 err = ip_route_output_key(&rt, &fl);
2779 }
2780 if (err)
2781 goto out_free;
2782
2783 skb->dst = &rt->u.dst;
2784 if (rtm->rtm_flags & RTM_F_NOTIFY)
2785 rt->rt_flags |= RTCF_NOTIFY;
2786
2787 NETLINK_CB(skb).dst_pid = NETLINK_CB(in_skb).pid;
2788
2789 err = rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
Jamal Hadi Salimb6544c02005-06-18 22:54:12 -07002790 RTM_NEWROUTE, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 if (!err)
2792 goto out_free;
2793 if (err < 0) {
2794 err = -EMSGSIZE;
2795 goto out_free;
2796 }
2797
2798 err = netlink_unicast(rtnl, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
2799 if (err > 0)
2800 err = 0;
2801out: return err;
2802
2803out_free:
2804 kfree_skb(skb);
2805 goto out;
2806}
2807
2808int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2809{
2810 struct rtable *rt;
2811 int h, s_h;
2812 int idx, s_idx;
2813
2814 s_h = cb->args[0];
2815 s_idx = idx = cb->args[1];
2816 for (h = 0; h <= rt_hash_mask; h++) {
2817 if (h < s_h) continue;
2818 if (h > s_h)
2819 s_idx = 0;
2820 rcu_read_lock_bh();
2821 for (rt = rcu_dereference(rt_hash_table[h].chain), idx = 0; rt;
2822 rt = rcu_dereference(rt->u.rt_next), idx++) {
2823 if (idx < s_idx)
2824 continue;
2825 skb->dst = dst_clone(&rt->u.dst);
2826 if (rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
Jamal Hadi Salimb6544c02005-06-18 22:54:12 -07002827 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2828 1, NLM_F_MULTI) <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829 dst_release(xchg(&skb->dst, NULL));
2830 rcu_read_unlock_bh();
2831 goto done;
2832 }
2833 dst_release(xchg(&skb->dst, NULL));
2834 }
2835 rcu_read_unlock_bh();
2836 }
2837
2838done:
2839 cb->args[0] = h;
2840 cb->args[1] = idx;
2841 return skb->len;
2842}
2843
2844void ip_rt_multicast_event(struct in_device *in_dev)
2845{
2846 rt_cache_flush(0);
2847}
2848
2849#ifdef CONFIG_SYSCTL
2850static int flush_delay;
2851
2852static int ipv4_sysctl_rtcache_flush(ctl_table *ctl, int write,
2853 struct file *filp, void __user *buffer,
2854 size_t *lenp, loff_t *ppos)
2855{
2856 if (write) {
2857 proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
2858 rt_cache_flush(flush_delay);
2859 return 0;
2860 }
2861
2862 return -EINVAL;
2863}
2864
2865static int ipv4_sysctl_rtcache_flush_strategy(ctl_table *table,
2866 int __user *name,
2867 int nlen,
2868 void __user *oldval,
2869 size_t __user *oldlenp,
2870 void __user *newval,
2871 size_t newlen,
2872 void **context)
2873{
2874 int delay;
2875 if (newlen != sizeof(int))
2876 return -EINVAL;
2877 if (get_user(delay, (int __user *)newval))
2878 return -EFAULT;
2879 rt_cache_flush(delay);
2880 return 0;
2881}
2882
2883ctl_table ipv4_route_table[] = {
2884 {
2885 .ctl_name = NET_IPV4_ROUTE_FLUSH,
2886 .procname = "flush",
2887 .data = &flush_delay,
2888 .maxlen = sizeof(int),
Dave Jones7e3e0362005-04-28 12:11:03 -07002889 .mode = 0200,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890 .proc_handler = &ipv4_sysctl_rtcache_flush,
2891 .strategy = &ipv4_sysctl_rtcache_flush_strategy,
2892 },
2893 {
2894 .ctl_name = NET_IPV4_ROUTE_MIN_DELAY,
2895 .procname = "min_delay",
2896 .data = &ip_rt_min_delay,
2897 .maxlen = sizeof(int),
2898 .mode = 0644,
2899 .proc_handler = &proc_dointvec_jiffies,
2900 .strategy = &sysctl_jiffies,
2901 },
2902 {
2903 .ctl_name = NET_IPV4_ROUTE_MAX_DELAY,
2904 .procname = "max_delay",
2905 .data = &ip_rt_max_delay,
2906 .maxlen = sizeof(int),
2907 .mode = 0644,
2908 .proc_handler = &proc_dointvec_jiffies,
2909 .strategy = &sysctl_jiffies,
2910 },
2911 {
2912 .ctl_name = NET_IPV4_ROUTE_GC_THRESH,
2913 .procname = "gc_thresh",
2914 .data = &ipv4_dst_ops.gc_thresh,
2915 .maxlen = sizeof(int),
2916 .mode = 0644,
2917 .proc_handler = &proc_dointvec,
2918 },
2919 {
2920 .ctl_name = NET_IPV4_ROUTE_MAX_SIZE,
2921 .procname = "max_size",
2922 .data = &ip_rt_max_size,
2923 .maxlen = sizeof(int),
2924 .mode = 0644,
2925 .proc_handler = &proc_dointvec,
2926 },
2927 {
2928 /* Deprecated. Use gc_min_interval_ms */
2929
2930 .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL,
2931 .procname = "gc_min_interval",
2932 .data = &ip_rt_gc_min_interval,
2933 .maxlen = sizeof(int),
2934 .mode = 0644,
2935 .proc_handler = &proc_dointvec_jiffies,
2936 .strategy = &sysctl_jiffies,
2937 },
2938 {
2939 .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL_MS,
2940 .procname = "gc_min_interval_ms",
2941 .data = &ip_rt_gc_min_interval,
2942 .maxlen = sizeof(int),
2943 .mode = 0644,
2944 .proc_handler = &proc_dointvec_ms_jiffies,
2945 .strategy = &sysctl_ms_jiffies,
2946 },
2947 {
2948 .ctl_name = NET_IPV4_ROUTE_GC_TIMEOUT,
2949 .procname = "gc_timeout",
2950 .data = &ip_rt_gc_timeout,
2951 .maxlen = sizeof(int),
2952 .mode = 0644,
2953 .proc_handler = &proc_dointvec_jiffies,
2954 .strategy = &sysctl_jiffies,
2955 },
2956 {
2957 .ctl_name = NET_IPV4_ROUTE_GC_INTERVAL,
2958 .procname = "gc_interval",
2959 .data = &ip_rt_gc_interval,
2960 .maxlen = sizeof(int),
2961 .mode = 0644,
2962 .proc_handler = &proc_dointvec_jiffies,
2963 .strategy = &sysctl_jiffies,
2964 },
2965 {
2966 .ctl_name = NET_IPV4_ROUTE_REDIRECT_LOAD,
2967 .procname = "redirect_load",
2968 .data = &ip_rt_redirect_load,
2969 .maxlen = sizeof(int),
2970 .mode = 0644,
2971 .proc_handler = &proc_dointvec,
2972 },
2973 {
2974 .ctl_name = NET_IPV4_ROUTE_REDIRECT_NUMBER,
2975 .procname = "redirect_number",
2976 .data = &ip_rt_redirect_number,
2977 .maxlen = sizeof(int),
2978 .mode = 0644,
2979 .proc_handler = &proc_dointvec,
2980 },
2981 {
2982 .ctl_name = NET_IPV4_ROUTE_REDIRECT_SILENCE,
2983 .procname = "redirect_silence",
2984 .data = &ip_rt_redirect_silence,
2985 .maxlen = sizeof(int),
2986 .mode = 0644,
2987 .proc_handler = &proc_dointvec,
2988 },
2989 {
2990 .ctl_name = NET_IPV4_ROUTE_ERROR_COST,
2991 .procname = "error_cost",
2992 .data = &ip_rt_error_cost,
2993 .maxlen = sizeof(int),
2994 .mode = 0644,
2995 .proc_handler = &proc_dointvec,
2996 },
2997 {
2998 .ctl_name = NET_IPV4_ROUTE_ERROR_BURST,
2999 .procname = "error_burst",
3000 .data = &ip_rt_error_burst,
3001 .maxlen = sizeof(int),
3002 .mode = 0644,
3003 .proc_handler = &proc_dointvec,
3004 },
3005 {
3006 .ctl_name = NET_IPV4_ROUTE_GC_ELASTICITY,
3007 .procname = "gc_elasticity",
3008 .data = &ip_rt_gc_elasticity,
3009 .maxlen = sizeof(int),
3010 .mode = 0644,
3011 .proc_handler = &proc_dointvec,
3012 },
3013 {
3014 .ctl_name = NET_IPV4_ROUTE_MTU_EXPIRES,
3015 .procname = "mtu_expires",
3016 .data = &ip_rt_mtu_expires,
3017 .maxlen = sizeof(int),
3018 .mode = 0644,
3019 .proc_handler = &proc_dointvec_jiffies,
3020 .strategy = &sysctl_jiffies,
3021 },
3022 {
3023 .ctl_name = NET_IPV4_ROUTE_MIN_PMTU,
3024 .procname = "min_pmtu",
3025 .data = &ip_rt_min_pmtu,
3026 .maxlen = sizeof(int),
3027 .mode = 0644,
3028 .proc_handler = &proc_dointvec,
3029 },
3030 {
3031 .ctl_name = NET_IPV4_ROUTE_MIN_ADVMSS,
3032 .procname = "min_adv_mss",
3033 .data = &ip_rt_min_advmss,
3034 .maxlen = sizeof(int),
3035 .mode = 0644,
3036 .proc_handler = &proc_dointvec,
3037 },
3038 {
3039 .ctl_name = NET_IPV4_ROUTE_SECRET_INTERVAL,
3040 .procname = "secret_interval",
3041 .data = &ip_rt_secret_interval,
3042 .maxlen = sizeof(int),
3043 .mode = 0644,
3044 .proc_handler = &proc_dointvec_jiffies,
3045 .strategy = &sysctl_jiffies,
3046 },
3047 { .ctl_name = 0 }
3048};
3049#endif
3050
3051#ifdef CONFIG_NET_CLS_ROUTE
3052struct ip_rt_acct *ip_rt_acct;
3053
3054/* This code sucks. But you should have seen it before! --RR */
3055
3056/* IP route accounting ptr for this logical cpu number. */
3057#define IP_RT_ACCT_CPU(i) (ip_rt_acct + i * 256)
3058
3059#ifdef CONFIG_PROC_FS
3060static int ip_rt_acct_read(char *buffer, char **start, off_t offset,
3061 int length, int *eof, void *data)
3062{
3063 unsigned int i;
3064
3065 if ((offset & 3) || (length & 3))
3066 return -EIO;
3067
3068 if (offset >= sizeof(struct ip_rt_acct) * 256) {
3069 *eof = 1;
3070 return 0;
3071 }
3072
3073 if (offset + length >= sizeof(struct ip_rt_acct) * 256) {
3074 length = sizeof(struct ip_rt_acct) * 256 - offset;
3075 *eof = 1;
3076 }
3077
3078 offset /= sizeof(u32);
3079
3080 if (length > 0) {
3081 u32 *src = ((u32 *) IP_RT_ACCT_CPU(0)) + offset;
3082 u32 *dst = (u32 *) buffer;
3083
3084 /* Copy first cpu. */
3085 *start = buffer;
3086 memcpy(dst, src, length);
3087
3088 /* Add the other cpus in, one int at a time */
KAMEZAWA Hiroyuki6f912042006-04-10 22:52:50 -07003089 for_each_possible_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 unsigned int j;
3091
3092 src = ((u32 *) IP_RT_ACCT_CPU(i)) + offset;
3093
3094 for (j = 0; j < length/4; j++)
3095 dst[j] += src[j];
3096 }
3097 }
3098 return length;
3099}
3100#endif /* CONFIG_PROC_FS */
3101#endif /* CONFIG_NET_CLS_ROUTE */
3102
3103static __initdata unsigned long rhash_entries;
3104static int __init set_rhash_entries(char *str)
3105{
3106 if (!str)
3107 return 0;
3108 rhash_entries = simple_strtoul(str, &str, 0);
3109 return 1;
3110}
3111__setup("rhash_entries=", set_rhash_entries);
3112
3113int __init ip_rt_init(void)
3114{
Eric Dumazet424c4b72005-07-05 14:58:19 -07003115 int rc = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003116
3117 rt_hash_rnd = (int) ((num_physpages ^ (num_physpages>>8)) ^
3118 (jiffies ^ (jiffies >> 7)));
3119
3120#ifdef CONFIG_NET_CLS_ROUTE
Eric Dumazet424c4b72005-07-05 14:58:19 -07003121 {
3122 int order;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123 for (order = 0;
3124 (PAGE_SIZE << order) < 256 * sizeof(struct ip_rt_acct) * NR_CPUS; order++)
3125 /* NOTHING */;
3126 ip_rt_acct = (struct ip_rt_acct *)__get_free_pages(GFP_KERNEL, order);
3127 if (!ip_rt_acct)
3128 panic("IP: failed to allocate ip_rt_acct\n");
3129 memset(ip_rt_acct, 0, PAGE_SIZE << order);
Eric Dumazet424c4b72005-07-05 14:58:19 -07003130 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131#endif
3132
3133 ipv4_dst_ops.kmem_cachep = kmem_cache_create("ip_dst_cache",
3134 sizeof(struct rtable),
3135 0, SLAB_HWCACHE_ALIGN,
3136 NULL, NULL);
3137
3138 if (!ipv4_dst_ops.kmem_cachep)
3139 panic("IP: failed to allocate ip_dst_cache\n");
3140
Eric Dumazet424c4b72005-07-05 14:58:19 -07003141 rt_hash_table = (struct rt_hash_bucket *)
3142 alloc_large_system_hash("IP route cache",
3143 sizeof(struct rt_hash_bucket),
3144 rhash_entries,
3145 (num_physpages >= 128 * 1024) ?
Mike Stroyan18955cf2005-11-29 16:12:55 -08003146 15 : 17,
Eric Dumazet424c4b72005-07-05 14:58:19 -07003147 HASH_HIGHMEM,
3148 &rt_hash_log,
3149 &rt_hash_mask,
3150 0);
Eric Dumazet22c047c2005-07-05 14:55:24 -07003151 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3152 rt_hash_lock_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153
3154 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3155 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3156
Linus Torvalds1da177e2005-04-16 15:20:36 -07003157 devinet_init();
3158 ip_fib_init();
3159
3160 init_timer(&rt_flush_timer);
3161 rt_flush_timer.function = rt_run_flush;
3162 init_timer(&rt_periodic_timer);
3163 rt_periodic_timer.function = rt_check_expire;
3164 init_timer(&rt_secret_timer);
3165 rt_secret_timer.function = rt_secret_rebuild;
3166
3167 /* All the timers, started at system startup tend
3168 to synchronize. Perturb it a bit.
3169 */
3170 rt_periodic_timer.expires = jiffies + net_random() % ip_rt_gc_interval +
3171 ip_rt_gc_interval;
3172 add_timer(&rt_periodic_timer);
3173
3174 rt_secret_timer.expires = jiffies + net_random() % ip_rt_secret_interval +
3175 ip_rt_secret_interval;
3176 add_timer(&rt_secret_timer);
3177
3178#ifdef CONFIG_PROC_FS
3179 {
3180 struct proc_dir_entry *rtstat_pde = NULL; /* keep gcc happy */
3181 if (!proc_net_fops_create("rt_cache", S_IRUGO, &rt_cache_seq_fops) ||
3182 !(rtstat_pde = create_proc_entry("rt_cache", S_IRUGO,
3183 proc_net_stat))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 return -ENOMEM;
3185 }
3186 rtstat_pde->proc_fops = &rt_cpu_seq_fops;
3187 }
3188#ifdef CONFIG_NET_CLS_ROUTE
3189 create_proc_read_entry("rt_acct", 0, proc_net, ip_rt_acct_read, NULL);
3190#endif
3191#endif
3192#ifdef CONFIG_XFRM
3193 xfrm_init();
3194 xfrm4_init();
3195#endif
3196 return rc;
3197}
3198
3199EXPORT_SYMBOL(__ip_select_ident);
3200EXPORT_SYMBOL(ip_route_input);
3201EXPORT_SYMBOL(ip_route_output_key);