]> bbs.cooldavid.org Git - net-next-2.6.git/blame_incremental - net/ipv4/route.c
git://bbs.cooldavid.org / net-next-2.6.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Linux 2.6.27-rc2
[net-next-2.6.git] / net / ipv4 / route.c
... / ...
CommitLineData
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * ROUTE - implementation of the IP router.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
12 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
13 *
14 * Fixes:
15 * Alan Cox : Verify area fixes.
16 * Alan Cox : cli() protects routing changes
17 * Rui Oliveira : ICMP routing table updates
18 * (rco@di.uminho.pt) Routing table insertion and update
19 * Linus Torvalds : Rewrote bits to be sensible
20 * Alan Cox : Added BSD route gw semantics
21 * Alan Cox : Super /proc >4K
22 * Alan Cox : MTU in route table
23 * Alan Cox : MSS actually. Also added the window
24 * clamper.
25 * Sam Lantinga : Fixed route matching in rt_del()
26 * Alan Cox : Routing cache support.
27 * Alan Cox : Removed compatibility cruft.
28 * Alan Cox : RTF_REJECT support.
29 * Alan Cox : TCP irtt support.
30 * Jonathan Naylor : Added Metric support.
31 * Miquel van Smoorenburg : BSD API fixes.
32 * Miquel van Smoorenburg : Metrics.
33 * Alan Cox : Use __u32 properly
34 * Alan Cox : Aligned routing errors more closely with BSD
35 * our system is still very different.
36 * Alan Cox : Faster /proc handling
37 * Alexey Kuznetsov : Massive rework to support tree based routing,
38 * routing caches and better behaviour.
39 *
40 * Olaf Erb : irtt wasn't being copied right.
41 * Bjorn Ekwall : Kerneld route support.
42 * Alan Cox : Multicast fixed (I hope)
43 * Pavel Krauz : Limited broadcast fixed
44 * Mike McLagan : Routing by source
45 * Alexey Kuznetsov : End of old history. Split to fib.c and
46 * route.c and rewritten from scratch.
47 * Andi Kleen : Load-limit warning messages.
48 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
49 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
50 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
51 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
52 * Marc Boucher : routing by fwmark
53 * Robert Olsson : Added rt_cache statistics
54 * Arnaldo C. Melo : Convert proc stuff to seq_file
55 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
56 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
57 * Ilia Sotnikov : Removed TOS from hash calculations
58 *
59 * This program is free software; you can redistribute it and/or
60 * modify it under the terms of the GNU General Public License
61 * as published by the Free Software Foundation; either version
62 * 2 of the License, or (at your option) any later version.
63 */
64
65#include <linux/module.h>
66#include <asm/uaccess.h>
67#include <asm/system.h>
68#include <linux/bitops.h>
69#include <linux/types.h>
70#include <linux/kernel.h>
71#include <linux/mm.h>
72#include <linux/bootmem.h>
73#include <linux/string.h>
74#include <linux/socket.h>
75#include <linux/sockios.h>
76#include <linux/errno.h>
77#include <linux/in.h>
78#include <linux/inet.h>
79#include <linux/netdevice.h>
80#include <linux/proc_fs.h>
81#include <linux/init.h>
82#include <linux/workqueue.h>
83#include <linux/skbuff.h>
84#include <linux/inetdevice.h>
85#include <linux/igmp.h>
86#include <linux/pkt_sched.h>
87#include <linux/mroute.h>
88#include <linux/netfilter_ipv4.h>
89#include <linux/random.h>
90#include <linux/jhash.h>
91#include <linux/rcupdate.h>
92#include <linux/times.h>
93#include <net/dst.h>
94#include <net/net_namespace.h>
95#include <net/protocol.h>
96#include <net/ip.h>
97#include <net/route.h>
98#include <net/inetpeer.h>
99#include <net/sock.h>
100#include <net/ip_fib.h>
101#include <net/arp.h>
102#include <net/tcp.h>
103#include <net/icmp.h>
104#include <net/xfrm.h>
105#include <net/netevent.h>
106#include <net/rtnetlink.h>
107#ifdef CONFIG_SYSCTL
108#include <linux/sysctl.h>
109#endif
110
111#define RT_FL_TOS(oldflp) \
112 ((u32)(oldflp->fl4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
113
114#define IP_MAX_MTU 0xFFF0
115
116#define RT_GC_TIMEOUT (300*HZ)
117
118static int ip_rt_max_size;
119static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
120static int ip_rt_gc_interval __read_mostly = 60 * HZ;
121static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
122static int ip_rt_redirect_number __read_mostly = 9;
123static int ip_rt_redirect_load __read_mostly = HZ / 50;
124static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
125static int ip_rt_error_cost __read_mostly = HZ;
126static int ip_rt_error_burst __read_mostly = 5 * HZ;
127static int ip_rt_gc_elasticity __read_mostly = 8;
128static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
129static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
130static int ip_rt_min_advmss __read_mostly = 256;
131static int ip_rt_secret_interval __read_mostly = 10 * 60 * HZ;
132
133static void rt_worker_func(struct work_struct *work);
134static DECLARE_DELAYED_WORK(expires_work, rt_worker_func);
135
136/*
137 * Interface to generic destination cache.
138 */
139
140static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
141static void ipv4_dst_destroy(struct dst_entry *dst);
142static void ipv4_dst_ifdown(struct dst_entry *dst,
143 struct net_device *dev, int how);
144static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
145static void ipv4_link_failure(struct sk_buff *skb);
146static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
147static int rt_garbage_collect(struct dst_ops *ops);
148
149
150static struct dst_ops ipv4_dst_ops = {
151 .family = AF_INET,
152 .protocol = __constant_htons(ETH_P_IP),
153 .gc = rt_garbage_collect,
154 .check = ipv4_dst_check,
155 .destroy = ipv4_dst_destroy,
156 .ifdown = ipv4_dst_ifdown,
157 .negative_advice = ipv4_negative_advice,
158 .link_failure = ipv4_link_failure,
159 .update_pmtu = ip_rt_update_pmtu,
160 .local_out = __ip_local_out,
161 .entry_size = sizeof(struct rtable),
162 .entries = ATOMIC_INIT(0),
163};
164
165#define ECN_OR_COST(class) TC_PRIO_##class
166
167const __u8 ip_tos2prio[16] = {
168 TC_PRIO_BESTEFFORT,
169 ECN_OR_COST(FILLER),
170 TC_PRIO_BESTEFFORT,
171 ECN_OR_COST(BESTEFFORT),
172 TC_PRIO_BULK,
173 ECN_OR_COST(BULK),
174 TC_PRIO_BULK,
175 ECN_OR_COST(BULK),
176 TC_PRIO_INTERACTIVE,
177 ECN_OR_COST(INTERACTIVE),
178 TC_PRIO_INTERACTIVE,
179 ECN_OR_COST(INTERACTIVE),
180 TC_PRIO_INTERACTIVE_BULK,
181 ECN_OR_COST(INTERACTIVE_BULK),
182 TC_PRIO_INTERACTIVE_BULK,
183 ECN_OR_COST(INTERACTIVE_BULK)
184};
185
186
187/*
188 * Route cache.
189 */
190
191/* The locking scheme is rather straight forward:
192 *
193 * 1) Read-Copy Update protects the buckets of the central route hash.
194 * 2) Only writers remove entries, and they hold the lock
195 * as they look at rtable reference counts.
196 * 3) Only readers acquire references to rtable entries,
197 * they do so with atomic increments and with the
198 * lock held.
199 */
200
201struct rt_hash_bucket {
202 struct rtable *chain;
203};
204#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
205 defined(CONFIG_PROVE_LOCKING)
206/*
207 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
208 * The size of this table is a power of two and depends on the number of CPUS.
209 * (on lockdep we have a quite big spinlock_t, so keep the size down there)
210 */
211#ifdef CONFIG_LOCKDEP
212# define RT_HASH_LOCK_SZ 256
213#else
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#endif
226
227static spinlock_t *rt_hash_locks;
228# define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
229
230static __init void rt_hash_lock_init(void)
231{
232 int i;
233
234 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
235 GFP_KERNEL);
236 if (!rt_hash_locks)
237 panic("IP: failed to allocate rt_hash_locks\n");
238
239 for (i = 0; i < RT_HASH_LOCK_SZ; i++)
240 spin_lock_init(&rt_hash_locks[i]);
241}
242#else
243# define rt_hash_lock_addr(slot) NULL
244
245static inline void rt_hash_lock_init(void)
246{
247}
248#endif
249
250static struct rt_hash_bucket *rt_hash_table __read_mostly;
251static unsigned rt_hash_mask __read_mostly;
252static unsigned int rt_hash_log __read_mostly;
253
254static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
255#define RT_CACHE_STAT_INC(field) \
256 (__raw_get_cpu_var(rt_cache_stat).field++)
257
258static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
259 int genid)
260{
261 return jhash_3words((__force u32)(__be32)(daddr),
262 (__force u32)(__be32)(saddr),
263 idx, genid)
264 & rt_hash_mask;
265}
266
267static inline int rt_genid(struct net *net)
268{
269 return atomic_read(&net->ipv4.rt_genid);
270}
271
272#ifdef CONFIG_PROC_FS
273struct rt_cache_iter_state {
274 struct seq_net_private p;
275 int bucket;
276 int genid;
277};
278
279static struct rtable *rt_cache_get_first(struct seq_file *seq)
280{
281 struct rt_cache_iter_state *st = seq->private;
282 struct rtable *r = NULL;
283
284 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
285 rcu_read_lock_bh();
286 r = rcu_dereference(rt_hash_table[st->bucket].chain);
287 while (r) {
288 if (dev_net(r->u.dst.dev) == seq_file_net(seq) &&
289 r->rt_genid == st->genid)
290 return r;
291 r = rcu_dereference(r->u.dst.rt_next);
292 }
293 rcu_read_unlock_bh();
294 }
295 return r;
296}
297
298static struct rtable *__rt_cache_get_next(struct seq_file *seq,
299 struct rtable *r)
300{
301 struct rt_cache_iter_state *st = seq->private;
302 r = r->u.dst.rt_next;
303 while (!r) {
304 rcu_read_unlock_bh();
305 if (--st->bucket < 0)
306 break;
307 rcu_read_lock_bh();
308 r = rt_hash_table[st->bucket].chain;
309 }
310 return rcu_dereference(r);
311}
312
313static struct rtable *rt_cache_get_next(struct seq_file *seq,
314 struct rtable *r)
315{
316 struct rt_cache_iter_state *st = seq->private;
317 while ((r = __rt_cache_get_next(seq, r)) != NULL) {
318 if (dev_net(r->u.dst.dev) != seq_file_net(seq))
319 continue;
320 if (r->rt_genid == st->genid)
321 break;
322 }
323 return r;
324}
325
326static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
327{
328 struct rtable *r = rt_cache_get_first(seq);
329
330 if (r)
331 while (pos && (r = rt_cache_get_next(seq, r)))
332 --pos;
333 return pos ? NULL : r;
334}
335
336static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
337{
338 struct rt_cache_iter_state *st = seq->private;
339 if (*pos)
340 return rt_cache_get_idx(seq, *pos - 1);
341 st->genid = rt_genid(seq_file_net(seq));
342 return SEQ_START_TOKEN;
343}
344
345static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
346{
347 struct rtable *r;
348
349 if (v == SEQ_START_TOKEN)
350 r = rt_cache_get_first(seq);
351 else
352 r = rt_cache_get_next(seq, v);
353 ++*pos;
354 return r;
355}
356
357static void rt_cache_seq_stop(struct seq_file *seq, void *v)
358{
359 if (v && v != SEQ_START_TOKEN)
360 rcu_read_unlock_bh();
361}
362
363static int rt_cache_seq_show(struct seq_file *seq, void *v)
364{
365 if (v == SEQ_START_TOKEN)
366 seq_printf(seq, "%-127s\n",
367 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
368 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
369 "HHUptod\tSpecDst");
370 else {
371 struct rtable *r = v;
372 int len;
373
374 seq_printf(seq, "%s\t%08lX\t%08lX\t%8X\t%d\t%u\t%d\t"
375 "%08lX\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
376 r->u.dst.dev ? r->u.dst.dev->name : "*",
377 (unsigned long)r->rt_dst, (unsigned long)r->rt_gateway,
378 r->rt_flags, atomic_read(&r->u.dst.__refcnt),
379 r->u.dst.__use, 0, (unsigned long)r->rt_src,
380 (dst_metric(&r->u.dst, RTAX_ADVMSS) ?
381 (int)dst_metric(&r->u.dst, RTAX_ADVMSS) + 40 : 0),
382 dst_metric(&r->u.dst, RTAX_WINDOW),
383 (int)((dst_metric(&r->u.dst, RTAX_RTT) >> 3) +
384 dst_metric(&r->u.dst, RTAX_RTTVAR)),
385 r->fl.fl4_tos,
386 r->u.dst.hh ? atomic_read(&r->u.dst.hh->hh_refcnt) : -1,
387 r->u.dst.hh ? (r->u.dst.hh->hh_output ==
388 dev_queue_xmit) : 0,
389 r->rt_spec_dst, &len);
390
391 seq_printf(seq, "%*s\n", 127 - len, "");
392 }
393 return 0;
394}
395
396static const struct seq_operations rt_cache_seq_ops = {
397 .start = rt_cache_seq_start,
398 .next = rt_cache_seq_next,
399 .stop = rt_cache_seq_stop,
400 .show = rt_cache_seq_show,
401};
402
403static int rt_cache_seq_open(struct inode *inode, struct file *file)
404{
405 return seq_open_net(inode, file, &rt_cache_seq_ops,
406 sizeof(struct rt_cache_iter_state));
407}
408
409static const struct file_operations rt_cache_seq_fops = {
410 .owner = THIS_MODULE,
411 .open = rt_cache_seq_open,
412 .read = seq_read,
413 .llseek = seq_lseek,
414 .release = seq_release_net,
415};
416
417
418static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
419{
420 int cpu;
421
422 if (*pos == 0)
423 return SEQ_START_TOKEN;
424
425 for (cpu = *pos-1; cpu < NR_CPUS; ++cpu) {
426 if (!cpu_possible(cpu))
427 continue;
428 *pos = cpu+1;
429 return &per_cpu(rt_cache_stat, cpu);
430 }
431 return NULL;
432}
433
434static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
435{
436 int cpu;
437
438 for (cpu = *pos; cpu < NR_CPUS; ++cpu) {
439 if (!cpu_possible(cpu))
440 continue;
441 *pos = cpu+1;
442 return &per_cpu(rt_cache_stat, cpu);
443 }
444 return NULL;
445
446}
447
448static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
449{
450
451}
452
453static int rt_cpu_seq_show(struct seq_file *seq, void *v)
454{
455 struct rt_cache_stat *st = v;
456
457 if (v == SEQ_START_TOKEN) {
458 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");
459 return 0;
460 }
461
462 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
463 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
464 atomic_read(&ipv4_dst_ops.entries),
465 st->in_hit,
466 st->in_slow_tot,
467 st->in_slow_mc,
468 st->in_no_route,
469 st->in_brd,
470 st->in_martian_dst,
471 st->in_martian_src,
472
473 st->out_hit,
474 st->out_slow_tot,
475 st->out_slow_mc,
476
477 st->gc_total,
478 st->gc_ignored,
479 st->gc_goal_miss,
480 st->gc_dst_overflow,
481 st->in_hlist_search,
482 st->out_hlist_search
483 );
484 return 0;
485}
486
487static const struct seq_operations rt_cpu_seq_ops = {
488 .start = rt_cpu_seq_start,
489 .next = rt_cpu_seq_next,
490 .stop = rt_cpu_seq_stop,
491 .show = rt_cpu_seq_show,
492};
493
494
495static int rt_cpu_seq_open(struct inode *inode, struct file *file)
496{
497 return seq_open(file, &rt_cpu_seq_ops);
498}
499
500static const struct file_operations rt_cpu_seq_fops = {
501 .owner = THIS_MODULE,
502 .open = rt_cpu_seq_open,
503 .read = seq_read,
504 .llseek = seq_lseek,
505 .release = seq_release,
506};
507
508#ifdef CONFIG_NET_CLS_ROUTE
509static int ip_rt_acct_read(char *buffer, char **start, off_t offset,
510 int length, int *eof, void *data)
511{
512 unsigned int i;
513
514 if ((offset & 3) || (length & 3))
515 return -EIO;
516
517 if (offset >= sizeof(struct ip_rt_acct) * 256) {
518 *eof = 1;
519 return 0;
520 }
521
522 if (offset + length >= sizeof(struct ip_rt_acct) * 256) {
523 length = sizeof(struct ip_rt_acct) * 256 - offset;
524 *eof = 1;
525 }
526
527 offset /= sizeof(u32);
528
529 if (length > 0) {
530 u32 *dst = (u32 *) buffer;
531
532 *start = buffer;
533 memset(dst, 0, length);
534
535 for_each_possible_cpu(i) {
536 unsigned int j;
537 u32 *src;
538
539 src = ((u32 *) per_cpu_ptr(ip_rt_acct, i)) + offset;
540 for (j = 0; j < length/4; j++)
541 dst[j] += src[j];
542 }
543 }
544 return length;
545}
546#endif
547
548static int __net_init ip_rt_do_proc_init(struct net *net)
549{
550 struct proc_dir_entry *pde;
551
552 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
553 &rt_cache_seq_fops);
554 if (!pde)
555 goto err1;
556
557 pde = proc_create("rt_cache", S_IRUGO,
558 net->proc_net_stat, &rt_cpu_seq_fops);
559 if (!pde)
560 goto err2;
561
562#ifdef CONFIG_NET_CLS_ROUTE
563 pde = create_proc_read_entry("rt_acct", 0, net->proc_net,
564 ip_rt_acct_read, NULL);
565 if (!pde)
566 goto err3;
567#endif
568 return 0;
569
570#ifdef CONFIG_NET_CLS_ROUTE
571err3:
572 remove_proc_entry("rt_cache", net->proc_net_stat);
573#endif
574err2:
575 remove_proc_entry("rt_cache", net->proc_net);
576err1:
577 return -ENOMEM;
578}
579
580static void __net_exit ip_rt_do_proc_exit(struct net *net)
581{
582 remove_proc_entry("rt_cache", net->proc_net_stat);
583 remove_proc_entry("rt_cache", net->proc_net);
584 remove_proc_entry("rt_acct", net->proc_net);
585}
586
587static struct pernet_operations ip_rt_proc_ops __net_initdata = {
588 .init = ip_rt_do_proc_init,
589 .exit = ip_rt_do_proc_exit,
590};
591
592static int __init ip_rt_proc_init(void)
593{
594 return register_pernet_subsys(&ip_rt_proc_ops);
595}
596
597#else
598static inline int ip_rt_proc_init(void)
599{
600 return 0;
601}
602#endif /* CONFIG_PROC_FS */
603
604static inline void rt_free(struct rtable *rt)
605{
606 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
607}
608
609static inline void rt_drop(struct rtable *rt)
610{
611 ip_rt_put(rt);
612 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
613}
614
615static inline int rt_fast_clean(struct rtable *rth)
616{
617 /* Kill broadcast/multicast entries very aggresively, if they
618 collide in hash table with more useful entries */
619 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
620 rth->fl.iif && rth->u.dst.rt_next;
621}
622
623static inline int rt_valuable(struct rtable *rth)
624{
625 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
626 rth->u.dst.expires;
627}
628
629static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
630{
631 unsigned long age;
632 int ret = 0;
633
634 if (atomic_read(&rth->u.dst.__refcnt))
635 goto out;
636
637 ret = 1;
638 if (rth->u.dst.expires &&
639 time_after_eq(jiffies, rth->u.dst.expires))
640 goto out;
641
642 age = jiffies - rth->u.dst.lastuse;
643 ret = 0;
644 if ((age <= tmo1 && !rt_fast_clean(rth)) ||
645 (age <= tmo2 && rt_valuable(rth)))
646 goto out;
647 ret = 1;
648out: return ret;
649}
650
651/* Bits of score are:
652 * 31: very valuable
653 * 30: not quite useless
654 * 29..0: usage counter
655 */
656static inline u32 rt_score(struct rtable *rt)
657{
658 u32 score = jiffies - rt->u.dst.lastuse;
659
660 score = ~score & ~(3<<30);
661
662 if (rt_valuable(rt))
663 score |= (1<<31);
664
665 if (!rt->fl.iif ||
666 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
667 score |= (1<<30);
668
669 return score;
670}
671
672static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
673{
674 return ((__force u32)((fl1->nl_u.ip4_u.daddr ^ fl2->nl_u.ip4_u.daddr) |
675 (fl1->nl_u.ip4_u.saddr ^ fl2->nl_u.ip4_u.saddr)) |
676 (fl1->mark ^ fl2->mark) |
677 (*(u16 *)&fl1->nl_u.ip4_u.tos ^
678 *(u16 *)&fl2->nl_u.ip4_u.tos) |
679 (fl1->oif ^ fl2->oif) |
680 (fl1->iif ^ fl2->iif)) == 0;
681}
682
683static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
684{
685 return dev_net(rt1->u.dst.dev) == dev_net(rt2->u.dst.dev);
686}
687
688static inline int rt_is_expired(struct rtable *rth)
689{
690 return rth->rt_genid != rt_genid(dev_net(rth->u.dst.dev));
691}
692
693/*
694 * Perform a full scan of hash table and free all entries.
695 * Can be called by a softirq or a process.
696 * In the later case, we want to be reschedule if necessary
697 */
698static void rt_do_flush(int process_context)
699{
700 unsigned int i;
701 struct rtable *rth, *next;
702 struct rtable * tail;
703
704 for (i = 0; i <= rt_hash_mask; i++) {
705 if (process_context && need_resched())
706 cond_resched();
707 rth = rt_hash_table[i].chain;
708 if (!rth)
709 continue;
710
711 spin_lock_bh(rt_hash_lock_addr(i));
712#ifdef CONFIG_NET_NS
713 {
714 struct rtable ** prev, * p;
715
716 rth = rt_hash_table[i].chain;
717
718 /* defer releasing the head of the list after spin_unlock */
719 for (tail = rth; tail; tail = tail->u.dst.rt_next)
720 if (!rt_is_expired(tail))
721 break;
722 if (rth != tail)
723 rt_hash_table[i].chain = tail;
724
725 /* call rt_free on entries after the tail requiring flush */
726 prev = &rt_hash_table[i].chain;
727 for (p = *prev; p; p = next) {
728 next = p->u.dst.rt_next;
729 if (!rt_is_expired(p)) {
730 prev = &p->u.dst.rt_next;
731 } else {
732 *prev = next;
733 rt_free(p);
734 }
735 }
736 }
737#else
738 rth = rt_hash_table[i].chain;
739 rt_hash_table[i].chain = NULL;
740 tail = NULL;
741#endif
742 spin_unlock_bh(rt_hash_lock_addr(i));
743
744 for (; rth != tail; rth = next) {
745 next = rth->u.dst.rt_next;
746 rt_free(rth);
747 }
748 }
749}
750
751static void rt_check_expire(void)
752{
753 static unsigned int rover;
754 unsigned int i = rover, goal;
755 struct rtable *rth, **rthp;
756 u64 mult;
757
758 mult = ((u64)ip_rt_gc_interval) << rt_hash_log;
759 if (ip_rt_gc_timeout > 1)
760 do_div(mult, ip_rt_gc_timeout);
761 goal = (unsigned int)mult;
762 if (goal > rt_hash_mask)
763 goal = rt_hash_mask + 1;
764 for (; goal > 0; goal--) {
765 unsigned long tmo = ip_rt_gc_timeout;
766
767 i = (i + 1) & rt_hash_mask;
768 rthp = &rt_hash_table[i].chain;
769
770 if (need_resched())
771 cond_resched();
772
773 if (*rthp == NULL)
774 continue;
775 spin_lock_bh(rt_hash_lock_addr(i));
776 while ((rth = *rthp) != NULL) {
777 if (rt_is_expired(rth)) {
778 *rthp = rth->u.dst.rt_next;
779 rt_free(rth);
780 continue;
781 }
782 if (rth->u.dst.expires) {
783 /* Entry is expired even if it is in use */
784 if (time_before_eq(jiffies, rth->u.dst.expires)) {
785 tmo >>= 1;
786 rthp = &rth->u.dst.rt_next;
787 continue;
788 }
789 } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout)) {
790 tmo >>= 1;
791 rthp = &rth->u.dst.rt_next;
792 continue;
793 }
794
795 /* Cleanup aged off entries. */
796 *rthp = rth->u.dst.rt_next;
797 rt_free(rth);
798 }
799 spin_unlock_bh(rt_hash_lock_addr(i));
800 }
801 rover = i;
802}
803
804/*
805 * rt_worker_func() is run in process context.
806 * we call rt_check_expire() to scan part of the hash table
807 */
808static void rt_worker_func(struct work_struct *work)
809{
810 rt_check_expire();
811 schedule_delayed_work(&expires_work, ip_rt_gc_interval);
812}
813
814/*
815 * Pertubation of rt_genid by a small quantity [1..256]
816 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
817 * many times (2^24) without giving recent rt_genid.
818 * Jenkins hash is strong enough that litle changes of rt_genid are OK.
819 */
820static void rt_cache_invalidate(struct net *net)
821{
822 unsigned char shuffle;
823
824 get_random_bytes(&shuffle, sizeof(shuffle));
825 atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
826}
827
828/*
829 * delay < 0 : invalidate cache (fast : entries will be deleted later)
830 * delay >= 0 : invalidate & flush cache (can be long)
831 */
832void rt_cache_flush(struct net *net, int delay)
833{
834 rt_cache_invalidate(net);
835 if (delay >= 0)
836 rt_do_flush(!in_softirq());
837}
838
839/*
840 * We change rt_genid and let gc do the cleanup
841 */
842static void rt_secret_rebuild(unsigned long __net)
843{
844 struct net *net = (struct net *)__net;
845 rt_cache_invalidate(net);
846 mod_timer(&net->ipv4.rt_secret_timer, jiffies + ip_rt_secret_interval);
847}
848
849/*
850 Short description of GC goals.
851
852 We want to build algorithm, which will keep routing cache
853 at some equilibrium point, when number of aged off entries
854 is kept approximately equal to newly generated ones.
855
856 Current expiration strength is variable "expire".
857 We try to adjust it dynamically, so that if networking
858 is idle expires is large enough to keep enough of warm entries,
859 and when load increases it reduces to limit cache size.
860 */
861
862static int rt_garbage_collect(struct dst_ops *ops)
863{
864 static unsigned long expire = RT_GC_TIMEOUT;
865 static unsigned long last_gc;
866 static int rover;
867 static int equilibrium;
868 struct rtable *rth, **rthp;
869 unsigned long now = jiffies;
870 int goal;
871
872 /*
873 * Garbage collection is pretty expensive,
874 * do not make it too frequently.
875 */
876
877 RT_CACHE_STAT_INC(gc_total);
878
879 if (now - last_gc < ip_rt_gc_min_interval &&
880 atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) {
881 RT_CACHE_STAT_INC(gc_ignored);
882 goto out;
883 }
884
885 /* Calculate number of entries, which we want to expire now. */
886 goal = atomic_read(&ipv4_dst_ops.entries) -
887 (ip_rt_gc_elasticity << rt_hash_log);
888 if (goal <= 0) {
889 if (equilibrium < ipv4_dst_ops.gc_thresh)
890 equilibrium = ipv4_dst_ops.gc_thresh;
891 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
892 if (goal > 0) {
893 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
894 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
895 }
896 } else {
897 /* We are in dangerous area. Try to reduce cache really
898 * aggressively.
899 */
900 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
901 equilibrium = atomic_read(&ipv4_dst_ops.entries) - goal;
902 }
903
904 if (now - last_gc >= ip_rt_gc_min_interval)
905 last_gc = now;
906
907 if (goal <= 0) {
908 equilibrium += goal;
909 goto work_done;
910 }
911
912 do {
913 int i, k;
914
915 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
916 unsigned long tmo = expire;
917
918 k = (k + 1) & rt_hash_mask;
919 rthp = &rt_hash_table[k].chain;
920 spin_lock_bh(rt_hash_lock_addr(k));
921 while ((rth = *rthp) != NULL) {
922 if (!rt_is_expired(rth) &&
923 !rt_may_expire(rth, tmo, expire)) {
924 tmo >>= 1;
925 rthp = &rth->u.dst.rt_next;
926 continue;
927 }
928 *rthp = rth->u.dst.rt_next;
929 rt_free(rth);
930 goal--;
931 }
932 spin_unlock_bh(rt_hash_lock_addr(k));
933 if (goal <= 0)
934 break;
935 }
936 rover = k;
937
938 if (goal <= 0)
939 goto work_done;
940
941 /* Goal is not achieved. We stop process if:
942
943 - if expire reduced to zero. Otherwise, expire is halfed.
944 - if table is not full.
945 - if we are called from interrupt.
946 - jiffies check is just fallback/debug loop breaker.
947 We will not spin here for long time in any case.
948 */
949
950 RT_CACHE_STAT_INC(gc_goal_miss);
951
952 if (expire == 0)
953 break;
954
955 expire >>= 1;
956#if RT_CACHE_DEBUG >= 2
957 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
958 atomic_read(&ipv4_dst_ops.entries), goal, i);
959#endif
960
961 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
962 goto out;
963 } while (!in_softirq() && time_before_eq(jiffies, now));
964
965 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
966 goto out;
967 if (net_ratelimit())
968 printk(KERN_WARNING "dst cache overflow\n");
969 RT_CACHE_STAT_INC(gc_dst_overflow);
970 return 1;
971
972work_done:
973 expire += ip_rt_gc_min_interval;
974 if (expire > ip_rt_gc_timeout ||
975 atomic_read(&ipv4_dst_ops.entries) < ipv4_dst_ops.gc_thresh)
976 expire = ip_rt_gc_timeout;
977#if RT_CACHE_DEBUG >= 2
978 printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
979 atomic_read(&ipv4_dst_ops.entries), goal, rover);
980#endif
981out: return 0;
982}
983
984static int rt_intern_hash(unsigned hash, struct rtable *rt, struct rtable **rp)
985{
986 struct rtable *rth, **rthp;
987 unsigned long now;
988 struct rtable *cand, **candp;
989 u32 min_score;
990 int chain_length;
991 int attempts = !in_softirq();
992
993restart:
994 chain_length = 0;
995 min_score = ~(u32)0;
996 cand = NULL;
997 candp = NULL;
998 now = jiffies;
999
1000 rthp = &rt_hash_table[hash].chain;
1001
1002 spin_lock_bh(rt_hash_lock_addr(hash));
1003 while ((rth = *rthp) != NULL) {
1004 if (rt_is_expired(rth)) {
1005 *rthp = rth->u.dst.rt_next;
1006 rt_free(rth);
1007 continue;
1008 }
1009 if (compare_keys(&rth->fl, &rt->fl) && compare_netns(rth, rt)) {
1010 /* Put it first */
1011 *rthp = rth->u.dst.rt_next;
1012 /*
1013 * Since lookup is lockfree, the deletion
1014 * must be visible to another weakly ordered CPU before
1015 * the insertion at the start of the hash chain.
1016 */
1017 rcu_assign_pointer(rth->u.dst.rt_next,
1018 rt_hash_table[hash].chain);
1019 /*
1020 * Since lookup is lockfree, the update writes
1021 * must be ordered for consistency on SMP.
1022 */
1023 rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1024
1025 dst_use(&rth->u.dst, now);
1026 spin_unlock_bh(rt_hash_lock_addr(hash));
1027
1028 rt_drop(rt);
1029 *rp = rth;
1030 return 0;
1031 }
1032
1033 if (!atomic_read(&rth->u.dst.__refcnt)) {
1034 u32 score = rt_score(rth);
1035
1036 if (score <= min_score) {
1037 cand = rth;
1038 candp = rthp;
1039 min_score = score;
1040 }
1041 }
1042
1043 chain_length++;
1044
1045 rthp = &rth->u.dst.rt_next;
1046 }
1047
1048 if (cand) {
1049 /* ip_rt_gc_elasticity used to be average length of chain
1050 * length, when exceeded gc becomes really aggressive.
1051 *
1052 * The second limit is less certain. At the moment it allows
1053 * only 2 entries per bucket. We will see.
1054 */
1055 if (chain_length > ip_rt_gc_elasticity) {
1056 *candp = cand->u.dst.rt_next;
1057 rt_free(cand);
1058 }
1059 }
1060
1061 /* Try to bind route to arp only if it is output
1062 route or unicast forwarding path.
1063 */
1064 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1065 int err = arp_bind_neighbour(&rt->u.dst);
1066 if (err) {
1067 spin_unlock_bh(rt_hash_lock_addr(hash));
1068
1069 if (err != -ENOBUFS) {
1070 rt_drop(rt);
1071 return err;
1072 }
1073
1074 /* Neighbour tables are full and nothing
1075 can be released. Try to shrink route cache,
1076 it is most likely it holds some neighbour records.
1077 */
1078 if (attempts-- > 0) {
1079 int saved_elasticity = ip_rt_gc_elasticity;
1080 int saved_int = ip_rt_gc_min_interval;
1081 ip_rt_gc_elasticity = 1;
1082 ip_rt_gc_min_interval = 0;
1083 rt_garbage_collect(&ipv4_dst_ops);
1084 ip_rt_gc_min_interval = saved_int;
1085 ip_rt_gc_elasticity = saved_elasticity;
1086 goto restart;
1087 }
1088
1089 if (net_ratelimit())
1090 printk(KERN_WARNING "Neighbour table overflow.\n");
1091 rt_drop(rt);
1092 return -ENOBUFS;
1093 }
1094 }
1095
1096 rt->u.dst.rt_next = rt_hash_table[hash].chain;
1097#if RT_CACHE_DEBUG >= 2
1098 if (rt->u.dst.rt_next) {
1099 struct rtable *trt;
1100 printk(KERN_DEBUG "rt_cache @%02x: " NIPQUAD_FMT, hash,
1101 NIPQUAD(rt->rt_dst));
1102 for (trt = rt->u.dst.rt_next; trt; trt = trt->u.dst.rt_next)
1103 printk(" . " NIPQUAD_FMT, NIPQUAD(trt->rt_dst));
1104 printk("\n");
1105 }
1106#endif
1107 rt_hash_table[hash].chain = rt;
1108 spin_unlock_bh(rt_hash_lock_addr(hash));
1109 *rp = rt;
1110 return 0;
1111}
1112
1113void rt_bind_peer(struct rtable *rt, int create)
1114{
1115 static DEFINE_SPINLOCK(rt_peer_lock);
1116 struct inet_peer *peer;
1117
1118 peer = inet_getpeer(rt->rt_dst, create);
1119
1120 spin_lock_bh(&rt_peer_lock);
1121 if (rt->peer == NULL) {
1122 rt->peer = peer;
1123 peer = NULL;
1124 }
1125 spin_unlock_bh(&rt_peer_lock);
1126 if (peer)
1127 inet_putpeer(peer);
1128}
1129
1130/*
1131 * Peer allocation may fail only in serious out-of-memory conditions. However
1132 * we still can generate some output.
1133 * Random ID selection looks a bit dangerous because we have no chances to
1134 * select ID being unique in a reasonable period of time.
1135 * But broken packet identifier may be better than no packet at all.
1136 */
1137static void ip_select_fb_ident(struct iphdr *iph)
1138{
1139 static DEFINE_SPINLOCK(ip_fb_id_lock);
1140 static u32 ip_fallback_id;
1141 u32 salt;
1142
1143 spin_lock_bh(&ip_fb_id_lock);
1144 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1145 iph->id = htons(salt & 0xFFFF);
1146 ip_fallback_id = salt;
1147 spin_unlock_bh(&ip_fb_id_lock);
1148}
1149
1150void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1151{
1152 struct rtable *rt = (struct rtable *) dst;
1153
1154 if (rt) {
1155 if (rt->peer == NULL)
1156 rt_bind_peer(rt, 1);
1157
1158 /* If peer is attached to destination, it is never detached,
1159 so that we need not to grab a lock to dereference it.
1160 */
1161 if (rt->peer) {
1162 iph->id = htons(inet_getid(rt->peer, more));
1163 return;
1164 }
1165 } else
1166 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1167 __builtin_return_address(0));
1168
1169 ip_select_fb_ident(iph);
1170}
1171
1172static void rt_del(unsigned hash, struct rtable *rt)
1173{
1174 struct rtable **rthp, *aux;
1175
1176 rthp = &rt_hash_table[hash].chain;
1177 spin_lock_bh(rt_hash_lock_addr(hash));
1178 ip_rt_put(rt);
1179 while ((aux = *rthp) != NULL) {
1180 if (aux == rt || rt_is_expired(aux)) {
1181 *rthp = aux->u.dst.rt_next;
1182 rt_free(aux);
1183 continue;
1184 }
1185 rthp = &aux->u.dst.rt_next;
1186 }
1187 spin_unlock_bh(rt_hash_lock_addr(hash));
1188}
1189
1190void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1191 __be32 saddr, struct net_device *dev)
1192{
1193 int i, k;
1194 struct in_device *in_dev = in_dev_get(dev);
1195 struct rtable *rth, **rthp;
1196 __be32 skeys[2] = { saddr, 0 };
1197 int ikeys[2] = { dev->ifindex, 0 };
1198 struct netevent_redirect netevent;
1199 struct net *net;
1200
1201 if (!in_dev)
1202 return;
1203
1204 net = dev_net(dev);
1205 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev)
1206 || ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw)
1207 || ipv4_is_zeronet(new_gw))
1208 goto reject_redirect;
1209
1210 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1211 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1212 goto reject_redirect;
1213 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1214 goto reject_redirect;
1215 } else {
1216 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1217 goto reject_redirect;
1218 }
1219
1220 for (i = 0; i < 2; i++) {
1221 for (k = 0; k < 2; k++) {
1222 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1223 rt_genid(net));
1224
1225 rthp=&rt_hash_table[hash].chain;
1226
1227 rcu_read_lock();
1228 while ((rth = rcu_dereference(*rthp)) != NULL) {
1229 struct rtable *rt;
1230
1231 if (rth->fl.fl4_dst != daddr ||
1232 rth->fl.fl4_src != skeys[i] ||
1233 rth->fl.oif != ikeys[k] ||
1234 rth->fl.iif != 0 ||
1235 rt_is_expired(rth) ||
1236 !net_eq(dev_net(rth->u.dst.dev), net)) {
1237 rthp = &rth->u.dst.rt_next;
1238 continue;
1239 }
1240
1241 if (rth->rt_dst != daddr ||
1242 rth->rt_src != saddr ||
1243 rth->u.dst.error ||
1244 rth->rt_gateway != old_gw ||
1245 rth->u.dst.dev != dev)
1246 break;
1247
1248 dst_hold(&rth->u.dst);
1249 rcu_read_unlock();
1250
1251 rt = dst_alloc(&ipv4_dst_ops);
1252 if (rt == NULL) {
1253 ip_rt_put(rth);
1254 in_dev_put(in_dev);
1255 return;
1256 }
1257
1258 /* Copy all the information. */
1259 *rt = *rth;
1260 INIT_RCU_HEAD(&rt->u.dst.rcu_head);
1261 rt->u.dst.__use = 1;
1262 atomic_set(&rt->u.dst.__refcnt, 1);
1263 rt->u.dst.child = NULL;
1264 if (rt->u.dst.dev)
1265 dev_hold(rt->u.dst.dev);
1266 if (rt->idev)
1267 in_dev_hold(rt->idev);
1268 rt->u.dst.obsolete = 0;
1269 rt->u.dst.lastuse = jiffies;
1270 rt->u.dst.path = &rt->u.dst;
1271 rt->u.dst.neighbour = NULL;
1272 rt->u.dst.hh = NULL;
1273 rt->u.dst.xfrm = NULL;
1274 rt->rt_genid = rt_genid(net);
1275 rt->rt_flags |= RTCF_REDIRECTED;
1276
1277 /* Gateway is different ... */
1278 rt->rt_gateway = new_gw;
1279
1280 /* Redirect received -> path was valid */
1281 dst_confirm(&rth->u.dst);
1282
1283 if (rt->peer)
1284 atomic_inc(&rt->peer->refcnt);
1285
1286 if (arp_bind_neighbour(&rt->u.dst) ||
1287 !(rt->u.dst.neighbour->nud_state &
1288 NUD_VALID)) {
1289 if (rt->u.dst.neighbour)
1290 neigh_event_send(rt->u.dst.neighbour, NULL);
1291 ip_rt_put(rth);
1292 rt_drop(rt);
1293 goto do_next;
1294 }
1295
1296 netevent.old = &rth->u.dst;
1297 netevent.new = &rt->u.dst;
1298 call_netevent_notifiers(NETEVENT_REDIRECT,
1299 &netevent);
1300
1301 rt_del(hash, rth);
1302 if (!rt_intern_hash(hash, rt, &rt))
1303 ip_rt_put(rt);
1304 goto do_next;
1305 }
1306 rcu_read_unlock();
1307 do_next:
1308 ;
1309 }
1310 }
1311 in_dev_put(in_dev);
1312 return;
1313
1314reject_redirect:
1315#ifdef CONFIG_IP_ROUTE_VERBOSE
1316 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1317 printk(KERN_INFO "Redirect from " NIPQUAD_FMT " on %s about "
1318 NIPQUAD_FMT " ignored.\n"
1319 " Advised path = " NIPQUAD_FMT " -> " NIPQUAD_FMT "\n",
1320 NIPQUAD(old_gw), dev->name, NIPQUAD(new_gw),
1321 NIPQUAD(saddr), NIPQUAD(daddr));
1322#endif
1323 in_dev_put(in_dev);
1324}
1325
1326static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1327{
1328 struct rtable *rt = (struct rtable *)dst;
1329 struct dst_entry *ret = dst;
1330
1331 if (rt) {
1332 if (dst->obsolete) {
1333 ip_rt_put(rt);
1334 ret = NULL;
1335 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1336 rt->u.dst.expires) {
1337 unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1338 rt->fl.oif,
1339 rt_genid(dev_net(dst->dev)));
1340#if RT_CACHE_DEBUG >= 1
1341 printk(KERN_DEBUG "ipv4_negative_advice: redirect to "
1342 NIPQUAD_FMT "/%02x dropped\n",
1343 NIPQUAD(rt->rt_dst), rt->fl.fl4_tos);
1344#endif
1345 rt_del(hash, rt);
1346 ret = NULL;
1347 }
1348 }
1349 return ret;
1350}
1351
1352/*
1353 * Algorithm:
1354 * 1. The first ip_rt_redirect_number redirects are sent
1355 * with exponential backoff, then we stop sending them at all,
1356 * assuming that the host ignores our redirects.
1357 * 2. If we did not see packets requiring redirects
1358 * during ip_rt_redirect_silence, we assume that the host
1359 * forgot redirected route and start to send redirects again.
1360 *
1361 * This algorithm is much cheaper and more intelligent than dumb load limiting
1362 * in icmp.c.
1363 *
1364 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1365 * and "frag. need" (breaks PMTU discovery) in icmp.c.
1366 */
1367
1368void ip_rt_send_redirect(struct sk_buff *skb)
1369{
1370 struct rtable *rt = skb->rtable;
1371 struct in_device *in_dev = in_dev_get(rt->u.dst.dev);
1372
1373 if (!in_dev)
1374 return;
1375
1376 if (!IN_DEV_TX_REDIRECTS(in_dev))
1377 goto out;
1378
1379 /* No redirected packets during ip_rt_redirect_silence;
1380 * reset the algorithm.
1381 */
1382 if (time_after(jiffies, rt->u.dst.rate_last + ip_rt_redirect_silence))
1383 rt->u.dst.rate_tokens = 0;
1384
1385 /* Too many ignored redirects; do not send anything
1386 * set u.dst.rate_last to the last seen redirected packet.
1387 */
1388 if (rt->u.dst.rate_tokens >= ip_rt_redirect_number) {
1389 rt->u.dst.rate_last = jiffies;
1390 goto out;
1391 }
1392
1393 /* Check for load limit; set rate_last to the latest sent
1394 * redirect.
1395 */
1396 if (rt->u.dst.rate_tokens == 0 ||
1397 time_after(jiffies,
1398 (rt->u.dst.rate_last +
1399 (ip_rt_redirect_load << rt->u.dst.rate_tokens)))) {
1400 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1401 rt->u.dst.rate_last = jiffies;
1402 ++rt->u.dst.rate_tokens;
1403#ifdef CONFIG_IP_ROUTE_VERBOSE
1404 if (IN_DEV_LOG_MARTIANS(in_dev) &&
1405 rt->u.dst.rate_tokens == ip_rt_redirect_number &&
1406 net_ratelimit())
1407 printk(KERN_WARNING "host " NIPQUAD_FMT "/if%d ignores "
1408 "redirects for " NIPQUAD_FMT " to " NIPQUAD_FMT ".\n",
1409 NIPQUAD(rt->rt_src), rt->rt_iif,
1410 NIPQUAD(rt->rt_dst), NIPQUAD(rt->rt_gateway));
1411#endif
1412 }
1413out:
1414 in_dev_put(in_dev);
1415}
1416
1417static int ip_error(struct sk_buff *skb)
1418{
1419 struct rtable *rt = skb->rtable;
1420 unsigned long now;
1421 int code;
1422
1423 switch (rt->u.dst.error) {
1424 case EINVAL:
1425 default:
1426 goto out;
1427 case EHOSTUNREACH:
1428 code = ICMP_HOST_UNREACH;
1429 break;
1430 case ENETUNREACH:
1431 code = ICMP_NET_UNREACH;
1432 IP_INC_STATS_BH(dev_net(rt->u.dst.dev),
1433 IPSTATS_MIB_INNOROUTES);
1434 break;
1435 case EACCES:
1436 code = ICMP_PKT_FILTERED;
1437 break;
1438 }
1439
1440 now = jiffies;
1441 rt->u.dst.rate_tokens += now - rt->u.dst.rate_last;
1442 if (rt->u.dst.rate_tokens > ip_rt_error_burst)
1443 rt->u.dst.rate_tokens = ip_rt_error_burst;
1444 rt->u.dst.rate_last = now;
1445 if (rt->u.dst.rate_tokens >= ip_rt_error_cost) {
1446 rt->u.dst.rate_tokens -= ip_rt_error_cost;
1447 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1448 }
1449
1450out: kfree_skb(skb);
1451 return 0;
1452}
1453
1454/*
1455 * The last two values are not from the RFC but
1456 * are needed for AMPRnet AX.25 paths.
1457 */
1458
1459static const unsigned short mtu_plateau[] =
1460{32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1461
1462static inline unsigned short guess_mtu(unsigned short old_mtu)
1463{
1464 int i;
1465
1466 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1467 if (old_mtu > mtu_plateau[i])
1468 return mtu_plateau[i];
1469 return 68;
1470}
1471
1472unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1473 unsigned short new_mtu,
1474 struct net_device *dev)
1475{
1476 int i, k;
1477 unsigned short old_mtu = ntohs(iph->tot_len);
1478 struct rtable *rth;
1479 int ikeys[2] = { dev->ifindex, 0 };
1480 __be32 skeys[2] = { iph->saddr, 0, };
1481 __be32 daddr = iph->daddr;
1482 unsigned short est_mtu = 0;
1483
1484 if (ipv4_config.no_pmtu_disc)
1485 return 0;
1486
1487 for (k = 0; k < 2; k++) {
1488 for (i = 0; i < 2; i++) {
1489 unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1490 rt_genid(net));
1491
1492 rcu_read_lock();
1493 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1494 rth = rcu_dereference(rth->u.dst.rt_next)) {
1495 unsigned short mtu = new_mtu;
1496
1497 if (rth->fl.fl4_dst != daddr ||
1498 rth->fl.fl4_src != skeys[i] ||
1499 rth->rt_dst != daddr ||
1500 rth->rt_src != iph->saddr ||
1501 rth->fl.oif != ikeys[k] ||
1502 rth->fl.iif != 0 ||
1503 dst_metric_locked(&rth->u.dst, RTAX_MTU) ||
1504 !net_eq(dev_net(rth->u.dst.dev), net) ||
1505 rt_is_expired(rth))
1506 continue;
1507
1508 if (new_mtu < 68 || new_mtu >= old_mtu) {
1509
1510 /* BSD 4.2 compatibility hack :-( */
1511 if (mtu == 0 &&
1512 old_mtu >= dst_metric(&rth->u.dst, RTAX_MTU) &&
1513 old_mtu >= 68 + (iph->ihl << 2))
1514 old_mtu -= iph->ihl << 2;
1515
1516 mtu = guess_mtu(old_mtu);
1517 }
1518 if (mtu <= dst_metric(&rth->u.dst, RTAX_MTU)) {
1519 if (mtu < dst_metric(&rth->u.dst, RTAX_MTU)) {
1520 dst_confirm(&rth->u.dst);
1521 if (mtu < ip_rt_min_pmtu) {
1522 mtu = ip_rt_min_pmtu;
1523 rth->u.dst.metrics[RTAX_LOCK-1] |=
1524 (1 << RTAX_MTU);
1525 }
1526 rth->u.dst.metrics[RTAX_MTU-1] = mtu;
1527 dst_set_expires(&rth->u.dst,
1528 ip_rt_mtu_expires);
1529 }
1530 est_mtu = mtu;
1531 }
1532 }
1533 rcu_read_unlock();
1534 }
1535 }
1536 return est_mtu ? : new_mtu;
1537}
1538
1539static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1540{
1541 if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= 68 &&
1542 !(dst_metric_locked(dst, RTAX_MTU))) {
1543 if (mtu < ip_rt_min_pmtu) {
1544 mtu = ip_rt_min_pmtu;
1545 dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1546 }
1547 dst->metrics[RTAX_MTU-1] = mtu;
1548 dst_set_expires(dst, ip_rt_mtu_expires);
1549 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1550 }
1551}
1552
1553static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1554{
1555 return NULL;
1556}
1557
1558static void ipv4_dst_destroy(struct dst_entry *dst)
1559{
1560 struct rtable *rt = (struct rtable *) dst;
1561 struct inet_peer *peer = rt->peer;
1562 struct in_device *idev = rt->idev;
1563
1564 if (peer) {
1565 rt->peer = NULL;
1566 inet_putpeer(peer);
1567 }
1568
1569 if (idev) {
1570 rt->idev = NULL;
1571 in_dev_put(idev);
1572 }
1573}
1574
1575static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1576 int how)
1577{
1578 struct rtable *rt = (struct rtable *) dst;
1579 struct in_device *idev = rt->idev;
1580 if (dev != dev_net(dev)->loopback_dev && idev && idev->dev == dev) {
1581 struct in_device *loopback_idev =
1582 in_dev_get(dev_net(dev)->loopback_dev);
1583 if (loopback_idev) {
1584 rt->idev = loopback_idev;
1585 in_dev_put(idev);
1586 }
1587 }
1588}
1589
1590static void ipv4_link_failure(struct sk_buff *skb)
1591{
1592 struct rtable *rt;
1593
1594 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1595
1596 rt = skb->rtable;
1597 if (rt)
1598 dst_set_expires(&rt->u.dst, 0);
1599}
1600
1601static int ip_rt_bug(struct sk_buff *skb)
1602{
1603 printk(KERN_DEBUG "ip_rt_bug: " NIPQUAD_FMT " -> " NIPQUAD_FMT ", %s\n",
1604 NIPQUAD(ip_hdr(skb)->saddr), NIPQUAD(ip_hdr(skb)->daddr),
1605 skb->dev ? skb->dev->name : "?");
1606 kfree_skb(skb);
1607 return 0;
1608}
1609
1610/*
1611 We do not cache source address of outgoing interface,
1612 because it is used only by IP RR, TS and SRR options,
1613 so that it out of fast path.
1614
1615 BTW remember: "addr" is allowed to be not aligned
1616 in IP options!
1617 */
1618
1619void ip_rt_get_source(u8 *addr, struct rtable *rt)
1620{
1621 __be32 src;
1622 struct fib_result res;
1623
1624 if (rt->fl.iif == 0)
1625 src = rt->rt_src;
1626 else if (fib_lookup(dev_net(rt->u.dst.dev), &rt->fl, &res) == 0) {
1627 src = FIB_RES_PREFSRC(res);
1628 fib_res_put(&res);
1629 } else
1630 src = inet_select_addr(rt->u.dst.dev, rt->rt_gateway,
1631 RT_SCOPE_UNIVERSE);
1632 memcpy(addr, &src, 4);
1633}
1634
1635#ifdef CONFIG_NET_CLS_ROUTE
1636static void set_class_tag(struct rtable *rt, u32 tag)
1637{
1638 if (!(rt->u.dst.tclassid & 0xFFFF))
1639 rt->u.dst.tclassid |= tag & 0xFFFF;
1640 if (!(rt->u.dst.tclassid & 0xFFFF0000))
1641 rt->u.dst.tclassid |= tag & 0xFFFF0000;
1642}
1643#endif
1644
1645static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1646{
1647 struct fib_info *fi = res->fi;
1648
1649 if (fi) {
1650 if (FIB_RES_GW(*res) &&
1651 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1652 rt->rt_gateway = FIB_RES_GW(*res);
1653 memcpy(rt->u.dst.metrics, fi->fib_metrics,
1654 sizeof(rt->u.dst.metrics));
1655 if (fi->fib_mtu == 0) {
1656 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
1657 if (dst_metric_locked(&rt->u.dst, RTAX_MTU) &&
1658 rt->rt_gateway != rt->rt_dst &&
1659 rt->u.dst.dev->mtu > 576)
1660 rt->u.dst.metrics[RTAX_MTU-1] = 576;
1661 }
1662#ifdef CONFIG_NET_CLS_ROUTE
1663 rt->u.dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1664#endif
1665 } else
1666 rt->u.dst.metrics[RTAX_MTU-1]= rt->u.dst.dev->mtu;
1667
1668 if (dst_metric(&rt->u.dst, RTAX_HOPLIMIT) == 0)
1669 rt->u.dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1670 if (dst_metric(&rt->u.dst, RTAX_MTU) > IP_MAX_MTU)
1671 rt->u.dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1672 if (dst_metric(&rt->u.dst, RTAX_ADVMSS) == 0)
1673 rt->u.dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->u.dst.dev->mtu - 40,
1674 ip_rt_min_advmss);
1675 if (dst_metric(&rt->u.dst, RTAX_ADVMSS) > 65535 - 40)
1676 rt->u.dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1677
1678#ifdef CONFIG_NET_CLS_ROUTE
1679#ifdef CONFIG_IP_MULTIPLE_TABLES
1680 set_class_tag(rt, fib_rules_tclass(res));
1681#endif
1682 set_class_tag(rt, itag);
1683#endif
1684 rt->rt_type = res->type;
1685}
1686
1687static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1688 u8 tos, struct net_device *dev, int our)
1689{
1690 unsigned hash;
1691 struct rtable *rth;
1692 __be32 spec_dst;
1693 struct in_device *in_dev = in_dev_get(dev);
1694 u32 itag = 0;
1695
1696 /* Primary sanity checks. */
1697
1698 if (in_dev == NULL)
1699 return -EINVAL;
1700
1701 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1702 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1703 goto e_inval;
1704
1705 if (ipv4_is_zeronet(saddr)) {
1706 if (!ipv4_is_local_multicast(daddr))
1707 goto e_inval;
1708 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1709 } else if (fib_validate_source(saddr, 0, tos, 0,
1710 dev, &spec_dst, &itag) < 0)
1711 goto e_inval;
1712
1713 rth = dst_alloc(&ipv4_dst_ops);
1714 if (!rth)
1715 goto e_nobufs;
1716
1717 rth->u.dst.output= ip_rt_bug;
1718
1719 atomic_set(&rth->u.dst.__refcnt, 1);
1720 rth->u.dst.flags= DST_HOST;
1721 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1722 rth->u.dst.flags |= DST_NOPOLICY;
1723 rth->fl.fl4_dst = daddr;
1724 rth->rt_dst = daddr;
1725 rth->fl.fl4_tos = tos;
1726 rth->fl.mark = skb->mark;
1727 rth->fl.fl4_src = saddr;
1728 rth->rt_src = saddr;
1729#ifdef CONFIG_NET_CLS_ROUTE
1730 rth->u.dst.tclassid = itag;
1731#endif
1732 rth->rt_iif =
1733 rth->fl.iif = dev->ifindex;
1734 rth->u.dst.dev = init_net.loopback_dev;
1735 dev_hold(rth->u.dst.dev);
1736 rth->idev = in_dev_get(rth->u.dst.dev);
1737 rth->fl.oif = 0;
1738 rth->rt_gateway = daddr;
1739 rth->rt_spec_dst= spec_dst;
1740 rth->rt_genid = rt_genid(dev_net(dev));
1741 rth->rt_flags = RTCF_MULTICAST;
1742 rth->rt_type = RTN_MULTICAST;
1743 if (our) {
1744 rth->u.dst.input= ip_local_deliver;
1745 rth->rt_flags |= RTCF_LOCAL;
1746 }
1747
1748#ifdef CONFIG_IP_MROUTE
1749 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1750 rth->u.dst.input = ip_mr_input;
1751#endif
1752 RT_CACHE_STAT_INC(in_slow_mc);
1753
1754 in_dev_put(in_dev);
1755 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1756 return rt_intern_hash(hash, rth, &skb->rtable);
1757
1758e_nobufs:
1759 in_dev_put(in_dev);
1760 return -ENOBUFS;
1761
1762e_inval:
1763 in_dev_put(in_dev);
1764 return -EINVAL;
1765}
1766
1767
1768static void ip_handle_martian_source(struct net_device *dev,
1769 struct in_device *in_dev,
1770 struct sk_buff *skb,
1771 __be32 daddr,
1772 __be32 saddr)
1773{
1774 RT_CACHE_STAT_INC(in_martian_src);
1775#ifdef CONFIG_IP_ROUTE_VERBOSE
1776 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1777 /*
1778 * RFC1812 recommendation, if source is martian,
1779 * the only hint is MAC header.
1780 */
1781 printk(KERN_WARNING "martian source " NIPQUAD_FMT " from "
1782 NIPQUAD_FMT", on dev %s\n",
1783 NIPQUAD(daddr), NIPQUAD(saddr), dev->name);
1784 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1785 int i;
1786 const unsigned char *p = skb_mac_header(skb);
1787 printk(KERN_WARNING "ll header: ");
1788 for (i = 0; i < dev->hard_header_len; i++, p++) {
1789 printk("%02x", *p);
1790 if (i < (dev->hard_header_len - 1))
1791 printk(":");
1792 }
1793 printk("\n");
1794 }
1795 }
1796#endif
1797}
1798
1799static int __mkroute_input(struct sk_buff *skb,
1800 struct fib_result *res,
1801 struct in_device *in_dev,
1802 __be32 daddr, __be32 saddr, u32 tos,
1803 struct rtable **result)
1804{
1805
1806 struct rtable *rth;
1807 int err;
1808 struct in_device *out_dev;
1809 unsigned flags = 0;
1810 __be32 spec_dst;
1811 u32 itag;
1812
1813 /* get a working reference to the output device */
1814 out_dev = in_dev_get(FIB_RES_DEV(*res));
1815 if (out_dev == NULL) {
1816 if (net_ratelimit())
1817 printk(KERN_CRIT "Bug in ip_route_input" \
1818 "_slow(). Please, report\n");
1819 return -EINVAL;
1820 }
1821
1822
1823 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1824 in_dev->dev, &spec_dst, &itag);
1825 if (err < 0) {
1826 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1827 saddr);
1828
1829 err = -EINVAL;
1830 goto cleanup;
1831 }
1832
1833 if (err)
1834 flags |= RTCF_DIRECTSRC;
1835
1836 if (out_dev == in_dev && err &&
1837 (IN_DEV_SHARED_MEDIA(out_dev) ||
1838 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1839 flags |= RTCF_DOREDIRECT;
1840
1841 if (skb->protocol != htons(ETH_P_IP)) {
1842 /* Not IP (i.e. ARP). Do not create route, if it is
1843 * invalid for proxy arp. DNAT routes are always valid.
1844 */
1845 if (out_dev == in_dev) {
1846 err = -EINVAL;
1847 goto cleanup;
1848 }
1849 }
1850
1851
1852 rth = dst_alloc(&ipv4_dst_ops);
1853 if (!rth) {
1854 err = -ENOBUFS;
1855 goto cleanup;
1856 }
1857
1858 atomic_set(&rth->u.dst.__refcnt, 1);
1859 rth->u.dst.flags= DST_HOST;
1860 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1861 rth->u.dst.flags |= DST_NOPOLICY;
1862 if (IN_DEV_CONF_GET(out_dev, NOXFRM))
1863 rth->u.dst.flags |= DST_NOXFRM;
1864 rth->fl.fl4_dst = daddr;
1865 rth->rt_dst = daddr;
1866 rth->fl.fl4_tos = tos;
1867 rth->fl.mark = skb->mark;
1868 rth->fl.fl4_src = saddr;
1869 rth->rt_src = saddr;
1870 rth->rt_gateway = daddr;
1871 rth->rt_iif =
1872 rth->fl.iif = in_dev->dev->ifindex;
1873 rth->u.dst.dev = (out_dev)->dev;
1874 dev_hold(rth->u.dst.dev);
1875 rth->idev = in_dev_get(rth->u.dst.dev);
1876 rth->fl.oif = 0;
1877 rth->rt_spec_dst= spec_dst;
1878
1879 rth->u.dst.input = ip_forward;
1880 rth->u.dst.output = ip_output;
1881 rth->rt_genid = rt_genid(dev_net(rth->u.dst.dev));
1882
1883 rt_set_nexthop(rth, res, itag);
1884
1885 rth->rt_flags = flags;
1886
1887 *result = rth;
1888 err = 0;
1889 cleanup:
1890 /* release the working reference to the output device */
1891 in_dev_put(out_dev);
1892 return err;
1893}
1894
1895static int ip_mkroute_input(struct sk_buff *skb,
1896 struct fib_result *res,
1897 const struct flowi *fl,
1898 struct in_device *in_dev,
1899 __be32 daddr, __be32 saddr, u32 tos)
1900{
1901 struct rtable* rth = NULL;
1902 int err;
1903 unsigned hash;
1904
1905#ifdef CONFIG_IP_ROUTE_MULTIPATH
1906 if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
1907 fib_select_multipath(fl, res);
1908#endif
1909
1910 /* create a routing cache entry */
1911 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
1912 if (err)
1913 return err;
1914
1915 /* put it into the cache */
1916 hash = rt_hash(daddr, saddr, fl->iif,
1917 rt_genid(dev_net(rth->u.dst.dev)));
1918 return rt_intern_hash(hash, rth, &skb->rtable);
1919}
1920
1921/*
1922 * NOTE. We drop all the packets that has local source
1923 * addresses, because every properly looped back packet
1924 * must have correct destination already attached by output routine.
1925 *
1926 * Such approach solves two big problems:
1927 * 1. Not simplex devices are handled properly.
1928 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1929 */
1930
1931static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1932 u8 tos, struct net_device *dev)
1933{
1934 struct fib_result res;
1935 struct in_device *in_dev = in_dev_get(dev);
1936 struct flowi fl = { .nl_u = { .ip4_u =
1937 { .daddr = daddr,
1938 .saddr = saddr,
1939 .tos = tos,
1940 .scope = RT_SCOPE_UNIVERSE,
1941 } },
1942 .mark = skb->mark,
1943 .iif = dev->ifindex };
1944 unsigned flags = 0;
1945 u32 itag = 0;
1946 struct rtable * rth;
1947 unsigned hash;
1948 __be32 spec_dst;
1949 int err = -EINVAL;
1950 int free_res = 0;
1951 struct net * net = dev_net(dev);
1952
1953 /* IP on this device is disabled. */
1954
1955 if (!in_dev)
1956 goto out;
1957
1958 /* Check for the most weird martians, which can be not detected
1959 by fib_lookup.
1960 */
1961
1962 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1963 ipv4_is_loopback(saddr))
1964 goto martian_source;
1965
1966 if (daddr == htonl(0xFFFFFFFF) || (saddr == 0 && daddr == 0))
1967 goto brd_input;
1968
1969 /* Accept zero addresses only to limited broadcast;
1970 * I even do not know to fix it or not. Waiting for complains :-)
1971 */
1972 if (ipv4_is_zeronet(saddr))
1973 goto martian_source;
1974
1975 if (ipv4_is_lbcast(daddr) || ipv4_is_zeronet(daddr) ||
1976 ipv4_is_loopback(daddr))
1977 goto martian_destination;
1978
1979 /*
1980 * Now we are ready to route packet.
1981 */
1982 if ((err = fib_lookup(net, &fl, &res)) != 0) {
1983 if (!IN_DEV_FORWARD(in_dev))
1984 goto e_hostunreach;
1985 goto no_route;
1986 }
1987 free_res = 1;
1988
1989 RT_CACHE_STAT_INC(in_slow_tot);
1990
1991 if (res.type == RTN_BROADCAST)
1992 goto brd_input;
1993
1994 if (res.type == RTN_LOCAL) {
1995 int result;
1996 result = fib_validate_source(saddr, daddr, tos,
1997 net->loopback_dev->ifindex,
1998 dev, &spec_dst, &itag);
1999 if (result < 0)
2000 goto martian_source;
2001 if (result)
2002 flags |= RTCF_DIRECTSRC;
2003 spec_dst = daddr;
2004 goto local_input;
2005 }
2006
2007 if (!IN_DEV_FORWARD(in_dev))
2008 goto e_hostunreach;
2009 if (res.type != RTN_UNICAST)
2010 goto martian_destination;
2011
2012 err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
2013done:
2014 in_dev_put(in_dev);
2015 if (free_res)
2016 fib_res_put(&res);
2017out: return err;
2018
2019brd_input:
2020 if (skb->protocol != htons(ETH_P_IP))
2021 goto e_inval;
2022
2023 if (ipv4_is_zeronet(saddr))
2024 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2025 else {
2026 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2027 &itag);
2028 if (err < 0)
2029 goto martian_source;
2030 if (err)
2031 flags |= RTCF_DIRECTSRC;
2032 }
2033 flags |= RTCF_BROADCAST;
2034 res.type = RTN_BROADCAST;
2035 RT_CACHE_STAT_INC(in_brd);
2036
2037local_input:
2038 rth = dst_alloc(&ipv4_dst_ops);
2039 if (!rth)
2040 goto e_nobufs;
2041
2042 rth->u.dst.output= ip_rt_bug;
2043 rth->rt_genid = rt_genid(net);
2044
2045 atomic_set(&rth->u.dst.__refcnt, 1);
2046 rth->u.dst.flags= DST_HOST;
2047 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2048 rth->u.dst.flags |= DST_NOPOLICY;
2049 rth->fl.fl4_dst = daddr;
2050 rth->rt_dst = daddr;
2051 rth->fl.fl4_tos = tos;
2052 rth->fl.mark = skb->mark;
2053 rth->fl.fl4_src = saddr;
2054 rth->rt_src = saddr;
2055#ifdef CONFIG_NET_CLS_ROUTE
2056 rth->u.dst.tclassid = itag;
2057#endif
2058 rth->rt_iif =
2059 rth->fl.iif = dev->ifindex;
2060 rth->u.dst.dev = net->loopback_dev;
2061 dev_hold(rth->u.dst.dev);
2062 rth->idev = in_dev_get(rth->u.dst.dev);
2063 rth->rt_gateway = daddr;
2064 rth->rt_spec_dst= spec_dst;
2065 rth->u.dst.input= ip_local_deliver;
2066 rth->rt_flags = flags|RTCF_LOCAL;
2067 if (res.type == RTN_UNREACHABLE) {
2068 rth->u.dst.input= ip_error;
2069 rth->u.dst.error= -err;
2070 rth->rt_flags &= ~RTCF_LOCAL;
2071 }
2072 rth->rt_type = res.type;
2073 hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));
2074 err = rt_intern_hash(hash, rth, &skb->rtable);
2075 goto done;
2076
2077no_route:
2078 RT_CACHE_STAT_INC(in_no_route);
2079 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2080 res.type = RTN_UNREACHABLE;
2081 if (err == -ESRCH)
2082 err = -ENETUNREACH;
2083 goto local_input;
2084
2085 /*
2086 * Do not cache martian addresses: they should be logged (RFC1812)
2087 */
2088martian_destination:
2089 RT_CACHE_STAT_INC(in_martian_dst);
2090#ifdef CONFIG_IP_ROUTE_VERBOSE
2091 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2092 printk(KERN_WARNING "martian destination " NIPQUAD_FMT " from "
2093 NIPQUAD_FMT ", dev %s\n",
2094 NIPQUAD(daddr), NIPQUAD(saddr), dev->name);
2095#endif
2096
2097e_hostunreach:
2098 err = -EHOSTUNREACH;
2099 goto done;
2100
2101e_inval:
2102 err = -EINVAL;
2103 goto done;
2104
2105e_nobufs:
2106 err = -ENOBUFS;
2107 goto done;
2108
2109martian_source:
2110 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2111 goto e_inval;
2112}
2113
2114int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2115 u8 tos, struct net_device *dev)
2116{
2117 struct rtable * rth;
2118 unsigned hash;
2119 int iif = dev->ifindex;
2120 struct net *net;
2121
2122 net = dev_net(dev);
2123 tos &= IPTOS_RT_MASK;
2124 hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2125
2126 rcu_read_lock();
2127 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2128 rth = rcu_dereference(rth->u.dst.rt_next)) {
2129 if (((rth->fl.fl4_dst ^ daddr) |
2130 (rth->fl.fl4_src ^ saddr) |
2131 (rth->fl.iif ^ iif) |
2132 rth->fl.oif |
2133 (rth->fl.fl4_tos ^ tos)) == 0 &&
2134 rth->fl.mark == skb->mark &&
2135 net_eq(dev_net(rth->u.dst.dev), net) &&
2136 !rt_is_expired(rth)) {
2137 dst_use(&rth->u.dst, jiffies);
2138 RT_CACHE_STAT_INC(in_hit);
2139 rcu_read_unlock();
2140 skb->rtable = rth;
2141 return 0;
2142 }
2143 RT_CACHE_STAT_INC(in_hlist_search);
2144 }
2145 rcu_read_unlock();
2146
2147 /* Multicast recognition logic is moved from route cache to here.
2148 The problem was that too many Ethernet cards have broken/missing
2149 hardware multicast filters :-( As result the host on multicasting
2150 network acquires a lot of useless route cache entries, sort of
2151 SDR messages from all the world. Now we try to get rid of them.
2152 Really, provided software IP multicast filter is organized
2153 reasonably (at least, hashed), it does not result in a slowdown
2154 comparing with route cache reject entries.
2155 Note, that multicast routers are not affected, because
2156 route cache entry is created eventually.
2157 */
2158 if (ipv4_is_multicast(daddr)) {
2159 struct in_device *in_dev;
2160
2161 rcu_read_lock();
2162 if ((in_dev = __in_dev_get_rcu(dev)) != NULL) {
2163 int our = ip_check_mc(in_dev, daddr, saddr,
2164 ip_hdr(skb)->protocol);
2165 if (our
2166#ifdef CONFIG_IP_MROUTE
2167 || (!ipv4_is_local_multicast(daddr) &&
2168 IN_DEV_MFORWARD(in_dev))
2169#endif
2170 ) {
2171 rcu_read_unlock();
2172 return ip_route_input_mc(skb, daddr, saddr,
2173 tos, dev, our);
2174 }
2175 }
2176 rcu_read_unlock();
2177 return -EINVAL;
2178 }
2179 return ip_route_input_slow(skb, daddr, saddr, tos, dev);
2180}
2181
2182static int __mkroute_output(struct rtable **result,
2183 struct fib_result *res,
2184 const struct flowi *fl,
2185 const struct flowi *oldflp,
2186 struct net_device *dev_out,
2187 unsigned flags)
2188{
2189 struct rtable *rth;
2190 struct in_device *in_dev;
2191 u32 tos = RT_FL_TOS(oldflp);
2192 int err = 0;
2193
2194 if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags&IFF_LOOPBACK))
2195 return -EINVAL;
2196
2197 if (fl->fl4_dst == htonl(0xFFFFFFFF))
2198 res->type = RTN_BROADCAST;
2199 else if (ipv4_is_multicast(fl->fl4_dst))
2200 res->type = RTN_MULTICAST;
2201 else if (ipv4_is_lbcast(fl->fl4_dst) || ipv4_is_zeronet(fl->fl4_dst))
2202 return -EINVAL;
2203
2204 if (dev_out->flags & IFF_LOOPBACK)
2205 flags |= RTCF_LOCAL;
2206
2207 /* get work reference to inet device */
2208 in_dev = in_dev_get(dev_out);
2209 if (!in_dev)
2210 return -EINVAL;
2211
2212 if (res->type == RTN_BROADCAST) {
2213 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2214 if (res->fi) {
2215 fib_info_put(res->fi);
2216 res->fi = NULL;
2217 }
2218 } else if (res->type == RTN_MULTICAST) {
2219 flags |= RTCF_MULTICAST|RTCF_LOCAL;
2220 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2221 oldflp->proto))
2222 flags &= ~RTCF_LOCAL;
2223 /* If multicast route do not exist use
2224 default one, but do not gateway in this case.
2225 Yes, it is hack.
2226 */
2227 if (res->fi && res->prefixlen < 4) {
2228 fib_info_put(res->fi);
2229 res->fi = NULL;
2230 }
2231 }
2232
2233
2234 rth = dst_alloc(&ipv4_dst_ops);
2235 if (!rth) {
2236 err = -ENOBUFS;
2237 goto cleanup;
2238 }
2239
2240 atomic_set(&rth->u.dst.__refcnt, 1);
2241 rth->u.dst.flags= DST_HOST;
2242 if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2243 rth->u.dst.flags |= DST_NOXFRM;
2244 if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2245 rth->u.dst.flags |= DST_NOPOLICY;
2246
2247 rth->fl.fl4_dst = oldflp->fl4_dst;
2248 rth->fl.fl4_tos = tos;
2249 rth->fl.fl4_src = oldflp->fl4_src;
2250 rth->fl.oif = oldflp->oif;
2251 rth->fl.mark = oldflp->mark;
2252 rth->rt_dst = fl->fl4_dst;
2253 rth->rt_src = fl->fl4_src;
2254 rth->rt_iif = oldflp->oif ? : dev_out->ifindex;
2255 /* get references to the devices that are to be hold by the routing
2256 cache entry */
2257 rth->u.dst.dev = dev_out;
2258 dev_hold(dev_out);
2259 rth->idev = in_dev_get(dev_out);
2260 rth->rt_gateway = fl->fl4_dst;
2261 rth->rt_spec_dst= fl->fl4_src;
2262
2263 rth->u.dst.output=ip_output;
2264 rth->rt_genid = rt_genid(dev_net(dev_out));
2265
2266 RT_CACHE_STAT_INC(out_slow_tot);
2267
2268 if (flags & RTCF_LOCAL) {
2269 rth->u.dst.input = ip_local_deliver;
2270 rth->rt_spec_dst = fl->fl4_dst;
2271 }
2272 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2273 rth->rt_spec_dst = fl->fl4_src;
2274 if (flags & RTCF_LOCAL &&
2275 !(dev_out->flags & IFF_LOOPBACK)) {
2276 rth->u.dst.output = ip_mc_output;
2277 RT_CACHE_STAT_INC(out_slow_mc);
2278 }
2279#ifdef CONFIG_IP_MROUTE
2280 if (res->type == RTN_MULTICAST) {
2281 if (IN_DEV_MFORWARD(in_dev) &&
2282 !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2283 rth->u.dst.input = ip_mr_input;
2284 rth->u.dst.output = ip_mc_output;
2285 }
2286 }
2287#endif
2288 }
2289
2290 rt_set_nexthop(rth, res, 0);
2291
2292 rth->rt_flags = flags;
2293
2294 *result = rth;
2295 cleanup:
2296 /* release work reference to inet device */
2297 in_dev_put(in_dev);
2298
2299 return err;
2300}
2301
2302static int ip_mkroute_output(struct rtable **rp,
2303 struct fib_result *res,
2304 const struct flowi *fl,
2305 const struct flowi *oldflp,
2306 struct net_device *dev_out,
2307 unsigned flags)
2308{
2309 struct rtable *rth = NULL;
2310 int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2311 unsigned hash;
2312 if (err == 0) {
2313 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif,
2314 rt_genid(dev_net(dev_out)));
2315 err = rt_intern_hash(hash, rth, rp);
2316 }
2317
2318 return err;
2319}
2320
2321/*
2322 * Major route resolver routine.
2323 */
2324
2325static int ip_route_output_slow(struct net *net, struct rtable **rp,
2326 const struct flowi *oldflp)
2327{
2328 u32 tos = RT_FL_TOS(oldflp);
2329 struct flowi fl = { .nl_u = { .ip4_u =
2330 { .daddr = oldflp->fl4_dst,
2331 .saddr = oldflp->fl4_src,
2332 .tos = tos & IPTOS_RT_MASK,
2333 .scope = ((tos & RTO_ONLINK) ?
2334 RT_SCOPE_LINK :
2335 RT_SCOPE_UNIVERSE),
2336 } },
2337 .mark = oldflp->mark,
2338 .iif = net->loopback_dev->ifindex,
2339 .oif = oldflp->oif };
2340 struct fib_result res;
2341 unsigned flags = 0;
2342 struct net_device *dev_out = NULL;
2343 int free_res = 0;
2344 int err;
2345
2346
2347 res.fi = NULL;
2348#ifdef CONFIG_IP_MULTIPLE_TABLES
2349 res.r = NULL;
2350#endif
2351
2352 if (oldflp->fl4_src) {
2353 err = -EINVAL;
2354 if (ipv4_is_multicast(oldflp->fl4_src) ||
2355 ipv4_is_lbcast(oldflp->fl4_src) ||
2356 ipv4_is_zeronet(oldflp->fl4_src))
2357 goto out;
2358
2359 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2360 dev_out = ip_dev_find(net, oldflp->fl4_src);
2361 if (dev_out == NULL)
2362 goto out;
2363
2364 /* I removed check for oif == dev_out->oif here.
2365 It was wrong for two reasons:
2366 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2367 is assigned to multiple interfaces.
2368 2. Moreover, we are allowed to send packets with saddr
2369 of another iface. --ANK
2370 */
2371
2372 if (oldflp->oif == 0
2373 && (ipv4_is_multicast(oldflp->fl4_dst) ||
2374 oldflp->fl4_dst == htonl(0xFFFFFFFF))) {
2375 /* Special hack: user can direct multicasts
2376 and limited broadcast via necessary interface
2377 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2378 This hack is not just for fun, it allows
2379 vic,vat and friends to work.
2380 They bind socket to loopback, set ttl to zero
2381 and expect that it will work.
2382 From the viewpoint of routing cache they are broken,
2383 because we are not allowed to build multicast path
2384 with loopback source addr (look, routing cache
2385 cannot know, that ttl is zero, so that packet
2386 will not leave this host and route is valid).
2387 Luckily, this hack is good workaround.
2388 */
2389
2390 fl.oif = dev_out->ifindex;
2391 goto make_route;
2392 }
2393 if (dev_out)
2394 dev_put(dev_out);
2395 dev_out = NULL;
2396 }
2397
2398
2399 if (oldflp->oif) {
2400 dev_out = dev_get_by_index(net, oldflp->oif);
2401 err = -ENODEV;
2402 if (dev_out == NULL)
2403 goto out;
2404
2405 /* RACE: Check return value of inet_select_addr instead. */
2406 if (__in_dev_get_rtnl(dev_out) == NULL) {
2407 dev_put(dev_out);
2408 goto out; /* Wrong error code */
2409 }
2410
2411 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2412 oldflp->fl4_dst == htonl(0xFFFFFFFF)) {
2413 if (!fl.fl4_src)
2414 fl.fl4_src = inet_select_addr(dev_out, 0,
2415 RT_SCOPE_LINK);
2416 goto make_route;
2417 }
2418 if (!fl.fl4_src) {
2419 if (ipv4_is_multicast(oldflp->fl4_dst))
2420 fl.fl4_src = inet_select_addr(dev_out, 0,
2421 fl.fl4_scope);
2422 else if (!oldflp->fl4_dst)
2423 fl.fl4_src = inet_select_addr(dev_out, 0,
2424 RT_SCOPE_HOST);
2425 }
2426 }
2427
2428 if (!fl.fl4_dst) {
2429 fl.fl4_dst = fl.fl4_src;
2430 if (!fl.fl4_dst)
2431 fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2432 if (dev_out)
2433 dev_put(dev_out);
2434 dev_out = net->loopback_dev;
2435 dev_hold(dev_out);
2436 fl.oif = net->loopback_dev->ifindex;
2437 res.type = RTN_LOCAL;
2438 flags |= RTCF_LOCAL;
2439 goto make_route;
2440 }
2441
2442 if (fib_lookup(net, &fl, &res)) {
2443 res.fi = NULL;
2444 if (oldflp->oif) {
2445 /* Apparently, routing tables are wrong. Assume,
2446 that the destination is on link.
2447
2448 WHY? DW.
2449 Because we are allowed to send to iface
2450 even if it has NO routes and NO assigned
2451 addresses. When oif is specified, routing
2452 tables are looked up with only one purpose:
2453 to catch if destination is gatewayed, rather than
2454 direct. Moreover, if MSG_DONTROUTE is set,
2455 we send packet, ignoring both routing tables
2456 and ifaddr state. --ANK
2457
2458
2459 We could make it even if oif is unknown,
2460 likely IPv6, but we do not.
2461 */
2462
2463 if (fl.fl4_src == 0)
2464 fl.fl4_src = inet_select_addr(dev_out, 0,
2465 RT_SCOPE_LINK);
2466 res.type = RTN_UNICAST;
2467 goto make_route;
2468 }
2469 if (dev_out)
2470 dev_put(dev_out);
2471 err = -ENETUNREACH;
2472 goto out;
2473 }
2474 free_res = 1;
2475
2476 if (res.type == RTN_LOCAL) {
2477 if (!fl.fl4_src)
2478 fl.fl4_src = fl.fl4_dst;
2479 if (dev_out)
2480 dev_put(dev_out);
2481 dev_out = net->loopback_dev;
2482 dev_hold(dev_out);
2483 fl.oif = dev_out->ifindex;
2484 if (res.fi)
2485 fib_info_put(res.fi);
2486 res.fi = NULL;
2487 flags |= RTCF_LOCAL;
2488 goto make_route;
2489 }
2490
2491#ifdef CONFIG_IP_ROUTE_MULTIPATH
2492 if (res.fi->fib_nhs > 1 && fl.oif == 0)
2493 fib_select_multipath(&fl, &res);
2494 else
2495#endif
2496 if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2497 fib_select_default(net, &fl, &res);
2498
2499 if (!fl.fl4_src)
2500 fl.fl4_src = FIB_RES_PREFSRC(res);
2501
2502 if (dev_out)
2503 dev_put(dev_out);
2504 dev_out = FIB_RES_DEV(res);
2505 dev_hold(dev_out);
2506 fl.oif = dev_out->ifindex;
2507
2508
2509make_route:
2510 err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2511
2512
2513 if (free_res)
2514 fib_res_put(&res);
2515 if (dev_out)
2516 dev_put(dev_out);
2517out: return err;
2518}
2519
2520int __ip_route_output_key(struct net *net, struct rtable **rp,
2521 const struct flowi *flp)
2522{
2523 unsigned hash;
2524 struct rtable *rth;
2525
2526 hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, rt_genid(net));
2527
2528 rcu_read_lock_bh();
2529 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2530 rth = rcu_dereference(rth->u.dst.rt_next)) {
2531 if (rth->fl.fl4_dst == flp->fl4_dst &&
2532 rth->fl.fl4_src == flp->fl4_src &&
2533 rth->fl.iif == 0 &&
2534 rth->fl.oif == flp->oif &&
2535 rth->fl.mark == flp->mark &&
2536 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2537 (IPTOS_RT_MASK | RTO_ONLINK)) &&
2538 net_eq(dev_net(rth->u.dst.dev), net) &&
2539 !rt_is_expired(rth)) {
2540 dst_use(&rth->u.dst, jiffies);
2541 RT_CACHE_STAT_INC(out_hit);
2542 rcu_read_unlock_bh();
2543 *rp = rth;
2544 return 0;
2545 }
2546 RT_CACHE_STAT_INC(out_hlist_search);
2547 }
2548 rcu_read_unlock_bh();
2549
2550 return ip_route_output_slow(net, rp, flp);
2551}
2552
2553EXPORT_SYMBOL_GPL(__ip_route_output_key);
2554
2555static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2556{
2557}
2558
2559static struct dst_ops ipv4_dst_blackhole_ops = {
2560 .family = AF_INET,
2561 .protocol = __constant_htons(ETH_P_IP),
2562 .destroy = ipv4_dst_destroy,
2563 .check = ipv4_dst_check,
2564 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2565 .entry_size = sizeof(struct rtable),
2566 .entries = ATOMIC_INIT(0),
2567};
2568
2569
2570static int ipv4_dst_blackhole(struct net *net, struct rtable **rp, struct flowi *flp)
2571{
2572 struct rtable *ort = *rp;
2573 struct rtable *rt = (struct rtable *)
2574 dst_alloc(&ipv4_dst_blackhole_ops);
2575
2576 if (rt) {
2577 struct dst_entry *new = &rt->u.dst;
2578
2579 atomic_set(&new->__refcnt, 1);
2580 new->__use = 1;
2581 new->input = dst_discard;
2582 new->output = dst_discard;
2583 memcpy(new->metrics, ort->u.dst.metrics, RTAX_MAX*sizeof(u32));
2584
2585 new->dev = ort->u.dst.dev;
2586 if (new->dev)
2587 dev_hold(new->dev);
2588
2589 rt->fl = ort->fl;
2590
2591 rt->idev = ort->idev;
2592 if (rt->idev)
2593 in_dev_hold(rt->idev);
2594 rt->rt_genid = rt_genid(net);
2595 rt->rt_flags = ort->rt_flags;
2596 rt->rt_type = ort->rt_type;
2597 rt->rt_dst = ort->rt_dst;
2598 rt->rt_src = ort->rt_src;
2599 rt->rt_iif = ort->rt_iif;
2600 rt->rt_gateway = ort->rt_gateway;
2601 rt->rt_spec_dst = ort->rt_spec_dst;
2602 rt->peer = ort->peer;
2603 if (rt->peer)
2604 atomic_inc(&rt->peer->refcnt);
2605
2606 dst_free(new);
2607 }
2608
2609 dst_release(&(*rp)->u.dst);
2610 *rp = rt;
2611 return (rt ? 0 : -ENOMEM);
2612}
2613
2614int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2615 struct sock *sk, int flags)
2616{
2617 int err;
2618
2619 if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2620 return err;
2621
2622 if (flp->proto) {
2623 if (!flp->fl4_src)
2624 flp->fl4_src = (*rp)->rt_src;
2625 if (!flp->fl4_dst)
2626 flp->fl4_dst = (*rp)->rt_dst;
2627 err = __xfrm_lookup((struct dst_entry **)rp, flp, sk,
2628 flags ? XFRM_LOOKUP_WAIT : 0);
2629 if (err == -EREMOTE)
2630 err = ipv4_dst_blackhole(net, rp, flp);
2631
2632 return err;
2633 }
2634
2635 return 0;
2636}
2637
2638EXPORT_SYMBOL_GPL(ip_route_output_flow);
2639
2640int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2641{
2642 return ip_route_output_flow(net, rp, flp, NULL, 0);
2643}
2644
2645static int rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
2646 int nowait, unsigned int flags)
2647{
2648 struct rtable *rt = skb->rtable;
2649 struct rtmsg *r;
2650 struct nlmsghdr *nlh;
2651 long expires;
2652 u32 id = 0, ts = 0, tsage = 0, error;
2653
2654 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2655 if (nlh == NULL)
2656 return -EMSGSIZE;
2657
2658 r = nlmsg_data(nlh);
2659 r->rtm_family = AF_INET;
2660 r->rtm_dst_len = 32;
2661 r->rtm_src_len = 0;
2662 r->rtm_tos = rt->fl.fl4_tos;
2663 r->rtm_table = RT_TABLE_MAIN;
2664 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2665 r->rtm_type = rt->rt_type;
2666 r->rtm_scope = RT_SCOPE_UNIVERSE;
2667 r->rtm_protocol = RTPROT_UNSPEC;
2668 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2669 if (rt->rt_flags & RTCF_NOTIFY)
2670 r->rtm_flags |= RTM_F_NOTIFY;
2671
2672 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2673
2674 if (rt->fl.fl4_src) {
2675 r->rtm_src_len = 32;
2676 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2677 }
2678 if (rt->u.dst.dev)
2679 NLA_PUT_U32(skb, RTA_OIF, rt->u.dst.dev->ifindex);
2680#ifdef CONFIG_NET_CLS_ROUTE
2681 if (rt->u.dst.tclassid)
2682 NLA_PUT_U32(skb, RTA_FLOW, rt->u.dst.tclassid);
2683#endif
2684 if (rt->fl.iif)
2685 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2686 else if (rt->rt_src != rt->fl.fl4_src)
2687 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2688
2689 if (rt->rt_dst != rt->rt_gateway)
2690 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2691
2692 if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
2693 goto nla_put_failure;
2694
2695 error = rt->u.dst.error;
2696 expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
2697 if (rt->peer) {
2698 id = rt->peer->ip_id_count;
2699 if (rt->peer->tcp_ts_stamp) {
2700 ts = rt->peer->tcp_ts;
2701 tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2702 }
2703 }
2704
2705 if (rt->fl.iif) {
2706#ifdef CONFIG_IP_MROUTE
2707 __be32 dst = rt->rt_dst;
2708
2709 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2710 IPV4_DEVCONF_ALL(&init_net, MC_FORWARDING)) {
2711 int err = ipmr_get_route(skb, r, nowait);
2712 if (err <= 0) {
2713 if (!nowait) {
2714 if (err == 0)
2715 return 0;
2716 goto nla_put_failure;
2717 } else {
2718 if (err == -EMSGSIZE)
2719 goto nla_put_failure;
2720 error = err;
2721 }
2722 }
2723 } else
2724#endif
2725 NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2726 }
2727
2728 if (rtnl_put_cacheinfo(skb, &rt->u.dst, id, ts, tsage,
2729 expires, error) < 0)
2730 goto nla_put_failure;
2731
2732 return nlmsg_end(skb, nlh);
2733
2734nla_put_failure:
2735 nlmsg_cancel(skb, nlh);
2736 return -EMSGSIZE;
2737}
2738
2739static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2740{
2741 struct net *net = sock_net(in_skb->sk);
2742 struct rtmsg *rtm;
2743 struct nlattr *tb[RTA_MAX+1];
2744 struct rtable *rt = NULL;
2745 __be32 dst = 0;
2746 __be32 src = 0;
2747 u32 iif;
2748 int err;
2749 struct sk_buff *skb;
2750
2751 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2752 if (err < 0)
2753 goto errout;
2754
2755 rtm = nlmsg_data(nlh);
2756
2757 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2758 if (skb == NULL) {
2759 err = -ENOBUFS;
2760 goto errout;
2761 }
2762
2763 /* Reserve room for dummy headers, this skb can pass
2764 through good chunk of routing engine.
2765 */
2766 skb_reset_mac_header(skb);
2767 skb_reset_network_header(skb);
2768
2769 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2770 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2771 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2772
2773 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2774 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2775 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2776
2777 if (iif) {
2778 struct net_device *dev;
2779
2780 dev = __dev_get_by_index(net, iif);
2781 if (dev == NULL) {
2782 err = -ENODEV;
2783 goto errout_free;
2784 }
2785
2786 skb->protocol = htons(ETH_P_IP);
2787 skb->dev = dev;
2788 local_bh_disable();
2789 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2790 local_bh_enable();
2791
2792 rt = skb->rtable;
2793 if (err == 0 && rt->u.dst.error)
2794 err = -rt->u.dst.error;
2795 } else {
2796 struct flowi fl = {
2797 .nl_u = {
2798 .ip4_u = {
2799 .daddr = dst,
2800 .saddr = src,
2801 .tos = rtm->rtm_tos,
2802 },
2803 },
2804 .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2805 };
2806 err = ip_route_output_key(net, &rt, &fl);
2807 }
2808
2809 if (err)
2810 goto errout_free;
2811
2812 skb->rtable = rt;
2813 if (rtm->rtm_flags & RTM_F_NOTIFY)
2814 rt->rt_flags |= RTCF_NOTIFY;
2815
2816 err = rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2817 RTM_NEWROUTE, 0, 0);
2818 if (err <= 0)
2819 goto errout_free;
2820
2821 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2822errout:
2823 return err;
2824
2825errout_free:
2826 kfree_skb(skb);
2827 goto errout;
2828}
2829
2830int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2831{
2832 struct rtable *rt;
2833 int h, s_h;
2834 int idx, s_idx;
2835 struct net *net;
2836
2837 net = sock_net(skb->sk);
2838
2839 s_h = cb->args[0];
2840 if (s_h < 0)
2841 s_h = 0;
2842 s_idx = idx = cb->args[1];
2843 for (h = s_h; h <= rt_hash_mask; h++) {
2844 rcu_read_lock_bh();
2845 for (rt = rcu_dereference(rt_hash_table[h].chain), idx = 0; rt;
2846 rt = rcu_dereference(rt->u.dst.rt_next), idx++) {
2847 if (!net_eq(dev_net(rt->u.dst.dev), net) || idx < s_idx)
2848 continue;
2849 if (rt_is_expired(rt))
2850 continue;
2851 skb->dst = dst_clone(&rt->u.dst);
2852 if (rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
2853 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
2854 1, NLM_F_MULTI) <= 0) {
2855 dst_release(xchg(&skb->dst, NULL));
2856 rcu_read_unlock_bh();
2857 goto done;
2858 }
2859 dst_release(xchg(&skb->dst, NULL));
2860 }
2861 rcu_read_unlock_bh();
2862 s_idx = 0;
2863 }
2864
2865done:
2866 cb->args[0] = h;
2867 cb->args[1] = idx;
2868 return skb->len;
2869}
2870
2871void ip_rt_multicast_event(struct in_device *in_dev)
2872{
2873 rt_cache_flush(dev_net(in_dev->dev), 0);
2874}
2875
2876#ifdef CONFIG_SYSCTL
2877static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
2878 struct file *filp, void __user *buffer,
2879 size_t *lenp, loff_t *ppos)
2880{
2881 if (write) {
2882 int flush_delay;
2883 ctl_table ctl;
2884 struct net *net;
2885
2886 memcpy(&ctl, __ctl, sizeof(ctl));
2887 ctl.data = &flush_delay;
2888 proc_dointvec(&ctl, write, filp, buffer, lenp, ppos);
2889
2890 net = (struct net *)__ctl->extra1;
2891 rt_cache_flush(net, flush_delay);
2892 return 0;
2893 }
2894
2895 return -EINVAL;
2896}
2897
2898static int ipv4_sysctl_rtcache_flush_strategy(ctl_table *table,
2899 int __user *name,
2900 int nlen,
2901 void __user *oldval,
2902 size_t __user *oldlenp,
2903 void __user *newval,
2904 size_t newlen)
2905{
2906 int delay;
2907 struct net *net;
2908 if (newlen != sizeof(int))
2909 return -EINVAL;
2910 if (get_user(delay, (int __user *)newval))
2911 return -EFAULT;
2912 net = (struct net *)table->extra1;
2913 rt_cache_flush(net, delay);
2914 return 0;
2915}
2916
2917static ctl_table ipv4_route_table[] = {
2918 {
2919 .ctl_name = NET_IPV4_ROUTE_GC_THRESH,
2920 .procname = "gc_thresh",
2921 .data = &ipv4_dst_ops.gc_thresh,
2922 .maxlen = sizeof(int),
2923 .mode = 0644,
2924 .proc_handler = &proc_dointvec,
2925 },
2926 {
2927 .ctl_name = NET_IPV4_ROUTE_MAX_SIZE,
2928 .procname = "max_size",
2929 .data = &ip_rt_max_size,
2930 .maxlen = sizeof(int),
2931 .mode = 0644,
2932 .proc_handler = &proc_dointvec,
2933 },
2934 {
2935 /* Deprecated. Use gc_min_interval_ms */
2936
2937 .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL,
2938 .procname = "gc_min_interval",
2939 .data = &ip_rt_gc_min_interval,
2940 .maxlen = sizeof(int),
2941 .mode = 0644,
2942 .proc_handler = &proc_dointvec_jiffies,
2943 .strategy = &sysctl_jiffies,
2944 },
2945 {
2946 .ctl_name = NET_IPV4_ROUTE_GC_MIN_INTERVAL_MS,
2947 .procname = "gc_min_interval_ms",
2948 .data = &ip_rt_gc_min_interval,
2949 .maxlen = sizeof(int),
2950 .mode = 0644,
2951 .proc_handler = &proc_dointvec_ms_jiffies,
2952 .strategy = &sysctl_ms_jiffies,
2953 },
2954 {
2955 .ctl_name = NET_IPV4_ROUTE_GC_TIMEOUT,
2956 .procname = "gc_timeout",
2957 .data = &ip_rt_gc_timeout,
2958 .maxlen = sizeof(int),
2959 .mode = 0644,
2960 .proc_handler = &proc_dointvec_jiffies,
2961 .strategy = &sysctl_jiffies,
2962 },
2963 {
2964 .ctl_name = NET_IPV4_ROUTE_GC_INTERVAL,
2965 .procname = "gc_interval",
2966 .data = &ip_rt_gc_interval,
2967 .maxlen = sizeof(int),
2968 .mode = 0644,
2969 .proc_handler = &proc_dointvec_jiffies,
2970 .strategy = &sysctl_jiffies,
2971 },
2972 {
2973 .ctl_name = NET_IPV4_ROUTE_REDIRECT_LOAD,
2974 .procname = "redirect_load",
2975 .data = &ip_rt_redirect_load,
2976 .maxlen = sizeof(int),
2977 .mode = 0644,
2978 .proc_handler = &proc_dointvec,
2979 },
2980 {
2981 .ctl_name = NET_IPV4_ROUTE_REDIRECT_NUMBER,
2982 .procname = "redirect_number",
2983 .data = &ip_rt_redirect_number,
2984 .maxlen = sizeof(int),
2985 .mode = 0644,
2986 .proc_handler = &proc_dointvec,
2987 },
2988 {
2989 .ctl_name = NET_IPV4_ROUTE_REDIRECT_SILENCE,
2990 .procname = "redirect_silence",
2991 .data = &ip_rt_redirect_silence,
2992 .maxlen = sizeof(int),
2993 .mode = 0644,
2994 .proc_handler = &proc_dointvec,
2995 },
2996 {
2997 .ctl_name = NET_IPV4_ROUTE_ERROR_COST,
2998 .procname = "error_cost",
2999 .data = &ip_rt_error_cost,
3000 .maxlen = sizeof(int),
3001 .mode = 0644,
3002 .proc_handler = &proc_dointvec,
3003 },
3004 {
3005 .ctl_name = NET_IPV4_ROUTE_ERROR_BURST,
3006 .procname = "error_burst",
3007 .data = &ip_rt_error_burst,
3008 .maxlen = sizeof(int),
3009 .mode = 0644,
3010 .proc_handler = &proc_dointvec,
3011 },
3012 {
3013 .ctl_name = NET_IPV4_ROUTE_GC_ELASTICITY,
3014 .procname = "gc_elasticity",
3015 .data = &ip_rt_gc_elasticity,
3016 .maxlen = sizeof(int),
3017 .mode = 0644,
3018 .proc_handler = &proc_dointvec,
3019 },
3020 {
3021 .ctl_name = NET_IPV4_ROUTE_MTU_EXPIRES,
3022 .procname = "mtu_expires",
3023 .data = &ip_rt_mtu_expires,
3024 .maxlen = sizeof(int),
3025 .mode = 0644,
3026 .proc_handler = &proc_dointvec_jiffies,
3027 .strategy = &sysctl_jiffies,
3028 },
3029 {
3030 .ctl_name = NET_IPV4_ROUTE_MIN_PMTU,
3031 .procname = "min_pmtu",
3032 .data = &ip_rt_min_pmtu,
3033 .maxlen = sizeof(int),
3034 .mode = 0644,
3035 .proc_handler = &proc_dointvec,
3036 },
3037 {
3038 .ctl_name = NET_IPV4_ROUTE_MIN_ADVMSS,
3039 .procname = "min_adv_mss",
3040 .data = &ip_rt_min_advmss,
3041 .maxlen = sizeof(int),
3042 .mode = 0644,
3043 .proc_handler = &proc_dointvec,
3044 },
3045 {
3046 .ctl_name = NET_IPV4_ROUTE_SECRET_INTERVAL,
3047 .procname = "secret_interval",
3048 .data = &ip_rt_secret_interval,
3049 .maxlen = sizeof(int),
3050 .mode = 0644,
3051 .proc_handler = &proc_dointvec_jiffies,
3052 .strategy = &sysctl_jiffies,
3053 },
3054 { .ctl_name = 0 }
3055};
3056
3057static __net_initdata struct ctl_path ipv4_route_path[] = {
3058 { .procname = "net", .ctl_name = CTL_NET, },
3059 { .procname = "ipv4", .ctl_name = NET_IPV4, },
3060 { .procname = "route", .ctl_name = NET_IPV4_ROUTE, },
3061 { },
3062};
3063
3064
3065static struct ctl_table ipv4_route_flush_table[] = {
3066 {
3067 .ctl_name = NET_IPV4_ROUTE_FLUSH,
3068 .procname = "flush",
3069 .maxlen = sizeof(int),
3070 .mode = 0200,
3071 .proc_handler = &ipv4_sysctl_rtcache_flush,
3072 .strategy = &ipv4_sysctl_rtcache_flush_strategy,
3073 },
3074 { .ctl_name = 0 },
3075};
3076
3077static __net_init int sysctl_route_net_init(struct net *net)
3078{
3079 struct ctl_table *tbl;
3080
3081 tbl = ipv4_route_flush_table;
3082 if (net != &init_net) {
3083 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3084 if (tbl == NULL)
3085 goto err_dup;
3086 }
3087 tbl[0].extra1 = net;
3088
3089 net->ipv4.route_hdr =
3090 register_net_sysctl_table(net, ipv4_route_path, tbl);
3091 if (net->ipv4.route_hdr == NULL)
3092 goto err_reg;
3093 return 0;
3094
3095err_reg:
3096 if (tbl != ipv4_route_flush_table)
3097 kfree(tbl);
3098err_dup:
3099 return -ENOMEM;
3100}
3101
3102static __net_exit void sysctl_route_net_exit(struct net *net)
3103{
3104 struct ctl_table *tbl;
3105
3106 tbl = net->ipv4.route_hdr->ctl_table_arg;
3107 unregister_net_sysctl_table(net->ipv4.route_hdr);
3108 BUG_ON(tbl == ipv4_route_flush_table);
3109 kfree(tbl);
3110}
3111
3112static __net_initdata struct pernet_operations sysctl_route_ops = {
3113 .init = sysctl_route_net_init,
3114 .exit = sysctl_route_net_exit,
3115};
3116#endif
3117
3118
3119static __net_init int rt_secret_timer_init(struct net *net)
3120{
3121 atomic_set(&net->ipv4.rt_genid,
3122 (int) ((num_physpages ^ (num_physpages>>8)) ^
3123 (jiffies ^ (jiffies >> 7))));
3124
3125 net->ipv4.rt_secret_timer.function = rt_secret_rebuild;
3126 net->ipv4.rt_secret_timer.data = (unsigned long)net;
3127 init_timer_deferrable(&net->ipv4.rt_secret_timer);
3128
3129 net->ipv4.rt_secret_timer.expires =
3130 jiffies + net_random() % ip_rt_secret_interval +
3131 ip_rt_secret_interval;
3132 add_timer(&net->ipv4.rt_secret_timer);
3133 return 0;
3134}
3135
3136static __net_exit void rt_secret_timer_exit(struct net *net)
3137{
3138 del_timer_sync(&net->ipv4.rt_secret_timer);
3139}
3140
3141static __net_initdata struct pernet_operations rt_secret_timer_ops = {
3142 .init = rt_secret_timer_init,
3143 .exit = rt_secret_timer_exit,
3144};
3145
3146
3147#ifdef CONFIG_NET_CLS_ROUTE
3148struct ip_rt_acct *ip_rt_acct __read_mostly;
3149#endif /* CONFIG_NET_CLS_ROUTE */
3150
3151static __initdata unsigned long rhash_entries;
3152static int __init set_rhash_entries(char *str)
3153{
3154 if (!str)
3155 return 0;
3156 rhash_entries = simple_strtoul(str, &str, 0);
3157 return 1;
3158}
3159__setup("rhash_entries=", set_rhash_entries);
3160
3161int __init ip_rt_init(void)
3162{
3163 int rc = 0;
3164
3165#ifdef CONFIG_NET_CLS_ROUTE
3166 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct));
3167 if (!ip_rt_acct)
3168 panic("IP: failed to allocate ip_rt_acct\n");
3169#endif
3170
3171 ipv4_dst_ops.kmem_cachep =
3172 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3173 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3174
3175 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3176
3177 rt_hash_table = (struct rt_hash_bucket *)
3178 alloc_large_system_hash("IP route cache",
3179 sizeof(struct rt_hash_bucket),
3180 rhash_entries,
3181 (num_physpages >= 128 * 1024) ?
3182 15 : 17,
3183 0,
3184 &rt_hash_log,
3185 &rt_hash_mask,
3186 0);
3187 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3188 rt_hash_lock_init();
3189
3190 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3191 ip_rt_max_size = (rt_hash_mask + 1) * 16;
3192
3193 devinet_init();
3194 ip_fib_init();
3195
3196 /* All the timers, started at system startup tend
3197 to synchronize. Perturb it a bit.
3198 */
3199 schedule_delayed_work(&expires_work,
3200 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3201
3202 if (register_pernet_subsys(&rt_secret_timer_ops))
3203 printk(KERN_ERR "Unable to setup rt_secret_timer\n");
3204
3205 if (ip_rt_proc_init())
3206 printk(KERN_ERR "Unable to create route proc files\n");
3207#ifdef CONFIG_XFRM
3208 xfrm_init();
3209 xfrm4_init();
3210#endif
3211 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3212
3213#ifdef CONFIG_SYSCTL
3214 register_pernet_subsys(&sysctl_route_ops);
3215#endif
3216 return rc;
3217}
3218
3219#ifdef CONFIG_SYSCTL
3220/*
3221 * We really need to sanitize the damn ipv4 init order, then all
3222 * this nonsense will go away.
3223 */
3224void __init ip_static_sysctl_init(void)
3225{
3226#ifdef CONFIG_SYSCTL
3227 register_sysctl_paths(ipv4_route_path, ipv4_route_table);
3228#endif
3229}
3230#endif
3231
3232EXPORT_SYMBOL(__ip_select_ident);
3233EXPORT_SYMBOL(ip_route_input);
3234EXPORT_SYMBOL(ip_route_output_key);
Clone of davem's net-next-2.6 tree