ipv4 routing: Ensure that route cache entries are usable and reclaimable with caching...
[linux-2.6.git] / net / ipv4 / route.c
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
118 static int ip_rt_max_size;
119 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
120 static int ip_rt_gc_interval __read_mostly      = 60 * HZ;
121 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
122 static int ip_rt_redirect_number __read_mostly  = 9;
123 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
124 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
125 static int ip_rt_error_cost __read_mostly       = HZ;
126 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
127 static int ip_rt_gc_elasticity __read_mostly    = 8;
128 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
129 static int ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
130 static int ip_rt_min_advmss __read_mostly       = 256;
131 static int ip_rt_secret_interval __read_mostly  = 10 * 60 * HZ;
132 static int rt_chain_length_max __read_mostly    = 20;
133
134 static struct delayed_work expires_work;
135 static unsigned long expires_ljiffies;
136
137 /*
138  *      Interface to generic destination cache.
139  */
140
141 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
142 static void              ipv4_dst_destroy(struct dst_entry *dst);
143 static void              ipv4_dst_ifdown(struct dst_entry *dst,
144                                          struct net_device *dev, int how);
145 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
146 static void              ipv4_link_failure(struct sk_buff *skb);
147 static void              ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
148 static int rt_garbage_collect(struct dst_ops *ops);
149 static void rt_emergency_hash_rebuild(struct net *net);
150
151
152 static struct dst_ops ipv4_dst_ops = {
153         .family =               AF_INET,
154         .protocol =             cpu_to_be16(ETH_P_IP),
155         .gc =                   rt_garbage_collect,
156         .check =                ipv4_dst_check,
157         .destroy =              ipv4_dst_destroy,
158         .ifdown =               ipv4_dst_ifdown,
159         .negative_advice =      ipv4_negative_advice,
160         .link_failure =         ipv4_link_failure,
161         .update_pmtu =          ip_rt_update_pmtu,
162         .local_out =            __ip_local_out,
163         .entries =              ATOMIC_INIT(0),
164 };
165
166 #define ECN_OR_COST(class)      TC_PRIO_##class
167
168 const __u8 ip_tos2prio[16] = {
169         TC_PRIO_BESTEFFORT,
170         ECN_OR_COST(FILLER),
171         TC_PRIO_BESTEFFORT,
172         ECN_OR_COST(BESTEFFORT),
173         TC_PRIO_BULK,
174         ECN_OR_COST(BULK),
175         TC_PRIO_BULK,
176         ECN_OR_COST(BULK),
177         TC_PRIO_INTERACTIVE,
178         ECN_OR_COST(INTERACTIVE),
179         TC_PRIO_INTERACTIVE,
180         ECN_OR_COST(INTERACTIVE),
181         TC_PRIO_INTERACTIVE_BULK,
182         ECN_OR_COST(INTERACTIVE_BULK),
183         TC_PRIO_INTERACTIVE_BULK,
184         ECN_OR_COST(INTERACTIVE_BULK)
185 };
186
187
188 /*
189  * Route cache.
190  */
191
192 /* The locking scheme is rather straight forward:
193  *
194  * 1) Read-Copy Update protects the buckets of the central route hash.
195  * 2) Only writers remove entries, and they hold the lock
196  *    as they look at rtable reference counts.
197  * 3) Only readers acquire references to rtable entries,
198  *    they do so with atomic increments and with the
199  *    lock held.
200  */
201
202 struct rt_hash_bucket {
203         struct rtable   *chain;
204 };
205
206 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
207         defined(CONFIG_PROVE_LOCKING)
208 /*
209  * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
210  * The size of this table is a power of two and depends on the number of CPUS.
211  * (on lockdep we have a quite big spinlock_t, so keep the size down there)
212  */
213 #ifdef CONFIG_LOCKDEP
214 # define RT_HASH_LOCK_SZ        256
215 #else
216 # if NR_CPUS >= 32
217 #  define RT_HASH_LOCK_SZ       4096
218 # elif NR_CPUS >= 16
219 #  define RT_HASH_LOCK_SZ       2048
220 # elif NR_CPUS >= 8
221 #  define RT_HASH_LOCK_SZ       1024
222 # elif NR_CPUS >= 4
223 #  define RT_HASH_LOCK_SZ       512
224 # else
225 #  define RT_HASH_LOCK_SZ       256
226 # endif
227 #endif
228
229 static spinlock_t       *rt_hash_locks;
230 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
231
232 static __init void rt_hash_lock_init(void)
233 {
234         int i;
235
236         rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
237                         GFP_KERNEL);
238         if (!rt_hash_locks)
239                 panic("IP: failed to allocate rt_hash_locks\n");
240
241         for (i = 0; i < RT_HASH_LOCK_SZ; i++)
242                 spin_lock_init(&rt_hash_locks[i]);
243 }
244 #else
245 # define rt_hash_lock_addr(slot) NULL
246
247 static inline void rt_hash_lock_init(void)
248 {
249 }
250 #endif
251
252 static struct rt_hash_bucket    *rt_hash_table __read_mostly;
253 static unsigned                 rt_hash_mask __read_mostly;
254 static unsigned int             rt_hash_log  __read_mostly;
255
256 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
257 #define RT_CACHE_STAT_INC(field) \
258         (__raw_get_cpu_var(rt_cache_stat).field++)
259
260 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
261                 int genid)
262 {
263         return jhash_3words((__force u32)(__be32)(daddr),
264                             (__force u32)(__be32)(saddr),
265                             idx, genid)
266                 & rt_hash_mask;
267 }
268
269 static inline int rt_genid(struct net *net)
270 {
271         return atomic_read(&net->ipv4.rt_genid);
272 }
273
274 #ifdef CONFIG_PROC_FS
275 struct rt_cache_iter_state {
276         struct seq_net_private p;
277         int bucket;
278         int genid;
279 };
280
281 static struct rtable *rt_cache_get_first(struct seq_file *seq)
282 {
283         struct rt_cache_iter_state *st = seq->private;
284         struct rtable *r = NULL;
285
286         for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
287                 if (!rt_hash_table[st->bucket].chain)
288                         continue;
289                 rcu_read_lock_bh();
290                 r = rcu_dereference(rt_hash_table[st->bucket].chain);
291                 while (r) {
292                         if (dev_net(r->u.dst.dev) == seq_file_net(seq) &&
293                             r->rt_genid == st->genid)
294                                 return r;
295                         r = rcu_dereference(r->u.dst.rt_next);
296                 }
297                 rcu_read_unlock_bh();
298         }
299         return r;
300 }
301
302 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
303                                           struct rtable *r)
304 {
305         struct rt_cache_iter_state *st = seq->private;
306
307         r = r->u.dst.rt_next;
308         while (!r) {
309                 rcu_read_unlock_bh();
310                 do {
311                         if (--st->bucket < 0)
312                                 return NULL;
313                 } while (!rt_hash_table[st->bucket].chain);
314                 rcu_read_lock_bh();
315                 r = rt_hash_table[st->bucket].chain;
316         }
317         return rcu_dereference(r);
318 }
319
320 static struct rtable *rt_cache_get_next(struct seq_file *seq,
321                                         struct rtable *r)
322 {
323         struct rt_cache_iter_state *st = seq->private;
324         while ((r = __rt_cache_get_next(seq, r)) != NULL) {
325                 if (dev_net(r->u.dst.dev) != seq_file_net(seq))
326                         continue;
327                 if (r->rt_genid == st->genid)
328                         break;
329         }
330         return r;
331 }
332
333 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
334 {
335         struct rtable *r = rt_cache_get_first(seq);
336
337         if (r)
338                 while (pos && (r = rt_cache_get_next(seq, r)))
339                         --pos;
340         return pos ? NULL : r;
341 }
342
343 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
344 {
345         struct rt_cache_iter_state *st = seq->private;
346         if (*pos)
347                 return rt_cache_get_idx(seq, *pos - 1);
348         st->genid = rt_genid(seq_file_net(seq));
349         return SEQ_START_TOKEN;
350 }
351
352 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
353 {
354         struct rtable *r;
355
356         if (v == SEQ_START_TOKEN)
357                 r = rt_cache_get_first(seq);
358         else
359                 r = rt_cache_get_next(seq, v);
360         ++*pos;
361         return r;
362 }
363
364 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
365 {
366         if (v && v != SEQ_START_TOKEN)
367                 rcu_read_unlock_bh();
368 }
369
370 static int rt_cache_seq_show(struct seq_file *seq, void *v)
371 {
372         if (v == SEQ_START_TOKEN)
373                 seq_printf(seq, "%-127s\n",
374                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
375                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
376                            "HHUptod\tSpecDst");
377         else {
378                 struct rtable *r = v;
379                 int len;
380
381                 seq_printf(seq, "%s\t%08lX\t%08lX\t%8X\t%d\t%u\t%d\t"
382                               "%08lX\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
383                         r->u.dst.dev ? r->u.dst.dev->name : "*",
384                         (unsigned long)r->rt_dst, (unsigned long)r->rt_gateway,
385                         r->rt_flags, atomic_read(&r->u.dst.__refcnt),
386                         r->u.dst.__use, 0, (unsigned long)r->rt_src,
387                         (dst_metric(&r->u.dst, RTAX_ADVMSS) ?
388                              (int)dst_metric(&r->u.dst, RTAX_ADVMSS) + 40 : 0),
389                         dst_metric(&r->u.dst, RTAX_WINDOW),
390                         (int)((dst_metric(&r->u.dst, RTAX_RTT) >> 3) +
391                               dst_metric(&r->u.dst, RTAX_RTTVAR)),
392                         r->fl.fl4_tos,
393                         r->u.dst.hh ? atomic_read(&r->u.dst.hh->hh_refcnt) : -1,
394                         r->u.dst.hh ? (r->u.dst.hh->hh_output ==
395                                        dev_queue_xmit) : 0,
396                         r->rt_spec_dst, &len);
397
398                 seq_printf(seq, "%*s\n", 127 - len, "");
399         }
400         return 0;
401 }
402
403 static const struct seq_operations rt_cache_seq_ops = {
404         .start  = rt_cache_seq_start,
405         .next   = rt_cache_seq_next,
406         .stop   = rt_cache_seq_stop,
407         .show   = rt_cache_seq_show,
408 };
409
410 static int rt_cache_seq_open(struct inode *inode, struct file *file)
411 {
412         return seq_open_net(inode, file, &rt_cache_seq_ops,
413                         sizeof(struct rt_cache_iter_state));
414 }
415
416 static const struct file_operations rt_cache_seq_fops = {
417         .owner   = THIS_MODULE,
418         .open    = rt_cache_seq_open,
419         .read    = seq_read,
420         .llseek  = seq_lseek,
421         .release = seq_release_net,
422 };
423
424
425 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
426 {
427         int cpu;
428
429         if (*pos == 0)
430                 return SEQ_START_TOKEN;
431
432         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
433                 if (!cpu_possible(cpu))
434                         continue;
435                 *pos = cpu+1;
436                 return &per_cpu(rt_cache_stat, cpu);
437         }
438         return NULL;
439 }
440
441 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
442 {
443         int cpu;
444
445         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
446                 if (!cpu_possible(cpu))
447                         continue;
448                 *pos = cpu+1;
449                 return &per_cpu(rt_cache_stat, cpu);
450         }
451         return NULL;
452
453 }
454
455 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
456 {
457
458 }
459
460 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
461 {
462         struct rt_cache_stat *st = v;
463
464         if (v == SEQ_START_TOKEN) {
465                 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");
466                 return 0;
467         }
468
469         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
470                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
471                    atomic_read(&ipv4_dst_ops.entries),
472                    st->in_hit,
473                    st->in_slow_tot,
474                    st->in_slow_mc,
475                    st->in_no_route,
476                    st->in_brd,
477                    st->in_martian_dst,
478                    st->in_martian_src,
479
480                    st->out_hit,
481                    st->out_slow_tot,
482                    st->out_slow_mc,
483
484                    st->gc_total,
485                    st->gc_ignored,
486                    st->gc_goal_miss,
487                    st->gc_dst_overflow,
488                    st->in_hlist_search,
489                    st->out_hlist_search
490                 );
491         return 0;
492 }
493
494 static const struct seq_operations rt_cpu_seq_ops = {
495         .start  = rt_cpu_seq_start,
496         .next   = rt_cpu_seq_next,
497         .stop   = rt_cpu_seq_stop,
498         .show   = rt_cpu_seq_show,
499 };
500
501
502 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
503 {
504         return seq_open(file, &rt_cpu_seq_ops);
505 }
506
507 static const struct file_operations rt_cpu_seq_fops = {
508         .owner   = THIS_MODULE,
509         .open    = rt_cpu_seq_open,
510         .read    = seq_read,
511         .llseek  = seq_lseek,
512         .release = seq_release,
513 };
514
515 #ifdef CONFIG_NET_CLS_ROUTE
516 static int ip_rt_acct_read(char *buffer, char **start, off_t offset,
517                            int length, int *eof, void *data)
518 {
519         unsigned int i;
520
521         if ((offset & 3) || (length & 3))
522                 return -EIO;
523
524         if (offset >= sizeof(struct ip_rt_acct) * 256) {
525                 *eof = 1;
526                 return 0;
527         }
528
529         if (offset + length >= sizeof(struct ip_rt_acct) * 256) {
530                 length = sizeof(struct ip_rt_acct) * 256 - offset;
531                 *eof = 1;
532         }
533
534         offset /= sizeof(u32);
535
536         if (length > 0) {
537                 u32 *dst = (u32 *) buffer;
538
539                 *start = buffer;
540                 memset(dst, 0, length);
541
542                 for_each_possible_cpu(i) {
543                         unsigned int j;
544                         u32 *src;
545
546                         src = ((u32 *) per_cpu_ptr(ip_rt_acct, i)) + offset;
547                         for (j = 0; j < length/4; j++)
548                                 dst[j] += src[j];
549                 }
550         }
551         return length;
552 }
553 #endif
554
555 static int __net_init ip_rt_do_proc_init(struct net *net)
556 {
557         struct proc_dir_entry *pde;
558
559         pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
560                         &rt_cache_seq_fops);
561         if (!pde)
562                 goto err1;
563
564         pde = proc_create("rt_cache", S_IRUGO,
565                           net->proc_net_stat, &rt_cpu_seq_fops);
566         if (!pde)
567                 goto err2;
568
569 #ifdef CONFIG_NET_CLS_ROUTE
570         pde = create_proc_read_entry("rt_acct", 0, net->proc_net,
571                         ip_rt_acct_read, NULL);
572         if (!pde)
573                 goto err3;
574 #endif
575         return 0;
576
577 #ifdef CONFIG_NET_CLS_ROUTE
578 err3:
579         remove_proc_entry("rt_cache", net->proc_net_stat);
580 #endif
581 err2:
582         remove_proc_entry("rt_cache", net->proc_net);
583 err1:
584         return -ENOMEM;
585 }
586
587 static void __net_exit ip_rt_do_proc_exit(struct net *net)
588 {
589         remove_proc_entry("rt_cache", net->proc_net_stat);
590         remove_proc_entry("rt_cache", net->proc_net);
591         remove_proc_entry("rt_acct", net->proc_net);
592 }
593
594 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
595         .init = ip_rt_do_proc_init,
596         .exit = ip_rt_do_proc_exit,
597 };
598
599 static int __init ip_rt_proc_init(void)
600 {
601         return register_pernet_subsys(&ip_rt_proc_ops);
602 }
603
604 #else
605 static inline int ip_rt_proc_init(void)
606 {
607         return 0;
608 }
609 #endif /* CONFIG_PROC_FS */
610
611 static inline void rt_free(struct rtable *rt)
612 {
613         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
614 }
615
616 static inline void rt_drop(struct rtable *rt)
617 {
618         ip_rt_put(rt);
619         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
620 }
621
622 static inline int rt_fast_clean(struct rtable *rth)
623 {
624         /* Kill broadcast/multicast entries very aggresively, if they
625            collide in hash table with more useful entries */
626         return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
627                 rth->fl.iif && rth->u.dst.rt_next;
628 }
629
630 static inline int rt_valuable(struct rtable *rth)
631 {
632         return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
633                 rth->u.dst.expires;
634 }
635
636 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
637 {
638         unsigned long age;
639         int ret = 0;
640
641         if (atomic_read(&rth->u.dst.__refcnt))
642                 goto out;
643
644         ret = 1;
645         if (rth->u.dst.expires &&
646             time_after_eq(jiffies, rth->u.dst.expires))
647                 goto out;
648
649         age = jiffies - rth->u.dst.lastuse;
650         ret = 0;
651         if ((age <= tmo1 && !rt_fast_clean(rth)) ||
652             (age <= tmo2 && rt_valuable(rth)))
653                 goto out;
654         ret = 1;
655 out:    return ret;
656 }
657
658 /* Bits of score are:
659  * 31: very valuable
660  * 30: not quite useless
661  * 29..0: usage counter
662  */
663 static inline u32 rt_score(struct rtable *rt)
664 {
665         u32 score = jiffies - rt->u.dst.lastuse;
666
667         score = ~score & ~(3<<30);
668
669         if (rt_valuable(rt))
670                 score |= (1<<31);
671
672         if (!rt->fl.iif ||
673             !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
674                 score |= (1<<30);
675
676         return score;
677 }
678
679 static inline bool rt_caching(const struct net *net)
680 {
681         return net->ipv4.current_rt_cache_rebuild_count <=
682                 net->ipv4.sysctl_rt_cache_rebuild_count;
683 }
684
685 static inline bool compare_hash_inputs(const struct flowi *fl1,
686                                         const struct flowi *fl2)
687 {
688         return (__force u32)(((fl1->nl_u.ip4_u.daddr ^ fl2->nl_u.ip4_u.daddr) |
689                 (fl1->nl_u.ip4_u.saddr ^ fl2->nl_u.ip4_u.saddr) |
690                 (fl1->iif ^ fl2->iif)) == 0);
691 }
692
693 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
694 {
695         return ((__force u32)((fl1->nl_u.ip4_u.daddr ^ fl2->nl_u.ip4_u.daddr) |
696                 (fl1->nl_u.ip4_u.saddr ^ fl2->nl_u.ip4_u.saddr)) |
697                 (fl1->mark ^ fl2->mark) |
698                 (*(u16 *)&fl1->nl_u.ip4_u.tos ^
699                  *(u16 *)&fl2->nl_u.ip4_u.tos) |
700                 (fl1->oif ^ fl2->oif) |
701                 (fl1->iif ^ fl2->iif)) == 0;
702 }
703
704 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
705 {
706         return dev_net(rt1->u.dst.dev) == dev_net(rt2->u.dst.dev);
707 }
708
709 static inline int rt_is_expired(struct rtable *rth)
710 {
711         return rth->rt_genid != rt_genid(dev_net(rth->u.dst.dev));
712 }
713
714 /*
715  * Perform a full scan of hash table and free all entries.
716  * Can be called by a softirq or a process.
717  * In the later case, we want to be reschedule if necessary
718  */
719 static void rt_do_flush(int process_context)
720 {
721         unsigned int i;
722         struct rtable *rth, *next;
723         struct rtable * tail;
724
725         for (i = 0; i <= rt_hash_mask; i++) {
726                 if (process_context && need_resched())
727                         cond_resched();
728                 rth = rt_hash_table[i].chain;
729                 if (!rth)
730                         continue;
731
732                 spin_lock_bh(rt_hash_lock_addr(i));
733 #ifdef CONFIG_NET_NS
734                 {
735                 struct rtable ** prev, * p;
736
737                 rth = rt_hash_table[i].chain;
738
739                 /* defer releasing the head of the list after spin_unlock */
740                 for (tail = rth; tail; tail = tail->u.dst.rt_next)
741                         if (!rt_is_expired(tail))
742                                 break;
743                 if (rth != tail)
744                         rt_hash_table[i].chain = tail;
745
746                 /* call rt_free on entries after the tail requiring flush */
747                 prev = &rt_hash_table[i].chain;
748                 for (p = *prev; p; p = next) {
749                         next = p->u.dst.rt_next;
750                         if (!rt_is_expired(p)) {
751                                 prev = &p->u.dst.rt_next;
752                         } else {
753                                 *prev = next;
754                                 rt_free(p);
755                         }
756                 }
757                 }
758 #else
759                 rth = rt_hash_table[i].chain;
760                 rt_hash_table[i].chain = NULL;
761                 tail = NULL;
762 #endif
763                 spin_unlock_bh(rt_hash_lock_addr(i));
764
765                 for (; rth != tail; rth = next) {
766                         next = rth->u.dst.rt_next;
767                         rt_free(rth);
768                 }
769         }
770 }
771
772 /*
773  * While freeing expired entries, we compute average chain length
774  * and standard deviation, using fixed-point arithmetic.
775  * This to have an estimation of rt_chain_length_max
776  *  rt_chain_length_max = max(elasticity, AVG + 4*SD)
777  * We use 3 bits for frational part, and 29 (or 61) for magnitude.
778  */
779
780 #define FRACT_BITS 3
781 #define ONE (1UL << FRACT_BITS)
782
783 static void rt_check_expire(void)
784 {
785         static unsigned int rover;
786         unsigned int i = rover, goal;
787         struct rtable *rth, *aux, **rthp;
788         unsigned long samples = 0;
789         unsigned long sum = 0, sum2 = 0;
790         unsigned long delta;
791         u64 mult;
792
793         delta = jiffies - expires_ljiffies;
794         expires_ljiffies = jiffies;
795         mult = ((u64)delta) << rt_hash_log;
796         if (ip_rt_gc_timeout > 1)
797                 do_div(mult, ip_rt_gc_timeout);
798         goal = (unsigned int)mult;
799         if (goal > rt_hash_mask)
800                 goal = rt_hash_mask + 1;
801         for (; goal > 0; goal--) {
802                 unsigned long tmo = ip_rt_gc_timeout;
803                 unsigned long length;
804
805                 i = (i + 1) & rt_hash_mask;
806                 rthp = &rt_hash_table[i].chain;
807
808                 if (need_resched())
809                         cond_resched();
810
811                 samples++;
812
813                 if (*rthp == NULL)
814                         continue;
815                 length = 0;
816                 spin_lock_bh(rt_hash_lock_addr(i));
817                 while ((rth = *rthp) != NULL) {
818                         prefetch(rth->u.dst.rt_next);
819                         if (rt_is_expired(rth)) {
820                                 *rthp = rth->u.dst.rt_next;
821                                 rt_free(rth);
822                                 continue;
823                         }
824                         if (rth->u.dst.expires) {
825                                 /* Entry is expired even if it is in use */
826                                 if (time_before_eq(jiffies, rth->u.dst.expires)) {
827 nofree:
828                                         tmo >>= 1;
829                                         rthp = &rth->u.dst.rt_next;
830                                         /*
831                                          * We only count entries on
832                                          * a chain with equal hash inputs once
833                                          * so that entries for different QOS
834                                          * levels, and other non-hash input
835                                          * attributes don't unfairly skew
836                                          * the length computation
837                                          */
838                                         for (aux = rt_hash_table[i].chain;;) {
839                                                 if (aux == rth) {
840                                                         length += ONE;
841                                                         break;
842                                                 }
843                                                 if (compare_hash_inputs(&aux->fl, &rth->fl))
844                                                         break;
845                                                 aux = aux->u.dst.rt_next;
846                                         }
847                                         continue;
848                                 }
849                         } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout))
850                                 goto nofree;
851
852                         /* Cleanup aged off entries. */
853                         *rthp = rth->u.dst.rt_next;
854                         rt_free(rth);
855                 }
856                 spin_unlock_bh(rt_hash_lock_addr(i));
857                 sum += length;
858                 sum2 += length*length;
859         }
860         if (samples) {
861                 unsigned long avg = sum / samples;
862                 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
863                 rt_chain_length_max = max_t(unsigned long,
864                                         ip_rt_gc_elasticity,
865                                         (avg + 4*sd) >> FRACT_BITS);
866         }
867         rover = i;
868 }
869
870 /*
871  * rt_worker_func() is run in process context.
872  * we call rt_check_expire() to scan part of the hash table
873  */
874 static void rt_worker_func(struct work_struct *work)
875 {
876         rt_check_expire();
877         schedule_delayed_work(&expires_work, ip_rt_gc_interval);
878 }
879
880 /*
881  * Pertubation of rt_genid by a small quantity [1..256]
882  * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
883  * many times (2^24) without giving recent rt_genid.
884  * Jenkins hash is strong enough that litle changes of rt_genid are OK.
885  */
886 static void rt_cache_invalidate(struct net *net)
887 {
888         unsigned char shuffle;
889
890         get_random_bytes(&shuffle, sizeof(shuffle));
891         atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
892 }
893
894 /*
895  * delay < 0  : invalidate cache (fast : entries will be deleted later)
896  * delay >= 0 : invalidate & flush cache (can be long)
897  */
898 void rt_cache_flush(struct net *net, int delay)
899 {
900         rt_cache_invalidate(net);
901         if (delay >= 0)
902                 rt_do_flush(!in_softirq());
903 }
904
905 /*
906  * We change rt_genid and let gc do the cleanup
907  */
908 static void rt_secret_rebuild(unsigned long __net)
909 {
910         struct net *net = (struct net *)__net;
911         rt_cache_invalidate(net);
912         mod_timer(&net->ipv4.rt_secret_timer, jiffies + ip_rt_secret_interval);
913 }
914
915 static void rt_secret_rebuild_oneshot(struct net *net)
916 {
917         del_timer_sync(&net->ipv4.rt_secret_timer);
918         rt_cache_invalidate(net);
919         if (ip_rt_secret_interval) {
920                 net->ipv4.rt_secret_timer.expires += ip_rt_secret_interval;
921                 add_timer(&net->ipv4.rt_secret_timer);
922         }
923 }
924
925 static void rt_emergency_hash_rebuild(struct net *net)
926 {
927         if (net_ratelimit()) {
928                 printk(KERN_WARNING "Route hash chain too long!\n");
929                 printk(KERN_WARNING "Adjust your secret_interval!\n");
930         }
931
932         rt_secret_rebuild_oneshot(net);
933 }
934
935 /*
936    Short description of GC goals.
937
938    We want to build algorithm, which will keep routing cache
939    at some equilibrium point, when number of aged off entries
940    is kept approximately equal to newly generated ones.
941
942    Current expiration strength is variable "expire".
943    We try to adjust it dynamically, so that if networking
944    is idle expires is large enough to keep enough of warm entries,
945    and when load increases it reduces to limit cache size.
946  */
947
948 static int rt_garbage_collect(struct dst_ops *ops)
949 {
950         static unsigned long expire = RT_GC_TIMEOUT;
951         static unsigned long last_gc;
952         static int rover;
953         static int equilibrium;
954         struct rtable *rth, **rthp;
955         unsigned long now = jiffies;
956         int goal;
957
958         /*
959          * Garbage collection is pretty expensive,
960          * do not make it too frequently.
961          */
962
963         RT_CACHE_STAT_INC(gc_total);
964
965         if (now - last_gc < ip_rt_gc_min_interval &&
966             atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size) {
967                 RT_CACHE_STAT_INC(gc_ignored);
968                 goto out;
969         }
970
971         /* Calculate number of entries, which we want to expire now. */
972         goal = atomic_read(&ipv4_dst_ops.entries) -
973                 (ip_rt_gc_elasticity << rt_hash_log);
974         if (goal <= 0) {
975                 if (equilibrium < ipv4_dst_ops.gc_thresh)
976                         equilibrium = ipv4_dst_ops.gc_thresh;
977                 goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
978                 if (goal > 0) {
979                         equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
980                         goal = atomic_read(&ipv4_dst_ops.entries) - equilibrium;
981                 }
982         } else {
983                 /* We are in dangerous area. Try to reduce cache really
984                  * aggressively.
985                  */
986                 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
987                 equilibrium = atomic_read(&ipv4_dst_ops.entries) - goal;
988         }
989
990         if (now - last_gc >= ip_rt_gc_min_interval)
991                 last_gc = now;
992
993         if (goal <= 0) {
994                 equilibrium += goal;
995                 goto work_done;
996         }
997
998         do {
999                 int i, k;
1000
1001                 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
1002                         unsigned long tmo = expire;
1003
1004                         k = (k + 1) & rt_hash_mask;
1005                         rthp = &rt_hash_table[k].chain;
1006                         spin_lock_bh(rt_hash_lock_addr(k));
1007                         while ((rth = *rthp) != NULL) {
1008                                 if (!rt_is_expired(rth) &&
1009                                         !rt_may_expire(rth, tmo, expire)) {
1010                                         tmo >>= 1;
1011                                         rthp = &rth->u.dst.rt_next;
1012                                         continue;
1013                                 }
1014                                 *rthp = rth->u.dst.rt_next;
1015                                 rt_free(rth);
1016                                 goal--;
1017                         }
1018                         spin_unlock_bh(rt_hash_lock_addr(k));
1019                         if (goal <= 0)
1020                                 break;
1021                 }
1022                 rover = k;
1023
1024                 if (goal <= 0)
1025                         goto work_done;
1026
1027                 /* Goal is not achieved. We stop process if:
1028
1029                    - if expire reduced to zero. Otherwise, expire is halfed.
1030                    - if table is not full.
1031                    - if we are called from interrupt.
1032                    - jiffies check is just fallback/debug loop breaker.
1033                      We will not spin here for long time in any case.
1034                  */
1035
1036                 RT_CACHE_STAT_INC(gc_goal_miss);
1037
1038                 if (expire == 0)
1039                         break;
1040
1041                 expire >>= 1;
1042 #if RT_CACHE_DEBUG >= 2
1043                 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
1044                                 atomic_read(&ipv4_dst_ops.entries), goal, i);
1045 #endif
1046
1047                 if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
1048                         goto out;
1049         } while (!in_softirq() && time_before_eq(jiffies, now));
1050
1051         if (atomic_read(&ipv4_dst_ops.entries) < ip_rt_max_size)
1052                 goto out;
1053         if (net_ratelimit())
1054                 printk(KERN_WARNING "dst cache overflow\n");
1055         RT_CACHE_STAT_INC(gc_dst_overflow);
1056         return 1;
1057
1058 work_done:
1059         expire += ip_rt_gc_min_interval;
1060         if (expire > ip_rt_gc_timeout ||
1061             atomic_read(&ipv4_dst_ops.entries) < ipv4_dst_ops.gc_thresh)
1062                 expire = ip_rt_gc_timeout;
1063 #if RT_CACHE_DEBUG >= 2
1064         printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
1065                         atomic_read(&ipv4_dst_ops.entries), goal, rover);
1066 #endif
1067 out:    return 0;
1068 }
1069
1070 static int rt_intern_hash(unsigned hash, struct rtable *rt,
1071                           struct rtable **rp, struct sk_buff *skb)
1072 {
1073         struct rtable   *rth, **rthp;
1074         unsigned long   now;
1075         struct rtable *cand, **candp;
1076         u32             min_score;
1077         int             chain_length;
1078         int attempts = !in_softirq();
1079
1080 restart:
1081         chain_length = 0;
1082         min_score = ~(u32)0;
1083         cand = NULL;
1084         candp = NULL;
1085         now = jiffies;
1086
1087         if (!rt_caching(dev_net(rt->u.dst.dev))) {
1088                 /*
1089                  * If we're not caching, just tell the caller we
1090                  * were successful and don't touch the route.  The
1091                  * caller hold the sole reference to the cache entry, and
1092                  * it will be released when the caller is done with it.
1093                  * If we drop it here, the callers have no way to resolve routes
1094                  * when we're not caching.  Instead, just point *rp at rt, so
1095                  * the caller gets a single use out of the route
1096                  * Note that we do rt_free on this new route entry, so that
1097                  * once its refcount hits zero, we are still able to reap it
1098                  * (Thanks Alexey)
1099                  * Note also the rt_free uses call_rcu.  We don't actually
1100                  * need rcu protection here, this is just our path to get
1101                  * on the route gc list.
1102                  */
1103
1104                 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1105                         int err = arp_bind_neighbour(&rt->u.dst);
1106                         if (err) {
1107                                 if (net_ratelimit())
1108                                         printk(KERN_WARNING
1109                                             "Neighbour table failure & not caching routes.\n");
1110                                 rt_drop(rt);
1111                                 return err;
1112                         }
1113                 }
1114
1115                 rt_free(rt);
1116                 goto skip_hashing;
1117         }
1118
1119         rthp = &rt_hash_table[hash].chain;
1120
1121         spin_lock_bh(rt_hash_lock_addr(hash));
1122         while ((rth = *rthp) != NULL) {
1123                 if (rt_is_expired(rth)) {
1124                         *rthp = rth->u.dst.rt_next;
1125                         rt_free(rth);
1126                         continue;
1127                 }
1128                 if (compare_keys(&rth->fl, &rt->fl) && compare_netns(rth, rt)) {
1129                         /* Put it first */
1130                         *rthp = rth->u.dst.rt_next;
1131                         /*
1132                          * Since lookup is lockfree, the deletion
1133                          * must be visible to another weakly ordered CPU before
1134                          * the insertion at the start of the hash chain.
1135                          */
1136                         rcu_assign_pointer(rth->u.dst.rt_next,
1137                                            rt_hash_table[hash].chain);
1138                         /*
1139                          * Since lookup is lockfree, the update writes
1140                          * must be ordered for consistency on SMP.
1141                          */
1142                         rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1143
1144                         dst_use(&rth->u.dst, now);
1145                         spin_unlock_bh(rt_hash_lock_addr(hash));
1146
1147                         rt_drop(rt);
1148                         if (rp)
1149                                 *rp = rth;
1150                         else
1151                                 skb_dst_set(skb, &rth->u.dst);
1152                         return 0;
1153                 }
1154
1155                 if (!atomic_read(&rth->u.dst.__refcnt)) {
1156                         u32 score = rt_score(rth);
1157
1158                         if (score <= min_score) {
1159                                 cand = rth;
1160                                 candp = rthp;
1161                                 min_score = score;
1162                         }
1163                 }
1164
1165                 chain_length++;
1166
1167                 rthp = &rth->u.dst.rt_next;
1168         }
1169
1170         if (cand) {
1171                 /* ip_rt_gc_elasticity used to be average length of chain
1172                  * length, when exceeded gc becomes really aggressive.
1173                  *
1174                  * The second limit is less certain. At the moment it allows
1175                  * only 2 entries per bucket. We will see.
1176                  */
1177                 if (chain_length > ip_rt_gc_elasticity) {
1178                         *candp = cand->u.dst.rt_next;
1179                         rt_free(cand);
1180                 }
1181         } else {
1182                 if (chain_length > rt_chain_length_max) {
1183                         struct net *net = dev_net(rt->u.dst.dev);
1184                         int num = ++net->ipv4.current_rt_cache_rebuild_count;
1185                         if (!rt_caching(dev_net(rt->u.dst.dev))) {
1186                                 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1187                                         rt->u.dst.dev->name, num);
1188                         }
1189                         rt_emergency_hash_rebuild(dev_net(rt->u.dst.dev));
1190                 }
1191         }
1192
1193         /* Try to bind route to arp only if it is output
1194            route or unicast forwarding path.
1195          */
1196         if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1197                 int err = arp_bind_neighbour(&rt->u.dst);
1198                 if (err) {
1199                         spin_unlock_bh(rt_hash_lock_addr(hash));
1200
1201                         if (err != -ENOBUFS) {
1202                                 rt_drop(rt);
1203                                 return err;
1204                         }
1205
1206                         /* Neighbour tables are full and nothing
1207                            can be released. Try to shrink route cache,
1208                            it is most likely it holds some neighbour records.
1209                          */
1210                         if (attempts-- > 0) {
1211                                 int saved_elasticity = ip_rt_gc_elasticity;
1212                                 int saved_int = ip_rt_gc_min_interval;
1213                                 ip_rt_gc_elasticity     = 1;
1214                                 ip_rt_gc_min_interval   = 0;
1215                                 rt_garbage_collect(&ipv4_dst_ops);
1216                                 ip_rt_gc_min_interval   = saved_int;
1217                                 ip_rt_gc_elasticity     = saved_elasticity;
1218                                 goto restart;
1219                         }
1220
1221                         if (net_ratelimit())
1222                                 printk(KERN_WARNING "Neighbour table overflow.\n");
1223                         rt_drop(rt);
1224                         return -ENOBUFS;
1225                 }
1226         }
1227
1228         rt->u.dst.rt_next = rt_hash_table[hash].chain;
1229
1230 #if RT_CACHE_DEBUG >= 2
1231         if (rt->u.dst.rt_next) {
1232                 struct rtable *trt;
1233                 printk(KERN_DEBUG "rt_cache @%02x: %pI4",
1234                        hash, &rt->rt_dst);
1235                 for (trt = rt->u.dst.rt_next; trt; trt = trt->u.dst.rt_next)
1236                         printk(" . %pI4", &trt->rt_dst);
1237                 printk("\n");
1238         }
1239 #endif
1240         /*
1241          * Since lookup is lockfree, we must make sure
1242          * previous writes to rt are comitted to memory
1243          * before making rt visible to other CPUS.
1244          */
1245         rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1246
1247         spin_unlock_bh(rt_hash_lock_addr(hash));
1248
1249 skip_hashing:
1250         if (rp)
1251                 *rp = rt;
1252         else
1253                 skb_dst_set(skb, &rt->u.dst);
1254         return 0;
1255 }
1256
1257 void rt_bind_peer(struct rtable *rt, int create)
1258 {
1259         static DEFINE_SPINLOCK(rt_peer_lock);
1260         struct inet_peer *peer;
1261
1262         peer = inet_getpeer(rt->rt_dst, create);
1263
1264         spin_lock_bh(&rt_peer_lock);
1265         if (rt->peer == NULL) {
1266                 rt->peer = peer;
1267                 peer = NULL;
1268         }
1269         spin_unlock_bh(&rt_peer_lock);
1270         if (peer)
1271                 inet_putpeer(peer);
1272 }
1273
1274 /*
1275  * Peer allocation may fail only in serious out-of-memory conditions.  However
1276  * we still can generate some output.
1277  * Random ID selection looks a bit dangerous because we have no chances to
1278  * select ID being unique in a reasonable period of time.
1279  * But broken packet identifier may be better than no packet at all.
1280  */
1281 static void ip_select_fb_ident(struct iphdr *iph)
1282 {
1283         static DEFINE_SPINLOCK(ip_fb_id_lock);
1284         static u32 ip_fallback_id;
1285         u32 salt;
1286
1287         spin_lock_bh(&ip_fb_id_lock);
1288         salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1289         iph->id = htons(salt & 0xFFFF);
1290         ip_fallback_id = salt;
1291         spin_unlock_bh(&ip_fb_id_lock);
1292 }
1293
1294 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1295 {
1296         struct rtable *rt = (struct rtable *) dst;
1297
1298         if (rt) {
1299                 if (rt->peer == NULL)
1300                         rt_bind_peer(rt, 1);
1301
1302                 /* If peer is attached to destination, it is never detached,
1303                    so that we need not to grab a lock to dereference it.
1304                  */
1305                 if (rt->peer) {
1306                         iph->id = htons(inet_getid(rt->peer, more));
1307                         return;
1308                 }
1309         } else
1310                 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1311                        __builtin_return_address(0));
1312
1313         ip_select_fb_ident(iph);
1314 }
1315
1316 static void rt_del(unsigned hash, struct rtable *rt)
1317 {
1318         struct rtable **rthp, *aux;
1319
1320         rthp = &rt_hash_table[hash].chain;
1321         spin_lock_bh(rt_hash_lock_addr(hash));
1322         ip_rt_put(rt);
1323         while ((aux = *rthp) != NULL) {
1324                 if (aux == rt || rt_is_expired(aux)) {
1325                         *rthp = aux->u.dst.rt_next;
1326                         rt_free(aux);
1327                         continue;
1328                 }
1329                 rthp = &aux->u.dst.rt_next;
1330         }
1331         spin_unlock_bh(rt_hash_lock_addr(hash));
1332 }
1333
1334 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1335                     __be32 saddr, struct net_device *dev)
1336 {
1337         int i, k;
1338         struct in_device *in_dev = in_dev_get(dev);
1339         struct rtable *rth, **rthp;
1340         __be32  skeys[2] = { saddr, 0 };
1341         int  ikeys[2] = { dev->ifindex, 0 };
1342         struct netevent_redirect netevent;
1343         struct net *net;
1344
1345         if (!in_dev)
1346                 return;
1347
1348         net = dev_net(dev);
1349         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev)
1350             || ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw)
1351             || ipv4_is_zeronet(new_gw))
1352                 goto reject_redirect;
1353
1354         if (!rt_caching(net))
1355                 goto reject_redirect;
1356
1357         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1358                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1359                         goto reject_redirect;
1360                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1361                         goto reject_redirect;
1362         } else {
1363                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1364                         goto reject_redirect;
1365         }
1366
1367         for (i = 0; i < 2; i++) {
1368                 for (k = 0; k < 2; k++) {
1369                         unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1370                                                 rt_genid(net));
1371
1372                         rthp=&rt_hash_table[hash].chain;
1373
1374                         rcu_read_lock();
1375                         while ((rth = rcu_dereference(*rthp)) != NULL) {
1376                                 struct rtable *rt;
1377
1378                                 if (rth->fl.fl4_dst != daddr ||
1379                                     rth->fl.fl4_src != skeys[i] ||
1380                                     rth->fl.oif != ikeys[k] ||
1381                                     rth->fl.iif != 0 ||
1382                                     rt_is_expired(rth) ||
1383                                     !net_eq(dev_net(rth->u.dst.dev), net)) {
1384                                         rthp = &rth->u.dst.rt_next;
1385                                         continue;
1386                                 }
1387
1388                                 if (rth->rt_dst != daddr ||
1389                                     rth->rt_src != saddr ||
1390                                     rth->u.dst.error ||
1391                                     rth->rt_gateway != old_gw ||
1392                                     rth->u.dst.dev != dev)
1393                                         break;
1394
1395                                 dst_hold(&rth->u.dst);
1396                                 rcu_read_unlock();
1397
1398                                 rt = dst_alloc(&ipv4_dst_ops);
1399                                 if (rt == NULL) {
1400                                         ip_rt_put(rth);
1401                                         in_dev_put(in_dev);
1402                                         return;
1403                                 }
1404
1405                                 /* Copy all the information. */
1406                                 *rt = *rth;
1407                                 rt->u.dst.__use         = 1;
1408                                 atomic_set(&rt->u.dst.__refcnt, 1);
1409                                 rt->u.dst.child         = NULL;
1410                                 if (rt->u.dst.dev)
1411                                         dev_hold(rt->u.dst.dev);
1412                                 if (rt->idev)
1413                                         in_dev_hold(rt->idev);
1414                                 rt->u.dst.obsolete      = 0;
1415                                 rt->u.dst.lastuse       = jiffies;
1416                                 rt->u.dst.path          = &rt->u.dst;
1417                                 rt->u.dst.neighbour     = NULL;
1418                                 rt->u.dst.hh            = NULL;
1419 #ifdef CONFIG_XFRM
1420                                 rt->u.dst.xfrm          = NULL;
1421 #endif
1422                                 rt->rt_genid            = rt_genid(net);
1423                                 rt->rt_flags            |= RTCF_REDIRECTED;
1424
1425                                 /* Gateway is different ... */
1426                                 rt->rt_gateway          = new_gw;
1427
1428                                 /* Redirect received -> path was valid */
1429                                 dst_confirm(&rth->u.dst);
1430
1431                                 if (rt->peer)
1432                                         atomic_inc(&rt->peer->refcnt);
1433
1434                                 if (arp_bind_neighbour(&rt->u.dst) ||
1435                                     !(rt->u.dst.neighbour->nud_state &
1436                                             NUD_VALID)) {
1437                                         if (rt->u.dst.neighbour)
1438                                                 neigh_event_send(rt->u.dst.neighbour, NULL);
1439                                         ip_rt_put(rth);
1440                                         rt_drop(rt);
1441                                         goto do_next;
1442                                 }
1443
1444                                 netevent.old = &rth->u.dst;
1445                                 netevent.new = &rt->u.dst;
1446                                 call_netevent_notifiers(NETEVENT_REDIRECT,
1447                                                         &netevent);
1448
1449                                 rt_del(hash, rth);
1450                                 if (!rt_intern_hash(hash, rt, &rt, NULL))
1451                                         ip_rt_put(rt);
1452                                 goto do_next;
1453                         }
1454                         rcu_read_unlock();
1455                 do_next:
1456                         ;
1457                 }
1458         }
1459         in_dev_put(in_dev);
1460         return;
1461
1462 reject_redirect:
1463 #ifdef CONFIG_IP_ROUTE_VERBOSE
1464         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1465                 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1466                         "  Advised path = %pI4 -> %pI4\n",
1467                        &old_gw, dev->name, &new_gw,
1468                        &saddr, &daddr);
1469 #endif
1470         in_dev_put(in_dev);
1471 }
1472
1473 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1474 {
1475         struct rtable *rt = (struct rtable *)dst;
1476         struct dst_entry *ret = dst;
1477
1478         if (rt) {
1479                 if (dst->obsolete) {
1480                         ip_rt_put(rt);
1481                         ret = NULL;
1482                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1483                            rt->u.dst.expires) {
1484                         unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1485                                                 rt->fl.oif,
1486                                                 rt_genid(dev_net(dst->dev)));
1487 #if RT_CACHE_DEBUG >= 1
1488                         printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1489                                 &rt->rt_dst, rt->fl.fl4_tos);
1490 #endif
1491                         rt_del(hash, rt);
1492                         ret = NULL;
1493                 }
1494         }
1495         return ret;
1496 }
1497
1498 /*
1499  * Algorithm:
1500  *      1. The first ip_rt_redirect_number redirects are sent
1501  *         with exponential backoff, then we stop sending them at all,
1502  *         assuming that the host ignores our redirects.
1503  *      2. If we did not see packets requiring redirects
1504  *         during ip_rt_redirect_silence, we assume that the host
1505  *         forgot redirected route and start to send redirects again.
1506  *
1507  * This algorithm is much cheaper and more intelligent than dumb load limiting
1508  * in icmp.c.
1509  *
1510  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1511  * and "frag. need" (breaks PMTU discovery) in icmp.c.
1512  */
1513
1514 void ip_rt_send_redirect(struct sk_buff *skb)
1515 {
1516         struct rtable *rt = skb_rtable(skb);
1517         struct in_device *in_dev = in_dev_get(rt->u.dst.dev);
1518
1519         if (!in_dev)
1520                 return;
1521
1522         if (!IN_DEV_TX_REDIRECTS(in_dev))
1523                 goto out;
1524
1525         /* No redirected packets during ip_rt_redirect_silence;
1526          * reset the algorithm.
1527          */
1528         if (time_after(jiffies, rt->u.dst.rate_last + ip_rt_redirect_silence))
1529                 rt->u.dst.rate_tokens = 0;
1530
1531         /* Too many ignored redirects; do not send anything
1532          * set u.dst.rate_last to the last seen redirected packet.
1533          */
1534         if (rt->u.dst.rate_tokens >= ip_rt_redirect_number) {
1535                 rt->u.dst.rate_last = jiffies;
1536                 goto out;
1537         }
1538
1539         /* Check for load limit; set rate_last to the latest sent
1540          * redirect.
1541          */
1542         if (rt->u.dst.rate_tokens == 0 ||
1543             time_after(jiffies,
1544                        (rt->u.dst.rate_last +
1545                         (ip_rt_redirect_load << rt->u.dst.rate_tokens)))) {
1546                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1547                 rt->u.dst.rate_last = jiffies;
1548                 ++rt->u.dst.rate_tokens;
1549 #ifdef CONFIG_IP_ROUTE_VERBOSE
1550                 if (IN_DEV_LOG_MARTIANS(in_dev) &&
1551                     rt->u.dst.rate_tokens == ip_rt_redirect_number &&
1552                     net_ratelimit())
1553                         printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1554                                 &rt->rt_src, rt->rt_iif,
1555                                 &rt->rt_dst, &rt->rt_gateway);
1556 #endif
1557         }
1558 out:
1559         in_dev_put(in_dev);
1560 }
1561
1562 static int ip_error(struct sk_buff *skb)
1563 {
1564         struct rtable *rt = skb_rtable(skb);
1565         unsigned long now;
1566         int code;
1567
1568         switch (rt->u.dst.error) {
1569                 case EINVAL:
1570                 default:
1571                         goto out;
1572                 case EHOSTUNREACH:
1573                         code = ICMP_HOST_UNREACH;
1574                         break;
1575                 case ENETUNREACH:
1576                         code = ICMP_NET_UNREACH;
1577                         IP_INC_STATS_BH(dev_net(rt->u.dst.dev),
1578                                         IPSTATS_MIB_INNOROUTES);
1579                         break;
1580                 case EACCES:
1581                         code = ICMP_PKT_FILTERED;
1582                         break;
1583         }
1584
1585         now = jiffies;
1586         rt->u.dst.rate_tokens += now - rt->u.dst.rate_last;
1587         if (rt->u.dst.rate_tokens > ip_rt_error_burst)
1588                 rt->u.dst.rate_tokens = ip_rt_error_burst;
1589         rt->u.dst.rate_last = now;
1590         if (rt->u.dst.rate_tokens >= ip_rt_error_cost) {
1591                 rt->u.dst.rate_tokens -= ip_rt_error_cost;
1592                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1593         }
1594
1595 out:    kfree_skb(skb);
1596         return 0;
1597 }
1598
1599 /*
1600  *      The last two values are not from the RFC but
1601  *      are needed for AMPRnet AX.25 paths.
1602  */
1603
1604 static const unsigned short mtu_plateau[] =
1605 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1606
1607 static inline unsigned short guess_mtu(unsigned short old_mtu)
1608 {
1609         int i;
1610
1611         for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1612                 if (old_mtu > mtu_plateau[i])
1613                         return mtu_plateau[i];
1614         return 68;
1615 }
1616
1617 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1618                                  unsigned short new_mtu,
1619                                  struct net_device *dev)
1620 {
1621         int i, k;
1622         unsigned short old_mtu = ntohs(iph->tot_len);
1623         struct rtable *rth;
1624         int  ikeys[2] = { dev->ifindex, 0 };
1625         __be32  skeys[2] = { iph->saddr, 0, };
1626         __be32  daddr = iph->daddr;
1627         unsigned short est_mtu = 0;
1628
1629         if (ipv4_config.no_pmtu_disc)
1630                 return 0;
1631
1632         for (k = 0; k < 2; k++) {
1633                 for (i = 0; i < 2; i++) {
1634                         unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1635                                                 rt_genid(net));
1636
1637                         rcu_read_lock();
1638                         for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1639                              rth = rcu_dereference(rth->u.dst.rt_next)) {
1640                                 unsigned short mtu = new_mtu;
1641
1642                                 if (rth->fl.fl4_dst != daddr ||
1643                                     rth->fl.fl4_src != skeys[i] ||
1644                                     rth->rt_dst != daddr ||
1645                                     rth->rt_src != iph->saddr ||
1646                                     rth->fl.oif != ikeys[k] ||
1647                                     rth->fl.iif != 0 ||
1648                                     dst_metric_locked(&rth->u.dst, RTAX_MTU) ||
1649                                     !net_eq(dev_net(rth->u.dst.dev), net) ||
1650                                     rt_is_expired(rth))
1651                                         continue;
1652
1653                                 if (new_mtu < 68 || new_mtu >= old_mtu) {
1654
1655                                         /* BSD 4.2 compatibility hack :-( */
1656                                         if (mtu == 0 &&
1657                                             old_mtu >= dst_mtu(&rth->u.dst) &&
1658                                             old_mtu >= 68 + (iph->ihl << 2))
1659                                                 old_mtu -= iph->ihl << 2;
1660
1661                                         mtu = guess_mtu(old_mtu);
1662                                 }
1663                                 if (mtu <= dst_mtu(&rth->u.dst)) {
1664                                         if (mtu < dst_mtu(&rth->u.dst)) {
1665                                                 dst_confirm(&rth->u.dst);
1666                                                 if (mtu < ip_rt_min_pmtu) {
1667                                                         mtu = ip_rt_min_pmtu;
1668                                                         rth->u.dst.metrics[RTAX_LOCK-1] |=
1669                                                                 (1 << RTAX_MTU);
1670                                                 }
1671                                                 rth->u.dst.metrics[RTAX_MTU-1] = mtu;
1672                                                 dst_set_expires(&rth->u.dst,
1673                                                         ip_rt_mtu_expires);
1674                                         }
1675                                         est_mtu = mtu;
1676                                 }
1677                         }
1678                         rcu_read_unlock();
1679                 }
1680         }
1681         return est_mtu ? : new_mtu;
1682 }
1683
1684 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1685 {
1686         if (dst_mtu(dst) > mtu && mtu >= 68 &&
1687             !(dst_metric_locked(dst, RTAX_MTU))) {
1688                 if (mtu < ip_rt_min_pmtu) {
1689                         mtu = ip_rt_min_pmtu;
1690                         dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1691                 }
1692                 dst->metrics[RTAX_MTU-1] = mtu;
1693                 dst_set_expires(dst, ip_rt_mtu_expires);
1694                 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1695         }
1696 }
1697
1698 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1699 {
1700         return NULL;
1701 }
1702
1703 static void ipv4_dst_destroy(struct dst_entry *dst)
1704 {
1705         struct rtable *rt = (struct rtable *) dst;
1706         struct inet_peer *peer = rt->peer;
1707         struct in_device *idev = rt->idev;
1708
1709         if (peer) {
1710                 rt->peer = NULL;
1711                 inet_putpeer(peer);
1712         }
1713
1714         if (idev) {
1715                 rt->idev = NULL;
1716                 in_dev_put(idev);
1717         }
1718 }
1719
1720 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1721                             int how)
1722 {
1723         struct rtable *rt = (struct rtable *) dst;
1724         struct in_device *idev = rt->idev;
1725         if (dev != dev_net(dev)->loopback_dev && idev && idev->dev == dev) {
1726                 struct in_device *loopback_idev =
1727                         in_dev_get(dev_net(dev)->loopback_dev);
1728                 if (loopback_idev) {
1729                         rt->idev = loopback_idev;
1730                         in_dev_put(idev);
1731                 }
1732         }
1733 }
1734
1735 static void ipv4_link_failure(struct sk_buff *skb)
1736 {
1737         struct rtable *rt;
1738
1739         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1740
1741         rt = skb_rtable(skb);
1742         if (rt)
1743                 dst_set_expires(&rt->u.dst, 0);
1744 }
1745
1746 static int ip_rt_bug(struct sk_buff *skb)
1747 {
1748         printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1749                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1750                 skb->dev ? skb->dev->name : "?");
1751         kfree_skb(skb);
1752         return 0;
1753 }
1754
1755 /*
1756    We do not cache source address of outgoing interface,
1757    because it is used only by IP RR, TS and SRR options,
1758    so that it out of fast path.
1759
1760    BTW remember: "addr" is allowed to be not aligned
1761    in IP options!
1762  */
1763
1764 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1765 {
1766         __be32 src;
1767         struct fib_result res;
1768
1769         if (rt->fl.iif == 0)
1770                 src = rt->rt_src;
1771         else if (fib_lookup(dev_net(rt->u.dst.dev), &rt->fl, &res) == 0) {
1772                 src = FIB_RES_PREFSRC(res);
1773                 fib_res_put(&res);
1774         } else
1775                 src = inet_select_addr(rt->u.dst.dev, rt->rt_gateway,
1776                                         RT_SCOPE_UNIVERSE);
1777         memcpy(addr, &src, 4);
1778 }
1779
1780 #ifdef CONFIG_NET_CLS_ROUTE
1781 static void set_class_tag(struct rtable *rt, u32 tag)
1782 {
1783         if (!(rt->u.dst.tclassid & 0xFFFF))
1784                 rt->u.dst.tclassid |= tag & 0xFFFF;
1785         if (!(rt->u.dst.tclassid & 0xFFFF0000))
1786                 rt->u.dst.tclassid |= tag & 0xFFFF0000;
1787 }
1788 #endif
1789
1790 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1791 {
1792         struct fib_info *fi = res->fi;
1793
1794         if (fi) {
1795                 if (FIB_RES_GW(*res) &&
1796                     FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1797                         rt->rt_gateway = FIB_RES_GW(*res);
1798                 memcpy(rt->u.dst.metrics, fi->fib_metrics,
1799                        sizeof(rt->u.dst.metrics));
1800                 if (fi->fib_mtu == 0) {
1801                         rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
1802                         if (dst_metric_locked(&rt->u.dst, RTAX_MTU) &&
1803                             rt->rt_gateway != rt->rt_dst &&
1804                             rt->u.dst.dev->mtu > 576)
1805                                 rt->u.dst.metrics[RTAX_MTU-1] = 576;
1806                 }
1807 #ifdef CONFIG_NET_CLS_ROUTE
1808                 rt->u.dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1809 #endif
1810         } else
1811                 rt->u.dst.metrics[RTAX_MTU-1]= rt->u.dst.dev->mtu;
1812
1813         if (dst_metric(&rt->u.dst, RTAX_HOPLIMIT) == 0)
1814                 rt->u.dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1815         if (dst_mtu(&rt->u.dst) > IP_MAX_MTU)
1816                 rt->u.dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1817         if (dst_metric(&rt->u.dst, RTAX_ADVMSS) == 0)
1818                 rt->u.dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->u.dst.dev->mtu - 40,
1819                                        ip_rt_min_advmss);
1820         if (dst_metric(&rt->u.dst, RTAX_ADVMSS) > 65535 - 40)
1821                 rt->u.dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1822
1823 #ifdef CONFIG_NET_CLS_ROUTE
1824 #ifdef CONFIG_IP_MULTIPLE_TABLES
1825         set_class_tag(rt, fib_rules_tclass(res));
1826 #endif
1827         set_class_tag(rt, itag);
1828 #endif
1829         rt->rt_type = res->type;
1830 }
1831
1832 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1833                                 u8 tos, struct net_device *dev, int our)
1834 {
1835         unsigned hash;
1836         struct rtable *rth;
1837         __be32 spec_dst;
1838         struct in_device *in_dev = in_dev_get(dev);
1839         u32 itag = 0;
1840
1841         /* Primary sanity checks. */
1842
1843         if (in_dev == NULL)
1844                 return -EINVAL;
1845
1846         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1847             ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1848                 goto e_inval;
1849
1850         if (ipv4_is_zeronet(saddr)) {
1851                 if (!ipv4_is_local_multicast(daddr))
1852                         goto e_inval;
1853                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1854         } else if (fib_validate_source(saddr, 0, tos, 0,
1855                                         dev, &spec_dst, &itag) < 0)
1856                 goto e_inval;
1857
1858         rth = dst_alloc(&ipv4_dst_ops);
1859         if (!rth)
1860                 goto e_nobufs;
1861
1862         rth->u.dst.output= ip_rt_bug;
1863
1864         atomic_set(&rth->u.dst.__refcnt, 1);
1865         rth->u.dst.flags= DST_HOST;
1866         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1867                 rth->u.dst.flags |= DST_NOPOLICY;
1868         rth->fl.fl4_dst = daddr;
1869         rth->rt_dst     = daddr;
1870         rth->fl.fl4_tos = tos;
1871         rth->fl.mark    = skb->mark;
1872         rth->fl.fl4_src = saddr;
1873         rth->rt_src     = saddr;
1874 #ifdef CONFIG_NET_CLS_ROUTE
1875         rth->u.dst.tclassid = itag;
1876 #endif
1877         rth->rt_iif     =
1878         rth->fl.iif     = dev->ifindex;
1879         rth->u.dst.dev  = init_net.loopback_dev;
1880         dev_hold(rth->u.dst.dev);
1881         rth->idev       = in_dev_get(rth->u.dst.dev);
1882         rth->fl.oif     = 0;
1883         rth->rt_gateway = daddr;
1884         rth->rt_spec_dst= spec_dst;
1885         rth->rt_genid   = rt_genid(dev_net(dev));
1886         rth->rt_flags   = RTCF_MULTICAST;
1887         rth->rt_type    = RTN_MULTICAST;
1888         if (our) {
1889                 rth->u.dst.input= ip_local_deliver;
1890                 rth->rt_flags |= RTCF_LOCAL;
1891         }
1892
1893 #ifdef CONFIG_IP_MROUTE
1894         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1895                 rth->u.dst.input = ip_mr_input;
1896 #endif
1897         RT_CACHE_STAT_INC(in_slow_mc);
1898
1899         in_dev_put(in_dev);
1900         hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1901         return rt_intern_hash(hash, rth, NULL, skb);
1902
1903 e_nobufs:
1904         in_dev_put(in_dev);
1905         return -ENOBUFS;
1906
1907 e_inval:
1908         in_dev_put(in_dev);
1909         return -EINVAL;
1910 }
1911
1912
1913 static void ip_handle_martian_source(struct net_device *dev,
1914                                      struct in_device *in_dev,
1915                                      struct sk_buff *skb,
1916                                      __be32 daddr,
1917                                      __be32 saddr)
1918 {
1919         RT_CACHE_STAT_INC(in_martian_src);
1920 #ifdef CONFIG_IP_ROUTE_VERBOSE
1921         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1922                 /*
1923                  *      RFC1812 recommendation, if source is martian,
1924                  *      the only hint is MAC header.
1925                  */
1926                 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1927                         &daddr, &saddr, dev->name);
1928                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1929                         int i;
1930                         const unsigned char *p = skb_mac_header(skb);
1931                         printk(KERN_WARNING "ll header: ");
1932                         for (i = 0; i < dev->hard_header_len; i++, p++) {
1933                                 printk("%02x", *p);
1934                                 if (i < (dev->hard_header_len - 1))
1935                                         printk(":");
1936                         }
1937                         printk("\n");
1938                 }
1939         }
1940 #endif
1941 }
1942
1943 static int __mkroute_input(struct sk_buff *skb,
1944                            struct fib_result *res,
1945                            struct in_device *in_dev,
1946                            __be32 daddr, __be32 saddr, u32 tos,
1947                            struct rtable **result)
1948 {
1949
1950         struct rtable *rth;
1951         int err;
1952         struct in_device *out_dev;
1953         unsigned flags = 0;
1954         __be32 spec_dst;
1955         u32 itag;
1956
1957         /* get a working reference to the output device */
1958         out_dev = in_dev_get(FIB_RES_DEV(*res));
1959         if (out_dev == NULL) {
1960                 if (net_ratelimit())
1961                         printk(KERN_CRIT "Bug in ip_route_input" \
1962                                "_slow(). Please, report\n");
1963                 return -EINVAL;
1964         }
1965
1966
1967         err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1968                                   in_dev->dev, &spec_dst, &itag);
1969         if (err < 0) {
1970                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1971                                          saddr);
1972
1973                 err = -EINVAL;
1974                 goto cleanup;
1975         }
1976
1977         if (err)
1978                 flags |= RTCF_DIRECTSRC;
1979
1980         if (out_dev == in_dev && err &&
1981             (IN_DEV_SHARED_MEDIA(out_dev) ||
1982              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1983                 flags |= RTCF_DOREDIRECT;
1984
1985         if (skb->protocol != htons(ETH_P_IP)) {
1986                 /* Not IP (i.e. ARP). Do not create route, if it is
1987                  * invalid for proxy arp. DNAT routes are always valid.
1988                  */
1989                 if (out_dev == in_dev) {
1990                         err = -EINVAL;
1991                         goto cleanup;
1992                 }
1993         }
1994
1995
1996         rth = dst_alloc(&ipv4_dst_ops);
1997         if (!rth) {
1998                 err = -ENOBUFS;
1999                 goto cleanup;
2000         }
2001
2002         atomic_set(&rth->u.dst.__refcnt, 1);
2003         rth->u.dst.flags= DST_HOST;
2004         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2005                 rth->u.dst.flags |= DST_NOPOLICY;
2006         if (IN_DEV_CONF_GET(out_dev, NOXFRM))
2007                 rth->u.dst.flags |= DST_NOXFRM;
2008         rth->fl.fl4_dst = daddr;
2009         rth->rt_dst     = daddr;
2010         rth->fl.fl4_tos = tos;
2011         rth->fl.mark    = skb->mark;
2012         rth->fl.fl4_src = saddr;
2013         rth->rt_src     = saddr;
2014         rth->rt_gateway = daddr;
2015         rth->rt_iif     =
2016                 rth->fl.iif     = in_dev->dev->ifindex;
2017         rth->u.dst.dev  = (out_dev)->dev;
2018         dev_hold(rth->u.dst.dev);
2019         rth->idev       = in_dev_get(rth->u.dst.dev);
2020         rth->fl.oif     = 0;
2021         rth->rt_spec_dst= spec_dst;
2022
2023         rth->u.dst.input = ip_forward;
2024         rth->u.dst.output = ip_output;
2025         rth->rt_genid = rt_genid(dev_net(rth->u.dst.dev));
2026
2027         rt_set_nexthop(rth, res, itag);
2028
2029         rth->rt_flags = flags;
2030
2031         *result = rth;
2032         err = 0;
2033  cleanup:
2034         /* release the working reference to the output device */
2035         in_dev_put(out_dev);
2036         return err;
2037 }
2038
2039 static int ip_mkroute_input(struct sk_buff *skb,
2040                             struct fib_result *res,
2041                             const struct flowi *fl,
2042                             struct in_device *in_dev,
2043                             __be32 daddr, __be32 saddr, u32 tos)
2044 {
2045         struct rtable* rth = NULL;
2046         int err;
2047         unsigned hash;
2048
2049 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2050         if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
2051                 fib_select_multipath(fl, res);
2052 #endif
2053
2054         /* create a routing cache entry */
2055         err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2056         if (err)
2057                 return err;
2058
2059         /* put it into the cache */
2060         hash = rt_hash(daddr, saddr, fl->iif,
2061                        rt_genid(dev_net(rth->u.dst.dev)));
2062         return rt_intern_hash(hash, rth, NULL, skb);
2063 }
2064
2065 /*
2066  *      NOTE. We drop all the packets that has local source
2067  *      addresses, because every properly looped back packet
2068  *      must have correct destination already attached by output routine.
2069  *
2070  *      Such approach solves two big problems:
2071  *      1. Not simplex devices are handled properly.
2072  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2073  */
2074
2075 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2076                                u8 tos, struct net_device *dev)
2077 {
2078         struct fib_result res;
2079         struct in_device *in_dev = in_dev_get(dev);
2080         struct flowi fl = { .nl_u = { .ip4_u =
2081                                       { .daddr = daddr,
2082                                         .saddr = saddr,
2083                                         .tos = tos,
2084                                         .scope = RT_SCOPE_UNIVERSE,
2085                                       } },
2086                             .mark = skb->mark,
2087                             .iif = dev->ifindex };
2088         unsigned        flags = 0;
2089         u32             itag = 0;
2090         struct rtable * rth;
2091         unsigned        hash;
2092         __be32          spec_dst;
2093         int             err = -EINVAL;
2094         int             free_res = 0;
2095         struct net    * net = dev_net(dev);
2096
2097         /* IP on this device is disabled. */
2098
2099         if (!in_dev)
2100                 goto out;
2101
2102         /* Check for the most weird martians, which can be not detected
2103            by fib_lookup.
2104          */
2105
2106         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2107             ipv4_is_loopback(saddr))
2108                 goto martian_source;
2109
2110         if (daddr == htonl(0xFFFFFFFF) || (saddr == 0 && daddr == 0))
2111                 goto brd_input;
2112
2113         /* Accept zero addresses only to limited broadcast;
2114          * I even do not know to fix it or not. Waiting for complains :-)
2115          */
2116         if (ipv4_is_zeronet(saddr))
2117                 goto martian_source;
2118
2119         if (ipv4_is_lbcast(daddr) || ipv4_is_zeronet(daddr) ||
2120             ipv4_is_loopback(daddr))
2121                 goto martian_destination;
2122
2123         /*
2124          *      Now we are ready to route packet.
2125          */
2126         if ((err = fib_lookup(net, &fl, &res)) != 0) {
2127                 if (!IN_DEV_FORWARD(in_dev))
2128                         goto e_hostunreach;
2129                 goto no_route;
2130         }
2131         free_res = 1;
2132
2133         RT_CACHE_STAT_INC(in_slow_tot);
2134
2135         if (res.type == RTN_BROADCAST)
2136                 goto brd_input;
2137
2138         if (res.type == RTN_LOCAL) {
2139                 int result;
2140                 result = fib_validate_source(saddr, daddr, tos,
2141                                              net->loopback_dev->ifindex,
2142                                              dev, &spec_dst, &itag);
2143                 if (result < 0)
2144                         goto martian_source;
2145                 if (result)
2146                         flags |= RTCF_DIRECTSRC;
2147                 spec_dst = daddr;
2148                 goto local_input;
2149         }
2150
2151         if (!IN_DEV_FORWARD(in_dev))
2152                 goto e_hostunreach;
2153         if (res.type != RTN_UNICAST)
2154                 goto martian_destination;
2155
2156         err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
2157 done:
2158         in_dev_put(in_dev);
2159         if (free_res)
2160                 fib_res_put(&res);
2161 out:    return err;
2162
2163 brd_input:
2164         if (skb->protocol != htons(ETH_P_IP))
2165                 goto e_inval;
2166
2167         if (ipv4_is_zeronet(saddr))
2168                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2169         else {
2170                 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2171                                           &itag);
2172                 if (err < 0)
2173                         goto martian_source;
2174                 if (err)
2175                         flags |= RTCF_DIRECTSRC;
2176         }
2177         flags |= RTCF_BROADCAST;
2178         res.type = RTN_BROADCAST;
2179         RT_CACHE_STAT_INC(in_brd);
2180
2181 local_input:
2182         rth = dst_alloc(&ipv4_dst_ops);
2183         if (!rth)
2184                 goto e_nobufs;
2185
2186         rth->u.dst.output= ip_rt_bug;
2187         rth->rt_genid = rt_genid(net);
2188
2189         atomic_set(&rth->u.dst.__refcnt, 1);
2190         rth->u.dst.flags= DST_HOST;
2191         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2192                 rth->u.dst.flags |= DST_NOPOLICY;
2193         rth->fl.fl4_dst = daddr;
2194         rth->rt_dst     = daddr;
2195         rth->fl.fl4_tos = tos;
2196         rth->fl.mark    = skb->mark;
2197         rth->fl.fl4_src = saddr;
2198         rth->rt_src     = saddr;
2199 #ifdef CONFIG_NET_CLS_ROUTE
2200         rth->u.dst.tclassid = itag;
2201 #endif
2202         rth->rt_iif     =
2203         rth->fl.iif     = dev->ifindex;
2204         rth->u.dst.dev  = net->loopback_dev;
2205         dev_hold(rth->u.dst.dev);
2206         rth->idev       = in_dev_get(rth->u.dst.dev);
2207         rth->rt_gateway = daddr;
2208         rth->rt_spec_dst= spec_dst;
2209         rth->u.dst.input= ip_local_deliver;
2210         rth->rt_flags   = flags|RTCF_LOCAL;
2211         if (res.type == RTN_UNREACHABLE) {
2212                 rth->u.dst.input= ip_error;
2213                 rth->u.dst.error= -err;
2214                 rth->rt_flags   &= ~RTCF_LOCAL;
2215         }
2216         rth->rt_type    = res.type;
2217         hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));
2218         err = rt_intern_hash(hash, rth, NULL, skb);
2219         goto done;
2220
2221 no_route:
2222         RT_CACHE_STAT_INC(in_no_route);
2223         spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2224         res.type = RTN_UNREACHABLE;
2225         if (err == -ESRCH)
2226                 err = -ENETUNREACH;
2227         goto local_input;
2228
2229         /*
2230          *      Do not cache martian addresses: they should be logged (RFC1812)
2231          */
2232 martian_destination:
2233         RT_CACHE_STAT_INC(in_martian_dst);
2234 #ifdef CONFIG_IP_ROUTE_VERBOSE
2235         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2236                 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2237                         &daddr, &saddr, dev->name);
2238 #endif
2239
2240 e_hostunreach:
2241         err = -EHOSTUNREACH;
2242         goto done;
2243
2244 e_inval:
2245         err = -EINVAL;
2246         goto done;
2247
2248 e_nobufs:
2249         err = -ENOBUFS;
2250         goto done;
2251
2252 martian_source:
2253         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2254         goto e_inval;
2255 }
2256
2257 int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2258                    u8 tos, struct net_device *dev)
2259 {
2260         struct rtable * rth;
2261         unsigned        hash;
2262         int iif = dev->ifindex;
2263         struct net *net;
2264
2265         net = dev_net(dev);
2266
2267         if (!rt_caching(net))
2268                 goto skip_cache;
2269
2270         tos &= IPTOS_RT_MASK;
2271         hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2272
2273         rcu_read_lock();
2274         for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2275              rth = rcu_dereference(rth->u.dst.rt_next)) {
2276                 if (((rth->fl.fl4_dst ^ daddr) |
2277                      (rth->fl.fl4_src ^ saddr) |
2278                      (rth->fl.iif ^ iif) |
2279                      rth->fl.oif |
2280                      (rth->fl.fl4_tos ^ tos)) == 0 &&
2281                     rth->fl.mark == skb->mark &&
2282                     net_eq(dev_net(rth->u.dst.dev), net) &&
2283                     !rt_is_expired(rth)) {
2284                         dst_use(&rth->u.dst, jiffies);
2285                         RT_CACHE_STAT_INC(in_hit);
2286                         rcu_read_unlock();
2287                         skb_dst_set(skb, &rth->u.dst);
2288                         return 0;
2289                 }
2290                 RT_CACHE_STAT_INC(in_hlist_search);
2291         }
2292         rcu_read_unlock();
2293
2294 skip_cache:
2295         /* Multicast recognition logic is moved from route cache to here.
2296            The problem was that too many Ethernet cards have broken/missing
2297            hardware multicast filters :-( As result the host on multicasting
2298            network acquires a lot of useless route cache entries, sort of
2299            SDR messages from all the world. Now we try to get rid of them.
2300            Really, provided software IP multicast filter is organized
2301            reasonably (at least, hashed), it does not result in a slowdown
2302            comparing with route cache reject entries.
2303            Note, that multicast routers are not affected, because
2304            route cache entry is created eventually.
2305          */
2306         if (ipv4_is_multicast(daddr)) {
2307                 struct in_device *in_dev;
2308
2309                 rcu_read_lock();
2310                 if ((in_dev = __in_dev_get_rcu(dev)) != NULL) {
2311                         int our = ip_check_mc(in_dev, daddr, saddr,
2312                                 ip_hdr(skb)->protocol);
2313                         if (our
2314 #ifdef CONFIG_IP_MROUTE
2315                             || (!ipv4_is_local_multicast(daddr) &&
2316                                 IN_DEV_MFORWARD(in_dev))
2317 #endif
2318                             ) {
2319                                 rcu_read_unlock();
2320                                 return ip_route_input_mc(skb, daddr, saddr,
2321                                                          tos, dev, our);
2322                         }
2323                 }
2324                 rcu_read_unlock();
2325                 return -EINVAL;
2326         }
2327         return ip_route_input_slow(skb, daddr, saddr, tos, dev);
2328 }
2329
2330 static int __mkroute_output(struct rtable **result,
2331                             struct fib_result *res,
2332                             const struct flowi *fl,
2333                             const struct flowi *oldflp,
2334                             struct net_device *dev_out,
2335                             unsigned flags)
2336 {
2337         struct rtable *rth;
2338         struct in_device *in_dev;
2339         u32 tos = RT_FL_TOS(oldflp);
2340         int err = 0;
2341
2342         if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags&IFF_LOOPBACK))
2343                 return -EINVAL;
2344
2345         if (fl->fl4_dst == htonl(0xFFFFFFFF))
2346                 res->type = RTN_BROADCAST;
2347         else if (ipv4_is_multicast(fl->fl4_dst))
2348                 res->type = RTN_MULTICAST;
2349         else if (ipv4_is_lbcast(fl->fl4_dst) || ipv4_is_zeronet(fl->fl4_dst))
2350                 return -EINVAL;
2351
2352         if (dev_out->flags & IFF_LOOPBACK)
2353                 flags |= RTCF_LOCAL;
2354
2355         /* get work reference to inet device */
2356         in_dev = in_dev_get(dev_out);
2357         if (!in_dev)
2358                 return -EINVAL;
2359
2360         if (res->type == RTN_BROADCAST) {
2361                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2362                 if (res->fi) {
2363                         fib_info_put(res->fi);
2364                         res->fi = NULL;
2365                 }
2366         } else if (res->type == RTN_MULTICAST) {
2367                 flags |= RTCF_MULTICAST|RTCF_LOCAL;
2368                 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2369                                  oldflp->proto))
2370                         flags &= ~RTCF_LOCAL;
2371                 /* If multicast route do not exist use
2372                    default one, but do not gateway in this case.
2373                    Yes, it is hack.
2374                  */
2375                 if (res->fi && res->prefixlen < 4) {
2376                         fib_info_put(res->fi);
2377                         res->fi = NULL;
2378                 }
2379         }
2380
2381
2382         rth = dst_alloc(&ipv4_dst_ops);
2383         if (!rth) {
2384                 err = -ENOBUFS;
2385                 goto cleanup;
2386         }
2387
2388         atomic_set(&rth->u.dst.__refcnt, 1);
2389         rth->u.dst.flags= DST_HOST;
2390         if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2391                 rth->u.dst.flags |= DST_NOXFRM;
2392         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2393                 rth->u.dst.flags |= DST_NOPOLICY;
2394
2395         rth->fl.fl4_dst = oldflp->fl4_dst;
2396         rth->fl.fl4_tos = tos;
2397         rth->fl.fl4_src = oldflp->fl4_src;
2398         rth->fl.oif     = oldflp->oif;
2399         rth->fl.mark    = oldflp->mark;
2400         rth->rt_dst     = fl->fl4_dst;
2401         rth->rt_src     = fl->fl4_src;
2402         rth->rt_iif     = oldflp->oif ? : dev_out->ifindex;
2403         /* get references to the devices that are to be hold by the routing
2404            cache entry */
2405         rth->u.dst.dev  = dev_out;
2406         dev_hold(dev_out);
2407         rth->idev       = in_dev_get(dev_out);
2408         rth->rt_gateway = fl->fl4_dst;
2409         rth->rt_spec_dst= fl->fl4_src;
2410
2411         rth->u.dst.output=ip_output;
2412         rth->rt_genid = rt_genid(dev_net(dev_out));
2413
2414         RT_CACHE_STAT_INC(out_slow_tot);
2415
2416         if (flags & RTCF_LOCAL) {
2417                 rth->u.dst.input = ip_local_deliver;
2418                 rth->rt_spec_dst = fl->fl4_dst;
2419         }
2420         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2421                 rth->rt_spec_dst = fl->fl4_src;
2422                 if (flags & RTCF_LOCAL &&
2423                     !(dev_out->flags & IFF_LOOPBACK)) {
2424                         rth->u.dst.output = ip_mc_output;
2425                         RT_CACHE_STAT_INC(out_slow_mc);
2426                 }
2427 #ifdef CONFIG_IP_MROUTE
2428                 if (res->type == RTN_MULTICAST) {
2429                         if (IN_DEV_MFORWARD(in_dev) &&
2430                             !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2431                                 rth->u.dst.input = ip_mr_input;
2432                                 rth->u.dst.output = ip_mc_output;
2433                         }
2434                 }
2435 #endif
2436         }
2437
2438         rt_set_nexthop(rth, res, 0);
2439
2440         rth->rt_flags = flags;
2441
2442         *result = rth;
2443  cleanup:
2444         /* release work reference to inet device */
2445         in_dev_put(in_dev);
2446
2447         return err;
2448 }
2449
2450 static int ip_mkroute_output(struct rtable **rp,
2451                              struct fib_result *res,
2452                              const struct flowi *fl,
2453                              const struct flowi *oldflp,
2454                              struct net_device *dev_out,
2455                              unsigned flags)
2456 {
2457         struct rtable *rth = NULL;
2458         int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2459         unsigned hash;
2460         if (err == 0) {
2461                 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif,
2462                                rt_genid(dev_net(dev_out)));
2463                 err = rt_intern_hash(hash, rth, rp, NULL);
2464         }
2465
2466         return err;
2467 }
2468
2469 /*
2470  * Major route resolver routine.
2471  */
2472
2473 static int ip_route_output_slow(struct net *net, struct rtable **rp,
2474                                 const struct flowi *oldflp)
2475 {
2476         u32 tos = RT_FL_TOS(oldflp);
2477         struct flowi fl = { .nl_u = { .ip4_u =
2478                                       { .daddr = oldflp->fl4_dst,
2479                                         .saddr = oldflp->fl4_src,
2480                                         .tos = tos & IPTOS_RT_MASK,
2481                                         .scope = ((tos & RTO_ONLINK) ?
2482                                                   RT_SCOPE_LINK :
2483                                                   RT_SCOPE_UNIVERSE),
2484                                       } },
2485                             .mark = oldflp->mark,
2486                             .iif = net->loopback_dev->ifindex,
2487                             .oif = oldflp->oif };
2488         struct fib_result res;
2489         unsigned flags = 0;
2490         struct net_device *dev_out = NULL;
2491         int free_res = 0;
2492         int err;
2493
2494
2495         res.fi          = NULL;
2496 #ifdef CONFIG_IP_MULTIPLE_TABLES
2497         res.r           = NULL;
2498 #endif
2499
2500         if (oldflp->fl4_src) {
2501                 err = -EINVAL;
2502                 if (ipv4_is_multicast(oldflp->fl4_src) ||
2503                     ipv4_is_lbcast(oldflp->fl4_src) ||
2504                     ipv4_is_zeronet(oldflp->fl4_src))
2505                         goto out;
2506
2507                 /* I removed check for oif == dev_out->oif here.
2508                    It was wrong for two reasons:
2509                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2510                       is assigned to multiple interfaces.
2511                    2. Moreover, we are allowed to send packets with saddr
2512                       of another iface. --ANK
2513                  */
2514
2515                 if (oldflp->oif == 0
2516                     && (ipv4_is_multicast(oldflp->fl4_dst) ||
2517                         oldflp->fl4_dst == htonl(0xFFFFFFFF))) {
2518                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2519                         dev_out = ip_dev_find(net, oldflp->fl4_src);
2520                         if (dev_out == NULL)
2521                                 goto out;
2522
2523                         /* Special hack: user can direct multicasts
2524                            and limited broadcast via necessary interface
2525                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2526                            This hack is not just for fun, it allows
2527                            vic,vat and friends to work.
2528                            They bind socket to loopback, set ttl to zero
2529                            and expect that it will work.
2530                            From the viewpoint of routing cache they are broken,
2531                            because we are not allowed to build multicast path
2532                            with loopback source addr (look, routing cache
2533                            cannot know, that ttl is zero, so that packet
2534                            will not leave this host and route is valid).
2535                            Luckily, this hack is good workaround.
2536                          */
2537
2538                         fl.oif = dev_out->ifindex;
2539                         goto make_route;
2540                 }
2541
2542                 if (!(oldflp->flags & FLOWI_FLAG_ANYSRC)) {
2543                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2544                         dev_out = ip_dev_find(net, oldflp->fl4_src);
2545                         if (dev_out == NULL)
2546                                 goto out;
2547                         dev_put(dev_out);
2548                         dev_out = NULL;
2549                 }
2550         }
2551
2552
2553         if (oldflp->oif) {
2554                 dev_out = dev_get_by_index(net, oldflp->oif);
2555                 err = -ENODEV;
2556                 if (dev_out == NULL)
2557                         goto out;
2558
2559                 /* RACE: Check return value of inet_select_addr instead. */
2560                 if (__in_dev_get_rtnl(dev_out) == NULL) {
2561                         dev_put(dev_out);
2562                         goto out;       /* Wrong error code */
2563                 }
2564
2565                 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2566                     oldflp->fl4_dst == htonl(0xFFFFFFFF)) {
2567                         if (!fl.fl4_src)
2568                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2569                                                               RT_SCOPE_LINK);
2570                         goto make_route;
2571                 }
2572                 if (!fl.fl4_src) {
2573                         if (ipv4_is_multicast(oldflp->fl4_dst))
2574                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2575                                                               fl.fl4_scope);
2576                         else if (!oldflp->fl4_dst)
2577                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2578                                                               RT_SCOPE_HOST);
2579                 }
2580         }
2581
2582         if (!fl.fl4_dst) {
2583                 fl.fl4_dst = fl.fl4_src;
2584                 if (!fl.fl4_dst)
2585                         fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2586                 if (dev_out)
2587                         dev_put(dev_out);
2588                 dev_out = net->loopback_dev;
2589                 dev_hold(dev_out);
2590                 fl.oif = net->loopback_dev->ifindex;
2591                 res.type = RTN_LOCAL;
2592                 flags |= RTCF_LOCAL;
2593                 goto make_route;
2594         }
2595
2596         if (fib_lookup(net, &fl, &res)) {
2597                 res.fi = NULL;
2598                 if (oldflp->oif) {
2599                         /* Apparently, routing tables are wrong. Assume,
2600                            that the destination is on link.
2601
2602                            WHY? DW.
2603                            Because we are allowed to send to iface
2604                            even if it has NO routes and NO assigned
2605                            addresses. When oif is specified, routing
2606                            tables are looked up with only one purpose:
2607                            to catch if destination is gatewayed, rather than
2608                            direct. Moreover, if MSG_DONTROUTE is set,
2609                            we send packet, ignoring both routing tables
2610                            and ifaddr state. --ANK
2611
2612
2613                            We could make it even if oif is unknown,
2614                            likely IPv6, but we do not.
2615                          */
2616
2617                         if (fl.fl4_src == 0)
2618                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2619                                                               RT_SCOPE_LINK);
2620                         res.type = RTN_UNICAST;
2621                         goto make_route;
2622                 }
2623                 if (dev_out)
2624                         dev_put(dev_out);
2625                 err = -ENETUNREACH;
2626                 goto out;
2627         }
2628         free_res = 1;
2629
2630         if (res.type == RTN_LOCAL) {
2631                 if (!fl.fl4_src)
2632                         fl.fl4_src = fl.fl4_dst;
2633                 if (dev_out)
2634                         dev_put(dev_out);
2635                 dev_out = net->loopback_dev;
2636                 dev_hold(dev_out);
2637                 fl.oif = dev_out->ifindex;
2638                 if (res.fi)
2639                         fib_info_put(res.fi);
2640                 res.fi = NULL;
2641                 flags |= RTCF_LOCAL;
2642                 goto make_route;
2643         }
2644
2645 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2646         if (res.fi->fib_nhs > 1 && fl.oif == 0)
2647                 fib_select_multipath(&fl, &res);
2648         else
2649 #endif
2650         if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2651                 fib_select_default(net, &fl, &res);
2652
2653         if (!fl.fl4_src)
2654                 fl.fl4_src = FIB_RES_PREFSRC(res);
2655
2656         if (dev_out)
2657                 dev_put(dev_out);
2658         dev_out = FIB_RES_DEV(res);
2659         dev_hold(dev_out);
2660         fl.oif = dev_out->ifindex;
2661
2662
2663 make_route:
2664         err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2665
2666
2667         if (free_res)
2668                 fib_res_put(&res);
2669         if (dev_out)
2670                 dev_put(dev_out);
2671 out:    return err;
2672 }
2673
2674 int __ip_route_output_key(struct net *net, struct rtable **rp,
2675                           const struct flowi *flp)
2676 {
2677         unsigned hash;
2678         struct rtable *rth;
2679
2680         if (!rt_caching(net))
2681                 goto slow_output;
2682
2683         hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, rt_genid(net));
2684
2685         rcu_read_lock_bh();
2686         for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2687                 rth = rcu_dereference(rth->u.dst.rt_next)) {
2688                 if (rth->fl.fl4_dst == flp->fl4_dst &&
2689                     rth->fl.fl4_src == flp->fl4_src &&
2690                     rth->fl.iif == 0 &&
2691                     rth->fl.oif == flp->oif &&
2692                     rth->fl.mark == flp->mark &&
2693                     !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2694                             (IPTOS_RT_MASK | RTO_ONLINK)) &&
2695                     net_eq(dev_net(rth->u.dst.dev), net) &&
2696                     !rt_is_expired(rth)) {
2697                         dst_use(&rth->u.dst, jiffies);
2698                         RT_CACHE_STAT_INC(out_hit);
2699                         rcu_read_unlock_bh();
2700                         *rp = rth;
2701                         return 0;
2702                 }
2703                 RT_CACHE_STAT_INC(out_hlist_search);
2704         }
2705         rcu_read_unlock_bh();
2706
2707 slow_output:
2708         return ip_route_output_slow(net, rp, flp);
2709 }
2710
2711 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2712
2713 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2714 {
2715 }
2716
2717 static struct dst_ops ipv4_dst_blackhole_ops = {
2718         .family                 =       AF_INET,
2719         .protocol               =       cpu_to_be16(ETH_P_IP),
2720         .destroy                =       ipv4_dst_destroy,
2721         .check                  =       ipv4_dst_check,
2722         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2723         .entries                =       ATOMIC_INIT(0),
2724 };
2725
2726
2727 static int ipv4_dst_blackhole(struct net *net, struct rtable **rp, struct flowi *flp)
2728 {
2729         struct rtable *ort = *rp;
2730         struct rtable *rt = (struct rtable *)
2731                 dst_alloc(&ipv4_dst_blackhole_ops);
2732
2733         if (rt) {
2734                 struct dst_entry *new = &rt->u.dst;
2735
2736                 atomic_set(&new->__refcnt, 1);
2737                 new->__use = 1;
2738                 new->input = dst_discard;
2739                 new->output = dst_discard;
2740                 memcpy(new->metrics, ort->u.dst.metrics, RTAX_MAX*sizeof(u32));
2741
2742                 new->dev = ort->u.dst.dev;
2743                 if (new->dev)
2744                         dev_hold(new->dev);
2745
2746                 rt->fl = ort->fl;
2747
2748                 rt->idev = ort->idev;
2749                 if (rt->idev)
2750                         in_dev_hold(rt->idev);
2751                 rt->rt_genid = rt_genid(net);
2752                 rt->rt_flags = ort->rt_flags;
2753                 rt->rt_type = ort->rt_type;
2754                 rt->rt_dst = ort->rt_dst;
2755                 rt->rt_src = ort->rt_src;
2756                 rt->rt_iif = ort->rt_iif;
2757                 rt->rt_gateway = ort->rt_gateway;
2758                 rt->rt_spec_dst = ort->rt_spec_dst;
2759                 rt->peer = ort->peer;
2760                 if (rt->peer)
2761                         atomic_inc(&rt->peer->refcnt);
2762
2763                 dst_free(new);
2764         }
2765
2766         dst_release(&(*rp)->u.dst);
2767         *rp = rt;
2768         return (rt ? 0 : -ENOMEM);
2769 }
2770
2771 int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2772                          struct sock *sk, int flags)
2773 {
2774         int err;
2775
2776         if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2777                 return err;
2778
2779         if (flp->proto) {
2780                 if (!flp->fl4_src)
2781                         flp->fl4_src = (*rp)->rt_src;
2782                 if (!flp->fl4_dst)
2783                         flp->fl4_dst = (*rp)->rt_dst;
2784                 err = __xfrm_lookup(net, (struct dst_entry **)rp, flp, sk,
2785                                     flags ? XFRM_LOOKUP_WAIT : 0);
2786                 if (err == -EREMOTE)
2787                         err = ipv4_dst_blackhole(net, rp, flp);
2788
2789                 return err;
2790         }
2791
2792         return 0;
2793 }
2794
2795 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2796
2797 int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2798 {
2799         return ip_route_output_flow(net, rp, flp, NULL, 0);
2800 }
2801
2802 static int rt_fill_info(struct net *net,
2803                         struct sk_buff *skb, u32 pid, u32 seq, int event,
2804                         int nowait, unsigned int flags)
2805 {
2806         struct rtable *rt = skb_rtable(skb);
2807         struct rtmsg *r;
2808         struct nlmsghdr *nlh;
2809         long expires;
2810         u32 id = 0, ts = 0, tsage = 0, error;
2811
2812         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2813         if (nlh == NULL)
2814                 return -EMSGSIZE;
2815
2816         r = nlmsg_data(nlh);
2817         r->rtm_family    = AF_INET;
2818         r->rtm_dst_len  = 32;
2819         r->rtm_src_len  = 0;
2820         r->rtm_tos      = rt->fl.fl4_tos;
2821         r->rtm_table    = RT_TABLE_MAIN;
2822         NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2823         r->rtm_type     = rt->rt_type;
2824         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2825         r->rtm_protocol = RTPROT_UNSPEC;
2826         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2827         if (rt->rt_flags & RTCF_NOTIFY)
2828                 r->rtm_flags |= RTM_F_NOTIFY;
2829
2830         NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2831
2832         if (rt->fl.fl4_src) {
2833                 r->rtm_src_len = 32;
2834                 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2835         }
2836         if (rt->u.dst.dev)
2837                 NLA_PUT_U32(skb, RTA_OIF, rt->u.dst.dev->ifindex);
2838 #ifdef CONFIG_NET_CLS_ROUTE
2839         if (rt->u.dst.tclassid)
2840                 NLA_PUT_U32(skb, RTA_FLOW, rt->u.dst.tclassid);
2841 #endif
2842         if (rt->fl.iif)
2843                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2844         else if (rt->rt_src != rt->fl.fl4_src)
2845                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2846
2847         if (rt->rt_dst != rt->rt_gateway)
2848                 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2849
2850         if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
2851                 goto nla_put_failure;
2852
2853         error = rt->u.dst.error;
2854         expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
2855         if (rt->peer) {
2856                 id = rt->peer->ip_id_count;
2857                 if (rt->peer->tcp_ts_stamp) {
2858                         ts = rt->peer->tcp_ts;
2859                         tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2860                 }
2861         }
2862
2863         if (rt->fl.iif) {
2864 #ifdef CONFIG_IP_MROUTE
2865                 __be32 dst = rt->rt_dst;
2866
2867                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2868                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2869                         int err = ipmr_get_route(net, skb, r, nowait);
2870                         if (err <= 0) {
2871                                 if (!nowait) {
2872                                         if (err == 0)
2873                                                 return 0;
2874                                         goto nla_put_failure;
2875                                 } else {
2876                                         if (err == -EMSGSIZE)
2877                                                 goto nla_put_failure;
2878                                         error = err;
2879                                 }
2880                         }
2881                 } else
2882 #endif
2883                         NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2884         }
2885
2886         if (rtnl_put_cacheinfo(skb, &rt->u.dst, id, ts, tsage,
2887                                expires, error) < 0)
2888                 goto nla_put_failure;
2889
2890         return nlmsg_end(skb, nlh);
2891
2892 nla_put_failure:
2893         nlmsg_cancel(skb, nlh);
2894         return -EMSGSIZE;
2895 }
2896
2897 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2898 {
2899         struct net *net = sock_net(in_skb->sk);
2900         struct rtmsg *rtm;
2901         struct nlattr *tb[RTA_MAX+1];
2902         struct rtable *rt = NULL;
2903         __be32 dst = 0;
2904         __be32 src = 0;
2905         u32 iif;
2906         int err;
2907         struct sk_buff *skb;
2908
2909         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2910         if (err < 0)
2911                 goto errout;
2912
2913         rtm = nlmsg_data(nlh);
2914
2915         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2916         if (skb == NULL) {
2917                 err = -ENOBUFS;
2918                 goto errout;
2919         }
2920
2921         /* Reserve room for dummy headers, this skb can pass
2922            through good chunk of routing engine.
2923          */
2924         skb_reset_mac_header(skb);
2925         skb_reset_network_header(skb);
2926
2927         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2928         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2929         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2930
2931         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2932         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2933         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2934
2935         if (iif) {
2936                 struct net_device *dev;
2937
2938                 dev = __dev_get_by_index(net, iif);
2939                 if (dev == NULL) {
2940                         err = -ENODEV;
2941                         goto errout_free;
2942                 }
2943
2944                 skb->protocol   = htons(ETH_P_IP);
2945                 skb->dev        = dev;
2946                 local_bh_disable();
2947                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2948                 local_bh_enable();
2949
2950                 rt = skb_rtable(skb);
2951                 if (err == 0 && rt->u.dst.error)
2952                         err = -rt->u.dst.error;
2953         } else {
2954                 struct flowi fl = {
2955                         .nl_u = {
2956                                 .ip4_u = {
2957                                         .daddr = dst,
2958                                         .saddr = src,
2959                                         .tos = rtm->rtm_tos,
2960                                 },
2961                         },
2962                         .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2963                 };
2964                 err = ip_route_output_key(net, &rt, &fl);
2965         }
2966
2967         if (err)
2968                 goto errout_free;
2969
2970         skb_dst_set(skb, &rt->u.dst);
2971         if (rtm->rtm_flags & RTM_F_NOTIFY)
2972                 rt->rt_flags |= RTCF_NOTIFY;
2973
2974         err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2975                            RTM_NEWROUTE, 0, 0);
2976         if (err <= 0)
2977                 goto errout_free;
2978
2979         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2980 errout:
2981         return err;
2982
2983 errout_free:
2984         kfree_skb(skb);
2985         goto errout;
2986 }
2987
2988 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
2989 {
2990         struct rtable *rt;
2991         int h, s_h;
2992         int idx, s_idx;
2993         struct net *net;
2994
2995         net = sock_net(skb->sk);
2996
2997         s_h = cb->args[0];
2998         if (s_h < 0)
2999                 s_h = 0;
3000         s_idx = idx = cb->args[1];
3001         for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3002                 if (!rt_hash_table[h].chain)
3003                         continue;
3004                 rcu_read_lock_bh();
3005                 for (rt = rcu_dereference(rt_hash_table[h].chain), idx = 0; rt;
3006                      rt = rcu_dereference(rt->u.dst.rt_next), idx++) {
3007                         if (!net_eq(dev_net(rt->u.dst.dev), net) || idx < s_idx)
3008                                 continue;
3009                         if (rt_is_expired(rt))
3010                                 continue;
3011                         skb_dst_set(skb, dst_clone(&rt->u.dst));
3012                         if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3013                                          cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3014                                          1, NLM_F_MULTI) <= 0) {
3015                                 skb_dst_drop(skb);
3016                                 rcu_read_unlock_bh();
3017                                 goto done;
3018                         }
3019                         skb_dst_drop(skb);
3020                 }
3021                 rcu_read_unlock_bh();
3022         }
3023
3024 done:
3025         cb->args[0] = h;
3026         cb->args[1] = idx;
3027         return skb->len;
3028 }
3029
3030 void ip_rt_multicast_event(struct in_device *in_dev)
3031 {
3032         rt_cache_flush(dev_net(in_dev->dev), 0);
3033 }
3034
3035 #ifdef CONFIG_SYSCTL
3036 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3037                                         struct file *filp, void __user *buffer,
3038                                         size_t *lenp, loff_t *ppos)
3039 {
3040         if (write) {
3041                 int flush_delay;
3042                 ctl_table ctl;
3043                 struct net *net;
3044
3045                 memcpy(&ctl, __ctl, sizeof(ctl));
3046                 ctl.data = &flush_delay;
3047                 proc_dointvec(&ctl, write, filp, buffer, lenp, ppos);
3048
3049                 net = (struct net *)__ctl->extra1;
3050                 rt_cache_flush(net, flush_delay);
3051                 return 0;
3052         }
3053
3054         return -EINVAL;
3055 }
3056
3057 static int ipv4_sysctl_rtcache_flush_strategy(ctl_table *table,
3058                                                 void __user *oldval,
3059                                                 size_t __user *oldlenp,
3060                                                 void __user *newval,
3061                                                 size_t newlen)
3062 {
3063         int delay;
3064         struct net *net;
3065         if (newlen != sizeof(int))
3066                 return -EINVAL;
3067         if (get_user(delay, (int __user *)newval))
3068                 return -EFAULT;
3069         net = (struct net *)table->extra1;
3070         rt_cache_flush(net, delay);
3071         return 0;
3072 }
3073
3074 static void rt_secret_reschedule(int old)
3075 {
3076         struct net *net;
3077         int new = ip_rt_secret_interval;
3078         int diff = new - old;
3079
3080         if (!diff)
3081                 return;
3082
3083         rtnl_lock();
3084         for_each_net(net) {
3085                 int deleted = del_timer_sync(&net->ipv4.rt_secret_timer);
3086
3087                 if (!new)
3088                         continue;
3089
3090                 if (deleted) {
3091                         long time = net->ipv4.rt_secret_timer.expires - jiffies;
3092
3093                         if (time <= 0 || (time += diff) <= 0)
3094                                 time = 0;
3095
3096                         net->ipv4.rt_secret_timer.expires = time;
3097                 } else
3098                         net->ipv4.rt_secret_timer.expires = new;
3099
3100                 net->ipv4.rt_secret_timer.expires += jiffies;
3101                 add_timer(&net->ipv4.rt_secret_timer);
3102         }
3103         rtnl_unlock();
3104 }
3105
3106 static int ipv4_sysctl_rt_secret_interval(ctl_table *ctl, int write,
3107                                           struct file *filp,
3108                                           void __user *buffer, size_t *lenp,
3109                                           loff_t *ppos)
3110 {
3111         int old = ip_rt_secret_interval;
3112         int ret = proc_dointvec_jiffies(ctl, write, filp, buffer, lenp, ppos);
3113
3114         rt_secret_reschedule(old);
3115
3116         return ret;
3117 }
3118
3119 static int ipv4_sysctl_rt_secret_interval_strategy(ctl_table *table,
3120                                                    void __user *oldval,
3121                                                    size_t __user *oldlenp,
3122                                                    void __user *newval,
3123                                                    size_t newlen)
3124 {
3125         int old = ip_rt_secret_interval;
3126         int ret = sysctl_jiffies(table, oldval, oldlenp, newval, newlen);
3127
3128         rt_secret_reschedule(old);
3129
3130         return ret;
3131 }
3132
3133 static ctl_table ipv4_route_table[] = {
3134         {
3135                 .ctl_name       = NET_IPV4_ROUTE_GC_THRESH,
3136                 .procname       = "gc_thresh",
3137                 .data           = &ipv4_dst_ops.gc_thresh,
3138                 .maxlen         = sizeof(int),
3139                 .mode           = 0644,
3140                 .proc_handler   = proc_dointvec,
3141         },
3142         {
3143                 .ctl_name       = NET_IPV4_ROUTE_MAX_SIZE,
3144                 .procname       = "max_size",
3145                 .data           = &ip_rt_max_size,
3146                 .maxlen         = sizeof(int),
3147                 .mode           = 0644,
3148                 .proc_handler   = proc_dointvec,
3149         },
3150         {
3151                 /*  Deprecated. Use gc_min_interval_ms */
3152
3153                 .ctl_name       = NET_IPV4_ROUTE_GC_MIN_INTERVAL,
3154                 .procname       = "gc_min_interval",
3155                 .data           = &ip_rt_gc_min_interval,
3156                 .maxlen         = sizeof(int),
3157                 .mode           = 0644,
3158                 .proc_handler   = proc_dointvec_jiffies,
3159                 .strategy       = sysctl_jiffies,
3160         },
3161         {
3162                 .ctl_name       = NET_IPV4_ROUTE_GC_MIN_INTERVAL_MS,
3163                 .procname       = "gc_min_interval_ms",
3164                 .data           = &ip_rt_gc_min_interval,
3165                 .maxlen         = sizeof(int),
3166                 .mode           = 0644,
3167                 .proc_handler   = proc_dointvec_ms_jiffies,
3168                 .strategy       = sysctl_ms_jiffies,
3169         },
3170         {
3171                 .ctl_name       = NET_IPV4_ROUTE_GC_TIMEOUT,
3172                 .procname       = "gc_timeout",
3173                 .data           = &ip_rt_gc_timeout,
3174                 .maxlen         = sizeof(int),
3175                 .mode           = 0644,
3176                 .proc_handler   = proc_dointvec_jiffies,
3177                 .strategy       = sysctl_jiffies,
3178         },
3179         {
3180                 .ctl_name       = NET_IPV4_ROUTE_GC_INTERVAL,
3181                 .procname       = "gc_interval",
3182                 .data           = &ip_rt_gc_interval,
3183                 .maxlen         = sizeof(int),
3184                 .mode           = 0644,
3185                 .proc_handler   = proc_dointvec_jiffies,
3186                 .strategy       = sysctl_jiffies,
3187         },
3188         {
3189                 .ctl_name       = NET_IPV4_ROUTE_REDIRECT_LOAD,
3190                 .procname       = "redirect_load",
3191                 .data           = &ip_rt_redirect_load,
3192                 .maxlen         = sizeof(int),
3193                 .mode           = 0644,
3194                 .proc_handler   = proc_dointvec,
3195         },
3196         {
3197                 .ctl_name       = NET_IPV4_ROUTE_REDIRECT_NUMBER,
3198                 .procname       = "redirect_number",
3199                 .data           = &ip_rt_redirect_number,
3200                 .maxlen         = sizeof(int),
3201                 .mode           = 0644,
3202                 .proc_handler   = proc_dointvec,
3203         },
3204         {
3205                 .ctl_name       = NET_IPV4_ROUTE_REDIRECT_SILENCE,
3206                 .procname       = "redirect_silence",
3207                 .data           = &ip_rt_redirect_silence,
3208                 .maxlen         = sizeof(int),
3209                 .mode           = 0644,
3210                 .proc_handler   = proc_dointvec,
3211         },
3212         {
3213                 .ctl_name       = NET_IPV4_ROUTE_ERROR_COST,
3214                 .procname       = "error_cost",
3215                 .data           = &ip_rt_error_cost,
3216                 .maxlen         = sizeof(int),
3217                 .mode           = 0644,
3218                 .proc_handler   = proc_dointvec,
3219         },
3220         {
3221                 .ctl_name       = NET_IPV4_ROUTE_ERROR_BURST,
3222                 .procname       = "error_burst",
3223                 .data           = &ip_rt_error_burst,
3224                 .maxlen         = sizeof(int),
3225                 .mode           = 0644,
3226                 .proc_handler   = proc_dointvec,
3227         },
3228         {
3229                 .ctl_name       = NET_IPV4_ROUTE_GC_ELASTICITY,
3230                 .procname       = "gc_elasticity",
3231                 .data           = &ip_rt_gc_elasticity,
3232                 .maxlen         = sizeof(int),
3233                 .mode           = 0644,
3234                 .proc_handler   = proc_dointvec,
3235         },
3236         {
3237                 .ctl_name       = NET_IPV4_ROUTE_MTU_EXPIRES,
3238                 .procname       = "mtu_expires",
3239                 .data           = &ip_rt_mtu_expires,
3240                 .maxlen         = sizeof(int),
3241                 .mode           = 0644,
3242                 .proc_handler   = proc_dointvec_jiffies,
3243                 .strategy       = sysctl_jiffies,
3244         },
3245         {
3246                 .ctl_name       = NET_IPV4_ROUTE_MIN_PMTU,
3247                 .procname       = "min_pmtu",
3248                 .data           = &ip_rt_min_pmtu,
3249                 .maxlen         = sizeof(int),
3250                 .mode           = 0644,
3251                 .proc_handler   = proc_dointvec,
3252         },
3253         {
3254                 .ctl_name       = NET_IPV4_ROUTE_MIN_ADVMSS,
3255                 .procname       = "min_adv_mss",
3256                 .data           = &ip_rt_min_advmss,
3257                 .maxlen         = sizeof(int),
3258                 .mode           = 0644,
3259                 .proc_handler   = proc_dointvec,
3260         },
3261         {
3262                 .ctl_name       = NET_IPV4_ROUTE_SECRET_INTERVAL,
3263                 .procname       = "secret_interval",
3264                 .data           = &ip_rt_secret_interval,
3265                 .maxlen         = sizeof(int),
3266                 .mode           = 0644,
3267                 .proc_handler   = ipv4_sysctl_rt_secret_interval,
3268                 .strategy       = ipv4_sysctl_rt_secret_interval_strategy,
3269         },
3270         { .ctl_name = 0 }
3271 };
3272
3273 static struct ctl_table empty[1];
3274
3275 static struct ctl_table ipv4_skeleton[] =
3276 {
3277         { .procname = "route", .ctl_name = NET_IPV4_ROUTE,
3278           .mode = 0555, .child = ipv4_route_table},
3279         { .procname = "neigh", .ctl_name = NET_IPV4_NEIGH,
3280           .mode = 0555, .child = empty},
3281         { }
3282 };
3283
3284 static __net_initdata struct ctl_path ipv4_path[] = {
3285         { .procname = "net", .ctl_name = CTL_NET, },
3286         { .procname = "ipv4", .ctl_name = NET_IPV4, },
3287         { },
3288 };
3289
3290 static struct ctl_table ipv4_route_flush_table[] = {
3291         {
3292                 .ctl_name       = NET_IPV4_ROUTE_FLUSH,
3293                 .procname       = "flush",
3294                 .maxlen         = sizeof(int),
3295                 .mode           = 0200,
3296                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3297                 .strategy       = ipv4_sysctl_rtcache_flush_strategy,
3298         },
3299         { .ctl_name = 0 },
3300 };
3301
3302 static __net_initdata struct ctl_path ipv4_route_path[] = {
3303         { .procname = "net", .ctl_name = CTL_NET, },
3304         { .procname = "ipv4", .ctl_name = NET_IPV4, },
3305         { .procname = "route", .ctl_name = NET_IPV4_ROUTE, },
3306         { },
3307 };
3308
3309 static __net_init int sysctl_route_net_init(struct net *net)
3310 {
3311         struct ctl_table *tbl;
3312
3313         tbl = ipv4_route_flush_table;
3314         if (net != &init_net) {
3315                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3316                 if (tbl == NULL)
3317                         goto err_dup;
3318         }
3319         tbl[0].extra1 = net;
3320
3321         net->ipv4.route_hdr =
3322                 register_net_sysctl_table(net, ipv4_route_path, tbl);
3323         if (net->ipv4.route_hdr == NULL)
3324                 goto err_reg;
3325         return 0;
3326
3327 err_reg:
3328         if (tbl != ipv4_route_flush_table)
3329                 kfree(tbl);
3330 err_dup:
3331         return -ENOMEM;
3332 }
3333
3334 static __net_exit void sysctl_route_net_exit(struct net *net)
3335 {
3336         struct ctl_table *tbl;
3337
3338         tbl = net->ipv4.route_hdr->ctl_table_arg;
3339         unregister_net_sysctl_table(net->ipv4.route_hdr);
3340         BUG_ON(tbl == ipv4_route_flush_table);
3341         kfree(tbl);
3342 }
3343
3344 static __net_initdata struct pernet_operations sysctl_route_ops = {
3345         .init = sysctl_route_net_init,
3346         .exit = sysctl_route_net_exit,
3347 };
3348 #endif
3349
3350
3351 static __net_init int rt_secret_timer_init(struct net *net)
3352 {
3353         atomic_set(&net->ipv4.rt_genid,
3354                         (int) ((num_physpages ^ (num_physpages>>8)) ^
3355                         (jiffies ^ (jiffies >> 7))));
3356
3357         net->ipv4.rt_secret_timer.function = rt_secret_rebuild;
3358         net->ipv4.rt_secret_timer.data = (unsigned long)net;
3359         init_timer_deferrable(&net->ipv4.rt_secret_timer);
3360
3361         if (ip_rt_secret_interval) {
3362                 net->ipv4.rt_secret_timer.expires =
3363                         jiffies + net_random() % ip_rt_secret_interval +
3364                         ip_rt_secret_interval;
3365                 add_timer(&net->ipv4.rt_secret_timer);
3366         }
3367         return 0;
3368 }
3369
3370 static __net_exit void rt_secret_timer_exit(struct net *net)
3371 {
3372         del_timer_sync(&net->ipv4.rt_secret_timer);
3373 }
3374
3375 static __net_initdata struct pernet_operations rt_secret_timer_ops = {
3376         .init = rt_secret_timer_init,
3377         .exit = rt_secret_timer_exit,
3378 };
3379
3380
3381 #ifdef CONFIG_NET_CLS_ROUTE
3382 struct ip_rt_acct *ip_rt_acct __read_mostly;
3383 #endif /* CONFIG_NET_CLS_ROUTE */
3384
3385 static __initdata unsigned long rhash_entries;
3386 static int __init set_rhash_entries(char *str)
3387 {
3388         if (!str)
3389                 return 0;
3390         rhash_entries = simple_strtoul(str, &str, 0);
3391         return 1;
3392 }
3393 __setup("rhash_entries=", set_rhash_entries);
3394
3395 int __init ip_rt_init(void)
3396 {
3397         int rc = 0;
3398
3399 #ifdef CONFIG_NET_CLS_ROUTE
3400         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3401         if (!ip_rt_acct)
3402                 panic("IP: failed to allocate ip_rt_acct\n");
3403 #endif
3404
3405         ipv4_dst_ops.kmem_cachep =
3406                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3407                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3408
3409         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3410
3411         rt_hash_table = (struct rt_hash_bucket *)
3412                 alloc_large_system_hash("IP route cache",
3413                                         sizeof(struct rt_hash_bucket),
3414                                         rhash_entries,
3415                                         (num_physpages >= 128 * 1024) ?
3416                                         15 : 17,
3417                                         0,
3418                                         &rt_hash_log,
3419                                         &rt_hash_mask,
3420                                         rhash_entries ? 0 : 512 * 1024);
3421         memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3422         rt_hash_lock_init();
3423
3424         ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3425         ip_rt_max_size = (rt_hash_mask + 1) * 16;
3426
3427         devinet_init();
3428         ip_fib_init();
3429
3430         /* All the timers, started at system startup tend
3431            to synchronize. Perturb it a bit.
3432          */
3433         INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3434         expires_ljiffies = jiffies;
3435         schedule_delayed_work(&expires_work,
3436                 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3437
3438         if (register_pernet_subsys(&rt_secret_timer_ops))
3439                 printk(KERN_ERR "Unable to setup rt_secret_timer\n");
3440
3441         if (ip_rt_proc_init())
3442                 printk(KERN_ERR "Unable to create route proc files\n");
3443 #ifdef CONFIG_XFRM
3444         xfrm_init();
3445         xfrm4_init();
3446 #endif
3447         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3448
3449 #ifdef CONFIG_SYSCTL
3450         register_pernet_subsys(&sysctl_route_ops);
3451 #endif
3452         return rc;
3453 }
3454
3455 #ifdef CONFIG_SYSCTL
3456 /*
3457  * We really need to sanitize the damn ipv4 init order, then all
3458  * this nonsense will go away.
3459  */
3460 void __init ip_static_sysctl_init(void)
3461 {
3462         register_sysctl_paths(ipv4_path, ipv4_skeleton);
3463 }
3464 #endif
3465
3466 EXPORT_SYMBOL(__ip_select_ident);
3467 EXPORT_SYMBOL(ip_route_input);
3468 EXPORT_SYMBOL(ip_route_output_key);