ipv4/route.c: respect prefsrc for local routes
[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 <linux/slab.h>
94 #include <net/dst.h>
95 #include <net/net_namespace.h>
96 #include <net/protocol.h>
97 #include <net/ip.h>
98 #include <net/route.h>
99 #include <net/inetpeer.h>
100 #include <net/sock.h>
101 #include <net/ip_fib.h>
102 #include <net/arp.h>
103 #include <net/tcp.h>
104 #include <net/icmp.h>
105 #include <net/xfrm.h>
106 #include <net/netevent.h>
107 #include <net/rtnetlink.h>
108 #ifdef CONFIG_SYSCTL
109 #include <linux/sysctl.h>
110 #endif
111
112 #define RT_FL_TOS(oldflp) \
113     ((u32)(oldflp->fl4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
114
115 #define IP_MAX_MTU      0xFFF0
116
117 #define RT_GC_TIMEOUT (300*HZ)
118
119 static int ip_rt_max_size;
120 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
121 static int ip_rt_gc_interval __read_mostly      = 60 * HZ;
122 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
123 static int ip_rt_redirect_number __read_mostly  = 9;
124 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
125 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
126 static int ip_rt_error_cost __read_mostly       = HZ;
127 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
128 static int ip_rt_gc_elasticity __read_mostly    = 8;
129 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
130 static int ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
131 static int ip_rt_min_advmss __read_mostly       = 256;
132 static int rt_chain_length_max __read_mostly    = 20;
133
134 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
150
151 static struct dst_ops ipv4_dst_ops = {
152         .family =               AF_INET,
153         .protocol =             cpu_to_be16(ETH_P_IP),
154         .gc =                   rt_garbage_collect,
155         .check =                ipv4_dst_check,
156         .destroy =              ipv4_dst_destroy,
157         .ifdown =               ipv4_dst_ifdown,
158         .negative_advice =      ipv4_negative_advice,
159         .link_failure =         ipv4_link_failure,
160         .update_pmtu =          ip_rt_update_pmtu,
161         .local_out =            __ip_local_out,
162 };
163
164 #define ECN_OR_COST(class)      TC_PRIO_##class
165
166 const __u8 ip_tos2prio[16] = {
167         TC_PRIO_BESTEFFORT,
168         ECN_OR_COST(FILLER),
169         TC_PRIO_BESTEFFORT,
170         ECN_OR_COST(BESTEFFORT),
171         TC_PRIO_BULK,
172         ECN_OR_COST(BULK),
173         TC_PRIO_BULK,
174         ECN_OR_COST(BULK),
175         TC_PRIO_INTERACTIVE,
176         ECN_OR_COST(INTERACTIVE),
177         TC_PRIO_INTERACTIVE,
178         ECN_OR_COST(INTERACTIVE),
179         TC_PRIO_INTERACTIVE_BULK,
180         ECN_OR_COST(INTERACTIVE_BULK),
181         TC_PRIO_INTERACTIVE_BULK,
182         ECN_OR_COST(INTERACTIVE_BULK)
183 };
184
185
186 /*
187  * Route cache.
188  */
189
190 /* The locking scheme is rather straight forward:
191  *
192  * 1) Read-Copy Update protects the buckets of the central route hash.
193  * 2) Only writers remove entries, and they hold the lock
194  *    as they look at rtable reference counts.
195  * 3) Only readers acquire references to rtable entries,
196  *    they do so with atomic increments and with the
197  *    lock held.
198  */
199
200 struct rt_hash_bucket {
201         struct rtable __rcu     *chain;
202 };
203
204 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
205         defined(CONFIG_PROVE_LOCKING)
206 /*
207  * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
208  * The size of this table is a power of two and depends on the number of CPUS.
209  * (on lockdep we have a quite big spinlock_t, so keep the size down there)
210  */
211 #ifdef CONFIG_LOCKDEP
212 # define RT_HASH_LOCK_SZ        256
213 #else
214 # if NR_CPUS >= 32
215 #  define RT_HASH_LOCK_SZ       4096
216 # elif NR_CPUS >= 16
217 #  define RT_HASH_LOCK_SZ       2048
218 # elif NR_CPUS >= 8
219 #  define RT_HASH_LOCK_SZ       1024
220 # elif NR_CPUS >= 4
221 #  define RT_HASH_LOCK_SZ       512
222 # else
223 #  define RT_HASH_LOCK_SZ       256
224 # endif
225 #endif
226
227 static spinlock_t       *rt_hash_locks;
228 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
229
230 static __init void rt_hash_lock_init(void)
231 {
232         int i;
233
234         rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
235                         GFP_KERNEL);
236         if (!rt_hash_locks)
237                 panic("IP: failed to allocate rt_hash_locks\n");
238
239         for (i = 0; i < RT_HASH_LOCK_SZ; i++)
240                 spin_lock_init(&rt_hash_locks[i]);
241 }
242 #else
243 # define rt_hash_lock_addr(slot) NULL
244
245 static inline void rt_hash_lock_init(void)
246 {
247 }
248 #endif
249
250 static struct rt_hash_bucket    *rt_hash_table __read_mostly;
251 static unsigned                 rt_hash_mask __read_mostly;
252 static unsigned int             rt_hash_log  __read_mostly;
253
254 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
255 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
256
257 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
258                                    int genid)
259 {
260         return jhash_3words((__force u32)daddr, (__force u32)saddr,
261                             idx, genid)
262                 & rt_hash_mask;
263 }
264
265 static inline int rt_genid(struct net *net)
266 {
267         return atomic_read(&net->ipv4.rt_genid);
268 }
269
270 #ifdef CONFIG_PROC_FS
271 struct rt_cache_iter_state {
272         struct seq_net_private p;
273         int bucket;
274         int genid;
275 };
276
277 static struct rtable *rt_cache_get_first(struct seq_file *seq)
278 {
279         struct rt_cache_iter_state *st = seq->private;
280         struct rtable *r = NULL;
281
282         for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
283                 if (!rcu_dereference_raw(rt_hash_table[st->bucket].chain))
284                         continue;
285                 rcu_read_lock_bh();
286                 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
287                 while (r) {
288                         if (dev_net(r->dst.dev) == seq_file_net(seq) &&
289                             r->rt_genid == st->genid)
290                                 return r;
291                         r = rcu_dereference_bh(r->dst.rt_next);
292                 }
293                 rcu_read_unlock_bh();
294         }
295         return r;
296 }
297
298 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
299                                           struct rtable *r)
300 {
301         struct rt_cache_iter_state *st = seq->private;
302
303         r = rcu_dereference_bh(r->dst.rt_next);
304         while (!r) {
305                 rcu_read_unlock_bh();
306                 do {
307                         if (--st->bucket < 0)
308                                 return NULL;
309                 } while (!rcu_dereference_raw(rt_hash_table[st->bucket].chain));
310                 rcu_read_lock_bh();
311                 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
312         }
313         return r;
314 }
315
316 static struct rtable *rt_cache_get_next(struct seq_file *seq,
317                                         struct rtable *r)
318 {
319         struct rt_cache_iter_state *st = seq->private;
320         while ((r = __rt_cache_get_next(seq, r)) != NULL) {
321                 if (dev_net(r->dst.dev) != seq_file_net(seq))
322                         continue;
323                 if (r->rt_genid == st->genid)
324                         break;
325         }
326         return r;
327 }
328
329 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
330 {
331         struct rtable *r = rt_cache_get_first(seq);
332
333         if (r)
334                 while (pos && (r = rt_cache_get_next(seq, r)))
335                         --pos;
336         return pos ? NULL : r;
337 }
338
339 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
340 {
341         struct rt_cache_iter_state *st = seq->private;
342         if (*pos)
343                 return rt_cache_get_idx(seq, *pos - 1);
344         st->genid = rt_genid(seq_file_net(seq));
345         return SEQ_START_TOKEN;
346 }
347
348 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
349 {
350         struct rtable *r;
351
352         if (v == SEQ_START_TOKEN)
353                 r = rt_cache_get_first(seq);
354         else
355                 r = rt_cache_get_next(seq, v);
356         ++*pos;
357         return r;
358 }
359
360 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
361 {
362         if (v && v != SEQ_START_TOKEN)
363                 rcu_read_unlock_bh();
364 }
365
366 static int rt_cache_seq_show(struct seq_file *seq, void *v)
367 {
368         if (v == SEQ_START_TOKEN)
369                 seq_printf(seq, "%-127s\n",
370                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
371                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
372                            "HHUptod\tSpecDst");
373         else {
374                 struct rtable *r = v;
375                 int len;
376
377                 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
378                               "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
379                         r->dst.dev ? r->dst.dev->name : "*",
380                         (__force u32)r->rt_dst,
381                         (__force u32)r->rt_gateway,
382                         r->rt_flags, atomic_read(&r->dst.__refcnt),
383                         r->dst.__use, 0, (__force u32)r->rt_src,
384                         (dst_metric(&r->dst, RTAX_ADVMSS) ?
385                              (int)dst_metric(&r->dst, RTAX_ADVMSS) + 40 : 0),
386                         dst_metric(&r->dst, RTAX_WINDOW),
387                         (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
388                               dst_metric(&r->dst, RTAX_RTTVAR)),
389                         r->fl.fl4_tos,
390                         r->dst.hh ? atomic_read(&r->dst.hh->hh_refcnt) : -1,
391                         r->dst.hh ? (r->dst.hh->hh_output ==
392                                        dev_queue_xmit) : 0,
393                         r->rt_spec_dst, &len);
394
395                 seq_printf(seq, "%*s\n", 127 - len, "");
396         }
397         return 0;
398 }
399
400 static const struct seq_operations rt_cache_seq_ops = {
401         .start  = rt_cache_seq_start,
402         .next   = rt_cache_seq_next,
403         .stop   = rt_cache_seq_stop,
404         .show   = rt_cache_seq_show,
405 };
406
407 static int rt_cache_seq_open(struct inode *inode, struct file *file)
408 {
409         return seq_open_net(inode, file, &rt_cache_seq_ops,
410                         sizeof(struct rt_cache_iter_state));
411 }
412
413 static const struct file_operations rt_cache_seq_fops = {
414         .owner   = THIS_MODULE,
415         .open    = rt_cache_seq_open,
416         .read    = seq_read,
417         .llseek  = seq_lseek,
418         .release = seq_release_net,
419 };
420
421
422 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
423 {
424         int cpu;
425
426         if (*pos == 0)
427                 return SEQ_START_TOKEN;
428
429         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
430                 if (!cpu_possible(cpu))
431                         continue;
432                 *pos = cpu+1;
433                 return &per_cpu(rt_cache_stat, cpu);
434         }
435         return NULL;
436 }
437
438 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
439 {
440         int cpu;
441
442         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
443                 if (!cpu_possible(cpu))
444                         continue;
445                 *pos = cpu+1;
446                 return &per_cpu(rt_cache_stat, cpu);
447         }
448         return NULL;
449
450 }
451
452 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
453 {
454
455 }
456
457 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
458 {
459         struct rt_cache_stat *st = v;
460
461         if (v == SEQ_START_TOKEN) {
462                 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");
463                 return 0;
464         }
465
466         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
467                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
468                    dst_entries_get_slow(&ipv4_dst_ops),
469                    st->in_hit,
470                    st->in_slow_tot,
471                    st->in_slow_mc,
472                    st->in_no_route,
473                    st->in_brd,
474                    st->in_martian_dst,
475                    st->in_martian_src,
476
477                    st->out_hit,
478                    st->out_slow_tot,
479                    st->out_slow_mc,
480
481                    st->gc_total,
482                    st->gc_ignored,
483                    st->gc_goal_miss,
484                    st->gc_dst_overflow,
485                    st->in_hlist_search,
486                    st->out_hlist_search
487                 );
488         return 0;
489 }
490
491 static const struct seq_operations rt_cpu_seq_ops = {
492         .start  = rt_cpu_seq_start,
493         .next   = rt_cpu_seq_next,
494         .stop   = rt_cpu_seq_stop,
495         .show   = rt_cpu_seq_show,
496 };
497
498
499 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
500 {
501         return seq_open(file, &rt_cpu_seq_ops);
502 }
503
504 static const struct file_operations rt_cpu_seq_fops = {
505         .owner   = THIS_MODULE,
506         .open    = rt_cpu_seq_open,
507         .read    = seq_read,
508         .llseek  = seq_lseek,
509         .release = seq_release,
510 };
511
512 #ifdef CONFIG_NET_CLS_ROUTE
513 static int rt_acct_proc_show(struct seq_file *m, void *v)
514 {
515         struct ip_rt_acct *dst, *src;
516         unsigned int i, j;
517
518         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
519         if (!dst)
520                 return -ENOMEM;
521
522         for_each_possible_cpu(i) {
523                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
524                 for (j = 0; j < 256; j++) {
525                         dst[j].o_bytes   += src[j].o_bytes;
526                         dst[j].o_packets += src[j].o_packets;
527                         dst[j].i_bytes   += src[j].i_bytes;
528                         dst[j].i_packets += src[j].i_packets;
529                 }
530         }
531
532         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
533         kfree(dst);
534         return 0;
535 }
536
537 static int rt_acct_proc_open(struct inode *inode, struct file *file)
538 {
539         return single_open(file, rt_acct_proc_show, NULL);
540 }
541
542 static const struct file_operations rt_acct_proc_fops = {
543         .owner          = THIS_MODULE,
544         .open           = rt_acct_proc_open,
545         .read           = seq_read,
546         .llseek         = seq_lseek,
547         .release        = single_release,
548 };
549 #endif
550
551 static int __net_init ip_rt_do_proc_init(struct net *net)
552 {
553         struct proc_dir_entry *pde;
554
555         pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
556                         &rt_cache_seq_fops);
557         if (!pde)
558                 goto err1;
559
560         pde = proc_create("rt_cache", S_IRUGO,
561                           net->proc_net_stat, &rt_cpu_seq_fops);
562         if (!pde)
563                 goto err2;
564
565 #ifdef CONFIG_NET_CLS_ROUTE
566         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
567         if (!pde)
568                 goto err3;
569 #endif
570         return 0;
571
572 #ifdef CONFIG_NET_CLS_ROUTE
573 err3:
574         remove_proc_entry("rt_cache", net->proc_net_stat);
575 #endif
576 err2:
577         remove_proc_entry("rt_cache", net->proc_net);
578 err1:
579         return -ENOMEM;
580 }
581
582 static void __net_exit ip_rt_do_proc_exit(struct net *net)
583 {
584         remove_proc_entry("rt_cache", net->proc_net_stat);
585         remove_proc_entry("rt_cache", net->proc_net);
586 #ifdef CONFIG_NET_CLS_ROUTE
587         remove_proc_entry("rt_acct", net->proc_net);
588 #endif
589 }
590
591 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
592         .init = ip_rt_do_proc_init,
593         .exit = ip_rt_do_proc_exit,
594 };
595
596 static int __init ip_rt_proc_init(void)
597 {
598         return register_pernet_subsys(&ip_rt_proc_ops);
599 }
600
601 #else
602 static inline int ip_rt_proc_init(void)
603 {
604         return 0;
605 }
606 #endif /* CONFIG_PROC_FS */
607
608 static inline void rt_free(struct rtable *rt)
609 {
610         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
611 }
612
613 static inline void rt_drop(struct rtable *rt)
614 {
615         ip_rt_put(rt);
616         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
617 }
618
619 static inline int rt_fast_clean(struct rtable *rth)
620 {
621         /* Kill broadcast/multicast entries very aggresively, if they
622            collide in hash table with more useful entries */
623         return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
624                 rth->fl.iif && rth->dst.rt_next;
625 }
626
627 static inline int rt_valuable(struct rtable *rth)
628 {
629         return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
630                 rth->dst.expires;
631 }
632
633 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
634 {
635         unsigned long age;
636         int ret = 0;
637
638         if (atomic_read(&rth->dst.__refcnt))
639                 goto out;
640
641         ret = 1;
642         if (rth->dst.expires &&
643             time_after_eq(jiffies, rth->dst.expires))
644                 goto out;
645
646         age = jiffies - rth->dst.lastuse;
647         ret = 0;
648         if ((age <= tmo1 && !rt_fast_clean(rth)) ||
649             (age <= tmo2 && rt_valuable(rth)))
650                 goto out;
651         ret = 1;
652 out:    return ret;
653 }
654
655 /* Bits of score are:
656  * 31: very valuable
657  * 30: not quite useless
658  * 29..0: usage counter
659  */
660 static inline u32 rt_score(struct rtable *rt)
661 {
662         u32 score = jiffies - rt->dst.lastuse;
663
664         score = ~score & ~(3<<30);
665
666         if (rt_valuable(rt))
667                 score |= (1<<31);
668
669         if (!rt->fl.iif ||
670             !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
671                 score |= (1<<30);
672
673         return score;
674 }
675
676 static inline bool rt_caching(const struct net *net)
677 {
678         return net->ipv4.current_rt_cache_rebuild_count <=
679                 net->ipv4.sysctl_rt_cache_rebuild_count;
680 }
681
682 static inline bool compare_hash_inputs(const struct flowi *fl1,
683                                         const struct flowi *fl2)
684 {
685         return ((((__force u32)fl1->nl_u.ip4_u.daddr ^ (__force u32)fl2->nl_u.ip4_u.daddr) |
686                 ((__force u32)fl1->nl_u.ip4_u.saddr ^ (__force u32)fl2->nl_u.ip4_u.saddr) |
687                 (fl1->iif ^ fl2->iif)) == 0);
688 }
689
690 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
691 {
692         return (((__force u32)fl1->nl_u.ip4_u.daddr ^ (__force u32)fl2->nl_u.ip4_u.daddr) |
693                 ((__force u32)fl1->nl_u.ip4_u.saddr ^ (__force u32)fl2->nl_u.ip4_u.saddr) |
694                 (fl1->mark ^ fl2->mark) |
695                 (*(u16 *)&fl1->nl_u.ip4_u.tos ^ *(u16 *)&fl2->nl_u.ip4_u.tos) |
696                 (fl1->oif ^ fl2->oif) |
697                 (fl1->iif ^ fl2->iif)) == 0;
698 }
699
700 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
701 {
702         return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
703 }
704
705 static inline int rt_is_expired(struct rtable *rth)
706 {
707         return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
708 }
709
710 /*
711  * Perform a full scan of hash table and free all entries.
712  * Can be called by a softirq or a process.
713  * In the later case, we want to be reschedule if necessary
714  */
715 static void rt_do_flush(int process_context)
716 {
717         unsigned int i;
718         struct rtable *rth, *next;
719         struct rtable * tail;
720
721         for (i = 0; i <= rt_hash_mask; i++) {
722                 if (process_context && need_resched())
723                         cond_resched();
724                 rth = rcu_dereference_raw(rt_hash_table[i].chain);
725                 if (!rth)
726                         continue;
727
728                 spin_lock_bh(rt_hash_lock_addr(i));
729 #ifdef CONFIG_NET_NS
730                 {
731                 struct rtable __rcu **prev;
732                 struct rtable *p;
733
734                 rth = rcu_dereference_protected(rt_hash_table[i].chain,
735                         lockdep_is_held(rt_hash_lock_addr(i)));
736
737                 /* defer releasing the head of the list after spin_unlock */
738                 for (tail = rth; tail;
739                      tail = rcu_dereference_protected(tail->dst.rt_next,
740                                 lockdep_is_held(rt_hash_lock_addr(i))))
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 = rcu_dereference_protected(*prev,
749                                 lockdep_is_held(rt_hash_lock_addr(i)));
750                      p != NULL;
751                      p = next) {
752                         next = rcu_dereference_protected(p->dst.rt_next,
753                                 lockdep_is_held(rt_hash_lock_addr(i)));
754                         if (!rt_is_expired(p)) {
755                                 prev = &p->dst.rt_next;
756                         } else {
757                                 *prev = next;
758                                 rt_free(p);
759                         }
760                 }
761                 }
762 #else
763                 rth = rcu_dereference_protected(rt_hash_table[i].chain,
764                         lockdep_is_held(rt_hash_lock_addr(i)));
765                 rcu_assign_pointer(rt_hash_table[i].chain, NULL);
766                 tail = NULL;
767 #endif
768                 spin_unlock_bh(rt_hash_lock_addr(i));
769
770                 for (; rth != tail; rth = next) {
771                         next = rcu_dereference_protected(rth->dst.rt_next, 1);
772                         rt_free(rth);
773                 }
774         }
775 }
776
777 /*
778  * While freeing expired entries, we compute average chain length
779  * and standard deviation, using fixed-point arithmetic.
780  * This to have an estimation of rt_chain_length_max
781  *  rt_chain_length_max = max(elasticity, AVG + 4*SD)
782  * We use 3 bits for frational part, and 29 (or 61) for magnitude.
783  */
784
785 #define FRACT_BITS 3
786 #define ONE (1UL << FRACT_BITS)
787
788 /*
789  * Given a hash chain and an item in this hash chain,
790  * find if a previous entry has the same hash_inputs
791  * (but differs on tos, mark or oif)
792  * Returns 0 if an alias is found.
793  * Returns ONE if rth has no alias before itself.
794  */
795 static int has_noalias(const struct rtable *head, const struct rtable *rth)
796 {
797         const struct rtable *aux = head;
798
799         while (aux != rth) {
800                 if (compare_hash_inputs(&aux->fl, &rth->fl))
801                         return 0;
802                 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
803         }
804         return ONE;
805 }
806
807 static void rt_check_expire(void)
808 {
809         static unsigned int rover;
810         unsigned int i = rover, goal;
811         struct rtable *rth;
812         struct rtable __rcu **rthp;
813         unsigned long samples = 0;
814         unsigned long sum = 0, sum2 = 0;
815         unsigned long delta;
816         u64 mult;
817
818         delta = jiffies - expires_ljiffies;
819         expires_ljiffies = jiffies;
820         mult = ((u64)delta) << rt_hash_log;
821         if (ip_rt_gc_timeout > 1)
822                 do_div(mult, ip_rt_gc_timeout);
823         goal = (unsigned int)mult;
824         if (goal > rt_hash_mask)
825                 goal = rt_hash_mask + 1;
826         for (; goal > 0; goal--) {
827                 unsigned long tmo = ip_rt_gc_timeout;
828                 unsigned long length;
829
830                 i = (i + 1) & rt_hash_mask;
831                 rthp = &rt_hash_table[i].chain;
832
833                 if (need_resched())
834                         cond_resched();
835
836                 samples++;
837
838                 if (rcu_dereference_raw(*rthp) == NULL)
839                         continue;
840                 length = 0;
841                 spin_lock_bh(rt_hash_lock_addr(i));
842                 while ((rth = rcu_dereference_protected(*rthp,
843                                         lockdep_is_held(rt_hash_lock_addr(i)))) != NULL) {
844                         prefetch(rth->dst.rt_next);
845                         if (rt_is_expired(rth)) {
846                                 *rthp = rth->dst.rt_next;
847                                 rt_free(rth);
848                                 continue;
849                         }
850                         if (rth->dst.expires) {
851                                 /* Entry is expired even if it is in use */
852                                 if (time_before_eq(jiffies, rth->dst.expires)) {
853 nofree:
854                                         tmo >>= 1;
855                                         rthp = &rth->dst.rt_next;
856                                         /*
857                                          * We only count entries on
858                                          * a chain with equal hash inputs once
859                                          * so that entries for different QOS
860                                          * levels, and other non-hash input
861                                          * attributes don't unfairly skew
862                                          * the length computation
863                                          */
864                                         length += has_noalias(rt_hash_table[i].chain, rth);
865                                         continue;
866                                 }
867                         } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout))
868                                 goto nofree;
869
870                         /* Cleanup aged off entries. */
871                         *rthp = rth->dst.rt_next;
872                         rt_free(rth);
873                 }
874                 spin_unlock_bh(rt_hash_lock_addr(i));
875                 sum += length;
876                 sum2 += length*length;
877         }
878         if (samples) {
879                 unsigned long avg = sum / samples;
880                 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
881                 rt_chain_length_max = max_t(unsigned long,
882                                         ip_rt_gc_elasticity,
883                                         (avg + 4*sd) >> FRACT_BITS);
884         }
885         rover = i;
886 }
887
888 /*
889  * rt_worker_func() is run in process context.
890  * we call rt_check_expire() to scan part of the hash table
891  */
892 static void rt_worker_func(struct work_struct *work)
893 {
894         rt_check_expire();
895         schedule_delayed_work(&expires_work, ip_rt_gc_interval);
896 }
897
898 /*
899  * Pertubation of rt_genid by a small quantity [1..256]
900  * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
901  * many times (2^24) without giving recent rt_genid.
902  * Jenkins hash is strong enough that litle changes of rt_genid are OK.
903  */
904 static void rt_cache_invalidate(struct net *net)
905 {
906         unsigned char shuffle;
907
908         get_random_bytes(&shuffle, sizeof(shuffle));
909         atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
910 }
911
912 /*
913  * delay < 0  : invalidate cache (fast : entries will be deleted later)
914  * delay >= 0 : invalidate & flush cache (can be long)
915  */
916 void rt_cache_flush(struct net *net, int delay)
917 {
918         rt_cache_invalidate(net);
919         if (delay >= 0)
920                 rt_do_flush(!in_softirq());
921 }
922
923 /* Flush previous cache invalidated entries from the cache */
924 void rt_cache_flush_batch(void)
925 {
926         rt_do_flush(!in_softirq());
927 }
928
929 static void rt_emergency_hash_rebuild(struct net *net)
930 {
931         if (net_ratelimit())
932                 printk(KERN_WARNING "Route hash chain too long!\n");
933         rt_cache_invalidate(net);
934 }
935
936 /*
937    Short description of GC goals.
938
939    We want to build algorithm, which will keep routing cache
940    at some equilibrium point, when number of aged off entries
941    is kept approximately equal to newly generated ones.
942
943    Current expiration strength is variable "expire".
944    We try to adjust it dynamically, so that if networking
945    is idle expires is large enough to keep enough of warm entries,
946    and when load increases it reduces to limit cache size.
947  */
948
949 static int rt_garbage_collect(struct dst_ops *ops)
950 {
951         static unsigned long expire = RT_GC_TIMEOUT;
952         static unsigned long last_gc;
953         static int rover;
954         static int equilibrium;
955         struct rtable *rth;
956         struct rtable __rcu **rthp;
957         unsigned long now = jiffies;
958         int goal;
959         int entries = dst_entries_get_fast(&ipv4_dst_ops);
960
961         /*
962          * Garbage collection is pretty expensive,
963          * do not make it too frequently.
964          */
965
966         RT_CACHE_STAT_INC(gc_total);
967
968         if (now - last_gc < ip_rt_gc_min_interval &&
969             entries < ip_rt_max_size) {
970                 RT_CACHE_STAT_INC(gc_ignored);
971                 goto out;
972         }
973
974         entries = dst_entries_get_slow(&ipv4_dst_ops);
975         /* Calculate number of entries, which we want to expire now. */
976         goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
977         if (goal <= 0) {
978                 if (equilibrium < ipv4_dst_ops.gc_thresh)
979                         equilibrium = ipv4_dst_ops.gc_thresh;
980                 goal = entries - equilibrium;
981                 if (goal > 0) {
982                         equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
983                         goal = entries - equilibrium;
984                 }
985         } else {
986                 /* We are in dangerous area. Try to reduce cache really
987                  * aggressively.
988                  */
989                 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
990                 equilibrium = entries - goal;
991         }
992
993         if (now - last_gc >= ip_rt_gc_min_interval)
994                 last_gc = now;
995
996         if (goal <= 0) {
997                 equilibrium += goal;
998                 goto work_done;
999         }
1000
1001         do {
1002                 int i, k;
1003
1004                 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
1005                         unsigned long tmo = expire;
1006
1007                         k = (k + 1) & rt_hash_mask;
1008                         rthp = &rt_hash_table[k].chain;
1009                         spin_lock_bh(rt_hash_lock_addr(k));
1010                         while ((rth = rcu_dereference_protected(*rthp,
1011                                         lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
1012                                 if (!rt_is_expired(rth) &&
1013                                         !rt_may_expire(rth, tmo, expire)) {
1014                                         tmo >>= 1;
1015                                         rthp = &rth->dst.rt_next;
1016                                         continue;
1017                                 }
1018                                 *rthp = rth->dst.rt_next;
1019                                 rt_free(rth);
1020                                 goal--;
1021                         }
1022                         spin_unlock_bh(rt_hash_lock_addr(k));
1023                         if (goal <= 0)
1024                                 break;
1025                 }
1026                 rover = k;
1027
1028                 if (goal <= 0)
1029                         goto work_done;
1030
1031                 /* Goal is not achieved. We stop process if:
1032
1033                    - if expire reduced to zero. Otherwise, expire is halfed.
1034                    - if table is not full.
1035                    - if we are called from interrupt.
1036                    - jiffies check is just fallback/debug loop breaker.
1037                      We will not spin here for long time in any case.
1038                  */
1039
1040                 RT_CACHE_STAT_INC(gc_goal_miss);
1041
1042                 if (expire == 0)
1043                         break;
1044
1045                 expire >>= 1;
1046 #if RT_CACHE_DEBUG >= 2
1047                 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
1048                                 dst_entries_get_fast(&ipv4_dst_ops), goal, i);
1049 #endif
1050
1051                 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1052                         goto out;
1053         } while (!in_softirq() && time_before_eq(jiffies, now));
1054
1055         if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1056                 goto out;
1057         if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
1058                 goto out;
1059         if (net_ratelimit())
1060                 printk(KERN_WARNING "dst cache overflow\n");
1061         RT_CACHE_STAT_INC(gc_dst_overflow);
1062         return 1;
1063
1064 work_done:
1065         expire += ip_rt_gc_min_interval;
1066         if (expire > ip_rt_gc_timeout ||
1067             dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
1068             dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1069                 expire = ip_rt_gc_timeout;
1070 #if RT_CACHE_DEBUG >= 2
1071         printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
1072                         dst_entries_get_fast(&ipv4_dst_ops), goal, rover);
1073 #endif
1074 out:    return 0;
1075 }
1076
1077 /*
1078  * Returns number of entries in a hash chain that have different hash_inputs
1079  */
1080 static int slow_chain_length(const struct rtable *head)
1081 {
1082         int length = 0;
1083         const struct rtable *rth = head;
1084
1085         while (rth) {
1086                 length += has_noalias(head, rth);
1087                 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1088         }
1089         return length >> FRACT_BITS;
1090 }
1091
1092 static int rt_intern_hash(unsigned hash, struct rtable *rt,
1093                           struct rtable **rp, struct sk_buff *skb, int ifindex)
1094 {
1095         struct rtable   *rth, *cand;
1096         struct rtable __rcu **rthp, **candp;
1097         unsigned long   now;
1098         u32             min_score;
1099         int             chain_length;
1100         int attempts = !in_softirq();
1101
1102 restart:
1103         chain_length = 0;
1104         min_score = ~(u32)0;
1105         cand = NULL;
1106         candp = NULL;
1107         now = jiffies;
1108
1109         if (!rt_caching(dev_net(rt->dst.dev))) {
1110                 /*
1111                  * If we're not caching, just tell the caller we
1112                  * were successful and don't touch the route.  The
1113                  * caller hold the sole reference to the cache entry, and
1114                  * it will be released when the caller is done with it.
1115                  * If we drop it here, the callers have no way to resolve routes
1116                  * when we're not caching.  Instead, just point *rp at rt, so
1117                  * the caller gets a single use out of the route
1118                  * Note that we do rt_free on this new route entry, so that
1119                  * once its refcount hits zero, we are still able to reap it
1120                  * (Thanks Alexey)
1121                  * Note: To avoid expensive rcu stuff for this uncached dst,
1122                  * we set DST_NOCACHE so that dst_release() can free dst without
1123                  * waiting a grace period.
1124                  */
1125
1126                 rt->dst.flags |= DST_NOCACHE;
1127                 if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1128                         int err = arp_bind_neighbour(&rt->dst);
1129                         if (err) {
1130                                 if (net_ratelimit())
1131                                         printk(KERN_WARNING
1132                                             "Neighbour table failure & not caching routes.\n");
1133                                 ip_rt_put(rt);
1134                                 return err;
1135                         }
1136                 }
1137
1138                 goto skip_hashing;
1139         }
1140
1141         rthp = &rt_hash_table[hash].chain;
1142
1143         spin_lock_bh(rt_hash_lock_addr(hash));
1144         while ((rth = rcu_dereference_protected(*rthp,
1145                         lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1146                 if (rt_is_expired(rth)) {
1147                         *rthp = rth->dst.rt_next;
1148                         rt_free(rth);
1149                         continue;
1150                 }
1151                 if (compare_keys(&rth->fl, &rt->fl) && compare_netns(rth, rt)) {
1152                         /* Put it first */
1153                         *rthp = rth->dst.rt_next;
1154                         /*
1155                          * Since lookup is lockfree, the deletion
1156                          * must be visible to another weakly ordered CPU before
1157                          * the insertion at the start of the hash chain.
1158                          */
1159                         rcu_assign_pointer(rth->dst.rt_next,
1160                                            rt_hash_table[hash].chain);
1161                         /*
1162                          * Since lookup is lockfree, the update writes
1163                          * must be ordered for consistency on SMP.
1164                          */
1165                         rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1166
1167                         dst_use(&rth->dst, now);
1168                         spin_unlock_bh(rt_hash_lock_addr(hash));
1169
1170                         rt_drop(rt);
1171                         if (rp)
1172                                 *rp = rth;
1173                         else
1174                                 skb_dst_set(skb, &rth->dst);
1175                         return 0;
1176                 }
1177
1178                 if (!atomic_read(&rth->dst.__refcnt)) {
1179                         u32 score = rt_score(rth);
1180
1181                         if (score <= min_score) {
1182                                 cand = rth;
1183                                 candp = rthp;
1184                                 min_score = score;
1185                         }
1186                 }
1187
1188                 chain_length++;
1189
1190                 rthp = &rth->dst.rt_next;
1191         }
1192
1193         if (cand) {
1194                 /* ip_rt_gc_elasticity used to be average length of chain
1195                  * length, when exceeded gc becomes really aggressive.
1196                  *
1197                  * The second limit is less certain. At the moment it allows
1198                  * only 2 entries per bucket. We will see.
1199                  */
1200                 if (chain_length > ip_rt_gc_elasticity) {
1201                         *candp = cand->dst.rt_next;
1202                         rt_free(cand);
1203                 }
1204         } else {
1205                 if (chain_length > rt_chain_length_max &&
1206                     slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1207                         struct net *net = dev_net(rt->dst.dev);
1208                         int num = ++net->ipv4.current_rt_cache_rebuild_count;
1209                         if (!rt_caching(net)) {
1210                                 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1211                                         rt->dst.dev->name, num);
1212                         }
1213                         rt_emergency_hash_rebuild(net);
1214                         spin_unlock_bh(rt_hash_lock_addr(hash));
1215
1216                         hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1217                                         ifindex, rt_genid(net));
1218                         goto restart;
1219                 }
1220         }
1221
1222         /* Try to bind route to arp only if it is output
1223            route or unicast forwarding path.
1224          */
1225         if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
1226                 int err = arp_bind_neighbour(&rt->dst);
1227                 if (err) {
1228                         spin_unlock_bh(rt_hash_lock_addr(hash));
1229
1230                         if (err != -ENOBUFS) {
1231                                 rt_drop(rt);
1232                                 return err;
1233                         }
1234
1235                         /* Neighbour tables are full and nothing
1236                            can be released. Try to shrink route cache,
1237                            it is most likely it holds some neighbour records.
1238                          */
1239                         if (attempts-- > 0) {
1240                                 int saved_elasticity = ip_rt_gc_elasticity;
1241                                 int saved_int = ip_rt_gc_min_interval;
1242                                 ip_rt_gc_elasticity     = 1;
1243                                 ip_rt_gc_min_interval   = 0;
1244                                 rt_garbage_collect(&ipv4_dst_ops);
1245                                 ip_rt_gc_min_interval   = saved_int;
1246                                 ip_rt_gc_elasticity     = saved_elasticity;
1247                                 goto restart;
1248                         }
1249
1250                         if (net_ratelimit())
1251                                 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1252                         rt_drop(rt);
1253                         return -ENOBUFS;
1254                 }
1255         }
1256
1257         rt->dst.rt_next = rt_hash_table[hash].chain;
1258
1259 #if RT_CACHE_DEBUG >= 2
1260         if (rt->dst.rt_next) {
1261                 struct rtable *trt;
1262                 printk(KERN_DEBUG "rt_cache @%02x: %pI4",
1263                        hash, &rt->rt_dst);
1264                 for (trt = rt->dst.rt_next; trt; trt = trt->dst.rt_next)
1265                         printk(" . %pI4", &trt->rt_dst);
1266                 printk("\n");
1267         }
1268 #endif
1269         /*
1270          * Since lookup is lockfree, we must make sure
1271          * previous writes to rt are comitted to memory
1272          * before making rt visible to other CPUS.
1273          */
1274         rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1275
1276         spin_unlock_bh(rt_hash_lock_addr(hash));
1277
1278 skip_hashing:
1279         if (rp)
1280                 *rp = rt;
1281         else
1282                 skb_dst_set(skb, &rt->dst);
1283         return 0;
1284 }
1285
1286 void rt_bind_peer(struct rtable *rt, int create)
1287 {
1288         struct inet_peer *peer;
1289
1290         peer = inet_getpeer(rt->rt_dst, create);
1291
1292         if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1293                 inet_putpeer(peer);
1294 }
1295
1296 /*
1297  * Peer allocation may fail only in serious out-of-memory conditions.  However
1298  * we still can generate some output.
1299  * Random ID selection looks a bit dangerous because we have no chances to
1300  * select ID being unique in a reasonable period of time.
1301  * But broken packet identifier may be better than no packet at all.
1302  */
1303 static void ip_select_fb_ident(struct iphdr *iph)
1304 {
1305         static DEFINE_SPINLOCK(ip_fb_id_lock);
1306         static u32 ip_fallback_id;
1307         u32 salt;
1308
1309         spin_lock_bh(&ip_fb_id_lock);
1310         salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1311         iph->id = htons(salt & 0xFFFF);
1312         ip_fallback_id = salt;
1313         spin_unlock_bh(&ip_fb_id_lock);
1314 }
1315
1316 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1317 {
1318         struct rtable *rt = (struct rtable *) dst;
1319
1320         if (rt) {
1321                 if (rt->peer == NULL)
1322                         rt_bind_peer(rt, 1);
1323
1324                 /* If peer is attached to destination, it is never detached,
1325                    so that we need not to grab a lock to dereference it.
1326                  */
1327                 if (rt->peer) {
1328                         iph->id = htons(inet_getid(rt->peer, more));
1329                         return;
1330                 }
1331         } else
1332                 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1333                        __builtin_return_address(0));
1334
1335         ip_select_fb_ident(iph);
1336 }
1337 EXPORT_SYMBOL(__ip_select_ident);
1338
1339 static void rt_del(unsigned hash, struct rtable *rt)
1340 {
1341         struct rtable __rcu **rthp;
1342         struct rtable *aux;
1343
1344         rthp = &rt_hash_table[hash].chain;
1345         spin_lock_bh(rt_hash_lock_addr(hash));
1346         ip_rt_put(rt);
1347         while ((aux = rcu_dereference_protected(*rthp,
1348                         lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1349                 if (aux == rt || rt_is_expired(aux)) {
1350                         *rthp = aux->dst.rt_next;
1351                         rt_free(aux);
1352                         continue;
1353                 }
1354                 rthp = &aux->dst.rt_next;
1355         }
1356         spin_unlock_bh(rt_hash_lock_addr(hash));
1357 }
1358
1359 /* called in rcu_read_lock() section */
1360 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1361                     __be32 saddr, struct net_device *dev)
1362 {
1363         int i, k;
1364         struct in_device *in_dev = __in_dev_get_rcu(dev);
1365         struct rtable *rth;
1366         struct rtable __rcu **rthp;
1367         __be32  skeys[2] = { saddr, 0 };
1368         int  ikeys[2] = { dev->ifindex, 0 };
1369         struct netevent_redirect netevent;
1370         struct net *net;
1371
1372         if (!in_dev)
1373                 return;
1374
1375         net = dev_net(dev);
1376         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1377             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1378             ipv4_is_zeronet(new_gw))
1379                 goto reject_redirect;
1380
1381         if (!rt_caching(net))
1382                 goto reject_redirect;
1383
1384         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1385                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1386                         goto reject_redirect;
1387                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1388                         goto reject_redirect;
1389         } else {
1390                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1391                         goto reject_redirect;
1392         }
1393
1394         for (i = 0; i < 2; i++) {
1395                 for (k = 0; k < 2; k++) {
1396                         unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1397                                                 rt_genid(net));
1398
1399                         rthp = &rt_hash_table[hash].chain;
1400
1401                         while ((rth = rcu_dereference(*rthp)) != NULL) {
1402                                 struct rtable *rt;
1403
1404                                 if (rth->fl.fl4_dst != daddr ||
1405                                     rth->fl.fl4_src != skeys[i] ||
1406                                     rth->fl.oif != ikeys[k] ||
1407                                     rth->fl.iif != 0 ||
1408                                     rt_is_expired(rth) ||
1409                                     !net_eq(dev_net(rth->dst.dev), net)) {
1410                                         rthp = &rth->dst.rt_next;
1411                                         continue;
1412                                 }
1413
1414                                 if (rth->rt_dst != daddr ||
1415                                     rth->rt_src != saddr ||
1416                                     rth->dst.error ||
1417                                     rth->rt_gateway != old_gw ||
1418                                     rth->dst.dev != dev)
1419                                         break;
1420
1421                                 dst_hold(&rth->dst);
1422
1423                                 rt = dst_alloc(&ipv4_dst_ops);
1424                                 if (rt == NULL) {
1425                                         ip_rt_put(rth);
1426                                         return;
1427                                 }
1428
1429                                 /* Copy all the information. */
1430                                 *rt = *rth;
1431                                 rt->dst.__use           = 1;
1432                                 atomic_set(&rt->dst.__refcnt, 1);
1433                                 rt->dst.child           = NULL;
1434                                 if (rt->dst.dev)
1435                                         dev_hold(rt->dst.dev);
1436                                 if (rt->idev)
1437                                         in_dev_hold(rt->idev);
1438                                 rt->dst.obsolete        = -1;
1439                                 rt->dst.lastuse = jiffies;
1440                                 rt->dst.path            = &rt->dst;
1441                                 rt->dst.neighbour       = NULL;
1442                                 rt->dst.hh              = NULL;
1443 #ifdef CONFIG_XFRM
1444                                 rt->dst.xfrm            = NULL;
1445 #endif
1446                                 rt->rt_genid            = rt_genid(net);
1447                                 rt->rt_flags            |= RTCF_REDIRECTED;
1448
1449                                 /* Gateway is different ... */
1450                                 rt->rt_gateway          = new_gw;
1451
1452                                 /* Redirect received -> path was valid */
1453                                 dst_confirm(&rth->dst);
1454
1455                                 if (rt->peer)
1456                                         atomic_inc(&rt->peer->refcnt);
1457
1458                                 if (arp_bind_neighbour(&rt->dst) ||
1459                                     !(rt->dst.neighbour->nud_state &
1460                                             NUD_VALID)) {
1461                                         if (rt->dst.neighbour)
1462                                                 neigh_event_send(rt->dst.neighbour, NULL);
1463                                         ip_rt_put(rth);
1464                                         rt_drop(rt);
1465                                         goto do_next;
1466                                 }
1467
1468                                 netevent.old = &rth->dst;
1469                                 netevent.new = &rt->dst;
1470                                 call_netevent_notifiers(NETEVENT_REDIRECT,
1471                                                         &netevent);
1472
1473                                 rt_del(hash, rth);
1474                                 if (!rt_intern_hash(hash, rt, &rt, NULL, rt->fl.oif))
1475                                         ip_rt_put(rt);
1476                                 goto do_next;
1477                         }
1478                 do_next:
1479                         ;
1480                 }
1481         }
1482         return;
1483
1484 reject_redirect:
1485 #ifdef CONFIG_IP_ROUTE_VERBOSE
1486         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1487                 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1488                         "  Advised path = %pI4 -> %pI4\n",
1489                        &old_gw, dev->name, &new_gw,
1490                        &saddr, &daddr);
1491 #endif
1492         ;
1493 }
1494
1495 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1496 {
1497         struct rtable *rt = (struct rtable *)dst;
1498         struct dst_entry *ret = dst;
1499
1500         if (rt) {
1501                 if (dst->obsolete > 0) {
1502                         ip_rt_put(rt);
1503                         ret = NULL;
1504                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1505                            (rt->dst.expires &&
1506                             time_after_eq(jiffies, rt->dst.expires))) {
1507                         unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1508                                                 rt->fl.oif,
1509                                                 rt_genid(dev_net(dst->dev)));
1510 #if RT_CACHE_DEBUG >= 1
1511                         printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1512                                 &rt->rt_dst, rt->fl.fl4_tos);
1513 #endif
1514                         rt_del(hash, rt);
1515                         ret = NULL;
1516                 }
1517         }
1518         return ret;
1519 }
1520
1521 /*
1522  * Algorithm:
1523  *      1. The first ip_rt_redirect_number redirects are sent
1524  *         with exponential backoff, then we stop sending them at all,
1525  *         assuming that the host ignores our redirects.
1526  *      2. If we did not see packets requiring redirects
1527  *         during ip_rt_redirect_silence, we assume that the host
1528  *         forgot redirected route and start to send redirects again.
1529  *
1530  * This algorithm is much cheaper and more intelligent than dumb load limiting
1531  * in icmp.c.
1532  *
1533  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1534  * and "frag. need" (breaks PMTU discovery) in icmp.c.
1535  */
1536
1537 void ip_rt_send_redirect(struct sk_buff *skb)
1538 {
1539         struct rtable *rt = skb_rtable(skb);
1540         struct in_device *in_dev;
1541         int log_martians;
1542
1543         rcu_read_lock();
1544         in_dev = __in_dev_get_rcu(rt->dst.dev);
1545         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1546                 rcu_read_unlock();
1547                 return;
1548         }
1549         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1550         rcu_read_unlock();
1551
1552         /* No redirected packets during ip_rt_redirect_silence;
1553          * reset the algorithm.
1554          */
1555         if (time_after(jiffies, rt->dst.rate_last + ip_rt_redirect_silence))
1556                 rt->dst.rate_tokens = 0;
1557
1558         /* Too many ignored redirects; do not send anything
1559          * set dst.rate_last to the last seen redirected packet.
1560          */
1561         if (rt->dst.rate_tokens >= ip_rt_redirect_number) {
1562                 rt->dst.rate_last = jiffies;
1563                 return;
1564         }
1565
1566         /* Check for load limit; set rate_last to the latest sent
1567          * redirect.
1568          */
1569         if (rt->dst.rate_tokens == 0 ||
1570             time_after(jiffies,
1571                        (rt->dst.rate_last +
1572                         (ip_rt_redirect_load << rt->dst.rate_tokens)))) {
1573                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1574                 rt->dst.rate_last = jiffies;
1575                 ++rt->dst.rate_tokens;
1576 #ifdef CONFIG_IP_ROUTE_VERBOSE
1577                 if (log_martians &&
1578                     rt->dst.rate_tokens == ip_rt_redirect_number &&
1579                     net_ratelimit())
1580                         printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1581                                 &rt->rt_src, rt->rt_iif,
1582                                 &rt->rt_dst, &rt->rt_gateway);
1583 #endif
1584         }
1585 }
1586
1587 static int ip_error(struct sk_buff *skb)
1588 {
1589         struct rtable *rt = skb_rtable(skb);
1590         unsigned long now;
1591         int code;
1592
1593         switch (rt->dst.error) {
1594                 case EINVAL:
1595                 default:
1596                         goto out;
1597                 case EHOSTUNREACH:
1598                         code = ICMP_HOST_UNREACH;
1599                         break;
1600                 case ENETUNREACH:
1601                         code = ICMP_NET_UNREACH;
1602                         IP_INC_STATS_BH(dev_net(rt->dst.dev),
1603                                         IPSTATS_MIB_INNOROUTES);
1604                         break;
1605                 case EACCES:
1606                         code = ICMP_PKT_FILTERED;
1607                         break;
1608         }
1609
1610         now = jiffies;
1611         rt->dst.rate_tokens += now - rt->dst.rate_last;
1612         if (rt->dst.rate_tokens > ip_rt_error_burst)
1613                 rt->dst.rate_tokens = ip_rt_error_burst;
1614         rt->dst.rate_last = now;
1615         if (rt->dst.rate_tokens >= ip_rt_error_cost) {
1616                 rt->dst.rate_tokens -= ip_rt_error_cost;
1617                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1618         }
1619
1620 out:    kfree_skb(skb);
1621         return 0;
1622 }
1623
1624 /*
1625  *      The last two values are not from the RFC but
1626  *      are needed for AMPRnet AX.25 paths.
1627  */
1628
1629 static const unsigned short mtu_plateau[] =
1630 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1631
1632 static inline unsigned short guess_mtu(unsigned short old_mtu)
1633 {
1634         int i;
1635
1636         for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1637                 if (old_mtu > mtu_plateau[i])
1638                         return mtu_plateau[i];
1639         return 68;
1640 }
1641
1642 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1643                                  unsigned short new_mtu,
1644                                  struct net_device *dev)
1645 {
1646         int i, k;
1647         unsigned short old_mtu = ntohs(iph->tot_len);
1648         struct rtable *rth;
1649         int  ikeys[2] = { dev->ifindex, 0 };
1650         __be32  skeys[2] = { iph->saddr, 0, };
1651         __be32  daddr = iph->daddr;
1652         unsigned short est_mtu = 0;
1653
1654         for (k = 0; k < 2; k++) {
1655                 for (i = 0; i < 2; i++) {
1656                         unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1657                                                 rt_genid(net));
1658
1659                         rcu_read_lock();
1660                         for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1661                              rth = rcu_dereference(rth->dst.rt_next)) {
1662                                 unsigned short mtu = new_mtu;
1663
1664                                 if (rth->fl.fl4_dst != daddr ||
1665                                     rth->fl.fl4_src != skeys[i] ||
1666                                     rth->rt_dst != daddr ||
1667                                     rth->rt_src != iph->saddr ||
1668                                     rth->fl.oif != ikeys[k] ||
1669                                     rth->fl.iif != 0 ||
1670                                     dst_metric_locked(&rth->dst, RTAX_MTU) ||
1671                                     !net_eq(dev_net(rth->dst.dev), net) ||
1672                                     rt_is_expired(rth))
1673                                         continue;
1674
1675                                 if (new_mtu < 68 || new_mtu >= old_mtu) {
1676
1677                                         /* BSD 4.2 compatibility hack :-( */
1678                                         if (mtu == 0 &&
1679                                             old_mtu >= dst_mtu(&rth->dst) &&
1680                                             old_mtu >= 68 + (iph->ihl << 2))
1681                                                 old_mtu -= iph->ihl << 2;
1682
1683                                         mtu = guess_mtu(old_mtu);
1684                                 }
1685                                 if (mtu <= dst_mtu(&rth->dst)) {
1686                                         if (mtu < dst_mtu(&rth->dst)) {
1687                                                 dst_confirm(&rth->dst);
1688                                                 if (mtu < ip_rt_min_pmtu) {
1689                                                         mtu = ip_rt_min_pmtu;
1690                                                         rth->dst.metrics[RTAX_LOCK-1] |=
1691                                                                 (1 << RTAX_MTU);
1692                                                 }
1693                                                 rth->dst.metrics[RTAX_MTU-1] = mtu;
1694                                                 dst_set_expires(&rth->dst,
1695                                                         ip_rt_mtu_expires);
1696                                         }
1697                                         est_mtu = mtu;
1698                                 }
1699                         }
1700                         rcu_read_unlock();
1701                 }
1702         }
1703         return est_mtu ? : new_mtu;
1704 }
1705
1706 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1707 {
1708         if (dst_mtu(dst) > mtu && mtu >= 68 &&
1709             !(dst_metric_locked(dst, RTAX_MTU))) {
1710                 if (mtu < ip_rt_min_pmtu) {
1711                         mtu = ip_rt_min_pmtu;
1712                         dst->metrics[RTAX_LOCK-1] |= (1 << RTAX_MTU);
1713                 }
1714                 dst->metrics[RTAX_MTU-1] = mtu;
1715                 dst_set_expires(dst, ip_rt_mtu_expires);
1716                 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1717         }
1718 }
1719
1720 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1721 {
1722         if (rt_is_expired((struct rtable *)dst))
1723                 return NULL;
1724         return dst;
1725 }
1726
1727 static void ipv4_dst_destroy(struct dst_entry *dst)
1728 {
1729         struct rtable *rt = (struct rtable *) dst;
1730         struct inet_peer *peer = rt->peer;
1731         struct in_device *idev = rt->idev;
1732
1733         if (peer) {
1734                 rt->peer = NULL;
1735                 inet_putpeer(peer);
1736         }
1737
1738         if (idev) {
1739                 rt->idev = NULL;
1740                 in_dev_put(idev);
1741         }
1742 }
1743
1744 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
1745                             int how)
1746 {
1747         struct rtable *rt = (struct rtable *) dst;
1748         struct in_device *idev = rt->idev;
1749         if (dev != dev_net(dev)->loopback_dev && idev && idev->dev == dev) {
1750                 struct in_device *loopback_idev =
1751                         in_dev_get(dev_net(dev)->loopback_dev);
1752                 if (loopback_idev) {
1753                         rt->idev = loopback_idev;
1754                         in_dev_put(idev);
1755                 }
1756         }
1757 }
1758
1759 static void ipv4_link_failure(struct sk_buff *skb)
1760 {
1761         struct rtable *rt;
1762
1763         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1764
1765         rt = skb_rtable(skb);
1766         if (rt)
1767                 dst_set_expires(&rt->dst, 0);
1768 }
1769
1770 static int ip_rt_bug(struct sk_buff *skb)
1771 {
1772         printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1773                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1774                 skb->dev ? skb->dev->name : "?");
1775         kfree_skb(skb);
1776         return 0;
1777 }
1778
1779 /*
1780    We do not cache source address of outgoing interface,
1781    because it is used only by IP RR, TS and SRR options,
1782    so that it out of fast path.
1783
1784    BTW remember: "addr" is allowed to be not aligned
1785    in IP options!
1786  */
1787
1788 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1789 {
1790         __be32 src;
1791         struct fib_result res;
1792
1793         if (rt->fl.iif == 0)
1794                 src = rt->rt_src;
1795         else {
1796                 rcu_read_lock();
1797                 if (fib_lookup(dev_net(rt->dst.dev), &rt->fl, &res) == 0)
1798                         src = FIB_RES_PREFSRC(res);
1799                 else
1800                         src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1801                                         RT_SCOPE_UNIVERSE);
1802                 rcu_read_unlock();
1803         }
1804         memcpy(addr, &src, 4);
1805 }
1806
1807 #ifdef CONFIG_NET_CLS_ROUTE
1808 static void set_class_tag(struct rtable *rt, u32 tag)
1809 {
1810         if (!(rt->dst.tclassid & 0xFFFF))
1811                 rt->dst.tclassid |= tag & 0xFFFF;
1812         if (!(rt->dst.tclassid & 0xFFFF0000))
1813                 rt->dst.tclassid |= tag & 0xFFFF0000;
1814 }
1815 #endif
1816
1817 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1818 {
1819         struct fib_info *fi = res->fi;
1820
1821         if (fi) {
1822                 if (FIB_RES_GW(*res) &&
1823                     FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1824                         rt->rt_gateway = FIB_RES_GW(*res);
1825                 memcpy(rt->dst.metrics, fi->fib_metrics,
1826                        sizeof(rt->dst.metrics));
1827                 if (fi->fib_mtu == 0) {
1828                         rt->dst.metrics[RTAX_MTU-1] = rt->dst.dev->mtu;
1829                         if (dst_metric_locked(&rt->dst, RTAX_MTU) &&
1830                             rt->rt_gateway != rt->rt_dst &&
1831                             rt->dst.dev->mtu > 576)
1832                                 rt->dst.metrics[RTAX_MTU-1] = 576;
1833                 }
1834 #ifdef CONFIG_NET_CLS_ROUTE
1835                 rt->dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1836 #endif
1837         } else
1838                 rt->dst.metrics[RTAX_MTU-1]= rt->dst.dev->mtu;
1839
1840         if (dst_metric(&rt->dst, RTAX_HOPLIMIT) == 0)
1841                 rt->dst.metrics[RTAX_HOPLIMIT-1] = sysctl_ip_default_ttl;
1842         if (dst_mtu(&rt->dst) > IP_MAX_MTU)
1843                 rt->dst.metrics[RTAX_MTU-1] = IP_MAX_MTU;
1844         if (dst_metric(&rt->dst, RTAX_ADVMSS) == 0)
1845                 rt->dst.metrics[RTAX_ADVMSS-1] = max_t(unsigned int, rt->dst.dev->mtu - 40,
1846                                        ip_rt_min_advmss);
1847         if (dst_metric(&rt->dst, RTAX_ADVMSS) > 65535 - 40)
1848                 rt->dst.metrics[RTAX_ADVMSS-1] = 65535 - 40;
1849
1850 #ifdef CONFIG_NET_CLS_ROUTE
1851 #ifdef CONFIG_IP_MULTIPLE_TABLES
1852         set_class_tag(rt, fib_rules_tclass(res));
1853 #endif
1854         set_class_tag(rt, itag);
1855 #endif
1856         rt->rt_type = res->type;
1857 }
1858
1859 /* called in rcu_read_lock() section */
1860 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1861                                 u8 tos, struct net_device *dev, int our)
1862 {
1863         unsigned int hash;
1864         struct rtable *rth;
1865         __be32 spec_dst;
1866         struct in_device *in_dev = __in_dev_get_rcu(dev);
1867         u32 itag = 0;
1868         int err;
1869
1870         /* Primary sanity checks. */
1871
1872         if (in_dev == NULL)
1873                 return -EINVAL;
1874
1875         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1876             ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1877                 goto e_inval;
1878
1879         if (ipv4_is_zeronet(saddr)) {
1880                 if (!ipv4_is_local_multicast(daddr))
1881                         goto e_inval;
1882                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1883         } else {
1884                 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1885                                           &itag, 0);
1886                 if (err < 0)
1887                         goto e_err;
1888         }
1889         rth = dst_alloc(&ipv4_dst_ops);
1890         if (!rth)
1891                 goto e_nobufs;
1892
1893         rth->dst.output = ip_rt_bug;
1894         rth->dst.obsolete = -1;
1895
1896         atomic_set(&rth->dst.__refcnt, 1);
1897         rth->dst.flags= DST_HOST;
1898         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1899                 rth->dst.flags |= DST_NOPOLICY;
1900         rth->fl.fl4_dst = daddr;
1901         rth->rt_dst     = daddr;
1902         rth->fl.fl4_tos = tos;
1903         rth->fl.mark    = skb->mark;
1904         rth->fl.fl4_src = saddr;
1905         rth->rt_src     = saddr;
1906 #ifdef CONFIG_NET_CLS_ROUTE
1907         rth->dst.tclassid = itag;
1908 #endif
1909         rth->rt_iif     =
1910         rth->fl.iif     = dev->ifindex;
1911         rth->dst.dev    = init_net.loopback_dev;
1912         dev_hold(rth->dst.dev);
1913         rth->idev       = in_dev_get(rth->dst.dev);
1914         rth->fl.oif     = 0;
1915         rth->rt_gateway = daddr;
1916         rth->rt_spec_dst= spec_dst;
1917         rth->rt_genid   = rt_genid(dev_net(dev));
1918         rth->rt_flags   = RTCF_MULTICAST;
1919         rth->rt_type    = RTN_MULTICAST;
1920         if (our) {
1921                 rth->dst.input= ip_local_deliver;
1922                 rth->rt_flags |= RTCF_LOCAL;
1923         }
1924
1925 #ifdef CONFIG_IP_MROUTE
1926         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1927                 rth->dst.input = ip_mr_input;
1928 #endif
1929         RT_CACHE_STAT_INC(in_slow_mc);
1930
1931         hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1932         return rt_intern_hash(hash, rth, NULL, skb, dev->ifindex);
1933
1934 e_nobufs:
1935         return -ENOBUFS;
1936 e_inval:
1937         return -EINVAL;
1938 e_err:
1939         return err;
1940 }
1941
1942
1943 static void ip_handle_martian_source(struct net_device *dev,
1944                                      struct in_device *in_dev,
1945                                      struct sk_buff *skb,
1946                                      __be32 daddr,
1947                                      __be32 saddr)
1948 {
1949         RT_CACHE_STAT_INC(in_martian_src);
1950 #ifdef CONFIG_IP_ROUTE_VERBOSE
1951         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1952                 /*
1953                  *      RFC1812 recommendation, if source is martian,
1954                  *      the only hint is MAC header.
1955                  */
1956                 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1957                         &daddr, &saddr, dev->name);
1958                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1959                         int i;
1960                         const unsigned char *p = skb_mac_header(skb);
1961                         printk(KERN_WARNING "ll header: ");
1962                         for (i = 0; i < dev->hard_header_len; i++, p++) {
1963                                 printk("%02x", *p);
1964                                 if (i < (dev->hard_header_len - 1))
1965                                         printk(":");
1966                         }
1967                         printk("\n");
1968                 }
1969         }
1970 #endif
1971 }
1972
1973 /* called in rcu_read_lock() section */
1974 static int __mkroute_input(struct sk_buff *skb,
1975                            struct fib_result *res,
1976                            struct in_device *in_dev,
1977                            __be32 daddr, __be32 saddr, u32 tos,
1978                            struct rtable **result)
1979 {
1980         struct rtable *rth;
1981         int err;
1982         struct in_device *out_dev;
1983         unsigned int flags = 0;
1984         __be32 spec_dst;
1985         u32 itag;
1986
1987         /* get a working reference to the output device */
1988         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1989         if (out_dev == NULL) {
1990                 if (net_ratelimit())
1991                         printk(KERN_CRIT "Bug in ip_route_input" \
1992                                "_slow(). Please, report\n");
1993                 return -EINVAL;
1994         }
1995
1996
1997         err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1998                                   in_dev->dev, &spec_dst, &itag, skb->mark);
1999         if (err < 0) {
2000                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
2001                                          saddr);
2002
2003                 goto cleanup;
2004         }
2005
2006         if (err)
2007                 flags |= RTCF_DIRECTSRC;
2008
2009         if (out_dev == in_dev && err &&
2010             (IN_DEV_SHARED_MEDIA(out_dev) ||
2011              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2012                 flags |= RTCF_DOREDIRECT;
2013
2014         if (skb->protocol != htons(ETH_P_IP)) {
2015                 /* Not IP (i.e. ARP). Do not create route, if it is
2016                  * invalid for proxy arp. DNAT routes are always valid.
2017                  *
2018                  * Proxy arp feature have been extended to allow, ARP
2019                  * replies back to the same interface, to support
2020                  * Private VLAN switch technologies. See arp.c.
2021                  */
2022                 if (out_dev == in_dev &&
2023                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2024                         err = -EINVAL;
2025                         goto cleanup;
2026                 }
2027         }
2028
2029
2030         rth = dst_alloc(&ipv4_dst_ops);
2031         if (!rth) {
2032                 err = -ENOBUFS;
2033                 goto cleanup;
2034         }
2035
2036         atomic_set(&rth->dst.__refcnt, 1);
2037         rth->dst.flags= DST_HOST;
2038         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2039                 rth->dst.flags |= DST_NOPOLICY;
2040         if (IN_DEV_CONF_GET(out_dev, NOXFRM))
2041                 rth->dst.flags |= DST_NOXFRM;
2042         rth->fl.fl4_dst = daddr;
2043         rth->rt_dst     = daddr;
2044         rth->fl.fl4_tos = tos;
2045         rth->fl.mark    = skb->mark;
2046         rth->fl.fl4_src = saddr;
2047         rth->rt_src     = saddr;
2048         rth->rt_gateway = daddr;
2049         rth->rt_iif     =
2050                 rth->fl.iif     = in_dev->dev->ifindex;
2051         rth->dst.dev    = (out_dev)->dev;
2052         dev_hold(rth->dst.dev);
2053         rth->idev       = in_dev_get(rth->dst.dev);
2054         rth->fl.oif     = 0;
2055         rth->rt_spec_dst= spec_dst;
2056
2057         rth->dst.obsolete = -1;
2058         rth->dst.input = ip_forward;
2059         rth->dst.output = ip_output;
2060         rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2061
2062         rt_set_nexthop(rth, res, itag);
2063
2064         rth->rt_flags = flags;
2065
2066         *result = rth;
2067         err = 0;
2068  cleanup:
2069         return err;
2070 }
2071
2072 static int ip_mkroute_input(struct sk_buff *skb,
2073                             struct fib_result *res,
2074                             const struct flowi *fl,
2075                             struct in_device *in_dev,
2076                             __be32 daddr, __be32 saddr, u32 tos)
2077 {
2078         struct rtable* rth = NULL;
2079         int err;
2080         unsigned hash;
2081
2082 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2083         if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
2084                 fib_select_multipath(fl, res);
2085 #endif
2086
2087         /* create a routing cache entry */
2088         err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2089         if (err)
2090                 return err;
2091
2092         /* put it into the cache */
2093         hash = rt_hash(daddr, saddr, fl->iif,
2094                        rt_genid(dev_net(rth->dst.dev)));
2095         return rt_intern_hash(hash, rth, NULL, skb, fl->iif);
2096 }
2097
2098 /*
2099  *      NOTE. We drop all the packets that has local source
2100  *      addresses, because every properly looped back packet
2101  *      must have correct destination already attached by output routine.
2102  *
2103  *      Such approach solves two big problems:
2104  *      1. Not simplex devices are handled properly.
2105  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2106  *      called with rcu_read_lock()
2107  */
2108
2109 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2110                                u8 tos, struct net_device *dev)
2111 {
2112         struct fib_result res;
2113         struct in_device *in_dev = __in_dev_get_rcu(dev);
2114         struct flowi fl = { .nl_u = { .ip4_u =
2115                                       { .daddr = daddr,
2116                                         .saddr = saddr,
2117                                         .tos = tos,
2118                                         .scope = RT_SCOPE_UNIVERSE,
2119                                       } },
2120                             .mark = skb->mark,
2121                             .iif = dev->ifindex };
2122         unsigned        flags = 0;
2123         u32             itag = 0;
2124         struct rtable * rth;
2125         unsigned        hash;
2126         __be32          spec_dst;
2127         int             err = -EINVAL;
2128         struct net    * net = dev_net(dev);
2129
2130         /* IP on this device is disabled. */
2131
2132         if (!in_dev)
2133                 goto out;
2134
2135         /* Check for the most weird martians, which can be not detected
2136            by fib_lookup.
2137          */
2138
2139         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2140             ipv4_is_loopback(saddr))
2141                 goto martian_source;
2142
2143         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2144                 goto brd_input;
2145
2146         /* Accept zero addresses only to limited broadcast;
2147          * I even do not know to fix it or not. Waiting for complains :-)
2148          */
2149         if (ipv4_is_zeronet(saddr))
2150                 goto martian_source;
2151
2152         if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2153                 goto martian_destination;
2154
2155         /*
2156          *      Now we are ready to route packet.
2157          */
2158         err = fib_lookup(net, &fl, &res);
2159         if (err != 0) {
2160                 if (!IN_DEV_FORWARD(in_dev))
2161                         goto e_hostunreach;
2162                 goto no_route;
2163         }
2164
2165         RT_CACHE_STAT_INC(in_slow_tot);
2166
2167         if (res.type == RTN_BROADCAST)
2168                 goto brd_input;
2169
2170         if (res.type == RTN_LOCAL) {
2171                 err = fib_validate_source(saddr, daddr, tos,
2172                                           net->loopback_dev->ifindex,
2173                                           dev, &spec_dst, &itag, skb->mark);
2174                 if (err < 0)
2175                         goto martian_source_keep_err;
2176                 if (err)
2177                         flags |= RTCF_DIRECTSRC;
2178                 spec_dst = daddr;
2179                 goto local_input;
2180         }
2181
2182         if (!IN_DEV_FORWARD(in_dev))
2183                 goto e_hostunreach;
2184         if (res.type != RTN_UNICAST)
2185                 goto martian_destination;
2186
2187         err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
2188 out:    return err;
2189
2190 brd_input:
2191         if (skb->protocol != htons(ETH_P_IP))
2192                 goto e_inval;
2193
2194         if (ipv4_is_zeronet(saddr))
2195                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2196         else {
2197                 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2198                                           &itag, skb->mark);
2199                 if (err < 0)
2200                         goto martian_source_keep_err;
2201                 if (err)
2202                         flags |= RTCF_DIRECTSRC;
2203         }
2204         flags |= RTCF_BROADCAST;
2205         res.type = RTN_BROADCAST;
2206         RT_CACHE_STAT_INC(in_brd);
2207
2208 local_input:
2209         rth = dst_alloc(&ipv4_dst_ops);
2210         if (!rth)
2211                 goto e_nobufs;
2212
2213         rth->dst.output= ip_rt_bug;
2214         rth->dst.obsolete = -1;
2215         rth->rt_genid = rt_genid(net);
2216
2217         atomic_set(&rth->dst.__refcnt, 1);
2218         rth->dst.flags= DST_HOST;
2219         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2220                 rth->dst.flags |= DST_NOPOLICY;
2221         rth->fl.fl4_dst = daddr;
2222         rth->rt_dst     = daddr;
2223         rth->fl.fl4_tos = tos;
2224         rth->fl.mark    = skb->mark;
2225         rth->fl.fl4_src = saddr;
2226         rth->rt_src     = saddr;
2227 #ifdef CONFIG_NET_CLS_ROUTE
2228         rth->dst.tclassid = itag;
2229 #endif
2230         rth->rt_iif     =
2231         rth->fl.iif     = dev->ifindex;
2232         rth->dst.dev    = net->loopback_dev;
2233         dev_hold(rth->dst.dev);
2234         rth->idev       = in_dev_get(rth->dst.dev);
2235         rth->rt_gateway = daddr;
2236         rth->rt_spec_dst= spec_dst;
2237         rth->dst.input= ip_local_deliver;
2238         rth->rt_flags   = flags|RTCF_LOCAL;
2239         if (res.type == RTN_UNREACHABLE) {
2240                 rth->dst.input= ip_error;
2241                 rth->dst.error= -err;
2242                 rth->rt_flags   &= ~RTCF_LOCAL;
2243         }
2244         rth->rt_type    = res.type;
2245         hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));
2246         err = rt_intern_hash(hash, rth, NULL, skb, fl.iif);
2247         goto out;
2248
2249 no_route:
2250         RT_CACHE_STAT_INC(in_no_route);
2251         spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2252         res.type = RTN_UNREACHABLE;
2253         if (err == -ESRCH)
2254                 err = -ENETUNREACH;
2255         goto local_input;
2256
2257         /*
2258          *      Do not cache martian addresses: they should be logged (RFC1812)
2259          */
2260 martian_destination:
2261         RT_CACHE_STAT_INC(in_martian_dst);
2262 #ifdef CONFIG_IP_ROUTE_VERBOSE
2263         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2264                 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2265                         &daddr, &saddr, dev->name);
2266 #endif
2267
2268 e_hostunreach:
2269         err = -EHOSTUNREACH;
2270         goto out;
2271
2272 e_inval:
2273         err = -EINVAL;
2274         goto out;
2275
2276 e_nobufs:
2277         err = -ENOBUFS;
2278         goto out;
2279
2280 martian_source:
2281         err = -EINVAL;
2282 martian_source_keep_err:
2283         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2284         goto out;
2285 }
2286
2287 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2288                            u8 tos, struct net_device *dev, bool noref)
2289 {
2290         struct rtable * rth;
2291         unsigned        hash;
2292         int iif = dev->ifindex;
2293         struct net *net;
2294         int res;
2295
2296         net = dev_net(dev);
2297
2298         rcu_read_lock();
2299
2300         if (!rt_caching(net))
2301                 goto skip_cache;
2302
2303         tos &= IPTOS_RT_MASK;
2304         hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2305
2306         for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2307              rth = rcu_dereference(rth->dst.rt_next)) {
2308                 if ((((__force u32)rth->fl.fl4_dst ^ (__force u32)daddr) |
2309                      ((__force u32)rth->fl.fl4_src ^ (__force u32)saddr) |
2310                      (rth->fl.iif ^ iif) |
2311                      rth->fl.oif |
2312                      (rth->fl.fl4_tos ^ tos)) == 0 &&
2313                     rth->fl.mark == skb->mark &&
2314                     net_eq(dev_net(rth->dst.dev), net) &&
2315                     !rt_is_expired(rth)) {
2316                         if (noref) {
2317                                 dst_use_noref(&rth->dst, jiffies);
2318                                 skb_dst_set_noref(skb, &rth->dst);
2319                         } else {
2320                                 dst_use(&rth->dst, jiffies);
2321                                 skb_dst_set(skb, &rth->dst);
2322                         }
2323                         RT_CACHE_STAT_INC(in_hit);
2324                         rcu_read_unlock();
2325                         return 0;
2326                 }
2327                 RT_CACHE_STAT_INC(in_hlist_search);
2328         }
2329
2330 skip_cache:
2331         /* Multicast recognition logic is moved from route cache to here.
2332            The problem was that too many Ethernet cards have broken/missing
2333            hardware multicast filters :-( As result the host on multicasting
2334            network acquires a lot of useless route cache entries, sort of
2335            SDR messages from all the world. Now we try to get rid of them.
2336            Really, provided software IP multicast filter is organized
2337            reasonably (at least, hashed), it does not result in a slowdown
2338            comparing with route cache reject entries.
2339            Note, that multicast routers are not affected, because
2340            route cache entry is created eventually.
2341          */
2342         if (ipv4_is_multicast(daddr)) {
2343                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2344
2345                 if (in_dev) {
2346                         int our = ip_check_mc(in_dev, daddr, saddr,
2347                                               ip_hdr(skb)->protocol);
2348                         if (our
2349 #ifdef CONFIG_IP_MROUTE
2350                                 ||
2351                             (!ipv4_is_local_multicast(daddr) &&
2352                              IN_DEV_MFORWARD(in_dev))
2353 #endif
2354                            ) {
2355                                 int res = ip_route_input_mc(skb, daddr, saddr,
2356                                                             tos, dev, our);
2357                                 rcu_read_unlock();
2358                                 return res;
2359                         }
2360                 }
2361                 rcu_read_unlock();
2362                 return -EINVAL;
2363         }
2364         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2365         rcu_read_unlock();
2366         return res;
2367 }
2368 EXPORT_SYMBOL(ip_route_input_common);
2369
2370 /* called with rcu_read_lock() */
2371 static int __mkroute_output(struct rtable **result,
2372                             struct fib_result *res,
2373                             const struct flowi *fl,
2374                             const struct flowi *oldflp,
2375                             struct net_device *dev_out,
2376                             unsigned flags)
2377 {
2378         struct rtable *rth;
2379         struct in_device *in_dev;
2380         u32 tos = RT_FL_TOS(oldflp);
2381
2382         if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags & IFF_LOOPBACK))
2383                 return -EINVAL;
2384
2385         if (ipv4_is_lbcast(fl->fl4_dst))
2386                 res->type = RTN_BROADCAST;
2387         else if (ipv4_is_multicast(fl->fl4_dst))
2388                 res->type = RTN_MULTICAST;
2389         else if (ipv4_is_zeronet(fl->fl4_dst))
2390                 return -EINVAL;
2391
2392         if (dev_out->flags & IFF_LOOPBACK)
2393                 flags |= RTCF_LOCAL;
2394
2395         in_dev = __in_dev_get_rcu(dev_out);
2396         if (!in_dev)
2397                 return -EINVAL;
2398
2399         if (res->type == RTN_BROADCAST) {
2400                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2401                 res->fi = NULL;
2402         } else if (res->type == RTN_MULTICAST) {
2403                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2404                 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2405                                  oldflp->proto))
2406                         flags &= ~RTCF_LOCAL;
2407                 /* If multicast route do not exist use
2408                  * default one, but do not gateway in this case.
2409                  * Yes, it is hack.
2410                  */
2411                 if (res->fi && res->prefixlen < 4)
2412                         res->fi = NULL;
2413         }
2414
2415
2416         rth = dst_alloc(&ipv4_dst_ops);
2417         if (!rth)
2418                 return -ENOBUFS;
2419
2420         in_dev_hold(in_dev);
2421         rth->idev = in_dev;
2422
2423         atomic_set(&rth->dst.__refcnt, 1);
2424         rth->dst.flags= DST_HOST;
2425         if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2426                 rth->dst.flags |= DST_NOXFRM;
2427         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2428                 rth->dst.flags |= DST_NOPOLICY;
2429
2430         rth->fl.fl4_dst = oldflp->fl4_dst;
2431         rth->fl.fl4_tos = tos;
2432         rth->fl.fl4_src = oldflp->fl4_src;
2433         rth->fl.oif     = oldflp->oif;
2434         rth->fl.mark    = oldflp->mark;
2435         rth->rt_dst     = fl->fl4_dst;
2436         rth->rt_src     = fl->fl4_src;
2437         rth->rt_iif     = oldflp->oif ? : dev_out->ifindex;
2438         /* get references to the devices that are to be hold by the routing
2439            cache entry */
2440         rth->dst.dev    = dev_out;
2441         dev_hold(dev_out);
2442         rth->rt_gateway = fl->fl4_dst;
2443         rth->rt_spec_dst= fl->fl4_src;
2444
2445         rth->dst.output=ip_output;
2446         rth->dst.obsolete = -1;
2447         rth->rt_genid = rt_genid(dev_net(dev_out));
2448
2449         RT_CACHE_STAT_INC(out_slow_tot);
2450
2451         if (flags & RTCF_LOCAL) {
2452                 rth->dst.input = ip_local_deliver;
2453                 rth->rt_spec_dst = fl->fl4_dst;
2454         }
2455         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2456                 rth->rt_spec_dst = fl->fl4_src;
2457                 if (flags & RTCF_LOCAL &&
2458                     !(dev_out->flags & IFF_LOOPBACK)) {
2459                         rth->dst.output = ip_mc_output;
2460                         RT_CACHE_STAT_INC(out_slow_mc);
2461                 }
2462 #ifdef CONFIG_IP_MROUTE
2463                 if (res->type == RTN_MULTICAST) {
2464                         if (IN_DEV_MFORWARD(in_dev) &&
2465                             !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2466                                 rth->dst.input = ip_mr_input;
2467                                 rth->dst.output = ip_mc_output;
2468                         }
2469                 }
2470 #endif
2471         }
2472
2473         rt_set_nexthop(rth, res, 0);
2474
2475         rth->rt_flags = flags;
2476         *result = rth;
2477         return 0;
2478 }
2479
2480 /* called with rcu_read_lock() */
2481 static int ip_mkroute_output(struct rtable **rp,
2482                              struct fib_result *res,
2483                              const struct flowi *fl,
2484                              const struct flowi *oldflp,
2485                              struct net_device *dev_out,
2486                              unsigned flags)
2487 {
2488         struct rtable *rth = NULL;
2489         int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2490         unsigned hash;
2491         if (err == 0) {
2492                 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif,
2493                                rt_genid(dev_net(dev_out)));
2494                 err = rt_intern_hash(hash, rth, rp, NULL, oldflp->oif);
2495         }
2496
2497         return err;
2498 }
2499
2500 /*
2501  * Major route resolver routine.
2502  * called with rcu_read_lock();
2503  */
2504
2505 static int ip_route_output_slow(struct net *net, struct rtable **rp,
2506                                 const struct flowi *oldflp)
2507 {
2508         u32 tos = RT_FL_TOS(oldflp);
2509         struct flowi fl = { .nl_u = { .ip4_u =
2510                                       { .daddr = oldflp->fl4_dst,
2511                                         .saddr = oldflp->fl4_src,
2512                                         .tos = tos & IPTOS_RT_MASK,
2513                                         .scope = ((tos & RTO_ONLINK) ?
2514                                                   RT_SCOPE_LINK :
2515                                                   RT_SCOPE_UNIVERSE),
2516                                       } },
2517                             .mark = oldflp->mark,
2518                             .iif = net->loopback_dev->ifindex,
2519                             .oif = oldflp->oif };
2520         struct fib_result res;
2521         unsigned int flags = 0;
2522         struct net_device *dev_out = NULL;
2523         int err;
2524
2525
2526         res.fi          = NULL;
2527 #ifdef CONFIG_IP_MULTIPLE_TABLES
2528         res.r           = NULL;
2529 #endif
2530
2531         if (oldflp->fl4_src) {
2532                 err = -EINVAL;
2533                 if (ipv4_is_multicast(oldflp->fl4_src) ||
2534                     ipv4_is_lbcast(oldflp->fl4_src) ||
2535                     ipv4_is_zeronet(oldflp->fl4_src))
2536                         goto out;
2537
2538                 /* I removed check for oif == dev_out->oif here.
2539                    It was wrong for two reasons:
2540                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2541                       is assigned to multiple interfaces.
2542                    2. Moreover, we are allowed to send packets with saddr
2543                       of another iface. --ANK
2544                  */
2545
2546                 if (oldflp->oif == 0 &&
2547                     (ipv4_is_multicast(oldflp->fl4_dst) ||
2548                      ipv4_is_lbcast(oldflp->fl4_dst))) {
2549                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2550                         dev_out = __ip_dev_find(net, oldflp->fl4_src, false);
2551                         if (dev_out == NULL)
2552                                 goto out;
2553
2554                         /* Special hack: user can direct multicasts
2555                            and limited broadcast via necessary interface
2556                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2557                            This hack is not just for fun, it allows
2558                            vic,vat and friends to work.
2559                            They bind socket to loopback, set ttl to zero
2560                            and expect that it will work.
2561                            From the viewpoint of routing cache they are broken,
2562                            because we are not allowed to build multicast path
2563                            with loopback source addr (look, routing cache
2564                            cannot know, that ttl is zero, so that packet
2565                            will not leave this host and route is valid).
2566                            Luckily, this hack is good workaround.
2567                          */
2568
2569                         fl.oif = dev_out->ifindex;
2570                         goto make_route;
2571                 }
2572
2573                 if (!(oldflp->flags & FLOWI_FLAG_ANYSRC)) {
2574                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2575                         if (!__ip_dev_find(net, oldflp->fl4_src, false))
2576                                 goto out;
2577                 }
2578         }
2579
2580
2581         if (oldflp->oif) {
2582                 dev_out = dev_get_by_index_rcu(net, oldflp->oif);
2583                 err = -ENODEV;
2584                 if (dev_out == NULL)
2585                         goto out;
2586
2587                 /* RACE: Check return value of inet_select_addr instead. */
2588                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2589                         err = -ENETUNREACH;
2590                         goto out;
2591                 }
2592                 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2593                     ipv4_is_lbcast(oldflp->fl4_dst)) {
2594                         if (!fl.fl4_src)
2595                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2596                                                               RT_SCOPE_LINK);
2597                         goto make_route;
2598                 }
2599                 if (!fl.fl4_src) {
2600                         if (ipv4_is_multicast(oldflp->fl4_dst))
2601                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2602                                                               fl.fl4_scope);
2603                         else if (!oldflp->fl4_dst)
2604                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2605                                                               RT_SCOPE_HOST);
2606                 }
2607         }
2608
2609         if (!fl.fl4_dst) {
2610                 fl.fl4_dst = fl.fl4_src;
2611                 if (!fl.fl4_dst)
2612                         fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2613                 dev_out = net->loopback_dev;
2614                 fl.oif = net->loopback_dev->ifindex;
2615                 res.type = RTN_LOCAL;
2616                 flags |= RTCF_LOCAL;
2617                 goto make_route;
2618         }
2619
2620         if (fib_lookup(net, &fl, &res)) {
2621                 res.fi = NULL;
2622                 if (oldflp->oif) {
2623                         /* Apparently, routing tables are wrong. Assume,
2624                            that the destination is on link.
2625
2626                            WHY? DW.
2627                            Because we are allowed to send to iface
2628                            even if it has NO routes and NO assigned
2629                            addresses. When oif is specified, routing
2630                            tables are looked up with only one purpose:
2631                            to catch if destination is gatewayed, rather than
2632                            direct. Moreover, if MSG_DONTROUTE is set,
2633                            we send packet, ignoring both routing tables
2634                            and ifaddr state. --ANK
2635
2636
2637                            We could make it even if oif is unknown,
2638                            likely IPv6, but we do not.
2639                          */
2640
2641                         if (fl.fl4_src == 0)
2642                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2643                                                               RT_SCOPE_LINK);
2644                         res.type = RTN_UNICAST;
2645                         goto make_route;
2646                 }
2647                 err = -ENETUNREACH;
2648                 goto out;
2649         }
2650
2651         if (res.type == RTN_LOCAL) {
2652                 if (!fl.fl4_src) {
2653                         if (res.fi->fib_prefsrc)
2654                                 fl.fl4_src = res.fi->fib_prefsrc;
2655                         else
2656                                 fl.fl4_src = fl.fl4_dst;
2657                 }
2658                 dev_out = net->loopback_dev;
2659                 fl.oif = dev_out->ifindex;
2660                 res.fi = NULL;
2661                 flags |= RTCF_LOCAL;
2662                 goto make_route;
2663         }
2664
2665 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2666         if (res.fi->fib_nhs > 1 && fl.oif == 0)
2667                 fib_select_multipath(&fl, &res);
2668         else
2669 #endif
2670         if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2671                 fib_select_default(net, &fl, &res);
2672
2673         if (!fl.fl4_src)
2674                 fl.fl4_src = FIB_RES_PREFSRC(res);
2675
2676         dev_out = FIB_RES_DEV(res);
2677         fl.oif = dev_out->ifindex;
2678
2679
2680 make_route:
2681         err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2682
2683 out:    return err;
2684 }
2685
2686 int __ip_route_output_key(struct net *net, struct rtable **rp,
2687                           const struct flowi *flp)
2688 {
2689         unsigned int hash;
2690         int res;
2691         struct rtable *rth;
2692
2693         if (!rt_caching(net))
2694                 goto slow_output;
2695
2696         hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, rt_genid(net));
2697
2698         rcu_read_lock_bh();
2699         for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2700                 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2701                 if (rth->fl.fl4_dst == flp->fl4_dst &&
2702                     rth->fl.fl4_src == flp->fl4_src &&
2703                     rth->fl.iif == 0 &&
2704                     rth->fl.oif == flp->oif &&
2705                     rth->fl.mark == flp->mark &&
2706                     !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2707                             (IPTOS_RT_MASK | RTO_ONLINK)) &&
2708                     net_eq(dev_net(rth->dst.dev), net) &&
2709                     !rt_is_expired(rth)) {
2710                         dst_use(&rth->dst, jiffies);
2711                         RT_CACHE_STAT_INC(out_hit);
2712                         rcu_read_unlock_bh();
2713                         *rp = rth;
2714                         return 0;
2715                 }
2716                 RT_CACHE_STAT_INC(out_hlist_search);
2717         }
2718         rcu_read_unlock_bh();
2719
2720 slow_output:
2721         rcu_read_lock();
2722         res = ip_route_output_slow(net, rp, flp);
2723         rcu_read_unlock();
2724         return res;
2725 }
2726 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2727
2728 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2729 {
2730         return NULL;
2731 }
2732
2733 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2734 {
2735 }
2736
2737 static struct dst_ops ipv4_dst_blackhole_ops = {
2738         .family                 =       AF_INET,
2739         .protocol               =       cpu_to_be16(ETH_P_IP),
2740         .destroy                =       ipv4_dst_destroy,
2741         .check                  =       ipv4_blackhole_dst_check,
2742         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2743 };
2744
2745
2746 static int ipv4_dst_blackhole(struct net *net, struct rtable **rp, struct flowi *flp)
2747 {
2748         struct rtable *ort = *rp;
2749         struct rtable *rt = (struct rtable *)
2750                 dst_alloc(&ipv4_dst_blackhole_ops);
2751
2752         if (rt) {
2753                 struct dst_entry *new = &rt->dst;
2754
2755                 atomic_set(&new->__refcnt, 1);
2756                 new->__use = 1;
2757                 new->input = dst_discard;
2758                 new->output = dst_discard;
2759                 memcpy(new->metrics, ort->dst.metrics, RTAX_MAX*sizeof(u32));
2760
2761                 new->dev = ort->dst.dev;
2762                 if (new->dev)
2763                         dev_hold(new->dev);
2764
2765                 rt->fl = ort->fl;
2766
2767                 rt->idev = ort->idev;
2768                 if (rt->idev)
2769                         in_dev_hold(rt->idev);
2770                 rt->rt_genid = rt_genid(net);
2771                 rt->rt_flags = ort->rt_flags;
2772                 rt->rt_type = ort->rt_type;
2773                 rt->rt_dst = ort->rt_dst;
2774                 rt->rt_src = ort->rt_src;
2775                 rt->rt_iif = ort->rt_iif;
2776                 rt->rt_gateway = ort->rt_gateway;
2777                 rt->rt_spec_dst = ort->rt_spec_dst;
2778                 rt->peer = ort->peer;
2779                 if (rt->peer)
2780                         atomic_inc(&rt->peer->refcnt);
2781
2782                 dst_free(new);
2783         }
2784
2785         dst_release(&(*rp)->dst);
2786         *rp = rt;
2787         return rt ? 0 : -ENOMEM;
2788 }
2789
2790 int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2791                          struct sock *sk, int flags)
2792 {
2793         int err;
2794
2795         if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2796                 return err;
2797
2798         if (flp->proto) {
2799                 if (!flp->fl4_src)
2800                         flp->fl4_src = (*rp)->rt_src;
2801                 if (!flp->fl4_dst)
2802                         flp->fl4_dst = (*rp)->rt_dst;
2803                 err = __xfrm_lookup(net, (struct dst_entry **)rp, flp, sk,
2804                                     flags ? XFRM_LOOKUP_WAIT : 0);
2805                 if (err == -EREMOTE)
2806                         err = ipv4_dst_blackhole(net, rp, flp);
2807
2808                 return err;
2809         }
2810
2811         return 0;
2812 }
2813 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2814
2815 int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2816 {
2817         return ip_route_output_flow(net, rp, flp, NULL, 0);
2818 }
2819 EXPORT_SYMBOL(ip_route_output_key);
2820
2821 static int rt_fill_info(struct net *net,
2822                         struct sk_buff *skb, u32 pid, u32 seq, int event,
2823                         int nowait, unsigned int flags)
2824 {
2825         struct rtable *rt = skb_rtable(skb);
2826         struct rtmsg *r;
2827         struct nlmsghdr *nlh;
2828         long expires;
2829         u32 id = 0, ts = 0, tsage = 0, error;
2830
2831         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2832         if (nlh == NULL)
2833                 return -EMSGSIZE;
2834
2835         r = nlmsg_data(nlh);
2836         r->rtm_family    = AF_INET;
2837         r->rtm_dst_len  = 32;
2838         r->rtm_src_len  = 0;
2839         r->rtm_tos      = rt->fl.fl4_tos;
2840         r->rtm_table    = RT_TABLE_MAIN;
2841         NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2842         r->rtm_type     = rt->rt_type;
2843         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2844         r->rtm_protocol = RTPROT_UNSPEC;
2845         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2846         if (rt->rt_flags & RTCF_NOTIFY)
2847                 r->rtm_flags |= RTM_F_NOTIFY;
2848
2849         NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2850
2851         if (rt->fl.fl4_src) {
2852                 r->rtm_src_len = 32;
2853                 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2854         }
2855         if (rt->dst.dev)
2856                 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2857 #ifdef CONFIG_NET_CLS_ROUTE
2858         if (rt->dst.tclassid)
2859                 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2860 #endif
2861         if (rt->fl.iif)
2862                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2863         else if (rt->rt_src != rt->fl.fl4_src)
2864                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2865
2866         if (rt->rt_dst != rt->rt_gateway)
2867                 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2868
2869         if (rtnetlink_put_metrics(skb, rt->dst.metrics) < 0)
2870                 goto nla_put_failure;
2871
2872         if (rt->fl.mark)
2873                 NLA_PUT_BE32(skb, RTA_MARK, rt->fl.mark);
2874
2875         error = rt->dst.error;
2876         expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
2877         if (rt->peer) {
2878                 inet_peer_refcheck(rt->peer);
2879                 id = atomic_read(&rt->peer->ip_id_count) & 0xffff;
2880                 if (rt->peer->tcp_ts_stamp) {
2881                         ts = rt->peer->tcp_ts;
2882                         tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2883                 }
2884         }
2885
2886         if (rt->fl.iif) {
2887 #ifdef CONFIG_IP_MROUTE
2888                 __be32 dst = rt->rt_dst;
2889
2890                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2891                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2892                         int err = ipmr_get_route(net, skb, r, nowait);
2893                         if (err <= 0) {
2894                                 if (!nowait) {
2895                                         if (err == 0)
2896                                                 return 0;
2897                                         goto nla_put_failure;
2898                                 } else {
2899                                         if (err == -EMSGSIZE)
2900                                                 goto nla_put_failure;
2901                                         error = err;
2902                                 }
2903                         }
2904                 } else
2905 #endif
2906                         NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2907         }
2908
2909         if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2910                                expires, error) < 0)
2911                 goto nla_put_failure;
2912
2913         return nlmsg_end(skb, nlh);
2914
2915 nla_put_failure:
2916         nlmsg_cancel(skb, nlh);
2917         return -EMSGSIZE;
2918 }
2919
2920 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2921 {
2922         struct net *net = sock_net(in_skb->sk);
2923         struct rtmsg *rtm;
2924         struct nlattr *tb[RTA_MAX+1];
2925         struct rtable *rt = NULL;
2926         __be32 dst = 0;
2927         __be32 src = 0;
2928         u32 iif;
2929         int err;
2930         int mark;
2931         struct sk_buff *skb;
2932
2933         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2934         if (err < 0)
2935                 goto errout;
2936
2937         rtm = nlmsg_data(nlh);
2938
2939         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2940         if (skb == NULL) {
2941                 err = -ENOBUFS;
2942                 goto errout;
2943         }
2944
2945         /* Reserve room for dummy headers, this skb can pass
2946            through good chunk of routing engine.
2947          */
2948         skb_reset_mac_header(skb);
2949         skb_reset_network_header(skb);
2950
2951         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2952         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2953         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2954
2955         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2956         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2957         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2958         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2959
2960         if (iif) {
2961                 struct net_device *dev;
2962
2963                 dev = __dev_get_by_index(net, iif);
2964                 if (dev == NULL) {
2965                         err = -ENODEV;
2966                         goto errout_free;
2967                 }
2968
2969                 skb->protocol   = htons(ETH_P_IP);
2970                 skb->dev        = dev;
2971                 skb->mark       = mark;
2972                 local_bh_disable();
2973                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2974                 local_bh_enable();
2975
2976                 rt = skb_rtable(skb);
2977                 if (err == 0 && rt->dst.error)
2978                         err = -rt->dst.error;
2979         } else {
2980                 struct flowi fl = {
2981                         .nl_u = {
2982                                 .ip4_u = {
2983                                         .daddr = dst,
2984                                         .saddr = src,
2985                                         .tos = rtm->rtm_tos,
2986                                 },
2987                         },
2988                         .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2989                         .mark = mark,
2990                 };
2991                 err = ip_route_output_key(net, &rt, &fl);
2992         }
2993
2994         if (err)
2995                 goto errout_free;
2996
2997         skb_dst_set(skb, &rt->dst);
2998         if (rtm->rtm_flags & RTM_F_NOTIFY)
2999                 rt->rt_flags |= RTCF_NOTIFY;
3000
3001         err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
3002                            RTM_NEWROUTE, 0, 0);
3003         if (err <= 0)
3004                 goto errout_free;
3005
3006         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
3007 errout:
3008         return err;
3009
3010 errout_free:
3011         kfree_skb(skb);
3012         goto errout;
3013 }
3014
3015 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
3016 {
3017         struct rtable *rt;
3018         int h, s_h;
3019         int idx, s_idx;
3020         struct net *net;
3021
3022         net = sock_net(skb->sk);
3023
3024         s_h = cb->args[0];
3025         if (s_h < 0)
3026                 s_h = 0;
3027         s_idx = idx = cb->args[1];
3028         for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
3029                 if (!rt_hash_table[h].chain)
3030                         continue;
3031                 rcu_read_lock_bh();
3032                 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3033                      rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3034                         if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3035                                 continue;
3036                         if (rt_is_expired(rt))
3037                                 continue;
3038                         skb_dst_set_noref(skb, &rt->dst);
3039                         if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3040                                          cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3041                                          1, NLM_F_MULTI) <= 0) {
3042                                 skb_dst_drop(skb);
3043                                 rcu_read_unlock_bh();
3044                                 goto done;
3045                         }
3046                         skb_dst_drop(skb);
3047                 }
3048                 rcu_read_unlock_bh();
3049         }
3050
3051 done:
3052         cb->args[0] = h;
3053         cb->args[1] = idx;
3054         return skb->len;
3055 }
3056
3057 void ip_rt_multicast_event(struct in_device *in_dev)
3058 {
3059         rt_cache_flush(dev_net(in_dev->dev), 0);
3060 }
3061
3062 #ifdef CONFIG_SYSCTL
3063 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3064                                         void __user *buffer,
3065                                         size_t *lenp, loff_t *ppos)
3066 {
3067         if (write) {
3068                 int flush_delay;
3069                 ctl_table ctl;
3070                 struct net *net;
3071
3072                 memcpy(&ctl, __ctl, sizeof(ctl));
3073                 ctl.data = &flush_delay;
3074                 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3075
3076                 net = (struct net *)__ctl->extra1;
3077                 rt_cache_flush(net, flush_delay);
3078                 return 0;
3079         }
3080
3081         return -EINVAL;
3082 }
3083
3084 static ctl_table ipv4_route_table[] = {
3085         {
3086                 .procname       = "gc_thresh",
3087                 .data           = &ipv4_dst_ops.gc_thresh,
3088                 .maxlen         = sizeof(int),
3089                 .mode           = 0644,
3090                 .proc_handler   = proc_dointvec,
3091         },
3092         {
3093                 .procname       = "max_size",
3094                 .data           = &ip_rt_max_size,
3095                 .maxlen         = sizeof(int),
3096                 .mode           = 0644,
3097                 .proc_handler   = proc_dointvec,
3098         },
3099         {
3100                 /*  Deprecated. Use gc_min_interval_ms */
3101
3102                 .procname       = "gc_min_interval",
3103                 .data           = &ip_rt_gc_min_interval,
3104                 .maxlen         = sizeof(int),
3105                 .mode           = 0644,
3106                 .proc_handler   = proc_dointvec_jiffies,
3107         },
3108         {
3109                 .procname       = "gc_min_interval_ms",
3110                 .data           = &ip_rt_gc_min_interval,
3111                 .maxlen         = sizeof(int),
3112                 .mode           = 0644,
3113                 .proc_handler   = proc_dointvec_ms_jiffies,
3114         },
3115         {
3116                 .procname       = "gc_timeout",
3117                 .data           = &ip_rt_gc_timeout,
3118                 .maxlen         = sizeof(int),
3119                 .mode           = 0644,
3120                 .proc_handler   = proc_dointvec_jiffies,
3121         },
3122         {
3123                 .procname       = "gc_interval",
3124                 .data           = &ip_rt_gc_interval,
3125                 .maxlen         = sizeof(int),
3126                 .mode           = 0644,
3127                 .proc_handler   = proc_dointvec_jiffies,
3128         },
3129         {
3130                 .procname       = "redirect_load",
3131                 .data           = &ip_rt_redirect_load,
3132                 .maxlen         = sizeof(int),
3133                 .mode           = 0644,
3134                 .proc_handler   = proc_dointvec,
3135         },
3136         {
3137                 .procname       = "redirect_number",
3138                 .data           = &ip_rt_redirect_number,
3139                 .maxlen         = sizeof(int),
3140                 .mode           = 0644,
3141                 .proc_handler   = proc_dointvec,
3142         },
3143         {
3144                 .procname       = "redirect_silence",
3145                 .data           = &ip_rt_redirect_silence,
3146                 .maxlen         = sizeof(int),
3147                 .mode           = 0644,
3148                 .proc_handler   = proc_dointvec,
3149         },
3150         {
3151                 .procname       = "error_cost",
3152                 .data           = &ip_rt_error_cost,
3153                 .maxlen         = sizeof(int),
3154                 .mode           = 0644,
3155                 .proc_handler   = proc_dointvec,
3156         },
3157         {
3158                 .procname       = "error_burst",
3159                 .data           = &ip_rt_error_burst,
3160                 .maxlen         = sizeof(int),
3161                 .mode           = 0644,
3162                 .proc_handler   = proc_dointvec,
3163         },
3164         {
3165                 .procname       = "gc_elasticity",
3166                 .data           = &ip_rt_gc_elasticity,
3167                 .maxlen         = sizeof(int),
3168                 .mode           = 0644,
3169                 .proc_handler   = proc_dointvec,
3170         },
3171         {
3172                 .procname       = "mtu_expires",
3173                 .data           = &ip_rt_mtu_expires,
3174                 .maxlen         = sizeof(int),
3175                 .mode           = 0644,
3176                 .proc_handler   = proc_dointvec_jiffies,
3177         },
3178         {
3179                 .procname       = "min_pmtu",
3180                 .data           = &ip_rt_min_pmtu,
3181                 .maxlen         = sizeof(int),
3182                 .mode           = 0644,
3183                 .proc_handler   = proc_dointvec,
3184         },
3185         {
3186                 .procname       = "min_adv_mss",
3187                 .data           = &ip_rt_min_advmss,
3188                 .maxlen         = sizeof(int),
3189                 .mode           = 0644,
3190                 .proc_handler   = proc_dointvec,
3191         },
3192         { }
3193 };
3194
3195 static struct ctl_table empty[1];
3196
3197 static struct ctl_table ipv4_skeleton[] =
3198 {
3199         { .procname = "route", 
3200           .mode = 0555, .child = ipv4_route_table},
3201         { .procname = "neigh", 
3202           .mode = 0555, .child = empty},
3203         { }
3204 };
3205
3206 static __net_initdata struct ctl_path ipv4_path[] = {
3207         { .procname = "net", },
3208         { .procname = "ipv4", },
3209         { },
3210 };
3211
3212 static struct ctl_table ipv4_route_flush_table[] = {
3213         {
3214                 .procname       = "flush",
3215                 .maxlen         = sizeof(int),
3216                 .mode           = 0200,
3217                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3218         },
3219         { },
3220 };
3221
3222 static __net_initdata struct ctl_path ipv4_route_path[] = {
3223         { .procname = "net", },
3224         { .procname = "ipv4", },
3225         { .procname = "route", },
3226         { },
3227 };
3228
3229 static __net_init int sysctl_route_net_init(struct net *net)
3230 {
3231         struct ctl_table *tbl;
3232
3233         tbl = ipv4_route_flush_table;
3234         if (!net_eq(net, &init_net)) {
3235                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3236                 if (tbl == NULL)
3237                         goto err_dup;
3238         }
3239         tbl[0].extra1 = net;
3240
3241         net->ipv4.route_hdr =
3242                 register_net_sysctl_table(net, ipv4_route_path, tbl);
3243         if (net->ipv4.route_hdr == NULL)
3244                 goto err_reg;
3245         return 0;
3246
3247 err_reg:
3248         if (tbl != ipv4_route_flush_table)
3249                 kfree(tbl);
3250 err_dup:
3251         return -ENOMEM;
3252 }
3253
3254 static __net_exit void sysctl_route_net_exit(struct net *net)
3255 {
3256         struct ctl_table *tbl;
3257
3258         tbl = net->ipv4.route_hdr->ctl_table_arg;
3259         unregister_net_sysctl_table(net->ipv4.route_hdr);
3260         BUG_ON(tbl == ipv4_route_flush_table);
3261         kfree(tbl);
3262 }
3263
3264 static __net_initdata struct pernet_operations sysctl_route_ops = {
3265         .init = sysctl_route_net_init,
3266         .exit = sysctl_route_net_exit,
3267 };
3268 #endif
3269
3270 static __net_init int rt_genid_init(struct net *net)
3271 {
3272         get_random_bytes(&net->ipv4.rt_genid,
3273                          sizeof(net->ipv4.rt_genid));
3274         return 0;
3275 }
3276
3277 static __net_initdata struct pernet_operations rt_genid_ops = {
3278         .init = rt_genid_init,
3279 };
3280
3281
3282 #ifdef CONFIG_NET_CLS_ROUTE
3283 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3284 #endif /* CONFIG_NET_CLS_ROUTE */
3285
3286 static __initdata unsigned long rhash_entries;
3287 static int __init set_rhash_entries(char *str)
3288 {
3289         if (!str)
3290                 return 0;
3291         rhash_entries = simple_strtoul(str, &str, 0);
3292         return 1;
3293 }
3294 __setup("rhash_entries=", set_rhash_entries);
3295
3296 int __init ip_rt_init(void)
3297 {
3298         int rc = 0;
3299
3300 #ifdef CONFIG_NET_CLS_ROUTE
3301         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3302         if (!ip_rt_acct)
3303                 panic("IP: failed to allocate ip_rt_acct\n");
3304 #endif
3305
3306         ipv4_dst_ops.kmem_cachep =
3307                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3308                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3309
3310         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3311
3312         if (dst_entries_init(&ipv4_dst_ops) < 0)
3313                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3314
3315         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3316                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3317
3318         rt_hash_table = (struct rt_hash_bucket *)
3319                 alloc_large_system_hash("IP route cache",
3320                                         sizeof(struct rt_hash_bucket),
3321                                         rhash_entries,
3322                                         (totalram_pages >= 128 * 1024) ?
3323                                         15 : 17,
3324                                         0,
3325                                         &rt_hash_log,
3326                                         &rt_hash_mask,
3327                                         rhash_entries ? 0 : 512 * 1024);
3328         memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3329         rt_hash_lock_init();
3330
3331         ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3332         ip_rt_max_size = (rt_hash_mask + 1) * 16;
3333
3334         devinet_init();
3335         ip_fib_init();
3336
3337         /* All the timers, started at system startup tend
3338            to synchronize. Perturb it a bit.
3339          */
3340         INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3341         expires_ljiffies = jiffies;
3342         schedule_delayed_work(&expires_work,
3343                 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3344
3345         if (ip_rt_proc_init())
3346                 printk(KERN_ERR "Unable to create route proc files\n");
3347 #ifdef CONFIG_XFRM
3348         xfrm_init();
3349         xfrm4_init(ip_rt_max_size);
3350 #endif
3351         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3352
3353 #ifdef CONFIG_SYSCTL
3354         register_pernet_subsys(&sysctl_route_ops);
3355 #endif
3356         register_pernet_subsys(&rt_genid_ops);
3357         return rc;
3358 }
3359
3360 #ifdef CONFIG_SYSCTL
3361 /*
3362  * We really need to sanitize the damn ipv4 init order, then all
3363  * this nonsense will go away.
3364  */
3365 void __init ip_static_sysctl_init(void)
3366 {
3367         register_sysctl_paths(ipv4_path, ipv4_skeleton);
3368 }
3369 #endif