ipv4: fix multicast losses
[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(oldflp4) \
113     ((u32)(oldflp4->flowi4_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 /*
135  *      Interface to generic destination cache.
136  */
137
138 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
139 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
140 static unsigned int      ipv4_default_mtu(const struct dst_entry *dst);
141 static void              ipv4_dst_destroy(struct dst_entry *dst);
142 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
143 static void              ipv4_link_failure(struct sk_buff *skb);
144 static void              ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
145 static int rt_garbage_collect(struct dst_ops *ops);
146
147 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
148                             int how)
149 {
150 }
151
152 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
153 {
154         struct rtable *rt = (struct rtable *) dst;
155         struct inet_peer *peer;
156         u32 *p = NULL;
157
158         if (!rt->peer)
159                 rt_bind_peer(rt, rt->rt_dst, 1);
160
161         peer = rt->peer;
162         if (peer) {
163                 u32 *old_p = __DST_METRICS_PTR(old);
164                 unsigned long prev, new;
165
166                 p = peer->metrics;
167                 if (inet_metrics_new(peer))
168                         memcpy(p, old_p, sizeof(u32) * RTAX_MAX);
169
170                 new = (unsigned long) p;
171                 prev = cmpxchg(&dst->_metrics, old, new);
172
173                 if (prev != old) {
174                         p = __DST_METRICS_PTR(prev);
175                         if (prev & DST_METRICS_READ_ONLY)
176                                 p = NULL;
177                 } else {
178                         if (rt->fi) {
179                                 fib_info_put(rt->fi);
180                                 rt->fi = NULL;
181                         }
182                 }
183         }
184         return p;
185 }
186
187 static struct dst_ops ipv4_dst_ops = {
188         .family =               AF_INET,
189         .protocol =             cpu_to_be16(ETH_P_IP),
190         .gc =                   rt_garbage_collect,
191         .check =                ipv4_dst_check,
192         .default_advmss =       ipv4_default_advmss,
193         .default_mtu =          ipv4_default_mtu,
194         .cow_metrics =          ipv4_cow_metrics,
195         .destroy =              ipv4_dst_destroy,
196         .ifdown =               ipv4_dst_ifdown,
197         .negative_advice =      ipv4_negative_advice,
198         .link_failure =         ipv4_link_failure,
199         .update_pmtu =          ip_rt_update_pmtu,
200         .local_out =            __ip_local_out,
201 };
202
203 #define ECN_OR_COST(class)      TC_PRIO_##class
204
205 const __u8 ip_tos2prio[16] = {
206         TC_PRIO_BESTEFFORT,
207         ECN_OR_COST(BESTEFFORT),
208         TC_PRIO_BESTEFFORT,
209         ECN_OR_COST(BESTEFFORT),
210         TC_PRIO_BULK,
211         ECN_OR_COST(BULK),
212         TC_PRIO_BULK,
213         ECN_OR_COST(BULK),
214         TC_PRIO_INTERACTIVE,
215         ECN_OR_COST(INTERACTIVE),
216         TC_PRIO_INTERACTIVE,
217         ECN_OR_COST(INTERACTIVE),
218         TC_PRIO_INTERACTIVE_BULK,
219         ECN_OR_COST(INTERACTIVE_BULK),
220         TC_PRIO_INTERACTIVE_BULK,
221         ECN_OR_COST(INTERACTIVE_BULK)
222 };
223
224
225 /*
226  * Route cache.
227  */
228
229 /* The locking scheme is rather straight forward:
230  *
231  * 1) Read-Copy Update protects the buckets of the central route hash.
232  * 2) Only writers remove entries, and they hold the lock
233  *    as they look at rtable reference counts.
234  * 3) Only readers acquire references to rtable entries,
235  *    they do so with atomic increments and with the
236  *    lock held.
237  */
238
239 struct rt_hash_bucket {
240         struct rtable __rcu     *chain;
241 };
242
243 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
244         defined(CONFIG_PROVE_LOCKING)
245 /*
246  * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
247  * The size of this table is a power of two and depends on the number of CPUS.
248  * (on lockdep we have a quite big spinlock_t, so keep the size down there)
249  */
250 #ifdef CONFIG_LOCKDEP
251 # define RT_HASH_LOCK_SZ        256
252 #else
253 # if NR_CPUS >= 32
254 #  define RT_HASH_LOCK_SZ       4096
255 # elif NR_CPUS >= 16
256 #  define RT_HASH_LOCK_SZ       2048
257 # elif NR_CPUS >= 8
258 #  define RT_HASH_LOCK_SZ       1024
259 # elif NR_CPUS >= 4
260 #  define RT_HASH_LOCK_SZ       512
261 # else
262 #  define RT_HASH_LOCK_SZ       256
263 # endif
264 #endif
265
266 static spinlock_t       *rt_hash_locks;
267 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
268
269 static __init void rt_hash_lock_init(void)
270 {
271         int i;
272
273         rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
274                         GFP_KERNEL);
275         if (!rt_hash_locks)
276                 panic("IP: failed to allocate rt_hash_locks\n");
277
278         for (i = 0; i < RT_HASH_LOCK_SZ; i++)
279                 spin_lock_init(&rt_hash_locks[i]);
280 }
281 #else
282 # define rt_hash_lock_addr(slot) NULL
283
284 static inline void rt_hash_lock_init(void)
285 {
286 }
287 #endif
288
289 static struct rt_hash_bucket    *rt_hash_table __read_mostly;
290 static unsigned                 rt_hash_mask __read_mostly;
291 static unsigned int             rt_hash_log  __read_mostly;
292
293 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
294 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
295
296 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
297                                    int genid)
298 {
299         return jhash_3words((__force u32)daddr, (__force u32)saddr,
300                             idx, genid)
301                 & rt_hash_mask;
302 }
303
304 static inline int rt_genid(struct net *net)
305 {
306         return atomic_read(&net->ipv4.rt_genid);
307 }
308
309 #ifdef CONFIG_PROC_FS
310 struct rt_cache_iter_state {
311         struct seq_net_private p;
312         int bucket;
313         int genid;
314 };
315
316 static struct rtable *rt_cache_get_first(struct seq_file *seq)
317 {
318         struct rt_cache_iter_state *st = seq->private;
319         struct rtable *r = NULL;
320
321         for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
322                 if (!rcu_dereference_raw(rt_hash_table[st->bucket].chain))
323                         continue;
324                 rcu_read_lock_bh();
325                 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
326                 while (r) {
327                         if (dev_net(r->dst.dev) == seq_file_net(seq) &&
328                             r->rt_genid == st->genid)
329                                 return r;
330                         r = rcu_dereference_bh(r->dst.rt_next);
331                 }
332                 rcu_read_unlock_bh();
333         }
334         return r;
335 }
336
337 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
338                                           struct rtable *r)
339 {
340         struct rt_cache_iter_state *st = seq->private;
341
342         r = rcu_dereference_bh(r->dst.rt_next);
343         while (!r) {
344                 rcu_read_unlock_bh();
345                 do {
346                         if (--st->bucket < 0)
347                                 return NULL;
348                 } while (!rcu_dereference_raw(rt_hash_table[st->bucket].chain));
349                 rcu_read_lock_bh();
350                 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
351         }
352         return r;
353 }
354
355 static struct rtable *rt_cache_get_next(struct seq_file *seq,
356                                         struct rtable *r)
357 {
358         struct rt_cache_iter_state *st = seq->private;
359         while ((r = __rt_cache_get_next(seq, r)) != NULL) {
360                 if (dev_net(r->dst.dev) != seq_file_net(seq))
361                         continue;
362                 if (r->rt_genid == st->genid)
363                         break;
364         }
365         return r;
366 }
367
368 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
369 {
370         struct rtable *r = rt_cache_get_first(seq);
371
372         if (r)
373                 while (pos && (r = rt_cache_get_next(seq, r)))
374                         --pos;
375         return pos ? NULL : r;
376 }
377
378 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
379 {
380         struct rt_cache_iter_state *st = seq->private;
381         if (*pos)
382                 return rt_cache_get_idx(seq, *pos - 1);
383         st->genid = rt_genid(seq_file_net(seq));
384         return SEQ_START_TOKEN;
385 }
386
387 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
388 {
389         struct rtable *r;
390
391         if (v == SEQ_START_TOKEN)
392                 r = rt_cache_get_first(seq);
393         else
394                 r = rt_cache_get_next(seq, v);
395         ++*pos;
396         return r;
397 }
398
399 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
400 {
401         if (v && v != SEQ_START_TOKEN)
402                 rcu_read_unlock_bh();
403 }
404
405 static int rt_cache_seq_show(struct seq_file *seq, void *v)
406 {
407         if (v == SEQ_START_TOKEN)
408                 seq_printf(seq, "%-127s\n",
409                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
410                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
411                            "HHUptod\tSpecDst");
412         else {
413                 struct rtable *r = v;
414                 int len;
415
416                 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
417                               "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
418                         r->dst.dev ? r->dst.dev->name : "*",
419                         (__force u32)r->rt_dst,
420                         (__force u32)r->rt_gateway,
421                         r->rt_flags, atomic_read(&r->dst.__refcnt),
422                         r->dst.__use, 0, (__force u32)r->rt_src,
423                         dst_metric_advmss(&r->dst) + 40,
424                         dst_metric(&r->dst, RTAX_WINDOW),
425                         (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
426                               dst_metric(&r->dst, RTAX_RTTVAR)),
427                         r->rt_key_tos,
428                         r->dst.hh ? atomic_read(&r->dst.hh->hh_refcnt) : -1,
429                         r->dst.hh ? (r->dst.hh->hh_output ==
430                                        dev_queue_xmit) : 0,
431                         r->rt_spec_dst, &len);
432
433                 seq_printf(seq, "%*s\n", 127 - len, "");
434         }
435         return 0;
436 }
437
438 static const struct seq_operations rt_cache_seq_ops = {
439         .start  = rt_cache_seq_start,
440         .next   = rt_cache_seq_next,
441         .stop   = rt_cache_seq_stop,
442         .show   = rt_cache_seq_show,
443 };
444
445 static int rt_cache_seq_open(struct inode *inode, struct file *file)
446 {
447         return seq_open_net(inode, file, &rt_cache_seq_ops,
448                         sizeof(struct rt_cache_iter_state));
449 }
450
451 static const struct file_operations rt_cache_seq_fops = {
452         .owner   = THIS_MODULE,
453         .open    = rt_cache_seq_open,
454         .read    = seq_read,
455         .llseek  = seq_lseek,
456         .release = seq_release_net,
457 };
458
459
460 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
461 {
462         int cpu;
463
464         if (*pos == 0)
465                 return SEQ_START_TOKEN;
466
467         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
468                 if (!cpu_possible(cpu))
469                         continue;
470                 *pos = cpu+1;
471                 return &per_cpu(rt_cache_stat, cpu);
472         }
473         return NULL;
474 }
475
476 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
477 {
478         int cpu;
479
480         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
481                 if (!cpu_possible(cpu))
482                         continue;
483                 *pos = cpu+1;
484                 return &per_cpu(rt_cache_stat, cpu);
485         }
486         return NULL;
487
488 }
489
490 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
491 {
492
493 }
494
495 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
496 {
497         struct rt_cache_stat *st = v;
498
499         if (v == SEQ_START_TOKEN) {
500                 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");
501                 return 0;
502         }
503
504         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
505                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
506                    dst_entries_get_slow(&ipv4_dst_ops),
507                    st->in_hit,
508                    st->in_slow_tot,
509                    st->in_slow_mc,
510                    st->in_no_route,
511                    st->in_brd,
512                    st->in_martian_dst,
513                    st->in_martian_src,
514
515                    st->out_hit,
516                    st->out_slow_tot,
517                    st->out_slow_mc,
518
519                    st->gc_total,
520                    st->gc_ignored,
521                    st->gc_goal_miss,
522                    st->gc_dst_overflow,
523                    st->in_hlist_search,
524                    st->out_hlist_search
525                 );
526         return 0;
527 }
528
529 static const struct seq_operations rt_cpu_seq_ops = {
530         .start  = rt_cpu_seq_start,
531         .next   = rt_cpu_seq_next,
532         .stop   = rt_cpu_seq_stop,
533         .show   = rt_cpu_seq_show,
534 };
535
536
537 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
538 {
539         return seq_open(file, &rt_cpu_seq_ops);
540 }
541
542 static const struct file_operations rt_cpu_seq_fops = {
543         .owner   = THIS_MODULE,
544         .open    = rt_cpu_seq_open,
545         .read    = seq_read,
546         .llseek  = seq_lseek,
547         .release = seq_release,
548 };
549
550 #ifdef CONFIG_IP_ROUTE_CLASSID
551 static int rt_acct_proc_show(struct seq_file *m, void *v)
552 {
553         struct ip_rt_acct *dst, *src;
554         unsigned int i, j;
555
556         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
557         if (!dst)
558                 return -ENOMEM;
559
560         for_each_possible_cpu(i) {
561                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
562                 for (j = 0; j < 256; j++) {
563                         dst[j].o_bytes   += src[j].o_bytes;
564                         dst[j].o_packets += src[j].o_packets;
565                         dst[j].i_bytes   += src[j].i_bytes;
566                         dst[j].i_packets += src[j].i_packets;
567                 }
568         }
569
570         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
571         kfree(dst);
572         return 0;
573 }
574
575 static int rt_acct_proc_open(struct inode *inode, struct file *file)
576 {
577         return single_open(file, rt_acct_proc_show, NULL);
578 }
579
580 static const struct file_operations rt_acct_proc_fops = {
581         .owner          = THIS_MODULE,
582         .open           = rt_acct_proc_open,
583         .read           = seq_read,
584         .llseek         = seq_lseek,
585         .release        = single_release,
586 };
587 #endif
588
589 static int __net_init ip_rt_do_proc_init(struct net *net)
590 {
591         struct proc_dir_entry *pde;
592
593         pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
594                         &rt_cache_seq_fops);
595         if (!pde)
596                 goto err1;
597
598         pde = proc_create("rt_cache", S_IRUGO,
599                           net->proc_net_stat, &rt_cpu_seq_fops);
600         if (!pde)
601                 goto err2;
602
603 #ifdef CONFIG_IP_ROUTE_CLASSID
604         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
605         if (!pde)
606                 goto err3;
607 #endif
608         return 0;
609
610 #ifdef CONFIG_IP_ROUTE_CLASSID
611 err3:
612         remove_proc_entry("rt_cache", net->proc_net_stat);
613 #endif
614 err2:
615         remove_proc_entry("rt_cache", net->proc_net);
616 err1:
617         return -ENOMEM;
618 }
619
620 static void __net_exit ip_rt_do_proc_exit(struct net *net)
621 {
622         remove_proc_entry("rt_cache", net->proc_net_stat);
623         remove_proc_entry("rt_cache", net->proc_net);
624 #ifdef CONFIG_IP_ROUTE_CLASSID
625         remove_proc_entry("rt_acct", net->proc_net);
626 #endif
627 }
628
629 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
630         .init = ip_rt_do_proc_init,
631         .exit = ip_rt_do_proc_exit,
632 };
633
634 static int __init ip_rt_proc_init(void)
635 {
636         return register_pernet_subsys(&ip_rt_proc_ops);
637 }
638
639 #else
640 static inline int ip_rt_proc_init(void)
641 {
642         return 0;
643 }
644 #endif /* CONFIG_PROC_FS */
645
646 static inline void rt_free(struct rtable *rt)
647 {
648         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
649 }
650
651 static inline void rt_drop(struct rtable *rt)
652 {
653         ip_rt_put(rt);
654         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
655 }
656
657 static inline int rt_fast_clean(struct rtable *rth)
658 {
659         /* Kill broadcast/multicast entries very aggresively, if they
660            collide in hash table with more useful entries */
661         return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
662                 rt_is_input_route(rth) && rth->dst.rt_next;
663 }
664
665 static inline int rt_valuable(struct rtable *rth)
666 {
667         return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
668                 (rth->peer && rth->peer->pmtu_expires);
669 }
670
671 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
672 {
673         unsigned long age;
674         int ret = 0;
675
676         if (atomic_read(&rth->dst.__refcnt))
677                 goto out;
678
679         age = jiffies - rth->dst.lastuse;
680         if ((age <= tmo1 && !rt_fast_clean(rth)) ||
681             (age <= tmo2 && rt_valuable(rth)))
682                 goto out;
683         ret = 1;
684 out:    return ret;
685 }
686
687 /* Bits of score are:
688  * 31: very valuable
689  * 30: not quite useless
690  * 29..0: usage counter
691  */
692 static inline u32 rt_score(struct rtable *rt)
693 {
694         u32 score = jiffies - rt->dst.lastuse;
695
696         score = ~score & ~(3<<30);
697
698         if (rt_valuable(rt))
699                 score |= (1<<31);
700
701         if (rt_is_output_route(rt) ||
702             !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
703                 score |= (1<<30);
704
705         return score;
706 }
707
708 static inline bool rt_caching(const struct net *net)
709 {
710         return net->ipv4.current_rt_cache_rebuild_count <=
711                 net->ipv4.sysctl_rt_cache_rebuild_count;
712 }
713
714 static inline bool compare_hash_inputs(const struct rtable *rt1,
715                                        const struct rtable *rt2)
716 {
717         return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
718                 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
719                 (rt1->rt_iif ^ rt2->rt_iif)) == 0);
720 }
721
722 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2)
723 {
724         return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) |
725                 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) |
726                 (rt1->rt_mark ^ rt2->rt_mark) |
727                 (rt1->rt_key_tos ^ rt2->rt_key_tos) |
728                 (rt1->rt_oif ^ rt2->rt_oif) |
729                 (rt1->rt_iif ^ rt2->rt_iif)) == 0;
730 }
731
732 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
733 {
734         return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
735 }
736
737 static inline int rt_is_expired(struct rtable *rth)
738 {
739         return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
740 }
741
742 /*
743  * Perform a full scan of hash table and free all entries.
744  * Can be called by a softirq or a process.
745  * In the later case, we want to be reschedule if necessary
746  */
747 static void rt_do_flush(struct net *net, int process_context)
748 {
749         unsigned int i;
750         struct rtable *rth, *next;
751
752         for (i = 0; i <= rt_hash_mask; i++) {
753                 struct rtable __rcu **pprev;
754                 struct rtable *list;
755
756                 if (process_context && need_resched())
757                         cond_resched();
758                 rth = rcu_dereference_raw(rt_hash_table[i].chain);
759                 if (!rth)
760                         continue;
761
762                 spin_lock_bh(rt_hash_lock_addr(i));
763
764                 list = NULL;
765                 pprev = &rt_hash_table[i].chain;
766                 rth = rcu_dereference_protected(*pprev,
767                         lockdep_is_held(rt_hash_lock_addr(i)));
768
769                 while (rth) {
770                         next = rcu_dereference_protected(rth->dst.rt_next,
771                                 lockdep_is_held(rt_hash_lock_addr(i)));
772
773                         if (!net ||
774                             net_eq(dev_net(rth->dst.dev), net)) {
775                                 rcu_assign_pointer(*pprev, next);
776                                 rcu_assign_pointer(rth->dst.rt_next, list);
777                                 list = rth;
778                         } else {
779                                 pprev = &rth->dst.rt_next;
780                         }
781                         rth = next;
782                 }
783
784                 spin_unlock_bh(rt_hash_lock_addr(i));
785
786                 for (; list; list = next) {
787                         next = rcu_dereference_protected(list->dst.rt_next, 1);
788                         rt_free(list);
789                 }
790         }
791 }
792
793 /*
794  * While freeing expired entries, we compute average chain length
795  * and standard deviation, using fixed-point arithmetic.
796  * This to have an estimation of rt_chain_length_max
797  *  rt_chain_length_max = max(elasticity, AVG + 4*SD)
798  * We use 3 bits for frational part, and 29 (or 61) for magnitude.
799  */
800
801 #define FRACT_BITS 3
802 #define ONE (1UL << FRACT_BITS)
803
804 /*
805  * Given a hash chain and an item in this hash chain,
806  * find if a previous entry has the same hash_inputs
807  * (but differs on tos, mark or oif)
808  * Returns 0 if an alias is found.
809  * Returns ONE if rth has no alias before itself.
810  */
811 static int has_noalias(const struct rtable *head, const struct rtable *rth)
812 {
813         const struct rtable *aux = head;
814
815         while (aux != rth) {
816                 if (compare_hash_inputs(aux, rth))
817                         return 0;
818                 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
819         }
820         return ONE;
821 }
822
823 /*
824  * Perturbation of rt_genid by a small quantity [1..256]
825  * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
826  * many times (2^24) without giving recent rt_genid.
827  * Jenkins hash is strong enough that litle changes of rt_genid are OK.
828  */
829 static void rt_cache_invalidate(struct net *net)
830 {
831         unsigned char shuffle;
832
833         get_random_bytes(&shuffle, sizeof(shuffle));
834         atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
835 }
836
837 /*
838  * delay < 0  : invalidate cache (fast : entries will be deleted later)
839  * delay >= 0 : invalidate & flush cache (can be long)
840  */
841 void rt_cache_flush(struct net *net, int delay)
842 {
843         rt_cache_invalidate(net);
844         if (delay >= 0)
845                 rt_do_flush(net, !in_softirq());
846 }
847
848 /* Flush previous cache invalidated entries from the cache */
849 void rt_cache_flush_batch(struct net *net)
850 {
851         rt_do_flush(net, !in_softirq());
852 }
853
854 static void rt_emergency_hash_rebuild(struct net *net)
855 {
856         if (net_ratelimit())
857                 printk(KERN_WARNING "Route hash chain too long!\n");
858         rt_cache_invalidate(net);
859 }
860
861 /*
862    Short description of GC goals.
863
864    We want to build algorithm, which will keep routing cache
865    at some equilibrium point, when number of aged off entries
866    is kept approximately equal to newly generated ones.
867
868    Current expiration strength is variable "expire".
869    We try to adjust it dynamically, so that if networking
870    is idle expires is large enough to keep enough of warm entries,
871    and when load increases it reduces to limit cache size.
872  */
873
874 static int rt_garbage_collect(struct dst_ops *ops)
875 {
876         static unsigned long expire = RT_GC_TIMEOUT;
877         static unsigned long last_gc;
878         static int rover;
879         static int equilibrium;
880         struct rtable *rth;
881         struct rtable __rcu **rthp;
882         unsigned long now = jiffies;
883         int goal;
884         int entries = dst_entries_get_fast(&ipv4_dst_ops);
885
886         /*
887          * Garbage collection is pretty expensive,
888          * do not make it too frequently.
889          */
890
891         RT_CACHE_STAT_INC(gc_total);
892
893         if (now - last_gc < ip_rt_gc_min_interval &&
894             entries < ip_rt_max_size) {
895                 RT_CACHE_STAT_INC(gc_ignored);
896                 goto out;
897         }
898
899         entries = dst_entries_get_slow(&ipv4_dst_ops);
900         /* Calculate number of entries, which we want to expire now. */
901         goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
902         if (goal <= 0) {
903                 if (equilibrium < ipv4_dst_ops.gc_thresh)
904                         equilibrium = ipv4_dst_ops.gc_thresh;
905                 goal = entries - equilibrium;
906                 if (goal > 0) {
907                         equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
908                         goal = entries - equilibrium;
909                 }
910         } else {
911                 /* We are in dangerous area. Try to reduce cache really
912                  * aggressively.
913                  */
914                 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
915                 equilibrium = entries - goal;
916         }
917
918         if (now - last_gc >= ip_rt_gc_min_interval)
919                 last_gc = now;
920
921         if (goal <= 0) {
922                 equilibrium += goal;
923                 goto work_done;
924         }
925
926         do {
927                 int i, k;
928
929                 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
930                         unsigned long tmo = expire;
931
932                         k = (k + 1) & rt_hash_mask;
933                         rthp = &rt_hash_table[k].chain;
934                         spin_lock_bh(rt_hash_lock_addr(k));
935                         while ((rth = rcu_dereference_protected(*rthp,
936                                         lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
937                                 if (!rt_is_expired(rth) &&
938                                         !rt_may_expire(rth, tmo, expire)) {
939                                         tmo >>= 1;
940                                         rthp = &rth->dst.rt_next;
941                                         continue;
942                                 }
943                                 *rthp = rth->dst.rt_next;
944                                 rt_free(rth);
945                                 goal--;
946                         }
947                         spin_unlock_bh(rt_hash_lock_addr(k));
948                         if (goal <= 0)
949                                 break;
950                 }
951                 rover = k;
952
953                 if (goal <= 0)
954                         goto work_done;
955
956                 /* Goal is not achieved. We stop process if:
957
958                    - if expire reduced to zero. Otherwise, expire is halfed.
959                    - if table is not full.
960                    - if we are called from interrupt.
961                    - jiffies check is just fallback/debug loop breaker.
962                      We will not spin here for long time in any case.
963                  */
964
965                 RT_CACHE_STAT_INC(gc_goal_miss);
966
967                 if (expire == 0)
968                         break;
969
970                 expire >>= 1;
971
972                 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
973                         goto out;
974         } while (!in_softirq() && time_before_eq(jiffies, now));
975
976         if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
977                 goto out;
978         if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
979                 goto out;
980         if (net_ratelimit())
981                 printk(KERN_WARNING "dst cache overflow\n");
982         RT_CACHE_STAT_INC(gc_dst_overflow);
983         return 1;
984
985 work_done:
986         expire += ip_rt_gc_min_interval;
987         if (expire > ip_rt_gc_timeout ||
988             dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
989             dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
990                 expire = ip_rt_gc_timeout;
991 out:    return 0;
992 }
993
994 /*
995  * Returns number of entries in a hash chain that have different hash_inputs
996  */
997 static int slow_chain_length(const struct rtable *head)
998 {
999         int length = 0;
1000         const struct rtable *rth = head;
1001
1002         while (rth) {
1003                 length += has_noalias(head, rth);
1004                 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1005         }
1006         return length >> FRACT_BITS;
1007 }
1008
1009 static struct rtable *rt_intern_hash(unsigned hash, struct rtable *rt,
1010                                      struct sk_buff *skb, int ifindex)
1011 {
1012         struct rtable   *rth, *cand;
1013         struct rtable __rcu **rthp, **candp;
1014         unsigned long   now;
1015         u32             min_score;
1016         int             chain_length;
1017         int attempts = !in_softirq();
1018
1019 restart:
1020         chain_length = 0;
1021         min_score = ~(u32)0;
1022         cand = NULL;
1023         candp = NULL;
1024         now = jiffies;
1025
1026         if (!rt_caching(dev_net(rt->dst.dev))) {
1027                 /*
1028                  * If we're not caching, just tell the caller we
1029                  * were successful and don't touch the route.  The
1030                  * caller hold the sole reference to the cache entry, and
1031                  * it will be released when the caller is done with it.
1032                  * If we drop it here, the callers have no way to resolve routes
1033                  * when we're not caching.  Instead, just point *rp at rt, so
1034                  * the caller gets a single use out of the route
1035                  * Note that we do rt_free on this new route entry, so that
1036                  * once its refcount hits zero, we are still able to reap it
1037                  * (Thanks Alexey)
1038                  * Note: To avoid expensive rcu stuff for this uncached dst,
1039                  * we set DST_NOCACHE so that dst_release() can free dst without
1040                  * waiting a grace period.
1041                  */
1042
1043                 rt->dst.flags |= DST_NOCACHE;
1044                 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1045                         int err = arp_bind_neighbour(&rt->dst);
1046                         if (err) {
1047                                 if (net_ratelimit())
1048                                         printk(KERN_WARNING
1049                                             "Neighbour table failure & not caching routes.\n");
1050                                 ip_rt_put(rt);
1051                                 return ERR_PTR(err);
1052                         }
1053                 }
1054
1055                 goto skip_hashing;
1056         }
1057
1058         rthp = &rt_hash_table[hash].chain;
1059
1060         spin_lock_bh(rt_hash_lock_addr(hash));
1061         while ((rth = rcu_dereference_protected(*rthp,
1062                         lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1063                 if (rt_is_expired(rth)) {
1064                         *rthp = rth->dst.rt_next;
1065                         rt_free(rth);
1066                         continue;
1067                 }
1068                 if (compare_keys(rth, rt) && compare_netns(rth, rt)) {
1069                         /* Put it first */
1070                         *rthp = rth->dst.rt_next;
1071                         /*
1072                          * Since lookup is lockfree, the deletion
1073                          * must be visible to another weakly ordered CPU before
1074                          * the insertion at the start of the hash chain.
1075                          */
1076                         rcu_assign_pointer(rth->dst.rt_next,
1077                                            rt_hash_table[hash].chain);
1078                         /*
1079                          * Since lookup is lockfree, the update writes
1080                          * must be ordered for consistency on SMP.
1081                          */
1082                         rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1083
1084                         dst_use(&rth->dst, now);
1085                         spin_unlock_bh(rt_hash_lock_addr(hash));
1086
1087                         rt_drop(rt);
1088                         if (skb)
1089                                 skb_dst_set(skb, &rth->dst);
1090                         return rth;
1091                 }
1092
1093                 if (!atomic_read(&rth->dst.__refcnt)) {
1094                         u32 score = rt_score(rth);
1095
1096                         if (score <= min_score) {
1097                                 cand = rth;
1098                                 candp = rthp;
1099                                 min_score = score;
1100                         }
1101                 }
1102
1103                 chain_length++;
1104
1105                 rthp = &rth->dst.rt_next;
1106         }
1107
1108         if (cand) {
1109                 /* ip_rt_gc_elasticity used to be average length of chain
1110                  * length, when exceeded gc becomes really aggressive.
1111                  *
1112                  * The second limit is less certain. At the moment it allows
1113                  * only 2 entries per bucket. We will see.
1114                  */
1115                 if (chain_length > ip_rt_gc_elasticity) {
1116                         *candp = cand->dst.rt_next;
1117                         rt_free(cand);
1118                 }
1119         } else {
1120                 if (chain_length > rt_chain_length_max &&
1121                     slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1122                         struct net *net = dev_net(rt->dst.dev);
1123                         int num = ++net->ipv4.current_rt_cache_rebuild_count;
1124                         if (!rt_caching(net)) {
1125                                 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1126                                         rt->dst.dev->name, num);
1127                         }
1128                         rt_emergency_hash_rebuild(net);
1129                         spin_unlock_bh(rt_hash_lock_addr(hash));
1130
1131                         hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1132                                         ifindex, rt_genid(net));
1133                         goto restart;
1134                 }
1135         }
1136
1137         /* Try to bind route to arp only if it is output
1138            route or unicast forwarding path.
1139          */
1140         if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1141                 int err = arp_bind_neighbour(&rt->dst);
1142                 if (err) {
1143                         spin_unlock_bh(rt_hash_lock_addr(hash));
1144
1145                         if (err != -ENOBUFS) {
1146                                 rt_drop(rt);
1147                                 return ERR_PTR(err);
1148                         }
1149
1150                         /* Neighbour tables are full and nothing
1151                            can be released. Try to shrink route cache,
1152                            it is most likely it holds some neighbour records.
1153                          */
1154                         if (attempts-- > 0) {
1155                                 int saved_elasticity = ip_rt_gc_elasticity;
1156                                 int saved_int = ip_rt_gc_min_interval;
1157                                 ip_rt_gc_elasticity     = 1;
1158                                 ip_rt_gc_min_interval   = 0;
1159                                 rt_garbage_collect(&ipv4_dst_ops);
1160                                 ip_rt_gc_min_interval   = saved_int;
1161                                 ip_rt_gc_elasticity     = saved_elasticity;
1162                                 goto restart;
1163                         }
1164
1165                         if (net_ratelimit())
1166                                 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1167                         rt_drop(rt);
1168                         return ERR_PTR(-ENOBUFS);
1169                 }
1170         }
1171
1172         rt->dst.rt_next = rt_hash_table[hash].chain;
1173
1174         /*
1175          * Since lookup is lockfree, we must make sure
1176          * previous writes to rt are committed to memory
1177          * before making rt visible to other CPUS.
1178          */
1179         rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1180
1181         spin_unlock_bh(rt_hash_lock_addr(hash));
1182
1183 skip_hashing:
1184         if (skb)
1185                 skb_dst_set(skb, &rt->dst);
1186         return rt;
1187 }
1188
1189 static atomic_t __rt_peer_genid = ATOMIC_INIT(0);
1190
1191 static u32 rt_peer_genid(void)
1192 {
1193         return atomic_read(&__rt_peer_genid);
1194 }
1195
1196 void rt_bind_peer(struct rtable *rt, __be32 daddr, int create)
1197 {
1198         struct inet_peer *peer;
1199
1200         peer = inet_getpeer_v4(daddr, create);
1201
1202         if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1203                 inet_putpeer(peer);
1204         else
1205                 rt->rt_peer_genid = rt_peer_genid();
1206 }
1207
1208 /*
1209  * Peer allocation may fail only in serious out-of-memory conditions.  However
1210  * we still can generate some output.
1211  * Random ID selection looks a bit dangerous because we have no chances to
1212  * select ID being unique in a reasonable period of time.
1213  * But broken packet identifier may be better than no packet at all.
1214  */
1215 static void ip_select_fb_ident(struct iphdr *iph)
1216 {
1217         static DEFINE_SPINLOCK(ip_fb_id_lock);
1218         static u32 ip_fallback_id;
1219         u32 salt;
1220
1221         spin_lock_bh(&ip_fb_id_lock);
1222         salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1223         iph->id = htons(salt & 0xFFFF);
1224         ip_fallback_id = salt;
1225         spin_unlock_bh(&ip_fb_id_lock);
1226 }
1227
1228 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1229 {
1230         struct rtable *rt = (struct rtable *) dst;
1231
1232         if (rt) {
1233                 if (rt->peer == NULL)
1234                         rt_bind_peer(rt, rt->rt_dst, 1);
1235
1236                 /* If peer is attached to destination, it is never detached,
1237                    so that we need not to grab a lock to dereference it.
1238                  */
1239                 if (rt->peer) {
1240                         iph->id = htons(inet_getid(rt->peer, more));
1241                         return;
1242                 }
1243         } else
1244                 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1245                        __builtin_return_address(0));
1246
1247         ip_select_fb_ident(iph);
1248 }
1249 EXPORT_SYMBOL(__ip_select_ident);
1250
1251 static void rt_del(unsigned hash, struct rtable *rt)
1252 {
1253         struct rtable __rcu **rthp;
1254         struct rtable *aux;
1255
1256         rthp = &rt_hash_table[hash].chain;
1257         spin_lock_bh(rt_hash_lock_addr(hash));
1258         ip_rt_put(rt);
1259         while ((aux = rcu_dereference_protected(*rthp,
1260                         lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1261                 if (aux == rt || rt_is_expired(aux)) {
1262                         *rthp = aux->dst.rt_next;
1263                         rt_free(aux);
1264                         continue;
1265                 }
1266                 rthp = &aux->dst.rt_next;
1267         }
1268         spin_unlock_bh(rt_hash_lock_addr(hash));
1269 }
1270
1271 /* called in rcu_read_lock() section */
1272 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1273                     __be32 saddr, struct net_device *dev)
1274 {
1275         struct in_device *in_dev = __in_dev_get_rcu(dev);
1276         struct inet_peer *peer;
1277         struct net *net;
1278
1279         if (!in_dev)
1280                 return;
1281
1282         net = dev_net(dev);
1283         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1284             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1285             ipv4_is_zeronet(new_gw))
1286                 goto reject_redirect;
1287
1288         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1289                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1290                         goto reject_redirect;
1291                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1292                         goto reject_redirect;
1293         } else {
1294                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1295                         goto reject_redirect;
1296         }
1297
1298         peer = inet_getpeer_v4(daddr, 1);
1299         if (peer) {
1300                 peer->redirect_learned.a4 = new_gw;
1301
1302                 inet_putpeer(peer);
1303
1304                 atomic_inc(&__rt_peer_genid);
1305         }
1306         return;
1307
1308 reject_redirect:
1309 #ifdef CONFIG_IP_ROUTE_VERBOSE
1310         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1311                 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1312                         "  Advised path = %pI4 -> %pI4\n",
1313                        &old_gw, dev->name, &new_gw,
1314                        &saddr, &daddr);
1315 #endif
1316         ;
1317 }
1318
1319 static bool peer_pmtu_expired(struct inet_peer *peer)
1320 {
1321         unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1322
1323         return orig &&
1324                time_after_eq(jiffies, orig) &&
1325                cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1326 }
1327
1328 static bool peer_pmtu_cleaned(struct inet_peer *peer)
1329 {
1330         unsigned long orig = ACCESS_ONCE(peer->pmtu_expires);
1331
1332         return orig &&
1333                cmpxchg(&peer->pmtu_expires, orig, 0) == orig;
1334 }
1335
1336 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1337 {
1338         struct rtable *rt = (struct rtable *)dst;
1339         struct dst_entry *ret = dst;
1340
1341         if (rt) {
1342                 if (dst->obsolete > 0) {
1343                         ip_rt_put(rt);
1344                         ret = NULL;
1345                 } else if (rt->rt_flags & RTCF_REDIRECTED) {
1346                         unsigned hash = rt_hash(rt->rt_key_dst, rt->rt_key_src,
1347                                                 rt->rt_oif,
1348                                                 rt_genid(dev_net(dst->dev)));
1349                         rt_del(hash, rt);
1350                         ret = NULL;
1351                 } else if (rt->peer && peer_pmtu_expired(rt->peer)) {
1352                         dst_metric_set(dst, RTAX_MTU, rt->peer->pmtu_orig);
1353                 }
1354         }
1355         return ret;
1356 }
1357
1358 /*
1359  * Algorithm:
1360  *      1. The first ip_rt_redirect_number redirects are sent
1361  *         with exponential backoff, then we stop sending them at all,
1362  *         assuming that the host ignores our redirects.
1363  *      2. If we did not see packets requiring redirects
1364  *         during ip_rt_redirect_silence, we assume that the host
1365  *         forgot redirected route and start to send redirects again.
1366  *
1367  * This algorithm is much cheaper and more intelligent than dumb load limiting
1368  * in icmp.c.
1369  *
1370  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1371  * and "frag. need" (breaks PMTU discovery) in icmp.c.
1372  */
1373
1374 void ip_rt_send_redirect(struct sk_buff *skb)
1375 {
1376         struct rtable *rt = skb_rtable(skb);
1377         struct in_device *in_dev;
1378         struct inet_peer *peer;
1379         int log_martians;
1380
1381         rcu_read_lock();
1382         in_dev = __in_dev_get_rcu(rt->dst.dev);
1383         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1384                 rcu_read_unlock();
1385                 return;
1386         }
1387         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1388         rcu_read_unlock();
1389
1390         if (!rt->peer)
1391                 rt_bind_peer(rt, rt->rt_dst, 1);
1392         peer = rt->peer;
1393         if (!peer) {
1394                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1395                 return;
1396         }
1397
1398         /* No redirected packets during ip_rt_redirect_silence;
1399          * reset the algorithm.
1400          */
1401         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
1402                 peer->rate_tokens = 0;
1403
1404         /* Too many ignored redirects; do not send anything
1405          * set dst.rate_last to the last seen redirected packet.
1406          */
1407         if (peer->rate_tokens >= ip_rt_redirect_number) {
1408                 peer->rate_last = jiffies;
1409                 return;
1410         }
1411
1412         /* Check for load limit; set rate_last to the latest sent
1413          * redirect.
1414          */
1415         if (peer->rate_tokens == 0 ||
1416             time_after(jiffies,
1417                        (peer->rate_last +
1418                         (ip_rt_redirect_load << peer->rate_tokens)))) {
1419                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1420                 peer->rate_last = jiffies;
1421                 ++peer->rate_tokens;
1422 #ifdef CONFIG_IP_ROUTE_VERBOSE
1423                 if (log_martians &&
1424                     peer->rate_tokens == ip_rt_redirect_number &&
1425                     net_ratelimit())
1426                         printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1427                                &ip_hdr(skb)->saddr, rt->rt_iif,
1428                                 &rt->rt_dst, &rt->rt_gateway);
1429 #endif
1430         }
1431 }
1432
1433 static int ip_error(struct sk_buff *skb)
1434 {
1435         struct rtable *rt = skb_rtable(skb);
1436         struct inet_peer *peer;
1437         unsigned long now;
1438         bool send;
1439         int code;
1440
1441         switch (rt->dst.error) {
1442                 case EINVAL:
1443                 default:
1444                         goto out;
1445                 case EHOSTUNREACH:
1446                         code = ICMP_HOST_UNREACH;
1447                         break;
1448                 case ENETUNREACH:
1449                         code = ICMP_NET_UNREACH;
1450                         IP_INC_STATS_BH(dev_net(rt->dst.dev),
1451                                         IPSTATS_MIB_INNOROUTES);
1452                         break;
1453                 case EACCES:
1454                         code = ICMP_PKT_FILTERED;
1455                         break;
1456         }
1457
1458         if (!rt->peer)
1459                 rt_bind_peer(rt, rt->rt_dst, 1);
1460         peer = rt->peer;
1461
1462         send = true;
1463         if (peer) {
1464                 now = jiffies;
1465                 peer->rate_tokens += now - peer->rate_last;
1466                 if (peer->rate_tokens > ip_rt_error_burst)
1467                         peer->rate_tokens = ip_rt_error_burst;
1468                 peer->rate_last = now;
1469                 if (peer->rate_tokens >= ip_rt_error_cost)
1470                         peer->rate_tokens -= ip_rt_error_cost;
1471                 else
1472                         send = false;
1473         }
1474         if (send)
1475                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1476
1477 out:    kfree_skb(skb);
1478         return 0;
1479 }
1480
1481 /*
1482  *      The last two values are not from the RFC but
1483  *      are needed for AMPRnet AX.25 paths.
1484  */
1485
1486 static const unsigned short mtu_plateau[] =
1487 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1488
1489 static inline unsigned short guess_mtu(unsigned short old_mtu)
1490 {
1491         int i;
1492
1493         for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1494                 if (old_mtu > mtu_plateau[i])
1495                         return mtu_plateau[i];
1496         return 68;
1497 }
1498
1499 unsigned short ip_rt_frag_needed(struct net *net, const struct iphdr *iph,
1500                                  unsigned short new_mtu,
1501                                  struct net_device *dev)
1502 {
1503         unsigned short old_mtu = ntohs(iph->tot_len);
1504         unsigned short est_mtu = 0;
1505         struct inet_peer *peer;
1506
1507         peer = inet_getpeer_v4(iph->daddr, 1);
1508         if (peer) {
1509                 unsigned short mtu = new_mtu;
1510
1511                 if (new_mtu < 68 || new_mtu >= old_mtu) {
1512                         /* BSD 4.2 derived systems incorrectly adjust
1513                          * tot_len by the IP header length, and report
1514                          * a zero MTU in the ICMP message.
1515                          */
1516                         if (mtu == 0 &&
1517                             old_mtu >= 68 + (iph->ihl << 2))
1518                                 old_mtu -= iph->ihl << 2;
1519                         mtu = guess_mtu(old_mtu);
1520                 }
1521
1522                 if (mtu < ip_rt_min_pmtu)
1523                         mtu = ip_rt_min_pmtu;
1524                 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) {
1525                         unsigned long pmtu_expires;
1526
1527                         pmtu_expires = jiffies + ip_rt_mtu_expires;
1528                         if (!pmtu_expires)
1529                                 pmtu_expires = 1UL;
1530
1531                         est_mtu = mtu;
1532                         peer->pmtu_learned = mtu;
1533                         peer->pmtu_expires = pmtu_expires;
1534                 }
1535
1536                 inet_putpeer(peer);
1537
1538                 atomic_inc(&__rt_peer_genid);
1539         }
1540         return est_mtu ? : new_mtu;
1541 }
1542
1543 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer)
1544 {
1545         unsigned long expires = ACCESS_ONCE(peer->pmtu_expires);
1546
1547         if (!expires)
1548                 return;
1549         if (time_before(jiffies, expires)) {
1550                 u32 orig_dst_mtu = dst_mtu(dst);
1551                 if (peer->pmtu_learned < orig_dst_mtu) {
1552                         if (!peer->pmtu_orig)
1553                                 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU);
1554                         dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned);
1555                 }
1556         } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires)
1557                 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig);
1558 }
1559
1560 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1561 {
1562         struct rtable *rt = (struct rtable *) dst;
1563         struct inet_peer *peer;
1564
1565         dst_confirm(dst);
1566
1567         if (!rt->peer)
1568                 rt_bind_peer(rt, rt->rt_dst, 1);
1569         peer = rt->peer;
1570         if (peer) {
1571                 unsigned long pmtu_expires = ACCESS_ONCE(peer->pmtu_expires);
1572
1573                 if (mtu < ip_rt_min_pmtu)
1574                         mtu = ip_rt_min_pmtu;
1575                 if (!pmtu_expires || mtu < peer->pmtu_learned) {
1576
1577                         pmtu_expires = jiffies + ip_rt_mtu_expires;
1578                         if (!pmtu_expires)
1579                                 pmtu_expires = 1UL;
1580
1581                         peer->pmtu_learned = mtu;
1582                         peer->pmtu_expires = pmtu_expires;
1583
1584                         atomic_inc(&__rt_peer_genid);
1585                         rt->rt_peer_genid = rt_peer_genid();
1586                 }
1587                 check_peer_pmtu(dst, peer);
1588         }
1589 }
1590
1591 static int check_peer_redir(struct dst_entry *dst, struct inet_peer *peer)
1592 {
1593         struct rtable *rt = (struct rtable *) dst;
1594         __be32 orig_gw = rt->rt_gateway;
1595
1596         dst_confirm(&rt->dst);
1597
1598         neigh_release(rt->dst.neighbour);
1599         rt->dst.neighbour = NULL;
1600
1601         rt->rt_gateway = peer->redirect_learned.a4;
1602         if (arp_bind_neighbour(&rt->dst) ||
1603             !(rt->dst.neighbour->nud_state & NUD_VALID)) {
1604                 if (rt->dst.neighbour)
1605                         neigh_event_send(rt->dst.neighbour, NULL);
1606                 rt->rt_gateway = orig_gw;
1607                 return -EAGAIN;
1608         } else {
1609                 rt->rt_flags |= RTCF_REDIRECTED;
1610                 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE,
1611                                         rt->dst.neighbour);
1612         }
1613         return 0;
1614 }
1615
1616 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1617 {
1618         struct rtable *rt = (struct rtable *) dst;
1619
1620         if (rt_is_expired(rt))
1621                 return NULL;
1622         if (rt->rt_peer_genid != rt_peer_genid()) {
1623                 struct inet_peer *peer;
1624
1625                 if (!rt->peer)
1626                         rt_bind_peer(rt, rt->rt_dst, 0);
1627
1628                 peer = rt->peer;
1629                 if (peer) {
1630                         check_peer_pmtu(dst, peer);
1631
1632                         if (peer->redirect_learned.a4 &&
1633                             peer->redirect_learned.a4 != rt->rt_gateway) {
1634                                 if (check_peer_redir(dst, peer))
1635                                         return NULL;
1636                         }
1637                 }
1638
1639                 rt->rt_peer_genid = rt_peer_genid();
1640         }
1641         return dst;
1642 }
1643
1644 static void ipv4_dst_destroy(struct dst_entry *dst)
1645 {
1646         struct rtable *rt = (struct rtable *) dst;
1647         struct inet_peer *peer = rt->peer;
1648
1649         if (rt->fi) {
1650                 fib_info_put(rt->fi);
1651                 rt->fi = NULL;
1652         }
1653         if (peer) {
1654                 rt->peer = NULL;
1655                 inet_putpeer(peer);
1656         }
1657 }
1658
1659
1660 static void ipv4_link_failure(struct sk_buff *skb)
1661 {
1662         struct rtable *rt;
1663
1664         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1665
1666         rt = skb_rtable(skb);
1667         if (rt && rt->peer && peer_pmtu_cleaned(rt->peer))
1668                 dst_metric_set(&rt->dst, RTAX_MTU, rt->peer->pmtu_orig);
1669 }
1670
1671 static int ip_rt_bug(struct sk_buff *skb)
1672 {
1673         printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1674                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1675                 skb->dev ? skb->dev->name : "?");
1676         kfree_skb(skb);
1677         WARN_ON(1);
1678         return 0;
1679 }
1680
1681 /*
1682    We do not cache source address of outgoing interface,
1683    because it is used only by IP RR, TS and SRR options,
1684    so that it out of fast path.
1685
1686    BTW remember: "addr" is allowed to be not aligned
1687    in IP options!
1688  */
1689
1690 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1691 {
1692         __be32 src;
1693
1694         if (rt_is_output_route(rt))
1695                 src = ip_hdr(skb)->saddr;
1696         else {
1697                 struct fib_result res;
1698                 struct flowi4 fl4;
1699                 struct iphdr *iph;
1700
1701                 iph = ip_hdr(skb);
1702
1703                 memset(&fl4, 0, sizeof(fl4));
1704                 fl4.daddr = iph->daddr;
1705                 fl4.saddr = iph->saddr;
1706                 fl4.flowi4_tos = iph->tos;
1707                 fl4.flowi4_oif = rt->dst.dev->ifindex;
1708                 fl4.flowi4_iif = skb->dev->ifindex;
1709                 fl4.flowi4_mark = skb->mark;
1710
1711                 rcu_read_lock();
1712                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1713                         src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1714                 else
1715                         src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1716                                         RT_SCOPE_UNIVERSE);
1717                 rcu_read_unlock();
1718         }
1719         memcpy(addr, &src, 4);
1720 }
1721
1722 #ifdef CONFIG_IP_ROUTE_CLASSID
1723 static void set_class_tag(struct rtable *rt, u32 tag)
1724 {
1725         if (!(rt->dst.tclassid & 0xFFFF))
1726                 rt->dst.tclassid |= tag & 0xFFFF;
1727         if (!(rt->dst.tclassid & 0xFFFF0000))
1728                 rt->dst.tclassid |= tag & 0xFFFF0000;
1729 }
1730 #endif
1731
1732 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1733 {
1734         unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1735
1736         if (advmss == 0) {
1737                 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1738                                ip_rt_min_advmss);
1739                 if (advmss > 65535 - 40)
1740                         advmss = 65535 - 40;
1741         }
1742         return advmss;
1743 }
1744
1745 static unsigned int ipv4_default_mtu(const struct dst_entry *dst)
1746 {
1747         unsigned int mtu = dst->dev->mtu;
1748
1749         if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1750                 const struct rtable *rt = (const struct rtable *) dst;
1751
1752                 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1753                         mtu = 576;
1754         }
1755
1756         if (mtu > IP_MAX_MTU)
1757                 mtu = IP_MAX_MTU;
1758
1759         return mtu;
1760 }
1761
1762 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1763                             struct fib_info *fi)
1764 {
1765         struct inet_peer *peer;
1766         int create = 0;
1767
1768         /* If a peer entry exists for this destination, we must hook
1769          * it up in order to get at cached metrics.
1770          */
1771         if (fl4 && (fl4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS))
1772                 create = 1;
1773
1774         rt->peer = peer = inet_getpeer_v4(rt->rt_dst, create);
1775         if (peer) {
1776                 rt->rt_peer_genid = rt_peer_genid();
1777                 if (inet_metrics_new(peer))
1778                         memcpy(peer->metrics, fi->fib_metrics,
1779                                sizeof(u32) * RTAX_MAX);
1780                 dst_init_metrics(&rt->dst, peer->metrics, false);
1781
1782                 check_peer_pmtu(&rt->dst, peer);
1783                 if (peer->redirect_learned.a4 &&
1784                     peer->redirect_learned.a4 != rt->rt_gateway) {
1785                         rt->rt_gateway = peer->redirect_learned.a4;
1786                         rt->rt_flags |= RTCF_REDIRECTED;
1787                 }
1788         } else {
1789                 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1790                         rt->fi = fi;
1791                         atomic_inc(&fi->fib_clntref);
1792                 }
1793                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1794         }
1795 }
1796
1797 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1798                            const struct fib_result *res,
1799                            struct fib_info *fi, u16 type, u32 itag)
1800 {
1801         struct dst_entry *dst = &rt->dst;
1802
1803         if (fi) {
1804                 if (FIB_RES_GW(*res) &&
1805                     FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1806                         rt->rt_gateway = FIB_RES_GW(*res);
1807                 rt_init_metrics(rt, fl4, fi);
1808 #ifdef CONFIG_IP_ROUTE_CLASSID
1809                 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1810 #endif
1811         }
1812
1813         if (dst_mtu(dst) > IP_MAX_MTU)
1814                 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1815         if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40)
1816                 dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40);
1817
1818 #ifdef CONFIG_IP_ROUTE_CLASSID
1819 #ifdef CONFIG_IP_MULTIPLE_TABLES
1820         set_class_tag(rt, fib_rules_tclass(res));
1821 #endif
1822         set_class_tag(rt, itag);
1823 #endif
1824 }
1825
1826 static struct rtable *rt_dst_alloc(struct net_device *dev,
1827                                    bool nopolicy, bool noxfrm)
1828 {
1829         return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1830                          DST_HOST |
1831                          (nopolicy ? DST_NOPOLICY : 0) |
1832                          (noxfrm ? DST_NOXFRM : 0));
1833 }
1834
1835 /* called in rcu_read_lock() section */
1836 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1837                                 u8 tos, struct net_device *dev, int our)
1838 {
1839         unsigned int hash;
1840         struct rtable *rth;
1841         __be32 spec_dst;
1842         struct in_device *in_dev = __in_dev_get_rcu(dev);
1843         u32 itag = 0;
1844         int err;
1845
1846         /* Primary sanity checks. */
1847
1848         if (in_dev == NULL)
1849                 return -EINVAL;
1850
1851         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1852             ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1853                 goto e_inval;
1854
1855         if (ipv4_is_zeronet(saddr)) {
1856                 if (!ipv4_is_local_multicast(daddr))
1857                         goto e_inval;
1858                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1859         } else {
1860                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
1861                                           &itag);
1862                 if (err < 0)
1863                         goto e_err;
1864         }
1865         rth = rt_dst_alloc(init_net.loopback_dev,
1866                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1867         if (!rth)
1868                 goto e_nobufs;
1869
1870 #ifdef CONFIG_IP_ROUTE_CLASSID
1871         rth->dst.tclassid = itag;
1872 #endif
1873         rth->dst.output = ip_rt_bug;
1874
1875         rth->rt_key_dst = daddr;
1876         rth->rt_key_src = saddr;
1877         rth->rt_genid   = rt_genid(dev_net(dev));
1878         rth->rt_flags   = RTCF_MULTICAST;
1879         rth->rt_type    = RTN_MULTICAST;
1880         rth->rt_key_tos = tos;
1881         rth->rt_dst     = daddr;
1882         rth->rt_src     = saddr;
1883         rth->rt_route_iif = dev->ifindex;
1884         rth->rt_iif     = dev->ifindex;
1885         rth->rt_oif     = 0;
1886         rth->rt_mark    = skb->mark;
1887         rth->rt_gateway = daddr;
1888         rth->rt_spec_dst= spec_dst;
1889         rth->rt_peer_genid = 0;
1890         rth->peer = NULL;
1891         rth->fi = NULL;
1892         if (our) {
1893                 rth->dst.input= ip_local_deliver;
1894                 rth->rt_flags |= RTCF_LOCAL;
1895         }
1896
1897 #ifdef CONFIG_IP_MROUTE
1898         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1899                 rth->dst.input = ip_mr_input;
1900 #endif
1901         RT_CACHE_STAT_INC(in_slow_mc);
1902
1903         hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1904         rth = rt_intern_hash(hash, rth, skb, dev->ifindex);
1905         return IS_ERR(rth) ? PTR_ERR(rth) : 0;
1906
1907 e_nobufs:
1908         return -ENOBUFS;
1909 e_inval:
1910         return -EINVAL;
1911 e_err:
1912         return err;
1913 }
1914
1915
1916 static void ip_handle_martian_source(struct net_device *dev,
1917                                      struct in_device *in_dev,
1918                                      struct sk_buff *skb,
1919                                      __be32 daddr,
1920                                      __be32 saddr)
1921 {
1922         RT_CACHE_STAT_INC(in_martian_src);
1923 #ifdef CONFIG_IP_ROUTE_VERBOSE
1924         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1925                 /*
1926                  *      RFC1812 recommendation, if source is martian,
1927                  *      the only hint is MAC header.
1928                  */
1929                 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1930                         &daddr, &saddr, dev->name);
1931                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1932                         int i;
1933                         const unsigned char *p = skb_mac_header(skb);
1934                         printk(KERN_WARNING "ll header: ");
1935                         for (i = 0; i < dev->hard_header_len; i++, p++) {
1936                                 printk("%02x", *p);
1937                                 if (i < (dev->hard_header_len - 1))
1938                                         printk(":");
1939                         }
1940                         printk("\n");
1941                 }
1942         }
1943 #endif
1944 }
1945
1946 /* called in rcu_read_lock() section */
1947 static int __mkroute_input(struct sk_buff *skb,
1948                            const struct fib_result *res,
1949                            struct in_device *in_dev,
1950                            __be32 daddr, __be32 saddr, u32 tos,
1951                            struct rtable **result)
1952 {
1953         struct rtable *rth;
1954         int err;
1955         struct in_device *out_dev;
1956         unsigned int flags = 0;
1957         __be32 spec_dst;
1958         u32 itag;
1959
1960         /* get a working reference to the output device */
1961         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1962         if (out_dev == NULL) {
1963                 if (net_ratelimit())
1964                         printk(KERN_CRIT "Bug in ip_route_input" \
1965                                "_slow(). Please, report\n");
1966                 return -EINVAL;
1967         }
1968
1969
1970         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1971                                   in_dev->dev, &spec_dst, &itag);
1972         if (err < 0) {
1973                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1974                                          saddr);
1975
1976                 goto cleanup;
1977         }
1978
1979         if (err)
1980                 flags |= RTCF_DIRECTSRC;
1981
1982         if (out_dev == in_dev && err &&
1983             (IN_DEV_SHARED_MEDIA(out_dev) ||
1984              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1985                 flags |= RTCF_DOREDIRECT;
1986
1987         if (skb->protocol != htons(ETH_P_IP)) {
1988                 /* Not IP (i.e. ARP). Do not create route, if it is
1989                  * invalid for proxy arp. DNAT routes are always valid.
1990                  *
1991                  * Proxy arp feature have been extended to allow, ARP
1992                  * replies back to the same interface, to support
1993                  * Private VLAN switch technologies. See arp.c.
1994                  */
1995                 if (out_dev == in_dev &&
1996                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1997                         err = -EINVAL;
1998                         goto cleanup;
1999                 }
2000         }
2001
2002         rth = rt_dst_alloc(out_dev->dev,
2003                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
2004                            IN_DEV_CONF_GET(out_dev, NOXFRM));
2005         if (!rth) {
2006                 err = -ENOBUFS;
2007                 goto cleanup;
2008         }
2009
2010         rth->rt_key_dst = daddr;
2011         rth->rt_key_src = saddr;
2012         rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2013         rth->rt_flags = flags;
2014         rth->rt_type = res->type;
2015         rth->rt_key_tos = tos;
2016         rth->rt_dst     = daddr;
2017         rth->rt_src     = saddr;
2018         rth->rt_route_iif = in_dev->dev->ifindex;
2019         rth->rt_iif     = in_dev->dev->ifindex;
2020         rth->rt_oif     = 0;
2021         rth->rt_mark    = skb->mark;
2022         rth->rt_gateway = daddr;
2023         rth->rt_spec_dst= spec_dst;
2024         rth->rt_peer_genid = 0;
2025         rth->peer = NULL;
2026         rth->fi = NULL;
2027
2028         rth->dst.input = ip_forward;
2029         rth->dst.output = ip_output;
2030
2031         rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
2032
2033         *result = rth;
2034         err = 0;
2035  cleanup:
2036         return err;
2037 }
2038
2039 static int ip_mkroute_input(struct sk_buff *skb,
2040                             struct fib_result *res,
2041                             const struct flowi4 *fl4,
2042                             struct in_device *in_dev,
2043                             __be32 daddr, __be32 saddr, u32 tos)
2044 {
2045         struct rtable* rth = NULL;
2046         int err;
2047         unsigned hash;
2048
2049 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2050         if (res->fi && res->fi->fib_nhs > 1)
2051                 fib_select_multipath(res);
2052 #endif
2053
2054         /* create a routing cache entry */
2055         err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2056         if (err)
2057                 return err;
2058
2059         /* put it into the cache */
2060         hash = rt_hash(daddr, saddr, fl4->flowi4_iif,
2061                        rt_genid(dev_net(rth->dst.dev)));
2062         rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif);
2063         if (IS_ERR(rth))
2064                 return PTR_ERR(rth);
2065         return 0;
2066 }
2067
2068 /*
2069  *      NOTE. We drop all the packets that has local source
2070  *      addresses, because every properly looped back packet
2071  *      must have correct destination already attached by output routine.
2072  *
2073  *      Such approach solves two big problems:
2074  *      1. Not simplex devices are handled properly.
2075  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2076  *      called with rcu_read_lock()
2077  */
2078
2079 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2080                                u8 tos, struct net_device *dev)
2081 {
2082         struct fib_result res;
2083         struct in_device *in_dev = __in_dev_get_rcu(dev);
2084         struct flowi4   fl4;
2085         unsigned        flags = 0;
2086         u32             itag = 0;
2087         struct rtable * rth;
2088         unsigned        hash;
2089         __be32          spec_dst;
2090         int             err = -EINVAL;
2091         struct net    * net = dev_net(dev);
2092
2093         /* IP on this device is disabled. */
2094
2095         if (!in_dev)
2096                 goto out;
2097
2098         /* Check for the most weird martians, which can be not detected
2099            by fib_lookup.
2100          */
2101
2102         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2103             ipv4_is_loopback(saddr))
2104                 goto martian_source;
2105
2106         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2107                 goto brd_input;
2108
2109         /* Accept zero addresses only to limited broadcast;
2110          * I even do not know to fix it or not. Waiting for complains :-)
2111          */
2112         if (ipv4_is_zeronet(saddr))
2113                 goto martian_source;
2114
2115         if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2116                 goto martian_destination;
2117
2118         /*
2119          *      Now we are ready to route packet.
2120          */
2121         fl4.flowi4_oif = 0;
2122         fl4.flowi4_iif = dev->ifindex;
2123         fl4.flowi4_mark = skb->mark;
2124         fl4.flowi4_tos = tos;
2125         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
2126         fl4.daddr = daddr;
2127         fl4.saddr = saddr;
2128         err = fib_lookup(net, &fl4, &res);
2129         if (err != 0) {
2130                 if (!IN_DEV_FORWARD(in_dev))
2131                         goto e_hostunreach;
2132                 goto no_route;
2133         }
2134
2135         RT_CACHE_STAT_INC(in_slow_tot);
2136
2137         if (res.type == RTN_BROADCAST)
2138                 goto brd_input;
2139
2140         if (res.type == RTN_LOCAL) {
2141                 err = fib_validate_source(skb, saddr, daddr, tos,
2142                                           net->loopback_dev->ifindex,
2143                                           dev, &spec_dst, &itag);
2144                 if (err < 0)
2145                         goto martian_source_keep_err;
2146                 if (err)
2147                         flags |= RTCF_DIRECTSRC;
2148                 spec_dst = daddr;
2149                 goto local_input;
2150         }
2151
2152         if (!IN_DEV_FORWARD(in_dev))
2153                 goto e_hostunreach;
2154         if (res.type != RTN_UNICAST)
2155                 goto martian_destination;
2156
2157         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
2158 out:    return err;
2159
2160 brd_input:
2161         if (skb->protocol != htons(ETH_P_IP))
2162                 goto e_inval;
2163
2164         if (ipv4_is_zeronet(saddr))
2165                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2166         else {
2167                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, &spec_dst,
2168                                           &itag);
2169                 if (err < 0)
2170                         goto martian_source_keep_err;
2171                 if (err)
2172                         flags |= RTCF_DIRECTSRC;
2173         }
2174         flags |= RTCF_BROADCAST;
2175         res.type = RTN_BROADCAST;
2176         RT_CACHE_STAT_INC(in_brd);
2177
2178 local_input:
2179         rth = rt_dst_alloc(net->loopback_dev,
2180                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
2181         if (!rth)
2182                 goto e_nobufs;
2183
2184         rth->dst.input= ip_local_deliver;
2185         rth->dst.output= ip_rt_bug;
2186 #ifdef CONFIG_IP_ROUTE_CLASSID
2187         rth->dst.tclassid = itag;
2188 #endif
2189
2190         rth->rt_key_dst = daddr;
2191         rth->rt_key_src = saddr;
2192         rth->rt_genid = rt_genid(net);
2193         rth->rt_flags   = flags|RTCF_LOCAL;
2194         rth->rt_type    = res.type;
2195         rth->rt_key_tos = tos;
2196         rth->rt_dst     = daddr;
2197         rth->rt_src     = saddr;
2198 #ifdef CONFIG_IP_ROUTE_CLASSID
2199         rth->dst.tclassid = itag;
2200 #endif
2201         rth->rt_route_iif = dev->ifindex;
2202         rth->rt_iif     = dev->ifindex;
2203         rth->rt_oif     = 0;
2204         rth->rt_mark    = skb->mark;
2205         rth->rt_gateway = daddr;
2206         rth->rt_spec_dst= spec_dst;
2207         rth->rt_peer_genid = 0;
2208         rth->peer = NULL;
2209         rth->fi = NULL;
2210         if (res.type == RTN_UNREACHABLE) {
2211                 rth->dst.input= ip_error;
2212                 rth->dst.error= -err;
2213                 rth->rt_flags   &= ~RTCF_LOCAL;
2214         }
2215         hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net));
2216         rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif);
2217         err = 0;
2218         if (IS_ERR(rth))
2219                 err = PTR_ERR(rth);
2220         goto out;
2221
2222 no_route:
2223         RT_CACHE_STAT_INC(in_no_route);
2224         spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2225         res.type = RTN_UNREACHABLE;
2226         if (err == -ESRCH)
2227                 err = -ENETUNREACH;
2228         goto local_input;
2229
2230         /*
2231          *      Do not cache martian addresses: they should be logged (RFC1812)
2232          */
2233 martian_destination:
2234         RT_CACHE_STAT_INC(in_martian_dst);
2235 #ifdef CONFIG_IP_ROUTE_VERBOSE
2236         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2237                 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2238                         &daddr, &saddr, dev->name);
2239 #endif
2240
2241 e_hostunreach:
2242         err = -EHOSTUNREACH;
2243         goto out;
2244
2245 e_inval:
2246         err = -EINVAL;
2247         goto out;
2248
2249 e_nobufs:
2250         err = -ENOBUFS;
2251         goto out;
2252
2253 martian_source:
2254         err = -EINVAL;
2255 martian_source_keep_err:
2256         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2257         goto out;
2258 }
2259
2260 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2261                            u8 tos, struct net_device *dev, bool noref)
2262 {
2263         struct rtable * rth;
2264         unsigned        hash;
2265         int iif = dev->ifindex;
2266         struct net *net;
2267         int res;
2268
2269         net = dev_net(dev);
2270
2271         rcu_read_lock();
2272
2273         if (!rt_caching(net))
2274                 goto skip_cache;
2275
2276         tos &= IPTOS_RT_MASK;
2277         hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2278
2279         for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2280              rth = rcu_dereference(rth->dst.rt_next)) {
2281                 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) |
2282                      ((__force u32)rth->rt_key_src ^ (__force u32)saddr) |
2283                      (rth->rt_iif ^ iif) |
2284                      rth->rt_oif |
2285                      (rth->rt_key_tos ^ tos)) == 0 &&
2286                     rth->rt_mark == skb->mark &&
2287                     net_eq(dev_net(rth->dst.dev), net) &&
2288                     !rt_is_expired(rth)) {
2289                         if (noref) {
2290                                 dst_use_noref(&rth->dst, jiffies);
2291                                 skb_dst_set_noref(skb, &rth->dst);
2292                         } else {
2293                                 dst_use(&rth->dst, jiffies);
2294                                 skb_dst_set(skb, &rth->dst);
2295                         }
2296                         RT_CACHE_STAT_INC(in_hit);
2297                         rcu_read_unlock();
2298                         return 0;
2299                 }
2300                 RT_CACHE_STAT_INC(in_hlist_search);
2301         }
2302
2303 skip_cache:
2304         /* Multicast recognition logic is moved from route cache to here.
2305            The problem was that too many Ethernet cards have broken/missing
2306            hardware multicast filters :-( As result the host on multicasting
2307            network acquires a lot of useless route cache entries, sort of
2308            SDR messages from all the world. Now we try to get rid of them.
2309            Really, provided software IP multicast filter is organized
2310            reasonably (at least, hashed), it does not result in a slowdown
2311            comparing with route cache reject entries.
2312            Note, that multicast routers are not affected, because
2313            route cache entry is created eventually.
2314          */
2315         if (ipv4_is_multicast(daddr)) {
2316                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2317
2318                 if (in_dev) {
2319                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
2320                                                   ip_hdr(skb)->protocol);
2321                         if (our
2322 #ifdef CONFIG_IP_MROUTE
2323                                 ||
2324                             (!ipv4_is_local_multicast(daddr) &&
2325                              IN_DEV_MFORWARD(in_dev))
2326 #endif
2327                            ) {
2328                                 int res = ip_route_input_mc(skb, daddr, saddr,
2329                                                             tos, dev, our);
2330                                 rcu_read_unlock();
2331                                 return res;
2332                         }
2333                 }
2334                 rcu_read_unlock();
2335                 return -EINVAL;
2336         }
2337         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2338         rcu_read_unlock();
2339         return res;
2340 }
2341 EXPORT_SYMBOL(ip_route_input_common);
2342
2343 /* called with rcu_read_lock() */
2344 static struct rtable *__mkroute_output(const struct fib_result *res,
2345                                        const struct flowi4 *fl4,
2346                                        __be32 orig_daddr, __be32 orig_saddr,
2347                                        int orig_oif, struct net_device *dev_out,
2348                                        unsigned int flags)
2349 {
2350         struct fib_info *fi = res->fi;
2351         u32 tos = RT_FL_TOS(fl4);
2352         struct in_device *in_dev;
2353         u16 type = res->type;
2354         struct rtable *rth;
2355
2356         if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
2357                 return ERR_PTR(-EINVAL);
2358
2359         if (ipv4_is_lbcast(fl4->daddr))
2360                 type = RTN_BROADCAST;
2361         else if (ipv4_is_multicast(fl4->daddr))
2362                 type = RTN_MULTICAST;
2363         else if (ipv4_is_zeronet(fl4->daddr))
2364                 return ERR_PTR(-EINVAL);
2365
2366         if (dev_out->flags & IFF_LOOPBACK)
2367                 flags |= RTCF_LOCAL;
2368
2369         in_dev = __in_dev_get_rcu(dev_out);
2370         if (!in_dev)
2371                 return ERR_PTR(-EINVAL);
2372
2373         if (type == RTN_BROADCAST) {
2374                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2375                 fi = NULL;
2376         } else if (type == RTN_MULTICAST) {
2377                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2378                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2379                                      fl4->flowi4_proto))
2380                         flags &= ~RTCF_LOCAL;
2381                 /* If multicast route do not exist use
2382                  * default one, but do not gateway in this case.
2383                  * Yes, it is hack.
2384                  */
2385                 if (fi && res->prefixlen < 4)
2386                         fi = NULL;
2387         }
2388
2389         rth = rt_dst_alloc(dev_out,
2390                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
2391                            IN_DEV_CONF_GET(in_dev, NOXFRM));
2392         if (!rth)
2393                 return ERR_PTR(-ENOBUFS);
2394
2395         rth->dst.output = ip_output;
2396
2397         rth->rt_key_dst = orig_daddr;
2398         rth->rt_key_src = orig_saddr;
2399         rth->rt_genid = rt_genid(dev_net(dev_out));
2400         rth->rt_flags   = flags;
2401         rth->rt_type    = type;
2402         rth->rt_key_tos = tos;
2403         rth->rt_dst     = fl4->daddr;
2404         rth->rt_src     = fl4->saddr;
2405         rth->rt_route_iif = 0;
2406         rth->rt_iif     = orig_oif ? : dev_out->ifindex;
2407         rth->rt_oif     = orig_oif;
2408         rth->rt_mark    = fl4->flowi4_mark;
2409         rth->rt_gateway = fl4->daddr;
2410         rth->rt_spec_dst= fl4->saddr;
2411         rth->rt_peer_genid = 0;
2412         rth->peer = NULL;
2413         rth->fi = NULL;
2414
2415         RT_CACHE_STAT_INC(out_slow_tot);
2416
2417         if (flags & RTCF_LOCAL) {
2418                 rth->dst.input = ip_local_deliver;
2419                 rth->rt_spec_dst = fl4->daddr;
2420         }
2421         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2422                 rth->rt_spec_dst = fl4->saddr;
2423                 if (flags & RTCF_LOCAL &&
2424                     !(dev_out->flags & IFF_LOOPBACK)) {
2425                         rth->dst.output = ip_mc_output;
2426                         RT_CACHE_STAT_INC(out_slow_mc);
2427                 }
2428 #ifdef CONFIG_IP_MROUTE
2429                 if (type == RTN_MULTICAST) {
2430                         if (IN_DEV_MFORWARD(in_dev) &&
2431                             !ipv4_is_local_multicast(fl4->daddr)) {
2432                                 rth->dst.input = ip_mr_input;
2433                                 rth->dst.output = ip_mc_output;
2434                         }
2435                 }
2436 #endif
2437         }
2438
2439         rt_set_nexthop(rth, fl4, res, fi, type, 0);
2440
2441         return rth;
2442 }
2443
2444 /*
2445  * Major route resolver routine.
2446  * called with rcu_read_lock();
2447  */
2448
2449 static struct rtable *ip_route_output_slow(struct net *net, struct flowi4 *fl4)
2450 {
2451         struct net_device *dev_out = NULL;
2452         u32 tos = RT_FL_TOS(fl4);
2453         unsigned int flags = 0;
2454         struct fib_result res;
2455         struct rtable *rth;
2456         __be32 orig_daddr;
2457         __be32 orig_saddr;
2458         int orig_oif;
2459
2460         res.fi          = NULL;
2461 #ifdef CONFIG_IP_MULTIPLE_TABLES
2462         res.r           = NULL;
2463 #endif
2464
2465         orig_daddr = fl4->daddr;
2466         orig_saddr = fl4->saddr;
2467         orig_oif = fl4->flowi4_oif;
2468
2469         fl4->flowi4_iif = net->loopback_dev->ifindex;
2470         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2471         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2472                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2473
2474         rcu_read_lock();
2475         if (fl4->saddr) {
2476                 rth = ERR_PTR(-EINVAL);
2477                 if (ipv4_is_multicast(fl4->saddr) ||
2478                     ipv4_is_lbcast(fl4->saddr) ||
2479                     ipv4_is_zeronet(fl4->saddr))
2480                         goto out;
2481
2482                 /* I removed check for oif == dev_out->oif here.
2483                    It was wrong for two reasons:
2484                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2485                       is assigned to multiple interfaces.
2486                    2. Moreover, we are allowed to send packets with saddr
2487                       of another iface. --ANK
2488                  */
2489
2490                 if (fl4->flowi4_oif == 0 &&
2491                     (ipv4_is_multicast(fl4->daddr) ||
2492                      ipv4_is_lbcast(fl4->daddr))) {
2493                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2494                         dev_out = __ip_dev_find(net, fl4->saddr, false);
2495                         if (dev_out == NULL)
2496                                 goto out;
2497
2498                         /* Special hack: user can direct multicasts
2499                            and limited broadcast via necessary interface
2500                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2501                            This hack is not just for fun, it allows
2502                            vic,vat and friends to work.
2503                            They bind socket to loopback, set ttl to zero
2504                            and expect that it will work.
2505                            From the viewpoint of routing cache they are broken,
2506                            because we are not allowed to build multicast path
2507                            with loopback source addr (look, routing cache
2508                            cannot know, that ttl is zero, so that packet
2509                            will not leave this host and route is valid).
2510                            Luckily, this hack is good workaround.
2511                          */
2512
2513                         fl4->flowi4_oif = dev_out->ifindex;
2514                         goto make_route;
2515                 }
2516
2517                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2518                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2519                         if (!__ip_dev_find(net, fl4->saddr, false))
2520                                 goto out;
2521                 }
2522         }
2523
2524
2525         if (fl4->flowi4_oif) {
2526                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2527                 rth = ERR_PTR(-ENODEV);
2528                 if (dev_out == NULL)
2529                         goto out;
2530
2531                 /* RACE: Check return value of inet_select_addr instead. */
2532                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2533                         rth = ERR_PTR(-ENETUNREACH);
2534                         goto out;
2535                 }
2536                 if (ipv4_is_local_multicast(fl4->daddr) ||
2537                     ipv4_is_lbcast(fl4->daddr)) {
2538                         if (!fl4->saddr)
2539                                 fl4->saddr = inet_select_addr(dev_out, 0,
2540                                                               RT_SCOPE_LINK);
2541                         goto make_route;
2542                 }
2543                 if (fl4->saddr) {
2544                         if (ipv4_is_multicast(fl4->daddr))
2545                                 fl4->saddr = inet_select_addr(dev_out, 0,
2546                                                               fl4->flowi4_scope);
2547                         else if (!fl4->daddr)
2548                                 fl4->saddr = inet_select_addr(dev_out, 0,
2549                                                               RT_SCOPE_HOST);
2550                 }
2551         }
2552
2553         if (!fl4->daddr) {
2554                 fl4->daddr = fl4->saddr;
2555                 if (!fl4->daddr)
2556                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2557                 dev_out = net->loopback_dev;
2558                 fl4->flowi4_oif = net->loopback_dev->ifindex;
2559                 res.type = RTN_LOCAL;
2560                 flags |= RTCF_LOCAL;
2561                 goto make_route;
2562         }
2563
2564         if (fib_lookup(net, fl4, &res)) {
2565                 res.fi = NULL;
2566                 if (fl4->flowi4_oif) {
2567                         /* Apparently, routing tables are wrong. Assume,
2568                            that the destination is on link.
2569
2570                            WHY? DW.
2571                            Because we are allowed to send to iface
2572                            even if it has NO routes and NO assigned
2573                            addresses. When oif is specified, routing
2574                            tables are looked up with only one purpose:
2575                            to catch if destination is gatewayed, rather than
2576                            direct. Moreover, if MSG_DONTROUTE is set,
2577                            we send packet, ignoring both routing tables
2578                            and ifaddr state. --ANK
2579
2580
2581                            We could make it even if oif is unknown,
2582                            likely IPv6, but we do not.
2583                          */
2584
2585                         if (fl4->saddr == 0)
2586                                 fl4->saddr = inet_select_addr(dev_out, 0,
2587                                                               RT_SCOPE_LINK);
2588                         res.type = RTN_UNICAST;
2589                         goto make_route;
2590                 }
2591                 rth = ERR_PTR(-ENETUNREACH);
2592                 goto out;
2593         }
2594
2595         if (res.type == RTN_LOCAL) {
2596                 if (!fl4->saddr) {
2597                         if (res.fi->fib_prefsrc)
2598                                 fl4->saddr = res.fi->fib_prefsrc;
2599                         else
2600                                 fl4->saddr = fl4->daddr;
2601                 }
2602                 dev_out = net->loopback_dev;
2603                 fl4->flowi4_oif = dev_out->ifindex;
2604                 res.fi = NULL;
2605                 flags |= RTCF_LOCAL;
2606                 goto make_route;
2607         }
2608
2609 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2610         if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2611                 fib_select_multipath(&res);
2612         else
2613 #endif
2614         if (!res.prefixlen &&
2615             res.table->tb_num_default > 1 &&
2616             res.type == RTN_UNICAST && !fl4->flowi4_oif)
2617                 fib_select_default(&res);
2618
2619         if (!fl4->saddr)
2620                 fl4->saddr = FIB_RES_PREFSRC(net, res);
2621
2622         dev_out = FIB_RES_DEV(res);
2623         fl4->flowi4_oif = dev_out->ifindex;
2624
2625
2626 make_route:
2627         rth = __mkroute_output(&res, fl4, orig_daddr, orig_saddr, orig_oif,
2628                                dev_out, flags);
2629         if (!IS_ERR(rth)) {
2630                 unsigned int hash;
2631
2632                 hash = rt_hash(orig_daddr, orig_saddr, orig_oif,
2633                                rt_genid(dev_net(dev_out)));
2634                 rth = rt_intern_hash(hash, rth, NULL, orig_oif);
2635         }
2636
2637 out:
2638         rcu_read_unlock();
2639         return rth;
2640 }
2641
2642 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *flp4)
2643 {
2644         struct rtable *rth;
2645         unsigned int hash;
2646
2647         if (!rt_caching(net))
2648                 goto slow_output;
2649
2650         hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net));
2651
2652         rcu_read_lock_bh();
2653         for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2654                 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2655                 if (rth->rt_key_dst == flp4->daddr &&
2656                     rth->rt_key_src == flp4->saddr &&
2657                     rt_is_output_route(rth) &&
2658                     rth->rt_oif == flp4->flowi4_oif &&
2659                     rth->rt_mark == flp4->flowi4_mark &&
2660                     !((rth->rt_key_tos ^ flp4->flowi4_tos) &
2661                             (IPTOS_RT_MASK | RTO_ONLINK)) &&
2662                     net_eq(dev_net(rth->dst.dev), net) &&
2663                     !rt_is_expired(rth)) {
2664                         dst_use(&rth->dst, jiffies);
2665                         RT_CACHE_STAT_INC(out_hit);
2666                         rcu_read_unlock_bh();
2667                         if (!flp4->saddr)
2668                                 flp4->saddr = rth->rt_src;
2669                         if (!flp4->daddr)
2670                                 flp4->daddr = rth->rt_dst;
2671                         return rth;
2672                 }
2673                 RT_CACHE_STAT_INC(out_hlist_search);
2674         }
2675         rcu_read_unlock_bh();
2676
2677 slow_output:
2678         return ip_route_output_slow(net, flp4);
2679 }
2680 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2681
2682 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2683 {
2684         return NULL;
2685 }
2686
2687 static unsigned int ipv4_blackhole_default_mtu(const struct dst_entry *dst)
2688 {
2689         return 0;
2690 }
2691
2692 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2693 {
2694 }
2695
2696 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2697                                           unsigned long old)
2698 {
2699         return NULL;
2700 }
2701
2702 static struct dst_ops ipv4_dst_blackhole_ops = {
2703         .family                 =       AF_INET,
2704         .protocol               =       cpu_to_be16(ETH_P_IP),
2705         .destroy                =       ipv4_dst_destroy,
2706         .check                  =       ipv4_blackhole_dst_check,
2707         .default_mtu            =       ipv4_blackhole_default_mtu,
2708         .default_advmss         =       ipv4_default_advmss,
2709         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2710         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
2711 };
2712
2713 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2714 {
2715         struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
2716         struct rtable *ort = (struct rtable *) dst_orig;
2717
2718         if (rt) {
2719                 struct dst_entry *new = &rt->dst;
2720
2721                 new->__use = 1;
2722                 new->input = dst_discard;
2723                 new->output = dst_discard;
2724                 dst_copy_metrics(new, &ort->dst);
2725
2726                 new->dev = ort->dst.dev;
2727                 if (new->dev)
2728                         dev_hold(new->dev);
2729
2730                 rt->rt_key_dst = ort->rt_key_dst;
2731                 rt->rt_key_src = ort->rt_key_src;
2732                 rt->rt_key_tos = ort->rt_key_tos;
2733                 rt->rt_route_iif = ort->rt_route_iif;
2734                 rt->rt_iif = ort->rt_iif;
2735                 rt->rt_oif = ort->rt_oif;
2736                 rt->rt_mark = ort->rt_mark;
2737
2738                 rt->rt_genid = rt_genid(net);
2739                 rt->rt_flags = ort->rt_flags;
2740                 rt->rt_type = ort->rt_type;
2741                 rt->rt_dst = ort->rt_dst;
2742                 rt->rt_src = ort->rt_src;
2743                 rt->rt_gateway = ort->rt_gateway;
2744                 rt->rt_spec_dst = ort->rt_spec_dst;
2745                 rt->peer = ort->peer;
2746                 if (rt->peer)
2747                         atomic_inc(&rt->peer->refcnt);
2748                 rt->fi = ort->fi;
2749                 if (rt->fi)
2750                         atomic_inc(&rt->fi->fib_clntref);
2751
2752                 dst_free(new);
2753         }
2754
2755         dst_release(dst_orig);
2756
2757         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2758 }
2759
2760 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2761                                     struct sock *sk)
2762 {
2763         struct rtable *rt = __ip_route_output_key(net, flp4);
2764
2765         if (IS_ERR(rt))
2766                 return rt;
2767
2768         if (flp4->flowi4_proto)
2769                 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2770                                                    flowi4_to_flowi(flp4),
2771                                                    sk, 0);
2772
2773         return rt;
2774 }
2775 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2776
2777 static int rt_fill_info(struct net *net,
2778                         struct sk_buff *skb, u32 pid, u32 seq, int event,
2779                         int nowait, unsigned int flags)
2780 {
2781         struct rtable *rt = skb_rtable(skb);
2782         struct rtmsg *r;
2783         struct nlmsghdr *nlh;
2784         long expires = 0;
2785         const struct inet_peer *peer = rt->peer;
2786         u32 id = 0, ts = 0, tsage = 0, error;
2787
2788         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2789         if (nlh == NULL)
2790                 return -EMSGSIZE;
2791
2792         r = nlmsg_data(nlh);
2793         r->rtm_family    = AF_INET;
2794         r->rtm_dst_len  = 32;
2795         r->rtm_src_len  = 0;
2796         r->rtm_tos      = rt->rt_key_tos;
2797         r->rtm_table    = RT_TABLE_MAIN;
2798         NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2799         r->rtm_type     = rt->rt_type;
2800         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2801         r->rtm_protocol = RTPROT_UNSPEC;
2802         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2803         if (rt->rt_flags & RTCF_NOTIFY)
2804                 r->rtm_flags |= RTM_F_NOTIFY;
2805
2806         NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2807
2808         if (rt->rt_key_src) {
2809                 r->rtm_src_len = 32;
2810                 NLA_PUT_BE32(skb, RTA_SRC, rt->rt_key_src);
2811         }
2812         if (rt->dst.dev)
2813                 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2814 #ifdef CONFIG_IP_ROUTE_CLASSID
2815         if (rt->dst.tclassid)
2816                 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2817 #endif
2818         if (rt_is_input_route(rt))
2819                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2820         else if (rt->rt_src != rt->rt_key_src)
2821                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2822
2823         if (rt->rt_dst != rt->rt_gateway)
2824                 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2825
2826         if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2827                 goto nla_put_failure;
2828
2829         if (rt->rt_mark)
2830                 NLA_PUT_BE32(skb, RTA_MARK, rt->rt_mark);
2831
2832         error = rt->dst.error;
2833         if (peer) {
2834                 inet_peer_refcheck(rt->peer);
2835                 id = atomic_read(&peer->ip_id_count) & 0xffff;
2836                 if (peer->tcp_ts_stamp) {
2837                         ts = peer->tcp_ts;
2838                         tsage = get_seconds() - peer->tcp_ts_stamp;
2839                 }
2840                 expires = ACCESS_ONCE(peer->pmtu_expires);
2841                 if (expires)
2842                         expires -= jiffies;
2843         }
2844
2845         if (rt_is_input_route(rt)) {
2846 #ifdef CONFIG_IP_MROUTE
2847                 __be32 dst = rt->rt_dst;
2848
2849                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2850                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2851                         int err = ipmr_get_route(net, skb,
2852                                                  rt->rt_src, rt->rt_dst,
2853                                                  r, nowait);
2854                         if (err <= 0) {
2855                                 if (!nowait) {
2856                                         if (err == 0)
2857                                                 return 0;
2858                                         goto nla_put_failure;
2859                                 } else {
2860                                         if (err == -EMSGSIZE)
2861                                                 goto nla_put_failure;
2862                                         error = err;
2863                                 }
2864                         }
2865                 } else
2866 #endif
2867                         NLA_PUT_U32(skb, RTA_IIF, rt->rt_iif);
2868         }
2869
2870         if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2871                                expires, error) < 0)
2872                 goto nla_put_failure;
2873
2874         return nlmsg_end(skb, nlh);
2875
2876 nla_put_failure:
2877         nlmsg_cancel(skb, nlh);
2878         return -EMSGSIZE;
2879 }
2880
2881 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2882 {
2883         struct net *net = sock_net(in_skb->sk);
2884         struct rtmsg *rtm;
2885         struct nlattr *tb[RTA_MAX+1];
2886         struct rtable *rt = NULL;
2887         __be32 dst = 0;
2888         __be32 src = 0;
2889         u32 iif;
2890         int err;
2891         int mark;
2892         struct sk_buff *skb;
2893
2894         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2895         if (err < 0)
2896                 goto errout;
2897
2898         rtm = nlmsg_data(nlh);
2899
2900         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2901         if (skb == NULL) {
2902                 err = -ENOBUFS;
2903                 goto errout;
2904         }
2905
2906         /* Reserve room for dummy headers, this skb can pass
2907            through good chunk of routing engine.
2908          */
2909         skb_reset_mac_header(skb);
2910         skb_reset_network_header(skb);
2911
2912         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2913         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2914         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2915
2916         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2917         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2918         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2919         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2920
2921         if (iif) {
2922                 struct net_device *dev;
2923
2924                 dev = __dev_get_by_index(net, iif);
2925                 if (dev == NULL) {
2926                         err = -ENODEV;
2927                         goto errout_free;
2928                 }
2929
2930                 skb->protocol   = htons(ETH_P_IP);
2931                 skb->dev        = dev;
2932                 skb->mark       = mark;
2933                 local_bh_disable();
2934                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2935                 local_bh_enable();
2936
2937                 rt = skb_rtable(skb);
2938                 if (err == 0 && rt->dst.error)
2939                         err = -rt->dst.error;
2940         } else {
2941                 struct flowi4 fl4 = {
2942                         .daddr = dst,
2943                         .saddr = src,
2944                         .flowi4_tos = rtm->rtm_tos,
2945                         .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2946                         .flowi4_mark = mark,
2947                 };
2948                 rt = ip_route_output_key(net, &fl4);
2949
2950                 err = 0;
2951                 if (IS_ERR(rt))
2952                         err = PTR_ERR(rt);
2953         }
2954
2955         if (err)
2956                 goto errout_free;
2957
2958         skb_dst_set(skb, &rt->dst);
2959         if (rtm->rtm_flags & RTM_F_NOTIFY)
2960                 rt->rt_flags |= RTCF_NOTIFY;
2961
2962         err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2963                            RTM_NEWROUTE, 0, 0);
2964         if (err <= 0)
2965                 goto errout_free;
2966
2967         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2968 errout:
2969         return err;
2970
2971 errout_free:
2972         kfree_skb(skb);
2973         goto errout;
2974 }
2975
2976 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
2977 {
2978         struct rtable *rt;
2979         int h, s_h;
2980         int idx, s_idx;
2981         struct net *net;
2982
2983         net = sock_net(skb->sk);
2984
2985         s_h = cb->args[0];
2986         if (s_h < 0)
2987                 s_h = 0;
2988         s_idx = idx = cb->args[1];
2989         for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
2990                 if (!rt_hash_table[h].chain)
2991                         continue;
2992                 rcu_read_lock_bh();
2993                 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
2994                      rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
2995                         if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
2996                                 continue;
2997                         if (rt_is_expired(rt))
2998                                 continue;
2999                         skb_dst_set_noref(skb, &rt->dst);
3000                         if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3001                                          cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3002                                          1, NLM_F_MULTI) <= 0) {
3003                                 skb_dst_drop(skb);
3004                                 rcu_read_unlock_bh();
3005                                 goto done;
3006                         }
3007                         skb_dst_drop(skb);
3008                 }
3009                 rcu_read_unlock_bh();
3010         }
3011
3012 done:
3013         cb->args[0] = h;
3014         cb->args[1] = idx;
3015         return skb->len;
3016 }
3017
3018 void ip_rt_multicast_event(struct in_device *in_dev)
3019 {
3020         rt_cache_flush(dev_net(in_dev->dev), 0);
3021 }
3022
3023 #ifdef CONFIG_SYSCTL
3024 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3025                                         void __user *buffer,
3026                                         size_t *lenp, loff_t *ppos)
3027 {
3028         if (write) {
3029                 int flush_delay;
3030                 ctl_table ctl;
3031                 struct net *net;
3032
3033                 memcpy(&ctl, __ctl, sizeof(ctl));
3034                 ctl.data = &flush_delay;
3035                 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3036
3037                 net = (struct net *)__ctl->extra1;
3038                 rt_cache_flush(net, flush_delay);
3039                 return 0;
3040         }
3041
3042         return -EINVAL;
3043 }
3044
3045 static ctl_table ipv4_route_table[] = {
3046         {
3047                 .procname       = "gc_thresh",
3048                 .data           = &ipv4_dst_ops.gc_thresh,
3049                 .maxlen         = sizeof(int),
3050                 .mode           = 0644,
3051                 .proc_handler   = proc_dointvec,
3052         },
3053         {
3054                 .procname       = "max_size",
3055                 .data           = &ip_rt_max_size,
3056                 .maxlen         = sizeof(int),
3057                 .mode           = 0644,
3058                 .proc_handler   = proc_dointvec,
3059         },
3060         {
3061                 /*  Deprecated. Use gc_min_interval_ms */
3062
3063                 .procname       = "gc_min_interval",
3064                 .data           = &ip_rt_gc_min_interval,
3065                 .maxlen         = sizeof(int),
3066                 .mode           = 0644,
3067                 .proc_handler   = proc_dointvec_jiffies,
3068         },
3069         {
3070                 .procname       = "gc_min_interval_ms",
3071                 .data           = &ip_rt_gc_min_interval,
3072                 .maxlen         = sizeof(int),
3073                 .mode           = 0644,
3074                 .proc_handler   = proc_dointvec_ms_jiffies,
3075         },
3076         {
3077                 .procname       = "gc_timeout",
3078                 .data           = &ip_rt_gc_timeout,
3079                 .maxlen         = sizeof(int),
3080                 .mode           = 0644,
3081                 .proc_handler   = proc_dointvec_jiffies,
3082         },
3083         {
3084                 .procname       = "gc_interval",
3085                 .data           = &ip_rt_gc_interval,
3086                 .maxlen         = sizeof(int),
3087                 .mode           = 0644,
3088                 .proc_handler   = proc_dointvec_jiffies,
3089         },
3090         {
3091                 .procname       = "redirect_load",
3092                 .data           = &ip_rt_redirect_load,
3093                 .maxlen         = sizeof(int),
3094                 .mode           = 0644,
3095                 .proc_handler   = proc_dointvec,
3096         },
3097         {
3098                 .procname       = "redirect_number",
3099                 .data           = &ip_rt_redirect_number,
3100                 .maxlen         = sizeof(int),
3101                 .mode           = 0644,
3102                 .proc_handler   = proc_dointvec,
3103         },
3104         {
3105                 .procname       = "redirect_silence",
3106                 .data           = &ip_rt_redirect_silence,
3107                 .maxlen         = sizeof(int),
3108                 .mode           = 0644,
3109                 .proc_handler   = proc_dointvec,
3110         },
3111         {
3112                 .procname       = "error_cost",
3113                 .data           = &ip_rt_error_cost,
3114                 .maxlen         = sizeof(int),
3115                 .mode           = 0644,
3116                 .proc_handler   = proc_dointvec,
3117         },
3118         {
3119                 .procname       = "error_burst",
3120                 .data           = &ip_rt_error_burst,
3121                 .maxlen         = sizeof(int),
3122                 .mode           = 0644,
3123                 .proc_handler   = proc_dointvec,
3124         },
3125         {
3126                 .procname       = "gc_elasticity",
3127                 .data           = &ip_rt_gc_elasticity,
3128                 .maxlen         = sizeof(int),
3129                 .mode           = 0644,
3130                 .proc_handler   = proc_dointvec,
3131         },
3132         {
3133                 .procname       = "mtu_expires",
3134                 .data           = &ip_rt_mtu_expires,
3135                 .maxlen         = sizeof(int),
3136                 .mode           = 0644,
3137                 .proc_handler   = proc_dointvec_jiffies,
3138         },
3139         {
3140                 .procname       = "min_pmtu",
3141                 .data           = &ip_rt_min_pmtu,
3142                 .maxlen         = sizeof(int),
3143                 .mode           = 0644,
3144                 .proc_handler   = proc_dointvec,
3145         },
3146         {
3147                 .procname       = "min_adv_mss",
3148                 .data           = &ip_rt_min_advmss,
3149                 .maxlen         = sizeof(int),
3150                 .mode           = 0644,
3151                 .proc_handler   = proc_dointvec,
3152         },
3153         { }
3154 };
3155
3156 static struct ctl_table empty[1];
3157
3158 static struct ctl_table ipv4_skeleton[] =
3159 {
3160         { .procname = "route", 
3161           .mode = 0555, .child = ipv4_route_table},
3162         { .procname = "neigh", 
3163           .mode = 0555, .child = empty},
3164         { }
3165 };
3166
3167 static __net_initdata struct ctl_path ipv4_path[] = {
3168         { .procname = "net", },
3169         { .procname = "ipv4", },
3170         { },
3171 };
3172
3173 static struct ctl_table ipv4_route_flush_table[] = {
3174         {
3175                 .procname       = "flush",
3176                 .maxlen         = sizeof(int),
3177                 .mode           = 0200,
3178                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3179         },
3180         { },
3181 };
3182
3183 static __net_initdata struct ctl_path ipv4_route_path[] = {
3184         { .procname = "net", },
3185         { .procname = "ipv4", },
3186         { .procname = "route", },
3187         { },
3188 };
3189
3190 static __net_init int sysctl_route_net_init(struct net *net)
3191 {
3192         struct ctl_table *tbl;
3193
3194         tbl = ipv4_route_flush_table;
3195         if (!net_eq(net, &init_net)) {
3196                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3197                 if (tbl == NULL)
3198                         goto err_dup;
3199         }
3200         tbl[0].extra1 = net;
3201
3202         net->ipv4.route_hdr =
3203                 register_net_sysctl_table(net, ipv4_route_path, tbl);
3204         if (net->ipv4.route_hdr == NULL)
3205                 goto err_reg;
3206         return 0;
3207
3208 err_reg:
3209         if (tbl != ipv4_route_flush_table)
3210                 kfree(tbl);
3211 err_dup:
3212         return -ENOMEM;
3213 }
3214
3215 static __net_exit void sysctl_route_net_exit(struct net *net)
3216 {
3217         struct ctl_table *tbl;
3218
3219         tbl = net->ipv4.route_hdr->ctl_table_arg;
3220         unregister_net_sysctl_table(net->ipv4.route_hdr);
3221         BUG_ON(tbl == ipv4_route_flush_table);
3222         kfree(tbl);
3223 }
3224
3225 static __net_initdata struct pernet_operations sysctl_route_ops = {
3226         .init = sysctl_route_net_init,
3227         .exit = sysctl_route_net_exit,
3228 };
3229 #endif
3230
3231 static __net_init int rt_genid_init(struct net *net)
3232 {
3233         get_random_bytes(&net->ipv4.rt_genid,
3234                          sizeof(net->ipv4.rt_genid));
3235         get_random_bytes(&net->ipv4.dev_addr_genid,
3236                          sizeof(net->ipv4.dev_addr_genid));
3237         return 0;
3238 }
3239
3240 static __net_initdata struct pernet_operations rt_genid_ops = {
3241         .init = rt_genid_init,
3242 };
3243
3244
3245 #ifdef CONFIG_IP_ROUTE_CLASSID
3246 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3247 #endif /* CONFIG_IP_ROUTE_CLASSID */
3248
3249 static __initdata unsigned long rhash_entries;
3250 static int __init set_rhash_entries(char *str)
3251 {
3252         if (!str)
3253                 return 0;
3254         rhash_entries = simple_strtoul(str, &str, 0);
3255         return 1;
3256 }
3257 __setup("rhash_entries=", set_rhash_entries);
3258
3259 int __init ip_rt_init(void)
3260 {
3261         int rc = 0;
3262
3263 #ifdef CONFIG_IP_ROUTE_CLASSID
3264         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3265         if (!ip_rt_acct)
3266                 panic("IP: failed to allocate ip_rt_acct\n");
3267 #endif
3268
3269         ipv4_dst_ops.kmem_cachep =
3270                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3271                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3272
3273         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3274
3275         if (dst_entries_init(&ipv4_dst_ops) < 0)
3276                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3277
3278         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3279                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3280
3281         rt_hash_table = (struct rt_hash_bucket *)
3282                 alloc_large_system_hash("IP route cache",
3283                                         sizeof(struct rt_hash_bucket),
3284                                         rhash_entries,
3285                                         (totalram_pages >= 128 * 1024) ?
3286                                         15 : 17,
3287                                         0,
3288                                         &rt_hash_log,
3289                                         &rt_hash_mask,
3290                                         rhash_entries ? 0 : 512 * 1024);
3291         memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3292         rt_hash_lock_init();
3293
3294         ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3295         ip_rt_max_size = (rt_hash_mask + 1) * 16;
3296
3297         devinet_init();
3298         ip_fib_init();
3299
3300         if (ip_rt_proc_init())
3301                 printk(KERN_ERR "Unable to create route proc files\n");
3302 #ifdef CONFIG_XFRM
3303         xfrm_init();
3304         xfrm4_init(ip_rt_max_size);
3305 #endif
3306         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3307
3308 #ifdef CONFIG_SYSCTL
3309         register_pernet_subsys(&sysctl_route_ops);
3310 #endif
3311         register_pernet_subsys(&rt_genid_ops);
3312         return rc;
3313 }
3314
3315 #ifdef CONFIG_SYSCTL
3316 /*
3317  * We really need to sanitize the damn ipv4 init order, then all
3318  * this nonsense will go away.
3319  */
3320 void __init ip_static_sysctl_init(void)
3321 {
3322         register_sysctl_paths(ipv4_path, ipv4_skeleton);
3323 }
3324 #endif