net: Abstract default ADVMSS behind an accessor.
[linux-2.6.git] / net / ipv4 / route.c
1 /*
2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
3  *              operating system.  INET is implemented using the  BSD Socket
4  *              interface as the means of communication with the user level.
5  *
6  *              ROUTE - implementation of the IP router.
7  *
8  * Authors:     Ross Biro
9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
11  *              Linus Torvalds, <Linus.Torvalds@helsinki.fi>
12  *              Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
13  *
14  * Fixes:
15  *              Alan Cox        :       Verify area fixes.
16  *              Alan Cox        :       cli() protects routing changes
17  *              Rui Oliveira    :       ICMP routing table updates
18  *              (rco@di.uminho.pt)      Routing table insertion and update
19  *              Linus Torvalds  :       Rewrote bits to be sensible
20  *              Alan Cox        :       Added BSD route gw semantics
21  *              Alan Cox        :       Super /proc >4K
22  *              Alan Cox        :       MTU in route table
23  *              Alan Cox        :       MSS actually. Also added the window
24  *                                      clamper.
25  *              Sam Lantinga    :       Fixed route matching in rt_del()
26  *              Alan Cox        :       Routing cache support.
27  *              Alan Cox        :       Removed compatibility cruft.
28  *              Alan Cox        :       RTF_REJECT support.
29  *              Alan Cox        :       TCP irtt support.
30  *              Jonathan Naylor :       Added Metric support.
31  *      Miquel van Smoorenburg  :       BSD API fixes.
32  *      Miquel van Smoorenburg  :       Metrics.
33  *              Alan Cox        :       Use __u32 properly
34  *              Alan Cox        :       Aligned routing errors more closely with BSD
35  *                                      our system is still very different.
36  *              Alan Cox        :       Faster /proc handling
37  *      Alexey Kuznetsov        :       Massive rework to support tree based routing,
38  *                                      routing caches and better behaviour.
39  *
40  *              Olaf Erb        :       irtt wasn't being copied right.
41  *              Bjorn Ekwall    :       Kerneld route support.
42  *              Alan Cox        :       Multicast fixed (I hope)
43  *              Pavel Krauz     :       Limited broadcast fixed
44  *              Mike McLagan    :       Routing by source
45  *      Alexey Kuznetsov        :       End of old history. Split to fib.c and
46  *                                      route.c and rewritten from scratch.
47  *              Andi Kleen      :       Load-limit warning messages.
48  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
49  *      Vitaly E. Lavrov        :       Race condition in ip_route_input_slow.
50  *      Tobias Ringstrom        :       Uninitialized res.type in ip_route_output_slow.
51  *      Vladimir V. Ivanov      :       IP rule info (flowid) is really useful.
52  *              Marc Boucher    :       routing by fwmark
53  *      Robert Olsson           :       Added rt_cache statistics
54  *      Arnaldo C. Melo         :       Convert proc stuff to seq_file
55  *      Eric Dumazet            :       hashed spinlocks and rt_check_expire() fixes.
56  *      Ilia Sotnikov           :       Ignore TOS on PMTUD and Redirect
57  *      Ilia Sotnikov           :       Removed TOS from hash calculations
58  *
59  *              This program is free software; you can redistribute it and/or
60  *              modify it under the terms of the GNU General Public License
61  *              as published by the Free Software Foundation; either version
62  *              2 of the License, or (at your option) any later version.
63  */
64
65 #include <linux/module.h>
66 #include <asm/uaccess.h>
67 #include <asm/system.h>
68 #include <linux/bitops.h>
69 #include <linux/types.h>
70 #include <linux/kernel.h>
71 #include <linux/mm.h>
72 #include <linux/bootmem.h>
73 #include <linux/string.h>
74 #include <linux/socket.h>
75 #include <linux/sockios.h>
76 #include <linux/errno.h>
77 #include <linux/in.h>
78 #include <linux/inet.h>
79 #include <linux/netdevice.h>
80 #include <linux/proc_fs.h>
81 #include <linux/init.h>
82 #include <linux/workqueue.h>
83 #include <linux/skbuff.h>
84 #include <linux/inetdevice.h>
85 #include <linux/igmp.h>
86 #include <linux/pkt_sched.h>
87 #include <linux/mroute.h>
88 #include <linux/netfilter_ipv4.h>
89 #include <linux/random.h>
90 #include <linux/jhash.h>
91 #include <linux/rcupdate.h>
92 #include <linux/times.h>
93 #include <linux/slab.h>
94 #include <net/dst.h>
95 #include <net/net_namespace.h>
96 #include <net/protocol.h>
97 #include <net/ip.h>
98 #include <net/route.h>
99 #include <net/inetpeer.h>
100 #include <net/sock.h>
101 #include <net/ip_fib.h>
102 #include <net/arp.h>
103 #include <net/tcp.h>
104 #include <net/icmp.h>
105 #include <net/xfrm.h>
106 #include <net/netevent.h>
107 #include <net/rtnetlink.h>
108 #ifdef CONFIG_SYSCTL
109 #include <linux/sysctl.h>
110 #endif
111
112 #define RT_FL_TOS(oldflp) \
113     ((u32)(oldflp->fl4_tos & (IPTOS_RT_MASK | RTO_ONLINK)))
114
115 #define IP_MAX_MTU      0xFFF0
116
117 #define RT_GC_TIMEOUT (300*HZ)
118
119 static int ip_rt_max_size;
120 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
121 static int ip_rt_gc_interval __read_mostly      = 60 * HZ;
122 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
123 static int ip_rt_redirect_number __read_mostly  = 9;
124 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
125 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
126 static int ip_rt_error_cost __read_mostly       = HZ;
127 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
128 static int ip_rt_gc_elasticity __read_mostly    = 8;
129 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
130 static int ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
131 static int ip_rt_min_advmss __read_mostly       = 256;
132 static int rt_chain_length_max __read_mostly    = 20;
133
134 static struct delayed_work expires_work;
135 static unsigned long expires_ljiffies;
136
137 /*
138  *      Interface to generic destination cache.
139  */
140
141 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
142 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
143 static void              ipv4_dst_destroy(struct dst_entry *dst);
144 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
145 static void              ipv4_link_failure(struct sk_buff *skb);
146 static void              ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu);
147 static int rt_garbage_collect(struct dst_ops *ops);
148
149 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
150                             int how)
151 {
152 }
153
154 static struct dst_ops ipv4_dst_ops = {
155         .family =               AF_INET,
156         .protocol =             cpu_to_be16(ETH_P_IP),
157         .gc =                   rt_garbage_collect,
158         .check =                ipv4_dst_check,
159         .default_advmss =       ipv4_default_advmss,
160         .destroy =              ipv4_dst_destroy,
161         .ifdown =               ipv4_dst_ifdown,
162         .negative_advice =      ipv4_negative_advice,
163         .link_failure =         ipv4_link_failure,
164         .update_pmtu =          ip_rt_update_pmtu,
165         .local_out =            __ip_local_out,
166 };
167
168 #define ECN_OR_COST(class)      TC_PRIO_##class
169
170 const __u8 ip_tos2prio[16] = {
171         TC_PRIO_BESTEFFORT,
172         ECN_OR_COST(FILLER),
173         TC_PRIO_BESTEFFORT,
174         ECN_OR_COST(BESTEFFORT),
175         TC_PRIO_BULK,
176         ECN_OR_COST(BULK),
177         TC_PRIO_BULK,
178         ECN_OR_COST(BULK),
179         TC_PRIO_INTERACTIVE,
180         ECN_OR_COST(INTERACTIVE),
181         TC_PRIO_INTERACTIVE,
182         ECN_OR_COST(INTERACTIVE),
183         TC_PRIO_INTERACTIVE_BULK,
184         ECN_OR_COST(INTERACTIVE_BULK),
185         TC_PRIO_INTERACTIVE_BULK,
186         ECN_OR_COST(INTERACTIVE_BULK)
187 };
188
189
190 /*
191  * Route cache.
192  */
193
194 /* The locking scheme is rather straight forward:
195  *
196  * 1) Read-Copy Update protects the buckets of the central route hash.
197  * 2) Only writers remove entries, and they hold the lock
198  *    as they look at rtable reference counts.
199  * 3) Only readers acquire references to rtable entries,
200  *    they do so with atomic increments and with the
201  *    lock held.
202  */
203
204 struct rt_hash_bucket {
205         struct rtable __rcu     *chain;
206 };
207
208 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
209         defined(CONFIG_PROVE_LOCKING)
210 /*
211  * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks
212  * The size of this table is a power of two and depends on the number of CPUS.
213  * (on lockdep we have a quite big spinlock_t, so keep the size down there)
214  */
215 #ifdef CONFIG_LOCKDEP
216 # define RT_HASH_LOCK_SZ        256
217 #else
218 # if NR_CPUS >= 32
219 #  define RT_HASH_LOCK_SZ       4096
220 # elif NR_CPUS >= 16
221 #  define RT_HASH_LOCK_SZ       2048
222 # elif NR_CPUS >= 8
223 #  define RT_HASH_LOCK_SZ       1024
224 # elif NR_CPUS >= 4
225 #  define RT_HASH_LOCK_SZ       512
226 # else
227 #  define RT_HASH_LOCK_SZ       256
228 # endif
229 #endif
230
231 static spinlock_t       *rt_hash_locks;
232 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)]
233
234 static __init void rt_hash_lock_init(void)
235 {
236         int i;
237
238         rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ,
239                         GFP_KERNEL);
240         if (!rt_hash_locks)
241                 panic("IP: failed to allocate rt_hash_locks\n");
242
243         for (i = 0; i < RT_HASH_LOCK_SZ; i++)
244                 spin_lock_init(&rt_hash_locks[i]);
245 }
246 #else
247 # define rt_hash_lock_addr(slot) NULL
248
249 static inline void rt_hash_lock_init(void)
250 {
251 }
252 #endif
253
254 static struct rt_hash_bucket    *rt_hash_table __read_mostly;
255 static unsigned                 rt_hash_mask __read_mostly;
256 static unsigned int             rt_hash_log  __read_mostly;
257
258 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
259 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
260
261 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx,
262                                    int genid)
263 {
264         return jhash_3words((__force u32)daddr, (__force u32)saddr,
265                             idx, genid)
266                 & rt_hash_mask;
267 }
268
269 static inline int rt_genid(struct net *net)
270 {
271         return atomic_read(&net->ipv4.rt_genid);
272 }
273
274 #ifdef CONFIG_PROC_FS
275 struct rt_cache_iter_state {
276         struct seq_net_private p;
277         int bucket;
278         int genid;
279 };
280
281 static struct rtable *rt_cache_get_first(struct seq_file *seq)
282 {
283         struct rt_cache_iter_state *st = seq->private;
284         struct rtable *r = NULL;
285
286         for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) {
287                 if (!rcu_dereference_raw(rt_hash_table[st->bucket].chain))
288                         continue;
289                 rcu_read_lock_bh();
290                 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
291                 while (r) {
292                         if (dev_net(r->dst.dev) == seq_file_net(seq) &&
293                             r->rt_genid == st->genid)
294                                 return r;
295                         r = rcu_dereference_bh(r->dst.rt_next);
296                 }
297                 rcu_read_unlock_bh();
298         }
299         return r;
300 }
301
302 static struct rtable *__rt_cache_get_next(struct seq_file *seq,
303                                           struct rtable *r)
304 {
305         struct rt_cache_iter_state *st = seq->private;
306
307         r = rcu_dereference_bh(r->dst.rt_next);
308         while (!r) {
309                 rcu_read_unlock_bh();
310                 do {
311                         if (--st->bucket < 0)
312                                 return NULL;
313                 } while (!rcu_dereference_raw(rt_hash_table[st->bucket].chain));
314                 rcu_read_lock_bh();
315                 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain);
316         }
317         return r;
318 }
319
320 static struct rtable *rt_cache_get_next(struct seq_file *seq,
321                                         struct rtable *r)
322 {
323         struct rt_cache_iter_state *st = seq->private;
324         while ((r = __rt_cache_get_next(seq, r)) != NULL) {
325                 if (dev_net(r->dst.dev) != seq_file_net(seq))
326                         continue;
327                 if (r->rt_genid == st->genid)
328                         break;
329         }
330         return r;
331 }
332
333 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos)
334 {
335         struct rtable *r = rt_cache_get_first(seq);
336
337         if (r)
338                 while (pos && (r = rt_cache_get_next(seq, r)))
339                         --pos;
340         return pos ? NULL : r;
341 }
342
343 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
344 {
345         struct rt_cache_iter_state *st = seq->private;
346         if (*pos)
347                 return rt_cache_get_idx(seq, *pos - 1);
348         st->genid = rt_genid(seq_file_net(seq));
349         return SEQ_START_TOKEN;
350 }
351
352 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
353 {
354         struct rtable *r;
355
356         if (v == SEQ_START_TOKEN)
357                 r = rt_cache_get_first(seq);
358         else
359                 r = rt_cache_get_next(seq, v);
360         ++*pos;
361         return r;
362 }
363
364 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
365 {
366         if (v && v != SEQ_START_TOKEN)
367                 rcu_read_unlock_bh();
368 }
369
370 static int rt_cache_seq_show(struct seq_file *seq, void *v)
371 {
372         if (v == SEQ_START_TOKEN)
373                 seq_printf(seq, "%-127s\n",
374                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
375                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
376                            "HHUptod\tSpecDst");
377         else {
378                 struct rtable *r = v;
379                 int len;
380
381                 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t"
382                               "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n",
383                         r->dst.dev ? r->dst.dev->name : "*",
384                         (__force u32)r->rt_dst,
385                         (__force u32)r->rt_gateway,
386                         r->rt_flags, atomic_read(&r->dst.__refcnt),
387                         r->dst.__use, 0, (__force u32)r->rt_src,
388                         dst_metric_advmss(&r->dst) + 40,
389                         dst_metric(&r->dst, RTAX_WINDOW),
390                         (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) +
391                               dst_metric(&r->dst, RTAX_RTTVAR)),
392                         r->fl.fl4_tos,
393                         r->dst.hh ? atomic_read(&r->dst.hh->hh_refcnt) : -1,
394                         r->dst.hh ? (r->dst.hh->hh_output ==
395                                        dev_queue_xmit) : 0,
396                         r->rt_spec_dst, &len);
397
398                 seq_printf(seq, "%*s\n", 127 - len, "");
399         }
400         return 0;
401 }
402
403 static const struct seq_operations rt_cache_seq_ops = {
404         .start  = rt_cache_seq_start,
405         .next   = rt_cache_seq_next,
406         .stop   = rt_cache_seq_stop,
407         .show   = rt_cache_seq_show,
408 };
409
410 static int rt_cache_seq_open(struct inode *inode, struct file *file)
411 {
412         return seq_open_net(inode, file, &rt_cache_seq_ops,
413                         sizeof(struct rt_cache_iter_state));
414 }
415
416 static const struct file_operations rt_cache_seq_fops = {
417         .owner   = THIS_MODULE,
418         .open    = rt_cache_seq_open,
419         .read    = seq_read,
420         .llseek  = seq_lseek,
421         .release = seq_release_net,
422 };
423
424
425 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
426 {
427         int cpu;
428
429         if (*pos == 0)
430                 return SEQ_START_TOKEN;
431
432         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
433                 if (!cpu_possible(cpu))
434                         continue;
435                 *pos = cpu+1;
436                 return &per_cpu(rt_cache_stat, cpu);
437         }
438         return NULL;
439 }
440
441 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
442 {
443         int cpu;
444
445         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
446                 if (!cpu_possible(cpu))
447                         continue;
448                 *pos = cpu+1;
449                 return &per_cpu(rt_cache_stat, cpu);
450         }
451         return NULL;
452
453 }
454
455 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
456 {
457
458 }
459
460 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
461 {
462         struct rt_cache_stat *st = v;
463
464         if (v == SEQ_START_TOKEN) {
465                 seq_printf(seq, "entries  in_hit in_slow_tot in_slow_mc in_no_route in_brd in_martian_dst in_martian_src  out_hit out_slow_tot out_slow_mc  gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n");
466                 return 0;
467         }
468
469         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
470                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
471                    dst_entries_get_slow(&ipv4_dst_ops),
472                    st->in_hit,
473                    st->in_slow_tot,
474                    st->in_slow_mc,
475                    st->in_no_route,
476                    st->in_brd,
477                    st->in_martian_dst,
478                    st->in_martian_src,
479
480                    st->out_hit,
481                    st->out_slow_tot,
482                    st->out_slow_mc,
483
484                    st->gc_total,
485                    st->gc_ignored,
486                    st->gc_goal_miss,
487                    st->gc_dst_overflow,
488                    st->in_hlist_search,
489                    st->out_hlist_search
490                 );
491         return 0;
492 }
493
494 static const struct seq_operations rt_cpu_seq_ops = {
495         .start  = rt_cpu_seq_start,
496         .next   = rt_cpu_seq_next,
497         .stop   = rt_cpu_seq_stop,
498         .show   = rt_cpu_seq_show,
499 };
500
501
502 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
503 {
504         return seq_open(file, &rt_cpu_seq_ops);
505 }
506
507 static const struct file_operations rt_cpu_seq_fops = {
508         .owner   = THIS_MODULE,
509         .open    = rt_cpu_seq_open,
510         .read    = seq_read,
511         .llseek  = seq_lseek,
512         .release = seq_release,
513 };
514
515 #ifdef CONFIG_NET_CLS_ROUTE
516 static int rt_acct_proc_show(struct seq_file *m, void *v)
517 {
518         struct ip_rt_acct *dst, *src;
519         unsigned int i, j;
520
521         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
522         if (!dst)
523                 return -ENOMEM;
524
525         for_each_possible_cpu(i) {
526                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
527                 for (j = 0; j < 256; j++) {
528                         dst[j].o_bytes   += src[j].o_bytes;
529                         dst[j].o_packets += src[j].o_packets;
530                         dst[j].i_bytes   += src[j].i_bytes;
531                         dst[j].i_packets += src[j].i_packets;
532                 }
533         }
534
535         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
536         kfree(dst);
537         return 0;
538 }
539
540 static int rt_acct_proc_open(struct inode *inode, struct file *file)
541 {
542         return single_open(file, rt_acct_proc_show, NULL);
543 }
544
545 static const struct file_operations rt_acct_proc_fops = {
546         .owner          = THIS_MODULE,
547         .open           = rt_acct_proc_open,
548         .read           = seq_read,
549         .llseek         = seq_lseek,
550         .release        = single_release,
551 };
552 #endif
553
554 static int __net_init ip_rt_do_proc_init(struct net *net)
555 {
556         struct proc_dir_entry *pde;
557
558         pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
559                         &rt_cache_seq_fops);
560         if (!pde)
561                 goto err1;
562
563         pde = proc_create("rt_cache", S_IRUGO,
564                           net->proc_net_stat, &rt_cpu_seq_fops);
565         if (!pde)
566                 goto err2;
567
568 #ifdef CONFIG_NET_CLS_ROUTE
569         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
570         if (!pde)
571                 goto err3;
572 #endif
573         return 0;
574
575 #ifdef CONFIG_NET_CLS_ROUTE
576 err3:
577         remove_proc_entry("rt_cache", net->proc_net_stat);
578 #endif
579 err2:
580         remove_proc_entry("rt_cache", net->proc_net);
581 err1:
582         return -ENOMEM;
583 }
584
585 static void __net_exit ip_rt_do_proc_exit(struct net *net)
586 {
587         remove_proc_entry("rt_cache", net->proc_net_stat);
588         remove_proc_entry("rt_cache", net->proc_net);
589 #ifdef CONFIG_NET_CLS_ROUTE
590         remove_proc_entry("rt_acct", net->proc_net);
591 #endif
592 }
593
594 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
595         .init = ip_rt_do_proc_init,
596         .exit = ip_rt_do_proc_exit,
597 };
598
599 static int __init ip_rt_proc_init(void)
600 {
601         return register_pernet_subsys(&ip_rt_proc_ops);
602 }
603
604 #else
605 static inline int ip_rt_proc_init(void)
606 {
607         return 0;
608 }
609 #endif /* CONFIG_PROC_FS */
610
611 static inline void rt_free(struct rtable *rt)
612 {
613         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
614 }
615
616 static inline void rt_drop(struct rtable *rt)
617 {
618         ip_rt_put(rt);
619         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
620 }
621
622 static inline int rt_fast_clean(struct rtable *rth)
623 {
624         /* Kill broadcast/multicast entries very aggresively, if they
625            collide in hash table with more useful entries */
626         return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
627                 rt_is_input_route(rth) && rth->dst.rt_next;
628 }
629
630 static inline int rt_valuable(struct rtable *rth)
631 {
632         return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) ||
633                 rth->dst.expires;
634 }
635
636 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2)
637 {
638         unsigned long age;
639         int ret = 0;
640
641         if (atomic_read(&rth->dst.__refcnt))
642                 goto out;
643
644         ret = 1;
645         if (rth->dst.expires &&
646             time_after_eq(jiffies, rth->dst.expires))
647                 goto out;
648
649         age = jiffies - rth->dst.lastuse;
650         ret = 0;
651         if ((age <= tmo1 && !rt_fast_clean(rth)) ||
652             (age <= tmo2 && rt_valuable(rth)))
653                 goto out;
654         ret = 1;
655 out:    return ret;
656 }
657
658 /* Bits of score are:
659  * 31: very valuable
660  * 30: not quite useless
661  * 29..0: usage counter
662  */
663 static inline u32 rt_score(struct rtable *rt)
664 {
665         u32 score = jiffies - rt->dst.lastuse;
666
667         score = ~score & ~(3<<30);
668
669         if (rt_valuable(rt))
670                 score |= (1<<31);
671
672         if (rt_is_output_route(rt) ||
673             !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL)))
674                 score |= (1<<30);
675
676         return score;
677 }
678
679 static inline bool rt_caching(const struct net *net)
680 {
681         return net->ipv4.current_rt_cache_rebuild_count <=
682                 net->ipv4.sysctl_rt_cache_rebuild_count;
683 }
684
685 static inline bool compare_hash_inputs(const struct flowi *fl1,
686                                         const struct flowi *fl2)
687 {
688         return ((((__force u32)fl1->fl4_dst ^ (__force u32)fl2->fl4_dst) |
689                 ((__force u32)fl1->fl4_src ^ (__force u32)fl2->fl4_src) |
690                 (fl1->iif ^ fl2->iif)) == 0);
691 }
692
693 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
694 {
695         return (((__force u32)fl1->fl4_dst ^ (__force u32)fl2->fl4_dst) |
696                 ((__force u32)fl1->fl4_src ^ (__force u32)fl2->fl4_src) |
697                 (fl1->mark ^ fl2->mark) |
698                 (*(u16 *)&fl1->fl4_tos ^ *(u16 *)&fl2->fl4_tos) |
699                 (fl1->oif ^ fl2->oif) |
700                 (fl1->iif ^ fl2->iif)) == 0;
701 }
702
703 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2)
704 {
705         return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev));
706 }
707
708 static inline int rt_is_expired(struct rtable *rth)
709 {
710         return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
711 }
712
713 /*
714  * Perform a full scan of hash table and free all entries.
715  * Can be called by a softirq or a process.
716  * In the later case, we want to be reschedule if necessary
717  */
718 static void rt_do_flush(int process_context)
719 {
720         unsigned int i;
721         struct rtable *rth, *next;
722         struct rtable * tail;
723
724         for (i = 0; i <= rt_hash_mask; i++) {
725                 if (process_context && need_resched())
726                         cond_resched();
727                 rth = rcu_dereference_raw(rt_hash_table[i].chain);
728                 if (!rth)
729                         continue;
730
731                 spin_lock_bh(rt_hash_lock_addr(i));
732 #ifdef CONFIG_NET_NS
733                 {
734                 struct rtable __rcu **prev;
735                 struct rtable *p;
736
737                 rth = rcu_dereference_protected(rt_hash_table[i].chain,
738                         lockdep_is_held(rt_hash_lock_addr(i)));
739
740                 /* defer releasing the head of the list after spin_unlock */
741                 for (tail = rth; tail;
742                      tail = rcu_dereference_protected(tail->dst.rt_next,
743                                 lockdep_is_held(rt_hash_lock_addr(i))))
744                         if (!rt_is_expired(tail))
745                                 break;
746                 if (rth != tail)
747                         rt_hash_table[i].chain = tail;
748
749                 /* call rt_free on entries after the tail requiring flush */
750                 prev = &rt_hash_table[i].chain;
751                 for (p = rcu_dereference_protected(*prev,
752                                 lockdep_is_held(rt_hash_lock_addr(i)));
753                      p != NULL;
754                      p = next) {
755                         next = rcu_dereference_protected(p->dst.rt_next,
756                                 lockdep_is_held(rt_hash_lock_addr(i)));
757                         if (!rt_is_expired(p)) {
758                                 prev = &p->dst.rt_next;
759                         } else {
760                                 *prev = next;
761                                 rt_free(p);
762                         }
763                 }
764                 }
765 #else
766                 rth = rcu_dereference_protected(rt_hash_table[i].chain,
767                         lockdep_is_held(rt_hash_lock_addr(i)));
768                 rcu_assign_pointer(rt_hash_table[i].chain, NULL);
769                 tail = NULL;
770 #endif
771                 spin_unlock_bh(rt_hash_lock_addr(i));
772
773                 for (; rth != tail; rth = next) {
774                         next = rcu_dereference_protected(rth->dst.rt_next, 1);
775                         rt_free(rth);
776                 }
777         }
778 }
779
780 /*
781  * While freeing expired entries, we compute average chain length
782  * and standard deviation, using fixed-point arithmetic.
783  * This to have an estimation of rt_chain_length_max
784  *  rt_chain_length_max = max(elasticity, AVG + 4*SD)
785  * We use 3 bits for frational part, and 29 (or 61) for magnitude.
786  */
787
788 #define FRACT_BITS 3
789 #define ONE (1UL << FRACT_BITS)
790
791 /*
792  * Given a hash chain and an item in this hash chain,
793  * find if a previous entry has the same hash_inputs
794  * (but differs on tos, mark or oif)
795  * Returns 0 if an alias is found.
796  * Returns ONE if rth has no alias before itself.
797  */
798 static int has_noalias(const struct rtable *head, const struct rtable *rth)
799 {
800         const struct rtable *aux = head;
801
802         while (aux != rth) {
803                 if (compare_hash_inputs(&aux->fl, &rth->fl))
804                         return 0;
805                 aux = rcu_dereference_protected(aux->dst.rt_next, 1);
806         }
807         return ONE;
808 }
809
810 static void rt_check_expire(void)
811 {
812         static unsigned int rover;
813         unsigned int i = rover, goal;
814         struct rtable *rth;
815         struct rtable __rcu **rthp;
816         unsigned long samples = 0;
817         unsigned long sum = 0, sum2 = 0;
818         unsigned long delta;
819         u64 mult;
820
821         delta = jiffies - expires_ljiffies;
822         expires_ljiffies = jiffies;
823         mult = ((u64)delta) << rt_hash_log;
824         if (ip_rt_gc_timeout > 1)
825                 do_div(mult, ip_rt_gc_timeout);
826         goal = (unsigned int)mult;
827         if (goal > rt_hash_mask)
828                 goal = rt_hash_mask + 1;
829         for (; goal > 0; goal--) {
830                 unsigned long tmo = ip_rt_gc_timeout;
831                 unsigned long length;
832
833                 i = (i + 1) & rt_hash_mask;
834                 rthp = &rt_hash_table[i].chain;
835
836                 if (need_resched())
837                         cond_resched();
838
839                 samples++;
840
841                 if (rcu_dereference_raw(*rthp) == NULL)
842                         continue;
843                 length = 0;
844                 spin_lock_bh(rt_hash_lock_addr(i));
845                 while ((rth = rcu_dereference_protected(*rthp,
846                                         lockdep_is_held(rt_hash_lock_addr(i)))) != NULL) {
847                         prefetch(rth->dst.rt_next);
848                         if (rt_is_expired(rth)) {
849                                 *rthp = rth->dst.rt_next;
850                                 rt_free(rth);
851                                 continue;
852                         }
853                         if (rth->dst.expires) {
854                                 /* Entry is expired even if it is in use */
855                                 if (time_before_eq(jiffies, rth->dst.expires)) {
856 nofree:
857                                         tmo >>= 1;
858                                         rthp = &rth->dst.rt_next;
859                                         /*
860                                          * We only count entries on
861                                          * a chain with equal hash inputs once
862                                          * so that entries for different QOS
863                                          * levels, and other non-hash input
864                                          * attributes don't unfairly skew
865                                          * the length computation
866                                          */
867                                         length += has_noalias(rt_hash_table[i].chain, rth);
868                                         continue;
869                                 }
870                         } else if (!rt_may_expire(rth, tmo, ip_rt_gc_timeout))
871                                 goto nofree;
872
873                         /* Cleanup aged off entries. */
874                         *rthp = rth->dst.rt_next;
875                         rt_free(rth);
876                 }
877                 spin_unlock_bh(rt_hash_lock_addr(i));
878                 sum += length;
879                 sum2 += length*length;
880         }
881         if (samples) {
882                 unsigned long avg = sum / samples;
883                 unsigned long sd = int_sqrt(sum2 / samples - avg*avg);
884                 rt_chain_length_max = max_t(unsigned long,
885                                         ip_rt_gc_elasticity,
886                                         (avg + 4*sd) >> FRACT_BITS);
887         }
888         rover = i;
889 }
890
891 /*
892  * rt_worker_func() is run in process context.
893  * we call rt_check_expire() to scan part of the hash table
894  */
895 static void rt_worker_func(struct work_struct *work)
896 {
897         rt_check_expire();
898         schedule_delayed_work(&expires_work, ip_rt_gc_interval);
899 }
900
901 /*
902  * Pertubation of rt_genid by a small quantity [1..256]
903  * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
904  * many times (2^24) without giving recent rt_genid.
905  * Jenkins hash is strong enough that litle changes of rt_genid are OK.
906  */
907 static void rt_cache_invalidate(struct net *net)
908 {
909         unsigned char shuffle;
910
911         get_random_bytes(&shuffle, sizeof(shuffle));
912         atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
913 }
914
915 /*
916  * delay < 0  : invalidate cache (fast : entries will be deleted later)
917  * delay >= 0 : invalidate & flush cache (can be long)
918  */
919 void rt_cache_flush(struct net *net, int delay)
920 {
921         rt_cache_invalidate(net);
922         if (delay >= 0)
923                 rt_do_flush(!in_softirq());
924 }
925
926 /* Flush previous cache invalidated entries from the cache */
927 void rt_cache_flush_batch(void)
928 {
929         rt_do_flush(!in_softirq());
930 }
931
932 static void rt_emergency_hash_rebuild(struct net *net)
933 {
934         if (net_ratelimit())
935                 printk(KERN_WARNING "Route hash chain too long!\n");
936         rt_cache_invalidate(net);
937 }
938
939 /*
940    Short description of GC goals.
941
942    We want to build algorithm, which will keep routing cache
943    at some equilibrium point, when number of aged off entries
944    is kept approximately equal to newly generated ones.
945
946    Current expiration strength is variable "expire".
947    We try to adjust it dynamically, so that if networking
948    is idle expires is large enough to keep enough of warm entries,
949    and when load increases it reduces to limit cache size.
950  */
951
952 static int rt_garbage_collect(struct dst_ops *ops)
953 {
954         static unsigned long expire = RT_GC_TIMEOUT;
955         static unsigned long last_gc;
956         static int rover;
957         static int equilibrium;
958         struct rtable *rth;
959         struct rtable __rcu **rthp;
960         unsigned long now = jiffies;
961         int goal;
962         int entries = dst_entries_get_fast(&ipv4_dst_ops);
963
964         /*
965          * Garbage collection is pretty expensive,
966          * do not make it too frequently.
967          */
968
969         RT_CACHE_STAT_INC(gc_total);
970
971         if (now - last_gc < ip_rt_gc_min_interval &&
972             entries < ip_rt_max_size) {
973                 RT_CACHE_STAT_INC(gc_ignored);
974                 goto out;
975         }
976
977         entries = dst_entries_get_slow(&ipv4_dst_ops);
978         /* Calculate number of entries, which we want to expire now. */
979         goal = entries - (ip_rt_gc_elasticity << rt_hash_log);
980         if (goal <= 0) {
981                 if (equilibrium < ipv4_dst_ops.gc_thresh)
982                         equilibrium = ipv4_dst_ops.gc_thresh;
983                 goal = entries - equilibrium;
984                 if (goal > 0) {
985                         equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1);
986                         goal = entries - equilibrium;
987                 }
988         } else {
989                 /* We are in dangerous area. Try to reduce cache really
990                  * aggressively.
991                  */
992                 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1);
993                 equilibrium = entries - goal;
994         }
995
996         if (now - last_gc >= ip_rt_gc_min_interval)
997                 last_gc = now;
998
999         if (goal <= 0) {
1000                 equilibrium += goal;
1001                 goto work_done;
1002         }
1003
1004         do {
1005                 int i, k;
1006
1007                 for (i = rt_hash_mask, k = rover; i >= 0; i--) {
1008                         unsigned long tmo = expire;
1009
1010                         k = (k + 1) & rt_hash_mask;
1011                         rthp = &rt_hash_table[k].chain;
1012                         spin_lock_bh(rt_hash_lock_addr(k));
1013                         while ((rth = rcu_dereference_protected(*rthp,
1014                                         lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) {
1015                                 if (!rt_is_expired(rth) &&
1016                                         !rt_may_expire(rth, tmo, expire)) {
1017                                         tmo >>= 1;
1018                                         rthp = &rth->dst.rt_next;
1019                                         continue;
1020                                 }
1021                                 *rthp = rth->dst.rt_next;
1022                                 rt_free(rth);
1023                                 goal--;
1024                         }
1025                         spin_unlock_bh(rt_hash_lock_addr(k));
1026                         if (goal <= 0)
1027                                 break;
1028                 }
1029                 rover = k;
1030
1031                 if (goal <= 0)
1032                         goto work_done;
1033
1034                 /* Goal is not achieved. We stop process if:
1035
1036                    - if expire reduced to zero. Otherwise, expire is halfed.
1037                    - if table is not full.
1038                    - if we are called from interrupt.
1039                    - jiffies check is just fallback/debug loop breaker.
1040                      We will not spin here for long time in any case.
1041                  */
1042
1043                 RT_CACHE_STAT_INC(gc_goal_miss);
1044
1045                 if (expire == 0)
1046                         break;
1047
1048                 expire >>= 1;
1049 #if RT_CACHE_DEBUG >= 2
1050                 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire,
1051                                 dst_entries_get_fast(&ipv4_dst_ops), goal, i);
1052 #endif
1053
1054                 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1055                         goto out;
1056         } while (!in_softirq() && time_before_eq(jiffies, now));
1057
1058         if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size)
1059                 goto out;
1060         if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size)
1061                 goto out;
1062         if (net_ratelimit())
1063                 printk(KERN_WARNING "dst cache overflow\n");
1064         RT_CACHE_STAT_INC(gc_dst_overflow);
1065         return 1;
1066
1067 work_done:
1068         expire += ip_rt_gc_min_interval;
1069         if (expire > ip_rt_gc_timeout ||
1070             dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh ||
1071             dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh)
1072                 expire = ip_rt_gc_timeout;
1073 #if RT_CACHE_DEBUG >= 2
1074         printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire,
1075                         dst_entries_get_fast(&ipv4_dst_ops), goal, rover);
1076 #endif
1077 out:    return 0;
1078 }
1079
1080 /*
1081  * Returns number of entries in a hash chain that have different hash_inputs
1082  */
1083 static int slow_chain_length(const struct rtable *head)
1084 {
1085         int length = 0;
1086         const struct rtable *rth = head;
1087
1088         while (rth) {
1089                 length += has_noalias(head, rth);
1090                 rth = rcu_dereference_protected(rth->dst.rt_next, 1);
1091         }
1092         return length >> FRACT_BITS;
1093 }
1094
1095 static int rt_intern_hash(unsigned hash, struct rtable *rt,
1096                           struct rtable **rp, struct sk_buff *skb, int ifindex)
1097 {
1098         struct rtable   *rth, *cand;
1099         struct rtable __rcu **rthp, **candp;
1100         unsigned long   now;
1101         u32             min_score;
1102         int             chain_length;
1103         int attempts = !in_softirq();
1104
1105 restart:
1106         chain_length = 0;
1107         min_score = ~(u32)0;
1108         cand = NULL;
1109         candp = NULL;
1110         now = jiffies;
1111
1112         if (!rt_caching(dev_net(rt->dst.dev))) {
1113                 /*
1114                  * If we're not caching, just tell the caller we
1115                  * were successful and don't touch the route.  The
1116                  * caller hold the sole reference to the cache entry, and
1117                  * it will be released when the caller is done with it.
1118                  * If we drop it here, the callers have no way to resolve routes
1119                  * when we're not caching.  Instead, just point *rp at rt, so
1120                  * the caller gets a single use out of the route
1121                  * Note that we do rt_free on this new route entry, so that
1122                  * once its refcount hits zero, we are still able to reap it
1123                  * (Thanks Alexey)
1124                  * Note: To avoid expensive rcu stuff for this uncached dst,
1125                  * we set DST_NOCACHE so that dst_release() can free dst without
1126                  * waiting a grace period.
1127                  */
1128
1129                 rt->dst.flags |= DST_NOCACHE;
1130                 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1131                         int err = arp_bind_neighbour(&rt->dst);
1132                         if (err) {
1133                                 if (net_ratelimit())
1134                                         printk(KERN_WARNING
1135                                             "Neighbour table failure & not caching routes.\n");
1136                                 ip_rt_put(rt);
1137                                 return err;
1138                         }
1139                 }
1140
1141                 goto skip_hashing;
1142         }
1143
1144         rthp = &rt_hash_table[hash].chain;
1145
1146         spin_lock_bh(rt_hash_lock_addr(hash));
1147         while ((rth = rcu_dereference_protected(*rthp,
1148                         lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1149                 if (rt_is_expired(rth)) {
1150                         *rthp = rth->dst.rt_next;
1151                         rt_free(rth);
1152                         continue;
1153                 }
1154                 if (compare_keys(&rth->fl, &rt->fl) && compare_netns(rth, rt)) {
1155                         /* Put it first */
1156                         *rthp = rth->dst.rt_next;
1157                         /*
1158                          * Since lookup is lockfree, the deletion
1159                          * must be visible to another weakly ordered CPU before
1160                          * the insertion at the start of the hash chain.
1161                          */
1162                         rcu_assign_pointer(rth->dst.rt_next,
1163                                            rt_hash_table[hash].chain);
1164                         /*
1165                          * Since lookup is lockfree, the update writes
1166                          * must be ordered for consistency on SMP.
1167                          */
1168                         rcu_assign_pointer(rt_hash_table[hash].chain, rth);
1169
1170                         dst_use(&rth->dst, now);
1171                         spin_unlock_bh(rt_hash_lock_addr(hash));
1172
1173                         rt_drop(rt);
1174                         if (rp)
1175                                 *rp = rth;
1176                         else
1177                                 skb_dst_set(skb, &rth->dst);
1178                         return 0;
1179                 }
1180
1181                 if (!atomic_read(&rth->dst.__refcnt)) {
1182                         u32 score = rt_score(rth);
1183
1184                         if (score <= min_score) {
1185                                 cand = rth;
1186                                 candp = rthp;
1187                                 min_score = score;
1188                         }
1189                 }
1190
1191                 chain_length++;
1192
1193                 rthp = &rth->dst.rt_next;
1194         }
1195
1196         if (cand) {
1197                 /* ip_rt_gc_elasticity used to be average length of chain
1198                  * length, when exceeded gc becomes really aggressive.
1199                  *
1200                  * The second limit is less certain. At the moment it allows
1201                  * only 2 entries per bucket. We will see.
1202                  */
1203                 if (chain_length > ip_rt_gc_elasticity) {
1204                         *candp = cand->dst.rt_next;
1205                         rt_free(cand);
1206                 }
1207         } else {
1208                 if (chain_length > rt_chain_length_max &&
1209                     slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) {
1210                         struct net *net = dev_net(rt->dst.dev);
1211                         int num = ++net->ipv4.current_rt_cache_rebuild_count;
1212                         if (!rt_caching(net)) {
1213                                 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n",
1214                                         rt->dst.dev->name, num);
1215                         }
1216                         rt_emergency_hash_rebuild(net);
1217                         spin_unlock_bh(rt_hash_lock_addr(hash));
1218
1219                         hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1220                                         ifindex, rt_genid(net));
1221                         goto restart;
1222                 }
1223         }
1224
1225         /* Try to bind route to arp only if it is output
1226            route or unicast forwarding path.
1227          */
1228         if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) {
1229                 int err = arp_bind_neighbour(&rt->dst);
1230                 if (err) {
1231                         spin_unlock_bh(rt_hash_lock_addr(hash));
1232
1233                         if (err != -ENOBUFS) {
1234                                 rt_drop(rt);
1235                                 return err;
1236                         }
1237
1238                         /* Neighbour tables are full and nothing
1239                            can be released. Try to shrink route cache,
1240                            it is most likely it holds some neighbour records.
1241                          */
1242                         if (attempts-- > 0) {
1243                                 int saved_elasticity = ip_rt_gc_elasticity;
1244                                 int saved_int = ip_rt_gc_min_interval;
1245                                 ip_rt_gc_elasticity     = 1;
1246                                 ip_rt_gc_min_interval   = 0;
1247                                 rt_garbage_collect(&ipv4_dst_ops);
1248                                 ip_rt_gc_min_interval   = saved_int;
1249                                 ip_rt_gc_elasticity     = saved_elasticity;
1250                                 goto restart;
1251                         }
1252
1253                         if (net_ratelimit())
1254                                 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n");
1255                         rt_drop(rt);
1256                         return -ENOBUFS;
1257                 }
1258         }
1259
1260         rt->dst.rt_next = rt_hash_table[hash].chain;
1261
1262 #if RT_CACHE_DEBUG >= 2
1263         if (rt->dst.rt_next) {
1264                 struct rtable *trt;
1265                 printk(KERN_DEBUG "rt_cache @%02x: %pI4",
1266                        hash, &rt->rt_dst);
1267                 for (trt = rt->dst.rt_next; trt; trt = trt->dst.rt_next)
1268                         printk(" . %pI4", &trt->rt_dst);
1269                 printk("\n");
1270         }
1271 #endif
1272         /*
1273          * Since lookup is lockfree, we must make sure
1274          * previous writes to rt are comitted to memory
1275          * before making rt visible to other CPUS.
1276          */
1277         rcu_assign_pointer(rt_hash_table[hash].chain, rt);
1278
1279         spin_unlock_bh(rt_hash_lock_addr(hash));
1280
1281 skip_hashing:
1282         if (rp)
1283                 *rp = rt;
1284         else
1285                 skb_dst_set(skb, &rt->dst);
1286         return 0;
1287 }
1288
1289 void rt_bind_peer(struct rtable *rt, int create)
1290 {
1291         struct inet_peer *peer;
1292
1293         peer = inet_getpeer_v4(rt->rt_dst, create);
1294
1295         if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL)
1296                 inet_putpeer(peer);
1297 }
1298
1299 /*
1300  * Peer allocation may fail only in serious out-of-memory conditions.  However
1301  * we still can generate some output.
1302  * Random ID selection looks a bit dangerous because we have no chances to
1303  * select ID being unique in a reasonable period of time.
1304  * But broken packet identifier may be better than no packet at all.
1305  */
1306 static void ip_select_fb_ident(struct iphdr *iph)
1307 {
1308         static DEFINE_SPINLOCK(ip_fb_id_lock);
1309         static u32 ip_fallback_id;
1310         u32 salt;
1311
1312         spin_lock_bh(&ip_fb_id_lock);
1313         salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
1314         iph->id = htons(salt & 0xFFFF);
1315         ip_fallback_id = salt;
1316         spin_unlock_bh(&ip_fb_id_lock);
1317 }
1318
1319 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
1320 {
1321         struct rtable *rt = (struct rtable *) dst;
1322
1323         if (rt) {
1324                 if (rt->peer == NULL)
1325                         rt_bind_peer(rt, 1);
1326
1327                 /* If peer is attached to destination, it is never detached,
1328                    so that we need not to grab a lock to dereference it.
1329                  */
1330                 if (rt->peer) {
1331                         iph->id = htons(inet_getid(rt->peer, more));
1332                         return;
1333                 }
1334         } else
1335                 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n",
1336                        __builtin_return_address(0));
1337
1338         ip_select_fb_ident(iph);
1339 }
1340 EXPORT_SYMBOL(__ip_select_ident);
1341
1342 static void rt_del(unsigned hash, struct rtable *rt)
1343 {
1344         struct rtable __rcu **rthp;
1345         struct rtable *aux;
1346
1347         rthp = &rt_hash_table[hash].chain;
1348         spin_lock_bh(rt_hash_lock_addr(hash));
1349         ip_rt_put(rt);
1350         while ((aux = rcu_dereference_protected(*rthp,
1351                         lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) {
1352                 if (aux == rt || rt_is_expired(aux)) {
1353                         *rthp = aux->dst.rt_next;
1354                         rt_free(aux);
1355                         continue;
1356                 }
1357                 rthp = &aux->dst.rt_next;
1358         }
1359         spin_unlock_bh(rt_hash_lock_addr(hash));
1360 }
1361
1362 /* called in rcu_read_lock() section */
1363 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw,
1364                     __be32 saddr, struct net_device *dev)
1365 {
1366         int i, k;
1367         struct in_device *in_dev = __in_dev_get_rcu(dev);
1368         struct rtable *rth;
1369         struct rtable __rcu **rthp;
1370         __be32  skeys[2] = { saddr, 0 };
1371         int  ikeys[2] = { dev->ifindex, 0 };
1372         struct netevent_redirect netevent;
1373         struct net *net;
1374
1375         if (!in_dev)
1376                 return;
1377
1378         net = dev_net(dev);
1379         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
1380             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
1381             ipv4_is_zeronet(new_gw))
1382                 goto reject_redirect;
1383
1384         if (!rt_caching(net))
1385                 goto reject_redirect;
1386
1387         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
1388                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
1389                         goto reject_redirect;
1390                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
1391                         goto reject_redirect;
1392         } else {
1393                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
1394                         goto reject_redirect;
1395         }
1396
1397         for (i = 0; i < 2; i++) {
1398                 for (k = 0; k < 2; k++) {
1399                         unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1400                                                 rt_genid(net));
1401
1402                         rthp = &rt_hash_table[hash].chain;
1403
1404                         while ((rth = rcu_dereference(*rthp)) != NULL) {
1405                                 struct rtable *rt;
1406
1407                                 if (rth->fl.fl4_dst != daddr ||
1408                                     rth->fl.fl4_src != skeys[i] ||
1409                                     rth->fl.oif != ikeys[k] ||
1410                                     rt_is_input_route(rth) ||
1411                                     rt_is_expired(rth) ||
1412                                     !net_eq(dev_net(rth->dst.dev), net)) {
1413                                         rthp = &rth->dst.rt_next;
1414                                         continue;
1415                                 }
1416
1417                                 if (rth->rt_dst != daddr ||
1418                                     rth->rt_src != saddr ||
1419                                     rth->dst.error ||
1420                                     rth->rt_gateway != old_gw ||
1421                                     rth->dst.dev != dev)
1422                                         break;
1423
1424                                 dst_hold(&rth->dst);
1425
1426                                 rt = dst_alloc(&ipv4_dst_ops);
1427                                 if (rt == NULL) {
1428                                         ip_rt_put(rth);
1429                                         return;
1430                                 }
1431
1432                                 /* Copy all the information. */
1433                                 *rt = *rth;
1434                                 rt->dst.__use           = 1;
1435                                 atomic_set(&rt->dst.__refcnt, 1);
1436                                 rt->dst.child           = NULL;
1437                                 if (rt->dst.dev)
1438                                         dev_hold(rt->dst.dev);
1439                                 rt->dst.obsolete        = -1;
1440                                 rt->dst.lastuse = jiffies;
1441                                 rt->dst.path            = &rt->dst;
1442                                 rt->dst.neighbour       = NULL;
1443                                 rt->dst.hh              = NULL;
1444 #ifdef CONFIG_XFRM
1445                                 rt->dst.xfrm            = NULL;
1446 #endif
1447                                 rt->rt_genid            = rt_genid(net);
1448                                 rt->rt_flags            |= RTCF_REDIRECTED;
1449
1450                                 /* Gateway is different ... */
1451                                 rt->rt_gateway          = new_gw;
1452
1453                                 /* Redirect received -> path was valid */
1454                                 dst_confirm(&rth->dst);
1455
1456                                 if (rt->peer)
1457                                         atomic_inc(&rt->peer->refcnt);
1458
1459                                 if (arp_bind_neighbour(&rt->dst) ||
1460                                     !(rt->dst.neighbour->nud_state &
1461                                             NUD_VALID)) {
1462                                         if (rt->dst.neighbour)
1463                                                 neigh_event_send(rt->dst.neighbour, NULL);
1464                                         ip_rt_put(rth);
1465                                         rt_drop(rt);
1466                                         goto do_next;
1467                                 }
1468
1469                                 netevent.old = &rth->dst;
1470                                 netevent.new = &rt->dst;
1471                                 call_netevent_notifiers(NETEVENT_REDIRECT,
1472                                                         &netevent);
1473
1474                                 rt_del(hash, rth);
1475                                 if (!rt_intern_hash(hash, rt, &rt, NULL, rt->fl.oif))
1476                                         ip_rt_put(rt);
1477                                 goto do_next;
1478                         }
1479                 do_next:
1480                         ;
1481                 }
1482         }
1483         return;
1484
1485 reject_redirect:
1486 #ifdef CONFIG_IP_ROUTE_VERBOSE
1487         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
1488                 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n"
1489                         "  Advised path = %pI4 -> %pI4\n",
1490                        &old_gw, dev->name, &new_gw,
1491                        &saddr, &daddr);
1492 #endif
1493         ;
1494 }
1495
1496 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
1497 {
1498         struct rtable *rt = (struct rtable *)dst;
1499         struct dst_entry *ret = dst;
1500
1501         if (rt) {
1502                 if (dst->obsolete > 0) {
1503                         ip_rt_put(rt);
1504                         ret = NULL;
1505                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
1506                            (rt->dst.expires &&
1507                             time_after_eq(jiffies, rt->dst.expires))) {
1508                         unsigned hash = rt_hash(rt->fl.fl4_dst, rt->fl.fl4_src,
1509                                                 rt->fl.oif,
1510                                                 rt_genid(dev_net(dst->dev)));
1511 #if RT_CACHE_DEBUG >= 1
1512                         printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n",
1513                                 &rt->rt_dst, rt->fl.fl4_tos);
1514 #endif
1515                         rt_del(hash, rt);
1516                         ret = NULL;
1517                 }
1518         }
1519         return ret;
1520 }
1521
1522 /*
1523  * Algorithm:
1524  *      1. The first ip_rt_redirect_number redirects are sent
1525  *         with exponential backoff, then we stop sending them at all,
1526  *         assuming that the host ignores our redirects.
1527  *      2. If we did not see packets requiring redirects
1528  *         during ip_rt_redirect_silence, we assume that the host
1529  *         forgot redirected route and start to send redirects again.
1530  *
1531  * This algorithm is much cheaper and more intelligent than dumb load limiting
1532  * in icmp.c.
1533  *
1534  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
1535  * and "frag. need" (breaks PMTU discovery) in icmp.c.
1536  */
1537
1538 void ip_rt_send_redirect(struct sk_buff *skb)
1539 {
1540         struct rtable *rt = skb_rtable(skb);
1541         struct in_device *in_dev;
1542         int log_martians;
1543
1544         rcu_read_lock();
1545         in_dev = __in_dev_get_rcu(rt->dst.dev);
1546         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
1547                 rcu_read_unlock();
1548                 return;
1549         }
1550         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
1551         rcu_read_unlock();
1552
1553         /* No redirected packets during ip_rt_redirect_silence;
1554          * reset the algorithm.
1555          */
1556         if (time_after(jiffies, rt->dst.rate_last + ip_rt_redirect_silence))
1557                 rt->dst.rate_tokens = 0;
1558
1559         /* Too many ignored redirects; do not send anything
1560          * set dst.rate_last to the last seen redirected packet.
1561          */
1562         if (rt->dst.rate_tokens >= ip_rt_redirect_number) {
1563                 rt->dst.rate_last = jiffies;
1564                 return;
1565         }
1566
1567         /* Check for load limit; set rate_last to the latest sent
1568          * redirect.
1569          */
1570         if (rt->dst.rate_tokens == 0 ||
1571             time_after(jiffies,
1572                        (rt->dst.rate_last +
1573                         (ip_rt_redirect_load << rt->dst.rate_tokens)))) {
1574                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
1575                 rt->dst.rate_last = jiffies;
1576                 ++rt->dst.rate_tokens;
1577 #ifdef CONFIG_IP_ROUTE_VERBOSE
1578                 if (log_martians &&
1579                     rt->dst.rate_tokens == ip_rt_redirect_number &&
1580                     net_ratelimit())
1581                         printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
1582                                 &rt->rt_src, rt->rt_iif,
1583                                 &rt->rt_dst, &rt->rt_gateway);
1584 #endif
1585         }
1586 }
1587
1588 static int ip_error(struct sk_buff *skb)
1589 {
1590         struct rtable *rt = skb_rtable(skb);
1591         unsigned long now;
1592         int code;
1593
1594         switch (rt->dst.error) {
1595                 case EINVAL:
1596                 default:
1597                         goto out;
1598                 case EHOSTUNREACH:
1599                         code = ICMP_HOST_UNREACH;
1600                         break;
1601                 case ENETUNREACH:
1602                         code = ICMP_NET_UNREACH;
1603                         IP_INC_STATS_BH(dev_net(rt->dst.dev),
1604                                         IPSTATS_MIB_INNOROUTES);
1605                         break;
1606                 case EACCES:
1607                         code = ICMP_PKT_FILTERED;
1608                         break;
1609         }
1610
1611         now = jiffies;
1612         rt->dst.rate_tokens += now - rt->dst.rate_last;
1613         if (rt->dst.rate_tokens > ip_rt_error_burst)
1614                 rt->dst.rate_tokens = ip_rt_error_burst;
1615         rt->dst.rate_last = now;
1616         if (rt->dst.rate_tokens >= ip_rt_error_cost) {
1617                 rt->dst.rate_tokens -= ip_rt_error_cost;
1618                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1619         }
1620
1621 out:    kfree_skb(skb);
1622         return 0;
1623 }
1624
1625 /*
1626  *      The last two values are not from the RFC but
1627  *      are needed for AMPRnet AX.25 paths.
1628  */
1629
1630 static const unsigned short mtu_plateau[] =
1631 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 };
1632
1633 static inline unsigned short guess_mtu(unsigned short old_mtu)
1634 {
1635         int i;
1636
1637         for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++)
1638                 if (old_mtu > mtu_plateau[i])
1639                         return mtu_plateau[i];
1640         return 68;
1641 }
1642
1643 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph,
1644                                  unsigned short new_mtu,
1645                                  struct net_device *dev)
1646 {
1647         int i, k;
1648         unsigned short old_mtu = ntohs(iph->tot_len);
1649         struct rtable *rth;
1650         int  ikeys[2] = { dev->ifindex, 0 };
1651         __be32  skeys[2] = { iph->saddr, 0, };
1652         __be32  daddr = iph->daddr;
1653         unsigned short est_mtu = 0;
1654
1655         for (k = 0; k < 2; k++) {
1656                 for (i = 0; i < 2; i++) {
1657                         unsigned hash = rt_hash(daddr, skeys[i], ikeys[k],
1658                                                 rt_genid(net));
1659
1660                         rcu_read_lock();
1661                         for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
1662                              rth = rcu_dereference(rth->dst.rt_next)) {
1663                                 unsigned short mtu = new_mtu;
1664
1665                                 if (rth->fl.fl4_dst != daddr ||
1666                                     rth->fl.fl4_src != skeys[i] ||
1667                                     rth->rt_dst != daddr ||
1668                                     rth->rt_src != iph->saddr ||
1669                                     rth->fl.oif != ikeys[k] ||
1670                                     rt_is_input_route(rth) ||
1671                                     dst_metric_locked(&rth->dst, RTAX_MTU) ||
1672                                     !net_eq(dev_net(rth->dst.dev), net) ||
1673                                     rt_is_expired(rth))
1674                                         continue;
1675
1676                                 if (new_mtu < 68 || new_mtu >= old_mtu) {
1677
1678                                         /* BSD 4.2 compatibility hack :-( */
1679                                         if (mtu == 0 &&
1680                                             old_mtu >= dst_mtu(&rth->dst) &&
1681                                             old_mtu >= 68 + (iph->ihl << 2))
1682                                                 old_mtu -= iph->ihl << 2;
1683
1684                                         mtu = guess_mtu(old_mtu);
1685                                 }
1686                                 if (mtu <= dst_mtu(&rth->dst)) {
1687                                         if (mtu < dst_mtu(&rth->dst)) {
1688                                                 dst_confirm(&rth->dst);
1689                                                 if (mtu < ip_rt_min_pmtu) {
1690                                                         u32 lock = dst_metric(&rth->dst,
1691                                                                               RTAX_LOCK);
1692                                                         mtu = ip_rt_min_pmtu;
1693                                                         lock |= (1 << RTAX_MTU);
1694                                                         dst_metric_set(&rth->dst, RTAX_LOCK,
1695                                                                        lock);
1696                                                 }
1697                                                 dst_metric_set(&rth->dst, RTAX_MTU, mtu);
1698                                                 dst_set_expires(&rth->dst,
1699                                                         ip_rt_mtu_expires);
1700                                         }
1701                                         est_mtu = mtu;
1702                                 }
1703                         }
1704                         rcu_read_unlock();
1705                 }
1706         }
1707         return est_mtu ? : new_mtu;
1708 }
1709
1710 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu)
1711 {
1712         if (dst_mtu(dst) > mtu && mtu >= 68 &&
1713             !(dst_metric_locked(dst, RTAX_MTU))) {
1714                 if (mtu < ip_rt_min_pmtu) {
1715                         u32 lock = dst_metric(dst, RTAX_LOCK);
1716                         mtu = ip_rt_min_pmtu;
1717                         dst_metric_set(dst, RTAX_LOCK, lock | (1 << RTAX_MTU));
1718                 }
1719                 dst_metric_set(dst, RTAX_MTU, mtu);
1720                 dst_set_expires(dst, ip_rt_mtu_expires);
1721                 call_netevent_notifiers(NETEVENT_PMTU_UPDATE, dst);
1722         }
1723 }
1724
1725 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1726 {
1727         if (rt_is_expired((struct rtable *)dst))
1728                 return NULL;
1729         return dst;
1730 }
1731
1732 static void ipv4_dst_destroy(struct dst_entry *dst)
1733 {
1734         struct rtable *rt = (struct rtable *) dst;
1735         struct inet_peer *peer = rt->peer;
1736
1737         if (peer) {
1738                 rt->peer = NULL;
1739                 inet_putpeer(peer);
1740         }
1741 }
1742
1743
1744 static void ipv4_link_failure(struct sk_buff *skb)
1745 {
1746         struct rtable *rt;
1747
1748         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1749
1750         rt = skb_rtable(skb);
1751         if (rt)
1752                 dst_set_expires(&rt->dst, 0);
1753 }
1754
1755 static int ip_rt_bug(struct sk_buff *skb)
1756 {
1757         printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n",
1758                 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1759                 skb->dev ? skb->dev->name : "?");
1760         kfree_skb(skb);
1761         return 0;
1762 }
1763
1764 /*
1765    We do not cache source address of outgoing interface,
1766    because it is used only by IP RR, TS and SRR options,
1767    so that it out of fast path.
1768
1769    BTW remember: "addr" is allowed to be not aligned
1770    in IP options!
1771  */
1772
1773 void ip_rt_get_source(u8 *addr, struct rtable *rt)
1774 {
1775         __be32 src;
1776         struct fib_result res;
1777
1778         if (rt_is_output_route(rt))
1779                 src = rt->rt_src;
1780         else {
1781                 rcu_read_lock();
1782                 if (fib_lookup(dev_net(rt->dst.dev), &rt->fl, &res) == 0)
1783                         src = FIB_RES_PREFSRC(res);
1784                 else
1785                         src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1786                                         RT_SCOPE_UNIVERSE);
1787                 rcu_read_unlock();
1788         }
1789         memcpy(addr, &src, 4);
1790 }
1791
1792 #ifdef CONFIG_NET_CLS_ROUTE
1793 static void set_class_tag(struct rtable *rt, u32 tag)
1794 {
1795         if (!(rt->dst.tclassid & 0xFFFF))
1796                 rt->dst.tclassid |= tag & 0xFFFF;
1797         if (!(rt->dst.tclassid & 0xFFFF0000))
1798                 rt->dst.tclassid |= tag & 0xFFFF0000;
1799 }
1800 #endif
1801
1802 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1803 {
1804         unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1805
1806         if (advmss == 0) {
1807                 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1808                                ip_rt_min_advmss);
1809                 if (advmss > 65535 - 40)
1810                         advmss = 65535 - 40;
1811         }
1812         return advmss;
1813 }
1814
1815 static void rt_set_nexthop(struct rtable *rt, struct fib_result *res, u32 itag)
1816 {
1817         struct dst_entry *dst = &rt->dst;
1818         struct fib_info *fi = res->fi;
1819
1820         if (fi) {
1821                 if (FIB_RES_GW(*res) &&
1822                     FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
1823                         rt->rt_gateway = FIB_RES_GW(*res);
1824                 dst_import_metrics(dst, fi->fib_metrics);
1825                 if (fi->fib_mtu == 0) {
1826                         dst_metric_set(dst, RTAX_MTU, dst->dev->mtu);
1827                         if (dst_metric_locked(dst, RTAX_MTU) &&
1828                             rt->rt_gateway != rt->rt_dst &&
1829                             dst->dev->mtu > 576)
1830                                 dst_metric_set(dst, RTAX_MTU, 576);
1831                 }
1832 #ifdef CONFIG_NET_CLS_ROUTE
1833                 dst->tclassid = FIB_RES_NH(*res).nh_tclassid;
1834 #endif
1835         } else
1836                 dst_metric_set(dst, RTAX_MTU, dst->dev->mtu);
1837
1838         if (dst_mtu(dst) > IP_MAX_MTU)
1839                 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU);
1840         if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40)
1841                 dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40);
1842
1843 #ifdef CONFIG_NET_CLS_ROUTE
1844 #ifdef CONFIG_IP_MULTIPLE_TABLES
1845         set_class_tag(rt, fib_rules_tclass(res));
1846 #endif
1847         set_class_tag(rt, itag);
1848 #endif
1849         rt->rt_type = res->type;
1850 }
1851
1852 /* called in rcu_read_lock() section */
1853 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1854                                 u8 tos, struct net_device *dev, int our)
1855 {
1856         unsigned int hash;
1857         struct rtable *rth;
1858         __be32 spec_dst;
1859         struct in_device *in_dev = __in_dev_get_rcu(dev);
1860         u32 itag = 0;
1861         int err;
1862
1863         /* Primary sanity checks. */
1864
1865         if (in_dev == NULL)
1866                 return -EINVAL;
1867
1868         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1869             ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP))
1870                 goto e_inval;
1871
1872         if (ipv4_is_zeronet(saddr)) {
1873                 if (!ipv4_is_local_multicast(daddr))
1874                         goto e_inval;
1875                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1876         } else {
1877                 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
1878                                           &itag, 0);
1879                 if (err < 0)
1880                         goto e_err;
1881         }
1882         rth = dst_alloc(&ipv4_dst_ops);
1883         if (!rth)
1884                 goto e_nobufs;
1885
1886         rth->dst.output = ip_rt_bug;
1887         rth->dst.obsolete = -1;
1888
1889         atomic_set(&rth->dst.__refcnt, 1);
1890         rth->dst.flags= DST_HOST;
1891         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
1892                 rth->dst.flags |= DST_NOPOLICY;
1893         rth->fl.fl4_dst = daddr;
1894         rth->rt_dst     = daddr;
1895         rth->fl.fl4_tos = tos;
1896         rth->fl.mark    = skb->mark;
1897         rth->fl.fl4_src = saddr;
1898         rth->rt_src     = saddr;
1899 #ifdef CONFIG_NET_CLS_ROUTE
1900         rth->dst.tclassid = itag;
1901 #endif
1902         rth->rt_iif     =
1903         rth->fl.iif     = dev->ifindex;
1904         rth->dst.dev    = init_net.loopback_dev;
1905         dev_hold(rth->dst.dev);
1906         rth->fl.oif     = 0;
1907         rth->rt_gateway = daddr;
1908         rth->rt_spec_dst= spec_dst;
1909         rth->rt_genid   = rt_genid(dev_net(dev));
1910         rth->rt_flags   = RTCF_MULTICAST;
1911         rth->rt_type    = RTN_MULTICAST;
1912         if (our) {
1913                 rth->dst.input= ip_local_deliver;
1914                 rth->rt_flags |= RTCF_LOCAL;
1915         }
1916
1917 #ifdef CONFIG_IP_MROUTE
1918         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1919                 rth->dst.input = ip_mr_input;
1920 #endif
1921         RT_CACHE_STAT_INC(in_slow_mc);
1922
1923         hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev)));
1924         return rt_intern_hash(hash, rth, NULL, skb, dev->ifindex);
1925
1926 e_nobufs:
1927         return -ENOBUFS;
1928 e_inval:
1929         return -EINVAL;
1930 e_err:
1931         return err;
1932 }
1933
1934
1935 static void ip_handle_martian_source(struct net_device *dev,
1936                                      struct in_device *in_dev,
1937                                      struct sk_buff *skb,
1938                                      __be32 daddr,
1939                                      __be32 saddr)
1940 {
1941         RT_CACHE_STAT_INC(in_martian_src);
1942 #ifdef CONFIG_IP_ROUTE_VERBOSE
1943         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1944                 /*
1945                  *      RFC1812 recommendation, if source is martian,
1946                  *      the only hint is MAC header.
1947                  */
1948                 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n",
1949                         &daddr, &saddr, dev->name);
1950                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1951                         int i;
1952                         const unsigned char *p = skb_mac_header(skb);
1953                         printk(KERN_WARNING "ll header: ");
1954                         for (i = 0; i < dev->hard_header_len; i++, p++) {
1955                                 printk("%02x", *p);
1956                                 if (i < (dev->hard_header_len - 1))
1957                                         printk(":");
1958                         }
1959                         printk("\n");
1960                 }
1961         }
1962 #endif
1963 }
1964
1965 /* called in rcu_read_lock() section */
1966 static int __mkroute_input(struct sk_buff *skb,
1967                            struct fib_result *res,
1968                            struct in_device *in_dev,
1969                            __be32 daddr, __be32 saddr, u32 tos,
1970                            struct rtable **result)
1971 {
1972         struct rtable *rth;
1973         int err;
1974         struct in_device *out_dev;
1975         unsigned int flags = 0;
1976         __be32 spec_dst;
1977         u32 itag;
1978
1979         /* get a working reference to the output device */
1980         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1981         if (out_dev == NULL) {
1982                 if (net_ratelimit())
1983                         printk(KERN_CRIT "Bug in ip_route_input" \
1984                                "_slow(). Please, report\n");
1985                 return -EINVAL;
1986         }
1987
1988
1989         err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
1990                                   in_dev->dev, &spec_dst, &itag, skb->mark);
1991         if (err < 0) {
1992                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1993                                          saddr);
1994
1995                 goto cleanup;
1996         }
1997
1998         if (err)
1999                 flags |= RTCF_DIRECTSRC;
2000
2001         if (out_dev == in_dev && err &&
2002             (IN_DEV_SHARED_MEDIA(out_dev) ||
2003              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
2004                 flags |= RTCF_DOREDIRECT;
2005
2006         if (skb->protocol != htons(ETH_P_IP)) {
2007                 /* Not IP (i.e. ARP). Do not create route, if it is
2008                  * invalid for proxy arp. DNAT routes are always valid.
2009                  *
2010                  * Proxy arp feature have been extended to allow, ARP
2011                  * replies back to the same interface, to support
2012                  * Private VLAN switch technologies. See arp.c.
2013                  */
2014                 if (out_dev == in_dev &&
2015                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
2016                         err = -EINVAL;
2017                         goto cleanup;
2018                 }
2019         }
2020
2021
2022         rth = dst_alloc(&ipv4_dst_ops);
2023         if (!rth) {
2024                 err = -ENOBUFS;
2025                 goto cleanup;
2026         }
2027
2028         atomic_set(&rth->dst.__refcnt, 1);
2029         rth->dst.flags= DST_HOST;
2030         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2031                 rth->dst.flags |= DST_NOPOLICY;
2032         if (IN_DEV_CONF_GET(out_dev, NOXFRM))
2033                 rth->dst.flags |= DST_NOXFRM;
2034         rth->fl.fl4_dst = daddr;
2035         rth->rt_dst     = daddr;
2036         rth->fl.fl4_tos = tos;
2037         rth->fl.mark    = skb->mark;
2038         rth->fl.fl4_src = saddr;
2039         rth->rt_src     = saddr;
2040         rth->rt_gateway = daddr;
2041         rth->rt_iif     =
2042                 rth->fl.iif     = in_dev->dev->ifindex;
2043         rth->dst.dev    = (out_dev)->dev;
2044         dev_hold(rth->dst.dev);
2045         rth->fl.oif     = 0;
2046         rth->rt_spec_dst= spec_dst;
2047
2048         rth->dst.obsolete = -1;
2049         rth->dst.input = ip_forward;
2050         rth->dst.output = ip_output;
2051         rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
2052
2053         rt_set_nexthop(rth, res, itag);
2054
2055         rth->rt_flags = flags;
2056
2057         *result = rth;
2058         err = 0;
2059  cleanup:
2060         return err;
2061 }
2062
2063 static int ip_mkroute_input(struct sk_buff *skb,
2064                             struct fib_result *res,
2065                             const struct flowi *fl,
2066                             struct in_device *in_dev,
2067                             __be32 daddr, __be32 saddr, u32 tos)
2068 {
2069         struct rtable* rth = NULL;
2070         int err;
2071         unsigned hash;
2072
2073 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2074         if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
2075                 fib_select_multipath(fl, res);
2076 #endif
2077
2078         /* create a routing cache entry */
2079         err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
2080         if (err)
2081                 return err;
2082
2083         /* put it into the cache */
2084         hash = rt_hash(daddr, saddr, fl->iif,
2085                        rt_genid(dev_net(rth->dst.dev)));
2086         return rt_intern_hash(hash, rth, NULL, skb, fl->iif);
2087 }
2088
2089 /*
2090  *      NOTE. We drop all the packets that has local source
2091  *      addresses, because every properly looped back packet
2092  *      must have correct destination already attached by output routine.
2093  *
2094  *      Such approach solves two big problems:
2095  *      1. Not simplex devices are handled properly.
2096  *      2. IP spoofing attempts are filtered with 100% of guarantee.
2097  *      called with rcu_read_lock()
2098  */
2099
2100 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2101                                u8 tos, struct net_device *dev)
2102 {
2103         struct fib_result res;
2104         struct in_device *in_dev = __in_dev_get_rcu(dev);
2105         struct flowi fl = { .fl4_dst    = daddr,
2106                             .fl4_src    = saddr,
2107                             .fl4_tos    = tos,
2108                             .fl4_scope  = RT_SCOPE_UNIVERSE,
2109                             .mark = skb->mark,
2110                             .iif = dev->ifindex };
2111         unsigned        flags = 0;
2112         u32             itag = 0;
2113         struct rtable * rth;
2114         unsigned        hash;
2115         __be32          spec_dst;
2116         int             err = -EINVAL;
2117         struct net    * net = dev_net(dev);
2118
2119         /* IP on this device is disabled. */
2120
2121         if (!in_dev)
2122                 goto out;
2123
2124         /* Check for the most weird martians, which can be not detected
2125            by fib_lookup.
2126          */
2127
2128         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
2129             ipv4_is_loopback(saddr))
2130                 goto martian_source;
2131
2132         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
2133                 goto brd_input;
2134
2135         /* Accept zero addresses only to limited broadcast;
2136          * I even do not know to fix it or not. Waiting for complains :-)
2137          */
2138         if (ipv4_is_zeronet(saddr))
2139                 goto martian_source;
2140
2141         if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr))
2142                 goto martian_destination;
2143
2144         /*
2145          *      Now we are ready to route packet.
2146          */
2147         err = fib_lookup(net, &fl, &res);
2148         if (err != 0) {
2149                 if (!IN_DEV_FORWARD(in_dev))
2150                         goto e_hostunreach;
2151                 goto no_route;
2152         }
2153
2154         RT_CACHE_STAT_INC(in_slow_tot);
2155
2156         if (res.type == RTN_BROADCAST)
2157                 goto brd_input;
2158
2159         if (res.type == RTN_LOCAL) {
2160                 err = fib_validate_source(saddr, daddr, tos,
2161                                           net->loopback_dev->ifindex,
2162                                           dev, &spec_dst, &itag, skb->mark);
2163                 if (err < 0)
2164                         goto martian_source_keep_err;
2165                 if (err)
2166                         flags |= RTCF_DIRECTSRC;
2167                 spec_dst = daddr;
2168                 goto local_input;
2169         }
2170
2171         if (!IN_DEV_FORWARD(in_dev))
2172                 goto e_hostunreach;
2173         if (res.type != RTN_UNICAST)
2174                 goto martian_destination;
2175
2176         err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
2177 out:    return err;
2178
2179 brd_input:
2180         if (skb->protocol != htons(ETH_P_IP))
2181                 goto e_inval;
2182
2183         if (ipv4_is_zeronet(saddr))
2184                 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
2185         else {
2186                 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,
2187                                           &itag, skb->mark);
2188                 if (err < 0)
2189                         goto martian_source_keep_err;
2190                 if (err)
2191                         flags |= RTCF_DIRECTSRC;
2192         }
2193         flags |= RTCF_BROADCAST;
2194         res.type = RTN_BROADCAST;
2195         RT_CACHE_STAT_INC(in_brd);
2196
2197 local_input:
2198         rth = dst_alloc(&ipv4_dst_ops);
2199         if (!rth)
2200                 goto e_nobufs;
2201
2202         rth->dst.output= ip_rt_bug;
2203         rth->dst.obsolete = -1;
2204         rth->rt_genid = rt_genid(net);
2205
2206         atomic_set(&rth->dst.__refcnt, 1);
2207         rth->dst.flags= DST_HOST;
2208         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2209                 rth->dst.flags |= DST_NOPOLICY;
2210         rth->fl.fl4_dst = daddr;
2211         rth->rt_dst     = daddr;
2212         rth->fl.fl4_tos = tos;
2213         rth->fl.mark    = skb->mark;
2214         rth->fl.fl4_src = saddr;
2215         rth->rt_src     = saddr;
2216 #ifdef CONFIG_NET_CLS_ROUTE
2217         rth->dst.tclassid = itag;
2218 #endif
2219         rth->rt_iif     =
2220         rth->fl.iif     = dev->ifindex;
2221         rth->dst.dev    = net->loopback_dev;
2222         dev_hold(rth->dst.dev);
2223         rth->rt_gateway = daddr;
2224         rth->rt_spec_dst= spec_dst;
2225         rth->dst.input= ip_local_deliver;
2226         rth->rt_flags   = flags|RTCF_LOCAL;
2227         if (res.type == RTN_UNREACHABLE) {
2228                 rth->dst.input= ip_error;
2229                 rth->dst.error= -err;
2230                 rth->rt_flags   &= ~RTCF_LOCAL;
2231         }
2232         rth->rt_type    = res.type;
2233         hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));
2234         err = rt_intern_hash(hash, rth, NULL, skb, fl.iif);
2235         goto out;
2236
2237 no_route:
2238         RT_CACHE_STAT_INC(in_no_route);
2239         spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
2240         res.type = RTN_UNREACHABLE;
2241         if (err == -ESRCH)
2242                 err = -ENETUNREACH;
2243         goto local_input;
2244
2245         /*
2246          *      Do not cache martian addresses: they should be logged (RFC1812)
2247          */
2248 martian_destination:
2249         RT_CACHE_STAT_INC(in_martian_dst);
2250 #ifdef CONFIG_IP_ROUTE_VERBOSE
2251         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
2252                 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
2253                         &daddr, &saddr, dev->name);
2254 #endif
2255
2256 e_hostunreach:
2257         err = -EHOSTUNREACH;
2258         goto out;
2259
2260 e_inval:
2261         err = -EINVAL;
2262         goto out;
2263
2264 e_nobufs:
2265         err = -ENOBUFS;
2266         goto out;
2267
2268 martian_source:
2269         err = -EINVAL;
2270 martian_source_keep_err:
2271         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2272         goto out;
2273 }
2274
2275 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2276                            u8 tos, struct net_device *dev, bool noref)
2277 {
2278         struct rtable * rth;
2279         unsigned        hash;
2280         int iif = dev->ifindex;
2281         struct net *net;
2282         int res;
2283
2284         net = dev_net(dev);
2285
2286         rcu_read_lock();
2287
2288         if (!rt_caching(net))
2289                 goto skip_cache;
2290
2291         tos &= IPTOS_RT_MASK;
2292         hash = rt_hash(daddr, saddr, iif, rt_genid(net));
2293
2294         for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;
2295              rth = rcu_dereference(rth->dst.rt_next)) {
2296                 if ((((__force u32)rth->fl.fl4_dst ^ (__force u32)daddr) |
2297                      ((__force u32)rth->fl.fl4_src ^ (__force u32)saddr) |
2298                      (rth->fl.iif ^ iif) |
2299                      rth->fl.oif |
2300                      (rth->fl.fl4_tos ^ tos)) == 0 &&
2301                     rth->fl.mark == skb->mark &&
2302                     net_eq(dev_net(rth->dst.dev), net) &&
2303                     !rt_is_expired(rth)) {
2304                         if (noref) {
2305                                 dst_use_noref(&rth->dst, jiffies);
2306                                 skb_dst_set_noref(skb, &rth->dst);
2307                         } else {
2308                                 dst_use(&rth->dst, jiffies);
2309                                 skb_dst_set(skb, &rth->dst);
2310                         }
2311                         RT_CACHE_STAT_INC(in_hit);
2312                         rcu_read_unlock();
2313                         return 0;
2314                 }
2315                 RT_CACHE_STAT_INC(in_hlist_search);
2316         }
2317
2318 skip_cache:
2319         /* Multicast recognition logic is moved from route cache to here.
2320            The problem was that too many Ethernet cards have broken/missing
2321            hardware multicast filters :-( As result the host on multicasting
2322            network acquires a lot of useless route cache entries, sort of
2323            SDR messages from all the world. Now we try to get rid of them.
2324            Really, provided software IP multicast filter is organized
2325            reasonably (at least, hashed), it does not result in a slowdown
2326            comparing with route cache reject entries.
2327            Note, that multicast routers are not affected, because
2328            route cache entry is created eventually.
2329          */
2330         if (ipv4_is_multicast(daddr)) {
2331                 struct in_device *in_dev = __in_dev_get_rcu(dev);
2332
2333                 if (in_dev) {
2334                         int our = ip_check_mc(in_dev, daddr, saddr,
2335                                               ip_hdr(skb)->protocol);
2336                         if (our
2337 #ifdef CONFIG_IP_MROUTE
2338                                 ||
2339                             (!ipv4_is_local_multicast(daddr) &&
2340                              IN_DEV_MFORWARD(in_dev))
2341 #endif
2342                            ) {
2343                                 int res = ip_route_input_mc(skb, daddr, saddr,
2344                                                             tos, dev, our);
2345                                 rcu_read_unlock();
2346                                 return res;
2347                         }
2348                 }
2349                 rcu_read_unlock();
2350                 return -EINVAL;
2351         }
2352         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2353         rcu_read_unlock();
2354         return res;
2355 }
2356 EXPORT_SYMBOL(ip_route_input_common);
2357
2358 /* called with rcu_read_lock() */
2359 static int __mkroute_output(struct rtable **result,
2360                             struct fib_result *res,
2361                             const struct flowi *fl,
2362                             const struct flowi *oldflp,
2363                             struct net_device *dev_out,
2364                             unsigned flags)
2365 {
2366         struct rtable *rth;
2367         struct in_device *in_dev;
2368         u32 tos = RT_FL_TOS(oldflp);
2369
2370         if (ipv4_is_loopback(fl->fl4_src) && !(dev_out->flags & IFF_LOOPBACK))
2371                 return -EINVAL;
2372
2373         if (ipv4_is_lbcast(fl->fl4_dst))
2374                 res->type = RTN_BROADCAST;
2375         else if (ipv4_is_multicast(fl->fl4_dst))
2376                 res->type = RTN_MULTICAST;
2377         else if (ipv4_is_zeronet(fl->fl4_dst))
2378                 return -EINVAL;
2379
2380         if (dev_out->flags & IFF_LOOPBACK)
2381                 flags |= RTCF_LOCAL;
2382
2383         in_dev = __in_dev_get_rcu(dev_out);
2384         if (!in_dev)
2385                 return -EINVAL;
2386
2387         if (res->type == RTN_BROADCAST) {
2388                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2389                 res->fi = NULL;
2390         } else if (res->type == RTN_MULTICAST) {
2391                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2392                 if (!ip_check_mc(in_dev, oldflp->fl4_dst, oldflp->fl4_src,
2393                                  oldflp->proto))
2394                         flags &= ~RTCF_LOCAL;
2395                 /* If multicast route do not exist use
2396                  * default one, but do not gateway in this case.
2397                  * Yes, it is hack.
2398                  */
2399                 if (res->fi && res->prefixlen < 4)
2400                         res->fi = NULL;
2401         }
2402
2403
2404         rth = dst_alloc(&ipv4_dst_ops);
2405         if (!rth)
2406                 return -ENOBUFS;
2407
2408         atomic_set(&rth->dst.__refcnt, 1);
2409         rth->dst.flags= DST_HOST;
2410         if (IN_DEV_CONF_GET(in_dev, NOXFRM))
2411                 rth->dst.flags |= DST_NOXFRM;
2412         if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
2413                 rth->dst.flags |= DST_NOPOLICY;
2414
2415         rth->fl.fl4_dst = oldflp->fl4_dst;
2416         rth->fl.fl4_tos = tos;
2417         rth->fl.fl4_src = oldflp->fl4_src;
2418         rth->fl.oif     = oldflp->oif;
2419         rth->fl.mark    = oldflp->mark;
2420         rth->rt_dst     = fl->fl4_dst;
2421         rth->rt_src     = fl->fl4_src;
2422         rth->rt_iif     = oldflp->oif ? : dev_out->ifindex;
2423         /* get references to the devices that are to be hold by the routing
2424            cache entry */
2425         rth->dst.dev    = dev_out;
2426         dev_hold(dev_out);
2427         rth->rt_gateway = fl->fl4_dst;
2428         rth->rt_spec_dst= fl->fl4_src;
2429
2430         rth->dst.output=ip_output;
2431         rth->dst.obsolete = -1;
2432         rth->rt_genid = rt_genid(dev_net(dev_out));
2433
2434         RT_CACHE_STAT_INC(out_slow_tot);
2435
2436         if (flags & RTCF_LOCAL) {
2437                 rth->dst.input = ip_local_deliver;
2438                 rth->rt_spec_dst = fl->fl4_dst;
2439         }
2440         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2441                 rth->rt_spec_dst = fl->fl4_src;
2442                 if (flags & RTCF_LOCAL &&
2443                     !(dev_out->flags & IFF_LOOPBACK)) {
2444                         rth->dst.output = ip_mc_output;
2445                         RT_CACHE_STAT_INC(out_slow_mc);
2446                 }
2447 #ifdef CONFIG_IP_MROUTE
2448                 if (res->type == RTN_MULTICAST) {
2449                         if (IN_DEV_MFORWARD(in_dev) &&
2450                             !ipv4_is_local_multicast(oldflp->fl4_dst)) {
2451                                 rth->dst.input = ip_mr_input;
2452                                 rth->dst.output = ip_mc_output;
2453                         }
2454                 }
2455 #endif
2456         }
2457
2458         rt_set_nexthop(rth, res, 0);
2459
2460         rth->rt_flags = flags;
2461         *result = rth;
2462         return 0;
2463 }
2464
2465 /* called with rcu_read_lock() */
2466 static int ip_mkroute_output(struct rtable **rp,
2467                              struct fib_result *res,
2468                              const struct flowi *fl,
2469                              const struct flowi *oldflp,
2470                              struct net_device *dev_out,
2471                              unsigned flags)
2472 {
2473         struct rtable *rth = NULL;
2474         int err = __mkroute_output(&rth, res, fl, oldflp, dev_out, flags);
2475         unsigned hash;
2476         if (err == 0) {
2477                 hash = rt_hash(oldflp->fl4_dst, oldflp->fl4_src, oldflp->oif,
2478                                rt_genid(dev_net(dev_out)));
2479                 err = rt_intern_hash(hash, rth, rp, NULL, oldflp->oif);
2480         }
2481
2482         return err;
2483 }
2484
2485 /*
2486  * Major route resolver routine.
2487  * called with rcu_read_lock();
2488  */
2489
2490 static int ip_route_output_slow(struct net *net, struct rtable **rp,
2491                                 const struct flowi *oldflp)
2492 {
2493         u32 tos = RT_FL_TOS(oldflp);
2494         struct flowi fl = { .fl4_dst = oldflp->fl4_dst,
2495                             .fl4_src = oldflp->fl4_src,
2496                             .fl4_tos = tos & IPTOS_RT_MASK,
2497                             .fl4_scope = ((tos & RTO_ONLINK) ?
2498                                           RT_SCOPE_LINK : RT_SCOPE_UNIVERSE),
2499                             .mark = oldflp->mark,
2500                             .iif = net->loopback_dev->ifindex,
2501                             .oif = oldflp->oif };
2502         struct fib_result res;
2503         unsigned int flags = 0;
2504         struct net_device *dev_out = NULL;
2505         int err;
2506
2507
2508         res.fi          = NULL;
2509 #ifdef CONFIG_IP_MULTIPLE_TABLES
2510         res.r           = NULL;
2511 #endif
2512
2513         if (oldflp->fl4_src) {
2514                 err = -EINVAL;
2515                 if (ipv4_is_multicast(oldflp->fl4_src) ||
2516                     ipv4_is_lbcast(oldflp->fl4_src) ||
2517                     ipv4_is_zeronet(oldflp->fl4_src))
2518                         goto out;
2519
2520                 /* I removed check for oif == dev_out->oif here.
2521                    It was wrong for two reasons:
2522                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2523                       is assigned to multiple interfaces.
2524                    2. Moreover, we are allowed to send packets with saddr
2525                       of another iface. --ANK
2526                  */
2527
2528                 if (oldflp->oif == 0 &&
2529                     (ipv4_is_multicast(oldflp->fl4_dst) ||
2530                      ipv4_is_lbcast(oldflp->fl4_dst))) {
2531                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2532                         dev_out = __ip_dev_find(net, oldflp->fl4_src, false);
2533                         if (dev_out == NULL)
2534                                 goto out;
2535
2536                         /* Special hack: user can direct multicasts
2537                            and limited broadcast via necessary interface
2538                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2539                            This hack is not just for fun, it allows
2540                            vic,vat and friends to work.
2541                            They bind socket to loopback, set ttl to zero
2542                            and expect that it will work.
2543                            From the viewpoint of routing cache they are broken,
2544                            because we are not allowed to build multicast path
2545                            with loopback source addr (look, routing cache
2546                            cannot know, that ttl is zero, so that packet
2547                            will not leave this host and route is valid).
2548                            Luckily, this hack is good workaround.
2549                          */
2550
2551                         fl.oif = dev_out->ifindex;
2552                         goto make_route;
2553                 }
2554
2555                 if (!(oldflp->flags & FLOWI_FLAG_ANYSRC)) {
2556                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2557                         if (!__ip_dev_find(net, oldflp->fl4_src, false))
2558                                 goto out;
2559                 }
2560         }
2561
2562
2563         if (oldflp->oif) {
2564                 dev_out = dev_get_by_index_rcu(net, oldflp->oif);
2565                 err = -ENODEV;
2566                 if (dev_out == NULL)
2567                         goto out;
2568
2569                 /* RACE: Check return value of inet_select_addr instead. */
2570                 if (rcu_dereference(dev_out->ip_ptr) == NULL)
2571                         goto out;       /* Wrong error code */
2572
2573                 if (ipv4_is_local_multicast(oldflp->fl4_dst) ||
2574                     ipv4_is_lbcast(oldflp->fl4_dst)) {
2575                         if (!fl.fl4_src)
2576                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2577                                                               RT_SCOPE_LINK);
2578                         goto make_route;
2579                 }
2580                 if (!fl.fl4_src) {
2581                         if (ipv4_is_multicast(oldflp->fl4_dst))
2582                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2583                                                               fl.fl4_scope);
2584                         else if (!oldflp->fl4_dst)
2585                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2586                                                               RT_SCOPE_HOST);
2587                 }
2588         }
2589
2590         if (!fl.fl4_dst) {
2591                 fl.fl4_dst = fl.fl4_src;
2592                 if (!fl.fl4_dst)
2593                         fl.fl4_dst = fl.fl4_src = htonl(INADDR_LOOPBACK);
2594                 dev_out = net->loopback_dev;
2595                 fl.oif = net->loopback_dev->ifindex;
2596                 res.type = RTN_LOCAL;
2597                 flags |= RTCF_LOCAL;
2598                 goto make_route;
2599         }
2600
2601         if (fib_lookup(net, &fl, &res)) {
2602                 res.fi = NULL;
2603                 if (oldflp->oif) {
2604                         /* Apparently, routing tables are wrong. Assume,
2605                            that the destination is on link.
2606
2607                            WHY? DW.
2608                            Because we are allowed to send to iface
2609                            even if it has NO routes and NO assigned
2610                            addresses. When oif is specified, routing
2611                            tables are looked up with only one purpose:
2612                            to catch if destination is gatewayed, rather than
2613                            direct. Moreover, if MSG_DONTROUTE is set,
2614                            we send packet, ignoring both routing tables
2615                            and ifaddr state. --ANK
2616
2617
2618                            We could make it even if oif is unknown,
2619                            likely IPv6, but we do not.
2620                          */
2621
2622                         if (fl.fl4_src == 0)
2623                                 fl.fl4_src = inet_select_addr(dev_out, 0,
2624                                                               RT_SCOPE_LINK);
2625                         res.type = RTN_UNICAST;
2626                         goto make_route;
2627                 }
2628                 err = -ENETUNREACH;
2629                 goto out;
2630         }
2631
2632         if (res.type == RTN_LOCAL) {
2633                 if (!fl.fl4_src)
2634                         fl.fl4_src = fl.fl4_dst;
2635                 dev_out = net->loopback_dev;
2636                 fl.oif = dev_out->ifindex;
2637                 res.fi = NULL;
2638                 flags |= RTCF_LOCAL;
2639                 goto make_route;
2640         }
2641
2642 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2643         if (res.fi->fib_nhs > 1 && fl.oif == 0)
2644                 fib_select_multipath(&fl, &res);
2645         else
2646 #endif
2647         if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
2648                 fib_select_default(net, &fl, &res);
2649
2650         if (!fl.fl4_src)
2651                 fl.fl4_src = FIB_RES_PREFSRC(res);
2652
2653         dev_out = FIB_RES_DEV(res);
2654         fl.oif = dev_out->ifindex;
2655
2656
2657 make_route:
2658         err = ip_mkroute_output(rp, &res, &fl, oldflp, dev_out, flags);
2659
2660 out:    return err;
2661 }
2662
2663 int __ip_route_output_key(struct net *net, struct rtable **rp,
2664                           const struct flowi *flp)
2665 {
2666         unsigned int hash;
2667         int res;
2668         struct rtable *rth;
2669
2670         if (!rt_caching(net))
2671                 goto slow_output;
2672
2673         hash = rt_hash(flp->fl4_dst, flp->fl4_src, flp->oif, rt_genid(net));
2674
2675         rcu_read_lock_bh();
2676         for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth;
2677                 rth = rcu_dereference_bh(rth->dst.rt_next)) {
2678                 if (rth->fl.fl4_dst == flp->fl4_dst &&
2679                     rth->fl.fl4_src == flp->fl4_src &&
2680                     rt_is_output_route(rth) &&
2681                     rth->fl.oif == flp->oif &&
2682                     rth->fl.mark == flp->mark &&
2683                     !((rth->fl.fl4_tos ^ flp->fl4_tos) &
2684                             (IPTOS_RT_MASK | RTO_ONLINK)) &&
2685                     net_eq(dev_net(rth->dst.dev), net) &&
2686                     !rt_is_expired(rth)) {
2687                         dst_use(&rth->dst, jiffies);
2688                         RT_CACHE_STAT_INC(out_hit);
2689                         rcu_read_unlock_bh();
2690                         *rp = rth;
2691                         return 0;
2692                 }
2693                 RT_CACHE_STAT_INC(out_hlist_search);
2694         }
2695         rcu_read_unlock_bh();
2696
2697 slow_output:
2698         rcu_read_lock();
2699         res = ip_route_output_slow(net, rp, flp);
2700         rcu_read_unlock();
2701         return res;
2702 }
2703 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2704
2705 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2706 {
2707         return NULL;
2708 }
2709
2710 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu)
2711 {
2712 }
2713
2714 static struct dst_ops ipv4_dst_blackhole_ops = {
2715         .family                 =       AF_INET,
2716         .protocol               =       cpu_to_be16(ETH_P_IP),
2717         .destroy                =       ipv4_dst_destroy,
2718         .check                  =       ipv4_blackhole_dst_check,
2719         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2720 };
2721
2722
2723 static int ipv4_dst_blackhole(struct net *net, struct rtable **rp, struct flowi *flp)
2724 {
2725         struct rtable *ort = *rp;
2726         struct rtable *rt = (struct rtable *)
2727                 dst_alloc(&ipv4_dst_blackhole_ops);
2728
2729         if (rt) {
2730                 struct dst_entry *new = &rt->dst;
2731
2732                 atomic_set(&new->__refcnt, 1);
2733                 new->__use = 1;
2734                 new->input = dst_discard;
2735                 new->output = dst_discard;
2736                 dst_copy_metrics(new, &ort->dst);
2737
2738                 new->dev = ort->dst.dev;
2739                 if (new->dev)
2740                         dev_hold(new->dev);
2741
2742                 rt->fl = ort->fl;
2743
2744                 rt->rt_genid = rt_genid(net);
2745                 rt->rt_flags = ort->rt_flags;
2746                 rt->rt_type = ort->rt_type;
2747                 rt->rt_dst = ort->rt_dst;
2748                 rt->rt_src = ort->rt_src;
2749                 rt->rt_iif = ort->rt_iif;
2750                 rt->rt_gateway = ort->rt_gateway;
2751                 rt->rt_spec_dst = ort->rt_spec_dst;
2752                 rt->peer = ort->peer;
2753                 if (rt->peer)
2754                         atomic_inc(&rt->peer->refcnt);
2755
2756                 dst_free(new);
2757         }
2758
2759         dst_release(&(*rp)->dst);
2760         *rp = rt;
2761         return rt ? 0 : -ENOMEM;
2762 }
2763
2764 int ip_route_output_flow(struct net *net, struct rtable **rp, struct flowi *flp,
2765                          struct sock *sk, int flags)
2766 {
2767         int err;
2768
2769         if ((err = __ip_route_output_key(net, rp, flp)) != 0)
2770                 return err;
2771
2772         if (flp->proto) {
2773                 if (!flp->fl4_src)
2774                         flp->fl4_src = (*rp)->rt_src;
2775                 if (!flp->fl4_dst)
2776                         flp->fl4_dst = (*rp)->rt_dst;
2777                 err = __xfrm_lookup(net, (struct dst_entry **)rp, flp, sk,
2778                                     flags ? XFRM_LOOKUP_WAIT : 0);
2779                 if (err == -EREMOTE)
2780                         err = ipv4_dst_blackhole(net, rp, flp);
2781
2782                 return err;
2783         }
2784
2785         return 0;
2786 }
2787 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2788
2789 int ip_route_output_key(struct net *net, struct rtable **rp, struct flowi *flp)
2790 {
2791         return ip_route_output_flow(net, rp, flp, NULL, 0);
2792 }
2793 EXPORT_SYMBOL(ip_route_output_key);
2794
2795 static int rt_fill_info(struct net *net,
2796                         struct sk_buff *skb, u32 pid, u32 seq, int event,
2797                         int nowait, unsigned int flags)
2798 {
2799         struct rtable *rt = skb_rtable(skb);
2800         struct rtmsg *r;
2801         struct nlmsghdr *nlh;
2802         long expires;
2803         u32 id = 0, ts = 0, tsage = 0, error;
2804
2805         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2806         if (nlh == NULL)
2807                 return -EMSGSIZE;
2808
2809         r = nlmsg_data(nlh);
2810         r->rtm_family    = AF_INET;
2811         r->rtm_dst_len  = 32;
2812         r->rtm_src_len  = 0;
2813         r->rtm_tos      = rt->fl.fl4_tos;
2814         r->rtm_table    = RT_TABLE_MAIN;
2815         NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
2816         r->rtm_type     = rt->rt_type;
2817         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2818         r->rtm_protocol = RTPROT_UNSPEC;
2819         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2820         if (rt->rt_flags & RTCF_NOTIFY)
2821                 r->rtm_flags |= RTM_F_NOTIFY;
2822
2823         NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst);
2824
2825         if (rt->fl.fl4_src) {
2826                 r->rtm_src_len = 32;
2827                 NLA_PUT_BE32(skb, RTA_SRC, rt->fl.fl4_src);
2828         }
2829         if (rt->dst.dev)
2830                 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex);
2831 #ifdef CONFIG_NET_CLS_ROUTE
2832         if (rt->dst.tclassid)
2833                 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid);
2834 #endif
2835         if (rt_is_input_route(rt))
2836                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst);
2837         else if (rt->rt_src != rt->fl.fl4_src)
2838                 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src);
2839
2840         if (rt->rt_dst != rt->rt_gateway)
2841                 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway);
2842
2843         if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
2844                 goto nla_put_failure;
2845
2846         if (rt->fl.mark)
2847                 NLA_PUT_BE32(skb, RTA_MARK, rt->fl.mark);
2848
2849         error = rt->dst.error;
2850         expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
2851         if (rt->peer) {
2852                 inet_peer_refcheck(rt->peer);
2853                 id = atomic_read(&rt->peer->ip_id_count) & 0xffff;
2854                 if (rt->peer->tcp_ts_stamp) {
2855                         ts = rt->peer->tcp_ts;
2856                         tsage = get_seconds() - rt->peer->tcp_ts_stamp;
2857                 }
2858         }
2859
2860         if (rt_is_input_route(rt)) {
2861 #ifdef CONFIG_IP_MROUTE
2862                 __be32 dst = rt->rt_dst;
2863
2864                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2865                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2866                         int err = ipmr_get_route(net, skb, r, nowait);
2867                         if (err <= 0) {
2868                                 if (!nowait) {
2869                                         if (err == 0)
2870                                                 return 0;
2871                                         goto nla_put_failure;
2872                                 } else {
2873                                         if (err == -EMSGSIZE)
2874                                                 goto nla_put_failure;
2875                                         error = err;
2876                                 }
2877                         }
2878                 } else
2879 #endif
2880                         NLA_PUT_U32(skb, RTA_IIF, rt->fl.iif);
2881         }
2882
2883         if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage,
2884                                expires, error) < 0)
2885                 goto nla_put_failure;
2886
2887         return nlmsg_end(skb, nlh);
2888
2889 nla_put_failure:
2890         nlmsg_cancel(skb, nlh);
2891         return -EMSGSIZE;
2892 }
2893
2894 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg)
2895 {
2896         struct net *net = sock_net(in_skb->sk);
2897         struct rtmsg *rtm;
2898         struct nlattr *tb[RTA_MAX+1];
2899         struct rtable *rt = NULL;
2900         __be32 dst = 0;
2901         __be32 src = 0;
2902         u32 iif;
2903         int err;
2904         int mark;
2905         struct sk_buff *skb;
2906
2907         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2908         if (err < 0)
2909                 goto errout;
2910
2911         rtm = nlmsg_data(nlh);
2912
2913         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2914         if (skb == NULL) {
2915                 err = -ENOBUFS;
2916                 goto errout;
2917         }
2918
2919         /* Reserve room for dummy headers, this skb can pass
2920            through good chunk of routing engine.
2921          */
2922         skb_reset_mac_header(skb);
2923         skb_reset_network_header(skb);
2924
2925         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2926         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2927         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2928
2929         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2930         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2931         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2932         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2933
2934         if (iif) {
2935                 struct net_device *dev;
2936
2937                 dev = __dev_get_by_index(net, iif);
2938                 if (dev == NULL) {
2939                         err = -ENODEV;
2940                         goto errout_free;
2941                 }
2942
2943                 skb->protocol   = htons(ETH_P_IP);
2944                 skb->dev        = dev;
2945                 skb->mark       = mark;
2946                 local_bh_disable();
2947                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2948                 local_bh_enable();
2949
2950                 rt = skb_rtable(skb);
2951                 if (err == 0 && rt->dst.error)
2952                         err = -rt->dst.error;
2953         } else {
2954                 struct flowi fl = {
2955                         .fl4_dst = dst,
2956                         .fl4_src = src,
2957                         .fl4_tos = rtm->rtm_tos,
2958                         .oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2959                         .mark = mark,
2960                 };
2961                 err = ip_route_output_key(net, &rt, &fl);
2962         }
2963
2964         if (err)
2965                 goto errout_free;
2966
2967         skb_dst_set(skb, &rt->dst);
2968         if (rtm->rtm_flags & RTM_F_NOTIFY)
2969                 rt->rt_flags |= RTCF_NOTIFY;
2970
2971         err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2972                            RTM_NEWROUTE, 0, 0);
2973         if (err <= 0)
2974                 goto errout_free;
2975
2976         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2977 errout:
2978         return err;
2979
2980 errout_free:
2981         kfree_skb(skb);
2982         goto errout;
2983 }
2984
2985 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
2986 {
2987         struct rtable *rt;
2988         int h, s_h;
2989         int idx, s_idx;
2990         struct net *net;
2991
2992         net = sock_net(skb->sk);
2993
2994         s_h = cb->args[0];
2995         if (s_h < 0)
2996                 s_h = 0;
2997         s_idx = idx = cb->args[1];
2998         for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) {
2999                 if (!rt_hash_table[h].chain)
3000                         continue;
3001                 rcu_read_lock_bh();
3002                 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt;
3003                      rt = rcu_dereference_bh(rt->dst.rt_next), idx++) {
3004                         if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx)
3005                                 continue;
3006                         if (rt_is_expired(rt))
3007                                 continue;
3008                         skb_dst_set_noref(skb, &rt->dst);
3009                         if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
3010                                          cb->nlh->nlmsg_seq, RTM_NEWROUTE,
3011                                          1, NLM_F_MULTI) <= 0) {
3012                                 skb_dst_drop(skb);
3013                                 rcu_read_unlock_bh();
3014                                 goto done;
3015                         }
3016                         skb_dst_drop(skb);
3017                 }
3018                 rcu_read_unlock_bh();
3019         }
3020
3021 done:
3022         cb->args[0] = h;
3023         cb->args[1] = idx;
3024         return skb->len;
3025 }
3026
3027 void ip_rt_multicast_event(struct in_device *in_dev)
3028 {
3029         rt_cache_flush(dev_net(in_dev->dev), 0);
3030 }
3031
3032 #ifdef CONFIG_SYSCTL
3033 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
3034                                         void __user *buffer,
3035                                         size_t *lenp, loff_t *ppos)
3036 {
3037         if (write) {
3038                 int flush_delay;
3039                 ctl_table ctl;
3040                 struct net *net;
3041
3042                 memcpy(&ctl, __ctl, sizeof(ctl));
3043                 ctl.data = &flush_delay;
3044                 proc_dointvec(&ctl, write, buffer, lenp, ppos);
3045
3046                 net = (struct net *)__ctl->extra1;
3047                 rt_cache_flush(net, flush_delay);
3048                 return 0;
3049         }
3050
3051         return -EINVAL;
3052 }
3053
3054 static ctl_table ipv4_route_table[] = {
3055         {
3056                 .procname       = "gc_thresh",
3057                 .data           = &ipv4_dst_ops.gc_thresh,
3058                 .maxlen         = sizeof(int),
3059                 .mode           = 0644,
3060                 .proc_handler   = proc_dointvec,
3061         },
3062         {
3063                 .procname       = "max_size",
3064                 .data           = &ip_rt_max_size,
3065                 .maxlen         = sizeof(int),
3066                 .mode           = 0644,
3067                 .proc_handler   = proc_dointvec,
3068         },
3069         {
3070                 /*  Deprecated. Use gc_min_interval_ms */
3071
3072                 .procname       = "gc_min_interval",
3073                 .data           = &ip_rt_gc_min_interval,
3074                 .maxlen         = sizeof(int),
3075                 .mode           = 0644,
3076                 .proc_handler   = proc_dointvec_jiffies,
3077         },
3078         {
3079                 .procname       = "gc_min_interval_ms",
3080                 .data           = &ip_rt_gc_min_interval,
3081                 .maxlen         = sizeof(int),
3082                 .mode           = 0644,
3083                 .proc_handler   = proc_dointvec_ms_jiffies,
3084         },
3085         {
3086                 .procname       = "gc_timeout",
3087                 .data           = &ip_rt_gc_timeout,
3088                 .maxlen         = sizeof(int),
3089                 .mode           = 0644,
3090                 .proc_handler   = proc_dointvec_jiffies,
3091         },
3092         {
3093                 .procname       = "gc_interval",
3094                 .data           = &ip_rt_gc_interval,
3095                 .maxlen         = sizeof(int),
3096                 .mode           = 0644,
3097                 .proc_handler   = proc_dointvec_jiffies,
3098         },
3099         {
3100                 .procname       = "redirect_load",
3101                 .data           = &ip_rt_redirect_load,
3102                 .maxlen         = sizeof(int),
3103                 .mode           = 0644,
3104                 .proc_handler   = proc_dointvec,
3105         },
3106         {
3107                 .procname       = "redirect_number",
3108                 .data           = &ip_rt_redirect_number,
3109                 .maxlen         = sizeof(int),
3110                 .mode           = 0644,
3111                 .proc_handler   = proc_dointvec,
3112         },
3113         {
3114                 .procname       = "redirect_silence",
3115                 .data           = &ip_rt_redirect_silence,
3116                 .maxlen         = sizeof(int),
3117                 .mode           = 0644,
3118                 .proc_handler   = proc_dointvec,
3119         },
3120         {
3121                 .procname       = "error_cost",
3122                 .data           = &ip_rt_error_cost,
3123                 .maxlen         = sizeof(int),
3124                 .mode           = 0644,
3125                 .proc_handler   = proc_dointvec,
3126         },
3127         {
3128                 .procname       = "error_burst",
3129                 .data           = &ip_rt_error_burst,
3130                 .maxlen         = sizeof(int),
3131                 .mode           = 0644,
3132                 .proc_handler   = proc_dointvec,
3133         },
3134         {
3135                 .procname       = "gc_elasticity",
3136                 .data           = &ip_rt_gc_elasticity,
3137                 .maxlen         = sizeof(int),
3138                 .mode           = 0644,
3139                 .proc_handler   = proc_dointvec,
3140         },
3141         {
3142                 .procname       = "mtu_expires",
3143                 .data           = &ip_rt_mtu_expires,
3144                 .maxlen         = sizeof(int),
3145                 .mode           = 0644,
3146                 .proc_handler   = proc_dointvec_jiffies,
3147         },
3148         {
3149                 .procname       = "min_pmtu",
3150                 .data           = &ip_rt_min_pmtu,
3151                 .maxlen         = sizeof(int),
3152                 .mode           = 0644,
3153                 .proc_handler   = proc_dointvec,
3154         },
3155         {
3156                 .procname       = "min_adv_mss",
3157                 .data           = &ip_rt_min_advmss,
3158                 .maxlen         = sizeof(int),
3159                 .mode           = 0644,
3160                 .proc_handler   = proc_dointvec,
3161         },
3162         { }
3163 };
3164
3165 static struct ctl_table empty[1];
3166
3167 static struct ctl_table ipv4_skeleton[] =
3168 {
3169         { .procname = "route", 
3170           .mode = 0555, .child = ipv4_route_table},
3171         { .procname = "neigh", 
3172           .mode = 0555, .child = empty},
3173         { }
3174 };
3175
3176 static __net_initdata struct ctl_path ipv4_path[] = {
3177         { .procname = "net", },
3178         { .procname = "ipv4", },
3179         { },
3180 };
3181
3182 static struct ctl_table ipv4_route_flush_table[] = {
3183         {
3184                 .procname       = "flush",
3185                 .maxlen         = sizeof(int),
3186                 .mode           = 0200,
3187                 .proc_handler   = ipv4_sysctl_rtcache_flush,
3188         },
3189         { },
3190 };
3191
3192 static __net_initdata struct ctl_path ipv4_route_path[] = {
3193         { .procname = "net", },
3194         { .procname = "ipv4", },
3195         { .procname = "route", },
3196         { },
3197 };
3198
3199 static __net_init int sysctl_route_net_init(struct net *net)
3200 {
3201         struct ctl_table *tbl;
3202
3203         tbl = ipv4_route_flush_table;
3204         if (!net_eq(net, &init_net)) {
3205                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3206                 if (tbl == NULL)
3207                         goto err_dup;
3208         }
3209         tbl[0].extra1 = net;
3210
3211         net->ipv4.route_hdr =
3212                 register_net_sysctl_table(net, ipv4_route_path, tbl);
3213         if (net->ipv4.route_hdr == NULL)
3214                 goto err_reg;
3215         return 0;
3216
3217 err_reg:
3218         if (tbl != ipv4_route_flush_table)
3219                 kfree(tbl);
3220 err_dup:
3221         return -ENOMEM;
3222 }
3223
3224 static __net_exit void sysctl_route_net_exit(struct net *net)
3225 {
3226         struct ctl_table *tbl;
3227
3228         tbl = net->ipv4.route_hdr->ctl_table_arg;
3229         unregister_net_sysctl_table(net->ipv4.route_hdr);
3230         BUG_ON(tbl == ipv4_route_flush_table);
3231         kfree(tbl);
3232 }
3233
3234 static __net_initdata struct pernet_operations sysctl_route_ops = {
3235         .init = sysctl_route_net_init,
3236         .exit = sysctl_route_net_exit,
3237 };
3238 #endif
3239
3240 static __net_init int rt_genid_init(struct net *net)
3241 {
3242         get_random_bytes(&net->ipv4.rt_genid,
3243                          sizeof(net->ipv4.rt_genid));
3244         return 0;
3245 }
3246
3247 static __net_initdata struct pernet_operations rt_genid_ops = {
3248         .init = rt_genid_init,
3249 };
3250
3251
3252 #ifdef CONFIG_NET_CLS_ROUTE
3253 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3254 #endif /* CONFIG_NET_CLS_ROUTE */
3255
3256 static __initdata unsigned long rhash_entries;
3257 static int __init set_rhash_entries(char *str)
3258 {
3259         if (!str)
3260                 return 0;
3261         rhash_entries = simple_strtoul(str, &str, 0);
3262         return 1;
3263 }
3264 __setup("rhash_entries=", set_rhash_entries);
3265
3266 int __init ip_rt_init(void)
3267 {
3268         int rc = 0;
3269
3270 #ifdef CONFIG_NET_CLS_ROUTE
3271         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3272         if (!ip_rt_acct)
3273                 panic("IP: failed to allocate ip_rt_acct\n");
3274 #endif
3275
3276         ipv4_dst_ops.kmem_cachep =
3277                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3278                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3279
3280         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3281
3282         if (dst_entries_init(&ipv4_dst_ops) < 0)
3283                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3284
3285         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3286                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3287
3288         rt_hash_table = (struct rt_hash_bucket *)
3289                 alloc_large_system_hash("IP route cache",
3290                                         sizeof(struct rt_hash_bucket),
3291                                         rhash_entries,
3292                                         (totalram_pages >= 128 * 1024) ?
3293                                         15 : 17,
3294                                         0,
3295                                         &rt_hash_log,
3296                                         &rt_hash_mask,
3297                                         rhash_entries ? 0 : 512 * 1024);
3298         memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket));
3299         rt_hash_lock_init();
3300
3301         ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1);
3302         ip_rt_max_size = (rt_hash_mask + 1) * 16;
3303
3304         devinet_init();
3305         ip_fib_init();
3306
3307         /* All the timers, started at system startup tend
3308            to synchronize. Perturb it a bit.
3309          */
3310         INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
3311         expires_ljiffies = jiffies;
3312         schedule_delayed_work(&expires_work,
3313                 net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
3314
3315         if (ip_rt_proc_init())
3316                 printk(KERN_ERR "Unable to create route proc files\n");
3317 #ifdef CONFIG_XFRM
3318         xfrm_init();
3319         xfrm4_init(ip_rt_max_size);
3320 #endif
3321         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);
3322
3323 #ifdef CONFIG_SYSCTL
3324         register_pernet_subsys(&sysctl_route_ops);
3325 #endif
3326         register_pernet_subsys(&rt_genid_ops);
3327         return rc;
3328 }
3329
3330 #ifdef CONFIG_SYSCTL
3331 /*
3332  * We really need to sanitize the damn ipv4 init order, then all
3333  * this nonsense will go away.
3334  */
3335 void __init ip_static_sysctl_init(void)
3336 {
3337         register_sysctl_paths(ipv4_path, ipv4_skeleton);
3338 }
3339 #endif