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