ipv4: percpu nh_rth_output cache
[linux-3.10.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 #define pr_fmt(fmt) "IPv4: " fmt
66
67 #include <linux/module.h>
68 #include <asm/uaccess.h>
69 #include <linux/bitops.h>
70 #include <linux/types.h>
71 #include <linux/kernel.h>
72 #include <linux/mm.h>
73 #include <linux/bootmem.h>
74 #include <linux/string.h>
75 #include <linux/socket.h>
76 #include <linux/sockios.h>
77 #include <linux/errno.h>
78 #include <linux/in.h>
79 #include <linux/inet.h>
80 #include <linux/netdevice.h>
81 #include <linux/proc_fs.h>
82 #include <linux/init.h>
83 #include <linux/workqueue.h>
84 #include <linux/skbuff.h>
85 #include <linux/inetdevice.h>
86 #include <linux/igmp.h>
87 #include <linux/pkt_sched.h>
88 #include <linux/mroute.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/jhash.h>
92 #include <linux/rcupdate.h>
93 #include <linux/times.h>
94 #include <linux/slab.h>
95 #include <linux/prefetch.h>
96 #include <net/dst.h>
97 #include <net/net_namespace.h>
98 #include <net/protocol.h>
99 #include <net/ip.h>
100 #include <net/route.h>
101 #include <net/inetpeer.h>
102 #include <net/sock.h>
103 #include <net/ip_fib.h>
104 #include <net/arp.h>
105 #include <net/tcp.h>
106 #include <net/icmp.h>
107 #include <net/xfrm.h>
108 #include <net/netevent.h>
109 #include <net/rtnetlink.h>
110 #ifdef CONFIG_SYSCTL
111 #include <linux/sysctl.h>
112 #include <linux/kmemleak.h>
113 #endif
114 #include <net/secure_seq.h>
115
116 #define RT_FL_TOS(oldflp4) \
117         ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
118
119 #define IP_MAX_MTU      0xFFF0
120
121 #define RT_GC_TIMEOUT (300*HZ)
122
123 static int ip_rt_max_size;
124 static int ip_rt_gc_timeout __read_mostly       = RT_GC_TIMEOUT;
125 static int ip_rt_gc_interval __read_mostly  = 60 * HZ;
126 static int ip_rt_gc_min_interval __read_mostly  = HZ / 2;
127 static int ip_rt_redirect_number __read_mostly  = 9;
128 static int ip_rt_redirect_load __read_mostly    = HZ / 50;
129 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
130 static int ip_rt_error_cost __read_mostly       = HZ;
131 static int ip_rt_error_burst __read_mostly      = 5 * HZ;
132 static int ip_rt_gc_elasticity __read_mostly    = 8;
133 static int ip_rt_mtu_expires __read_mostly      = 10 * 60 * HZ;
134 static int ip_rt_min_pmtu __read_mostly         = 512 + 20 + 20;
135 static int ip_rt_min_advmss __read_mostly       = 256;
136
137 /*
138  *      Interface to generic destination cache.
139  */
140
141 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
142 static unsigned int      ipv4_default_advmss(const struct dst_entry *dst);
143 static unsigned int      ipv4_mtu(const struct dst_entry *dst);
144 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
145 static void              ipv4_link_failure(struct sk_buff *skb);
146 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
147                                            struct sk_buff *skb, u32 mtu);
148 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
149                                         struct sk_buff *skb);
150
151 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
152                             int how)
153 {
154 }
155
156 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
157 {
158         WARN_ON(1);
159         return NULL;
160 }
161
162 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
163                                            struct sk_buff *skb,
164                                            const void *daddr);
165
166 static struct dst_ops ipv4_dst_ops = {
167         .family =               AF_INET,
168         .protocol =             cpu_to_be16(ETH_P_IP),
169         .check =                ipv4_dst_check,
170         .default_advmss =       ipv4_default_advmss,
171         .mtu =                  ipv4_mtu,
172         .cow_metrics =          ipv4_cow_metrics,
173         .ifdown =               ipv4_dst_ifdown,
174         .negative_advice =      ipv4_negative_advice,
175         .link_failure =         ipv4_link_failure,
176         .update_pmtu =          ip_rt_update_pmtu,
177         .redirect =             ip_do_redirect,
178         .local_out =            __ip_local_out,
179         .neigh_lookup =         ipv4_neigh_lookup,
180 };
181
182 #define ECN_OR_COST(class)      TC_PRIO_##class
183
184 const __u8 ip_tos2prio[16] = {
185         TC_PRIO_BESTEFFORT,
186         ECN_OR_COST(BESTEFFORT),
187         TC_PRIO_BESTEFFORT,
188         ECN_OR_COST(BESTEFFORT),
189         TC_PRIO_BULK,
190         ECN_OR_COST(BULK),
191         TC_PRIO_BULK,
192         ECN_OR_COST(BULK),
193         TC_PRIO_INTERACTIVE,
194         ECN_OR_COST(INTERACTIVE),
195         TC_PRIO_INTERACTIVE,
196         ECN_OR_COST(INTERACTIVE),
197         TC_PRIO_INTERACTIVE_BULK,
198         ECN_OR_COST(INTERACTIVE_BULK),
199         TC_PRIO_INTERACTIVE_BULK,
200         ECN_OR_COST(INTERACTIVE_BULK)
201 };
202 EXPORT_SYMBOL(ip_tos2prio);
203
204 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
205 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
206
207 static inline int rt_genid(struct net *net)
208 {
209         return atomic_read(&net->ipv4.rt_genid);
210 }
211
212 #ifdef CONFIG_PROC_FS
213 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
214 {
215         if (*pos)
216                 return NULL;
217         return SEQ_START_TOKEN;
218 }
219
220 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
221 {
222         ++*pos;
223         return NULL;
224 }
225
226 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
227 {
228 }
229
230 static int rt_cache_seq_show(struct seq_file *seq, void *v)
231 {
232         if (v == SEQ_START_TOKEN)
233                 seq_printf(seq, "%-127s\n",
234                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
235                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
236                            "HHUptod\tSpecDst");
237         return 0;
238 }
239
240 static const struct seq_operations rt_cache_seq_ops = {
241         .start  = rt_cache_seq_start,
242         .next   = rt_cache_seq_next,
243         .stop   = rt_cache_seq_stop,
244         .show   = rt_cache_seq_show,
245 };
246
247 static int rt_cache_seq_open(struct inode *inode, struct file *file)
248 {
249         return seq_open(file, &rt_cache_seq_ops);
250 }
251
252 static const struct file_operations rt_cache_seq_fops = {
253         .owner   = THIS_MODULE,
254         .open    = rt_cache_seq_open,
255         .read    = seq_read,
256         .llseek  = seq_lseek,
257         .release = seq_release,
258 };
259
260
261 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
262 {
263         int cpu;
264
265         if (*pos == 0)
266                 return SEQ_START_TOKEN;
267
268         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
269                 if (!cpu_possible(cpu))
270                         continue;
271                 *pos = cpu+1;
272                 return &per_cpu(rt_cache_stat, cpu);
273         }
274         return NULL;
275 }
276
277 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
278 {
279         int cpu;
280
281         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
282                 if (!cpu_possible(cpu))
283                         continue;
284                 *pos = cpu+1;
285                 return &per_cpu(rt_cache_stat, cpu);
286         }
287         return NULL;
288
289 }
290
291 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
292 {
293
294 }
295
296 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
297 {
298         struct rt_cache_stat *st = v;
299
300         if (v == SEQ_START_TOKEN) {
301                 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");
302                 return 0;
303         }
304
305         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
306                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
307                    dst_entries_get_slow(&ipv4_dst_ops),
308                    st->in_hit,
309                    st->in_slow_tot,
310                    st->in_slow_mc,
311                    st->in_no_route,
312                    st->in_brd,
313                    st->in_martian_dst,
314                    st->in_martian_src,
315
316                    st->out_hit,
317                    st->out_slow_tot,
318                    st->out_slow_mc,
319
320                    st->gc_total,
321                    st->gc_ignored,
322                    st->gc_goal_miss,
323                    st->gc_dst_overflow,
324                    st->in_hlist_search,
325                    st->out_hlist_search
326                 );
327         return 0;
328 }
329
330 static const struct seq_operations rt_cpu_seq_ops = {
331         .start  = rt_cpu_seq_start,
332         .next   = rt_cpu_seq_next,
333         .stop   = rt_cpu_seq_stop,
334         .show   = rt_cpu_seq_show,
335 };
336
337
338 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
339 {
340         return seq_open(file, &rt_cpu_seq_ops);
341 }
342
343 static const struct file_operations rt_cpu_seq_fops = {
344         .owner   = THIS_MODULE,
345         .open    = rt_cpu_seq_open,
346         .read    = seq_read,
347         .llseek  = seq_lseek,
348         .release = seq_release,
349 };
350
351 #ifdef CONFIG_IP_ROUTE_CLASSID
352 static int rt_acct_proc_show(struct seq_file *m, void *v)
353 {
354         struct ip_rt_acct *dst, *src;
355         unsigned int i, j;
356
357         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
358         if (!dst)
359                 return -ENOMEM;
360
361         for_each_possible_cpu(i) {
362                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
363                 for (j = 0; j < 256; j++) {
364                         dst[j].o_bytes   += src[j].o_bytes;
365                         dst[j].o_packets += src[j].o_packets;
366                         dst[j].i_bytes   += src[j].i_bytes;
367                         dst[j].i_packets += src[j].i_packets;
368                 }
369         }
370
371         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
372         kfree(dst);
373         return 0;
374 }
375
376 static int rt_acct_proc_open(struct inode *inode, struct file *file)
377 {
378         return single_open(file, rt_acct_proc_show, NULL);
379 }
380
381 static const struct file_operations rt_acct_proc_fops = {
382         .owner          = THIS_MODULE,
383         .open           = rt_acct_proc_open,
384         .read           = seq_read,
385         .llseek         = seq_lseek,
386         .release        = single_release,
387 };
388 #endif
389
390 static int __net_init ip_rt_do_proc_init(struct net *net)
391 {
392         struct proc_dir_entry *pde;
393
394         pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
395                         &rt_cache_seq_fops);
396         if (!pde)
397                 goto err1;
398
399         pde = proc_create("rt_cache", S_IRUGO,
400                           net->proc_net_stat, &rt_cpu_seq_fops);
401         if (!pde)
402                 goto err2;
403
404 #ifdef CONFIG_IP_ROUTE_CLASSID
405         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
406         if (!pde)
407                 goto err3;
408 #endif
409         return 0;
410
411 #ifdef CONFIG_IP_ROUTE_CLASSID
412 err3:
413         remove_proc_entry("rt_cache", net->proc_net_stat);
414 #endif
415 err2:
416         remove_proc_entry("rt_cache", net->proc_net);
417 err1:
418         return -ENOMEM;
419 }
420
421 static void __net_exit ip_rt_do_proc_exit(struct net *net)
422 {
423         remove_proc_entry("rt_cache", net->proc_net_stat);
424         remove_proc_entry("rt_cache", net->proc_net);
425 #ifdef CONFIG_IP_ROUTE_CLASSID
426         remove_proc_entry("rt_acct", net->proc_net);
427 #endif
428 }
429
430 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
431         .init = ip_rt_do_proc_init,
432         .exit = ip_rt_do_proc_exit,
433 };
434
435 static int __init ip_rt_proc_init(void)
436 {
437         return register_pernet_subsys(&ip_rt_proc_ops);
438 }
439
440 #else
441 static inline int ip_rt_proc_init(void)
442 {
443         return 0;
444 }
445 #endif /* CONFIG_PROC_FS */
446
447 static inline bool rt_is_expired(const struct rtable *rth)
448 {
449         return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
450 }
451
452 /*
453  * Perturbation of rt_genid by a small quantity [1..256]
454  * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
455  * many times (2^24) without giving recent rt_genid.
456  * Jenkins hash is strong enough that litle changes of rt_genid are OK.
457  */
458 static void rt_cache_invalidate(struct net *net)
459 {
460         unsigned char shuffle;
461
462         get_random_bytes(&shuffle, sizeof(shuffle));
463         atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
464 }
465
466 /*
467  * delay < 0  : invalidate cache (fast : entries will be deleted later)
468  * delay >= 0 : invalidate & flush cache (can be long)
469  */
470 void rt_cache_flush(struct net *net, int delay)
471 {
472         rt_cache_invalidate(net);
473 }
474
475 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
476                                            struct sk_buff *skb,
477                                            const void *daddr)
478 {
479         struct net_device *dev = dst->dev;
480         const __be32 *pkey = daddr;
481         const struct rtable *rt;
482         struct neighbour *n;
483
484         rt = (const struct rtable *) dst;
485         if (rt->rt_gateway)
486                 pkey = (const __be32 *) &rt->rt_gateway;
487         else if (skb)
488                 pkey = &ip_hdr(skb)->daddr;
489
490         n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
491         if (n)
492                 return n;
493         return neigh_create(&arp_tbl, pkey, dev);
494 }
495
496 /*
497  * Peer allocation may fail only in serious out-of-memory conditions.  However
498  * we still can generate some output.
499  * Random ID selection looks a bit dangerous because we have no chances to
500  * select ID being unique in a reasonable period of time.
501  * But broken packet identifier may be better than no packet at all.
502  */
503 static void ip_select_fb_ident(struct iphdr *iph)
504 {
505         static DEFINE_SPINLOCK(ip_fb_id_lock);
506         static u32 ip_fallback_id;
507         u32 salt;
508
509         spin_lock_bh(&ip_fb_id_lock);
510         salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
511         iph->id = htons(salt & 0xFFFF);
512         ip_fallback_id = salt;
513         spin_unlock_bh(&ip_fb_id_lock);
514 }
515
516 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
517 {
518         struct net *net = dev_net(dst->dev);
519         struct inet_peer *peer;
520
521         peer = inet_getpeer_v4(net->ipv4.peers, iph->daddr, 1);
522         if (peer) {
523                 iph->id = htons(inet_getid(peer, more));
524                 inet_putpeer(peer);
525                 return;
526         }
527
528         ip_select_fb_ident(iph);
529 }
530 EXPORT_SYMBOL(__ip_select_ident);
531
532 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
533                              const struct iphdr *iph,
534                              int oif, u8 tos,
535                              u8 prot, u32 mark, int flow_flags)
536 {
537         if (sk) {
538                 const struct inet_sock *inet = inet_sk(sk);
539
540                 oif = sk->sk_bound_dev_if;
541                 mark = sk->sk_mark;
542                 tos = RT_CONN_FLAGS(sk);
543                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
544         }
545         flowi4_init_output(fl4, oif, mark, tos,
546                            RT_SCOPE_UNIVERSE, prot,
547                            flow_flags,
548                            iph->daddr, iph->saddr, 0, 0);
549 }
550
551 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
552                                const struct sock *sk)
553 {
554         const struct iphdr *iph = ip_hdr(skb);
555         int oif = skb->dev->ifindex;
556         u8 tos = RT_TOS(iph->tos);
557         u8 prot = iph->protocol;
558         u32 mark = skb->mark;
559
560         __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
561 }
562
563 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
564 {
565         const struct inet_sock *inet = inet_sk(sk);
566         const struct ip_options_rcu *inet_opt;
567         __be32 daddr = inet->inet_daddr;
568
569         rcu_read_lock();
570         inet_opt = rcu_dereference(inet->inet_opt);
571         if (inet_opt && inet_opt->opt.srr)
572                 daddr = inet_opt->opt.faddr;
573         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
574                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
575                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
576                            inet_sk_flowi_flags(sk),
577                            daddr, inet->inet_saddr, 0, 0);
578         rcu_read_unlock();
579 }
580
581 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
582                                  const struct sk_buff *skb)
583 {
584         if (skb)
585                 build_skb_flow_key(fl4, skb, sk);
586         else
587                 build_sk_flow_key(fl4, sk);
588 }
589
590 static DEFINE_SEQLOCK(fnhe_seqlock);
591
592 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
593 {
594         struct fib_nh_exception *fnhe, *oldest;
595
596         oldest = rcu_dereference(hash->chain);
597         for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
598              fnhe = rcu_dereference(fnhe->fnhe_next)) {
599                 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
600                         oldest = fnhe;
601         }
602         return oldest;
603 }
604
605 static inline u32 fnhe_hashfun(__be32 daddr)
606 {
607         u32 hval;
608
609         hval = (__force u32) daddr;
610         hval ^= (hval >> 11) ^ (hval >> 22);
611
612         return hval & (FNHE_HASH_SIZE - 1);
613 }
614
615 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
616                                   u32 pmtu, unsigned long expires)
617 {
618         struct fnhe_hash_bucket *hash;
619         struct fib_nh_exception *fnhe;
620         int depth;
621         u32 hval = fnhe_hashfun(daddr);
622
623         write_seqlock_bh(&fnhe_seqlock);
624
625         hash = nh->nh_exceptions;
626         if (!hash) {
627                 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
628                 if (!hash)
629                         goto out_unlock;
630                 nh->nh_exceptions = hash;
631         }
632
633         hash += hval;
634
635         depth = 0;
636         for (fnhe = rcu_dereference(hash->chain); fnhe;
637              fnhe = rcu_dereference(fnhe->fnhe_next)) {
638                 if (fnhe->fnhe_daddr == daddr)
639                         break;
640                 depth++;
641         }
642
643         if (fnhe) {
644                 if (gw)
645                         fnhe->fnhe_gw = gw;
646                 if (pmtu) {
647                         fnhe->fnhe_pmtu = pmtu;
648                         fnhe->fnhe_expires = expires;
649                 }
650         } else {
651                 if (depth > FNHE_RECLAIM_DEPTH)
652                         fnhe = fnhe_oldest(hash);
653                 else {
654                         fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
655                         if (!fnhe)
656                                 goto out_unlock;
657
658                         fnhe->fnhe_next = hash->chain;
659                         rcu_assign_pointer(hash->chain, fnhe);
660                 }
661                 fnhe->fnhe_daddr = daddr;
662                 fnhe->fnhe_gw = gw;
663                 fnhe->fnhe_pmtu = pmtu;
664                 fnhe->fnhe_expires = expires;
665         }
666
667         fnhe->fnhe_stamp = jiffies;
668
669 out_unlock:
670         write_sequnlock_bh(&fnhe_seqlock);
671         return;
672 }
673
674 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
675                              bool kill_route)
676 {
677         __be32 new_gw = icmp_hdr(skb)->un.gateway;
678         __be32 old_gw = ip_hdr(skb)->saddr;
679         struct net_device *dev = skb->dev;
680         struct in_device *in_dev;
681         struct fib_result res;
682         struct neighbour *n;
683         struct net *net;
684
685         switch (icmp_hdr(skb)->code & 7) {
686         case ICMP_REDIR_NET:
687         case ICMP_REDIR_NETTOS:
688         case ICMP_REDIR_HOST:
689         case ICMP_REDIR_HOSTTOS:
690                 break;
691
692         default:
693                 return;
694         }
695
696         if (rt->rt_gateway != old_gw)
697                 return;
698
699         in_dev = __in_dev_get_rcu(dev);
700         if (!in_dev)
701                 return;
702
703         net = dev_net(dev);
704         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
705             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
706             ipv4_is_zeronet(new_gw))
707                 goto reject_redirect;
708
709         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
710                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
711                         goto reject_redirect;
712                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
713                         goto reject_redirect;
714         } else {
715                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
716                         goto reject_redirect;
717         }
718
719         n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
720         if (n) {
721                 if (!(n->nud_state & NUD_VALID)) {
722                         neigh_event_send(n, NULL);
723                 } else {
724                         if (fib_lookup(net, fl4, &res) == 0) {
725                                 struct fib_nh *nh = &FIB_RES_NH(res);
726
727                                 update_or_create_fnhe(nh, fl4->daddr, new_gw,
728                                                       0, 0);
729                         }
730                         if (kill_route)
731                                 rt->dst.obsolete = DST_OBSOLETE_KILL;
732                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
733                 }
734                 neigh_release(n);
735         }
736         return;
737
738 reject_redirect:
739 #ifdef CONFIG_IP_ROUTE_VERBOSE
740         if (IN_DEV_LOG_MARTIANS(in_dev)) {
741                 const struct iphdr *iph = (const struct iphdr *) skb->data;
742                 __be32 daddr = iph->daddr;
743                 __be32 saddr = iph->saddr;
744
745                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
746                                      "  Advised path = %pI4 -> %pI4\n",
747                                      &old_gw, dev->name, &new_gw,
748                                      &saddr, &daddr);
749         }
750 #endif
751         ;
752 }
753
754 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
755 {
756         struct rtable *rt;
757         struct flowi4 fl4;
758
759         rt = (struct rtable *) dst;
760
761         ip_rt_build_flow_key(&fl4, sk, skb);
762         __ip_do_redirect(rt, skb, &fl4, true);
763 }
764
765 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
766 {
767         struct rtable *rt = (struct rtable *)dst;
768         struct dst_entry *ret = dst;
769
770         if (rt) {
771                 if (dst->obsolete > 0) {
772                         ip_rt_put(rt);
773                         ret = NULL;
774                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
775                            rt->dst.expires) {
776                         ip_rt_put(rt);
777                         ret = NULL;
778                 }
779         }
780         return ret;
781 }
782
783 /*
784  * Algorithm:
785  *      1. The first ip_rt_redirect_number redirects are sent
786  *         with exponential backoff, then we stop sending them at all,
787  *         assuming that the host ignores our redirects.
788  *      2. If we did not see packets requiring redirects
789  *         during ip_rt_redirect_silence, we assume that the host
790  *         forgot redirected route and start to send redirects again.
791  *
792  * This algorithm is much cheaper and more intelligent than dumb load limiting
793  * in icmp.c.
794  *
795  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
796  * and "frag. need" (breaks PMTU discovery) in icmp.c.
797  */
798
799 void ip_rt_send_redirect(struct sk_buff *skb)
800 {
801         struct rtable *rt = skb_rtable(skb);
802         struct in_device *in_dev;
803         struct inet_peer *peer;
804         struct net *net;
805         int log_martians;
806
807         rcu_read_lock();
808         in_dev = __in_dev_get_rcu(rt->dst.dev);
809         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
810                 rcu_read_unlock();
811                 return;
812         }
813         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
814         rcu_read_unlock();
815
816         net = dev_net(rt->dst.dev);
817         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
818         if (!peer) {
819                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
820                 return;
821         }
822
823         /* No redirected packets during ip_rt_redirect_silence;
824          * reset the algorithm.
825          */
826         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
827                 peer->rate_tokens = 0;
828
829         /* Too many ignored redirects; do not send anything
830          * set dst.rate_last to the last seen redirected packet.
831          */
832         if (peer->rate_tokens >= ip_rt_redirect_number) {
833                 peer->rate_last = jiffies;
834                 goto out_put_peer;
835         }
836
837         /* Check for load limit; set rate_last to the latest sent
838          * redirect.
839          */
840         if (peer->rate_tokens == 0 ||
841             time_after(jiffies,
842                        (peer->rate_last +
843                         (ip_rt_redirect_load << peer->rate_tokens)))) {
844                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
845                 peer->rate_last = jiffies;
846                 ++peer->rate_tokens;
847 #ifdef CONFIG_IP_ROUTE_VERBOSE
848                 if (log_martians &&
849                     peer->rate_tokens == ip_rt_redirect_number)
850                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
851                                              &ip_hdr(skb)->saddr, inet_iif(skb),
852                                              &ip_hdr(skb)->daddr, &rt->rt_gateway);
853 #endif
854         }
855 out_put_peer:
856         inet_putpeer(peer);
857 }
858
859 static int ip_error(struct sk_buff *skb)
860 {
861         struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
862         struct rtable *rt = skb_rtable(skb);
863         struct inet_peer *peer;
864         unsigned long now;
865         struct net *net;
866         bool send;
867         int code;
868
869         net = dev_net(rt->dst.dev);
870         if (!IN_DEV_FORWARD(in_dev)) {
871                 switch (rt->dst.error) {
872                 case EHOSTUNREACH:
873                         IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
874                         break;
875
876                 case ENETUNREACH:
877                         IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
878                         break;
879                 }
880                 goto out;
881         }
882
883         switch (rt->dst.error) {
884         case EINVAL:
885         default:
886                 goto out;
887         case EHOSTUNREACH:
888                 code = ICMP_HOST_UNREACH;
889                 break;
890         case ENETUNREACH:
891                 code = ICMP_NET_UNREACH;
892                 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
893                 break;
894         case EACCES:
895                 code = ICMP_PKT_FILTERED;
896                 break;
897         }
898
899         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
900
901         send = true;
902         if (peer) {
903                 now = jiffies;
904                 peer->rate_tokens += now - peer->rate_last;
905                 if (peer->rate_tokens > ip_rt_error_burst)
906                         peer->rate_tokens = ip_rt_error_burst;
907                 peer->rate_last = now;
908                 if (peer->rate_tokens >= ip_rt_error_cost)
909                         peer->rate_tokens -= ip_rt_error_cost;
910                 else
911                         send = false;
912                 inet_putpeer(peer);
913         }
914         if (send)
915                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
916
917 out:    kfree_skb(skb);
918         return 0;
919 }
920
921 static u32 __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
922 {
923         struct fib_result res;
924
925         if (mtu < ip_rt_min_pmtu)
926                 mtu = ip_rt_min_pmtu;
927
928         if (fib_lookup(dev_net(rt->dst.dev), fl4, &res) == 0) {
929                 struct fib_nh *nh = &FIB_RES_NH(res);
930
931                 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
932                                       jiffies + ip_rt_mtu_expires);
933         }
934         return mtu;
935 }
936
937 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
938                               struct sk_buff *skb, u32 mtu)
939 {
940         struct rtable *rt = (struct rtable *) dst;
941         struct flowi4 fl4;
942
943         ip_rt_build_flow_key(&fl4, sk, skb);
944         mtu = __ip_rt_update_pmtu(rt, &fl4, mtu);
945
946         if (!rt->rt_pmtu) {
947                 dst->obsolete = DST_OBSOLETE_KILL;
948         } else {
949                 rt->rt_pmtu = mtu;
950                 dst_set_expires(&rt->dst, ip_rt_mtu_expires);
951         }
952 }
953
954 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
955                       int oif, u32 mark, u8 protocol, int flow_flags)
956 {
957         const struct iphdr *iph = (const struct iphdr *) skb->data;
958         struct flowi4 fl4;
959         struct rtable *rt;
960
961         __build_flow_key(&fl4, NULL, iph, oif,
962                          RT_TOS(iph->tos), protocol, mark, flow_flags);
963         rt = __ip_route_output_key(net, &fl4);
964         if (!IS_ERR(rt)) {
965                 __ip_rt_update_pmtu(rt, &fl4, mtu);
966                 ip_rt_put(rt);
967         }
968 }
969 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
970
971 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
972 {
973         const struct iphdr *iph = (const struct iphdr *) skb->data;
974         struct flowi4 fl4;
975         struct rtable *rt;
976
977         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
978         rt = __ip_route_output_key(sock_net(sk), &fl4);
979         if (!IS_ERR(rt)) {
980                 __ip_rt_update_pmtu(rt, &fl4, mtu);
981                 ip_rt_put(rt);
982         }
983 }
984 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
985
986 void ipv4_redirect(struct sk_buff *skb, struct net *net,
987                    int oif, u32 mark, u8 protocol, int flow_flags)
988 {
989         const struct iphdr *iph = (const struct iphdr *) skb->data;
990         struct flowi4 fl4;
991         struct rtable *rt;
992
993         __build_flow_key(&fl4, NULL, iph, oif,
994                          RT_TOS(iph->tos), protocol, mark, flow_flags);
995         rt = __ip_route_output_key(net, &fl4);
996         if (!IS_ERR(rt)) {
997                 __ip_do_redirect(rt, skb, &fl4, false);
998                 ip_rt_put(rt);
999         }
1000 }
1001 EXPORT_SYMBOL_GPL(ipv4_redirect);
1002
1003 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1004 {
1005         const struct iphdr *iph = (const struct iphdr *) skb->data;
1006         struct flowi4 fl4;
1007         struct rtable *rt;
1008
1009         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1010         rt = __ip_route_output_key(sock_net(sk), &fl4);
1011         if (!IS_ERR(rt)) {
1012                 __ip_do_redirect(rt, skb, &fl4, false);
1013                 ip_rt_put(rt);
1014         }
1015 }
1016 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1017
1018 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1019 {
1020         struct rtable *rt = (struct rtable *) dst;
1021
1022         /* All IPV4 dsts are created with ->obsolete set to the value
1023          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1024          * into this function always.
1025          *
1026          * When a PMTU/redirect information update invalidates a
1027          * route, this is indicated by setting obsolete to
1028          * DST_OBSOLETE_KILL.
1029          */
1030         if (dst->obsolete == DST_OBSOLETE_KILL || rt_is_expired(rt))
1031                 return NULL;
1032         return dst;
1033 }
1034
1035 static void ipv4_link_failure(struct sk_buff *skb)
1036 {
1037         struct rtable *rt;
1038
1039         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1040
1041         rt = skb_rtable(skb);
1042         if (rt)
1043                 dst_set_expires(&rt->dst, 0);
1044 }
1045
1046 static int ip_rt_bug(struct sk_buff *skb)
1047 {
1048         pr_debug("%s: %pI4 -> %pI4, %s\n",
1049                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1050                  skb->dev ? skb->dev->name : "?");
1051         kfree_skb(skb);
1052         WARN_ON(1);
1053         return 0;
1054 }
1055
1056 /*
1057    We do not cache source address of outgoing interface,
1058    because it is used only by IP RR, TS and SRR options,
1059    so that it out of fast path.
1060
1061    BTW remember: "addr" is allowed to be not aligned
1062    in IP options!
1063  */
1064
1065 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1066 {
1067         __be32 src;
1068
1069         if (rt_is_output_route(rt))
1070                 src = ip_hdr(skb)->saddr;
1071         else {
1072                 struct fib_result res;
1073                 struct flowi4 fl4;
1074                 struct iphdr *iph;
1075
1076                 iph = ip_hdr(skb);
1077
1078                 memset(&fl4, 0, sizeof(fl4));
1079                 fl4.daddr = iph->daddr;
1080                 fl4.saddr = iph->saddr;
1081                 fl4.flowi4_tos = RT_TOS(iph->tos);
1082                 fl4.flowi4_oif = rt->dst.dev->ifindex;
1083                 fl4.flowi4_iif = skb->dev->ifindex;
1084                 fl4.flowi4_mark = skb->mark;
1085
1086                 rcu_read_lock();
1087                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1088                         src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1089                 else
1090                         src = inet_select_addr(rt->dst.dev,
1091                                                rt_nexthop(rt, iph->daddr),
1092                                                RT_SCOPE_UNIVERSE);
1093                 rcu_read_unlock();
1094         }
1095         memcpy(addr, &src, 4);
1096 }
1097
1098 #ifdef CONFIG_IP_ROUTE_CLASSID
1099 static void set_class_tag(struct rtable *rt, u32 tag)
1100 {
1101         if (!(rt->dst.tclassid & 0xFFFF))
1102                 rt->dst.tclassid |= tag & 0xFFFF;
1103         if (!(rt->dst.tclassid & 0xFFFF0000))
1104                 rt->dst.tclassid |= tag & 0xFFFF0000;
1105 }
1106 #endif
1107
1108 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1109 {
1110         unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1111
1112         if (advmss == 0) {
1113                 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1114                                ip_rt_min_advmss);
1115                 if (advmss > 65535 - 40)
1116                         advmss = 65535 - 40;
1117         }
1118         return advmss;
1119 }
1120
1121 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1122 {
1123         const struct rtable *rt = (const struct rtable *) dst;
1124         unsigned int mtu = rt->rt_pmtu;
1125
1126         if (mtu && time_after_eq(jiffies, rt->dst.expires))
1127                 mtu = 0;
1128
1129         if (!mtu)
1130                 mtu = dst_metric_raw(dst, RTAX_MTU);
1131
1132         if (mtu && rt_is_output_route(rt))
1133                 return mtu;
1134
1135         mtu = dst->dev->mtu;
1136
1137         if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1138                 if (rt->rt_gateway && mtu > 576)
1139                         mtu = 576;
1140         }
1141
1142         if (mtu > IP_MAX_MTU)
1143                 mtu = IP_MAX_MTU;
1144
1145         return mtu;
1146 }
1147
1148 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1149 {
1150         struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1151         struct fib_nh_exception *fnhe;
1152         u32 hval;
1153
1154         if (!hash)
1155                 return NULL;
1156
1157         hval = fnhe_hashfun(daddr);
1158
1159         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1160              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1161                 if (fnhe->fnhe_daddr == daddr)
1162                         return fnhe;
1163         }
1164         return NULL;
1165 }
1166
1167 static void rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1168                               __be32 daddr)
1169 {
1170         __be32 fnhe_daddr, gw;
1171         unsigned long expires;
1172         unsigned int seq;
1173         u32 pmtu;
1174
1175 restart:
1176         seq = read_seqbegin(&fnhe_seqlock);
1177         fnhe_daddr = fnhe->fnhe_daddr;
1178         gw = fnhe->fnhe_gw;
1179         pmtu = fnhe->fnhe_pmtu;
1180         expires = fnhe->fnhe_expires;
1181         if (read_seqretry(&fnhe_seqlock, seq))
1182                 goto restart;
1183
1184         if (daddr != fnhe_daddr)
1185                 return;
1186
1187         if (pmtu) {
1188                 unsigned long diff = expires - jiffies;
1189
1190                 if (time_before(jiffies, expires)) {
1191                         rt->rt_pmtu = pmtu;
1192                         dst_set_expires(&rt->dst, diff);
1193                 }
1194         }
1195         if (gw) {
1196                 rt->rt_flags |= RTCF_REDIRECTED;
1197                 rt->rt_gateway = gw;
1198         }
1199         fnhe->fnhe_stamp = jiffies;
1200 }
1201
1202 static inline void rt_free(struct rtable *rt)
1203 {
1204         call_rcu(&rt->dst.rcu_head, dst_rcu_free);
1205 }
1206
1207 static void rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1208 {
1209         struct rtable *orig, *prev, **p;
1210
1211         if (rt_is_input_route(rt)) {
1212                 p = (struct rtable **)&nh->nh_rth_input;
1213         } else {
1214                 if (!nh->nh_pcpu_rth_output)
1215                         goto nocache;
1216                 p = (struct rtable **)__this_cpu_ptr(nh->nh_pcpu_rth_output);
1217         }
1218         orig = *p;
1219
1220         prev = cmpxchg(p, orig, rt);
1221         if (prev == orig) {
1222                 if (orig)
1223                         rt_free(orig);
1224         } else {
1225                 /* Routes we intend to cache in the FIB nexthop have
1226                  * the DST_NOCACHE bit clear.  However, if we are
1227                  * unsuccessful at storing this route into the cache
1228                  * we really need to set it.
1229                  */
1230 nocache:
1231                 rt->dst.flags |= DST_NOCACHE;
1232         }
1233 }
1234
1235 static bool rt_cache_valid(const struct rtable *rt)
1236 {
1237         return  rt &&
1238                 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1239                 !rt_is_expired(rt);
1240 }
1241
1242 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1243                            const struct fib_result *res,
1244                            struct fib_nh_exception *fnhe,
1245                            struct fib_info *fi, u16 type, u32 itag)
1246 {
1247         if (fi) {
1248                 struct fib_nh *nh = &FIB_RES_NH(*res);
1249
1250                 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK)
1251                         rt->rt_gateway = nh->nh_gw;
1252                 if (unlikely(fnhe))
1253                         rt_bind_exception(rt, fnhe, daddr);
1254                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1255 #ifdef CONFIG_IP_ROUTE_CLASSID
1256                 rt->dst.tclassid = nh->nh_tclassid;
1257 #endif
1258                 if (!(rt->dst.flags & DST_NOCACHE))
1259                         rt_cache_route(nh, rt);
1260         }
1261
1262 #ifdef CONFIG_IP_ROUTE_CLASSID
1263 #ifdef CONFIG_IP_MULTIPLE_TABLES
1264         set_class_tag(rt, res->tclassid);
1265 #endif
1266         set_class_tag(rt, itag);
1267 #endif
1268 }
1269
1270 static struct rtable *rt_dst_alloc(struct net_device *dev,
1271                                    bool nopolicy, bool noxfrm, bool will_cache)
1272 {
1273         return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1274                          (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1275                          (nopolicy ? DST_NOPOLICY : 0) |
1276                          (noxfrm ? DST_NOXFRM : 0));
1277 }
1278
1279 /* called in rcu_read_lock() section */
1280 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1281                                 u8 tos, struct net_device *dev, int our)
1282 {
1283         struct rtable *rth;
1284         struct in_device *in_dev = __in_dev_get_rcu(dev);
1285         u32 itag = 0;
1286         int err;
1287
1288         /* Primary sanity checks. */
1289
1290         if (in_dev == NULL)
1291                 return -EINVAL;
1292
1293         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1294             skb->protocol != htons(ETH_P_IP))
1295                 goto e_inval;
1296
1297         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1298                 if (ipv4_is_loopback(saddr))
1299                         goto e_inval;
1300
1301         if (ipv4_is_zeronet(saddr)) {
1302                 if (!ipv4_is_local_multicast(daddr))
1303                         goto e_inval;
1304         } else {
1305                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1306                                           in_dev, &itag);
1307                 if (err < 0)
1308                         goto e_err;
1309         }
1310         rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1311                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1312         if (!rth)
1313                 goto e_nobufs;
1314
1315 #ifdef CONFIG_IP_ROUTE_CLASSID
1316         rth->dst.tclassid = itag;
1317 #endif
1318         rth->dst.output = ip_rt_bug;
1319
1320         rth->rt_genid   = rt_genid(dev_net(dev));
1321         rth->rt_flags   = RTCF_MULTICAST;
1322         rth->rt_type    = RTN_MULTICAST;
1323         rth->rt_is_input= 1;
1324         rth->rt_iif     = 0;
1325         rth->rt_pmtu    = 0;
1326         rth->rt_gateway = 0;
1327         if (our) {
1328                 rth->dst.input= ip_local_deliver;
1329                 rth->rt_flags |= RTCF_LOCAL;
1330         }
1331
1332 #ifdef CONFIG_IP_MROUTE
1333         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1334                 rth->dst.input = ip_mr_input;
1335 #endif
1336         RT_CACHE_STAT_INC(in_slow_mc);
1337
1338         skb_dst_set(skb, &rth->dst);
1339         return 0;
1340
1341 e_nobufs:
1342         return -ENOBUFS;
1343 e_inval:
1344         return -EINVAL;
1345 e_err:
1346         return err;
1347 }
1348
1349
1350 static void ip_handle_martian_source(struct net_device *dev,
1351                                      struct in_device *in_dev,
1352                                      struct sk_buff *skb,
1353                                      __be32 daddr,
1354                                      __be32 saddr)
1355 {
1356         RT_CACHE_STAT_INC(in_martian_src);
1357 #ifdef CONFIG_IP_ROUTE_VERBOSE
1358         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1359                 /*
1360                  *      RFC1812 recommendation, if source is martian,
1361                  *      the only hint is MAC header.
1362                  */
1363                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1364                         &daddr, &saddr, dev->name);
1365                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1366                         print_hex_dump(KERN_WARNING, "ll header: ",
1367                                        DUMP_PREFIX_OFFSET, 16, 1,
1368                                        skb_mac_header(skb),
1369                                        dev->hard_header_len, true);
1370                 }
1371         }
1372 #endif
1373 }
1374
1375 /* called in rcu_read_lock() section */
1376 static int __mkroute_input(struct sk_buff *skb,
1377                            const struct fib_result *res,
1378                            struct in_device *in_dev,
1379                            __be32 daddr, __be32 saddr, u32 tos)
1380 {
1381         struct rtable *rth;
1382         int err;
1383         struct in_device *out_dev;
1384         unsigned int flags = 0;
1385         bool do_cache;
1386         u32 itag;
1387
1388         /* get a working reference to the output device */
1389         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1390         if (out_dev == NULL) {
1391                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1392                 return -EINVAL;
1393         }
1394
1395
1396         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1397                                   in_dev->dev, in_dev, &itag);
1398         if (err < 0) {
1399                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1400                                          saddr);
1401
1402                 goto cleanup;
1403         }
1404
1405         if (out_dev == in_dev && err &&
1406             (IN_DEV_SHARED_MEDIA(out_dev) ||
1407              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1408                 flags |= RTCF_DOREDIRECT;
1409
1410         if (skb->protocol != htons(ETH_P_IP)) {
1411                 /* Not IP (i.e. ARP). Do not create route, if it is
1412                  * invalid for proxy arp. DNAT routes are always valid.
1413                  *
1414                  * Proxy arp feature have been extended to allow, ARP
1415                  * replies back to the same interface, to support
1416                  * Private VLAN switch technologies. See arp.c.
1417                  */
1418                 if (out_dev == in_dev &&
1419                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1420                         err = -EINVAL;
1421                         goto cleanup;
1422                 }
1423         }
1424
1425         do_cache = false;
1426         if (res->fi) {
1427                 if (!itag) {
1428                         rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1429                         if (rt_cache_valid(rth)) {
1430                                 skb_dst_set_noref(skb, &rth->dst);
1431                                 goto out;
1432                         }
1433                         do_cache = true;
1434                 }
1435         }
1436
1437         rth = rt_dst_alloc(out_dev->dev,
1438                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1439                            IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1440         if (!rth) {
1441                 err = -ENOBUFS;
1442                 goto cleanup;
1443         }
1444
1445         rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
1446         rth->rt_flags = flags;
1447         rth->rt_type = res->type;
1448         rth->rt_is_input = 1;
1449         rth->rt_iif     = 0;
1450         rth->rt_pmtu    = 0;
1451         rth->rt_gateway = 0;
1452
1453         rth->dst.input = ip_forward;
1454         rth->dst.output = ip_output;
1455
1456         rt_set_nexthop(rth, daddr, res, NULL, res->fi, res->type, itag);
1457         skb_dst_set(skb, &rth->dst);
1458 out:
1459         err = 0;
1460  cleanup:
1461         return err;
1462 }
1463
1464 static int ip_mkroute_input(struct sk_buff *skb,
1465                             struct fib_result *res,
1466                             const struct flowi4 *fl4,
1467                             struct in_device *in_dev,
1468                             __be32 daddr, __be32 saddr, u32 tos)
1469 {
1470 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1471         if (res->fi && res->fi->fib_nhs > 1)
1472                 fib_select_multipath(res);
1473 #endif
1474
1475         /* create a routing cache entry */
1476         return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1477 }
1478
1479 /*
1480  *      NOTE. We drop all the packets that has local source
1481  *      addresses, because every properly looped back packet
1482  *      must have correct destination already attached by output routine.
1483  *
1484  *      Such approach solves two big problems:
1485  *      1. Not simplex devices are handled properly.
1486  *      2. IP spoofing attempts are filtered with 100% of guarantee.
1487  *      called with rcu_read_lock()
1488  */
1489
1490 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1491                                u8 tos, struct net_device *dev)
1492 {
1493         struct fib_result res;
1494         struct in_device *in_dev = __in_dev_get_rcu(dev);
1495         struct flowi4   fl4;
1496         unsigned int    flags = 0;
1497         u32             itag = 0;
1498         struct rtable   *rth;
1499         int             err = -EINVAL;
1500         struct net    *net = dev_net(dev);
1501         bool do_cache;
1502
1503         /* IP on this device is disabled. */
1504
1505         if (!in_dev)
1506                 goto out;
1507
1508         /* Check for the most weird martians, which can be not detected
1509            by fib_lookup.
1510          */
1511
1512         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1513                 goto martian_source;
1514
1515         res.fi = NULL;
1516         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1517                 goto brd_input;
1518
1519         /* Accept zero addresses only to limited broadcast;
1520          * I even do not know to fix it or not. Waiting for complains :-)
1521          */
1522         if (ipv4_is_zeronet(saddr))
1523                 goto martian_source;
1524
1525         if (ipv4_is_zeronet(daddr))
1526                 goto martian_destination;
1527
1528         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) {
1529                 if (ipv4_is_loopback(daddr))
1530                         goto martian_destination;
1531
1532                 if (ipv4_is_loopback(saddr))
1533                         goto martian_source;
1534         }
1535
1536         /*
1537          *      Now we are ready to route packet.
1538          */
1539         fl4.flowi4_oif = 0;
1540         fl4.flowi4_iif = dev->ifindex;
1541         fl4.flowi4_mark = skb->mark;
1542         fl4.flowi4_tos = tos;
1543         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1544         fl4.daddr = daddr;
1545         fl4.saddr = saddr;
1546         err = fib_lookup(net, &fl4, &res);
1547         if (err != 0)
1548                 goto no_route;
1549
1550         RT_CACHE_STAT_INC(in_slow_tot);
1551
1552         if (res.type == RTN_BROADCAST)
1553                 goto brd_input;
1554
1555         if (res.type == RTN_LOCAL) {
1556                 err = fib_validate_source(skb, saddr, daddr, tos,
1557                                           net->loopback_dev->ifindex,
1558                                           dev, in_dev, &itag);
1559                 if (err < 0)
1560                         goto martian_source_keep_err;
1561                 goto local_input;
1562         }
1563
1564         if (!IN_DEV_FORWARD(in_dev))
1565                 goto no_route;
1566         if (res.type != RTN_UNICAST)
1567                 goto martian_destination;
1568
1569         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1570 out:    return err;
1571
1572 brd_input:
1573         if (skb->protocol != htons(ETH_P_IP))
1574                 goto e_inval;
1575
1576         if (!ipv4_is_zeronet(saddr)) {
1577                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1578                                           in_dev, &itag);
1579                 if (err < 0)
1580                         goto martian_source_keep_err;
1581         }
1582         flags |= RTCF_BROADCAST;
1583         res.type = RTN_BROADCAST;
1584         RT_CACHE_STAT_INC(in_brd);
1585
1586 local_input:
1587         do_cache = false;
1588         if (res.fi) {
1589                 if (!itag) {
1590                         rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1591                         if (rt_cache_valid(rth)) {
1592                                 skb_dst_set_noref(skb, &rth->dst);
1593                                 err = 0;
1594                                 goto out;
1595                         }
1596                         do_cache = true;
1597                 }
1598         }
1599
1600         rth = rt_dst_alloc(net->loopback_dev,
1601                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1602         if (!rth)
1603                 goto e_nobufs;
1604
1605         rth->dst.input= ip_local_deliver;
1606         rth->dst.output= ip_rt_bug;
1607 #ifdef CONFIG_IP_ROUTE_CLASSID
1608         rth->dst.tclassid = itag;
1609 #endif
1610
1611         rth->rt_genid = rt_genid(net);
1612         rth->rt_flags   = flags|RTCF_LOCAL;
1613         rth->rt_type    = res.type;
1614         rth->rt_is_input = 1;
1615         rth->rt_iif     = 0;
1616         rth->rt_pmtu    = 0;
1617         rth->rt_gateway = 0;
1618         if (res.type == RTN_UNREACHABLE) {
1619                 rth->dst.input= ip_error;
1620                 rth->dst.error= -err;
1621                 rth->rt_flags   &= ~RTCF_LOCAL;
1622         }
1623         if (do_cache)
1624                 rt_cache_route(&FIB_RES_NH(res), rth);
1625         skb_dst_set(skb, &rth->dst);
1626         err = 0;
1627         goto out;
1628
1629 no_route:
1630         RT_CACHE_STAT_INC(in_no_route);
1631         res.type = RTN_UNREACHABLE;
1632         if (err == -ESRCH)
1633                 err = -ENETUNREACH;
1634         goto local_input;
1635
1636         /*
1637          *      Do not cache martian addresses: they should be logged (RFC1812)
1638          */
1639 martian_destination:
1640         RT_CACHE_STAT_INC(in_martian_dst);
1641 #ifdef CONFIG_IP_ROUTE_VERBOSE
1642         if (IN_DEV_LOG_MARTIANS(in_dev))
1643                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1644                                      &daddr, &saddr, dev->name);
1645 #endif
1646
1647 e_inval:
1648         err = -EINVAL;
1649         goto out;
1650
1651 e_nobufs:
1652         err = -ENOBUFS;
1653         goto out;
1654
1655 martian_source:
1656         err = -EINVAL;
1657 martian_source_keep_err:
1658         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1659         goto out;
1660 }
1661
1662 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1663                          u8 tos, struct net_device *dev)
1664 {
1665         int res;
1666
1667         rcu_read_lock();
1668
1669         /* Multicast recognition logic is moved from route cache to here.
1670            The problem was that too many Ethernet cards have broken/missing
1671            hardware multicast filters :-( As result the host on multicasting
1672            network acquires a lot of useless route cache entries, sort of
1673            SDR messages from all the world. Now we try to get rid of them.
1674            Really, provided software IP multicast filter is organized
1675            reasonably (at least, hashed), it does not result in a slowdown
1676            comparing with route cache reject entries.
1677            Note, that multicast routers are not affected, because
1678            route cache entry is created eventually.
1679          */
1680         if (ipv4_is_multicast(daddr)) {
1681                 struct in_device *in_dev = __in_dev_get_rcu(dev);
1682
1683                 if (in_dev) {
1684                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1685                                                   ip_hdr(skb)->protocol);
1686                         if (our
1687 #ifdef CONFIG_IP_MROUTE
1688                                 ||
1689                             (!ipv4_is_local_multicast(daddr) &&
1690                              IN_DEV_MFORWARD(in_dev))
1691 #endif
1692                            ) {
1693                                 int res = ip_route_input_mc(skb, daddr, saddr,
1694                                                             tos, dev, our);
1695                                 rcu_read_unlock();
1696                                 return res;
1697                         }
1698                 }
1699                 rcu_read_unlock();
1700                 return -EINVAL;
1701         }
1702         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1703         rcu_read_unlock();
1704         return res;
1705 }
1706 EXPORT_SYMBOL(ip_route_input_noref);
1707
1708 /* called with rcu_read_lock() */
1709 static struct rtable *__mkroute_output(const struct fib_result *res,
1710                                        const struct flowi4 *fl4, int orig_oif,
1711                                        struct net_device *dev_out,
1712                                        unsigned int flags)
1713 {
1714         struct fib_info *fi = res->fi;
1715         struct fib_nh_exception *fnhe;
1716         struct in_device *in_dev;
1717         u16 type = res->type;
1718         struct rtable *rth;
1719
1720         in_dev = __in_dev_get_rcu(dev_out);
1721         if (!in_dev)
1722                 return ERR_PTR(-EINVAL);
1723
1724         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1725                 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1726                         return ERR_PTR(-EINVAL);
1727
1728         if (ipv4_is_lbcast(fl4->daddr))
1729                 type = RTN_BROADCAST;
1730         else if (ipv4_is_multicast(fl4->daddr))
1731                 type = RTN_MULTICAST;
1732         else if (ipv4_is_zeronet(fl4->daddr))
1733                 return ERR_PTR(-EINVAL);
1734
1735         if (dev_out->flags & IFF_LOOPBACK)
1736                 flags |= RTCF_LOCAL;
1737
1738         if (type == RTN_BROADCAST) {
1739                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1740                 fi = NULL;
1741         } else if (type == RTN_MULTICAST) {
1742                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1743                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1744                                      fl4->flowi4_proto))
1745                         flags &= ~RTCF_LOCAL;
1746                 /* If multicast route do not exist use
1747                  * default one, but do not gateway in this case.
1748                  * Yes, it is hack.
1749                  */
1750                 if (fi && res->prefixlen < 4)
1751                         fi = NULL;
1752         }
1753
1754         fnhe = NULL;
1755         if (fi) {
1756                 fnhe = find_exception(&FIB_RES_NH(*res), fl4->daddr);
1757                 if (!fnhe && FIB_RES_NH(*res).nh_pcpu_rth_output) {
1758                         struct rtable __rcu **prth;
1759
1760                         prth = __this_cpu_ptr(FIB_RES_NH(*res).nh_pcpu_rth_output);
1761                         rth = rcu_dereference(*prth);
1762                         if (rt_cache_valid(rth)) {
1763                                 dst_hold(&rth->dst);
1764                                 return rth;
1765                         }
1766                 }
1767         }
1768         rth = rt_dst_alloc(dev_out,
1769                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1770                            IN_DEV_CONF_GET(in_dev, NOXFRM),
1771                            fi && !fnhe);
1772         if (!rth)
1773                 return ERR_PTR(-ENOBUFS);
1774
1775         rth->dst.output = ip_output;
1776
1777         rth->rt_genid = rt_genid(dev_net(dev_out));
1778         rth->rt_flags   = flags;
1779         rth->rt_type    = type;
1780         rth->rt_is_input = 0;
1781         rth->rt_iif     = orig_oif ? : 0;
1782         rth->rt_pmtu    = 0;
1783         rth->rt_gateway = 0;
1784
1785         RT_CACHE_STAT_INC(out_slow_tot);
1786
1787         if (flags & RTCF_LOCAL)
1788                 rth->dst.input = ip_local_deliver;
1789         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1790                 if (flags & RTCF_LOCAL &&
1791                     !(dev_out->flags & IFF_LOOPBACK)) {
1792                         rth->dst.output = ip_mc_output;
1793                         RT_CACHE_STAT_INC(out_slow_mc);
1794                 }
1795 #ifdef CONFIG_IP_MROUTE
1796                 if (type == RTN_MULTICAST) {
1797                         if (IN_DEV_MFORWARD(in_dev) &&
1798                             !ipv4_is_local_multicast(fl4->daddr)) {
1799                                 rth->dst.input = ip_mr_input;
1800                                 rth->dst.output = ip_mc_output;
1801                         }
1802                 }
1803 #endif
1804         }
1805
1806         rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
1807
1808         return rth;
1809 }
1810
1811 /*
1812  * Major route resolver routine.
1813  */
1814
1815 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
1816 {
1817         struct net_device *dev_out = NULL;
1818         __u8 tos = RT_FL_TOS(fl4);
1819         unsigned int flags = 0;
1820         struct fib_result res;
1821         struct rtable *rth;
1822         int orig_oif;
1823
1824         res.tclassid    = 0;
1825         res.fi          = NULL;
1826         res.table       = NULL;
1827
1828         orig_oif = fl4->flowi4_oif;
1829
1830         fl4->flowi4_iif = net->loopback_dev->ifindex;
1831         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
1832         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
1833                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
1834
1835         rcu_read_lock();
1836         if (fl4->saddr) {
1837                 rth = ERR_PTR(-EINVAL);
1838                 if (ipv4_is_multicast(fl4->saddr) ||
1839                     ipv4_is_lbcast(fl4->saddr) ||
1840                     ipv4_is_zeronet(fl4->saddr))
1841                         goto out;
1842
1843                 /* I removed check for oif == dev_out->oif here.
1844                    It was wrong for two reasons:
1845                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
1846                       is assigned to multiple interfaces.
1847                    2. Moreover, we are allowed to send packets with saddr
1848                       of another iface. --ANK
1849                  */
1850
1851                 if (fl4->flowi4_oif == 0 &&
1852                     (ipv4_is_multicast(fl4->daddr) ||
1853                      ipv4_is_lbcast(fl4->daddr))) {
1854                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1855                         dev_out = __ip_dev_find(net, fl4->saddr, false);
1856                         if (dev_out == NULL)
1857                                 goto out;
1858
1859                         /* Special hack: user can direct multicasts
1860                            and limited broadcast via necessary interface
1861                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
1862                            This hack is not just for fun, it allows
1863                            vic,vat and friends to work.
1864                            They bind socket to loopback, set ttl to zero
1865                            and expect that it will work.
1866                            From the viewpoint of routing cache they are broken,
1867                            because we are not allowed to build multicast path
1868                            with loopback source addr (look, routing cache
1869                            cannot know, that ttl is zero, so that packet
1870                            will not leave this host and route is valid).
1871                            Luckily, this hack is good workaround.
1872                          */
1873
1874                         fl4->flowi4_oif = dev_out->ifindex;
1875                         goto make_route;
1876                 }
1877
1878                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
1879                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1880                         if (!__ip_dev_find(net, fl4->saddr, false))
1881                                 goto out;
1882                 }
1883         }
1884
1885
1886         if (fl4->flowi4_oif) {
1887                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
1888                 rth = ERR_PTR(-ENODEV);
1889                 if (dev_out == NULL)
1890                         goto out;
1891
1892                 /* RACE: Check return value of inet_select_addr instead. */
1893                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
1894                         rth = ERR_PTR(-ENETUNREACH);
1895                         goto out;
1896                 }
1897                 if (ipv4_is_local_multicast(fl4->daddr) ||
1898                     ipv4_is_lbcast(fl4->daddr)) {
1899                         if (!fl4->saddr)
1900                                 fl4->saddr = inet_select_addr(dev_out, 0,
1901                                                               RT_SCOPE_LINK);
1902                         goto make_route;
1903                 }
1904                 if (fl4->saddr) {
1905                         if (ipv4_is_multicast(fl4->daddr))
1906                                 fl4->saddr = inet_select_addr(dev_out, 0,
1907                                                               fl4->flowi4_scope);
1908                         else if (!fl4->daddr)
1909                                 fl4->saddr = inet_select_addr(dev_out, 0,
1910                                                               RT_SCOPE_HOST);
1911                 }
1912         }
1913
1914         if (!fl4->daddr) {
1915                 fl4->daddr = fl4->saddr;
1916                 if (!fl4->daddr)
1917                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
1918                 dev_out = net->loopback_dev;
1919                 fl4->flowi4_oif = net->loopback_dev->ifindex;
1920                 res.type = RTN_LOCAL;
1921                 flags |= RTCF_LOCAL;
1922                 goto make_route;
1923         }
1924
1925         if (fib_lookup(net, fl4, &res)) {
1926                 res.fi = NULL;
1927                 res.table = NULL;
1928                 if (fl4->flowi4_oif) {
1929                         /* Apparently, routing tables are wrong. Assume,
1930                            that the destination is on link.
1931
1932                            WHY? DW.
1933                            Because we are allowed to send to iface
1934                            even if it has NO routes and NO assigned
1935                            addresses. When oif is specified, routing
1936                            tables are looked up with only one purpose:
1937                            to catch if destination is gatewayed, rather than
1938                            direct. Moreover, if MSG_DONTROUTE is set,
1939                            we send packet, ignoring both routing tables
1940                            and ifaddr state. --ANK
1941
1942
1943                            We could make it even if oif is unknown,
1944                            likely IPv6, but we do not.
1945                          */
1946
1947                         if (fl4->saddr == 0)
1948                                 fl4->saddr = inet_select_addr(dev_out, 0,
1949                                                               RT_SCOPE_LINK);
1950                         res.type = RTN_UNICAST;
1951                         goto make_route;
1952                 }
1953                 rth = ERR_PTR(-ENETUNREACH);
1954                 goto out;
1955         }
1956
1957         if (res.type == RTN_LOCAL) {
1958                 if (!fl4->saddr) {
1959                         if (res.fi->fib_prefsrc)
1960                                 fl4->saddr = res.fi->fib_prefsrc;
1961                         else
1962                                 fl4->saddr = fl4->daddr;
1963                 }
1964                 dev_out = net->loopback_dev;
1965                 fl4->flowi4_oif = dev_out->ifindex;
1966                 res.fi = NULL;
1967                 flags |= RTCF_LOCAL;
1968                 goto make_route;
1969         }
1970
1971 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1972         if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
1973                 fib_select_multipath(&res);
1974         else
1975 #endif
1976         if (!res.prefixlen &&
1977             res.table->tb_num_default > 1 &&
1978             res.type == RTN_UNICAST && !fl4->flowi4_oif)
1979                 fib_select_default(&res);
1980
1981         if (!fl4->saddr)
1982                 fl4->saddr = FIB_RES_PREFSRC(net, res);
1983
1984         dev_out = FIB_RES_DEV(res);
1985         fl4->flowi4_oif = dev_out->ifindex;
1986
1987
1988 make_route:
1989         rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
1990
1991 out:
1992         rcu_read_unlock();
1993         return rth;
1994 }
1995 EXPORT_SYMBOL_GPL(__ip_route_output_key);
1996
1997 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
1998 {
1999         return NULL;
2000 }
2001
2002 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2003 {
2004         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2005
2006         return mtu ? : dst->dev->mtu;
2007 }
2008
2009 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2010                                           struct sk_buff *skb, u32 mtu)
2011 {
2012 }
2013
2014 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2015                                        struct sk_buff *skb)
2016 {
2017 }
2018
2019 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2020                                           unsigned long old)
2021 {
2022         return NULL;
2023 }
2024
2025 static struct dst_ops ipv4_dst_blackhole_ops = {
2026         .family                 =       AF_INET,
2027         .protocol               =       cpu_to_be16(ETH_P_IP),
2028         .check                  =       ipv4_blackhole_dst_check,
2029         .mtu                    =       ipv4_blackhole_mtu,
2030         .default_advmss         =       ipv4_default_advmss,
2031         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
2032         .redirect               =       ipv4_rt_blackhole_redirect,
2033         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
2034         .neigh_lookup           =       ipv4_neigh_lookup,
2035 };
2036
2037 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2038 {
2039         struct rtable *ort = (struct rtable *) dst_orig;
2040         struct rtable *rt;
2041
2042         rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2043         if (rt) {
2044                 struct dst_entry *new = &rt->dst;
2045
2046                 new->__use = 1;
2047                 new->input = dst_discard;
2048                 new->output = dst_discard;
2049
2050                 new->dev = ort->dst.dev;
2051                 if (new->dev)
2052                         dev_hold(new->dev);
2053
2054                 rt->rt_is_input = ort->rt_is_input;
2055                 rt->rt_iif = ort->rt_iif;
2056                 rt->rt_pmtu = ort->rt_pmtu;
2057
2058                 rt->rt_genid = rt_genid(net);
2059                 rt->rt_flags = ort->rt_flags;
2060                 rt->rt_type = ort->rt_type;
2061                 rt->rt_gateway = ort->rt_gateway;
2062
2063                 dst_free(new);
2064         }
2065
2066         dst_release(dst_orig);
2067
2068         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2069 }
2070
2071 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2072                                     struct sock *sk)
2073 {
2074         struct rtable *rt = __ip_route_output_key(net, flp4);
2075
2076         if (IS_ERR(rt))
2077                 return rt;
2078
2079         if (flp4->flowi4_proto)
2080                 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2081                                                    flowi4_to_flowi(flp4),
2082                                                    sk, 0);
2083
2084         return rt;
2085 }
2086 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2087
2088 static int rt_fill_info(struct net *net,  __be32 dst, __be32 src,
2089                         struct flowi4 *fl4, struct sk_buff *skb, u32 pid,
2090                         u32 seq, int event, int nowait, unsigned int flags)
2091 {
2092         struct rtable *rt = skb_rtable(skb);
2093         struct rtmsg *r;
2094         struct nlmsghdr *nlh;
2095         unsigned long expires = 0;
2096         u32 error;
2097         u32 metrics[RTAX_MAX];
2098
2099         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2100         if (nlh == NULL)
2101                 return -EMSGSIZE;
2102
2103         r = nlmsg_data(nlh);
2104         r->rtm_family    = AF_INET;
2105         r->rtm_dst_len  = 32;
2106         r->rtm_src_len  = 0;
2107         r->rtm_tos      = fl4->flowi4_tos;
2108         r->rtm_table    = RT_TABLE_MAIN;
2109         if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2110                 goto nla_put_failure;
2111         r->rtm_type     = rt->rt_type;
2112         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2113         r->rtm_protocol = RTPROT_UNSPEC;
2114         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2115         if (rt->rt_flags & RTCF_NOTIFY)
2116                 r->rtm_flags |= RTM_F_NOTIFY;
2117
2118         if (nla_put_be32(skb, RTA_DST, dst))
2119                 goto nla_put_failure;
2120         if (src) {
2121                 r->rtm_src_len = 32;
2122                 if (nla_put_be32(skb, RTA_SRC, src))
2123                         goto nla_put_failure;
2124         }
2125         if (rt->dst.dev &&
2126             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2127                 goto nla_put_failure;
2128 #ifdef CONFIG_IP_ROUTE_CLASSID
2129         if (rt->dst.tclassid &&
2130             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2131                 goto nla_put_failure;
2132 #endif
2133         if (!rt_is_input_route(rt) &&
2134             fl4->saddr != src) {
2135                 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2136                         goto nla_put_failure;
2137         }
2138         if (rt->rt_gateway &&
2139             nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2140                 goto nla_put_failure;
2141
2142         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2143         if (rt->rt_pmtu)
2144                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2145         if (rtnetlink_put_metrics(skb, metrics) < 0)
2146                 goto nla_put_failure;
2147
2148         if (fl4->flowi4_mark &&
2149             nla_put_be32(skb, RTA_MARK, fl4->flowi4_mark))
2150                 goto nla_put_failure;
2151
2152         error = rt->dst.error;
2153         expires = rt->dst.expires;
2154         if (expires) {
2155                 if (time_before(jiffies, expires))
2156                         expires -= jiffies;
2157                 else
2158                         expires = 0;
2159         }
2160
2161         if (rt_is_input_route(rt)) {
2162                 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
2163                         goto nla_put_failure;
2164         }
2165
2166         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2167                 goto nla_put_failure;
2168
2169         return nlmsg_end(skb, nlh);
2170
2171 nla_put_failure:
2172         nlmsg_cancel(skb, nlh);
2173         return -EMSGSIZE;
2174 }
2175
2176 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
2177 {
2178         struct net *net = sock_net(in_skb->sk);
2179         struct rtmsg *rtm;
2180         struct nlattr *tb[RTA_MAX+1];
2181         struct rtable *rt = NULL;
2182         struct flowi4 fl4;
2183         __be32 dst = 0;
2184         __be32 src = 0;
2185         u32 iif;
2186         int err;
2187         int mark;
2188         struct sk_buff *skb;
2189
2190         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2191         if (err < 0)
2192                 goto errout;
2193
2194         rtm = nlmsg_data(nlh);
2195
2196         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2197         if (skb == NULL) {
2198                 err = -ENOBUFS;
2199                 goto errout;
2200         }
2201
2202         /* Reserve room for dummy headers, this skb can pass
2203            through good chunk of routing engine.
2204          */
2205         skb_reset_mac_header(skb);
2206         skb_reset_network_header(skb);
2207
2208         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2209         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2210         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2211
2212         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2213         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2214         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2215         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2216
2217         memset(&fl4, 0, sizeof(fl4));
2218         fl4.daddr = dst;
2219         fl4.saddr = src;
2220         fl4.flowi4_tos = rtm->rtm_tos;
2221         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2222         fl4.flowi4_mark = mark;
2223
2224         if (iif) {
2225                 struct net_device *dev;
2226
2227                 dev = __dev_get_by_index(net, iif);
2228                 if (dev == NULL) {
2229                         err = -ENODEV;
2230                         goto errout_free;
2231                 }
2232
2233                 skb->protocol   = htons(ETH_P_IP);
2234                 skb->dev        = dev;
2235                 skb->mark       = mark;
2236                 local_bh_disable();
2237                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2238                 local_bh_enable();
2239
2240                 rt = skb_rtable(skb);
2241                 if (err == 0 && rt->dst.error)
2242                         err = -rt->dst.error;
2243         } else {
2244                 rt = ip_route_output_key(net, &fl4);
2245
2246                 err = 0;
2247                 if (IS_ERR(rt))
2248                         err = PTR_ERR(rt);
2249         }
2250
2251         if (err)
2252                 goto errout_free;
2253
2254         skb_dst_set(skb, &rt->dst);
2255         if (rtm->rtm_flags & RTM_F_NOTIFY)
2256                 rt->rt_flags |= RTCF_NOTIFY;
2257
2258         err = rt_fill_info(net, dst, src, &fl4, skb,
2259                            NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2260                            RTM_NEWROUTE, 0, 0);
2261         if (err <= 0)
2262                 goto errout_free;
2263
2264         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2265 errout:
2266         return err;
2267
2268 errout_free:
2269         kfree_skb(skb);
2270         goto errout;
2271 }
2272
2273 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
2274 {
2275         return skb->len;
2276 }
2277
2278 void ip_rt_multicast_event(struct in_device *in_dev)
2279 {
2280         rt_cache_flush(dev_net(in_dev->dev), 0);
2281 }
2282
2283 #ifdef CONFIG_SYSCTL
2284 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
2285                                         void __user *buffer,
2286                                         size_t *lenp, loff_t *ppos)
2287 {
2288         if (write) {
2289                 int flush_delay;
2290                 ctl_table ctl;
2291                 struct net *net;
2292
2293                 memcpy(&ctl, __ctl, sizeof(ctl));
2294                 ctl.data = &flush_delay;
2295                 proc_dointvec(&ctl, write, buffer, lenp, ppos);
2296
2297                 net = (struct net *)__ctl->extra1;
2298                 rt_cache_flush(net, flush_delay);
2299                 return 0;
2300         }
2301
2302         return -EINVAL;
2303 }
2304
2305 static ctl_table ipv4_route_table[] = {
2306         {
2307                 .procname       = "gc_thresh",
2308                 .data           = &ipv4_dst_ops.gc_thresh,
2309                 .maxlen         = sizeof(int),
2310                 .mode           = 0644,
2311                 .proc_handler   = proc_dointvec,
2312         },
2313         {
2314                 .procname       = "max_size",
2315                 .data           = &ip_rt_max_size,
2316                 .maxlen         = sizeof(int),
2317                 .mode           = 0644,
2318                 .proc_handler   = proc_dointvec,
2319         },
2320         {
2321                 /*  Deprecated. Use gc_min_interval_ms */
2322
2323                 .procname       = "gc_min_interval",
2324                 .data           = &ip_rt_gc_min_interval,
2325                 .maxlen         = sizeof(int),
2326                 .mode           = 0644,
2327                 .proc_handler   = proc_dointvec_jiffies,
2328         },
2329         {
2330                 .procname       = "gc_min_interval_ms",
2331                 .data           = &ip_rt_gc_min_interval,
2332                 .maxlen         = sizeof(int),
2333                 .mode           = 0644,
2334                 .proc_handler   = proc_dointvec_ms_jiffies,
2335         },
2336         {
2337                 .procname       = "gc_timeout",
2338                 .data           = &ip_rt_gc_timeout,
2339                 .maxlen         = sizeof(int),
2340                 .mode           = 0644,
2341                 .proc_handler   = proc_dointvec_jiffies,
2342         },
2343         {
2344                 .procname       = "gc_interval",
2345                 .data           = &ip_rt_gc_interval,
2346                 .maxlen         = sizeof(int),
2347                 .mode           = 0644,
2348                 .proc_handler   = proc_dointvec_jiffies,
2349         },
2350         {
2351                 .procname       = "redirect_load",
2352                 .data           = &ip_rt_redirect_load,
2353                 .maxlen         = sizeof(int),
2354                 .mode           = 0644,
2355                 .proc_handler   = proc_dointvec,
2356         },
2357         {
2358                 .procname       = "redirect_number",
2359                 .data           = &ip_rt_redirect_number,
2360                 .maxlen         = sizeof(int),
2361                 .mode           = 0644,
2362                 .proc_handler   = proc_dointvec,
2363         },
2364         {
2365                 .procname       = "redirect_silence",
2366                 .data           = &ip_rt_redirect_silence,
2367                 .maxlen         = sizeof(int),
2368                 .mode           = 0644,
2369                 .proc_handler   = proc_dointvec,
2370         },
2371         {
2372                 .procname       = "error_cost",
2373                 .data           = &ip_rt_error_cost,
2374                 .maxlen         = sizeof(int),
2375                 .mode           = 0644,
2376                 .proc_handler   = proc_dointvec,
2377         },
2378         {
2379                 .procname       = "error_burst",
2380                 .data           = &ip_rt_error_burst,
2381                 .maxlen         = sizeof(int),
2382                 .mode           = 0644,
2383                 .proc_handler   = proc_dointvec,
2384         },
2385         {
2386                 .procname       = "gc_elasticity",
2387                 .data           = &ip_rt_gc_elasticity,
2388                 .maxlen         = sizeof(int),
2389                 .mode           = 0644,
2390                 .proc_handler   = proc_dointvec,
2391         },
2392         {
2393                 .procname       = "mtu_expires",
2394                 .data           = &ip_rt_mtu_expires,
2395                 .maxlen         = sizeof(int),
2396                 .mode           = 0644,
2397                 .proc_handler   = proc_dointvec_jiffies,
2398         },
2399         {
2400                 .procname       = "min_pmtu",
2401                 .data           = &ip_rt_min_pmtu,
2402                 .maxlen         = sizeof(int),
2403                 .mode           = 0644,
2404                 .proc_handler   = proc_dointvec,
2405         },
2406         {
2407                 .procname       = "min_adv_mss",
2408                 .data           = &ip_rt_min_advmss,
2409                 .maxlen         = sizeof(int),
2410                 .mode           = 0644,
2411                 .proc_handler   = proc_dointvec,
2412         },
2413         { }
2414 };
2415
2416 static struct ctl_table ipv4_route_flush_table[] = {
2417         {
2418                 .procname       = "flush",
2419                 .maxlen         = sizeof(int),
2420                 .mode           = 0200,
2421                 .proc_handler   = ipv4_sysctl_rtcache_flush,
2422         },
2423         { },
2424 };
2425
2426 static __net_init int sysctl_route_net_init(struct net *net)
2427 {
2428         struct ctl_table *tbl;
2429
2430         tbl = ipv4_route_flush_table;
2431         if (!net_eq(net, &init_net)) {
2432                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2433                 if (tbl == NULL)
2434                         goto err_dup;
2435         }
2436         tbl[0].extra1 = net;
2437
2438         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2439         if (net->ipv4.route_hdr == NULL)
2440                 goto err_reg;
2441         return 0;
2442
2443 err_reg:
2444         if (tbl != ipv4_route_flush_table)
2445                 kfree(tbl);
2446 err_dup:
2447         return -ENOMEM;
2448 }
2449
2450 static __net_exit void sysctl_route_net_exit(struct net *net)
2451 {
2452         struct ctl_table *tbl;
2453
2454         tbl = net->ipv4.route_hdr->ctl_table_arg;
2455         unregister_net_sysctl_table(net->ipv4.route_hdr);
2456         BUG_ON(tbl == ipv4_route_flush_table);
2457         kfree(tbl);
2458 }
2459
2460 static __net_initdata struct pernet_operations sysctl_route_ops = {
2461         .init = sysctl_route_net_init,
2462         .exit = sysctl_route_net_exit,
2463 };
2464 #endif
2465
2466 static __net_init int rt_genid_init(struct net *net)
2467 {
2468         get_random_bytes(&net->ipv4.rt_genid,
2469                          sizeof(net->ipv4.rt_genid));
2470         get_random_bytes(&net->ipv4.dev_addr_genid,
2471                          sizeof(net->ipv4.dev_addr_genid));
2472         return 0;
2473 }
2474
2475 static __net_initdata struct pernet_operations rt_genid_ops = {
2476         .init = rt_genid_init,
2477 };
2478
2479 static int __net_init ipv4_inetpeer_init(struct net *net)
2480 {
2481         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2482
2483         if (!bp)
2484                 return -ENOMEM;
2485         inet_peer_base_init(bp);
2486         net->ipv4.peers = bp;
2487         return 0;
2488 }
2489
2490 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2491 {
2492         struct inet_peer_base *bp = net->ipv4.peers;
2493
2494         net->ipv4.peers = NULL;
2495         inetpeer_invalidate_tree(bp);
2496         kfree(bp);
2497 }
2498
2499 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2500         .init   =       ipv4_inetpeer_init,
2501         .exit   =       ipv4_inetpeer_exit,
2502 };
2503
2504 #ifdef CONFIG_IP_ROUTE_CLASSID
2505 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2506 #endif /* CONFIG_IP_ROUTE_CLASSID */
2507
2508 int __init ip_rt_init(void)
2509 {
2510         int rc = 0;
2511
2512 #ifdef CONFIG_IP_ROUTE_CLASSID
2513         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2514         if (!ip_rt_acct)
2515                 panic("IP: failed to allocate ip_rt_acct\n");
2516 #endif
2517
2518         ipv4_dst_ops.kmem_cachep =
2519                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2520                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2521
2522         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2523
2524         if (dst_entries_init(&ipv4_dst_ops) < 0)
2525                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2526
2527         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2528                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2529
2530         ipv4_dst_ops.gc_thresh = ~0;
2531         ip_rt_max_size = INT_MAX;
2532
2533         devinet_init();
2534         ip_fib_init();
2535
2536         if (ip_rt_proc_init())
2537                 pr_err("Unable to create route proc files\n");
2538 #ifdef CONFIG_XFRM
2539         xfrm_init();
2540         xfrm4_init(ip_rt_max_size);
2541 #endif
2542         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2543
2544 #ifdef CONFIG_SYSCTL
2545         register_pernet_subsys(&sysctl_route_ops);
2546 #endif
2547         register_pernet_subsys(&rt_genid_ops);
2548         register_pernet_subsys(&ipv4_inetpeer_ops);
2549         return rc;
2550 }
2551
2552 #ifdef CONFIG_SYSCTL
2553 /*
2554  * We really need to sanitize the damn ipv4 init order, then all
2555  * this nonsense will go away.
2556  */
2557 void __init ip_static_sysctl_init(void)
2558 {
2559         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
2560 }
2561 #endif