85d103fee88e4d31db26dc0726433344f6978367
[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 void              ipv4_dst_destroy(struct dst_entry *dst);
145 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
146 static void              ipv4_link_failure(struct sk_buff *skb);
147 static void              ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
148                                            struct sk_buff *skb, u32 mtu);
149 static void              ip_do_redirect(struct dst_entry *dst, struct sock *sk,
150                                         struct sk_buff *skb);
151
152 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
153                             int how)
154 {
155 }
156
157 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
158 {
159         WARN_ON(1);
160         return NULL;
161 }
162
163 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
164                                            struct sk_buff *skb,
165                                            const void *daddr);
166
167 static struct dst_ops ipv4_dst_ops = {
168         .family =               AF_INET,
169         .protocol =             cpu_to_be16(ETH_P_IP),
170         .check =                ipv4_dst_check,
171         .default_advmss =       ipv4_default_advmss,
172         .mtu =                  ipv4_mtu,
173         .cow_metrics =          ipv4_cow_metrics,
174         .destroy =              ipv4_dst_destroy,
175         .ifdown =               ipv4_dst_ifdown,
176         .negative_advice =      ipv4_negative_advice,
177         .link_failure =         ipv4_link_failure,
178         .update_pmtu =          ip_rt_update_pmtu,
179         .redirect =             ip_do_redirect,
180         .local_out =            __ip_local_out,
181         .neigh_lookup =         ipv4_neigh_lookup,
182 };
183
184 #define ECN_OR_COST(class)      TC_PRIO_##class
185
186 const __u8 ip_tos2prio[16] = {
187         TC_PRIO_BESTEFFORT,
188         ECN_OR_COST(BESTEFFORT),
189         TC_PRIO_BESTEFFORT,
190         ECN_OR_COST(BESTEFFORT),
191         TC_PRIO_BULK,
192         ECN_OR_COST(BULK),
193         TC_PRIO_BULK,
194         ECN_OR_COST(BULK),
195         TC_PRIO_INTERACTIVE,
196         ECN_OR_COST(INTERACTIVE),
197         TC_PRIO_INTERACTIVE,
198         ECN_OR_COST(INTERACTIVE),
199         TC_PRIO_INTERACTIVE_BULK,
200         ECN_OR_COST(INTERACTIVE_BULK),
201         TC_PRIO_INTERACTIVE_BULK,
202         ECN_OR_COST(INTERACTIVE_BULK)
203 };
204 EXPORT_SYMBOL(ip_tos2prio);
205
206 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
207 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field)
208
209 static inline int rt_genid(struct net *net)
210 {
211         return atomic_read(&net->ipv4.rt_genid);
212 }
213
214 #ifdef CONFIG_PROC_FS
215 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
216 {
217         if (*pos)
218                 return NULL;
219         return SEQ_START_TOKEN;
220 }
221
222 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
223 {
224         ++*pos;
225         return NULL;
226 }
227
228 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
229 {
230 }
231
232 static int rt_cache_seq_show(struct seq_file *seq, void *v)
233 {
234         if (v == SEQ_START_TOKEN)
235                 seq_printf(seq, "%-127s\n",
236                            "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
237                            "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
238                            "HHUptod\tSpecDst");
239         return 0;
240 }
241
242 static const struct seq_operations rt_cache_seq_ops = {
243         .start  = rt_cache_seq_start,
244         .next   = rt_cache_seq_next,
245         .stop   = rt_cache_seq_stop,
246         .show   = rt_cache_seq_show,
247 };
248
249 static int rt_cache_seq_open(struct inode *inode, struct file *file)
250 {
251         return seq_open(file, &rt_cache_seq_ops);
252 }
253
254 static const struct file_operations rt_cache_seq_fops = {
255         .owner   = THIS_MODULE,
256         .open    = rt_cache_seq_open,
257         .read    = seq_read,
258         .llseek  = seq_lseek,
259         .release = seq_release,
260 };
261
262
263 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
264 {
265         int cpu;
266
267         if (*pos == 0)
268                 return SEQ_START_TOKEN;
269
270         for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
271                 if (!cpu_possible(cpu))
272                         continue;
273                 *pos = cpu+1;
274                 return &per_cpu(rt_cache_stat, cpu);
275         }
276         return NULL;
277 }
278
279 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
280 {
281         int cpu;
282
283         for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
284                 if (!cpu_possible(cpu))
285                         continue;
286                 *pos = cpu+1;
287                 return &per_cpu(rt_cache_stat, cpu);
288         }
289         return NULL;
290
291 }
292
293 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
294 {
295
296 }
297
298 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
299 {
300         struct rt_cache_stat *st = v;
301
302         if (v == SEQ_START_TOKEN) {
303                 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");
304                 return 0;
305         }
306
307         seq_printf(seq,"%08x  %08x %08x %08x %08x %08x %08x %08x "
308                    " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
309                    dst_entries_get_slow(&ipv4_dst_ops),
310                    st->in_hit,
311                    st->in_slow_tot,
312                    st->in_slow_mc,
313                    st->in_no_route,
314                    st->in_brd,
315                    st->in_martian_dst,
316                    st->in_martian_src,
317
318                    st->out_hit,
319                    st->out_slow_tot,
320                    st->out_slow_mc,
321
322                    st->gc_total,
323                    st->gc_ignored,
324                    st->gc_goal_miss,
325                    st->gc_dst_overflow,
326                    st->in_hlist_search,
327                    st->out_hlist_search
328                 );
329         return 0;
330 }
331
332 static const struct seq_operations rt_cpu_seq_ops = {
333         .start  = rt_cpu_seq_start,
334         .next   = rt_cpu_seq_next,
335         .stop   = rt_cpu_seq_stop,
336         .show   = rt_cpu_seq_show,
337 };
338
339
340 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
341 {
342         return seq_open(file, &rt_cpu_seq_ops);
343 }
344
345 static const struct file_operations rt_cpu_seq_fops = {
346         .owner   = THIS_MODULE,
347         .open    = rt_cpu_seq_open,
348         .read    = seq_read,
349         .llseek  = seq_lseek,
350         .release = seq_release,
351 };
352
353 #ifdef CONFIG_IP_ROUTE_CLASSID
354 static int rt_acct_proc_show(struct seq_file *m, void *v)
355 {
356         struct ip_rt_acct *dst, *src;
357         unsigned int i, j;
358
359         dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
360         if (!dst)
361                 return -ENOMEM;
362
363         for_each_possible_cpu(i) {
364                 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
365                 for (j = 0; j < 256; j++) {
366                         dst[j].o_bytes   += src[j].o_bytes;
367                         dst[j].o_packets += src[j].o_packets;
368                         dst[j].i_bytes   += src[j].i_bytes;
369                         dst[j].i_packets += src[j].i_packets;
370                 }
371         }
372
373         seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
374         kfree(dst);
375         return 0;
376 }
377
378 static int rt_acct_proc_open(struct inode *inode, struct file *file)
379 {
380         return single_open(file, rt_acct_proc_show, NULL);
381 }
382
383 static const struct file_operations rt_acct_proc_fops = {
384         .owner          = THIS_MODULE,
385         .open           = rt_acct_proc_open,
386         .read           = seq_read,
387         .llseek         = seq_lseek,
388         .release        = single_release,
389 };
390 #endif
391
392 static int __net_init ip_rt_do_proc_init(struct net *net)
393 {
394         struct proc_dir_entry *pde;
395
396         pde = proc_net_fops_create(net, "rt_cache", S_IRUGO,
397                         &rt_cache_seq_fops);
398         if (!pde)
399                 goto err1;
400
401         pde = proc_create("rt_cache", S_IRUGO,
402                           net->proc_net_stat, &rt_cpu_seq_fops);
403         if (!pde)
404                 goto err2;
405
406 #ifdef CONFIG_IP_ROUTE_CLASSID
407         pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
408         if (!pde)
409                 goto err3;
410 #endif
411         return 0;
412
413 #ifdef CONFIG_IP_ROUTE_CLASSID
414 err3:
415         remove_proc_entry("rt_cache", net->proc_net_stat);
416 #endif
417 err2:
418         remove_proc_entry("rt_cache", net->proc_net);
419 err1:
420         return -ENOMEM;
421 }
422
423 static void __net_exit ip_rt_do_proc_exit(struct net *net)
424 {
425         remove_proc_entry("rt_cache", net->proc_net_stat);
426         remove_proc_entry("rt_cache", net->proc_net);
427 #ifdef CONFIG_IP_ROUTE_CLASSID
428         remove_proc_entry("rt_acct", net->proc_net);
429 #endif
430 }
431
432 static struct pernet_operations ip_rt_proc_ops __net_initdata =  {
433         .init = ip_rt_do_proc_init,
434         .exit = ip_rt_do_proc_exit,
435 };
436
437 static int __init ip_rt_proc_init(void)
438 {
439         return register_pernet_subsys(&ip_rt_proc_ops);
440 }
441
442 #else
443 static inline int ip_rt_proc_init(void)
444 {
445         return 0;
446 }
447 #endif /* CONFIG_PROC_FS */
448
449 static inline int rt_is_expired(struct rtable *rth)
450 {
451         return rth->rt_genid != rt_genid(dev_net(rth->dst.dev));
452 }
453
454 /*
455  * Perturbation of rt_genid by a small quantity [1..256]
456  * Using 8 bits of shuffling ensure we can call rt_cache_invalidate()
457  * many times (2^24) without giving recent rt_genid.
458  * Jenkins hash is strong enough that litle changes of rt_genid are OK.
459  */
460 static void rt_cache_invalidate(struct net *net)
461 {
462         unsigned char shuffle;
463
464         get_random_bytes(&shuffle, sizeof(shuffle));
465         atomic_add(shuffle + 1U, &net->ipv4.rt_genid);
466 }
467
468 /*
469  * delay < 0  : invalidate cache (fast : entries will be deleted later)
470  * delay >= 0 : invalidate & flush cache (can be long)
471  */
472 void rt_cache_flush(struct net *net, int delay)
473 {
474         rt_cache_invalidate(net);
475 }
476
477 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
478                                            struct sk_buff *skb,
479                                            const void *daddr)
480 {
481         struct net_device *dev = dst->dev;
482         const __be32 *pkey = daddr;
483         const struct rtable *rt;
484         struct neighbour *n;
485
486         rt = (const struct rtable *) dst;
487         if (rt->rt_gateway)
488                 pkey = (const __be32 *) &rt->rt_gateway;
489         else if (skb)
490                 pkey = &ip_hdr(skb)->daddr;
491
492         n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
493         if (n)
494                 return n;
495         return neigh_create(&arp_tbl, pkey, dev);
496 }
497
498 /*
499  * Peer allocation may fail only in serious out-of-memory conditions.  However
500  * we still can generate some output.
501  * Random ID selection looks a bit dangerous because we have no chances to
502  * select ID being unique in a reasonable period of time.
503  * But broken packet identifier may be better than no packet at all.
504  */
505 static void ip_select_fb_ident(struct iphdr *iph)
506 {
507         static DEFINE_SPINLOCK(ip_fb_id_lock);
508         static u32 ip_fallback_id;
509         u32 salt;
510
511         spin_lock_bh(&ip_fb_id_lock);
512         salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr);
513         iph->id = htons(salt & 0xFFFF);
514         ip_fallback_id = salt;
515         spin_unlock_bh(&ip_fb_id_lock);
516 }
517
518 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more)
519 {
520         struct net *net = dev_net(dst->dev);
521         struct inet_peer *peer;
522
523         peer = inet_getpeer_v4(net->ipv4.peers, iph->daddr, 1);
524         if (peer) {
525                 iph->id = htons(inet_getid(peer, more));
526                 inet_putpeer(peer);
527                 return;
528         }
529
530         ip_select_fb_ident(iph);
531 }
532 EXPORT_SYMBOL(__ip_select_ident);
533
534 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk,
535                              const struct iphdr *iph,
536                              int oif, u8 tos,
537                              u8 prot, u32 mark, int flow_flags)
538 {
539         if (sk) {
540                 const struct inet_sock *inet = inet_sk(sk);
541
542                 oif = sk->sk_bound_dev_if;
543                 mark = sk->sk_mark;
544                 tos = RT_CONN_FLAGS(sk);
545                 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
546         }
547         flowi4_init_output(fl4, oif, mark, tos,
548                            RT_SCOPE_UNIVERSE, prot,
549                            flow_flags,
550                            iph->daddr, iph->saddr, 0, 0);
551 }
552
553 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
554                                const struct sock *sk)
555 {
556         const struct iphdr *iph = ip_hdr(skb);
557         int oif = skb->dev->ifindex;
558         u8 tos = RT_TOS(iph->tos);
559         u8 prot = iph->protocol;
560         u32 mark = skb->mark;
561
562         __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0);
563 }
564
565 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
566 {
567         const struct inet_sock *inet = inet_sk(sk);
568         const struct ip_options_rcu *inet_opt;
569         __be32 daddr = inet->inet_daddr;
570
571         rcu_read_lock();
572         inet_opt = rcu_dereference(inet->inet_opt);
573         if (inet_opt && inet_opt->opt.srr)
574                 daddr = inet_opt->opt.faddr;
575         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
576                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
577                            inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
578                            inet_sk_flowi_flags(sk),
579                            daddr, inet->inet_saddr, 0, 0);
580         rcu_read_unlock();
581 }
582
583 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
584                                  const struct sk_buff *skb)
585 {
586         if (skb)
587                 build_skb_flow_key(fl4, skb, sk);
588         else
589                 build_sk_flow_key(fl4, sk);
590 }
591
592 static DEFINE_SEQLOCK(fnhe_seqlock);
593
594 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
595 {
596         struct fib_nh_exception *fnhe, *oldest;
597
598         oldest = rcu_dereference(hash->chain);
599         for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
600              fnhe = rcu_dereference(fnhe->fnhe_next)) {
601                 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
602                         oldest = fnhe;
603         }
604         return oldest;
605 }
606
607 static inline u32 fnhe_hashfun(__be32 daddr)
608 {
609         u32 hval;
610
611         hval = (__force u32) daddr;
612         hval ^= (hval >> 11) ^ (hval >> 22);
613
614         return hval & (FNHE_HASH_SIZE - 1);
615 }
616
617 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
618                                   u32 pmtu, unsigned long expires)
619 {
620         struct fnhe_hash_bucket *hash;
621         struct fib_nh_exception *fnhe;
622         int depth;
623         u32 hval = fnhe_hashfun(daddr);
624
625         write_seqlock_bh(&fnhe_seqlock);
626
627         hash = nh->nh_exceptions;
628         if (!hash) {
629                 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
630                 if (!hash)
631                         goto out_unlock;
632                 nh->nh_exceptions = hash;
633         }
634
635         hash += hval;
636
637         depth = 0;
638         for (fnhe = rcu_dereference(hash->chain); fnhe;
639              fnhe = rcu_dereference(fnhe->fnhe_next)) {
640                 if (fnhe->fnhe_daddr == daddr)
641                         break;
642                 depth++;
643         }
644
645         if (fnhe) {
646                 if (gw)
647                         fnhe->fnhe_gw = gw;
648                 if (pmtu) {
649                         fnhe->fnhe_pmtu = pmtu;
650                         fnhe->fnhe_expires = expires;
651                 }
652         } else {
653                 if (depth > FNHE_RECLAIM_DEPTH)
654                         fnhe = fnhe_oldest(hash);
655                 else {
656                         fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
657                         if (!fnhe)
658                                 goto out_unlock;
659
660                         fnhe->fnhe_next = hash->chain;
661                         rcu_assign_pointer(hash->chain, fnhe);
662                 }
663                 fnhe->fnhe_daddr = daddr;
664                 fnhe->fnhe_gw = gw;
665                 fnhe->fnhe_pmtu = pmtu;
666                 fnhe->fnhe_expires = expires;
667         }
668
669         fnhe->fnhe_stamp = jiffies;
670
671 out_unlock:
672         write_sequnlock_bh(&fnhe_seqlock);
673         return;
674 }
675
676 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4)
677 {
678         __be32 new_gw = icmp_hdr(skb)->un.gateway;
679         __be32 old_gw = ip_hdr(skb)->saddr;
680         struct net_device *dev = skb->dev;
681         struct in_device *in_dev;
682         struct fib_result res;
683         struct neighbour *n;
684         struct net *net;
685
686         switch (icmp_hdr(skb)->code & 7) {
687         case ICMP_REDIR_NET:
688         case ICMP_REDIR_NETTOS:
689         case ICMP_REDIR_HOST:
690         case ICMP_REDIR_HOSTTOS:
691                 break;
692
693         default:
694                 return;
695         }
696
697         if (rt->rt_gateway != old_gw)
698                 return;
699
700         in_dev = __in_dev_get_rcu(dev);
701         if (!in_dev)
702                 return;
703
704         net = dev_net(dev);
705         if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
706             ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
707             ipv4_is_zeronet(new_gw))
708                 goto reject_redirect;
709
710         if (!IN_DEV_SHARED_MEDIA(in_dev)) {
711                 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
712                         goto reject_redirect;
713                 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
714                         goto reject_redirect;
715         } else {
716                 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
717                         goto reject_redirect;
718         }
719
720         n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
721         if (n) {
722                 if (!(n->nud_state & NUD_VALID)) {
723                         neigh_event_send(n, NULL);
724                 } else {
725                         if (fib_lookup(net, fl4, &res) == 0) {
726                                 struct fib_nh *nh = &FIB_RES_NH(res);
727
728                                 update_or_create_fnhe(nh, fl4->daddr, new_gw,
729                                                       0, 0);
730                         }
731                         rt->rt_gateway = new_gw;
732                         rt->rt_flags |= RTCF_REDIRECTED;
733                         call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
734                 }
735                 neigh_release(n);
736         }
737         return;
738
739 reject_redirect:
740 #ifdef CONFIG_IP_ROUTE_VERBOSE
741         if (IN_DEV_LOG_MARTIANS(in_dev)) {
742                 const struct iphdr *iph = (const struct iphdr *) skb->data;
743                 __be32 daddr = iph->daddr;
744                 __be32 saddr = iph->saddr;
745
746                 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
747                                      "  Advised path = %pI4 -> %pI4\n",
748                                      &old_gw, dev->name, &new_gw,
749                                      &saddr, &daddr);
750         }
751 #endif
752         ;
753 }
754
755 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
756 {
757         struct rtable *rt;
758         struct flowi4 fl4;
759
760         rt = (struct rtable *) dst;
761
762         ip_rt_build_flow_key(&fl4, sk, skb);
763         __ip_do_redirect(rt, skb, &fl4);
764 }
765
766 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
767 {
768         struct rtable *rt = (struct rtable *)dst;
769         struct dst_entry *ret = dst;
770
771         if (rt) {
772                 if (dst->obsolete > 0) {
773                         ip_rt_put(rt);
774                         ret = NULL;
775                 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
776                            rt->dst.expires) {
777                         ip_rt_put(rt);
778                         ret = NULL;
779                 }
780         }
781         return ret;
782 }
783
784 /*
785  * Algorithm:
786  *      1. The first ip_rt_redirect_number redirects are sent
787  *         with exponential backoff, then we stop sending them at all,
788  *         assuming that the host ignores our redirects.
789  *      2. If we did not see packets requiring redirects
790  *         during ip_rt_redirect_silence, we assume that the host
791  *         forgot redirected route and start to send redirects again.
792  *
793  * This algorithm is much cheaper and more intelligent than dumb load limiting
794  * in icmp.c.
795  *
796  * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
797  * and "frag. need" (breaks PMTU discovery) in icmp.c.
798  */
799
800 void ip_rt_send_redirect(struct sk_buff *skb)
801 {
802         struct rtable *rt = skb_rtable(skb);
803         struct in_device *in_dev;
804         struct inet_peer *peer;
805         struct net *net;
806         int log_martians;
807
808         rcu_read_lock();
809         in_dev = __in_dev_get_rcu(rt->dst.dev);
810         if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
811                 rcu_read_unlock();
812                 return;
813         }
814         log_martians = IN_DEV_LOG_MARTIANS(in_dev);
815         rcu_read_unlock();
816
817         net = dev_net(rt->dst.dev);
818         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
819         if (!peer) {
820                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
821                 return;
822         }
823
824         /* No redirected packets during ip_rt_redirect_silence;
825          * reset the algorithm.
826          */
827         if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
828                 peer->rate_tokens = 0;
829
830         /* Too many ignored redirects; do not send anything
831          * set dst.rate_last to the last seen redirected packet.
832          */
833         if (peer->rate_tokens >= ip_rt_redirect_number) {
834                 peer->rate_last = jiffies;
835                 goto out_put_peer;
836         }
837
838         /* Check for load limit; set rate_last to the latest sent
839          * redirect.
840          */
841         if (peer->rate_tokens == 0 ||
842             time_after(jiffies,
843                        (peer->rate_last +
844                         (ip_rt_redirect_load << peer->rate_tokens)))) {
845                 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
846                 peer->rate_last = jiffies;
847                 ++peer->rate_tokens;
848 #ifdef CONFIG_IP_ROUTE_VERBOSE
849                 if (log_martians &&
850                     peer->rate_tokens == ip_rt_redirect_number)
851                         net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
852                                              &ip_hdr(skb)->saddr, rt->rt_iif,
853                                              &ip_hdr(skb)->daddr, &rt->rt_gateway);
854 #endif
855         }
856 out_put_peer:
857         inet_putpeer(peer);
858 }
859
860 static int ip_error(struct sk_buff *skb)
861 {
862         struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
863         struct rtable *rt = skb_rtable(skb);
864         struct inet_peer *peer;
865         unsigned long now;
866         struct net *net;
867         bool send;
868         int code;
869
870         net = dev_net(rt->dst.dev);
871         if (!IN_DEV_FORWARD(in_dev)) {
872                 switch (rt->dst.error) {
873                 case EHOSTUNREACH:
874                         IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
875                         break;
876
877                 case ENETUNREACH:
878                         IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
879                         break;
880                 }
881                 goto out;
882         }
883
884         switch (rt->dst.error) {
885         case EINVAL:
886         default:
887                 goto out;
888         case EHOSTUNREACH:
889                 code = ICMP_HOST_UNREACH;
890                 break;
891         case ENETUNREACH:
892                 code = ICMP_NET_UNREACH;
893                 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
894                 break;
895         case EACCES:
896                 code = ICMP_PKT_FILTERED;
897                 break;
898         }
899
900         peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
901
902         send = true;
903         if (peer) {
904                 now = jiffies;
905                 peer->rate_tokens += now - peer->rate_last;
906                 if (peer->rate_tokens > ip_rt_error_burst)
907                         peer->rate_tokens = ip_rt_error_burst;
908                 peer->rate_last = now;
909                 if (peer->rate_tokens >= ip_rt_error_cost)
910                         peer->rate_tokens -= ip_rt_error_cost;
911                 else
912                         send = false;
913                 inet_putpeer(peer);
914         }
915         if (send)
916                 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
917
918 out:    kfree_skb(skb);
919         return 0;
920 }
921
922 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
923 {
924         struct fib_result res;
925
926         if (mtu < ip_rt_min_pmtu)
927                 mtu = ip_rt_min_pmtu;
928
929         if (fib_lookup(dev_net(rt->dst.dev), fl4, &res) == 0) {
930                 struct fib_nh *nh = &FIB_RES_NH(res);
931
932                 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
933                                       jiffies + ip_rt_mtu_expires);
934         }
935         rt->rt_pmtu = mtu;
936         dst_set_expires(&rt->dst, ip_rt_mtu_expires);
937 }
938
939 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
940                               struct sk_buff *skb, u32 mtu)
941 {
942         struct rtable *rt = (struct rtable *) dst;
943         struct flowi4 fl4;
944
945         ip_rt_build_flow_key(&fl4, sk, skb);
946         __ip_rt_update_pmtu(rt, &fl4, mtu);
947 }
948
949 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
950                       int oif, u32 mark, u8 protocol, int flow_flags)
951 {
952         const struct iphdr *iph = (const struct iphdr *) skb->data;
953         struct flowi4 fl4;
954         struct rtable *rt;
955
956         __build_flow_key(&fl4, NULL, iph, oif,
957                          RT_TOS(iph->tos), protocol, mark, flow_flags);
958         rt = __ip_route_output_key(net, &fl4);
959         if (!IS_ERR(rt)) {
960                 __ip_rt_update_pmtu(rt, &fl4, mtu);
961                 ip_rt_put(rt);
962         }
963 }
964 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
965
966 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
967 {
968         const struct iphdr *iph = (const struct iphdr *) skb->data;
969         struct flowi4 fl4;
970         struct rtable *rt;
971
972         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
973         rt = __ip_route_output_key(sock_net(sk), &fl4);
974         if (!IS_ERR(rt)) {
975                 __ip_rt_update_pmtu(rt, &fl4, mtu);
976                 ip_rt_put(rt);
977         }
978 }
979 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
980
981 void ipv4_redirect(struct sk_buff *skb, struct net *net,
982                    int oif, u32 mark, u8 protocol, int flow_flags)
983 {
984         const struct iphdr *iph = (const struct iphdr *) skb->data;
985         struct flowi4 fl4;
986         struct rtable *rt;
987
988         __build_flow_key(&fl4, NULL, iph, oif,
989                          RT_TOS(iph->tos), protocol, mark, flow_flags);
990         rt = __ip_route_output_key(net, &fl4);
991         if (!IS_ERR(rt)) {
992                 __ip_do_redirect(rt, skb, &fl4);
993                 ip_rt_put(rt);
994         }
995 }
996 EXPORT_SYMBOL_GPL(ipv4_redirect);
997
998 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
999 {
1000         const struct iphdr *iph = (const struct iphdr *) skb->data;
1001         struct flowi4 fl4;
1002         struct rtable *rt;
1003
1004         __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1005         rt = __ip_route_output_key(sock_net(sk), &fl4);
1006         if (!IS_ERR(rt)) {
1007                 __ip_do_redirect(rt, skb, &fl4);
1008                 ip_rt_put(rt);
1009         }
1010 }
1011 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1012
1013 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1014 {
1015         struct rtable *rt = (struct rtable *) dst;
1016
1017         if (rt_is_expired(rt))
1018                 return NULL;
1019         return dst;
1020 }
1021
1022 static void ipv4_dst_destroy(struct dst_entry *dst)
1023 {
1024         struct rtable *rt = (struct rtable *) dst;
1025
1026         if (rt->fi) {
1027                 fib_info_put(rt->fi);
1028                 rt->fi = NULL;
1029         }
1030 }
1031
1032
1033 static void ipv4_link_failure(struct sk_buff *skb)
1034 {
1035         struct rtable *rt;
1036
1037         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1038
1039         rt = skb_rtable(skb);
1040         if (rt)
1041                 dst_set_expires(&rt->dst, 0);
1042 }
1043
1044 static int ip_rt_bug(struct sk_buff *skb)
1045 {
1046         pr_debug("%s: %pI4 -> %pI4, %s\n",
1047                  __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1048                  skb->dev ? skb->dev->name : "?");
1049         kfree_skb(skb);
1050         WARN_ON(1);
1051         return 0;
1052 }
1053
1054 /*
1055    We do not cache source address of outgoing interface,
1056    because it is used only by IP RR, TS and SRR options,
1057    so that it out of fast path.
1058
1059    BTW remember: "addr" is allowed to be not aligned
1060    in IP options!
1061  */
1062
1063 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1064 {
1065         __be32 src;
1066
1067         if (rt_is_output_route(rt))
1068                 src = ip_hdr(skb)->saddr;
1069         else {
1070                 struct fib_result res;
1071                 struct flowi4 fl4;
1072                 struct iphdr *iph;
1073
1074                 iph = ip_hdr(skb);
1075
1076                 memset(&fl4, 0, sizeof(fl4));
1077                 fl4.daddr = iph->daddr;
1078                 fl4.saddr = iph->saddr;
1079                 fl4.flowi4_tos = RT_TOS(iph->tos);
1080                 fl4.flowi4_oif = rt->dst.dev->ifindex;
1081                 fl4.flowi4_iif = skb->dev->ifindex;
1082                 fl4.flowi4_mark = skb->mark;
1083
1084                 rcu_read_lock();
1085                 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1086                         src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1087                 else
1088                         src = inet_select_addr(rt->dst.dev, rt->rt_gateway,
1089                                         RT_SCOPE_UNIVERSE);
1090                 rcu_read_unlock();
1091         }
1092         memcpy(addr, &src, 4);
1093 }
1094
1095 #ifdef CONFIG_IP_ROUTE_CLASSID
1096 static void set_class_tag(struct rtable *rt, u32 tag)
1097 {
1098         if (!(rt->dst.tclassid & 0xFFFF))
1099                 rt->dst.tclassid |= tag & 0xFFFF;
1100         if (!(rt->dst.tclassid & 0xFFFF0000))
1101                 rt->dst.tclassid |= tag & 0xFFFF0000;
1102 }
1103 #endif
1104
1105 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1106 {
1107         unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1108
1109         if (advmss == 0) {
1110                 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1111                                ip_rt_min_advmss);
1112                 if (advmss > 65535 - 40)
1113                         advmss = 65535 - 40;
1114         }
1115         return advmss;
1116 }
1117
1118 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1119 {
1120         const struct rtable *rt = (const struct rtable *) dst;
1121         unsigned int mtu = rt->rt_pmtu;
1122
1123         if (mtu && time_after_eq(jiffies, rt->dst.expires))
1124                 mtu = 0;
1125
1126         if (!mtu)
1127                 mtu = dst_metric_raw(dst, RTAX_MTU);
1128
1129         if (mtu && rt_is_output_route(rt))
1130                 return mtu;
1131
1132         mtu = dst->dev->mtu;
1133
1134         if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1135                 if (rt->rt_gateway != 0 && mtu > 576)
1136                         mtu = 576;
1137         }
1138
1139         if (mtu > IP_MAX_MTU)
1140                 mtu = IP_MAX_MTU;
1141
1142         return mtu;
1143 }
1144
1145 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1146                             struct fib_info *fi)
1147 {
1148         if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1149                 rt->fi = fi;
1150                 atomic_inc(&fi->fib_clntref);
1151         }
1152         dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1153 }
1154
1155 static void rt_bind_exception(struct rtable *rt, struct fib_nh *nh, __be32 daddr)
1156 {
1157         struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1158         struct fib_nh_exception *fnhe;
1159         u32 hval;
1160
1161         hval = fnhe_hashfun(daddr);
1162
1163 restart:
1164         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1165              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1166                 __be32 fnhe_daddr, gw;
1167                 unsigned long expires;
1168                 unsigned int seq;
1169                 u32 pmtu;
1170
1171                 seq = read_seqbegin(&fnhe_seqlock);
1172                 fnhe_daddr = fnhe->fnhe_daddr;
1173                 gw = fnhe->fnhe_gw;
1174                 pmtu = fnhe->fnhe_pmtu;
1175                 expires = fnhe->fnhe_expires;
1176                 if (read_seqretry(&fnhe_seqlock, seq))
1177                         goto restart;
1178                 if (daddr != fnhe_daddr)
1179                         continue;
1180                 if (pmtu) {
1181                         unsigned long diff = expires - jiffies;
1182
1183                         if (time_before(jiffies, expires)) {
1184                                 rt->rt_pmtu = pmtu;
1185                                 dst_set_expires(&rt->dst, diff);
1186                         }
1187                 }
1188                 if (gw)
1189                         rt->rt_gateway = gw;
1190                 fnhe->fnhe_stamp = jiffies;
1191                 break;
1192         }
1193 }
1194
1195 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1196                            const struct fib_result *res,
1197                            struct fib_info *fi, u16 type, u32 itag)
1198 {
1199         if (fi) {
1200                 struct fib_nh *nh = &FIB_RES_NH(*res);
1201
1202                 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK)
1203                         rt->rt_gateway = nh->nh_gw;
1204                 if (unlikely(nh->nh_exceptions))
1205                         rt_bind_exception(rt, nh, fl4->daddr);
1206                 rt_init_metrics(rt, fl4, fi);
1207 #ifdef CONFIG_IP_ROUTE_CLASSID
1208                 rt->dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1209 #endif
1210         }
1211
1212 #ifdef CONFIG_IP_ROUTE_CLASSID
1213 #ifdef CONFIG_IP_MULTIPLE_TABLES
1214         set_class_tag(rt, res->tclassid);
1215 #endif
1216         set_class_tag(rt, itag);
1217 #endif
1218 }
1219
1220 static struct rtable *rt_dst_alloc(struct net_device *dev,
1221                                    bool nopolicy, bool noxfrm)
1222 {
1223         return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1224                          DST_HOST | DST_NOCACHE |
1225                          (nopolicy ? DST_NOPOLICY : 0) |
1226                          (noxfrm ? DST_NOXFRM : 0));
1227 }
1228
1229 /* called in rcu_read_lock() section */
1230 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1231                                 u8 tos, struct net_device *dev, int our)
1232 {
1233         struct rtable *rth;
1234         struct in_device *in_dev = __in_dev_get_rcu(dev);
1235         u32 itag = 0;
1236         int err;
1237
1238         /* Primary sanity checks. */
1239
1240         if (in_dev == NULL)
1241                 return -EINVAL;
1242
1243         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1244             skb->protocol != htons(ETH_P_IP))
1245                 goto e_inval;
1246
1247         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1248                 if (ipv4_is_loopback(saddr))
1249                         goto e_inval;
1250
1251         if (ipv4_is_zeronet(saddr)) {
1252                 if (!ipv4_is_local_multicast(daddr))
1253                         goto e_inval;
1254         } else {
1255                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1256                                           in_dev, &itag);
1257                 if (err < 0)
1258                         goto e_err;
1259         }
1260         rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1261                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1262         if (!rth)
1263                 goto e_nobufs;
1264
1265 #ifdef CONFIG_IP_ROUTE_CLASSID
1266         rth->dst.tclassid = itag;
1267 #endif
1268         rth->dst.output = ip_rt_bug;
1269
1270         rth->rt_genid   = rt_genid(dev_net(dev));
1271         rth->rt_flags   = RTCF_MULTICAST;
1272         rth->rt_type    = RTN_MULTICAST;
1273         rth->rt_route_iif = dev->ifindex;
1274         rth->rt_iif     = dev->ifindex;
1275         rth->rt_oif     = 0;
1276         rth->rt_pmtu    = 0;
1277         rth->rt_gateway = daddr;
1278         rth->fi = NULL;
1279         if (our) {
1280                 rth->dst.input= ip_local_deliver;
1281                 rth->rt_flags |= RTCF_LOCAL;
1282         }
1283
1284 #ifdef CONFIG_IP_MROUTE
1285         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1286                 rth->dst.input = ip_mr_input;
1287 #endif
1288         RT_CACHE_STAT_INC(in_slow_mc);
1289
1290         skb_dst_set(skb, &rth->dst);
1291         return 0;
1292
1293 e_nobufs:
1294         return -ENOBUFS;
1295 e_inval:
1296         return -EINVAL;
1297 e_err:
1298         return err;
1299 }
1300
1301
1302 static void ip_handle_martian_source(struct net_device *dev,
1303                                      struct in_device *in_dev,
1304                                      struct sk_buff *skb,
1305                                      __be32 daddr,
1306                                      __be32 saddr)
1307 {
1308         RT_CACHE_STAT_INC(in_martian_src);
1309 #ifdef CONFIG_IP_ROUTE_VERBOSE
1310         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1311                 /*
1312                  *      RFC1812 recommendation, if source is martian,
1313                  *      the only hint is MAC header.
1314                  */
1315                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1316                         &daddr, &saddr, dev->name);
1317                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1318                         print_hex_dump(KERN_WARNING, "ll header: ",
1319                                        DUMP_PREFIX_OFFSET, 16, 1,
1320                                        skb_mac_header(skb),
1321                                        dev->hard_header_len, true);
1322                 }
1323         }
1324 #endif
1325 }
1326
1327 /* called in rcu_read_lock() section */
1328 static int __mkroute_input(struct sk_buff *skb,
1329                            const struct fib_result *res,
1330                            struct in_device *in_dev,
1331                            __be32 daddr, __be32 saddr, u32 tos,
1332                            struct rtable **result)
1333 {
1334         struct rtable *rth;
1335         int err;
1336         struct in_device *out_dev;
1337         unsigned int flags = 0;
1338         u32 itag;
1339
1340         /* get a working reference to the output device */
1341         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1342         if (out_dev == NULL) {
1343                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1344                 return -EINVAL;
1345         }
1346
1347
1348         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1349                                   in_dev->dev, in_dev, &itag);
1350         if (err < 0) {
1351                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1352                                          saddr);
1353
1354                 goto cleanup;
1355         }
1356
1357         if (err)
1358                 flags |= RTCF_DIRECTSRC;
1359
1360         if (out_dev == in_dev && err &&
1361             (IN_DEV_SHARED_MEDIA(out_dev) ||
1362              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1363                 flags |= RTCF_DOREDIRECT;
1364
1365         if (skb->protocol != htons(ETH_P_IP)) {
1366                 /* Not IP (i.e. ARP). Do not create route, if it is
1367                  * invalid for proxy arp. DNAT routes are always valid.
1368                  *
1369                  * Proxy arp feature have been extended to allow, ARP
1370                  * replies back to the same interface, to support
1371                  * Private VLAN switch technologies. See arp.c.
1372                  */
1373                 if (out_dev == in_dev &&
1374                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1375                         err = -EINVAL;
1376                         goto cleanup;
1377                 }
1378         }
1379
1380         rth = rt_dst_alloc(out_dev->dev,
1381                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1382                            IN_DEV_CONF_GET(out_dev, NOXFRM));
1383         if (!rth) {
1384                 err = -ENOBUFS;
1385                 goto cleanup;
1386         }
1387
1388         rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
1389         rth->rt_flags = flags;
1390         rth->rt_type = res->type;
1391         rth->rt_route_iif = in_dev->dev->ifindex;
1392         rth->rt_iif     = in_dev->dev->ifindex;
1393         rth->rt_oif     = 0;
1394         rth->rt_pmtu    = 0;
1395         rth->rt_gateway = daddr;
1396         rth->fi = NULL;
1397
1398         rth->dst.input = ip_forward;
1399         rth->dst.output = ip_output;
1400
1401         rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
1402
1403         *result = rth;
1404         err = 0;
1405  cleanup:
1406         return err;
1407 }
1408
1409 static int ip_mkroute_input(struct sk_buff *skb,
1410                             struct fib_result *res,
1411                             const struct flowi4 *fl4,
1412                             struct in_device *in_dev,
1413                             __be32 daddr, __be32 saddr, u32 tos)
1414 {
1415         struct rtable *rth = NULL;
1416         int err;
1417
1418 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1419         if (res->fi && res->fi->fib_nhs > 1)
1420                 fib_select_multipath(res);
1421 #endif
1422
1423         /* create a routing cache entry */
1424         err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
1425         if (err)
1426                 return err;
1427
1428         skb_dst_set(skb, &rth->dst);
1429         return 0;
1430 }
1431
1432 /*
1433  *      NOTE. We drop all the packets that has local source
1434  *      addresses, because every properly looped back packet
1435  *      must have correct destination already attached by output routine.
1436  *
1437  *      Such approach solves two big problems:
1438  *      1. Not simplex devices are handled properly.
1439  *      2. IP spoofing attempts are filtered with 100% of guarantee.
1440  *      called with rcu_read_lock()
1441  */
1442
1443 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1444                                u8 tos, struct net_device *dev)
1445 {
1446         struct fib_result res;
1447         struct in_device *in_dev = __in_dev_get_rcu(dev);
1448         struct flowi4   fl4;
1449         unsigned int    flags = 0;
1450         u32             itag = 0;
1451         struct rtable   *rth;
1452         int             err = -EINVAL;
1453         struct net    *net = dev_net(dev);
1454
1455         /* IP on this device is disabled. */
1456
1457         if (!in_dev)
1458                 goto out;
1459
1460         /* Check for the most weird martians, which can be not detected
1461            by fib_lookup.
1462          */
1463
1464         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1465                 goto martian_source;
1466
1467         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1468                 goto brd_input;
1469
1470         /* Accept zero addresses only to limited broadcast;
1471          * I even do not know to fix it or not. Waiting for complains :-)
1472          */
1473         if (ipv4_is_zeronet(saddr))
1474                 goto martian_source;
1475
1476         if (ipv4_is_zeronet(daddr))
1477                 goto martian_destination;
1478
1479         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) {
1480                 if (ipv4_is_loopback(daddr))
1481                         goto martian_destination;
1482
1483                 if (ipv4_is_loopback(saddr))
1484                         goto martian_source;
1485         }
1486
1487         /*
1488          *      Now we are ready to route packet.
1489          */
1490         fl4.flowi4_oif = 0;
1491         fl4.flowi4_iif = dev->ifindex;
1492         fl4.flowi4_mark = skb->mark;
1493         fl4.flowi4_tos = tos;
1494         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1495         fl4.daddr = daddr;
1496         fl4.saddr = saddr;
1497         err = fib_lookup(net, &fl4, &res);
1498         if (err != 0)
1499                 goto no_route;
1500
1501         RT_CACHE_STAT_INC(in_slow_tot);
1502
1503         if (res.type == RTN_BROADCAST)
1504                 goto brd_input;
1505
1506         if (res.type == RTN_LOCAL) {
1507                 err = fib_validate_source(skb, saddr, daddr, tos,
1508                                           net->loopback_dev->ifindex,
1509                                           dev, in_dev, &itag);
1510                 if (err < 0)
1511                         goto martian_source_keep_err;
1512                 if (err)
1513                         flags |= RTCF_DIRECTSRC;
1514                 goto local_input;
1515         }
1516
1517         if (!IN_DEV_FORWARD(in_dev))
1518                 goto no_route;
1519         if (res.type != RTN_UNICAST)
1520                 goto martian_destination;
1521
1522         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1523 out:    return err;
1524
1525 brd_input:
1526         if (skb->protocol != htons(ETH_P_IP))
1527                 goto e_inval;
1528
1529         if (!ipv4_is_zeronet(saddr)) {
1530                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1531                                           in_dev, &itag);
1532                 if (err < 0)
1533                         goto martian_source_keep_err;
1534                 if (err)
1535                         flags |= RTCF_DIRECTSRC;
1536         }
1537         flags |= RTCF_BROADCAST;
1538         res.type = RTN_BROADCAST;
1539         RT_CACHE_STAT_INC(in_brd);
1540
1541 local_input:
1542         rth = rt_dst_alloc(net->loopback_dev,
1543                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1544         if (!rth)
1545                 goto e_nobufs;
1546
1547         rth->dst.input= ip_local_deliver;
1548         rth->dst.output= ip_rt_bug;
1549 #ifdef CONFIG_IP_ROUTE_CLASSID
1550         rth->dst.tclassid = itag;
1551 #endif
1552
1553         rth->rt_genid = rt_genid(net);
1554         rth->rt_flags   = flags|RTCF_LOCAL;
1555         rth->rt_type    = res.type;
1556         rth->rt_route_iif = dev->ifindex;
1557         rth->rt_iif     = dev->ifindex;
1558         rth->rt_oif     = 0;
1559         rth->rt_pmtu    = 0;
1560         rth->rt_gateway = daddr;
1561         rth->fi = NULL;
1562         if (res.type == RTN_UNREACHABLE) {
1563                 rth->dst.input= ip_error;
1564                 rth->dst.error= -err;
1565                 rth->rt_flags   &= ~RTCF_LOCAL;
1566         }
1567         skb_dst_set(skb, &rth->dst);
1568         err = 0;
1569         goto out;
1570
1571 no_route:
1572         RT_CACHE_STAT_INC(in_no_route);
1573         res.type = RTN_UNREACHABLE;
1574         if (err == -ESRCH)
1575                 err = -ENETUNREACH;
1576         goto local_input;
1577
1578         /*
1579          *      Do not cache martian addresses: they should be logged (RFC1812)
1580          */
1581 martian_destination:
1582         RT_CACHE_STAT_INC(in_martian_dst);
1583 #ifdef CONFIG_IP_ROUTE_VERBOSE
1584         if (IN_DEV_LOG_MARTIANS(in_dev))
1585                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1586                                      &daddr, &saddr, dev->name);
1587 #endif
1588
1589 e_inval:
1590         err = -EINVAL;
1591         goto out;
1592
1593 e_nobufs:
1594         err = -ENOBUFS;
1595         goto out;
1596
1597 martian_source:
1598         err = -EINVAL;
1599 martian_source_keep_err:
1600         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1601         goto out;
1602 }
1603
1604 int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1605                    u8 tos, struct net_device *dev)
1606 {
1607         int res;
1608
1609         rcu_read_lock();
1610
1611         /* Multicast recognition logic is moved from route cache to here.
1612            The problem was that too many Ethernet cards have broken/missing
1613            hardware multicast filters :-( As result the host on multicasting
1614            network acquires a lot of useless route cache entries, sort of
1615            SDR messages from all the world. Now we try to get rid of them.
1616            Really, provided software IP multicast filter is organized
1617            reasonably (at least, hashed), it does not result in a slowdown
1618            comparing with route cache reject entries.
1619            Note, that multicast routers are not affected, because
1620            route cache entry is created eventually.
1621          */
1622         if (ipv4_is_multicast(daddr)) {
1623                 struct in_device *in_dev = __in_dev_get_rcu(dev);
1624
1625                 if (in_dev) {
1626                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1627                                                   ip_hdr(skb)->protocol);
1628                         if (our
1629 #ifdef CONFIG_IP_MROUTE
1630                                 ||
1631                             (!ipv4_is_local_multicast(daddr) &&
1632                              IN_DEV_MFORWARD(in_dev))
1633 #endif
1634                            ) {
1635                                 int res = ip_route_input_mc(skb, daddr, saddr,
1636                                                             tos, dev, our);
1637                                 rcu_read_unlock();
1638                                 return res;
1639                         }
1640                 }
1641                 rcu_read_unlock();
1642                 return -EINVAL;
1643         }
1644         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1645         rcu_read_unlock();
1646         return res;
1647 }
1648 EXPORT_SYMBOL(ip_route_input);
1649
1650 /* called with rcu_read_lock() */
1651 static struct rtable *__mkroute_output(const struct fib_result *res,
1652                                        const struct flowi4 *fl4, int orig_oif,
1653                                        struct net_device *dev_out,
1654                                        unsigned int flags)
1655 {
1656         struct fib_info *fi = res->fi;
1657         struct in_device *in_dev;
1658         u16 type = res->type;
1659         struct rtable *rth;
1660
1661         in_dev = __in_dev_get_rcu(dev_out);
1662         if (!in_dev)
1663                 return ERR_PTR(-EINVAL);
1664
1665         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1666                 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1667                         return ERR_PTR(-EINVAL);
1668
1669         if (ipv4_is_lbcast(fl4->daddr))
1670                 type = RTN_BROADCAST;
1671         else if (ipv4_is_multicast(fl4->daddr))
1672                 type = RTN_MULTICAST;
1673         else if (ipv4_is_zeronet(fl4->daddr))
1674                 return ERR_PTR(-EINVAL);
1675
1676         if (dev_out->flags & IFF_LOOPBACK)
1677                 flags |= RTCF_LOCAL;
1678
1679         if (type == RTN_BROADCAST) {
1680                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1681                 fi = NULL;
1682         } else if (type == RTN_MULTICAST) {
1683                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1684                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1685                                      fl4->flowi4_proto))
1686                         flags &= ~RTCF_LOCAL;
1687                 /* If multicast route do not exist use
1688                  * default one, but do not gateway in this case.
1689                  * Yes, it is hack.
1690                  */
1691                 if (fi && res->prefixlen < 4)
1692                         fi = NULL;
1693         }
1694
1695         rth = rt_dst_alloc(dev_out,
1696                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1697                            IN_DEV_CONF_GET(in_dev, NOXFRM));
1698         if (!rth)
1699                 return ERR_PTR(-ENOBUFS);
1700
1701         rth->dst.output = ip_output;
1702
1703         rth->rt_genid = rt_genid(dev_net(dev_out));
1704         rth->rt_flags   = flags;
1705         rth->rt_type    = type;
1706         rth->rt_route_iif = 0;
1707         rth->rt_iif     = orig_oif ? : dev_out->ifindex;
1708         rth->rt_oif     = orig_oif;
1709         rth->rt_pmtu    = 0;
1710         rth->rt_gateway = fl4->daddr;
1711         rth->fi = NULL;
1712
1713         RT_CACHE_STAT_INC(out_slow_tot);
1714
1715         if (flags & RTCF_LOCAL)
1716                 rth->dst.input = ip_local_deliver;
1717         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1718                 if (flags & RTCF_LOCAL &&
1719                     !(dev_out->flags & IFF_LOOPBACK)) {
1720                         rth->dst.output = ip_mc_output;
1721                         RT_CACHE_STAT_INC(out_slow_mc);
1722                 }
1723 #ifdef CONFIG_IP_MROUTE
1724                 if (type == RTN_MULTICAST) {
1725                         if (IN_DEV_MFORWARD(in_dev) &&
1726                             !ipv4_is_local_multicast(fl4->daddr)) {
1727                                 rth->dst.input = ip_mr_input;
1728                                 rth->dst.output = ip_mc_output;
1729                         }
1730                 }
1731 #endif
1732         }
1733
1734         rt_set_nexthop(rth, fl4, res, fi, type, 0);
1735
1736         if (fl4->flowi4_flags & FLOWI_FLAG_RT_NOCACHE)
1737                 rth->dst.flags |= DST_NOCACHE;
1738
1739         return rth;
1740 }
1741
1742 /*
1743  * Major route resolver routine.
1744  */
1745
1746 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
1747 {
1748         struct net_device *dev_out = NULL;
1749         __u8 tos = RT_FL_TOS(fl4);
1750         unsigned int flags = 0;
1751         struct fib_result res;
1752         struct rtable *rth;
1753         int orig_oif;
1754
1755         res.tclassid    = 0;
1756         res.fi          = NULL;
1757         res.table       = NULL;
1758
1759         orig_oif = fl4->flowi4_oif;
1760
1761         fl4->flowi4_iif = net->loopback_dev->ifindex;
1762         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
1763         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
1764                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
1765
1766         rcu_read_lock();
1767         if (fl4->saddr) {
1768                 rth = ERR_PTR(-EINVAL);
1769                 if (ipv4_is_multicast(fl4->saddr) ||
1770                     ipv4_is_lbcast(fl4->saddr) ||
1771                     ipv4_is_zeronet(fl4->saddr))
1772                         goto out;
1773
1774                 /* I removed check for oif == dev_out->oif here.
1775                    It was wrong for two reasons:
1776                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
1777                       is assigned to multiple interfaces.
1778                    2. Moreover, we are allowed to send packets with saddr
1779                       of another iface. --ANK
1780                  */
1781
1782                 if (fl4->flowi4_oif == 0 &&
1783                     (ipv4_is_multicast(fl4->daddr) ||
1784                      ipv4_is_lbcast(fl4->daddr))) {
1785                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1786                         dev_out = __ip_dev_find(net, fl4->saddr, false);
1787                         if (dev_out == NULL)
1788                                 goto out;
1789
1790                         /* Special hack: user can direct multicasts
1791                            and limited broadcast via necessary interface
1792                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
1793                            This hack is not just for fun, it allows
1794                            vic,vat and friends to work.
1795                            They bind socket to loopback, set ttl to zero
1796                            and expect that it will work.
1797                            From the viewpoint of routing cache they are broken,
1798                            because we are not allowed to build multicast path
1799                            with loopback source addr (look, routing cache
1800                            cannot know, that ttl is zero, so that packet
1801                            will not leave this host and route is valid).
1802                            Luckily, this hack is good workaround.
1803                          */
1804
1805                         fl4->flowi4_oif = dev_out->ifindex;
1806                         goto make_route;
1807                 }
1808
1809                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
1810                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1811                         if (!__ip_dev_find(net, fl4->saddr, false))
1812                                 goto out;
1813                 }
1814         }
1815
1816
1817         if (fl4->flowi4_oif) {
1818                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
1819                 rth = ERR_PTR(-ENODEV);
1820                 if (dev_out == NULL)
1821                         goto out;
1822
1823                 /* RACE: Check return value of inet_select_addr instead. */
1824                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
1825                         rth = ERR_PTR(-ENETUNREACH);
1826                         goto out;
1827                 }
1828                 if (ipv4_is_local_multicast(fl4->daddr) ||
1829                     ipv4_is_lbcast(fl4->daddr)) {
1830                         if (!fl4->saddr)
1831                                 fl4->saddr = inet_select_addr(dev_out, 0,
1832                                                               RT_SCOPE_LINK);
1833                         goto make_route;
1834                 }
1835                 if (fl4->saddr) {
1836                         if (ipv4_is_multicast(fl4->daddr))
1837                                 fl4->saddr = inet_select_addr(dev_out, 0,
1838                                                               fl4->flowi4_scope);
1839                         else if (!fl4->daddr)
1840                                 fl4->saddr = inet_select_addr(dev_out, 0,
1841                                                               RT_SCOPE_HOST);
1842                 }
1843         }
1844
1845         if (!fl4->daddr) {
1846                 fl4->daddr = fl4->saddr;
1847                 if (!fl4->daddr)
1848                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
1849                 dev_out = net->loopback_dev;
1850                 fl4->flowi4_oif = net->loopback_dev->ifindex;
1851                 res.type = RTN_LOCAL;
1852                 flags |= RTCF_LOCAL;
1853                 goto make_route;
1854         }
1855
1856         if (fib_lookup(net, fl4, &res)) {
1857                 res.fi = NULL;
1858                 res.table = NULL;
1859                 if (fl4->flowi4_oif) {
1860                         /* Apparently, routing tables are wrong. Assume,
1861                            that the destination is on link.
1862
1863                            WHY? DW.
1864                            Because we are allowed to send to iface
1865                            even if it has NO routes and NO assigned
1866                            addresses. When oif is specified, routing
1867                            tables are looked up with only one purpose:
1868                            to catch if destination is gatewayed, rather than
1869                            direct. Moreover, if MSG_DONTROUTE is set,
1870                            we send packet, ignoring both routing tables
1871                            and ifaddr state. --ANK
1872
1873
1874                            We could make it even if oif is unknown,
1875                            likely IPv6, but we do not.
1876                          */
1877
1878                         if (fl4->saddr == 0)
1879                                 fl4->saddr = inet_select_addr(dev_out, 0,
1880                                                               RT_SCOPE_LINK);
1881                         res.type = RTN_UNICAST;
1882                         goto make_route;
1883                 }
1884                 rth = ERR_PTR(-ENETUNREACH);
1885                 goto out;
1886         }
1887
1888         if (res.type == RTN_LOCAL) {
1889                 if (!fl4->saddr) {
1890                         if (res.fi->fib_prefsrc)
1891                                 fl4->saddr = res.fi->fib_prefsrc;
1892                         else
1893                                 fl4->saddr = fl4->daddr;
1894                 }
1895                 dev_out = net->loopback_dev;
1896                 fl4->flowi4_oif = dev_out->ifindex;
1897                 res.fi = NULL;
1898                 flags |= RTCF_LOCAL;
1899                 goto make_route;
1900         }
1901
1902 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1903         if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
1904                 fib_select_multipath(&res);
1905         else
1906 #endif
1907         if (!res.prefixlen &&
1908             res.table->tb_num_default > 1 &&
1909             res.type == RTN_UNICAST && !fl4->flowi4_oif)
1910                 fib_select_default(&res);
1911
1912         if (!fl4->saddr)
1913                 fl4->saddr = FIB_RES_PREFSRC(net, res);
1914
1915         dev_out = FIB_RES_DEV(res);
1916         fl4->flowi4_oif = dev_out->ifindex;
1917
1918
1919 make_route:
1920         rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
1921
1922 out:
1923         rcu_read_unlock();
1924         return rth;
1925 }
1926 EXPORT_SYMBOL_GPL(__ip_route_output_key);
1927
1928 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
1929 {
1930         return NULL;
1931 }
1932
1933 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
1934 {
1935         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
1936
1937         return mtu ? : dst->dev->mtu;
1938 }
1939
1940 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
1941                                           struct sk_buff *skb, u32 mtu)
1942 {
1943 }
1944
1945 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
1946                                        struct sk_buff *skb)
1947 {
1948 }
1949
1950 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
1951                                           unsigned long old)
1952 {
1953         return NULL;
1954 }
1955
1956 static struct dst_ops ipv4_dst_blackhole_ops = {
1957         .family                 =       AF_INET,
1958         .protocol               =       cpu_to_be16(ETH_P_IP),
1959         .destroy                =       ipv4_dst_destroy,
1960         .check                  =       ipv4_blackhole_dst_check,
1961         .mtu                    =       ipv4_blackhole_mtu,
1962         .default_advmss         =       ipv4_default_advmss,
1963         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
1964         .redirect               =       ipv4_rt_blackhole_redirect,
1965         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
1966         .neigh_lookup           =       ipv4_neigh_lookup,
1967 };
1968
1969 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
1970 {
1971         struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
1972         struct rtable *ort = (struct rtable *) dst_orig;
1973
1974         if (rt) {
1975                 struct dst_entry *new = &rt->dst;
1976
1977                 new->__use = 1;
1978                 new->input = dst_discard;
1979                 new->output = dst_discard;
1980
1981                 new->dev = ort->dst.dev;
1982                 if (new->dev)
1983                         dev_hold(new->dev);
1984
1985                 rt->rt_route_iif = ort->rt_route_iif;
1986                 rt->rt_iif = ort->rt_iif;
1987                 rt->rt_oif = ort->rt_oif;
1988                 rt->rt_pmtu = ort->rt_pmtu;
1989
1990                 rt->rt_genid = rt_genid(net);
1991                 rt->rt_flags = ort->rt_flags;
1992                 rt->rt_type = ort->rt_type;
1993                 rt->rt_gateway = ort->rt_gateway;
1994                 rt->fi = ort->fi;
1995                 if (rt->fi)
1996                         atomic_inc(&rt->fi->fib_clntref);
1997
1998                 dst_free(new);
1999         }
2000
2001         dst_release(dst_orig);
2002
2003         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2004 }
2005
2006 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2007                                     struct sock *sk)
2008 {
2009         struct rtable *rt = __ip_route_output_key(net, flp4);
2010
2011         if (IS_ERR(rt))
2012                 return rt;
2013
2014         if (flp4->flowi4_proto)
2015                 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2016                                                    flowi4_to_flowi(flp4),
2017                                                    sk, 0);
2018
2019         return rt;
2020 }
2021 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2022
2023 static int rt_fill_info(struct net *net,  __be32 dst, __be32 src,
2024                         struct flowi4 *fl4, struct sk_buff *skb, u32 pid,
2025                         u32 seq, int event, int nowait, unsigned int flags)
2026 {
2027         struct rtable *rt = skb_rtable(skb);
2028         struct rtmsg *r;
2029         struct nlmsghdr *nlh;
2030         unsigned long expires = 0;
2031         u32 error;
2032         u32 metrics[RTAX_MAX];
2033
2034         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2035         if (nlh == NULL)
2036                 return -EMSGSIZE;
2037
2038         r = nlmsg_data(nlh);
2039         r->rtm_family    = AF_INET;
2040         r->rtm_dst_len  = 32;
2041         r->rtm_src_len  = 0;
2042         r->rtm_tos      = fl4->flowi4_tos;
2043         r->rtm_table    = RT_TABLE_MAIN;
2044         if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2045                 goto nla_put_failure;
2046         r->rtm_type     = rt->rt_type;
2047         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2048         r->rtm_protocol = RTPROT_UNSPEC;
2049         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2050         if (rt->rt_flags & RTCF_NOTIFY)
2051                 r->rtm_flags |= RTM_F_NOTIFY;
2052
2053         if (nla_put_be32(skb, RTA_DST, dst))
2054                 goto nla_put_failure;
2055         if (src) {
2056                 r->rtm_src_len = 32;
2057                 if (nla_put_be32(skb, RTA_SRC, src))
2058                         goto nla_put_failure;
2059         }
2060         if (rt->dst.dev &&
2061             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2062                 goto nla_put_failure;
2063 #ifdef CONFIG_IP_ROUTE_CLASSID
2064         if (rt->dst.tclassid &&
2065             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2066                 goto nla_put_failure;
2067 #endif
2068         if (!rt_is_input_route(rt) &&
2069             fl4->saddr != src) {
2070                 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2071                         goto nla_put_failure;
2072         }
2073         if (fl4->daddr != rt->rt_gateway &&
2074             nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2075                 goto nla_put_failure;
2076
2077         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2078         if (rt->rt_pmtu)
2079                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2080         if (rtnetlink_put_metrics(skb, metrics) < 0)
2081                 goto nla_put_failure;
2082
2083         if (fl4->flowi4_mark &&
2084             nla_put_be32(skb, RTA_MARK, fl4->flowi4_mark))
2085                 goto nla_put_failure;
2086
2087         error = rt->dst.error;
2088         expires = rt->dst.expires;
2089         if (expires) {
2090                 if (time_before(jiffies, expires))
2091                         expires -= jiffies;
2092                 else
2093                         expires = 0;
2094         }
2095
2096         if (rt_is_input_route(rt)) {
2097                 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
2098                         goto nla_put_failure;
2099         }
2100
2101         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2102                 goto nla_put_failure;
2103
2104         return nlmsg_end(skb, nlh);
2105
2106 nla_put_failure:
2107         nlmsg_cancel(skb, nlh);
2108         return -EMSGSIZE;
2109 }
2110
2111 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
2112 {
2113         struct net *net = sock_net(in_skb->sk);
2114         struct rtmsg *rtm;
2115         struct nlattr *tb[RTA_MAX+1];
2116         struct rtable *rt = NULL;
2117         struct flowi4 fl4;
2118         __be32 dst = 0;
2119         __be32 src = 0;
2120         u32 iif;
2121         int err;
2122         int mark;
2123         struct sk_buff *skb;
2124
2125         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2126         if (err < 0)
2127                 goto errout;
2128
2129         rtm = nlmsg_data(nlh);
2130
2131         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2132         if (skb == NULL) {
2133                 err = -ENOBUFS;
2134                 goto errout;
2135         }
2136
2137         /* Reserve room for dummy headers, this skb can pass
2138            through good chunk of routing engine.
2139          */
2140         skb_reset_mac_header(skb);
2141         skb_reset_network_header(skb);
2142
2143         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2144         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2145         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2146
2147         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2148         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2149         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2150         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2151
2152         memset(&fl4, 0, sizeof(fl4));
2153         fl4.daddr = dst;
2154         fl4.saddr = src;
2155         fl4.flowi4_tos = rtm->rtm_tos;
2156         fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2157         fl4.flowi4_mark = mark;
2158
2159         if (iif) {
2160                 struct net_device *dev;
2161
2162                 dev = __dev_get_by_index(net, iif);
2163                 if (dev == NULL) {
2164                         err = -ENODEV;
2165                         goto errout_free;
2166                 }
2167
2168                 skb->protocol   = htons(ETH_P_IP);
2169                 skb->dev        = dev;
2170                 skb->mark       = mark;
2171                 local_bh_disable();
2172                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2173                 local_bh_enable();
2174
2175                 rt = skb_rtable(skb);
2176                 if (err == 0 && rt->dst.error)
2177                         err = -rt->dst.error;
2178         } else {
2179                 rt = ip_route_output_key(net, &fl4);
2180
2181                 err = 0;
2182                 if (IS_ERR(rt))
2183                         err = PTR_ERR(rt);
2184         }
2185
2186         if (err)
2187                 goto errout_free;
2188
2189         skb_dst_set(skb, &rt->dst);
2190         if (rtm->rtm_flags & RTM_F_NOTIFY)
2191                 rt->rt_flags |= RTCF_NOTIFY;
2192
2193         err = rt_fill_info(net, dst, src, &fl4, skb,
2194                            NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2195                            RTM_NEWROUTE, 0, 0);
2196         if (err <= 0)
2197                 goto errout_free;
2198
2199         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2200 errout:
2201         return err;
2202
2203 errout_free:
2204         kfree_skb(skb);
2205         goto errout;
2206 }
2207
2208 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
2209 {
2210         return skb->len;
2211 }
2212
2213 void ip_rt_multicast_event(struct in_device *in_dev)
2214 {
2215         rt_cache_flush(dev_net(in_dev->dev), 0);
2216 }
2217
2218 #ifdef CONFIG_SYSCTL
2219 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
2220                                         void __user *buffer,
2221                                         size_t *lenp, loff_t *ppos)
2222 {
2223         if (write) {
2224                 int flush_delay;
2225                 ctl_table ctl;
2226                 struct net *net;
2227
2228                 memcpy(&ctl, __ctl, sizeof(ctl));
2229                 ctl.data = &flush_delay;
2230                 proc_dointvec(&ctl, write, buffer, lenp, ppos);
2231
2232                 net = (struct net *)__ctl->extra1;
2233                 rt_cache_flush(net, flush_delay);
2234                 return 0;
2235         }
2236
2237         return -EINVAL;
2238 }
2239
2240 static ctl_table ipv4_route_table[] = {
2241         {
2242                 .procname       = "gc_thresh",
2243                 .data           = &ipv4_dst_ops.gc_thresh,
2244                 .maxlen         = sizeof(int),
2245                 .mode           = 0644,
2246                 .proc_handler   = proc_dointvec,
2247         },
2248         {
2249                 .procname       = "max_size",
2250                 .data           = &ip_rt_max_size,
2251                 .maxlen         = sizeof(int),
2252                 .mode           = 0644,
2253                 .proc_handler   = proc_dointvec,
2254         },
2255         {
2256                 /*  Deprecated. Use gc_min_interval_ms */
2257
2258                 .procname       = "gc_min_interval",
2259                 .data           = &ip_rt_gc_min_interval,
2260                 .maxlen         = sizeof(int),
2261                 .mode           = 0644,
2262                 .proc_handler   = proc_dointvec_jiffies,
2263         },
2264         {
2265                 .procname       = "gc_min_interval_ms",
2266                 .data           = &ip_rt_gc_min_interval,
2267                 .maxlen         = sizeof(int),
2268                 .mode           = 0644,
2269                 .proc_handler   = proc_dointvec_ms_jiffies,
2270         },
2271         {
2272                 .procname       = "gc_timeout",
2273                 .data           = &ip_rt_gc_timeout,
2274                 .maxlen         = sizeof(int),
2275                 .mode           = 0644,
2276                 .proc_handler   = proc_dointvec_jiffies,
2277         },
2278         {
2279                 .procname       = "gc_interval",
2280                 .data           = &ip_rt_gc_interval,
2281                 .maxlen         = sizeof(int),
2282                 .mode           = 0644,
2283                 .proc_handler   = proc_dointvec_jiffies,
2284         },
2285         {
2286                 .procname       = "redirect_load",
2287                 .data           = &ip_rt_redirect_load,
2288                 .maxlen         = sizeof(int),
2289                 .mode           = 0644,
2290                 .proc_handler   = proc_dointvec,
2291         },
2292         {
2293                 .procname       = "redirect_number",
2294                 .data           = &ip_rt_redirect_number,
2295                 .maxlen         = sizeof(int),
2296                 .mode           = 0644,
2297                 .proc_handler   = proc_dointvec,
2298         },
2299         {
2300                 .procname       = "redirect_silence",
2301                 .data           = &ip_rt_redirect_silence,
2302                 .maxlen         = sizeof(int),
2303                 .mode           = 0644,
2304                 .proc_handler   = proc_dointvec,
2305         },
2306         {
2307                 .procname       = "error_cost",
2308                 .data           = &ip_rt_error_cost,
2309                 .maxlen         = sizeof(int),
2310                 .mode           = 0644,
2311                 .proc_handler   = proc_dointvec,
2312         },
2313         {
2314                 .procname       = "error_burst",
2315                 .data           = &ip_rt_error_burst,
2316                 .maxlen         = sizeof(int),
2317                 .mode           = 0644,
2318                 .proc_handler   = proc_dointvec,
2319         },
2320         {
2321                 .procname       = "gc_elasticity",
2322                 .data           = &ip_rt_gc_elasticity,
2323                 .maxlen         = sizeof(int),
2324                 .mode           = 0644,
2325                 .proc_handler   = proc_dointvec,
2326         },
2327         {
2328                 .procname       = "mtu_expires",
2329                 .data           = &ip_rt_mtu_expires,
2330                 .maxlen         = sizeof(int),
2331                 .mode           = 0644,
2332                 .proc_handler   = proc_dointvec_jiffies,
2333         },
2334         {
2335                 .procname       = "min_pmtu",
2336                 .data           = &ip_rt_min_pmtu,
2337                 .maxlen         = sizeof(int),
2338                 .mode           = 0644,
2339                 .proc_handler   = proc_dointvec,
2340         },
2341         {
2342                 .procname       = "min_adv_mss",
2343                 .data           = &ip_rt_min_advmss,
2344                 .maxlen         = sizeof(int),
2345                 .mode           = 0644,
2346                 .proc_handler   = proc_dointvec,
2347         },
2348         { }
2349 };
2350
2351 static struct ctl_table ipv4_route_flush_table[] = {
2352         {
2353                 .procname       = "flush",
2354                 .maxlen         = sizeof(int),
2355                 .mode           = 0200,
2356                 .proc_handler   = ipv4_sysctl_rtcache_flush,
2357         },
2358         { },
2359 };
2360
2361 static __net_init int sysctl_route_net_init(struct net *net)
2362 {
2363         struct ctl_table *tbl;
2364
2365         tbl = ipv4_route_flush_table;
2366         if (!net_eq(net, &init_net)) {
2367                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2368                 if (tbl == NULL)
2369                         goto err_dup;
2370         }
2371         tbl[0].extra1 = net;
2372
2373         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2374         if (net->ipv4.route_hdr == NULL)
2375                 goto err_reg;
2376         return 0;
2377
2378 err_reg:
2379         if (tbl != ipv4_route_flush_table)
2380                 kfree(tbl);
2381 err_dup:
2382         return -ENOMEM;
2383 }
2384
2385 static __net_exit void sysctl_route_net_exit(struct net *net)
2386 {
2387         struct ctl_table *tbl;
2388
2389         tbl = net->ipv4.route_hdr->ctl_table_arg;
2390         unregister_net_sysctl_table(net->ipv4.route_hdr);
2391         BUG_ON(tbl == ipv4_route_flush_table);
2392         kfree(tbl);
2393 }
2394
2395 static __net_initdata struct pernet_operations sysctl_route_ops = {
2396         .init = sysctl_route_net_init,
2397         .exit = sysctl_route_net_exit,
2398 };
2399 #endif
2400
2401 static __net_init int rt_genid_init(struct net *net)
2402 {
2403         get_random_bytes(&net->ipv4.rt_genid,
2404                          sizeof(net->ipv4.rt_genid));
2405         get_random_bytes(&net->ipv4.dev_addr_genid,
2406                          sizeof(net->ipv4.dev_addr_genid));
2407         return 0;
2408 }
2409
2410 static __net_initdata struct pernet_operations rt_genid_ops = {
2411         .init = rt_genid_init,
2412 };
2413
2414 static int __net_init ipv4_inetpeer_init(struct net *net)
2415 {
2416         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2417
2418         if (!bp)
2419                 return -ENOMEM;
2420         inet_peer_base_init(bp);
2421         net->ipv4.peers = bp;
2422         return 0;
2423 }
2424
2425 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2426 {
2427         struct inet_peer_base *bp = net->ipv4.peers;
2428
2429         net->ipv4.peers = NULL;
2430         inetpeer_invalidate_tree(bp);
2431         kfree(bp);
2432 }
2433
2434 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2435         .init   =       ipv4_inetpeer_init,
2436         .exit   =       ipv4_inetpeer_exit,
2437 };
2438
2439 #ifdef CONFIG_IP_ROUTE_CLASSID
2440 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2441 #endif /* CONFIG_IP_ROUTE_CLASSID */
2442
2443 int __init ip_rt_init(void)
2444 {
2445         int rc = 0;
2446
2447 #ifdef CONFIG_IP_ROUTE_CLASSID
2448         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2449         if (!ip_rt_acct)
2450                 panic("IP: failed to allocate ip_rt_acct\n");
2451 #endif
2452
2453         ipv4_dst_ops.kmem_cachep =
2454                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2455                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2456
2457         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2458
2459         if (dst_entries_init(&ipv4_dst_ops) < 0)
2460                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2461
2462         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2463                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2464
2465         ipv4_dst_ops.gc_thresh = ~0;
2466         ip_rt_max_size = INT_MAX;
2467
2468         devinet_init();
2469         ip_fib_init();
2470
2471         if (ip_rt_proc_init())
2472                 pr_err("Unable to create route proc files\n");
2473 #ifdef CONFIG_XFRM
2474         xfrm_init();
2475         xfrm4_init(ip_rt_max_size);
2476 #endif
2477         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2478
2479 #ifdef CONFIG_SYSCTL
2480         register_pernet_subsys(&sysctl_route_ops);
2481 #endif
2482         register_pernet_subsys(&rt_genid_ops);
2483         register_pernet_subsys(&ipv4_inetpeer_ops);
2484         return rc;
2485 }
2486
2487 #ifdef CONFIG_SYSCTL
2488 /*
2489  * We really need to sanitize the damn ipv4 init order, then all
2490  * this nonsense will go away.
2491  */
2492 void __init ip_static_sysctl_init(void)
2493 {
2494         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
2495 }
2496 #endif