c89d690acdfaf1e04c621fd14b6bbfb0c86d0137
[~shefty/rdma-dev.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                                              &rt->rt_dst, &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
1136                 if (rt->rt_gateway != rt->rt_dst && mtu > 576)
1137                         mtu = 576;
1138         }
1139
1140         if (mtu > IP_MAX_MTU)
1141                 mtu = IP_MAX_MTU;
1142
1143         return mtu;
1144 }
1145
1146 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *fl4,
1147                             struct fib_info *fi)
1148 {
1149         if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1150                 rt->fi = fi;
1151                 atomic_inc(&fi->fib_clntref);
1152         }
1153         dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1154 }
1155
1156 static void rt_bind_exception(struct rtable *rt, struct fib_nh *nh, __be32 daddr)
1157 {
1158         struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1159         struct fib_nh_exception *fnhe;
1160         u32 hval;
1161
1162         hval = fnhe_hashfun(daddr);
1163
1164 restart:
1165         for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1166              fnhe = rcu_dereference(fnhe->fnhe_next)) {
1167                 __be32 fnhe_daddr, gw;
1168                 unsigned long expires;
1169                 unsigned int seq;
1170                 u32 pmtu;
1171
1172                 seq = read_seqbegin(&fnhe_seqlock);
1173                 fnhe_daddr = fnhe->fnhe_daddr;
1174                 gw = fnhe->fnhe_gw;
1175                 pmtu = fnhe->fnhe_pmtu;
1176                 expires = fnhe->fnhe_expires;
1177                 if (read_seqretry(&fnhe_seqlock, seq))
1178                         goto restart;
1179                 if (daddr != fnhe_daddr)
1180                         continue;
1181                 if (pmtu) {
1182                         unsigned long diff = expires - jiffies;
1183
1184                         if (time_before(jiffies, expires)) {
1185                                 rt->rt_pmtu = pmtu;
1186                                 dst_set_expires(&rt->dst, diff);
1187                         }
1188                 }
1189                 if (gw)
1190                         rt->rt_gateway = gw;
1191                 fnhe->fnhe_stamp = jiffies;
1192                 break;
1193         }
1194 }
1195
1196 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *fl4,
1197                            const struct fib_result *res,
1198                            struct fib_info *fi, u16 type, u32 itag)
1199 {
1200         if (fi) {
1201                 struct fib_nh *nh = &FIB_RES_NH(*res);
1202
1203                 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK)
1204                         rt->rt_gateway = nh->nh_gw;
1205                 if (unlikely(nh->nh_exceptions))
1206                         rt_bind_exception(rt, nh, fl4->daddr);
1207                 rt_init_metrics(rt, fl4, fi);
1208 #ifdef CONFIG_IP_ROUTE_CLASSID
1209                 rt->dst.tclassid = FIB_RES_NH(*res).nh_tclassid;
1210 #endif
1211         }
1212
1213 #ifdef CONFIG_IP_ROUTE_CLASSID
1214 #ifdef CONFIG_IP_MULTIPLE_TABLES
1215         set_class_tag(rt, res->tclassid);
1216 #endif
1217         set_class_tag(rt, itag);
1218 #endif
1219 }
1220
1221 static struct rtable *rt_dst_alloc(struct net_device *dev,
1222                                    bool nopolicy, bool noxfrm)
1223 {
1224         return dst_alloc(&ipv4_dst_ops, dev, 1, -1,
1225                          DST_HOST | DST_NOCACHE |
1226                          (nopolicy ? DST_NOPOLICY : 0) |
1227                          (noxfrm ? DST_NOXFRM : 0));
1228 }
1229
1230 /* called in rcu_read_lock() section */
1231 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1232                                 u8 tos, struct net_device *dev, int our)
1233 {
1234         struct rtable *rth;
1235         struct in_device *in_dev = __in_dev_get_rcu(dev);
1236         u32 itag = 0;
1237         int err;
1238
1239         /* Primary sanity checks. */
1240
1241         if (in_dev == NULL)
1242                 return -EINVAL;
1243
1244         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1245             skb->protocol != htons(ETH_P_IP))
1246                 goto e_inval;
1247
1248         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1249                 if (ipv4_is_loopback(saddr))
1250                         goto e_inval;
1251
1252         if (ipv4_is_zeronet(saddr)) {
1253                 if (!ipv4_is_local_multicast(daddr))
1254                         goto e_inval;
1255         } else {
1256                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1257                                           in_dev, &itag);
1258                 if (err < 0)
1259                         goto e_err;
1260         }
1261         rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1262                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1263         if (!rth)
1264                 goto e_nobufs;
1265
1266 #ifdef CONFIG_IP_ROUTE_CLASSID
1267         rth->dst.tclassid = itag;
1268 #endif
1269         rth->dst.output = ip_rt_bug;
1270
1271         rth->rt_genid   = rt_genid(dev_net(dev));
1272         rth->rt_flags   = RTCF_MULTICAST;
1273         rth->rt_type    = RTN_MULTICAST;
1274         rth->rt_dst     = daddr;
1275         rth->rt_src     = saddr;
1276         rth->rt_route_iif = dev->ifindex;
1277         rth->rt_iif     = dev->ifindex;
1278         rth->rt_oif     = 0;
1279         rth->rt_mark    = skb->mark;
1280         rth->rt_pmtu    = 0;
1281         rth->rt_gateway = daddr;
1282         rth->fi = NULL;
1283         if (our) {
1284                 rth->dst.input= ip_local_deliver;
1285                 rth->rt_flags |= RTCF_LOCAL;
1286         }
1287
1288 #ifdef CONFIG_IP_MROUTE
1289         if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1290                 rth->dst.input = ip_mr_input;
1291 #endif
1292         RT_CACHE_STAT_INC(in_slow_mc);
1293
1294         skb_dst_set(skb, &rth->dst);
1295         return 0;
1296
1297 e_nobufs:
1298         return -ENOBUFS;
1299 e_inval:
1300         return -EINVAL;
1301 e_err:
1302         return err;
1303 }
1304
1305
1306 static void ip_handle_martian_source(struct net_device *dev,
1307                                      struct in_device *in_dev,
1308                                      struct sk_buff *skb,
1309                                      __be32 daddr,
1310                                      __be32 saddr)
1311 {
1312         RT_CACHE_STAT_INC(in_martian_src);
1313 #ifdef CONFIG_IP_ROUTE_VERBOSE
1314         if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1315                 /*
1316                  *      RFC1812 recommendation, if source is martian,
1317                  *      the only hint is MAC header.
1318                  */
1319                 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1320                         &daddr, &saddr, dev->name);
1321                 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1322                         print_hex_dump(KERN_WARNING, "ll header: ",
1323                                        DUMP_PREFIX_OFFSET, 16, 1,
1324                                        skb_mac_header(skb),
1325                                        dev->hard_header_len, true);
1326                 }
1327         }
1328 #endif
1329 }
1330
1331 /* called in rcu_read_lock() section */
1332 static int __mkroute_input(struct sk_buff *skb,
1333                            const struct fib_result *res,
1334                            struct in_device *in_dev,
1335                            __be32 daddr, __be32 saddr, u32 tos,
1336                            struct rtable **result)
1337 {
1338         struct rtable *rth;
1339         int err;
1340         struct in_device *out_dev;
1341         unsigned int flags = 0;
1342         u32 itag;
1343
1344         /* get a working reference to the output device */
1345         out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1346         if (out_dev == NULL) {
1347                 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1348                 return -EINVAL;
1349         }
1350
1351
1352         err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1353                                   in_dev->dev, in_dev, &itag);
1354         if (err < 0) {
1355                 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1356                                          saddr);
1357
1358                 goto cleanup;
1359         }
1360
1361         if (err)
1362                 flags |= RTCF_DIRECTSRC;
1363
1364         if (out_dev == in_dev && err &&
1365             (IN_DEV_SHARED_MEDIA(out_dev) ||
1366              inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1367                 flags |= RTCF_DOREDIRECT;
1368
1369         if (skb->protocol != htons(ETH_P_IP)) {
1370                 /* Not IP (i.e. ARP). Do not create route, if it is
1371                  * invalid for proxy arp. DNAT routes are always valid.
1372                  *
1373                  * Proxy arp feature have been extended to allow, ARP
1374                  * replies back to the same interface, to support
1375                  * Private VLAN switch technologies. See arp.c.
1376                  */
1377                 if (out_dev == in_dev &&
1378                     IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1379                         err = -EINVAL;
1380                         goto cleanup;
1381                 }
1382         }
1383
1384         rth = rt_dst_alloc(out_dev->dev,
1385                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1386                            IN_DEV_CONF_GET(out_dev, NOXFRM));
1387         if (!rth) {
1388                 err = -ENOBUFS;
1389                 goto cleanup;
1390         }
1391
1392         rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
1393         rth->rt_flags = flags;
1394         rth->rt_type = res->type;
1395         rth->rt_dst     = daddr;
1396         rth->rt_src     = saddr;
1397         rth->rt_route_iif = in_dev->dev->ifindex;
1398         rth->rt_iif     = in_dev->dev->ifindex;
1399         rth->rt_oif     = 0;
1400         rth->rt_mark    = skb->mark;
1401         rth->rt_pmtu    = 0;
1402         rth->rt_gateway = daddr;
1403         rth->fi = NULL;
1404
1405         rth->dst.input = ip_forward;
1406         rth->dst.output = ip_output;
1407
1408         rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag);
1409
1410         *result = rth;
1411         err = 0;
1412  cleanup:
1413         return err;
1414 }
1415
1416 static int ip_mkroute_input(struct sk_buff *skb,
1417                             struct fib_result *res,
1418                             const struct flowi4 *fl4,
1419                             struct in_device *in_dev,
1420                             __be32 daddr, __be32 saddr, u32 tos)
1421 {
1422         struct rtable *rth = NULL;
1423         int err;
1424
1425 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1426         if (res->fi && res->fi->fib_nhs > 1)
1427                 fib_select_multipath(res);
1428 #endif
1429
1430         /* create a routing cache entry */
1431         err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
1432         if (err)
1433                 return err;
1434
1435         skb_dst_set(skb, &rth->dst);
1436         return 0;
1437 }
1438
1439 /*
1440  *      NOTE. We drop all the packets that has local source
1441  *      addresses, because every properly looped back packet
1442  *      must have correct destination already attached by output routine.
1443  *
1444  *      Such approach solves two big problems:
1445  *      1. Not simplex devices are handled properly.
1446  *      2. IP spoofing attempts are filtered with 100% of guarantee.
1447  *      called with rcu_read_lock()
1448  */
1449
1450 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1451                                u8 tos, struct net_device *dev)
1452 {
1453         struct fib_result res;
1454         struct in_device *in_dev = __in_dev_get_rcu(dev);
1455         struct flowi4   fl4;
1456         unsigned int    flags = 0;
1457         u32             itag = 0;
1458         struct rtable   *rth;
1459         int             err = -EINVAL;
1460         struct net    *net = dev_net(dev);
1461
1462         /* IP on this device is disabled. */
1463
1464         if (!in_dev)
1465                 goto out;
1466
1467         /* Check for the most weird martians, which can be not detected
1468            by fib_lookup.
1469          */
1470
1471         if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1472                 goto martian_source;
1473
1474         if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1475                 goto brd_input;
1476
1477         /* Accept zero addresses only to limited broadcast;
1478          * I even do not know to fix it or not. Waiting for complains :-)
1479          */
1480         if (ipv4_is_zeronet(saddr))
1481                 goto martian_source;
1482
1483         if (ipv4_is_zeronet(daddr))
1484                 goto martian_destination;
1485
1486         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) {
1487                 if (ipv4_is_loopback(daddr))
1488                         goto martian_destination;
1489
1490                 if (ipv4_is_loopback(saddr))
1491                         goto martian_source;
1492         }
1493
1494         /*
1495          *      Now we are ready to route packet.
1496          */
1497         fl4.flowi4_oif = 0;
1498         fl4.flowi4_iif = dev->ifindex;
1499         fl4.flowi4_mark = skb->mark;
1500         fl4.flowi4_tos = tos;
1501         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1502         fl4.daddr = daddr;
1503         fl4.saddr = saddr;
1504         err = fib_lookup(net, &fl4, &res);
1505         if (err != 0)
1506                 goto no_route;
1507
1508         RT_CACHE_STAT_INC(in_slow_tot);
1509
1510         if (res.type == RTN_BROADCAST)
1511                 goto brd_input;
1512
1513         if (res.type == RTN_LOCAL) {
1514                 err = fib_validate_source(skb, saddr, daddr, tos,
1515                                           net->loopback_dev->ifindex,
1516                                           dev, in_dev, &itag);
1517                 if (err < 0)
1518                         goto martian_source_keep_err;
1519                 if (err)
1520                         flags |= RTCF_DIRECTSRC;
1521                 goto local_input;
1522         }
1523
1524         if (!IN_DEV_FORWARD(in_dev))
1525                 goto no_route;
1526         if (res.type != RTN_UNICAST)
1527                 goto martian_destination;
1528
1529         err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1530 out:    return err;
1531
1532 brd_input:
1533         if (skb->protocol != htons(ETH_P_IP))
1534                 goto e_inval;
1535
1536         if (!ipv4_is_zeronet(saddr)) {
1537                 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1538                                           in_dev, &itag);
1539                 if (err < 0)
1540                         goto martian_source_keep_err;
1541                 if (err)
1542                         flags |= RTCF_DIRECTSRC;
1543         }
1544         flags |= RTCF_BROADCAST;
1545         res.type = RTN_BROADCAST;
1546         RT_CACHE_STAT_INC(in_brd);
1547
1548 local_input:
1549         rth = rt_dst_alloc(net->loopback_dev,
1550                            IN_DEV_CONF_GET(in_dev, NOPOLICY), false);
1551         if (!rth)
1552                 goto e_nobufs;
1553
1554         rth->dst.input= ip_local_deliver;
1555         rth->dst.output= ip_rt_bug;
1556 #ifdef CONFIG_IP_ROUTE_CLASSID
1557         rth->dst.tclassid = itag;
1558 #endif
1559
1560         rth->rt_genid = rt_genid(net);
1561         rth->rt_flags   = flags|RTCF_LOCAL;
1562         rth->rt_type    = res.type;
1563         rth->rt_dst     = daddr;
1564         rth->rt_src     = saddr;
1565         rth->rt_route_iif = dev->ifindex;
1566         rth->rt_iif     = dev->ifindex;
1567         rth->rt_oif     = 0;
1568         rth->rt_mark    = skb->mark;
1569         rth->rt_pmtu    = 0;
1570         rth->rt_gateway = daddr;
1571         rth->fi = NULL;
1572         if (res.type == RTN_UNREACHABLE) {
1573                 rth->dst.input= ip_error;
1574                 rth->dst.error= -err;
1575                 rth->rt_flags   &= ~RTCF_LOCAL;
1576         }
1577         skb_dst_set(skb, &rth->dst);
1578         err = 0;
1579         goto out;
1580
1581 no_route:
1582         RT_CACHE_STAT_INC(in_no_route);
1583         res.type = RTN_UNREACHABLE;
1584         if (err == -ESRCH)
1585                 err = -ENETUNREACH;
1586         goto local_input;
1587
1588         /*
1589          *      Do not cache martian addresses: they should be logged (RFC1812)
1590          */
1591 martian_destination:
1592         RT_CACHE_STAT_INC(in_martian_dst);
1593 #ifdef CONFIG_IP_ROUTE_VERBOSE
1594         if (IN_DEV_LOG_MARTIANS(in_dev))
1595                 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1596                                      &daddr, &saddr, dev->name);
1597 #endif
1598
1599 e_inval:
1600         err = -EINVAL;
1601         goto out;
1602
1603 e_nobufs:
1604         err = -ENOBUFS;
1605         goto out;
1606
1607 martian_source:
1608         err = -EINVAL;
1609 martian_source_keep_err:
1610         ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1611         goto out;
1612 }
1613
1614 int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1615                    u8 tos, struct net_device *dev)
1616 {
1617         int res;
1618
1619         rcu_read_lock();
1620
1621         /* Multicast recognition logic is moved from route cache to here.
1622            The problem was that too many Ethernet cards have broken/missing
1623            hardware multicast filters :-( As result the host on multicasting
1624            network acquires a lot of useless route cache entries, sort of
1625            SDR messages from all the world. Now we try to get rid of them.
1626            Really, provided software IP multicast filter is organized
1627            reasonably (at least, hashed), it does not result in a slowdown
1628            comparing with route cache reject entries.
1629            Note, that multicast routers are not affected, because
1630            route cache entry is created eventually.
1631          */
1632         if (ipv4_is_multicast(daddr)) {
1633                 struct in_device *in_dev = __in_dev_get_rcu(dev);
1634
1635                 if (in_dev) {
1636                         int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1637                                                   ip_hdr(skb)->protocol);
1638                         if (our
1639 #ifdef CONFIG_IP_MROUTE
1640                                 ||
1641                             (!ipv4_is_local_multicast(daddr) &&
1642                              IN_DEV_MFORWARD(in_dev))
1643 #endif
1644                            ) {
1645                                 int res = ip_route_input_mc(skb, daddr, saddr,
1646                                                             tos, dev, our);
1647                                 rcu_read_unlock();
1648                                 return res;
1649                         }
1650                 }
1651                 rcu_read_unlock();
1652                 return -EINVAL;
1653         }
1654         res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1655         rcu_read_unlock();
1656         return res;
1657 }
1658 EXPORT_SYMBOL(ip_route_input);
1659
1660 /* called with rcu_read_lock() */
1661 static struct rtable *__mkroute_output(const struct fib_result *res,
1662                                        const struct flowi4 *fl4, int orig_oif,
1663                                        struct net_device *dev_out,
1664                                        unsigned int flags)
1665 {
1666         struct fib_info *fi = res->fi;
1667         struct in_device *in_dev;
1668         u16 type = res->type;
1669         struct rtable *rth;
1670
1671         in_dev = __in_dev_get_rcu(dev_out);
1672         if (!in_dev)
1673                 return ERR_PTR(-EINVAL);
1674
1675         if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1676                 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1677                         return ERR_PTR(-EINVAL);
1678
1679         if (ipv4_is_lbcast(fl4->daddr))
1680                 type = RTN_BROADCAST;
1681         else if (ipv4_is_multicast(fl4->daddr))
1682                 type = RTN_MULTICAST;
1683         else if (ipv4_is_zeronet(fl4->daddr))
1684                 return ERR_PTR(-EINVAL);
1685
1686         if (dev_out->flags & IFF_LOOPBACK)
1687                 flags |= RTCF_LOCAL;
1688
1689         if (type == RTN_BROADCAST) {
1690                 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1691                 fi = NULL;
1692         } else if (type == RTN_MULTICAST) {
1693                 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1694                 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1695                                      fl4->flowi4_proto))
1696                         flags &= ~RTCF_LOCAL;
1697                 /* If multicast route do not exist use
1698                  * default one, but do not gateway in this case.
1699                  * Yes, it is hack.
1700                  */
1701                 if (fi && res->prefixlen < 4)
1702                         fi = NULL;
1703         }
1704
1705         rth = rt_dst_alloc(dev_out,
1706                            IN_DEV_CONF_GET(in_dev, NOPOLICY),
1707                            IN_DEV_CONF_GET(in_dev, NOXFRM));
1708         if (!rth)
1709                 return ERR_PTR(-ENOBUFS);
1710
1711         rth->dst.output = ip_output;
1712
1713         rth->rt_genid = rt_genid(dev_net(dev_out));
1714         rth->rt_flags   = flags;
1715         rth->rt_type    = type;
1716         rth->rt_dst     = fl4->daddr;
1717         rth->rt_src     = fl4->saddr;
1718         rth->rt_route_iif = 0;
1719         rth->rt_iif     = orig_oif ? : dev_out->ifindex;
1720         rth->rt_oif     = orig_oif;
1721         rth->rt_mark    = fl4->flowi4_mark;
1722         rth->rt_pmtu    = 0;
1723         rth->rt_gateway = fl4->daddr;
1724         rth->fi = NULL;
1725
1726         RT_CACHE_STAT_INC(out_slow_tot);
1727
1728         if (flags & RTCF_LOCAL)
1729                 rth->dst.input = ip_local_deliver;
1730         if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1731                 if (flags & RTCF_LOCAL &&
1732                     !(dev_out->flags & IFF_LOOPBACK)) {
1733                         rth->dst.output = ip_mc_output;
1734                         RT_CACHE_STAT_INC(out_slow_mc);
1735                 }
1736 #ifdef CONFIG_IP_MROUTE
1737                 if (type == RTN_MULTICAST) {
1738                         if (IN_DEV_MFORWARD(in_dev) &&
1739                             !ipv4_is_local_multicast(fl4->daddr)) {
1740                                 rth->dst.input = ip_mr_input;
1741                                 rth->dst.output = ip_mc_output;
1742                         }
1743                 }
1744 #endif
1745         }
1746
1747         rt_set_nexthop(rth, fl4, res, fi, type, 0);
1748
1749         if (fl4->flowi4_flags & FLOWI_FLAG_RT_NOCACHE)
1750                 rth->dst.flags |= DST_NOCACHE;
1751
1752         return rth;
1753 }
1754
1755 /*
1756  * Major route resolver routine.
1757  */
1758
1759 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
1760 {
1761         struct net_device *dev_out = NULL;
1762         __u8 tos = RT_FL_TOS(fl4);
1763         unsigned int flags = 0;
1764         struct fib_result res;
1765         struct rtable *rth;
1766         int orig_oif;
1767
1768         res.tclassid    = 0;
1769         res.fi          = NULL;
1770         res.table       = NULL;
1771
1772         orig_oif = fl4->flowi4_oif;
1773
1774         fl4->flowi4_iif = net->loopback_dev->ifindex;
1775         fl4->flowi4_tos = tos & IPTOS_RT_MASK;
1776         fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
1777                          RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
1778
1779         rcu_read_lock();
1780         if (fl4->saddr) {
1781                 rth = ERR_PTR(-EINVAL);
1782                 if (ipv4_is_multicast(fl4->saddr) ||
1783                     ipv4_is_lbcast(fl4->saddr) ||
1784                     ipv4_is_zeronet(fl4->saddr))
1785                         goto out;
1786
1787                 /* I removed check for oif == dev_out->oif here.
1788                    It was wrong for two reasons:
1789                    1. ip_dev_find(net, saddr) can return wrong iface, if saddr
1790                       is assigned to multiple interfaces.
1791                    2. Moreover, we are allowed to send packets with saddr
1792                       of another iface. --ANK
1793                  */
1794
1795                 if (fl4->flowi4_oif == 0 &&
1796                     (ipv4_is_multicast(fl4->daddr) ||
1797                      ipv4_is_lbcast(fl4->daddr))) {
1798                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1799                         dev_out = __ip_dev_find(net, fl4->saddr, false);
1800                         if (dev_out == NULL)
1801                                 goto out;
1802
1803                         /* Special hack: user can direct multicasts
1804                            and limited broadcast via necessary interface
1805                            without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
1806                            This hack is not just for fun, it allows
1807                            vic,vat and friends to work.
1808                            They bind socket to loopback, set ttl to zero
1809                            and expect that it will work.
1810                            From the viewpoint of routing cache they are broken,
1811                            because we are not allowed to build multicast path
1812                            with loopback source addr (look, routing cache
1813                            cannot know, that ttl is zero, so that packet
1814                            will not leave this host and route is valid).
1815                            Luckily, this hack is good workaround.
1816                          */
1817
1818                         fl4->flowi4_oif = dev_out->ifindex;
1819                         goto make_route;
1820                 }
1821
1822                 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
1823                         /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1824                         if (!__ip_dev_find(net, fl4->saddr, false))
1825                                 goto out;
1826                 }
1827         }
1828
1829
1830         if (fl4->flowi4_oif) {
1831                 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
1832                 rth = ERR_PTR(-ENODEV);
1833                 if (dev_out == NULL)
1834                         goto out;
1835
1836                 /* RACE: Check return value of inet_select_addr instead. */
1837                 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
1838                         rth = ERR_PTR(-ENETUNREACH);
1839                         goto out;
1840                 }
1841                 if (ipv4_is_local_multicast(fl4->daddr) ||
1842                     ipv4_is_lbcast(fl4->daddr)) {
1843                         if (!fl4->saddr)
1844                                 fl4->saddr = inet_select_addr(dev_out, 0,
1845                                                               RT_SCOPE_LINK);
1846                         goto make_route;
1847                 }
1848                 if (fl4->saddr) {
1849                         if (ipv4_is_multicast(fl4->daddr))
1850                                 fl4->saddr = inet_select_addr(dev_out, 0,
1851                                                               fl4->flowi4_scope);
1852                         else if (!fl4->daddr)
1853                                 fl4->saddr = inet_select_addr(dev_out, 0,
1854                                                               RT_SCOPE_HOST);
1855                 }
1856         }
1857
1858         if (!fl4->daddr) {
1859                 fl4->daddr = fl4->saddr;
1860                 if (!fl4->daddr)
1861                         fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
1862                 dev_out = net->loopback_dev;
1863                 fl4->flowi4_oif = net->loopback_dev->ifindex;
1864                 res.type = RTN_LOCAL;
1865                 flags |= RTCF_LOCAL;
1866                 goto make_route;
1867         }
1868
1869         if (fib_lookup(net, fl4, &res)) {
1870                 res.fi = NULL;
1871                 res.table = NULL;
1872                 if (fl4->flowi4_oif) {
1873                         /* Apparently, routing tables are wrong. Assume,
1874                            that the destination is on link.
1875
1876                            WHY? DW.
1877                            Because we are allowed to send to iface
1878                            even if it has NO routes and NO assigned
1879                            addresses. When oif is specified, routing
1880                            tables are looked up with only one purpose:
1881                            to catch if destination is gatewayed, rather than
1882                            direct. Moreover, if MSG_DONTROUTE is set,
1883                            we send packet, ignoring both routing tables
1884                            and ifaddr state. --ANK
1885
1886
1887                            We could make it even if oif is unknown,
1888                            likely IPv6, but we do not.
1889                          */
1890
1891                         if (fl4->saddr == 0)
1892                                 fl4->saddr = inet_select_addr(dev_out, 0,
1893                                                               RT_SCOPE_LINK);
1894                         res.type = RTN_UNICAST;
1895                         goto make_route;
1896                 }
1897                 rth = ERR_PTR(-ENETUNREACH);
1898                 goto out;
1899         }
1900
1901         if (res.type == RTN_LOCAL) {
1902                 if (!fl4->saddr) {
1903                         if (res.fi->fib_prefsrc)
1904                                 fl4->saddr = res.fi->fib_prefsrc;
1905                         else
1906                                 fl4->saddr = fl4->daddr;
1907                 }
1908                 dev_out = net->loopback_dev;
1909                 fl4->flowi4_oif = dev_out->ifindex;
1910                 res.fi = NULL;
1911                 flags |= RTCF_LOCAL;
1912                 goto make_route;
1913         }
1914
1915 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1916         if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
1917                 fib_select_multipath(&res);
1918         else
1919 #endif
1920         if (!res.prefixlen &&
1921             res.table->tb_num_default > 1 &&
1922             res.type == RTN_UNICAST && !fl4->flowi4_oif)
1923                 fib_select_default(&res);
1924
1925         if (!fl4->saddr)
1926                 fl4->saddr = FIB_RES_PREFSRC(net, res);
1927
1928         dev_out = FIB_RES_DEV(res);
1929         fl4->flowi4_oif = dev_out->ifindex;
1930
1931
1932 make_route:
1933         rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
1934
1935 out:
1936         rcu_read_unlock();
1937         return rth;
1938 }
1939 EXPORT_SYMBOL_GPL(__ip_route_output_key);
1940
1941 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
1942 {
1943         return NULL;
1944 }
1945
1946 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
1947 {
1948         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
1949
1950         return mtu ? : dst->dev->mtu;
1951 }
1952
1953 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
1954                                           struct sk_buff *skb, u32 mtu)
1955 {
1956 }
1957
1958 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
1959                                        struct sk_buff *skb)
1960 {
1961 }
1962
1963 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
1964                                           unsigned long old)
1965 {
1966         return NULL;
1967 }
1968
1969 static struct dst_ops ipv4_dst_blackhole_ops = {
1970         .family                 =       AF_INET,
1971         .protocol               =       cpu_to_be16(ETH_P_IP),
1972         .destroy                =       ipv4_dst_destroy,
1973         .check                  =       ipv4_blackhole_dst_check,
1974         .mtu                    =       ipv4_blackhole_mtu,
1975         .default_advmss         =       ipv4_default_advmss,
1976         .update_pmtu            =       ipv4_rt_blackhole_update_pmtu,
1977         .redirect               =       ipv4_rt_blackhole_redirect,
1978         .cow_metrics            =       ipv4_rt_blackhole_cow_metrics,
1979         .neigh_lookup           =       ipv4_neigh_lookup,
1980 };
1981
1982 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
1983 {
1984         struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, 0, 0);
1985         struct rtable *ort = (struct rtable *) dst_orig;
1986
1987         if (rt) {
1988                 struct dst_entry *new = &rt->dst;
1989
1990                 new->__use = 1;
1991                 new->input = dst_discard;
1992                 new->output = dst_discard;
1993
1994                 new->dev = ort->dst.dev;
1995                 if (new->dev)
1996                         dev_hold(new->dev);
1997
1998                 rt->rt_route_iif = ort->rt_route_iif;
1999                 rt->rt_iif = ort->rt_iif;
2000                 rt->rt_oif = ort->rt_oif;
2001                 rt->rt_mark = ort->rt_mark;
2002                 rt->rt_pmtu = ort->rt_pmtu;
2003
2004                 rt->rt_genid = rt_genid(net);
2005                 rt->rt_flags = ort->rt_flags;
2006                 rt->rt_type = ort->rt_type;
2007                 rt->rt_dst = ort->rt_dst;
2008                 rt->rt_src = ort->rt_src;
2009                 rt->rt_gateway = ort->rt_gateway;
2010                 rt->fi = ort->fi;
2011                 if (rt->fi)
2012                         atomic_inc(&rt->fi->fib_clntref);
2013
2014                 dst_free(new);
2015         }
2016
2017         dst_release(dst_orig);
2018
2019         return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2020 }
2021
2022 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2023                                     struct sock *sk)
2024 {
2025         struct rtable *rt = __ip_route_output_key(net, flp4);
2026
2027         if (IS_ERR(rt))
2028                 return rt;
2029
2030         if (flp4->flowi4_proto)
2031                 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2032                                                    flowi4_to_flowi(flp4),
2033                                                    sk, 0);
2034
2035         return rt;
2036 }
2037 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2038
2039 static int rt_fill_info(struct net *net,  __be32 src, u8 tos,
2040                         struct sk_buff *skb, u32 pid, u32 seq, int event,
2041                         int nowait, unsigned int flags)
2042 {
2043         struct rtable *rt = skb_rtable(skb);
2044         struct rtmsg *r;
2045         struct nlmsghdr *nlh;
2046         unsigned long expires = 0;
2047         u32 error;
2048         u32 metrics[RTAX_MAX];
2049
2050         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2051         if (nlh == NULL)
2052                 return -EMSGSIZE;
2053
2054         r = nlmsg_data(nlh);
2055         r->rtm_family    = AF_INET;
2056         r->rtm_dst_len  = 32;
2057         r->rtm_src_len  = 0;
2058         r->rtm_tos      = tos;
2059         r->rtm_table    = RT_TABLE_MAIN;
2060         if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2061                 goto nla_put_failure;
2062         r->rtm_type     = rt->rt_type;
2063         r->rtm_scope    = RT_SCOPE_UNIVERSE;
2064         r->rtm_protocol = RTPROT_UNSPEC;
2065         r->rtm_flags    = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2066         if (rt->rt_flags & RTCF_NOTIFY)
2067                 r->rtm_flags |= RTM_F_NOTIFY;
2068
2069         if (nla_put_be32(skb, RTA_DST, rt->rt_dst))
2070                 goto nla_put_failure;
2071         if (src) {
2072                 r->rtm_src_len = 32;
2073                 if (nla_put_be32(skb, RTA_SRC, src))
2074                         goto nla_put_failure;
2075         }
2076         if (rt->dst.dev &&
2077             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2078                 goto nla_put_failure;
2079 #ifdef CONFIG_IP_ROUTE_CLASSID
2080         if (rt->dst.tclassid &&
2081             nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2082                 goto nla_put_failure;
2083 #endif
2084         if (!rt_is_input_route(rt) &&
2085             rt->rt_src != src) {
2086                 if (nla_put_be32(skb, RTA_PREFSRC, rt->rt_src))
2087                         goto nla_put_failure;
2088         }
2089         if (rt->rt_dst != rt->rt_gateway &&
2090             nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2091                 goto nla_put_failure;
2092
2093         memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2094         if (rt->rt_pmtu)
2095                 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2096         if (rtnetlink_put_metrics(skb, metrics) < 0)
2097                 goto nla_put_failure;
2098
2099         if (rt->rt_mark &&
2100             nla_put_be32(skb, RTA_MARK, rt->rt_mark))
2101                 goto nla_put_failure;
2102
2103         error = rt->dst.error;
2104         expires = rt->dst.expires;
2105         if (expires) {
2106                 if (time_before(jiffies, expires))
2107                         expires -= jiffies;
2108                 else
2109                         expires = 0;
2110         }
2111
2112         if (rt_is_input_route(rt)) {
2113 #ifdef CONFIG_IP_MROUTE
2114                 __be32 dst = rt->rt_dst;
2115
2116                 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2117                     IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2118                         int err = ipmr_get_route(net, skb,
2119                                                  rt->rt_src, rt->rt_dst,
2120                                                  r, nowait);
2121                         if (err <= 0) {
2122                                 if (!nowait) {
2123                                         if (err == 0)
2124                                                 return 0;
2125                                         goto nla_put_failure;
2126                                 } else {
2127                                         if (err == -EMSGSIZE)
2128                                                 goto nla_put_failure;
2129                                         error = err;
2130                                 }
2131                         }
2132                 } else
2133 #endif
2134                         if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
2135                                 goto nla_put_failure;
2136         }
2137
2138         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2139                 goto nla_put_failure;
2140
2141         return nlmsg_end(skb, nlh);
2142
2143 nla_put_failure:
2144         nlmsg_cancel(skb, nlh);
2145         return -EMSGSIZE;
2146 }
2147
2148 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
2149 {
2150         struct net *net = sock_net(in_skb->sk);
2151         struct rtmsg *rtm;
2152         struct nlattr *tb[RTA_MAX+1];
2153         struct rtable *rt = NULL;
2154         __be32 dst = 0;
2155         __be32 src = 0;
2156         u32 iif;
2157         int err;
2158         int mark;
2159         struct sk_buff *skb;
2160
2161         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2162         if (err < 0)
2163                 goto errout;
2164
2165         rtm = nlmsg_data(nlh);
2166
2167         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2168         if (skb == NULL) {
2169                 err = -ENOBUFS;
2170                 goto errout;
2171         }
2172
2173         /* Reserve room for dummy headers, this skb can pass
2174            through good chunk of routing engine.
2175          */
2176         skb_reset_mac_header(skb);
2177         skb_reset_network_header(skb);
2178
2179         /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2180         ip_hdr(skb)->protocol = IPPROTO_ICMP;
2181         skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2182
2183         src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2184         dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2185         iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2186         mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2187
2188         if (iif) {
2189                 struct net_device *dev;
2190
2191                 dev = __dev_get_by_index(net, iif);
2192                 if (dev == NULL) {
2193                         err = -ENODEV;
2194                         goto errout_free;
2195                 }
2196
2197                 skb->protocol   = htons(ETH_P_IP);
2198                 skb->dev        = dev;
2199                 skb->mark       = mark;
2200                 local_bh_disable();
2201                 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2202                 local_bh_enable();
2203
2204                 rt = skb_rtable(skb);
2205                 if (err == 0 && rt->dst.error)
2206                         err = -rt->dst.error;
2207         } else {
2208                 struct flowi4 fl4 = {
2209                         .daddr = dst,
2210                         .saddr = src,
2211                         .flowi4_tos = rtm->rtm_tos,
2212                         .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0,
2213                         .flowi4_mark = mark,
2214                 };
2215                 rt = ip_route_output_key(net, &fl4);
2216
2217                 err = 0;
2218                 if (IS_ERR(rt))
2219                         err = PTR_ERR(rt);
2220         }
2221
2222         if (err)
2223                 goto errout_free;
2224
2225         skb_dst_set(skb, &rt->dst);
2226         if (rtm->rtm_flags & RTM_F_NOTIFY)
2227                 rt->rt_flags |= RTCF_NOTIFY;
2228
2229         err = rt_fill_info(net, src, rtm->rtm_tos, skb,
2230                            NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2231                            RTM_NEWROUTE, 0, 0);
2232         if (err <= 0)
2233                 goto errout_free;
2234
2235         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2236 errout:
2237         return err;
2238
2239 errout_free:
2240         kfree_skb(skb);
2241         goto errout;
2242 }
2243
2244 int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
2245 {
2246         return skb->len;
2247 }
2248
2249 void ip_rt_multicast_event(struct in_device *in_dev)
2250 {
2251         rt_cache_flush(dev_net(in_dev->dev), 0);
2252 }
2253
2254 #ifdef CONFIG_SYSCTL
2255 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
2256                                         void __user *buffer,
2257                                         size_t *lenp, loff_t *ppos)
2258 {
2259         if (write) {
2260                 int flush_delay;
2261                 ctl_table ctl;
2262                 struct net *net;
2263
2264                 memcpy(&ctl, __ctl, sizeof(ctl));
2265                 ctl.data = &flush_delay;
2266                 proc_dointvec(&ctl, write, buffer, lenp, ppos);
2267
2268                 net = (struct net *)__ctl->extra1;
2269                 rt_cache_flush(net, flush_delay);
2270                 return 0;
2271         }
2272
2273         return -EINVAL;
2274 }
2275
2276 static ctl_table ipv4_route_table[] = {
2277         {
2278                 .procname       = "gc_thresh",
2279                 .data           = &ipv4_dst_ops.gc_thresh,
2280                 .maxlen         = sizeof(int),
2281                 .mode           = 0644,
2282                 .proc_handler   = proc_dointvec,
2283         },
2284         {
2285                 .procname       = "max_size",
2286                 .data           = &ip_rt_max_size,
2287                 .maxlen         = sizeof(int),
2288                 .mode           = 0644,
2289                 .proc_handler   = proc_dointvec,
2290         },
2291         {
2292                 /*  Deprecated. Use gc_min_interval_ms */
2293
2294                 .procname       = "gc_min_interval",
2295                 .data           = &ip_rt_gc_min_interval,
2296                 .maxlen         = sizeof(int),
2297                 .mode           = 0644,
2298                 .proc_handler   = proc_dointvec_jiffies,
2299         },
2300         {
2301                 .procname       = "gc_min_interval_ms",
2302                 .data           = &ip_rt_gc_min_interval,
2303                 .maxlen         = sizeof(int),
2304                 .mode           = 0644,
2305                 .proc_handler   = proc_dointvec_ms_jiffies,
2306         },
2307         {
2308                 .procname       = "gc_timeout",
2309                 .data           = &ip_rt_gc_timeout,
2310                 .maxlen         = sizeof(int),
2311                 .mode           = 0644,
2312                 .proc_handler   = proc_dointvec_jiffies,
2313         },
2314         {
2315                 .procname       = "gc_interval",
2316                 .data           = &ip_rt_gc_interval,
2317                 .maxlen         = sizeof(int),
2318                 .mode           = 0644,
2319                 .proc_handler   = proc_dointvec_jiffies,
2320         },
2321         {
2322                 .procname       = "redirect_load",
2323                 .data           = &ip_rt_redirect_load,
2324                 .maxlen         = sizeof(int),
2325                 .mode           = 0644,
2326                 .proc_handler   = proc_dointvec,
2327         },
2328         {
2329                 .procname       = "redirect_number",
2330                 .data           = &ip_rt_redirect_number,
2331                 .maxlen         = sizeof(int),
2332                 .mode           = 0644,
2333                 .proc_handler   = proc_dointvec,
2334         },
2335         {
2336                 .procname       = "redirect_silence",
2337                 .data           = &ip_rt_redirect_silence,
2338                 .maxlen         = sizeof(int),
2339                 .mode           = 0644,
2340                 .proc_handler   = proc_dointvec,
2341         },
2342         {
2343                 .procname       = "error_cost",
2344                 .data           = &ip_rt_error_cost,
2345                 .maxlen         = sizeof(int),
2346                 .mode           = 0644,
2347                 .proc_handler   = proc_dointvec,
2348         },
2349         {
2350                 .procname       = "error_burst",
2351                 .data           = &ip_rt_error_burst,
2352                 .maxlen         = sizeof(int),
2353                 .mode           = 0644,
2354                 .proc_handler   = proc_dointvec,
2355         },
2356         {
2357                 .procname       = "gc_elasticity",
2358                 .data           = &ip_rt_gc_elasticity,
2359                 .maxlen         = sizeof(int),
2360                 .mode           = 0644,
2361                 .proc_handler   = proc_dointvec,
2362         },
2363         {
2364                 .procname       = "mtu_expires",
2365                 .data           = &ip_rt_mtu_expires,
2366                 .maxlen         = sizeof(int),
2367                 .mode           = 0644,
2368                 .proc_handler   = proc_dointvec_jiffies,
2369         },
2370         {
2371                 .procname       = "min_pmtu",
2372                 .data           = &ip_rt_min_pmtu,
2373                 .maxlen         = sizeof(int),
2374                 .mode           = 0644,
2375                 .proc_handler   = proc_dointvec,
2376         },
2377         {
2378                 .procname       = "min_adv_mss",
2379                 .data           = &ip_rt_min_advmss,
2380                 .maxlen         = sizeof(int),
2381                 .mode           = 0644,
2382                 .proc_handler   = proc_dointvec,
2383         },
2384         { }
2385 };
2386
2387 static struct ctl_table ipv4_route_flush_table[] = {
2388         {
2389                 .procname       = "flush",
2390                 .maxlen         = sizeof(int),
2391                 .mode           = 0200,
2392                 .proc_handler   = ipv4_sysctl_rtcache_flush,
2393         },
2394         { },
2395 };
2396
2397 static __net_init int sysctl_route_net_init(struct net *net)
2398 {
2399         struct ctl_table *tbl;
2400
2401         tbl = ipv4_route_flush_table;
2402         if (!net_eq(net, &init_net)) {
2403                 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2404                 if (tbl == NULL)
2405                         goto err_dup;
2406         }
2407         tbl[0].extra1 = net;
2408
2409         net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2410         if (net->ipv4.route_hdr == NULL)
2411                 goto err_reg;
2412         return 0;
2413
2414 err_reg:
2415         if (tbl != ipv4_route_flush_table)
2416                 kfree(tbl);
2417 err_dup:
2418         return -ENOMEM;
2419 }
2420
2421 static __net_exit void sysctl_route_net_exit(struct net *net)
2422 {
2423         struct ctl_table *tbl;
2424
2425         tbl = net->ipv4.route_hdr->ctl_table_arg;
2426         unregister_net_sysctl_table(net->ipv4.route_hdr);
2427         BUG_ON(tbl == ipv4_route_flush_table);
2428         kfree(tbl);
2429 }
2430
2431 static __net_initdata struct pernet_operations sysctl_route_ops = {
2432         .init = sysctl_route_net_init,
2433         .exit = sysctl_route_net_exit,
2434 };
2435 #endif
2436
2437 static __net_init int rt_genid_init(struct net *net)
2438 {
2439         get_random_bytes(&net->ipv4.rt_genid,
2440                          sizeof(net->ipv4.rt_genid));
2441         get_random_bytes(&net->ipv4.dev_addr_genid,
2442                          sizeof(net->ipv4.dev_addr_genid));
2443         return 0;
2444 }
2445
2446 static __net_initdata struct pernet_operations rt_genid_ops = {
2447         .init = rt_genid_init,
2448 };
2449
2450 static int __net_init ipv4_inetpeer_init(struct net *net)
2451 {
2452         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2453
2454         if (!bp)
2455                 return -ENOMEM;
2456         inet_peer_base_init(bp);
2457         net->ipv4.peers = bp;
2458         return 0;
2459 }
2460
2461 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2462 {
2463         struct inet_peer_base *bp = net->ipv4.peers;
2464
2465         net->ipv4.peers = NULL;
2466         inetpeer_invalidate_tree(bp);
2467         kfree(bp);
2468 }
2469
2470 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2471         .init   =       ipv4_inetpeer_init,
2472         .exit   =       ipv4_inetpeer_exit,
2473 };
2474
2475 #ifdef CONFIG_IP_ROUTE_CLASSID
2476 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2477 #endif /* CONFIG_IP_ROUTE_CLASSID */
2478
2479 int __init ip_rt_init(void)
2480 {
2481         int rc = 0;
2482
2483 #ifdef CONFIG_IP_ROUTE_CLASSID
2484         ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2485         if (!ip_rt_acct)
2486                 panic("IP: failed to allocate ip_rt_acct\n");
2487 #endif
2488
2489         ipv4_dst_ops.kmem_cachep =
2490                 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2491                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2492
2493         ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2494
2495         if (dst_entries_init(&ipv4_dst_ops) < 0)
2496                 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2497
2498         if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2499                 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2500
2501         ipv4_dst_ops.gc_thresh = ~0;
2502         ip_rt_max_size = INT_MAX;
2503
2504         devinet_init();
2505         ip_fib_init();
2506
2507         if (ip_rt_proc_init())
2508                 pr_err("Unable to create route proc files\n");
2509 #ifdef CONFIG_XFRM
2510         xfrm_init();
2511         xfrm4_init(ip_rt_max_size);
2512 #endif
2513         rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2514
2515 #ifdef CONFIG_SYSCTL
2516         register_pernet_subsys(&sysctl_route_ops);
2517 #endif
2518         register_pernet_subsys(&rt_genid_ops);
2519         register_pernet_subsys(&ipv4_inetpeer_ops);
2520         return rc;
2521 }
2522
2523 #ifdef CONFIG_SYSCTL
2524 /*
2525  * We really need to sanitize the damn ipv4 init order, then all
2526  * this nonsense will go away.
2527  */
2528 void __init ip_static_sysctl_init(void)
2529 {
2530         register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
2531 }
2532 #endif