projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ipv4: Turn rt->rt_route_iif into rt->rt_is_input.
[deliverable/linux.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 bool kill_route)
678 {
679 __be32 new_gw = icmp_hdr(skb)->un.gateway;
680 __be32 old_gw = ip_hdr(skb)->saddr;
681 struct net_device *dev = skb->dev;
682 struct in_device *in_dev;
683 struct fib_result res;
684 struct neighbour *n;
685 struct net *net;
686
687 switch (icmp_hdr(skb)->code & 7) {
688 case ICMP_REDIR_NET:
689 case ICMP_REDIR_NETTOS:
690 case ICMP_REDIR_HOST:
691 case ICMP_REDIR_HOSTTOS:
692 break;
693
694 default:
695 return;
696 }
697
698 if (rt->rt_gateway != old_gw)
699 return;
700
701 in_dev = __in_dev_get_rcu(dev);
702 if (!in_dev)
703 return;
704
705 net = dev_net(dev);
706 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
707 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
708 ipv4_is_zeronet(new_gw))
709 goto reject_redirect;
710
711 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
712 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
713 goto reject_redirect;
714 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
715 goto reject_redirect;
716 } else {
717 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
718 goto reject_redirect;
719 }
720
721 n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw);
722 if (n) {
723 if (!(n->nud_state & NUD_VALID)) {
724 neigh_event_send(n, NULL);
725 } else {
726 if (fib_lookup(net, fl4, &res) == 0) {
727 struct fib_nh *nh = &FIB_RES_NH(res);
728
729 update_or_create_fnhe(nh, fl4->daddr, new_gw,
730 0, 0);
731 }
732 if (kill_route)
733 rt->dst.obsolete = DST_OBSOLETE_KILL;
734 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
735 }
736 neigh_release(n);
737 }
738 return;
739
740 reject_redirect:
741 #ifdef CONFIG_IP_ROUTE_VERBOSE
742 if (IN_DEV_LOG_MARTIANS(in_dev)) {
743 const struct iphdr *iph = (const struct iphdr *) skb->data;
744 __be32 daddr = iph->daddr;
745 __be32 saddr = iph->saddr;
746
747 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
748 " Advised path = %pI4 -> %pI4\n",
749 &old_gw, dev->name, &new_gw,
750 &saddr, &daddr);
751 }
752 #endif
753 ;
754 }
755
756 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
757 {
758 struct rtable *rt;
759 struct flowi4 fl4;
760
761 rt = (struct rtable *) dst;
762
763 ip_rt_build_flow_key(&fl4, sk, skb);
764 __ip_do_redirect(rt, skb, &fl4, true);
765 }
766
767 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
768 {
769 struct rtable *rt = (struct rtable *)dst;
770 struct dst_entry *ret = dst;
771
772 if (rt) {
773 if (dst->obsolete > 0) {
774 ip_rt_put(rt);
775 ret = NULL;
776 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
777 rt->dst.expires) {
778 ip_rt_put(rt);
779 ret = NULL;
780 }
781 }
782 return ret;
783 }
784
785 /*
786 * Algorithm:
787 * 1. The first ip_rt_redirect_number redirects are sent
788 * with exponential backoff, then we stop sending them at all,
789 * assuming that the host ignores our redirects.
790 * 2. If we did not see packets requiring redirects
791 * during ip_rt_redirect_silence, we assume that the host
792 * forgot redirected route and start to send redirects again.
793 *
794 * This algorithm is much cheaper and more intelligent than dumb load limiting
795 * in icmp.c.
796 *
797 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
798 * and "frag. need" (breaks PMTU discovery) in icmp.c.
799 */
800
801 void ip_rt_send_redirect(struct sk_buff *skb)
802 {
803 struct rtable *rt = skb_rtable(skb);
804 struct in_device *in_dev;
805 struct inet_peer *peer;
806 struct net *net;
807 int log_martians;
808
809 rcu_read_lock();
810 in_dev = __in_dev_get_rcu(rt->dst.dev);
811 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
812 rcu_read_unlock();
813 return;
814 }
815 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
816 rcu_read_unlock();
817
818 net = dev_net(rt->dst.dev);
819 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
820 if (!peer) {
821 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
822 return;
823 }
824
825 /* No redirected packets during ip_rt_redirect_silence;
826 * reset the algorithm.
827 */
828 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
829 peer->rate_tokens = 0;
830
831 /* Too many ignored redirects; do not send anything
832 * set dst.rate_last to the last seen redirected packet.
833 */
834 if (peer->rate_tokens >= ip_rt_redirect_number) {
835 peer->rate_last = jiffies;
836 goto out_put_peer;
837 }
838
839 /* Check for load limit; set rate_last to the latest sent
840 * redirect.
841 */
842 if (peer->rate_tokens == 0 ||
843 time_after(jiffies,
844 (peer->rate_last +
845 (ip_rt_redirect_load << peer->rate_tokens)))) {
846 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway);
847 peer->rate_last = jiffies;
848 ++peer->rate_tokens;
849 #ifdef CONFIG_IP_ROUTE_VERBOSE
850 if (log_martians &&
851 peer->rate_tokens == ip_rt_redirect_number)
852 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
853 &ip_hdr(skb)->saddr, rt->rt_iif,
854 &ip_hdr(skb)->daddr, &rt->rt_gateway);
855 #endif
856 }
857 out_put_peer:
858 inet_putpeer(peer);
859 }
860
861 static int ip_error(struct sk_buff *skb)
862 {
863 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
864 struct rtable *rt = skb_rtable(skb);
865 struct inet_peer *peer;
866 unsigned long now;
867 struct net *net;
868 bool send;
869 int code;
870
871 net = dev_net(rt->dst.dev);
872 if (!IN_DEV_FORWARD(in_dev)) {
873 switch (rt->dst.error) {
874 case EHOSTUNREACH:
875 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS);
876 break;
877
878 case ENETUNREACH:
879 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
880 break;
881 }
882 goto out;
883 }
884
885 switch (rt->dst.error) {
886 case EINVAL:
887 default:
888 goto out;
889 case EHOSTUNREACH:
890 code = ICMP_HOST_UNREACH;
891 break;
892 case ENETUNREACH:
893 code = ICMP_NET_UNREACH;
894 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES);
895 break;
896 case EACCES:
897 code = ICMP_PKT_FILTERED;
898 break;
899 }
900
901 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1);
902
903 send = true;
904 if (peer) {
905 now = jiffies;
906 peer->rate_tokens += now - peer->rate_last;
907 if (peer->rate_tokens > ip_rt_error_burst)
908 peer->rate_tokens = ip_rt_error_burst;
909 peer->rate_last = now;
910 if (peer->rate_tokens >= ip_rt_error_cost)
911 peer->rate_tokens -= ip_rt_error_cost;
912 else
913 send = false;
914 inet_putpeer(peer);
915 }
916 if (send)
917 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
918
919 out: kfree_skb(skb);
920 return 0;
921 }
922
923 static u32 __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
924 {
925 struct fib_result res;
926
927 if (mtu < ip_rt_min_pmtu)
928 mtu = ip_rt_min_pmtu;
929
930 if (fib_lookup(dev_net(rt->dst.dev), fl4, &res) == 0) {
931 struct fib_nh *nh = &FIB_RES_NH(res);
932
933 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
934 jiffies + ip_rt_mtu_expires);
935 }
936 return mtu;
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 mtu = __ip_rt_update_pmtu(rt, &fl4, mtu);
947
948 if (!rt->rt_pmtu) {
949 dst->obsolete = DST_OBSOLETE_KILL;
950 } else {
951 rt->rt_pmtu = mtu;
952 dst_set_expires(&rt->dst, ip_rt_mtu_expires);
953 }
954 }
955
956 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
957 int oif, u32 mark, u8 protocol, int flow_flags)
958 {
959 const struct iphdr *iph = (const struct iphdr *) skb->data;
960 struct flowi4 fl4;
961 struct rtable *rt;
962
963 __build_flow_key(&fl4, NULL, iph, oif,
964 RT_TOS(iph->tos), protocol, mark, flow_flags);
965 rt = __ip_route_output_key(net, &fl4);
966 if (!IS_ERR(rt)) {
967 __ip_rt_update_pmtu(rt, &fl4, mtu);
968 ip_rt_put(rt);
969 }
970 }
971 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
972
973 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
974 {
975 const struct iphdr *iph = (const struct iphdr *) skb->data;
976 struct flowi4 fl4;
977 struct rtable *rt;
978
979 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
980 rt = __ip_route_output_key(sock_net(sk), &fl4);
981 if (!IS_ERR(rt)) {
982 __ip_rt_update_pmtu(rt, &fl4, mtu);
983 ip_rt_put(rt);
984 }
985 }
986 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
987
988 void ipv4_redirect(struct sk_buff *skb, struct net *net,
989 int oif, u32 mark, u8 protocol, int flow_flags)
990 {
991 const struct iphdr *iph = (const struct iphdr *) skb->data;
992 struct flowi4 fl4;
993 struct rtable *rt;
994
995 __build_flow_key(&fl4, NULL, iph, oif,
996 RT_TOS(iph->tos), protocol, mark, flow_flags);
997 rt = __ip_route_output_key(net, &fl4);
998 if (!IS_ERR(rt)) {
999 __ip_do_redirect(rt, skb, &fl4, false);
1000 ip_rt_put(rt);
1001 }
1002 }
1003 EXPORT_SYMBOL_GPL(ipv4_redirect);
1004
1005 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1006 {
1007 const struct iphdr *iph = (const struct iphdr *) skb->data;
1008 struct flowi4 fl4;
1009 struct rtable *rt;
1010
1011 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0);
1012 rt = __ip_route_output_key(sock_net(sk), &fl4);
1013 if (!IS_ERR(rt)) {
1014 __ip_do_redirect(rt, skb, &fl4, false);
1015 ip_rt_put(rt);
1016 }
1017 }
1018 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1019
1020 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1021 {
1022 struct rtable *rt = (struct rtable *) dst;
1023
1024 /* All IPV4 dsts are created with ->obsolete set to the value
1025 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1026 * into this function always.
1027 *
1028 * When a PMTU/redirect information update invalidates a
1029 * route, this is indicated by setting obsolete to
1030 * DST_OBSOLETE_KILL.
1031 */
1032 if (dst->obsolete == DST_OBSOLETE_KILL || rt_is_expired(rt))
1033 return NULL;
1034 return dst;
1035 }
1036
1037 static void ipv4_dst_destroy(struct dst_entry *dst)
1038 {
1039 struct rtable *rt = (struct rtable *) dst;
1040
1041 if (rt->fi) {
1042 fib_info_put(rt->fi);
1043 rt->fi = NULL;
1044 }
1045 }
1046
1047
1048 static void ipv4_link_failure(struct sk_buff *skb)
1049 {
1050 struct rtable *rt;
1051
1052 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1053
1054 rt = skb_rtable(skb);
1055 if (rt)
1056 dst_set_expires(&rt->dst, 0);
1057 }
1058
1059 static int ip_rt_bug(struct sk_buff *skb)
1060 {
1061 pr_debug("%s: %pI4 -> %pI4, %s\n",
1062 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1063 skb->dev ? skb->dev->name : "?");
1064 kfree_skb(skb);
1065 WARN_ON(1);
1066 return 0;
1067 }
1068
1069 /*
1070 We do not cache source address of outgoing interface,
1071 because it is used only by IP RR, TS and SRR options,
1072 so that it out of fast path.
1073
1074 BTW remember: "addr" is allowed to be not aligned
1075 in IP options!
1076 */
1077
1078 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1079 {
1080 __be32 src;
1081
1082 if (rt_is_output_route(rt))
1083 src = ip_hdr(skb)->saddr;
1084 else {
1085 struct fib_result res;
1086 struct flowi4 fl4;
1087 struct iphdr *iph;
1088
1089 iph = ip_hdr(skb);
1090
1091 memset(&fl4, 0, sizeof(fl4));
1092 fl4.daddr = iph->daddr;
1093 fl4.saddr = iph->saddr;
1094 fl4.flowi4_tos = RT_TOS(iph->tos);
1095 fl4.flowi4_oif = rt->dst.dev->ifindex;
1096 fl4.flowi4_iif = skb->dev->ifindex;
1097 fl4.flowi4_mark = skb->mark;
1098
1099 rcu_read_lock();
1100 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0)
1101 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1102 else
1103 src = inet_select_addr(rt->dst.dev,
1104 rt_nexthop(rt, iph->daddr),
1105 RT_SCOPE_UNIVERSE);
1106 rcu_read_unlock();
1107 }
1108 memcpy(addr, &src, 4);
1109 }
1110
1111 #ifdef CONFIG_IP_ROUTE_CLASSID
1112 static void set_class_tag(struct rtable *rt, u32 tag)
1113 {
1114 if (!(rt->dst.tclassid & 0xFFFF))
1115 rt->dst.tclassid |= tag & 0xFFFF;
1116 if (!(rt->dst.tclassid & 0xFFFF0000))
1117 rt->dst.tclassid |= tag & 0xFFFF0000;
1118 }
1119 #endif
1120
1121 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1122 {
1123 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1124
1125 if (advmss == 0) {
1126 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1127 ip_rt_min_advmss);
1128 if (advmss > 65535 - 40)
1129 advmss = 65535 - 40;
1130 }
1131 return advmss;
1132 }
1133
1134 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1135 {
1136 const struct rtable *rt = (const struct rtable *) dst;
1137 unsigned int mtu = rt->rt_pmtu;
1138
1139 if (mtu && time_after_eq(jiffies, rt->dst.expires))
1140 mtu = 0;
1141
1142 if (!mtu)
1143 mtu = dst_metric_raw(dst, RTAX_MTU);
1144
1145 if (mtu && rt_is_output_route(rt))
1146 return mtu;
1147
1148 mtu = dst->dev->mtu;
1149
1150 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1151 if (rt->rt_gateway && mtu > 576)
1152 mtu = 576;
1153 }
1154
1155 if (mtu > IP_MAX_MTU)
1156 mtu = IP_MAX_MTU;
1157
1158 return mtu;
1159 }
1160
1161 static void rt_init_metrics(struct rtable *rt, struct fib_info *fi)
1162 {
1163 if (fi->fib_metrics != (u32 *) dst_default_metrics) {
1164 rt->fi = fi;
1165 atomic_inc(&fi->fib_clntref);
1166 }
1167 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1168 }
1169
1170 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1171 {
1172 struct fnhe_hash_bucket *hash = nh->nh_exceptions;
1173 struct fib_nh_exception *fnhe;
1174 u32 hval;
1175
1176 if (!hash)
1177 return NULL;
1178
1179 hval = fnhe_hashfun(daddr);
1180
1181 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1182 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1183 if (fnhe->fnhe_daddr == daddr)
1184 return fnhe;
1185 }
1186 return NULL;
1187 }
1188
1189 static void rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1190 __be32 daddr)
1191 {
1192 __be32 fnhe_daddr, gw;
1193 unsigned long expires;
1194 unsigned int seq;
1195 u32 pmtu;
1196
1197 restart:
1198 seq = read_seqbegin(&fnhe_seqlock);
1199 fnhe_daddr = fnhe->fnhe_daddr;
1200 gw = fnhe->fnhe_gw;
1201 pmtu = fnhe->fnhe_pmtu;
1202 expires = fnhe->fnhe_expires;
1203 if (read_seqretry(&fnhe_seqlock, seq))
1204 goto restart;
1205
1206 if (daddr != fnhe_daddr)
1207 return;
1208
1209 if (pmtu) {
1210 unsigned long diff = expires - jiffies;
1211
1212 if (time_before(jiffies, expires)) {
1213 rt->rt_pmtu = pmtu;
1214 dst_set_expires(&rt->dst, diff);
1215 }
1216 }
1217 if (gw) {
1218 rt->rt_flags |= RTCF_REDIRECTED;
1219 rt->rt_gateway = gw;
1220 }
1221 fnhe->fnhe_stamp = jiffies;
1222 }
1223
1224 static inline void rt_release_rcu(struct rcu_head *head)
1225 {
1226 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head);
1227 dst_release(dst);
1228 }
1229
1230 static void rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1231 {
1232 struct rtable *orig, *prev, **p = &nh->nh_rth_output;
1233
1234 if (rt_is_input_route(rt))
1235 p = &nh->nh_rth_input;
1236
1237 orig = *p;
1238
1239 prev = cmpxchg(p, orig, rt);
1240 if (prev == orig) {
1241 dst_clone(&rt->dst);
1242 if (orig)
1243 call_rcu_bh(&orig->dst.rcu_head, rt_release_rcu);
1244 }
1245 }
1246
1247 static bool rt_cache_valid(struct rtable *rt)
1248 {
1249 return (rt && rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK);
1250 }
1251
1252 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1253 const struct fib_result *res,
1254 struct fib_nh_exception *fnhe,
1255 struct fib_info *fi, u16 type, u32 itag)
1256 {
1257 if (fi) {
1258 struct fib_nh *nh = &FIB_RES_NH(*res);
1259
1260 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK)
1261 rt->rt_gateway = nh->nh_gw;
1262 if (unlikely(fnhe))
1263 rt_bind_exception(rt, fnhe, daddr);
1264 rt_init_metrics(rt, fi);
1265 #ifdef CONFIG_IP_ROUTE_CLASSID
1266 rt->dst.tclassid = nh->nh_tclassid;
1267 #endif
1268 if (!(rt->dst.flags & DST_HOST))
1269 rt_cache_route(nh, rt);
1270 }
1271
1272 #ifdef CONFIG_IP_ROUTE_CLASSID
1273 #ifdef CONFIG_IP_MULTIPLE_TABLES
1274 set_class_tag(rt, res->tclassid);
1275 #endif
1276 set_class_tag(rt, itag);
1277 #endif
1278 }
1279
1280 static struct rtable *rt_dst_alloc(struct net_device *dev,
1281 bool nopolicy, bool noxfrm, bool will_cache)
1282 {
1283 return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1284 (will_cache ? 0 : DST_HOST) | DST_NOCACHE |
1285 (nopolicy ? DST_NOPOLICY : 0) |
1286 (noxfrm ? DST_NOXFRM : 0));
1287 }
1288
1289 /* called in rcu_read_lock() section */
1290 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1291 u8 tos, struct net_device *dev, int our)
1292 {
1293 struct rtable *rth;
1294 struct in_device *in_dev = __in_dev_get_rcu(dev);
1295 u32 itag = 0;
1296 int err;
1297
1298 /* Primary sanity checks. */
1299
1300 if (in_dev == NULL)
1301 return -EINVAL;
1302
1303 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1304 skb->protocol != htons(ETH_P_IP))
1305 goto e_inval;
1306
1307 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1308 if (ipv4_is_loopback(saddr))
1309 goto e_inval;
1310
1311 if (ipv4_is_zeronet(saddr)) {
1312 if (!ipv4_is_local_multicast(daddr))
1313 goto e_inval;
1314 } else {
1315 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1316 in_dev, &itag);
1317 if (err < 0)
1318 goto e_err;
1319 }
1320 rth = rt_dst_alloc(dev_net(dev)->loopback_dev,
1321 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1322 if (!rth)
1323 goto e_nobufs;
1324
1325 #ifdef CONFIG_IP_ROUTE_CLASSID
1326 rth->dst.tclassid = itag;
1327 #endif
1328 rth->dst.output = ip_rt_bug;
1329
1330 rth->rt_genid = rt_genid(dev_net(dev));
1331 rth->rt_flags = RTCF_MULTICAST;
1332 rth->rt_type = RTN_MULTICAST;
1333 rth->rt_is_input= 1;
1334 rth->rt_iif = dev->ifindex;
1335 rth->rt_pmtu = 0;
1336 rth->rt_gateway = 0;
1337 rth->fi = NULL;
1338 if (our) {
1339 rth->dst.input= ip_local_deliver;
1340 rth->rt_flags |= RTCF_LOCAL;
1341 }
1342
1343 #ifdef CONFIG_IP_MROUTE
1344 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1345 rth->dst.input = ip_mr_input;
1346 #endif
1347 RT_CACHE_STAT_INC(in_slow_mc);
1348
1349 skb_dst_set(skb, &rth->dst);
1350 return 0;
1351
1352 e_nobufs:
1353 return -ENOBUFS;
1354 e_inval:
1355 return -EINVAL;
1356 e_err:
1357 return err;
1358 }
1359
1360
1361 static void ip_handle_martian_source(struct net_device *dev,
1362 struct in_device *in_dev,
1363 struct sk_buff *skb,
1364 __be32 daddr,
1365 __be32 saddr)
1366 {
1367 RT_CACHE_STAT_INC(in_martian_src);
1368 #ifdef CONFIG_IP_ROUTE_VERBOSE
1369 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1370 /*
1371 * RFC1812 recommendation, if source is martian,
1372 * the only hint is MAC header.
1373 */
1374 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1375 &daddr, &saddr, dev->name);
1376 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1377 print_hex_dump(KERN_WARNING, "ll header: ",
1378 DUMP_PREFIX_OFFSET, 16, 1,
1379 skb_mac_header(skb),
1380 dev->hard_header_len, true);
1381 }
1382 }
1383 #endif
1384 }
1385
1386 /* called in rcu_read_lock() section */
1387 static int __mkroute_input(struct sk_buff *skb,
1388 const struct fib_result *res,
1389 struct in_device *in_dev,
1390 __be32 daddr, __be32 saddr, u32 tos,
1391 struct rtable **result)
1392 {
1393 struct rtable *rth;
1394 int err;
1395 struct in_device *out_dev;
1396 unsigned int flags = 0;
1397 bool do_cache;
1398 u32 itag;
1399
1400 /* get a working reference to the output device */
1401 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1402 if (out_dev == NULL) {
1403 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1404 return -EINVAL;
1405 }
1406
1407
1408 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1409 in_dev->dev, in_dev, &itag);
1410 if (err < 0) {
1411 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1412 saddr);
1413
1414 goto cleanup;
1415 }
1416
1417 if (err)
1418 flags |= RTCF_DIRECTSRC;
1419
1420 if (out_dev == in_dev && err &&
1421 (IN_DEV_SHARED_MEDIA(out_dev) ||
1422 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1423 flags |= RTCF_DOREDIRECT;
1424
1425 if (skb->protocol != htons(ETH_P_IP)) {
1426 /* Not IP (i.e. ARP). Do not create route, if it is
1427 * invalid for proxy arp. DNAT routes are always valid.
1428 *
1429 * Proxy arp feature have been extended to allow, ARP
1430 * replies back to the same interface, to support
1431 * Private VLAN switch technologies. See arp.c.
1432 */
1433 if (out_dev == in_dev &&
1434 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1435 err = -EINVAL;
1436 goto cleanup;
1437 }
1438 }
1439
1440 do_cache = false;
1441 if (res->fi) {
1442 if (!(flags & RTCF_DIRECTSRC) && !itag) {
1443 rth = FIB_RES_NH(*res).nh_rth_input;
1444 if (rt_cache_valid(rth)) {
1445 dst_hold(&rth->dst);
1446 goto out;
1447 }
1448 do_cache = true;
1449 }
1450 }
1451
1452 rth = rt_dst_alloc(out_dev->dev,
1453 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1454 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1455 if (!rth) {
1456 err = -ENOBUFS;
1457 goto cleanup;
1458 }
1459
1460 rth->rt_genid = rt_genid(dev_net(rth->dst.dev));
1461 rth->rt_flags = flags;
1462 rth->rt_type = res->type;
1463 rth->rt_is_input = 1;
1464 rth->rt_iif = in_dev->dev->ifindex;
1465 rth->rt_pmtu = 0;
1466 rth->rt_gateway = 0;
1467 rth->fi = NULL;
1468
1469 rth->dst.input = ip_forward;
1470 rth->dst.output = ip_output;
1471
1472 rt_set_nexthop(rth, daddr, res, NULL, res->fi, res->type, itag);
1473 out:
1474 *result = rth;
1475 err = 0;
1476 cleanup:
1477 return err;
1478 }
1479
1480 static int ip_mkroute_input(struct sk_buff *skb,
1481 struct fib_result *res,
1482 const struct flowi4 *fl4,
1483 struct in_device *in_dev,
1484 __be32 daddr, __be32 saddr, u32 tos)
1485 {
1486 struct rtable *rth = NULL;
1487 int err;
1488
1489 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1490 if (res->fi && res->fi->fib_nhs > 1)
1491 fib_select_multipath(res);
1492 #endif
1493
1494 /* create a routing cache entry */
1495 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
1496 if (err)
1497 return err;
1498
1499 skb_dst_set(skb, &rth->dst);
1500 return 0;
1501 }
1502
1503 /*
1504 * NOTE. We drop all the packets that has local source
1505 * addresses, because every properly looped back packet
1506 * must have correct destination already attached by output routine.
1507 *
1508 * Such approach solves two big problems:
1509 * 1. Not simplex devices are handled properly.
1510 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1511 * called with rcu_read_lock()
1512 */
1513
1514 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1515 u8 tos, struct net_device *dev)
1516 {
1517 struct fib_result res;
1518 struct in_device *in_dev = __in_dev_get_rcu(dev);
1519 struct flowi4 fl4;
1520 unsigned int flags = 0;
1521 u32 itag = 0;
1522 struct rtable *rth;
1523 int err = -EINVAL;
1524 struct net *net = dev_net(dev);
1525 bool do_cache;
1526
1527 /* IP on this device is disabled. */
1528
1529 if (!in_dev)
1530 goto out;
1531
1532 /* Check for the most weird martians, which can be not detected
1533 by fib_lookup.
1534 */
1535
1536 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1537 goto martian_source;
1538
1539 res.fi = NULL;
1540 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1541 goto brd_input;
1542
1543 /* Accept zero addresses only to limited broadcast;
1544 * I even do not know to fix it or not. Waiting for complains :-)
1545 */
1546 if (ipv4_is_zeronet(saddr))
1547 goto martian_source;
1548
1549 if (ipv4_is_zeronet(daddr))
1550 goto martian_destination;
1551
1552 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) {
1553 if (ipv4_is_loopback(daddr))
1554 goto martian_destination;
1555
1556 if (ipv4_is_loopback(saddr))
1557 goto martian_source;
1558 }
1559
1560 /*
1561 * Now we are ready to route packet.
1562 */
1563 fl4.flowi4_oif = 0;
1564 fl4.flowi4_iif = dev->ifindex;
1565 fl4.flowi4_mark = skb->mark;
1566 fl4.flowi4_tos = tos;
1567 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1568 fl4.daddr = daddr;
1569 fl4.saddr = saddr;
1570 err = fib_lookup(net, &fl4, &res);
1571 if (err != 0)
1572 goto no_route;
1573
1574 RT_CACHE_STAT_INC(in_slow_tot);
1575
1576 if (res.type == RTN_BROADCAST)
1577 goto brd_input;
1578
1579 if (res.type == RTN_LOCAL) {
1580 err = fib_validate_source(skb, saddr, daddr, tos,
1581 net->loopback_dev->ifindex,
1582 dev, in_dev, &itag);
1583 if (err < 0)
1584 goto martian_source_keep_err;
1585 if (err)
1586 flags |= RTCF_DIRECTSRC;
1587 goto local_input;
1588 }
1589
1590 if (!IN_DEV_FORWARD(in_dev))
1591 goto no_route;
1592 if (res.type != RTN_UNICAST)
1593 goto martian_destination;
1594
1595 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos);
1596 out: return err;
1597
1598 brd_input:
1599 if (skb->protocol != htons(ETH_P_IP))
1600 goto e_inval;
1601
1602 if (!ipv4_is_zeronet(saddr)) {
1603 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1604 in_dev, &itag);
1605 if (err < 0)
1606 goto martian_source_keep_err;
1607 if (err)
1608 flags |= RTCF_DIRECTSRC;
1609 }
1610 flags |= RTCF_BROADCAST;
1611 res.type = RTN_BROADCAST;
1612 RT_CACHE_STAT_INC(in_brd);
1613
1614 local_input:
1615 do_cache = false;
1616 if (res.fi) {
1617 if (!(flags & RTCF_DIRECTSRC) && !itag) {
1618 rth = FIB_RES_NH(res).nh_rth_input;
1619 if (rt_cache_valid(rth)) {
1620 dst_hold(&rth->dst);
1621 goto set_and_out;
1622 }
1623 do_cache = true;
1624 }
1625 }
1626
1627 rth = rt_dst_alloc(net->loopback_dev,
1628 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1629 if (!rth)
1630 goto e_nobufs;
1631
1632 rth->dst.input= ip_local_deliver;
1633 rth->dst.output= ip_rt_bug;
1634 #ifdef CONFIG_IP_ROUTE_CLASSID
1635 rth->dst.tclassid = itag;
1636 #endif
1637
1638 rth->rt_genid = rt_genid(net);
1639 rth->rt_flags = flags|RTCF_LOCAL;
1640 rth->rt_type = res.type;
1641 rth->rt_is_input = 1;
1642 rth->rt_iif = dev->ifindex;
1643 rth->rt_pmtu = 0;
1644 rth->rt_gateway = 0;
1645 rth->fi = NULL;
1646 if (res.type == RTN_UNREACHABLE) {
1647 rth->dst.input= ip_error;
1648 rth->dst.error= -err;
1649 rth->rt_flags &= ~RTCF_LOCAL;
1650 }
1651 if (do_cache)
1652 rt_cache_route(&FIB_RES_NH(res), rth);
1653 set_and_out:
1654 skb_dst_set(skb, &rth->dst);
1655 err = 0;
1656 goto out;
1657
1658 no_route:
1659 RT_CACHE_STAT_INC(in_no_route);
1660 res.type = RTN_UNREACHABLE;
1661 if (err == -ESRCH)
1662 err = -ENETUNREACH;
1663 goto local_input;
1664
1665 /*
1666 * Do not cache martian addresses: they should be logged (RFC1812)
1667 */
1668 martian_destination:
1669 RT_CACHE_STAT_INC(in_martian_dst);
1670 #ifdef CONFIG_IP_ROUTE_VERBOSE
1671 if (IN_DEV_LOG_MARTIANS(in_dev))
1672 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1673 &daddr, &saddr, dev->name);
1674 #endif
1675
1676 e_inval:
1677 err = -EINVAL;
1678 goto out;
1679
1680 e_nobufs:
1681 err = -ENOBUFS;
1682 goto out;
1683
1684 martian_source:
1685 err = -EINVAL;
1686 martian_source_keep_err:
1687 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
1688 goto out;
1689 }
1690
1691 int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1692 u8 tos, struct net_device *dev)
1693 {
1694 int res;
1695
1696 rcu_read_lock();
1697
1698 /* Multicast recognition logic is moved from route cache to here.
1699 The problem was that too many Ethernet cards have broken/missing
1700 hardware multicast filters :-( As result the host on multicasting
1701 network acquires a lot of useless route cache entries, sort of
1702 SDR messages from all the world. Now we try to get rid of them.
1703 Really, provided software IP multicast filter is organized
1704 reasonably (at least, hashed), it does not result in a slowdown
1705 comparing with route cache reject entries.
1706 Note, that multicast routers are not affected, because
1707 route cache entry is created eventually.
1708 */
1709 if (ipv4_is_multicast(daddr)) {
1710 struct in_device *in_dev = __in_dev_get_rcu(dev);
1711
1712 if (in_dev) {
1713 int our = ip_check_mc_rcu(in_dev, daddr, saddr,
1714 ip_hdr(skb)->protocol);
1715 if (our
1716 #ifdef CONFIG_IP_MROUTE
1717 ||
1718 (!ipv4_is_local_multicast(daddr) &&
1719 IN_DEV_MFORWARD(in_dev))
1720 #endif
1721 ) {
1722 int res = ip_route_input_mc(skb, daddr, saddr,
1723 tos, dev, our);
1724 rcu_read_unlock();
1725 return res;
1726 }
1727 }
1728 rcu_read_unlock();
1729 return -EINVAL;
1730 }
1731 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
1732 rcu_read_unlock();
1733 return res;
1734 }
1735 EXPORT_SYMBOL(ip_route_input);
1736
1737 /* called with rcu_read_lock() */
1738 static struct rtable *__mkroute_output(const struct fib_result *res,
1739 const struct flowi4 *fl4, int orig_oif,
1740 struct net_device *dev_out,
1741 unsigned int flags)
1742 {
1743 struct fib_info *fi = res->fi;
1744 struct fib_nh_exception *fnhe;
1745 struct in_device *in_dev;
1746 u16 type = res->type;
1747 struct rtable *rth;
1748
1749 in_dev = __in_dev_get_rcu(dev_out);
1750 if (!in_dev)
1751 return ERR_PTR(-EINVAL);
1752
1753 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
1754 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK))
1755 return ERR_PTR(-EINVAL);
1756
1757 if (ipv4_is_lbcast(fl4->daddr))
1758 type = RTN_BROADCAST;
1759 else if (ipv4_is_multicast(fl4->daddr))
1760 type = RTN_MULTICAST;
1761 else if (ipv4_is_zeronet(fl4->daddr))
1762 return ERR_PTR(-EINVAL);
1763
1764 if (dev_out->flags & IFF_LOOPBACK)
1765 flags |= RTCF_LOCAL;
1766
1767 if (type == RTN_BROADCAST) {
1768 flags |= RTCF_BROADCAST | RTCF_LOCAL;
1769 fi = NULL;
1770 } else if (type == RTN_MULTICAST) {
1771 flags |= RTCF_MULTICAST | RTCF_LOCAL;
1772 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
1773 fl4->flowi4_proto))
1774 flags &= ~RTCF_LOCAL;
1775 /* If multicast route do not exist use
1776 * default one, but do not gateway in this case.
1777 * Yes, it is hack.
1778 */
1779 if (fi && res->prefixlen < 4)
1780 fi = NULL;
1781 }
1782
1783 fnhe = NULL;
1784 if (fi) {
1785 fnhe = find_exception(&FIB_RES_NH(*res), fl4->daddr);
1786 if (!fnhe) {
1787 rth = FIB_RES_NH(*res).nh_rth_output;
1788 if (rt_cache_valid(rth)) {
1789 dst_hold(&rth->dst);
1790 return rth;
1791 }
1792 }
1793 }
1794 rth = rt_dst_alloc(dev_out,
1795 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1796 IN_DEV_CONF_GET(in_dev, NOXFRM),
1797 fi && !fnhe);
1798 if (!rth)
1799 return ERR_PTR(-ENOBUFS);
1800
1801 rth->dst.output = ip_output;
1802
1803 rth->rt_genid = rt_genid(dev_net(dev_out));
1804 rth->rt_flags = flags;
1805 rth->rt_type = type;
1806 rth->rt_is_input = 0;
1807 rth->rt_iif = orig_oif ? : dev_out->ifindex;
1808 rth->rt_pmtu = 0;
1809 rth->rt_gateway = 0;
1810 rth->fi = NULL;
1811
1812 RT_CACHE_STAT_INC(out_slow_tot);
1813
1814 if (flags & RTCF_LOCAL)
1815 rth->dst.input = ip_local_deliver;
1816 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1817 if (flags & RTCF_LOCAL &&
1818 !(dev_out->flags & IFF_LOOPBACK)) {
1819 rth->dst.output = ip_mc_output;
1820 RT_CACHE_STAT_INC(out_slow_mc);
1821 }
1822 #ifdef CONFIG_IP_MROUTE
1823 if (type == RTN_MULTICAST) {
1824 if (IN_DEV_MFORWARD(in_dev) &&
1825 !ipv4_is_local_multicast(fl4->daddr)) {
1826 rth->dst.input = ip_mr_input;
1827 rth->dst.output = ip_mc_output;
1828 }
1829 }
1830 #endif
1831 }
1832
1833 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
1834
1835 return rth;
1836 }
1837
1838 /*
1839 * Major route resolver routine.
1840 */
1841
1842 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4)
1843 {
1844 struct net_device *dev_out = NULL;
1845 __u8 tos = RT_FL_TOS(fl4);
1846 unsigned int flags = 0;
1847 struct fib_result res;
1848 struct rtable *rth;
1849 int orig_oif;
1850
1851 res.tclassid = 0;
1852 res.fi = NULL;
1853 res.table = NULL;
1854
1855 orig_oif = fl4->flowi4_oif;
1856
1857 fl4->flowi4_iif = net->loopback_dev->ifindex;
1858 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
1859 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
1860 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
1861
1862 rcu_read_lock();
1863 if (fl4->saddr) {
1864 rth = ERR_PTR(-EINVAL);
1865 if (ipv4_is_multicast(fl4->saddr) ||
1866 ipv4_is_lbcast(fl4->saddr) ||
1867 ipv4_is_zeronet(fl4->saddr))
1868 goto out;
1869
1870 /* I removed check for oif == dev_out->oif here.
1871 It was wrong for two reasons:
1872 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
1873 is assigned to multiple interfaces.
1874 2. Moreover, we are allowed to send packets with saddr
1875 of another iface. --ANK
1876 */
1877
1878 if (fl4->flowi4_oif == 0 &&
1879 (ipv4_is_multicast(fl4->daddr) ||
1880 ipv4_is_lbcast(fl4->daddr))) {
1881 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1882 dev_out = __ip_dev_find(net, fl4->saddr, false);
1883 if (dev_out == NULL)
1884 goto out;
1885
1886 /* Special hack: user can direct multicasts
1887 and limited broadcast via necessary interface
1888 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
1889 This hack is not just for fun, it allows
1890 vic,vat and friends to work.
1891 They bind socket to loopback, set ttl to zero
1892 and expect that it will work.
1893 From the viewpoint of routing cache they are broken,
1894 because we are not allowed to build multicast path
1895 with loopback source addr (look, routing cache
1896 cannot know, that ttl is zero, so that packet
1897 will not leave this host and route is valid).
1898 Luckily, this hack is good workaround.
1899 */
1900
1901 fl4->flowi4_oif = dev_out->ifindex;
1902 goto make_route;
1903 }
1904
1905 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
1906 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
1907 if (!__ip_dev_find(net, fl4->saddr, false))
1908 goto out;
1909 }
1910 }
1911
1912
1913 if (fl4->flowi4_oif) {
1914 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
1915 rth = ERR_PTR(-ENODEV);
1916 if (dev_out == NULL)
1917 goto out;
1918
1919 /* RACE: Check return value of inet_select_addr instead. */
1920 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
1921 rth = ERR_PTR(-ENETUNREACH);
1922 goto out;
1923 }
1924 if (ipv4_is_local_multicast(fl4->daddr) ||
1925 ipv4_is_lbcast(fl4->daddr)) {
1926 if (!fl4->saddr)
1927 fl4->saddr = inet_select_addr(dev_out, 0,
1928 RT_SCOPE_LINK);
1929 goto make_route;
1930 }
1931 if (fl4->saddr) {
1932 if (ipv4_is_multicast(fl4->daddr))
1933 fl4->saddr = inet_select_addr(dev_out, 0,
1934 fl4->flowi4_scope);
1935 else if (!fl4->daddr)
1936 fl4->saddr = inet_select_addr(dev_out, 0,
1937 RT_SCOPE_HOST);
1938 }
1939 }
1940
1941 if (!fl4->daddr) {
1942 fl4->daddr = fl4->saddr;
1943 if (!fl4->daddr)
1944 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
1945 dev_out = net->loopback_dev;
1946 fl4->flowi4_oif = net->loopback_dev->ifindex;
1947 res.type = RTN_LOCAL;
1948 flags |= RTCF_LOCAL;
1949 goto make_route;
1950 }
1951
1952 if (fib_lookup(net, fl4, &res)) {
1953 res.fi = NULL;
1954 res.table = NULL;
1955 if (fl4->flowi4_oif) {
1956 /* Apparently, routing tables are wrong. Assume,
1957 that the destination is on link.
1958
1959 WHY? DW.
1960 Because we are allowed to send to iface
1961 even if it has NO routes and NO assigned
1962 addresses. When oif is specified, routing
1963 tables are looked up with only one purpose:
1964 to catch if destination is gatewayed, rather than
1965 direct. Moreover, if MSG_DONTROUTE is set,
1966 we send packet, ignoring both routing tables
1967 and ifaddr state. --ANK
1968
1969
1970 We could make it even if oif is unknown,
1971 likely IPv6, but we do not.
1972 */
1973
1974 if (fl4->saddr == 0)
1975 fl4->saddr = inet_select_addr(dev_out, 0,
1976 RT_SCOPE_LINK);
1977 res.type = RTN_UNICAST;
1978 goto make_route;
1979 }
1980 rth = ERR_PTR(-ENETUNREACH);
1981 goto out;
1982 }
1983
1984 if (res.type == RTN_LOCAL) {
1985 if (!fl4->saddr) {
1986 if (res.fi->fib_prefsrc)
1987 fl4->saddr = res.fi->fib_prefsrc;
1988 else
1989 fl4->saddr = fl4->daddr;
1990 }
1991 dev_out = net->loopback_dev;
1992 fl4->flowi4_oif = dev_out->ifindex;
1993 res.fi = NULL;
1994 flags |= RTCF_LOCAL;
1995 goto make_route;
1996 }
1997
1998 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1999 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
2000 fib_select_multipath(&res);
2001 else
2002 #endif
2003 if (!res.prefixlen &&
2004 res.table->tb_num_default > 1 &&
2005 res.type == RTN_UNICAST && !fl4->flowi4_oif)
2006 fib_select_default(&res);
2007
2008 if (!fl4->saddr)
2009 fl4->saddr = FIB_RES_PREFSRC(net, res);
2010
2011 dev_out = FIB_RES_DEV(res);
2012 fl4->flowi4_oif = dev_out->ifindex;
2013
2014
2015 make_route:
2016 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2017
2018 out:
2019 rcu_read_unlock();
2020 return rth;
2021 }
2022 EXPORT_SYMBOL_GPL(__ip_route_output_key);
2023
2024 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2025 {
2026 return NULL;
2027 }
2028
2029 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2030 {
2031 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2032
2033 return mtu ? : dst->dev->mtu;
2034 }
2035
2036 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2037 struct sk_buff *skb, u32 mtu)
2038 {
2039 }
2040
2041 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2042 struct sk_buff *skb)
2043 {
2044 }
2045
2046 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2047 unsigned long old)
2048 {
2049 return NULL;
2050 }
2051
2052 static struct dst_ops ipv4_dst_blackhole_ops = {
2053 .family = AF_INET,
2054 .protocol = cpu_to_be16(ETH_P_IP),
2055 .destroy = ipv4_dst_destroy,
2056 .check = ipv4_blackhole_dst_check,
2057 .mtu = ipv4_blackhole_mtu,
2058 .default_advmss = ipv4_default_advmss,
2059 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2060 .redirect = ipv4_rt_blackhole_redirect,
2061 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2062 .neigh_lookup = ipv4_neigh_lookup,
2063 };
2064
2065 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2066 {
2067 struct rtable *ort = (struct rtable *) dst_orig;
2068 struct rtable *rt;
2069
2070 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2071 if (rt) {
2072 struct dst_entry *new = &rt->dst;
2073
2074 new->__use = 1;
2075 new->input = dst_discard;
2076 new->output = dst_discard;
2077
2078 new->dev = ort->dst.dev;
2079 if (new->dev)
2080 dev_hold(new->dev);
2081
2082 rt->rt_is_input = ort->rt_is_input;
2083 rt->rt_iif = ort->rt_iif;
2084 rt->rt_pmtu = ort->rt_pmtu;
2085
2086 rt->rt_genid = rt_genid(net);
2087 rt->rt_flags = ort->rt_flags;
2088 rt->rt_type = ort->rt_type;
2089 rt->rt_gateway = ort->rt_gateway;
2090 rt->fi = ort->fi;
2091 if (rt->fi)
2092 atomic_inc(&rt->fi->fib_clntref);
2093
2094 dst_free(new);
2095 }
2096
2097 dst_release(dst_orig);
2098
2099 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2100 }
2101
2102 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2103 struct sock *sk)
2104 {
2105 struct rtable *rt = __ip_route_output_key(net, flp4);
2106
2107 if (IS_ERR(rt))
2108 return rt;
2109
2110 if (flp4->flowi4_proto)
2111 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
2112 flowi4_to_flowi(flp4),
2113 sk, 0);
2114
2115 return rt;
2116 }
2117 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2118
2119 static int rt_fill_info(struct net *net, __be32 dst, __be32 src,
2120 struct flowi4 *fl4, struct sk_buff *skb, u32 pid,
2121 u32 seq, int event, int nowait, unsigned int flags)
2122 {
2123 struct rtable *rt = skb_rtable(skb);
2124 struct rtmsg *r;
2125 struct nlmsghdr *nlh;
2126 unsigned long expires = 0;
2127 u32 error;
2128 u32 metrics[RTAX_MAX];
2129
2130 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags);
2131 if (nlh == NULL)
2132 return -EMSGSIZE;
2133
2134 r = nlmsg_data(nlh);
2135 r->rtm_family = AF_INET;
2136 r->rtm_dst_len = 32;
2137 r->rtm_src_len = 0;
2138 r->rtm_tos = fl4->flowi4_tos;
2139 r->rtm_table = RT_TABLE_MAIN;
2140 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN))
2141 goto nla_put_failure;
2142 r->rtm_type = rt->rt_type;
2143 r->rtm_scope = RT_SCOPE_UNIVERSE;
2144 r->rtm_protocol = RTPROT_UNSPEC;
2145 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2146 if (rt->rt_flags & RTCF_NOTIFY)
2147 r->rtm_flags |= RTM_F_NOTIFY;
2148
2149 if (nla_put_be32(skb, RTA_DST, dst))
2150 goto nla_put_failure;
2151 if (src) {
2152 r->rtm_src_len = 32;
2153 if (nla_put_be32(skb, RTA_SRC, src))
2154 goto nla_put_failure;
2155 }
2156 if (rt->dst.dev &&
2157 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2158 goto nla_put_failure;
2159 #ifdef CONFIG_IP_ROUTE_CLASSID
2160 if (rt->dst.tclassid &&
2161 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2162 goto nla_put_failure;
2163 #endif
2164 if (!rt_is_input_route(rt) &&
2165 fl4->saddr != src) {
2166 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr))
2167 goto nla_put_failure;
2168 }
2169 if (rt->rt_gateway &&
2170 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway))
2171 goto nla_put_failure;
2172
2173 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2174 if (rt->rt_pmtu)
2175 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2176 if (rtnetlink_put_metrics(skb, metrics) < 0)
2177 goto nla_put_failure;
2178
2179 if (fl4->flowi4_mark &&
2180 nla_put_be32(skb, RTA_MARK, fl4->flowi4_mark))
2181 goto nla_put_failure;
2182
2183 error = rt->dst.error;
2184 expires = rt->dst.expires;
2185 if (expires) {
2186 if (time_before(jiffies, expires))
2187 expires -= jiffies;
2188 else
2189 expires = 0;
2190 }
2191
2192 if (rt_is_input_route(rt)) {
2193 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif))
2194 goto nla_put_failure;
2195 }
2196
2197 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2198 goto nla_put_failure;
2199
2200 return nlmsg_end(skb, nlh);
2201
2202 nla_put_failure:
2203 nlmsg_cancel(skb, nlh);
2204 return -EMSGSIZE;
2205 }
2206
2207 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
2208 {
2209 struct net *net = sock_net(in_skb->sk);
2210 struct rtmsg *rtm;
2211 struct nlattr *tb[RTA_MAX+1];
2212 struct rtable *rt = NULL;
2213 struct flowi4 fl4;
2214 __be32 dst = 0;
2215 __be32 src = 0;
2216 u32 iif;
2217 int err;
2218 int mark;
2219 struct sk_buff *skb;
2220
2221 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2222 if (err < 0)
2223 goto errout;
2224
2225 rtm = nlmsg_data(nlh);
2226
2227 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2228 if (skb == NULL) {
2229 err = -ENOBUFS;
2230 goto errout;
2231 }
2232
2233 /* Reserve room for dummy headers, this skb can pass
2234 through good chunk of routing engine.
2235 */
2236 skb_reset_mac_header(skb);
2237 skb_reset_network_header(skb);
2238
2239 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2240 ip_hdr(skb)->protocol = IPPROTO_ICMP;
2241 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2242
2243 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0;
2244 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0;
2245 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2246 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2247
2248 memset(&fl4, 0, sizeof(fl4));
2249 fl4.daddr = dst;
2250 fl4.saddr = src;
2251 fl4.flowi4_tos = rtm->rtm_tos;
2252 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2253 fl4.flowi4_mark = mark;
2254
2255 if (iif) {
2256 struct net_device *dev;
2257
2258 dev = __dev_get_by_index(net, iif);
2259 if (dev == NULL) {
2260 err = -ENODEV;
2261 goto errout_free;
2262 }
2263
2264 skb->protocol = htons(ETH_P_IP);
2265 skb->dev = dev;
2266 skb->mark = mark;
2267 local_bh_disable();
2268 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2269 local_bh_enable();
2270
2271 rt = skb_rtable(skb);
2272 if (err == 0 && rt->dst.error)
2273 err = -rt->dst.error;
2274 } else {
2275 rt = ip_route_output_key(net, &fl4);
2276
2277 err = 0;
2278 if (IS_ERR(rt))
2279 err = PTR_ERR(rt);
2280 }
2281
2282 if (err)
2283 goto errout_free;
2284
2285 skb_dst_set(skb, &rt->dst);
2286 if (rtm->rtm_flags & RTM_F_NOTIFY)
2287 rt->rt_flags |= RTCF_NOTIFY;
2288
2289 err = rt_fill_info(net, dst, src, &fl4, skb,
2290 NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
2291 RTM_NEWROUTE, 0, 0);
2292 if (err <= 0)
2293 goto errout_free;
2294
2295 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid);
2296 errout:
2297 return err;
2298
2299 errout_free:
2300 kfree_skb(skb);
2301 goto errout;
2302 }
2303
2304 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb)
2305 {
2306 return skb->len;
2307 }
2308
2309 void ip_rt_multicast_event(struct in_device *in_dev)
2310 {
2311 rt_cache_flush(dev_net(in_dev->dev), 0);
2312 }
2313
2314 #ifdef CONFIG_SYSCTL
2315 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
2316 void __user *buffer,
2317 size_t *lenp, loff_t *ppos)
2318 {
2319 if (write) {
2320 int flush_delay;
2321 ctl_table ctl;
2322 struct net *net;
2323
2324 memcpy(&ctl, __ctl, sizeof(ctl));
2325 ctl.data = &flush_delay;
2326 proc_dointvec(&ctl, write, buffer, lenp, ppos);
2327
2328 net = (struct net *)__ctl->extra1;
2329 rt_cache_flush(net, flush_delay);
2330 return 0;
2331 }
2332
2333 return -EINVAL;
2334 }
2335
2336 static ctl_table ipv4_route_table[] = {
2337 {
2338 .procname = "gc_thresh",
2339 .data = &ipv4_dst_ops.gc_thresh,
2340 .maxlen = sizeof(int),
2341 .mode = 0644,
2342 .proc_handler = proc_dointvec,
2343 },
2344 {
2345 .procname = "max_size",
2346 .data = &ip_rt_max_size,
2347 .maxlen = sizeof(int),
2348 .mode = 0644,
2349 .proc_handler = proc_dointvec,
2350 },
2351 {
2352 /* Deprecated. Use gc_min_interval_ms */
2353
2354 .procname = "gc_min_interval",
2355 .data = &ip_rt_gc_min_interval,
2356 .maxlen = sizeof(int),
2357 .mode = 0644,
2358 .proc_handler = proc_dointvec_jiffies,
2359 },
2360 {
2361 .procname = "gc_min_interval_ms",
2362 .data = &ip_rt_gc_min_interval,
2363 .maxlen = sizeof(int),
2364 .mode = 0644,
2365 .proc_handler = proc_dointvec_ms_jiffies,
2366 },
2367 {
2368 .procname = "gc_timeout",
2369 .data = &ip_rt_gc_timeout,
2370 .maxlen = sizeof(int),
2371 .mode = 0644,
2372 .proc_handler = proc_dointvec_jiffies,
2373 },
2374 {
2375 .procname = "gc_interval",
2376 .data = &ip_rt_gc_interval,
2377 .maxlen = sizeof(int),
2378 .mode = 0644,
2379 .proc_handler = proc_dointvec_jiffies,
2380 },
2381 {
2382 .procname = "redirect_load",
2383 .data = &ip_rt_redirect_load,
2384 .maxlen = sizeof(int),
2385 .mode = 0644,
2386 .proc_handler = proc_dointvec,
2387 },
2388 {
2389 .procname = "redirect_number",
2390 .data = &ip_rt_redirect_number,
2391 .maxlen = sizeof(int),
2392 .mode = 0644,
2393 .proc_handler = proc_dointvec,
2394 },
2395 {
2396 .procname = "redirect_silence",
2397 .data = &ip_rt_redirect_silence,
2398 .maxlen = sizeof(int),
2399 .mode = 0644,
2400 .proc_handler = proc_dointvec,
2401 },
2402 {
2403 .procname = "error_cost",
2404 .data = &ip_rt_error_cost,
2405 .maxlen = sizeof(int),
2406 .mode = 0644,
2407 .proc_handler = proc_dointvec,
2408 },
2409 {
2410 .procname = "error_burst",
2411 .data = &ip_rt_error_burst,
2412 .maxlen = sizeof(int),
2413 .mode = 0644,
2414 .proc_handler = proc_dointvec,
2415 },
2416 {
2417 .procname = "gc_elasticity",
2418 .data = &ip_rt_gc_elasticity,
2419 .maxlen = sizeof(int),
2420 .mode = 0644,
2421 .proc_handler = proc_dointvec,
2422 },
2423 {
2424 .procname = "mtu_expires",
2425 .data = &ip_rt_mtu_expires,
2426 .maxlen = sizeof(int),
2427 .mode = 0644,
2428 .proc_handler = proc_dointvec_jiffies,
2429 },
2430 {
2431 .procname = "min_pmtu",
2432 .data = &ip_rt_min_pmtu,
2433 .maxlen = sizeof(int),
2434 .mode = 0644,
2435 .proc_handler = proc_dointvec,
2436 },
2437 {
2438 .procname = "min_adv_mss",
2439 .data = &ip_rt_min_advmss,
2440 .maxlen = sizeof(int),
2441 .mode = 0644,
2442 .proc_handler = proc_dointvec,
2443 },
2444 { }
2445 };
2446
2447 static struct ctl_table ipv4_route_flush_table[] = {
2448 {
2449 .procname = "flush",
2450 .maxlen = sizeof(int),
2451 .mode = 0200,
2452 .proc_handler = ipv4_sysctl_rtcache_flush,
2453 },
2454 { },
2455 };
2456
2457 static __net_init int sysctl_route_net_init(struct net *net)
2458 {
2459 struct ctl_table *tbl;
2460
2461 tbl = ipv4_route_flush_table;
2462 if (!net_eq(net, &init_net)) {
2463 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2464 if (tbl == NULL)
2465 goto err_dup;
2466 }
2467 tbl[0].extra1 = net;
2468
2469 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2470 if (net->ipv4.route_hdr == NULL)
2471 goto err_reg;
2472 return 0;
2473
2474 err_reg:
2475 if (tbl != ipv4_route_flush_table)
2476 kfree(tbl);
2477 err_dup:
2478 return -ENOMEM;
2479 }
2480
2481 static __net_exit void sysctl_route_net_exit(struct net *net)
2482 {
2483 struct ctl_table *tbl;
2484
2485 tbl = net->ipv4.route_hdr->ctl_table_arg;
2486 unregister_net_sysctl_table(net->ipv4.route_hdr);
2487 BUG_ON(tbl == ipv4_route_flush_table);
2488 kfree(tbl);
2489 }
2490
2491 static __net_initdata struct pernet_operations sysctl_route_ops = {
2492 .init = sysctl_route_net_init,
2493 .exit = sysctl_route_net_exit,
2494 };
2495 #endif
2496
2497 static __net_init int rt_genid_init(struct net *net)
2498 {
2499 get_random_bytes(&net->ipv4.rt_genid,
2500 sizeof(net->ipv4.rt_genid));
2501 get_random_bytes(&net->ipv4.dev_addr_genid,
2502 sizeof(net->ipv4.dev_addr_genid));
2503 return 0;
2504 }
2505
2506 static __net_initdata struct pernet_operations rt_genid_ops = {
2507 .init = rt_genid_init,
2508 };
2509
2510 static int __net_init ipv4_inetpeer_init(struct net *net)
2511 {
2512 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2513
2514 if (!bp)
2515 return -ENOMEM;
2516 inet_peer_base_init(bp);
2517 net->ipv4.peers = bp;
2518 return 0;
2519 }
2520
2521 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2522 {
2523 struct inet_peer_base *bp = net->ipv4.peers;
2524
2525 net->ipv4.peers = NULL;
2526 inetpeer_invalidate_tree(bp);
2527 kfree(bp);
2528 }
2529
2530 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2531 .init = ipv4_inetpeer_init,
2532 .exit = ipv4_inetpeer_exit,
2533 };
2534
2535 #ifdef CONFIG_IP_ROUTE_CLASSID
2536 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2537 #endif /* CONFIG_IP_ROUTE_CLASSID */
2538
2539 int __init ip_rt_init(void)
2540 {
2541 int rc = 0;
2542
2543 #ifdef CONFIG_IP_ROUTE_CLASSID
2544 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2545 if (!ip_rt_acct)
2546 panic("IP: failed to allocate ip_rt_acct\n");
2547 #endif
2548
2549 ipv4_dst_ops.kmem_cachep =
2550 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2551 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2552
2553 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2554
2555 if (dst_entries_init(&ipv4_dst_ops) < 0)
2556 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2557
2558 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2559 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2560
2561 ipv4_dst_ops.gc_thresh = ~0;
2562 ip_rt_max_size = INT_MAX;
2563
2564 devinet_init();
2565 ip_fib_init();
2566
2567 if (ip_rt_proc_init())
2568 pr_err("Unable to create route proc files\n");
2569 #ifdef CONFIG_XFRM
2570 xfrm_init();
2571 xfrm4_init(ip_rt_max_size);
2572 #endif
2573 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2574
2575 #ifdef CONFIG_SYSCTL
2576 register_pernet_subsys(&sysctl_route_ops);
2577 #endif
2578 register_pernet_subsys(&rt_genid_ops);
2579 register_pernet_subsys(&ipv4_inetpeer_ops);
2580 return rc;
2581 }
2582
2583 #ifdef CONFIG_SYSCTL
2584 /*
2585 * We really need to sanitize the damn ipv4 init order, then all
2586 * this nonsense will go away.
2587 */
2588 void __init ip_static_sysctl_init(void)
2589 {
2590 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
2591 }
2592 #endif
Linux kernel - Deliverables
This page took 0.09157 seconds and 5 git commands to generate.