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 <linux/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/string.h>
74 #include <linux/socket.h>
75 #include <linux/sockios.h>
76 #include <linux/errno.h>
77 #include <linux/in.h>
78 #include <linux/inet.h>
79 #include <linux/netdevice.h>
80 #include <linux/proc_fs.h>
81 #include <linux/init.h>
82 #include <linux/skbuff.h>
83 #include <linux/inetdevice.h>
84 #include <linux/igmp.h>
85 #include <linux/pkt_sched.h>
86 #include <linux/mroute.h>
87 #include <linux/netfilter_ipv4.h>
88 #include <linux/random.h>
89 #include <linux/rcupdate.h>
90 #include <linux/times.h>
91 #include <linux/slab.h>
92 #include <linux/jhash.h>
93 #include <net/dst.h>
94 #include <net/dst_metadata.h>
95 #include <net/net_namespace.h>
96 #include <net/protocol.h>
97 #include <net/ip.h>
98 #include <net/route.h>
99 #include <net/inetpeer.h>
100 #include <net/sock.h>
101 #include <net/ip_fib.h>
102 #include <net/arp.h>
103 #include <net/tcp.h>
104 #include <net/icmp.h>
105 #include <net/xfrm.h>
106 #include <net/lwtunnel.h>
107 #include <net/netevent.h>
108 #include <net/rtnetlink.h>
109 #ifdef CONFIG_SYSCTL
110 #include <linux/sysctl.h>
111 #include <linux/kmemleak.h>
112 #endif
113 #include <net/secure_seq.h>
114 #include <net/ip_tunnels.h>
115 #include <net/l3mdev.h>
116
117 #define RT_FL_TOS(oldflp4) \
118 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
119
120 #define RT_GC_TIMEOUT (300*HZ)
121
122 static int ip_rt_max_size;
123 static int ip_rt_redirect_number __read_mostly = 9;
124 static int ip_rt_redirect_load __read_mostly = HZ / 50;
125 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
126 static int ip_rt_error_cost __read_mostly = HZ;
127 static int ip_rt_error_burst __read_mostly = 5 * HZ;
128 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
129 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
130 static int ip_rt_min_advmss __read_mostly = 256;
131
132 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
133 /*
134 * Interface to generic destination cache.
135 */
136
137 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
138 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
139 static unsigned int ipv4_mtu(const struct dst_entry *dst);
140 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst);
141 static void ipv4_link_failure(struct sk_buff *skb);
142 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
143 struct sk_buff *skb, u32 mtu);
144 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk,
145 struct sk_buff *skb);
146 static void ipv4_dst_destroy(struct dst_entry *dst);
147
148 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
149 {
150 WARN_ON(1);
151 return NULL;
152 }
153
154 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
155 struct sk_buff *skb,
156 const void *daddr);
157 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
158
159 static struct dst_ops ipv4_dst_ops = {
160 .family = AF_INET,
161 .check = ipv4_dst_check,
162 .default_advmss = ipv4_default_advmss,
163 .mtu = ipv4_mtu,
164 .cow_metrics = ipv4_cow_metrics,
165 .destroy = ipv4_dst_destroy,
166 .negative_advice = ipv4_negative_advice,
167 .link_failure = ipv4_link_failure,
168 .update_pmtu = ip_rt_update_pmtu,
169 .redirect = ip_do_redirect,
170 .local_out = __ip_local_out,
171 .neigh_lookup = ipv4_neigh_lookup,
172 .confirm_neigh = ipv4_confirm_neigh,
173 };
174
175 #define ECN_OR_COST(class) TC_PRIO_##class
176
177 const __u8 ip_tos2prio[16] = {
178 TC_PRIO_BESTEFFORT,
179 ECN_OR_COST(BESTEFFORT),
180 TC_PRIO_BESTEFFORT,
181 ECN_OR_COST(BESTEFFORT),
182 TC_PRIO_BULK,
183 ECN_OR_COST(BULK),
184 TC_PRIO_BULK,
185 ECN_OR_COST(BULK),
186 TC_PRIO_INTERACTIVE,
187 ECN_OR_COST(INTERACTIVE),
188 TC_PRIO_INTERACTIVE,
189 ECN_OR_COST(INTERACTIVE),
190 TC_PRIO_INTERACTIVE_BULK,
191 ECN_OR_COST(INTERACTIVE_BULK),
192 TC_PRIO_INTERACTIVE_BULK,
193 ECN_OR_COST(INTERACTIVE_BULK)
194 };
195 EXPORT_SYMBOL(ip_tos2prio);
196
197 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
198 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
199
200 #ifdef CONFIG_PROC_FS
201 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
202 {
203 if (*pos)
204 return NULL;
205 return SEQ_START_TOKEN;
206 }
207
208 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
209 {
210 ++*pos;
211 return NULL;
212 }
213
214 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
215 {
216 }
217
218 static int rt_cache_seq_show(struct seq_file *seq, void *v)
219 {
220 if (v == SEQ_START_TOKEN)
221 seq_printf(seq, "%-127s\n",
222 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
223 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
224 "HHUptod\tSpecDst");
225 return 0;
226 }
227
228 static const struct seq_operations rt_cache_seq_ops = {
229 .start = rt_cache_seq_start,
230 .next = rt_cache_seq_next,
231 .stop = rt_cache_seq_stop,
232 .show = rt_cache_seq_show,
233 };
234
235 static int rt_cache_seq_open(struct inode *inode, struct file *file)
236 {
237 return seq_open(file, &rt_cache_seq_ops);
238 }
239
240 static const struct file_operations rt_cache_seq_fops = {
241 .owner = THIS_MODULE,
242 .open = rt_cache_seq_open,
243 .read = seq_read,
244 .llseek = seq_lseek,
245 .release = seq_release,
246 };
247
248
249 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
250 {
251 int cpu;
252
253 if (*pos == 0)
254 return SEQ_START_TOKEN;
255
256 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
257 if (!cpu_possible(cpu))
258 continue;
259 *pos = cpu+1;
260 return &per_cpu(rt_cache_stat, cpu);
261 }
262 return NULL;
263 }
264
265 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
266 {
267 int cpu;
268
269 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
270 if (!cpu_possible(cpu))
271 continue;
272 *pos = cpu+1;
273 return &per_cpu(rt_cache_stat, cpu);
274 }
275 return NULL;
276
277 }
278
279 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
280 {
281
282 }
283
284 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
285 {
286 struct rt_cache_stat *st = v;
287
288 if (v == SEQ_START_TOKEN) {
289 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");
290 return 0;
291 }
292
293 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
294 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
295 dst_entries_get_slow(&ipv4_dst_ops),
296 0, /* st->in_hit */
297 st->in_slow_tot,
298 st->in_slow_mc,
299 st->in_no_route,
300 st->in_brd,
301 st->in_martian_dst,
302 st->in_martian_src,
303
304 0, /* st->out_hit */
305 st->out_slow_tot,
306 st->out_slow_mc,
307
308 0, /* st->gc_total */
309 0, /* st->gc_ignored */
310 0, /* st->gc_goal_miss */
311 0, /* st->gc_dst_overflow */
312 0, /* st->in_hlist_search */
313 0 /* st->out_hlist_search */
314 );
315 return 0;
316 }
317
318 static const struct seq_operations rt_cpu_seq_ops = {
319 .start = rt_cpu_seq_start,
320 .next = rt_cpu_seq_next,
321 .stop = rt_cpu_seq_stop,
322 .show = rt_cpu_seq_show,
323 };
324
325
326 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
327 {
328 return seq_open(file, &rt_cpu_seq_ops);
329 }
330
331 static const struct file_operations rt_cpu_seq_fops = {
332 .owner = THIS_MODULE,
333 .open = rt_cpu_seq_open,
334 .read = seq_read,
335 .llseek = seq_lseek,
336 .release = seq_release,
337 };
338
339 #ifdef CONFIG_IP_ROUTE_CLASSID
340 static int rt_acct_proc_show(struct seq_file *m, void *v)
341 {
342 struct ip_rt_acct *dst, *src;
343 unsigned int i, j;
344
345 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
346 if (!dst)
347 return -ENOMEM;
348
349 for_each_possible_cpu(i) {
350 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
351 for (j = 0; j < 256; j++) {
352 dst[j].o_bytes += src[j].o_bytes;
353 dst[j].o_packets += src[j].o_packets;
354 dst[j].i_bytes += src[j].i_bytes;
355 dst[j].i_packets += src[j].i_packets;
356 }
357 }
358
359 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
360 kfree(dst);
361 return 0;
362 }
363
364 static int rt_acct_proc_open(struct inode *inode, struct file *file)
365 {
366 return single_open(file, rt_acct_proc_show, NULL);
367 }
368
369 static const struct file_operations rt_acct_proc_fops = {
370 .owner = THIS_MODULE,
371 .open = rt_acct_proc_open,
372 .read = seq_read,
373 .llseek = seq_lseek,
374 .release = single_release,
375 };
376 #endif
377
378 static int __net_init ip_rt_do_proc_init(struct net *net)
379 {
380 struct proc_dir_entry *pde;
381
382 pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
383 &rt_cache_seq_fops);
384 if (!pde)
385 goto err1;
386
387 pde = proc_create("rt_cache", S_IRUGO,
388 net->proc_net_stat, &rt_cpu_seq_fops);
389 if (!pde)
390 goto err2;
391
392 #ifdef CONFIG_IP_ROUTE_CLASSID
393 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
394 if (!pde)
395 goto err3;
396 #endif
397 return 0;
398
399 #ifdef CONFIG_IP_ROUTE_CLASSID
400 err3:
401 remove_proc_entry("rt_cache", net->proc_net_stat);
402 #endif
403 err2:
404 remove_proc_entry("rt_cache", net->proc_net);
405 err1:
406 return -ENOMEM;
407 }
408
409 static void __net_exit ip_rt_do_proc_exit(struct net *net)
410 {
411 remove_proc_entry("rt_cache", net->proc_net_stat);
412 remove_proc_entry("rt_cache", net->proc_net);
413 #ifdef CONFIG_IP_ROUTE_CLASSID
414 remove_proc_entry("rt_acct", net->proc_net);
415 #endif
416 }
417
418 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
419 .init = ip_rt_do_proc_init,
420 .exit = ip_rt_do_proc_exit,
421 };
422
423 static int __init ip_rt_proc_init(void)
424 {
425 return register_pernet_subsys(&ip_rt_proc_ops);
426 }
427
428 #else
429 static inline int ip_rt_proc_init(void)
430 {
431 return 0;
432 }
433 #endif /* CONFIG_PROC_FS */
434
435 static inline bool rt_is_expired(const struct rtable *rth)
436 {
437 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
438 }
439
440 void rt_cache_flush(struct net *net)
441 {
442 rt_genid_bump_ipv4(net);
443 }
444
445 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
446 struct sk_buff *skb,
447 const void *daddr)
448 {
449 struct net_device *dev = dst->dev;
450 const __be32 *pkey = daddr;
451 const struct rtable *rt;
452 struct neighbour *n;
453
454 rt = (const struct rtable *) dst;
455 if (rt->rt_gateway)
456 pkey = (const __be32 *) &rt->rt_gateway;
457 else if (skb)
458 pkey = &ip_hdr(skb)->daddr;
459
460 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
461 if (n)
462 return n;
463 return neigh_create(&arp_tbl, pkey, dev);
464 }
465
466 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
467 {
468 struct net_device *dev = dst->dev;
469 const __be32 *pkey = daddr;
470 const struct rtable *rt;
471
472 rt = (const struct rtable *)dst;
473 if (rt->rt_gateway)
474 pkey = (const __be32 *)&rt->rt_gateway;
475 else if (!daddr ||
476 (rt->rt_flags &
477 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL)))
478 return;
479
480 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
481 }
482
483 #define IP_IDENTS_SZ 2048u
484
485 static atomic_t *ip_idents __read_mostly;
486 static u32 *ip_tstamps __read_mostly;
487
488 /* In order to protect privacy, we add a perturbation to identifiers
489 * if one generator is seldom used. This makes hard for an attacker
490 * to infer how many packets were sent between two points in time.
491 */
492 u32 ip_idents_reserve(u32 hash, int segs)
493 {
494 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
495 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
496 u32 old = ACCESS_ONCE(*p_tstamp);
497 u32 now = (u32)jiffies;
498 u32 new, delta = 0;
499
500 if (old != now && cmpxchg(p_tstamp, old, now) == old)
501 delta = prandom_u32_max(now - old);
502
503 /* Do not use atomic_add_return() as it makes UBSAN unhappy */
504 do {
505 old = (u32)atomic_read(p_id);
506 new = old + delta + segs;
507 } while (atomic_cmpxchg(p_id, old, new) != old);
508
509 return new - segs;
510 }
511 EXPORT_SYMBOL(ip_idents_reserve);
512
513 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
514 {
515 static u32 ip_idents_hashrnd __read_mostly;
516 u32 hash, id;
517
518 net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
519
520 hash = jhash_3words((__force u32)iph->daddr,
521 (__force u32)iph->saddr,
522 iph->protocol ^ net_hash_mix(net),
523 ip_idents_hashrnd);
524 id = ip_idents_reserve(hash, segs);
525 iph->id = htons(id);
526 }
527 EXPORT_SYMBOL(__ip_select_ident);
528
529 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
530 const struct sock *sk,
531 const struct iphdr *iph,
532 int oif, u8 tos,
533 u8 prot, u32 mark, int flow_flags)
534 {
535 if (sk) {
536 const struct inet_sock *inet = inet_sk(sk);
537
538 oif = sk->sk_bound_dev_if;
539 mark = sk->sk_mark;
540 tos = RT_CONN_FLAGS(sk);
541 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
542 }
543 flowi4_init_output(fl4, oif, mark, tos,
544 RT_SCOPE_UNIVERSE, prot,
545 flow_flags,
546 iph->daddr, iph->saddr, 0, 0,
547 sock_net_uid(net, sk));
548 }
549
550 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
551 const struct sock *sk)
552 {
553 const struct net *net = dev_net(skb->dev);
554 const struct iphdr *iph = ip_hdr(skb);
555 int oif = skb->dev->ifindex;
556 u8 tos = RT_TOS(iph->tos);
557 u8 prot = iph->protocol;
558 u32 mark = skb->mark;
559
560 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
561 }
562
563 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
564 {
565 const struct inet_sock *inet = inet_sk(sk);
566 const struct ip_options_rcu *inet_opt;
567 __be32 daddr = inet->inet_daddr;
568
569 rcu_read_lock();
570 inet_opt = rcu_dereference(inet->inet_opt);
571 if (inet_opt && inet_opt->opt.srr)
572 daddr = inet_opt->opt.faddr;
573 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
574 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
575 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
576 inet_sk_flowi_flags(sk),
577 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
578 rcu_read_unlock();
579 }
580
581 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
582 const struct sk_buff *skb)
583 {
584 if (skb)
585 build_skb_flow_key(fl4, skb, sk);
586 else
587 build_sk_flow_key(fl4, sk);
588 }
589
590 static inline void rt_free(struct rtable *rt)
591 {
592 call_rcu(&rt->dst.rcu_head, dst_rcu_free);
593 }
594
595 static DEFINE_SPINLOCK(fnhe_lock);
596
597 static void fnhe_flush_routes(struct fib_nh_exception *fnhe)
598 {
599 struct rtable *rt;
600
601 rt = rcu_dereference(fnhe->fnhe_rth_input);
602 if (rt) {
603 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL);
604 rt_free(rt);
605 }
606 rt = rcu_dereference(fnhe->fnhe_rth_output);
607 if (rt) {
608 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
609 rt_free(rt);
610 }
611 }
612
613 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
614 {
615 struct fib_nh_exception *fnhe, *oldest;
616
617 oldest = rcu_dereference(hash->chain);
618 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
619 fnhe = rcu_dereference(fnhe->fnhe_next)) {
620 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
621 oldest = fnhe;
622 }
623 fnhe_flush_routes(oldest);
624 return oldest;
625 }
626
627 static inline u32 fnhe_hashfun(__be32 daddr)
628 {
629 static u32 fnhe_hashrnd __read_mostly;
630 u32 hval;
631
632 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
633 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
634 return hash_32(hval, FNHE_HASH_SHIFT);
635 }
636
637 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
638 {
639 rt->rt_pmtu = fnhe->fnhe_pmtu;
640 rt->dst.expires = fnhe->fnhe_expires;
641
642 if (fnhe->fnhe_gw) {
643 rt->rt_flags |= RTCF_REDIRECTED;
644 rt->rt_gateway = fnhe->fnhe_gw;
645 rt->rt_uses_gateway = 1;
646 }
647 }
648
649 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
650 u32 pmtu, unsigned long expires)
651 {
652 struct fnhe_hash_bucket *hash;
653 struct fib_nh_exception *fnhe;
654 struct rtable *rt;
655 unsigned int i;
656 int depth;
657 u32 hval = fnhe_hashfun(daddr);
658
659 spin_lock_bh(&fnhe_lock);
660
661 hash = rcu_dereference(nh->nh_exceptions);
662 if (!hash) {
663 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
664 if (!hash)
665 goto out_unlock;
666 rcu_assign_pointer(nh->nh_exceptions, hash);
667 }
668
669 hash += hval;
670
671 depth = 0;
672 for (fnhe = rcu_dereference(hash->chain); fnhe;
673 fnhe = rcu_dereference(fnhe->fnhe_next)) {
674 if (fnhe->fnhe_daddr == daddr)
675 break;
676 depth++;
677 }
678
679 if (fnhe) {
680 if (gw)
681 fnhe->fnhe_gw = gw;
682 if (pmtu) {
683 fnhe->fnhe_pmtu = pmtu;
684 fnhe->fnhe_expires = max(1UL, expires);
685 }
686 /* Update all cached dsts too */
687 rt = rcu_dereference(fnhe->fnhe_rth_input);
688 if (rt)
689 fill_route_from_fnhe(rt, fnhe);
690 rt = rcu_dereference(fnhe->fnhe_rth_output);
691 if (rt)
692 fill_route_from_fnhe(rt, fnhe);
693 } else {
694 if (depth > FNHE_RECLAIM_DEPTH)
695 fnhe = fnhe_oldest(hash);
696 else {
697 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
698 if (!fnhe)
699 goto out_unlock;
700
701 fnhe->fnhe_next = hash->chain;
702 rcu_assign_pointer(hash->chain, fnhe);
703 }
704 fnhe->fnhe_genid = fnhe_genid(dev_net(nh->nh_dev));
705 fnhe->fnhe_daddr = daddr;
706 fnhe->fnhe_gw = gw;
707 fnhe->fnhe_pmtu = pmtu;
708 fnhe->fnhe_expires = expires;
709
710 /* Exception created; mark the cached routes for the nexthop
711 * stale, so anyone caching it rechecks if this exception
712 * applies to them.
713 */
714 rt = rcu_dereference(nh->nh_rth_input);
715 if (rt)
716 rt->dst.obsolete = DST_OBSOLETE_KILL;
717
718 for_each_possible_cpu(i) {
719 struct rtable __rcu **prt;
720 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
721 rt = rcu_dereference(*prt);
722 if (rt)
723 rt->dst.obsolete = DST_OBSOLETE_KILL;
724 }
725 }
726
727 fnhe->fnhe_stamp = jiffies;
728
729 out_unlock:
730 spin_unlock_bh(&fnhe_lock);
731 }
732
733 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
734 bool kill_route)
735 {
736 __be32 new_gw = icmp_hdr(skb)->un.gateway;
737 __be32 old_gw = ip_hdr(skb)->saddr;
738 struct net_device *dev = skb->dev;
739 struct in_device *in_dev;
740 struct fib_result res;
741 struct neighbour *n;
742 struct net *net;
743
744 switch (icmp_hdr(skb)->code & 7) {
745 case ICMP_REDIR_NET:
746 case ICMP_REDIR_NETTOS:
747 case ICMP_REDIR_HOST:
748 case ICMP_REDIR_HOSTTOS:
749 break;
750
751 default:
752 return;
753 }
754
755 if (rt->rt_gateway != old_gw)
756 return;
757
758 in_dev = __in_dev_get_rcu(dev);
759 if (!in_dev)
760 return;
761
762 net = dev_net(dev);
763 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
764 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
765 ipv4_is_zeronet(new_gw))
766 goto reject_redirect;
767
768 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
769 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
770 goto reject_redirect;
771 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
772 goto reject_redirect;
773 } else {
774 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
775 goto reject_redirect;
776 }
777
778 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
779 if (!n)
780 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
781 if (!IS_ERR(n)) {
782 if (!(n->nud_state & NUD_VALID)) {
783 neigh_event_send(n, NULL);
784 } else {
785 if (fib_lookup(net, fl4, &res, 0) == 0) {
786 struct fib_nh *nh = &FIB_RES_NH(res);
787
788 update_or_create_fnhe(nh, fl4->daddr, new_gw,
789 0, jiffies + ip_rt_gc_timeout);
790 }
791 if (kill_route)
792 rt->dst.obsolete = DST_OBSOLETE_KILL;
793 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
794 }
795 neigh_release(n);
796 }
797 return;
798
799 reject_redirect:
800 #ifdef CONFIG_IP_ROUTE_VERBOSE
801 if (IN_DEV_LOG_MARTIANS(in_dev)) {
802 const struct iphdr *iph = (const struct iphdr *) skb->data;
803 __be32 daddr = iph->daddr;
804 __be32 saddr = iph->saddr;
805
806 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
807 " Advised path = %pI4 -> %pI4\n",
808 &old_gw, dev->name, &new_gw,
809 &saddr, &daddr);
810 }
811 #endif
812 ;
813 }
814
815 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
816 {
817 struct rtable *rt;
818 struct flowi4 fl4;
819 const struct iphdr *iph = (const struct iphdr *) skb->data;
820 struct net *net = dev_net(skb->dev);
821 int oif = skb->dev->ifindex;
822 u8 tos = RT_TOS(iph->tos);
823 u8 prot = iph->protocol;
824 u32 mark = skb->mark;
825
826 rt = (struct rtable *) dst;
827
828 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
829 __ip_do_redirect(rt, skb, &fl4, true);
830 }
831
832 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
833 {
834 struct rtable *rt = (struct rtable *)dst;
835 struct dst_entry *ret = dst;
836
837 if (rt) {
838 if (dst->obsolete > 0) {
839 ip_rt_put(rt);
840 ret = NULL;
841 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
842 rt->dst.expires) {
843 ip_rt_put(rt);
844 ret = NULL;
845 }
846 }
847 return ret;
848 }
849
850 /*
851 * Algorithm:
852 * 1. The first ip_rt_redirect_number redirects are sent
853 * with exponential backoff, then we stop sending them at all,
854 * assuming that the host ignores our redirects.
855 * 2. If we did not see packets requiring redirects
856 * during ip_rt_redirect_silence, we assume that the host
857 * forgot redirected route and start to send redirects again.
858 *
859 * This algorithm is much cheaper and more intelligent than dumb load limiting
860 * in icmp.c.
861 *
862 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
863 * and "frag. need" (breaks PMTU discovery) in icmp.c.
864 */
865
866 void ip_rt_send_redirect(struct sk_buff *skb)
867 {
868 struct rtable *rt = skb_rtable(skb);
869 struct in_device *in_dev;
870 struct inet_peer *peer;
871 struct net *net;
872 int log_martians;
873 int vif;
874
875 rcu_read_lock();
876 in_dev = __in_dev_get_rcu(rt->dst.dev);
877 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
878 rcu_read_unlock();
879 return;
880 }
881 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
882 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
883 rcu_read_unlock();
884
885 net = dev_net(rt->dst.dev);
886 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
887 if (!peer) {
888 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
889 rt_nexthop(rt, ip_hdr(skb)->daddr));
890 return;
891 }
892
893 /* No redirected packets during ip_rt_redirect_silence;
894 * reset the algorithm.
895 */
896 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence))
897 peer->rate_tokens = 0;
898
899 /* Too many ignored redirects; do not send anything
900 * set dst.rate_last to the last seen redirected packet.
901 */
902 if (peer->rate_tokens >= ip_rt_redirect_number) {
903 peer->rate_last = jiffies;
904 goto out_put_peer;
905 }
906
907 /* Check for load limit; set rate_last to the latest sent
908 * redirect.
909 */
910 if (peer->rate_tokens == 0 ||
911 time_after(jiffies,
912 (peer->rate_last +
913 (ip_rt_redirect_load << peer->rate_tokens)))) {
914 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
915
916 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
917 peer->rate_last = jiffies;
918 ++peer->rate_tokens;
919 #ifdef CONFIG_IP_ROUTE_VERBOSE
920 if (log_martians &&
921 peer->rate_tokens == ip_rt_redirect_number)
922 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
923 &ip_hdr(skb)->saddr, inet_iif(skb),
924 &ip_hdr(skb)->daddr, &gw);
925 #endif
926 }
927 out_put_peer:
928 inet_putpeer(peer);
929 }
930
931 static int ip_error(struct sk_buff *skb)
932 {
933 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
934 struct rtable *rt = skb_rtable(skb);
935 struct inet_peer *peer;
936 unsigned long now;
937 struct net *net;
938 bool send;
939 int code;
940
941 /* IP on this device is disabled. */
942 if (!in_dev)
943 goto out;
944
945 net = dev_net(rt->dst.dev);
946 if (!IN_DEV_FORWARD(in_dev)) {
947 switch (rt->dst.error) {
948 case EHOSTUNREACH:
949 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
950 break;
951
952 case ENETUNREACH:
953 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
954 break;
955 }
956 goto out;
957 }
958
959 switch (rt->dst.error) {
960 case EINVAL:
961 default:
962 goto out;
963 case EHOSTUNREACH:
964 code = ICMP_HOST_UNREACH;
965 break;
966 case ENETUNREACH:
967 code = ICMP_NET_UNREACH;
968 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
969 break;
970 case EACCES:
971 code = ICMP_PKT_FILTERED;
972 break;
973 }
974
975 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
976 l3mdev_master_ifindex(skb->dev), 1);
977
978 send = true;
979 if (peer) {
980 now = jiffies;
981 peer->rate_tokens += now - peer->rate_last;
982 if (peer->rate_tokens > ip_rt_error_burst)
983 peer->rate_tokens = ip_rt_error_burst;
984 peer->rate_last = now;
985 if (peer->rate_tokens >= ip_rt_error_cost)
986 peer->rate_tokens -= ip_rt_error_cost;
987 else
988 send = false;
989 inet_putpeer(peer);
990 }
991 if (send)
992 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
993
994 out: kfree_skb(skb);
995 return 0;
996 }
997
998 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
999 {
1000 struct dst_entry *dst = &rt->dst;
1001 struct fib_result res;
1002
1003 if (dst_metric_locked(dst, RTAX_MTU))
1004 return;
1005
1006 if (ipv4_mtu(dst) < mtu)
1007 return;
1008
1009 if (mtu < ip_rt_min_pmtu)
1010 mtu = ip_rt_min_pmtu;
1011
1012 if (rt->rt_pmtu == mtu &&
1013 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1014 return;
1015
1016 rcu_read_lock();
1017 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1018 struct fib_nh *nh = &FIB_RES_NH(res);
1019
1020 update_or_create_fnhe(nh, fl4->daddr, 0, mtu,
1021 jiffies + ip_rt_mtu_expires);
1022 }
1023 rcu_read_unlock();
1024 }
1025
1026 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1027 struct sk_buff *skb, u32 mtu)
1028 {
1029 struct rtable *rt = (struct rtable *) dst;
1030 struct flowi4 fl4;
1031
1032 ip_rt_build_flow_key(&fl4, sk, skb);
1033 __ip_rt_update_pmtu(rt, &fl4, mtu);
1034 }
1035
1036 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1037 int oif, u32 mark, u8 protocol, int flow_flags)
1038 {
1039 const struct iphdr *iph = (const struct iphdr *) skb->data;
1040 struct flowi4 fl4;
1041 struct rtable *rt;
1042
1043 if (!mark)
1044 mark = IP4_REPLY_MARK(net, skb->mark);
1045
1046 __build_flow_key(net, &fl4, NULL, iph, oif,
1047 RT_TOS(iph->tos), protocol, mark, flow_flags);
1048 rt = __ip_route_output_key(net, &fl4);
1049 if (!IS_ERR(rt)) {
1050 __ip_rt_update_pmtu(rt, &fl4, mtu);
1051 ip_rt_put(rt);
1052 }
1053 }
1054 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1055
1056 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1057 {
1058 const struct iphdr *iph = (const struct iphdr *) skb->data;
1059 struct flowi4 fl4;
1060 struct rtable *rt;
1061
1062 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1063
1064 if (!fl4.flowi4_mark)
1065 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1066
1067 rt = __ip_route_output_key(sock_net(sk), &fl4);
1068 if (!IS_ERR(rt)) {
1069 __ip_rt_update_pmtu(rt, &fl4, mtu);
1070 ip_rt_put(rt);
1071 }
1072 }
1073
1074 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1075 {
1076 const struct iphdr *iph = (const struct iphdr *) skb->data;
1077 struct flowi4 fl4;
1078 struct rtable *rt;
1079 struct dst_entry *odst = NULL;
1080 bool new = false;
1081 struct net *net = sock_net(sk);
1082
1083 bh_lock_sock(sk);
1084
1085 if (!ip_sk_accept_pmtu(sk))
1086 goto out;
1087
1088 odst = sk_dst_get(sk);
1089
1090 if (sock_owned_by_user(sk) || !odst) {
1091 __ipv4_sk_update_pmtu(skb, sk, mtu);
1092 goto out;
1093 }
1094
1095 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1096
1097 rt = (struct rtable *)odst;
1098 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1099 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1100 if (IS_ERR(rt))
1101 goto out;
1102
1103 new = true;
1104 }
1105
1106 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1107
1108 if (!dst_check(&rt->dst, 0)) {
1109 if (new)
1110 dst_release(&rt->dst);
1111
1112 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1113 if (IS_ERR(rt))
1114 goto out;
1115
1116 new = true;
1117 }
1118
1119 if (new)
1120 sk_dst_set(sk, &rt->dst);
1121
1122 out:
1123 bh_unlock_sock(sk);
1124 dst_release(odst);
1125 }
1126 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1127
1128 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1129 int oif, u32 mark, u8 protocol, int flow_flags)
1130 {
1131 const struct iphdr *iph = (const struct iphdr *) skb->data;
1132 struct flowi4 fl4;
1133 struct rtable *rt;
1134
1135 __build_flow_key(net, &fl4, NULL, iph, oif,
1136 RT_TOS(iph->tos), protocol, mark, flow_flags);
1137 rt = __ip_route_output_key(net, &fl4);
1138 if (!IS_ERR(rt)) {
1139 __ip_do_redirect(rt, skb, &fl4, false);
1140 ip_rt_put(rt);
1141 }
1142 }
1143 EXPORT_SYMBOL_GPL(ipv4_redirect);
1144
1145 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1146 {
1147 const struct iphdr *iph = (const struct iphdr *) skb->data;
1148 struct flowi4 fl4;
1149 struct rtable *rt;
1150 struct net *net = sock_net(sk);
1151
1152 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1153 rt = __ip_route_output_key(net, &fl4);
1154 if (!IS_ERR(rt)) {
1155 __ip_do_redirect(rt, skb, &fl4, false);
1156 ip_rt_put(rt);
1157 }
1158 }
1159 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1160
1161 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1162 {
1163 struct rtable *rt = (struct rtable *) dst;
1164
1165 /* All IPV4 dsts are created with ->obsolete set to the value
1166 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1167 * into this function always.
1168 *
1169 * When a PMTU/redirect information update invalidates a route,
1170 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1171 * DST_OBSOLETE_DEAD by dst_free().
1172 */
1173 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1174 return NULL;
1175 return dst;
1176 }
1177
1178 static void ipv4_link_failure(struct sk_buff *skb)
1179 {
1180 struct rtable *rt;
1181
1182 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1183
1184 rt = skb_rtable(skb);
1185 if (rt)
1186 dst_set_expires(&rt->dst, 0);
1187 }
1188
1189 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1190 {
1191 pr_debug("%s: %pI4 -> %pI4, %s\n",
1192 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1193 skb->dev ? skb->dev->name : "?");
1194 kfree_skb(skb);
1195 WARN_ON(1);
1196 return 0;
1197 }
1198
1199 /*
1200 We do not cache source address of outgoing interface,
1201 because it is used only by IP RR, TS and SRR options,
1202 so that it out of fast path.
1203
1204 BTW remember: "addr" is allowed to be not aligned
1205 in IP options!
1206 */
1207
1208 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1209 {
1210 __be32 src;
1211
1212 if (rt_is_output_route(rt))
1213 src = ip_hdr(skb)->saddr;
1214 else {
1215 struct fib_result res;
1216 struct flowi4 fl4;
1217 struct iphdr *iph;
1218
1219 iph = ip_hdr(skb);
1220
1221 memset(&fl4, 0, sizeof(fl4));
1222 fl4.daddr = iph->daddr;
1223 fl4.saddr = iph->saddr;
1224 fl4.flowi4_tos = RT_TOS(iph->tos);
1225 fl4.flowi4_oif = rt->dst.dev->ifindex;
1226 fl4.flowi4_iif = skb->dev->ifindex;
1227 fl4.flowi4_mark = skb->mark;
1228
1229 rcu_read_lock();
1230 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1231 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1232 else
1233 src = inet_select_addr(rt->dst.dev,
1234 rt_nexthop(rt, iph->daddr),
1235 RT_SCOPE_UNIVERSE);
1236 rcu_read_unlock();
1237 }
1238 memcpy(addr, &src, 4);
1239 }
1240
1241 #ifdef CONFIG_IP_ROUTE_CLASSID
1242 static void set_class_tag(struct rtable *rt, u32 tag)
1243 {
1244 if (!(rt->dst.tclassid & 0xFFFF))
1245 rt->dst.tclassid |= tag & 0xFFFF;
1246 if (!(rt->dst.tclassid & 0xFFFF0000))
1247 rt->dst.tclassid |= tag & 0xFFFF0000;
1248 }
1249 #endif
1250
1251 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1252 {
1253 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS);
1254
1255 if (advmss == 0) {
1256 advmss = max_t(unsigned int, dst->dev->mtu - 40,
1257 ip_rt_min_advmss);
1258 if (advmss > 65535 - 40)
1259 advmss = 65535 - 40;
1260 }
1261 return advmss;
1262 }
1263
1264 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1265 {
1266 const struct rtable *rt = (const struct rtable *) dst;
1267 unsigned int mtu = rt->rt_pmtu;
1268
1269 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1270 mtu = dst_metric_raw(dst, RTAX_MTU);
1271
1272 if (mtu)
1273 return mtu;
1274
1275 mtu = dst->dev->mtu;
1276
1277 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1278 if (rt->rt_uses_gateway && mtu > 576)
1279 mtu = 576;
1280 }
1281
1282 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1283
1284 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1285 }
1286
1287 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1288 {
1289 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1290 struct fib_nh_exception *fnhe;
1291 u32 hval;
1292
1293 if (!hash)
1294 return NULL;
1295
1296 hval = fnhe_hashfun(daddr);
1297
1298 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1299 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1300 if (fnhe->fnhe_daddr == daddr)
1301 return fnhe;
1302 }
1303 return NULL;
1304 }
1305
1306 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1307 __be32 daddr)
1308 {
1309 bool ret = false;
1310
1311 spin_lock_bh(&fnhe_lock);
1312
1313 if (daddr == fnhe->fnhe_daddr) {
1314 struct rtable __rcu **porig;
1315 struct rtable *orig;
1316 int genid = fnhe_genid(dev_net(rt->dst.dev));
1317
1318 if (rt_is_input_route(rt))
1319 porig = &fnhe->fnhe_rth_input;
1320 else
1321 porig = &fnhe->fnhe_rth_output;
1322 orig = rcu_dereference(*porig);
1323
1324 if (fnhe->fnhe_genid != genid) {
1325 fnhe->fnhe_genid = genid;
1326 fnhe->fnhe_gw = 0;
1327 fnhe->fnhe_pmtu = 0;
1328 fnhe->fnhe_expires = 0;
1329 fnhe_flush_routes(fnhe);
1330 orig = NULL;
1331 }
1332 fill_route_from_fnhe(rt, fnhe);
1333 if (!rt->rt_gateway)
1334 rt->rt_gateway = daddr;
1335
1336 if (!(rt->dst.flags & DST_NOCACHE)) {
1337 rcu_assign_pointer(*porig, rt);
1338 if (orig)
1339 rt_free(orig);
1340 ret = true;
1341 }
1342
1343 fnhe->fnhe_stamp = jiffies;
1344 }
1345 spin_unlock_bh(&fnhe_lock);
1346
1347 return ret;
1348 }
1349
1350 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1351 {
1352 struct rtable *orig, *prev, **p;
1353 bool ret = true;
1354
1355 if (rt_is_input_route(rt)) {
1356 p = (struct rtable **)&nh->nh_rth_input;
1357 } else {
1358 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1359 }
1360 orig = *p;
1361
1362 prev = cmpxchg(p, orig, rt);
1363 if (prev == orig) {
1364 if (orig)
1365 rt_free(orig);
1366 } else
1367 ret = false;
1368
1369 return ret;
1370 }
1371
1372 struct uncached_list {
1373 spinlock_t lock;
1374 struct list_head head;
1375 };
1376
1377 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1378
1379 static void rt_add_uncached_list(struct rtable *rt)
1380 {
1381 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1382
1383 rt->rt_uncached_list = ul;
1384
1385 spin_lock_bh(&ul->lock);
1386 list_add_tail(&rt->rt_uncached, &ul->head);
1387 spin_unlock_bh(&ul->lock);
1388 }
1389
1390 static void ipv4_dst_destroy(struct dst_entry *dst)
1391 {
1392 struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst);
1393 struct rtable *rt = (struct rtable *) dst;
1394
1395 if (p != &dst_default_metrics && atomic_dec_and_test(&p->refcnt))
1396 kfree(p);
1397
1398 if (!list_empty(&rt->rt_uncached)) {
1399 struct uncached_list *ul = rt->rt_uncached_list;
1400
1401 spin_lock_bh(&ul->lock);
1402 list_del(&rt->rt_uncached);
1403 spin_unlock_bh(&ul->lock);
1404 }
1405 }
1406
1407 void rt_flush_dev(struct net_device *dev)
1408 {
1409 struct net *net = dev_net(dev);
1410 struct rtable *rt;
1411 int cpu;
1412
1413 for_each_possible_cpu(cpu) {
1414 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1415
1416 spin_lock_bh(&ul->lock);
1417 list_for_each_entry(rt, &ul->head, rt_uncached) {
1418 if (rt->dst.dev != dev)
1419 continue;
1420 rt->dst.dev = net->loopback_dev;
1421 dev_hold(rt->dst.dev);
1422 dev_put(dev);
1423 }
1424 spin_unlock_bh(&ul->lock);
1425 }
1426 }
1427
1428 static bool rt_cache_valid(const struct rtable *rt)
1429 {
1430 return rt &&
1431 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1432 !rt_is_expired(rt);
1433 }
1434
1435 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1436 const struct fib_result *res,
1437 struct fib_nh_exception *fnhe,
1438 struct fib_info *fi, u16 type, u32 itag)
1439 {
1440 bool cached = false;
1441
1442 if (fi) {
1443 struct fib_nh *nh = &FIB_RES_NH(*res);
1444
1445 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1446 rt->rt_gateway = nh->nh_gw;
1447 rt->rt_uses_gateway = 1;
1448 }
1449 dst_init_metrics(&rt->dst, fi->fib_metrics->metrics, true);
1450 if (fi->fib_metrics != &dst_default_metrics) {
1451 rt->dst._metrics |= DST_METRICS_REFCOUNTED;
1452 atomic_inc(&fi->fib_metrics->refcnt);
1453 }
1454 #ifdef CONFIG_IP_ROUTE_CLASSID
1455 rt->dst.tclassid = nh->nh_tclassid;
1456 #endif
1457 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1458 if (unlikely(fnhe))
1459 cached = rt_bind_exception(rt, fnhe, daddr);
1460 else if (!(rt->dst.flags & DST_NOCACHE))
1461 cached = rt_cache_route(nh, rt);
1462 if (unlikely(!cached)) {
1463 /* Routes we intend to cache in nexthop exception or
1464 * FIB nexthop have the DST_NOCACHE bit clear.
1465 * However, if we are unsuccessful at storing this
1466 * route into the cache we really need to set it.
1467 */
1468 rt->dst.flags |= DST_NOCACHE;
1469 if (!rt->rt_gateway)
1470 rt->rt_gateway = daddr;
1471 rt_add_uncached_list(rt);
1472 }
1473 } else
1474 rt_add_uncached_list(rt);
1475
1476 #ifdef CONFIG_IP_ROUTE_CLASSID
1477 #ifdef CONFIG_IP_MULTIPLE_TABLES
1478 set_class_tag(rt, res->tclassid);
1479 #endif
1480 set_class_tag(rt, itag);
1481 #endif
1482 }
1483
1484 struct rtable *rt_dst_alloc(struct net_device *dev,
1485 unsigned int flags, u16 type,
1486 bool nopolicy, bool noxfrm, bool will_cache)
1487 {
1488 struct rtable *rt;
1489
1490 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1491 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1492 (nopolicy ? DST_NOPOLICY : 0) |
1493 (noxfrm ? DST_NOXFRM : 0));
1494
1495 if (rt) {
1496 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1497 rt->rt_flags = flags;
1498 rt->rt_type = type;
1499 rt->rt_is_input = 0;
1500 rt->rt_iif = 0;
1501 rt->rt_pmtu = 0;
1502 rt->rt_gateway = 0;
1503 rt->rt_uses_gateway = 0;
1504 rt->rt_table_id = 0;
1505 INIT_LIST_HEAD(&rt->rt_uncached);
1506
1507 rt->dst.output = ip_output;
1508 if (flags & RTCF_LOCAL)
1509 rt->dst.input = ip_local_deliver;
1510 }
1511
1512 return rt;
1513 }
1514 EXPORT_SYMBOL(rt_dst_alloc);
1515
1516 /* called in rcu_read_lock() section */
1517 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1518 u8 tos, struct net_device *dev, int our)
1519 {
1520 struct rtable *rth;
1521 struct in_device *in_dev = __in_dev_get_rcu(dev);
1522 unsigned int flags = RTCF_MULTICAST;
1523 u32 itag = 0;
1524 int err;
1525
1526 /* Primary sanity checks. */
1527
1528 if (!in_dev)
1529 return -EINVAL;
1530
1531 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1532 skb->protocol != htons(ETH_P_IP))
1533 goto e_inval;
1534
1535 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1536 goto e_inval;
1537
1538 if (ipv4_is_zeronet(saddr)) {
1539 if (!ipv4_is_local_multicast(daddr))
1540 goto e_inval;
1541 } else {
1542 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1543 in_dev, &itag);
1544 if (err < 0)
1545 goto e_err;
1546 }
1547 if (our)
1548 flags |= RTCF_LOCAL;
1549
1550 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1551 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1552 if (!rth)
1553 goto e_nobufs;
1554
1555 #ifdef CONFIG_IP_ROUTE_CLASSID
1556 rth->dst.tclassid = itag;
1557 #endif
1558 rth->dst.output = ip_rt_bug;
1559 rth->rt_is_input= 1;
1560
1561 #ifdef CONFIG_IP_MROUTE
1562 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1563 rth->dst.input = ip_mr_input;
1564 #endif
1565 RT_CACHE_STAT_INC(in_slow_mc);
1566
1567 skb_dst_set(skb, &rth->dst);
1568 return 0;
1569
1570 e_nobufs:
1571 return -ENOBUFS;
1572 e_inval:
1573 return -EINVAL;
1574 e_err:
1575 return err;
1576 }
1577
1578
1579 static void ip_handle_martian_source(struct net_device *dev,
1580 struct in_device *in_dev,
1581 struct sk_buff *skb,
1582 __be32 daddr,
1583 __be32 saddr)
1584 {
1585 RT_CACHE_STAT_INC(in_martian_src);
1586 #ifdef CONFIG_IP_ROUTE_VERBOSE
1587 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1588 /*
1589 * RFC1812 recommendation, if source is martian,
1590 * the only hint is MAC header.
1591 */
1592 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1593 &daddr, &saddr, dev->name);
1594 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1595 print_hex_dump(KERN_WARNING, "ll header: ",
1596 DUMP_PREFIX_OFFSET, 16, 1,
1597 skb_mac_header(skb),
1598 dev->hard_header_len, true);
1599 }
1600 }
1601 #endif
1602 }
1603
1604 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1605 {
1606 struct fnhe_hash_bucket *hash;
1607 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1608 u32 hval = fnhe_hashfun(daddr);
1609
1610 spin_lock_bh(&fnhe_lock);
1611
1612 hash = rcu_dereference_protected(nh->nh_exceptions,
1613 lockdep_is_held(&fnhe_lock));
1614 hash += hval;
1615
1616 fnhe_p = &hash->chain;
1617 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1618 while (fnhe) {
1619 if (fnhe->fnhe_daddr == daddr) {
1620 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1621 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1622 fnhe_flush_routes(fnhe);
1623 kfree_rcu(fnhe, rcu);
1624 break;
1625 }
1626 fnhe_p = &fnhe->fnhe_next;
1627 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1628 lockdep_is_held(&fnhe_lock));
1629 }
1630
1631 spin_unlock_bh(&fnhe_lock);
1632 }
1633
1634 static void set_lwt_redirect(struct rtable *rth)
1635 {
1636 if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1637 rth->dst.lwtstate->orig_output = rth->dst.output;
1638 rth->dst.output = lwtunnel_output;
1639 }
1640
1641 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1642 rth->dst.lwtstate->orig_input = rth->dst.input;
1643 rth->dst.input = lwtunnel_input;
1644 }
1645 }
1646
1647 /* called in rcu_read_lock() section */
1648 static int __mkroute_input(struct sk_buff *skb,
1649 const struct fib_result *res,
1650 struct in_device *in_dev,
1651 __be32 daddr, __be32 saddr, u32 tos)
1652 {
1653 struct fib_nh_exception *fnhe;
1654 struct rtable *rth;
1655 int err;
1656 struct in_device *out_dev;
1657 bool do_cache;
1658 u32 itag = 0;
1659
1660 /* get a working reference to the output device */
1661 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1662 if (!out_dev) {
1663 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1664 return -EINVAL;
1665 }
1666
1667 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1668 in_dev->dev, in_dev, &itag);
1669 if (err < 0) {
1670 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1671 saddr);
1672
1673 goto cleanup;
1674 }
1675
1676 do_cache = res->fi && !itag;
1677 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1678 skb->protocol == htons(ETH_P_IP) &&
1679 (IN_DEV_SHARED_MEDIA(out_dev) ||
1680 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1681 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1682
1683 if (skb->protocol != htons(ETH_P_IP)) {
1684 /* Not IP (i.e. ARP). Do not create route, if it is
1685 * invalid for proxy arp. DNAT routes are always valid.
1686 *
1687 * Proxy arp feature have been extended to allow, ARP
1688 * replies back to the same interface, to support
1689 * Private VLAN switch technologies. See arp.c.
1690 */
1691 if (out_dev == in_dev &&
1692 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1693 err = -EINVAL;
1694 goto cleanup;
1695 }
1696 }
1697
1698 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1699 if (do_cache) {
1700 if (fnhe) {
1701 rth = rcu_dereference(fnhe->fnhe_rth_input);
1702 if (rth && rth->dst.expires &&
1703 time_after(jiffies, rth->dst.expires)) {
1704 ip_del_fnhe(&FIB_RES_NH(*res), daddr);
1705 fnhe = NULL;
1706 } else {
1707 goto rt_cache;
1708 }
1709 }
1710
1711 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1712
1713 rt_cache:
1714 if (rt_cache_valid(rth)) {
1715 skb_dst_set_noref(skb, &rth->dst);
1716 goto out;
1717 }
1718 }
1719
1720 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1721 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1722 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1723 if (!rth) {
1724 err = -ENOBUFS;
1725 goto cleanup;
1726 }
1727
1728 rth->rt_is_input = 1;
1729 if (res->table)
1730 rth->rt_table_id = res->table->tb_id;
1731 RT_CACHE_STAT_INC(in_slow_tot);
1732
1733 rth->dst.input = ip_forward;
1734
1735 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1736 set_lwt_redirect(rth);
1737 skb_dst_set(skb, &rth->dst);
1738 out:
1739 err = 0;
1740 cleanup:
1741 return err;
1742 }
1743
1744 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1745
1746 /* To make ICMP packets follow the right flow, the multipath hash is
1747 * calculated from the inner IP addresses in reverse order.
1748 */
1749 static int ip_multipath_icmp_hash(struct sk_buff *skb)
1750 {
1751 const struct iphdr *outer_iph = ip_hdr(skb);
1752 struct icmphdr _icmph;
1753 const struct icmphdr *icmph;
1754 struct iphdr _inner_iph;
1755 const struct iphdr *inner_iph;
1756
1757 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1758 goto standard_hash;
1759
1760 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1761 &_icmph);
1762 if (!icmph)
1763 goto standard_hash;
1764
1765 if (icmph->type != ICMP_DEST_UNREACH &&
1766 icmph->type != ICMP_REDIRECT &&
1767 icmph->type != ICMP_TIME_EXCEEDED &&
1768 icmph->type != ICMP_PARAMETERPROB) {
1769 goto standard_hash;
1770 }
1771
1772 inner_iph = skb_header_pointer(skb,
1773 outer_iph->ihl * 4 + sizeof(_icmph),
1774 sizeof(_inner_iph), &_inner_iph);
1775 if (!inner_iph)
1776 goto standard_hash;
1777
1778 return fib_multipath_hash(inner_iph->daddr, inner_iph->saddr);
1779
1780 standard_hash:
1781 return fib_multipath_hash(outer_iph->saddr, outer_iph->daddr);
1782 }
1783
1784 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1785
1786 static int ip_mkroute_input(struct sk_buff *skb,
1787 struct fib_result *res,
1788 struct in_device *in_dev,
1789 __be32 daddr, __be32 saddr, u32 tos)
1790 {
1791 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1792 if (res->fi && res->fi->fib_nhs > 1) {
1793 int h;
1794
1795 if (unlikely(ip_hdr(skb)->protocol == IPPROTO_ICMP))
1796 h = ip_multipath_icmp_hash(skb);
1797 else
1798 h = fib_multipath_hash(saddr, daddr);
1799 fib_select_multipath(res, h);
1800 }
1801 #endif
1802
1803 /* create a routing cache entry */
1804 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1805 }
1806
1807 /*
1808 * NOTE. We drop all the packets that has local source
1809 * addresses, because every properly looped back packet
1810 * must have correct destination already attached by output routine.
1811 *
1812 * Such approach solves two big problems:
1813 * 1. Not simplex devices are handled properly.
1814 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1815 * called with rcu_read_lock()
1816 */
1817
1818 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1819 u8 tos, struct net_device *dev)
1820 {
1821 struct fib_result res;
1822 struct in_device *in_dev = __in_dev_get_rcu(dev);
1823 struct ip_tunnel_info *tun_info;
1824 struct flowi4 fl4;
1825 unsigned int flags = 0;
1826 u32 itag = 0;
1827 struct rtable *rth;
1828 int err = -EINVAL;
1829 struct net *net = dev_net(dev);
1830 bool do_cache;
1831
1832 /* IP on this device is disabled. */
1833
1834 if (!in_dev)
1835 goto out;
1836
1837 /* Check for the most weird martians, which can be not detected
1838 by fib_lookup.
1839 */
1840
1841 tun_info = skb_tunnel_info(skb);
1842 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1843 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1844 else
1845 fl4.flowi4_tun_key.tun_id = 0;
1846 skb_dst_drop(skb);
1847
1848 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1849 goto martian_source;
1850
1851 res.fi = NULL;
1852 res.table = NULL;
1853 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1854 goto brd_input;
1855
1856 /* Accept zero addresses only to limited broadcast;
1857 * I even do not know to fix it or not. Waiting for complains :-)
1858 */
1859 if (ipv4_is_zeronet(saddr))
1860 goto martian_source;
1861
1862 if (ipv4_is_zeronet(daddr))
1863 goto martian_destination;
1864
1865 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1866 * and call it once if daddr or/and saddr are loopback addresses
1867 */
1868 if (ipv4_is_loopback(daddr)) {
1869 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1870 goto martian_destination;
1871 } else if (ipv4_is_loopback(saddr)) {
1872 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1873 goto martian_source;
1874 }
1875
1876 /*
1877 * Now we are ready to route packet.
1878 */
1879 fl4.flowi4_oif = 0;
1880 fl4.flowi4_iif = dev->ifindex;
1881 fl4.flowi4_mark = skb->mark;
1882 fl4.flowi4_tos = tos;
1883 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1884 fl4.flowi4_flags = 0;
1885 fl4.daddr = daddr;
1886 fl4.saddr = saddr;
1887 fl4.flowi4_uid = sock_net_uid(net, NULL);
1888 err = fib_lookup(net, &fl4, &res, 0);
1889 if (err != 0) {
1890 if (!IN_DEV_FORWARD(in_dev))
1891 err = -EHOSTUNREACH;
1892 goto no_route;
1893 }
1894
1895 if (res.type == RTN_BROADCAST)
1896 goto brd_input;
1897
1898 if (res.type == RTN_LOCAL) {
1899 err = fib_validate_source(skb, saddr, daddr, tos,
1900 0, dev, in_dev, &itag);
1901 if (err < 0)
1902 goto martian_source;
1903 goto local_input;
1904 }
1905
1906 if (!IN_DEV_FORWARD(in_dev)) {
1907 err = -EHOSTUNREACH;
1908 goto no_route;
1909 }
1910 if (res.type != RTN_UNICAST)
1911 goto martian_destination;
1912
1913 err = ip_mkroute_input(skb, &res, in_dev, daddr, saddr, tos);
1914 out: return err;
1915
1916 brd_input:
1917 if (skb->protocol != htons(ETH_P_IP))
1918 goto e_inval;
1919
1920 if (!ipv4_is_zeronet(saddr)) {
1921 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1922 in_dev, &itag);
1923 if (err < 0)
1924 goto martian_source;
1925 }
1926 flags |= RTCF_BROADCAST;
1927 res.type = RTN_BROADCAST;
1928 RT_CACHE_STAT_INC(in_brd);
1929
1930 local_input:
1931 do_cache = false;
1932 if (res.fi) {
1933 if (!itag) {
1934 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1935 if (rt_cache_valid(rth)) {
1936 skb_dst_set_noref(skb, &rth->dst);
1937 err = 0;
1938 goto out;
1939 }
1940 do_cache = true;
1941 }
1942 }
1943
1944 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
1945 flags | RTCF_LOCAL, res.type,
1946 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1947 if (!rth)
1948 goto e_nobufs;
1949
1950 rth->dst.output= ip_rt_bug;
1951 #ifdef CONFIG_IP_ROUTE_CLASSID
1952 rth->dst.tclassid = itag;
1953 #endif
1954 rth->rt_is_input = 1;
1955 if (res.table)
1956 rth->rt_table_id = res.table->tb_id;
1957
1958 RT_CACHE_STAT_INC(in_slow_tot);
1959 if (res.type == RTN_UNREACHABLE) {
1960 rth->dst.input= ip_error;
1961 rth->dst.error= -err;
1962 rth->rt_flags &= ~RTCF_LOCAL;
1963 }
1964
1965 if (do_cache) {
1966 struct fib_nh *nh = &FIB_RES_NH(res);
1967
1968 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1969 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1970 WARN_ON(rth->dst.input == lwtunnel_input);
1971 rth->dst.lwtstate->orig_input = rth->dst.input;
1972 rth->dst.input = lwtunnel_input;
1973 }
1974
1975 if (unlikely(!rt_cache_route(nh, rth))) {
1976 rth->dst.flags |= DST_NOCACHE;
1977 rt_add_uncached_list(rth);
1978 }
1979 }
1980 skb_dst_set(skb, &rth->dst);
1981 err = 0;
1982 goto out;
1983
1984 no_route:
1985 RT_CACHE_STAT_INC(in_no_route);
1986 res.type = RTN_UNREACHABLE;
1987 res.fi = NULL;
1988 res.table = NULL;
1989 goto local_input;
1990
1991 /*
1992 * Do not cache martian addresses: they should be logged (RFC1812)
1993 */
1994 martian_destination:
1995 RT_CACHE_STAT_INC(in_martian_dst);
1996 #ifdef CONFIG_IP_ROUTE_VERBOSE
1997 if (IN_DEV_LOG_MARTIANS(in_dev))
1998 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
1999 &daddr, &saddr, dev->name);
2000 #endif
2001
2002 e_inval:
2003 err = -EINVAL;
2004 goto out;
2005
2006 e_nobufs:
2007 err = -ENOBUFS;
2008 goto out;
2009
2010 martian_source:
2011 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2012 goto out;
2013 }
2014
2015 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2016 u8 tos, struct net_device *dev)
2017 {
2018 int res;
2019
2020 tos &= IPTOS_RT_MASK;
2021 rcu_read_lock();
2022
2023 /* Multicast recognition logic is moved from route cache to here.
2024 The problem was that too many Ethernet cards have broken/missing
2025 hardware multicast filters :-( As result the host on multicasting
2026 network acquires a lot of useless route cache entries, sort of
2027 SDR messages from all the world. Now we try to get rid of them.
2028 Really, provided software IP multicast filter is organized
2029 reasonably (at least, hashed), it does not result in a slowdown
2030 comparing with route cache reject entries.
2031 Note, that multicast routers are not affected, because
2032 route cache entry is created eventually.
2033 */
2034 if (ipv4_is_multicast(daddr)) {
2035 struct in_device *in_dev = __in_dev_get_rcu(dev);
2036 int our = 0;
2037
2038 if (in_dev)
2039 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2040 ip_hdr(skb)->protocol);
2041
2042 /* check l3 master if no match yet */
2043 if ((!in_dev || !our) && netif_is_l3_slave(dev)) {
2044 struct in_device *l3_in_dev;
2045
2046 l3_in_dev = __in_dev_get_rcu(skb->dev);
2047 if (l3_in_dev)
2048 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2049 ip_hdr(skb)->protocol);
2050 }
2051
2052 res = -EINVAL;
2053 if (our
2054 #ifdef CONFIG_IP_MROUTE
2055 ||
2056 (!ipv4_is_local_multicast(daddr) &&
2057 IN_DEV_MFORWARD(in_dev))
2058 #endif
2059 ) {
2060 res = ip_route_input_mc(skb, daddr, saddr,
2061 tos, dev, our);
2062 }
2063 rcu_read_unlock();
2064 return res;
2065 }
2066 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2067 rcu_read_unlock();
2068 return res;
2069 }
2070 EXPORT_SYMBOL(ip_route_input_noref);
2071
2072 /* called with rcu_read_lock() */
2073 static struct rtable *__mkroute_output(const struct fib_result *res,
2074 const struct flowi4 *fl4, int orig_oif,
2075 struct net_device *dev_out,
2076 unsigned int flags)
2077 {
2078 struct fib_info *fi = res->fi;
2079 struct fib_nh_exception *fnhe;
2080 struct in_device *in_dev;
2081 u16 type = res->type;
2082 struct rtable *rth;
2083 bool do_cache;
2084
2085 in_dev = __in_dev_get_rcu(dev_out);
2086 if (!in_dev)
2087 return ERR_PTR(-EINVAL);
2088
2089 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2090 if (ipv4_is_loopback(fl4->saddr) &&
2091 !(dev_out->flags & IFF_LOOPBACK) &&
2092 !netif_is_l3_master(dev_out))
2093 return ERR_PTR(-EINVAL);
2094
2095 if (ipv4_is_lbcast(fl4->daddr))
2096 type = RTN_BROADCAST;
2097 else if (ipv4_is_multicast(fl4->daddr))
2098 type = RTN_MULTICAST;
2099 else if (ipv4_is_zeronet(fl4->daddr))
2100 return ERR_PTR(-EINVAL);
2101
2102 if (dev_out->flags & IFF_LOOPBACK)
2103 flags |= RTCF_LOCAL;
2104
2105 do_cache = true;
2106 if (type == RTN_BROADCAST) {
2107 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2108 fi = NULL;
2109 } else if (type == RTN_MULTICAST) {
2110 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2111 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2112 fl4->flowi4_proto))
2113 flags &= ~RTCF_LOCAL;
2114 else
2115 do_cache = false;
2116 /* If multicast route do not exist use
2117 * default one, but do not gateway in this case.
2118 * Yes, it is hack.
2119 */
2120 if (fi && res->prefixlen < 4)
2121 fi = NULL;
2122 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2123 (orig_oif != dev_out->ifindex)) {
2124 /* For local routes that require a particular output interface
2125 * we do not want to cache the result. Caching the result
2126 * causes incorrect behaviour when there are multiple source
2127 * addresses on the interface, the end result being that if the
2128 * intended recipient is waiting on that interface for the
2129 * packet he won't receive it because it will be delivered on
2130 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2131 * be set to the loopback interface as well.
2132 */
2133 fi = NULL;
2134 }
2135
2136 fnhe = NULL;
2137 do_cache &= fi != NULL;
2138 if (do_cache) {
2139 struct rtable __rcu **prth;
2140 struct fib_nh *nh = &FIB_RES_NH(*res);
2141
2142 fnhe = find_exception(nh, fl4->daddr);
2143 if (fnhe) {
2144 prth = &fnhe->fnhe_rth_output;
2145 rth = rcu_dereference(*prth);
2146 if (rth && rth->dst.expires &&
2147 time_after(jiffies, rth->dst.expires)) {
2148 ip_del_fnhe(nh, fl4->daddr);
2149 fnhe = NULL;
2150 } else {
2151 goto rt_cache;
2152 }
2153 }
2154
2155 if (unlikely(fl4->flowi4_flags &
2156 FLOWI_FLAG_KNOWN_NH &&
2157 !(nh->nh_gw &&
2158 nh->nh_scope == RT_SCOPE_LINK))) {
2159 do_cache = false;
2160 goto add;
2161 }
2162 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2163 rth = rcu_dereference(*prth);
2164
2165 rt_cache:
2166 if (rt_cache_valid(rth)) {
2167 dst_hold(&rth->dst);
2168 return rth;
2169 }
2170 }
2171
2172 add:
2173 rth = rt_dst_alloc(dev_out, flags, type,
2174 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2175 IN_DEV_CONF_GET(in_dev, NOXFRM),
2176 do_cache);
2177 if (!rth)
2178 return ERR_PTR(-ENOBUFS);
2179
2180 rth->rt_iif = orig_oif ? : 0;
2181 if (res->table)
2182 rth->rt_table_id = res->table->tb_id;
2183
2184 RT_CACHE_STAT_INC(out_slow_tot);
2185
2186 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2187 if (flags & RTCF_LOCAL &&
2188 !(dev_out->flags & IFF_LOOPBACK)) {
2189 rth->dst.output = ip_mc_output;
2190 RT_CACHE_STAT_INC(out_slow_mc);
2191 }
2192 #ifdef CONFIG_IP_MROUTE
2193 if (type == RTN_MULTICAST) {
2194 if (IN_DEV_MFORWARD(in_dev) &&
2195 !ipv4_is_local_multicast(fl4->daddr)) {
2196 rth->dst.input = ip_mr_input;
2197 rth->dst.output = ip_mc_output;
2198 }
2199 }
2200 #endif
2201 }
2202
2203 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2204 set_lwt_redirect(rth);
2205
2206 return rth;
2207 }
2208
2209 /*
2210 * Major route resolver routine.
2211 */
2212
2213 struct rtable *__ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2214 int mp_hash)
2215 {
2216 struct net_device *dev_out = NULL;
2217 __u8 tos = RT_FL_TOS(fl4);
2218 unsigned int flags = 0;
2219 struct fib_result res;
2220 struct rtable *rth;
2221 int orig_oif;
2222 int err = -ENETUNREACH;
2223
2224 res.tclassid = 0;
2225 res.fi = NULL;
2226 res.table = NULL;
2227
2228 orig_oif = fl4->flowi4_oif;
2229
2230 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2231 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2232 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2233 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2234
2235 rcu_read_lock();
2236 if (fl4->saddr) {
2237 rth = ERR_PTR(-EINVAL);
2238 if (ipv4_is_multicast(fl4->saddr) ||
2239 ipv4_is_lbcast(fl4->saddr) ||
2240 ipv4_is_zeronet(fl4->saddr))
2241 goto out;
2242
2243 /* I removed check for oif == dev_out->oif here.
2244 It was wrong for two reasons:
2245 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2246 is assigned to multiple interfaces.
2247 2. Moreover, we are allowed to send packets with saddr
2248 of another iface. --ANK
2249 */
2250
2251 if (fl4->flowi4_oif == 0 &&
2252 (ipv4_is_multicast(fl4->daddr) ||
2253 ipv4_is_lbcast(fl4->daddr))) {
2254 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2255 dev_out = __ip_dev_find(net, fl4->saddr, false);
2256 if (!dev_out)
2257 goto out;
2258
2259 /* Special hack: user can direct multicasts
2260 and limited broadcast via necessary interface
2261 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2262 This hack is not just for fun, it allows
2263 vic,vat and friends to work.
2264 They bind socket to loopback, set ttl to zero
2265 and expect that it will work.
2266 From the viewpoint of routing cache they are broken,
2267 because we are not allowed to build multicast path
2268 with loopback source addr (look, routing cache
2269 cannot know, that ttl is zero, so that packet
2270 will not leave this host and route is valid).
2271 Luckily, this hack is good workaround.
2272 */
2273
2274 fl4->flowi4_oif = dev_out->ifindex;
2275 goto make_route;
2276 }
2277
2278 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2279 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2280 if (!__ip_dev_find(net, fl4->saddr, false))
2281 goto out;
2282 }
2283 }
2284
2285
2286 if (fl4->flowi4_oif) {
2287 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2288 rth = ERR_PTR(-ENODEV);
2289 if (!dev_out)
2290 goto out;
2291
2292 /* RACE: Check return value of inet_select_addr instead. */
2293 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2294 rth = ERR_PTR(-ENETUNREACH);
2295 goto out;
2296 }
2297 if (ipv4_is_local_multicast(fl4->daddr) ||
2298 ipv4_is_lbcast(fl4->daddr) ||
2299 fl4->flowi4_proto == IPPROTO_IGMP) {
2300 if (!fl4->saddr)
2301 fl4->saddr = inet_select_addr(dev_out, 0,
2302 RT_SCOPE_LINK);
2303 goto make_route;
2304 }
2305 if (!fl4->saddr) {
2306 if (ipv4_is_multicast(fl4->daddr))
2307 fl4->saddr = inet_select_addr(dev_out, 0,
2308 fl4->flowi4_scope);
2309 else if (!fl4->daddr)
2310 fl4->saddr = inet_select_addr(dev_out, 0,
2311 RT_SCOPE_HOST);
2312 }
2313 }
2314
2315 if (!fl4->daddr) {
2316 fl4->daddr = fl4->saddr;
2317 if (!fl4->daddr)
2318 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2319 dev_out = net->loopback_dev;
2320 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2321 res.type = RTN_LOCAL;
2322 flags |= RTCF_LOCAL;
2323 goto make_route;
2324 }
2325
2326 err = fib_lookup(net, fl4, &res, 0);
2327 if (err) {
2328 res.fi = NULL;
2329 res.table = NULL;
2330 if (fl4->flowi4_oif &&
2331 (ipv4_is_multicast(fl4->daddr) ||
2332 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2333 /* Apparently, routing tables are wrong. Assume,
2334 that the destination is on link.
2335
2336 WHY? DW.
2337 Because we are allowed to send to iface
2338 even if it has NO routes and NO assigned
2339 addresses. When oif is specified, routing
2340 tables are looked up with only one purpose:
2341 to catch if destination is gatewayed, rather than
2342 direct. Moreover, if MSG_DONTROUTE is set,
2343 we send packet, ignoring both routing tables
2344 and ifaddr state. --ANK
2345
2346
2347 We could make it even if oif is unknown,
2348 likely IPv6, but we do not.
2349 */
2350
2351 if (fl4->saddr == 0)
2352 fl4->saddr = inet_select_addr(dev_out, 0,
2353 RT_SCOPE_LINK);
2354 res.type = RTN_UNICAST;
2355 goto make_route;
2356 }
2357 rth = ERR_PTR(err);
2358 goto out;
2359 }
2360
2361 if (res.type == RTN_LOCAL) {
2362 if (!fl4->saddr) {
2363 if (res.fi->fib_prefsrc)
2364 fl4->saddr = res.fi->fib_prefsrc;
2365 else
2366 fl4->saddr = fl4->daddr;
2367 }
2368
2369 /* L3 master device is the loopback for that domain */
2370 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(res)) ? :
2371 net->loopback_dev;
2372 fl4->flowi4_oif = dev_out->ifindex;
2373 flags |= RTCF_LOCAL;
2374 goto make_route;
2375 }
2376
2377 fib_select_path(net, &res, fl4, mp_hash);
2378
2379 dev_out = FIB_RES_DEV(res);
2380 fl4->flowi4_oif = dev_out->ifindex;
2381
2382
2383 make_route:
2384 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2385
2386 out:
2387 rcu_read_unlock();
2388 return rth;
2389 }
2390 EXPORT_SYMBOL_GPL(__ip_route_output_key_hash);
2391
2392 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2393 {
2394 return NULL;
2395 }
2396
2397 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2398 {
2399 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2400
2401 return mtu ? : dst->dev->mtu;
2402 }
2403
2404 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2405 struct sk_buff *skb, u32 mtu)
2406 {
2407 }
2408
2409 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2410 struct sk_buff *skb)
2411 {
2412 }
2413
2414 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2415 unsigned long old)
2416 {
2417 return NULL;
2418 }
2419
2420 static struct dst_ops ipv4_dst_blackhole_ops = {
2421 .family = AF_INET,
2422 .check = ipv4_blackhole_dst_check,
2423 .mtu = ipv4_blackhole_mtu,
2424 .default_advmss = ipv4_default_advmss,
2425 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2426 .redirect = ipv4_rt_blackhole_redirect,
2427 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2428 .neigh_lookup = ipv4_neigh_lookup,
2429 };
2430
2431 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2432 {
2433 struct rtable *ort = (struct rtable *) dst_orig;
2434 struct rtable *rt;
2435
2436 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2437 if (rt) {
2438 struct dst_entry *new = &rt->dst;
2439
2440 new->__use = 1;
2441 new->input = dst_discard;
2442 new->output = dst_discard_out;
2443
2444 new->dev = ort->dst.dev;
2445 if (new->dev)
2446 dev_hold(new->dev);
2447
2448 rt->rt_is_input = ort->rt_is_input;
2449 rt->rt_iif = ort->rt_iif;
2450 rt->rt_pmtu = ort->rt_pmtu;
2451
2452 rt->rt_genid = rt_genid_ipv4(net);
2453 rt->rt_flags = ort->rt_flags;
2454 rt->rt_type = ort->rt_type;
2455 rt->rt_gateway = ort->rt_gateway;
2456 rt->rt_uses_gateway = ort->rt_uses_gateway;
2457
2458 INIT_LIST_HEAD(&rt->rt_uncached);
2459 dst_free(new);
2460 }
2461
2462 dst_release(dst_orig);
2463
2464 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2465 }
2466
2467 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2468 const struct sock *sk)
2469 {
2470 struct rtable *rt = __ip_route_output_key(net, flp4);
2471
2472 if (IS_ERR(rt))
2473 return rt;
2474
2475 if (flp4->flowi4_proto)
2476 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2477 flowi4_to_flowi(flp4),
2478 sk, 0);
2479
2480 return rt;
2481 }
2482 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2483
2484 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2485 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2486 u32 seq, int event)
2487 {
2488 struct rtable *rt = skb_rtable(skb);
2489 struct rtmsg *r;
2490 struct nlmsghdr *nlh;
2491 unsigned long expires = 0;
2492 u32 error;
2493 u32 metrics[RTAX_MAX];
2494
2495 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), 0);
2496 if (!nlh)
2497 return -EMSGSIZE;
2498
2499 r = nlmsg_data(nlh);
2500 r->rtm_family = AF_INET;
2501 r->rtm_dst_len = 32;
2502 r->rtm_src_len = 0;
2503 r->rtm_tos = fl4->flowi4_tos;
2504 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2505 if (nla_put_u32(skb, RTA_TABLE, table_id))
2506 goto nla_put_failure;
2507 r->rtm_type = rt->rt_type;
2508 r->rtm_scope = RT_SCOPE_UNIVERSE;
2509 r->rtm_protocol = RTPROT_UNSPEC;
2510 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2511 if (rt->rt_flags & RTCF_NOTIFY)
2512 r->rtm_flags |= RTM_F_NOTIFY;
2513 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2514 r->rtm_flags |= RTCF_DOREDIRECT;
2515
2516 if (nla_put_in_addr(skb, RTA_DST, dst))
2517 goto nla_put_failure;
2518 if (src) {
2519 r->rtm_src_len = 32;
2520 if (nla_put_in_addr(skb, RTA_SRC, src))
2521 goto nla_put_failure;
2522 }
2523 if (rt->dst.dev &&
2524 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2525 goto nla_put_failure;
2526 #ifdef CONFIG_IP_ROUTE_CLASSID
2527 if (rt->dst.tclassid &&
2528 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2529 goto nla_put_failure;
2530 #endif
2531 if (!rt_is_input_route(rt) &&
2532 fl4->saddr != src) {
2533 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2534 goto nla_put_failure;
2535 }
2536 if (rt->rt_uses_gateway &&
2537 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2538 goto nla_put_failure;
2539
2540 expires = rt->dst.expires;
2541 if (expires) {
2542 unsigned long now = jiffies;
2543
2544 if (time_before(now, expires))
2545 expires -= now;
2546 else
2547 expires = 0;
2548 }
2549
2550 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2551 if (rt->rt_pmtu && expires)
2552 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2553 if (rtnetlink_put_metrics(skb, metrics) < 0)
2554 goto nla_put_failure;
2555
2556 if (fl4->flowi4_mark &&
2557 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2558 goto nla_put_failure;
2559
2560 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2561 nla_put_u32(skb, RTA_UID,
2562 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2563 goto nla_put_failure;
2564
2565 error = rt->dst.error;
2566
2567 if (rt_is_input_route(rt)) {
2568 #ifdef CONFIG_IP_MROUTE
2569 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2570 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2571 int err = ipmr_get_route(net, skb,
2572 fl4->saddr, fl4->daddr,
2573 r, portid);
2574
2575 if (err <= 0) {
2576 if (err == 0)
2577 return 0;
2578 goto nla_put_failure;
2579 }
2580 } else
2581 #endif
2582 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2583 goto nla_put_failure;
2584 }
2585
2586 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2587 goto nla_put_failure;
2588
2589 nlmsg_end(skb, nlh);
2590 return 0;
2591
2592 nla_put_failure:
2593 nlmsg_cancel(skb, nlh);
2594 return -EMSGSIZE;
2595 }
2596
2597 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
2598 {
2599 struct net *net = sock_net(in_skb->sk);
2600 struct rtmsg *rtm;
2601 struct nlattr *tb[RTA_MAX+1];
2602 struct rtable *rt = NULL;
2603 struct flowi4 fl4;
2604 __be32 dst = 0;
2605 __be32 src = 0;
2606 u32 iif;
2607 int err;
2608 int mark;
2609 struct sk_buff *skb;
2610 u32 table_id = RT_TABLE_MAIN;
2611 kuid_t uid;
2612
2613 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy);
2614 if (err < 0)
2615 goto errout;
2616
2617 rtm = nlmsg_data(nlh);
2618
2619 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2620 if (!skb) {
2621 err = -ENOBUFS;
2622 goto errout;
2623 }
2624
2625 /* Reserve room for dummy headers, this skb can pass
2626 through good chunk of routing engine.
2627 */
2628 skb_reset_mac_header(skb);
2629 skb_reset_network_header(skb);
2630
2631 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */
2632 ip_hdr(skb)->protocol = IPPROTO_UDP;
2633 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2634
2635 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2636 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2637 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2638 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2639 if (tb[RTA_UID])
2640 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2641 else
2642 uid = (iif ? INVALID_UID : current_uid());
2643
2644 memset(&fl4, 0, sizeof(fl4));
2645 fl4.daddr = dst;
2646 fl4.saddr = src;
2647 fl4.flowi4_tos = rtm->rtm_tos;
2648 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2649 fl4.flowi4_mark = mark;
2650 fl4.flowi4_uid = uid;
2651
2652 if (iif) {
2653 struct net_device *dev;
2654
2655 dev = __dev_get_by_index(net, iif);
2656 if (!dev) {
2657 err = -ENODEV;
2658 goto errout_free;
2659 }
2660
2661 skb->protocol = htons(ETH_P_IP);
2662 skb->dev = dev;
2663 skb->mark = mark;
2664 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2665
2666 rt = skb_rtable(skb);
2667 if (err == 0 && rt->dst.error)
2668 err = -rt->dst.error;
2669 } else {
2670 rt = ip_route_output_key(net, &fl4);
2671
2672 err = 0;
2673 if (IS_ERR(rt))
2674 err = PTR_ERR(rt);
2675 }
2676
2677 if (err)
2678 goto errout_free;
2679
2680 skb_dst_set(skb, &rt->dst);
2681 if (rtm->rtm_flags & RTM_F_NOTIFY)
2682 rt->rt_flags |= RTCF_NOTIFY;
2683
2684 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2685 table_id = rt->rt_table_id;
2686
2687 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2688 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2689 RTM_NEWROUTE);
2690 if (err < 0)
2691 goto errout_free;
2692
2693 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2694 errout:
2695 return err;
2696
2697 errout_free:
2698 kfree_skb(skb);
2699 goto errout;
2700 }
2701
2702 void ip_rt_multicast_event(struct in_device *in_dev)
2703 {
2704 rt_cache_flush(dev_net(in_dev->dev));
2705 }
2706
2707 #ifdef CONFIG_SYSCTL
2708 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2709 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2710 static int ip_rt_gc_elasticity __read_mostly = 8;
2711
2712 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2713 void __user *buffer,
2714 size_t *lenp, loff_t *ppos)
2715 {
2716 struct net *net = (struct net *)__ctl->extra1;
2717
2718 if (write) {
2719 rt_cache_flush(net);
2720 fnhe_genid_bump(net);
2721 return 0;
2722 }
2723
2724 return -EINVAL;
2725 }
2726
2727 static struct ctl_table ipv4_route_table[] = {
2728 {
2729 .procname = "gc_thresh",
2730 .data = &ipv4_dst_ops.gc_thresh,
2731 .maxlen = sizeof(int),
2732 .mode = 0644,
2733 .proc_handler = proc_dointvec,
2734 },
2735 {
2736 .procname = "max_size",
2737 .data = &ip_rt_max_size,
2738 .maxlen = sizeof(int),
2739 .mode = 0644,
2740 .proc_handler = proc_dointvec,
2741 },
2742 {
2743 /* Deprecated. Use gc_min_interval_ms */
2744
2745 .procname = "gc_min_interval",
2746 .data = &ip_rt_gc_min_interval,
2747 .maxlen = sizeof(int),
2748 .mode = 0644,
2749 .proc_handler = proc_dointvec_jiffies,
2750 },
2751 {
2752 .procname = "gc_min_interval_ms",
2753 .data = &ip_rt_gc_min_interval,
2754 .maxlen = sizeof(int),
2755 .mode = 0644,
2756 .proc_handler = proc_dointvec_ms_jiffies,
2757 },
2758 {
2759 .procname = "gc_timeout",
2760 .data = &ip_rt_gc_timeout,
2761 .maxlen = sizeof(int),
2762 .mode = 0644,
2763 .proc_handler = proc_dointvec_jiffies,
2764 },
2765 {
2766 .procname = "gc_interval",
2767 .data = &ip_rt_gc_interval,
2768 .maxlen = sizeof(int),
2769 .mode = 0644,
2770 .proc_handler = proc_dointvec_jiffies,
2771 },
2772 {
2773 .procname = "redirect_load",
2774 .data = &ip_rt_redirect_load,
2775 .maxlen = sizeof(int),
2776 .mode = 0644,
2777 .proc_handler = proc_dointvec,
2778 },
2779 {
2780 .procname = "redirect_number",
2781 .data = &ip_rt_redirect_number,
2782 .maxlen = sizeof(int),
2783 .mode = 0644,
2784 .proc_handler = proc_dointvec,
2785 },
2786 {
2787 .procname = "redirect_silence",
2788 .data = &ip_rt_redirect_silence,
2789 .maxlen = sizeof(int),
2790 .mode = 0644,
2791 .proc_handler = proc_dointvec,
2792 },
2793 {
2794 .procname = "error_cost",
2795 .data = &ip_rt_error_cost,
2796 .maxlen = sizeof(int),
2797 .mode = 0644,
2798 .proc_handler = proc_dointvec,
2799 },
2800 {
2801 .procname = "error_burst",
2802 .data = &ip_rt_error_burst,
2803 .maxlen = sizeof(int),
2804 .mode = 0644,
2805 .proc_handler = proc_dointvec,
2806 },
2807 {
2808 .procname = "gc_elasticity",
2809 .data = &ip_rt_gc_elasticity,
2810 .maxlen = sizeof(int),
2811 .mode = 0644,
2812 .proc_handler = proc_dointvec,
2813 },
2814 {
2815 .procname = "mtu_expires",
2816 .data = &ip_rt_mtu_expires,
2817 .maxlen = sizeof(int),
2818 .mode = 0644,
2819 .proc_handler = proc_dointvec_jiffies,
2820 },
2821 {
2822 .procname = "min_pmtu",
2823 .data = &ip_rt_min_pmtu,
2824 .maxlen = sizeof(int),
2825 .mode = 0644,
2826 .proc_handler = proc_dointvec,
2827 },
2828 {
2829 .procname = "min_adv_mss",
2830 .data = &ip_rt_min_advmss,
2831 .maxlen = sizeof(int),
2832 .mode = 0644,
2833 .proc_handler = proc_dointvec,
2834 },
2835 { }
2836 };
2837
2838 static struct ctl_table ipv4_route_flush_table[] = {
2839 {
2840 .procname = "flush",
2841 .maxlen = sizeof(int),
2842 .mode = 0200,
2843 .proc_handler = ipv4_sysctl_rtcache_flush,
2844 },
2845 { },
2846 };
2847
2848 static __net_init int sysctl_route_net_init(struct net *net)
2849 {
2850 struct ctl_table *tbl;
2851
2852 tbl = ipv4_route_flush_table;
2853 if (!net_eq(net, &init_net)) {
2854 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2855 if (!tbl)
2856 goto err_dup;
2857
2858 /* Don't export sysctls to unprivileged users */
2859 if (net->user_ns != &init_user_ns)
2860 tbl[0].procname = NULL;
2861 }
2862 tbl[0].extra1 = net;
2863
2864 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2865 if (!net->ipv4.route_hdr)
2866 goto err_reg;
2867 return 0;
2868
2869 err_reg:
2870 if (tbl != ipv4_route_flush_table)
2871 kfree(tbl);
2872 err_dup:
2873 return -ENOMEM;
2874 }
2875
2876 static __net_exit void sysctl_route_net_exit(struct net *net)
2877 {
2878 struct ctl_table *tbl;
2879
2880 tbl = net->ipv4.route_hdr->ctl_table_arg;
2881 unregister_net_sysctl_table(net->ipv4.route_hdr);
2882 BUG_ON(tbl == ipv4_route_flush_table);
2883 kfree(tbl);
2884 }
2885
2886 static __net_initdata struct pernet_operations sysctl_route_ops = {
2887 .init = sysctl_route_net_init,
2888 .exit = sysctl_route_net_exit,
2889 };
2890 #endif
2891
2892 static __net_init int rt_genid_init(struct net *net)
2893 {
2894 atomic_set(&net->ipv4.rt_genid, 0);
2895 atomic_set(&net->fnhe_genid, 0);
2896 get_random_bytes(&net->ipv4.dev_addr_genid,
2897 sizeof(net->ipv4.dev_addr_genid));
2898 return 0;
2899 }
2900
2901 static __net_initdata struct pernet_operations rt_genid_ops = {
2902 .init = rt_genid_init,
2903 };
2904
2905 static int __net_init ipv4_inetpeer_init(struct net *net)
2906 {
2907 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2908
2909 if (!bp)
2910 return -ENOMEM;
2911 inet_peer_base_init(bp);
2912 net->ipv4.peers = bp;
2913 return 0;
2914 }
2915
2916 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2917 {
2918 struct inet_peer_base *bp = net->ipv4.peers;
2919
2920 net->ipv4.peers = NULL;
2921 inetpeer_invalidate_tree(bp);
2922 kfree(bp);
2923 }
2924
2925 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2926 .init = ipv4_inetpeer_init,
2927 .exit = ipv4_inetpeer_exit,
2928 };
2929
2930 #ifdef CONFIG_IP_ROUTE_CLASSID
2931 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2932 #endif /* CONFIG_IP_ROUTE_CLASSID */
2933
2934 int __init ip_rt_init(void)
2935 {
2936 int rc = 0;
2937 int cpu;
2938
2939 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2940 if (!ip_idents)
2941 panic("IP: failed to allocate ip_idents\n");
2942
2943 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2944
2945 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
2946 if (!ip_tstamps)
2947 panic("IP: failed to allocate ip_tstamps\n");
2948
2949 for_each_possible_cpu(cpu) {
2950 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
2951
2952 INIT_LIST_HEAD(&ul->head);
2953 spin_lock_init(&ul->lock);
2954 }
2955 #ifdef CONFIG_IP_ROUTE_CLASSID
2956 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
2957 if (!ip_rt_acct)
2958 panic("IP: failed to allocate ip_rt_acct\n");
2959 #endif
2960
2961 ipv4_dst_ops.kmem_cachep =
2962 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
2963 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2964
2965 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
2966
2967 if (dst_entries_init(&ipv4_dst_ops) < 0)
2968 panic("IP: failed to allocate ipv4_dst_ops counter\n");
2969
2970 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
2971 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
2972
2973 ipv4_dst_ops.gc_thresh = ~0;
2974 ip_rt_max_size = INT_MAX;
2975
2976 devinet_init();
2977 ip_fib_init();
2978
2979 if (ip_rt_proc_init())
2980 pr_err("Unable to create route proc files\n");
2981 #ifdef CONFIG_XFRM
2982 xfrm_init();
2983 xfrm4_init();
2984 #endif
2985 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
2986
2987 #ifdef CONFIG_SYSCTL
2988 register_pernet_subsys(&sysctl_route_ops);
2989 #endif
2990 register_pernet_subsys(&rt_genid_ops);
2991 register_pernet_subsys(&ipv4_inetpeer_ops);
2992 return rc;
2993 }
2994
2995 #ifdef CONFIG_SYSCTL
2996 /*
2997 * We really need to sanitize the damn ipv4 init order, then all
2998 * this nonsense will go away.
2999 */
3000 void __init ip_static_sysctl_init(void)
3001 {
3002 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3003 }
3004 #endif
3005
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.