?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86/speculation/mds: Fix documentation typo
[linux/fpc-iii.git] / net / ipv4 / route.c
blob6a7e187dd0a9df355bef5de2e9565845cfdc685f
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 #include "fib_lookup.h"
118
119 #define RT_FL_TOS(oldflp4) \
120 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))
121
122 #define RT_GC_TIMEOUT (300*HZ)
123
124 static int ip_rt_max_size;
125 static int ip_rt_redirect_number __read_mostly = 9;
126 static int ip_rt_redirect_load __read_mostly = HZ / 50;
127 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1));
128 static int ip_rt_error_cost __read_mostly = HZ;
129 static int ip_rt_error_burst __read_mostly = 5 * HZ;
130 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ;
131 static u32 ip_rt_min_pmtu __read_mostly = 512 + 20 + 20;
132 static int ip_rt_min_advmss __read_mostly = 256;
133
134 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT;
135
136 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU;
137
138 /*
139 * Interface to generic destination cache.
140 */
141
142 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie);
143 static unsigned int ipv4_default_advmss(const struct dst_entry *dst);
144 static unsigned int ipv4_mtu(const 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 static void ipv4_dst_destroy(struct dst_entry *dst);
152
153 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old)
154 {
155 WARN_ON(1);
156 return NULL;
157 }
158
159 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
160 struct sk_buff *skb,
161 const void *daddr);
162 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr);
163
164 static struct dst_ops ipv4_dst_ops = {
165 .family = AF_INET,
166 .check = ipv4_dst_check,
167 .default_advmss = ipv4_default_advmss,
168 .mtu = ipv4_mtu,
169 .cow_metrics = ipv4_cow_metrics,
170 .destroy = ipv4_dst_destroy,
171 .negative_advice = ipv4_negative_advice,
172 .link_failure = ipv4_link_failure,
173 .update_pmtu = ip_rt_update_pmtu,
174 .redirect = ip_do_redirect,
175 .local_out = __ip_local_out,
176 .neigh_lookup = ipv4_neigh_lookup,
177 .confirm_neigh = ipv4_confirm_neigh,
178 };
179
180 #define ECN_OR_COST(class) TC_PRIO_##class
181
182 const __u8 ip_tos2prio[16] = {
183 TC_PRIO_BESTEFFORT,
184 ECN_OR_COST(BESTEFFORT),
185 TC_PRIO_BESTEFFORT,
186 ECN_OR_COST(BESTEFFORT),
187 TC_PRIO_BULK,
188 ECN_OR_COST(BULK),
189 TC_PRIO_BULK,
190 ECN_OR_COST(BULK),
191 TC_PRIO_INTERACTIVE,
192 ECN_OR_COST(INTERACTIVE),
193 TC_PRIO_INTERACTIVE,
194 ECN_OR_COST(INTERACTIVE),
195 TC_PRIO_INTERACTIVE_BULK,
196 ECN_OR_COST(INTERACTIVE_BULK),
197 TC_PRIO_INTERACTIVE_BULK,
198 ECN_OR_COST(INTERACTIVE_BULK)
199 };
200 EXPORT_SYMBOL(ip_tos2prio);
201
202 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat);
203 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field)
204
205 #ifdef CONFIG_PROC_FS
206 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
207 {
208 if (*pos)
209 return NULL;
210 return SEQ_START_TOKEN;
211 }
212
213 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
214 {
215 ++*pos;
216 return NULL;
217 }
218
219 static void rt_cache_seq_stop(struct seq_file *seq, void *v)
220 {
221 }
222
223 static int rt_cache_seq_show(struct seq_file *seq, void *v)
224 {
225 if (v == SEQ_START_TOKEN)
226 seq_printf(seq, "%-127s\n",
227 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t"
228 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t"
229 "HHUptod\tSpecDst");
230 return 0;
231 }
232
233 static const struct seq_operations rt_cache_seq_ops = {
234 .start = rt_cache_seq_start,
235 .next = rt_cache_seq_next,
236 .stop = rt_cache_seq_stop,
237 .show = rt_cache_seq_show,
238 };
239
240 static int rt_cache_seq_open(struct inode *inode, struct file *file)
241 {
242 return seq_open(file, &rt_cache_seq_ops);
243 }
244
245 static const struct file_operations rt_cache_seq_fops = {
246 .owner = THIS_MODULE,
247 .open = rt_cache_seq_open,
248 .read = seq_read,
249 .llseek = seq_lseek,
250 .release = seq_release,
251 };
252
253
254 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos)
255 {
256 int cpu;
257
258 if (*pos == 0)
259 return SEQ_START_TOKEN;
260
261 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) {
262 if (!cpu_possible(cpu))
263 continue;
264 *pos = cpu+1;
265 return &per_cpu(rt_cache_stat, cpu);
266 }
267 return NULL;
268 }
269
270 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
271 {
272 int cpu;
273
274 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) {
275 if (!cpu_possible(cpu))
276 continue;
277 *pos = cpu+1;
278 return &per_cpu(rt_cache_stat, cpu);
279 }
280 return NULL;
281
282 }
283
284 static void rt_cpu_seq_stop(struct seq_file *seq, void *v)
285 {
286
287 }
288
289 static int rt_cpu_seq_show(struct seq_file *seq, void *v)
290 {
291 struct rt_cache_stat *st = v;
292
293 if (v == SEQ_START_TOKEN) {
294 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");
295 return 0;
296 }
297
298 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x "
299 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n",
300 dst_entries_get_slow(&ipv4_dst_ops),
301 0, /* st->in_hit */
302 st->in_slow_tot,
303 st->in_slow_mc,
304 st->in_no_route,
305 st->in_brd,
306 st->in_martian_dst,
307 st->in_martian_src,
308
309 0, /* st->out_hit */
310 st->out_slow_tot,
311 st->out_slow_mc,
312
313 0, /* st->gc_total */
314 0, /* st->gc_ignored */
315 0, /* st->gc_goal_miss */
316 0, /* st->gc_dst_overflow */
317 0, /* st->in_hlist_search */
318 0 /* st->out_hlist_search */
319 );
320 return 0;
321 }
322
323 static const struct seq_operations rt_cpu_seq_ops = {
324 .start = rt_cpu_seq_start,
325 .next = rt_cpu_seq_next,
326 .stop = rt_cpu_seq_stop,
327 .show = rt_cpu_seq_show,
328 };
329
330
331 static int rt_cpu_seq_open(struct inode *inode, struct file *file)
332 {
333 return seq_open(file, &rt_cpu_seq_ops);
334 }
335
336 static const struct file_operations rt_cpu_seq_fops = {
337 .owner = THIS_MODULE,
338 .open = rt_cpu_seq_open,
339 .read = seq_read,
340 .llseek = seq_lseek,
341 .release = seq_release,
342 };
343
344 #ifdef CONFIG_IP_ROUTE_CLASSID
345 static int rt_acct_proc_show(struct seq_file *m, void *v)
346 {
347 struct ip_rt_acct *dst, *src;
348 unsigned int i, j;
349
350 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL);
351 if (!dst)
352 return -ENOMEM;
353
354 for_each_possible_cpu(i) {
355 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i);
356 for (j = 0; j < 256; j++) {
357 dst[j].o_bytes += src[j].o_bytes;
358 dst[j].o_packets += src[j].o_packets;
359 dst[j].i_bytes += src[j].i_bytes;
360 dst[j].i_packets += src[j].i_packets;
361 }
362 }
363
364 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct));
365 kfree(dst);
366 return 0;
367 }
368
369 static int rt_acct_proc_open(struct inode *inode, struct file *file)
370 {
371 return single_open(file, rt_acct_proc_show, NULL);
372 }
373
374 static const struct file_operations rt_acct_proc_fops = {
375 .owner = THIS_MODULE,
376 .open = rt_acct_proc_open,
377 .read = seq_read,
378 .llseek = seq_lseek,
379 .release = single_release,
380 };
381 #endif
382
383 static int __net_init ip_rt_do_proc_init(struct net *net)
384 {
385 struct proc_dir_entry *pde;
386
387 pde = proc_create("rt_cache", S_IRUGO, net->proc_net,
388 &rt_cache_seq_fops);
389 if (!pde)
390 goto err1;
391
392 pde = proc_create("rt_cache", S_IRUGO,
393 net->proc_net_stat, &rt_cpu_seq_fops);
394 if (!pde)
395 goto err2;
396
397 #ifdef CONFIG_IP_ROUTE_CLASSID
398 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops);
399 if (!pde)
400 goto err3;
401 #endif
402 return 0;
403
404 #ifdef CONFIG_IP_ROUTE_CLASSID
405 err3:
406 remove_proc_entry("rt_cache", net->proc_net_stat);
407 #endif
408 err2:
409 remove_proc_entry("rt_cache", net->proc_net);
410 err1:
411 return -ENOMEM;
412 }
413
414 static void __net_exit ip_rt_do_proc_exit(struct net *net)
415 {
416 remove_proc_entry("rt_cache", net->proc_net_stat);
417 remove_proc_entry("rt_cache", net->proc_net);
418 #ifdef CONFIG_IP_ROUTE_CLASSID
419 remove_proc_entry("rt_acct", net->proc_net);
420 #endif
421 }
422
423 static struct pernet_operations ip_rt_proc_ops __net_initdata = {
424 .init = ip_rt_do_proc_init,
425 .exit = ip_rt_do_proc_exit,
426 };
427
428 static int __init ip_rt_proc_init(void)
429 {
430 return register_pernet_subsys(&ip_rt_proc_ops);
431 }
432
433 #else
434 static inline int ip_rt_proc_init(void)
435 {
436 return 0;
437 }
438 #endif /* CONFIG_PROC_FS */
439
440 static inline bool rt_is_expired(const struct rtable *rth)
441 {
442 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev));
443 }
444
445 void rt_cache_flush(struct net *net)
446 {
447 rt_genid_bump_ipv4(net);
448 }
449
450 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst,
451 struct sk_buff *skb,
452 const void *daddr)
453 {
454 struct net_device *dev = dst->dev;
455 const __be32 *pkey = daddr;
456 const struct rtable *rt;
457 struct neighbour *n;
458
459 rt = (const struct rtable *) dst;
460 if (rt->rt_gateway)
461 pkey = (const __be32 *) &rt->rt_gateway;
462 else if (skb)
463 pkey = &ip_hdr(skb)->daddr;
464
465 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey);
466 if (n)
467 return n;
468 return neigh_create(&arp_tbl, pkey, dev);
469 }
470
471 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr)
472 {
473 struct net_device *dev = dst->dev;
474 const __be32 *pkey = daddr;
475 const struct rtable *rt;
476
477 rt = (const struct rtable *)dst;
478 if (rt->rt_gateway)
479 pkey = (const __be32 *)&rt->rt_gateway;
480 else if (!daddr ||
481 (rt->rt_flags &
482 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL)))
483 return;
484
485 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
486 }
487
488 #define IP_IDENTS_SZ 2048u
489
490 static atomic_t *ip_idents __read_mostly;
491 static u32 *ip_tstamps __read_mostly;
492
493 /* In order to protect privacy, we add a perturbation to identifiers
494 * if one generator is seldom used. This makes hard for an attacker
495 * to infer how many packets were sent between two points in time.
496 */
497 u32 ip_idents_reserve(u32 hash, int segs)
498 {
499 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
500 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
501 u32 old = ACCESS_ONCE(*p_tstamp);
502 u32 now = (u32)jiffies;
503 u32 new, delta = 0;
504
505 if (old != now && cmpxchg(p_tstamp, old, now) == old)
506 delta = prandom_u32_max(now - old);
507
508 /* Do not use atomic_add_return() as it makes UBSAN unhappy */
509 do {
510 old = (u32)atomic_read(p_id);
511 new = old + delta + segs;
512 } while (atomic_cmpxchg(p_id, old, new) != old);
513
514 return new - segs;
515 }
516 EXPORT_SYMBOL(ip_idents_reserve);
517
518 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs)
519 {
520 static u32 ip_idents_hashrnd __read_mostly;
521 u32 hash, id;
522
523 net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd));
524
525 hash = jhash_3words((__force u32)iph->daddr,
526 (__force u32)iph->saddr,
527 iph->protocol ^ net_hash_mix(net),
528 ip_idents_hashrnd);
529 id = ip_idents_reserve(hash, segs);
530 iph->id = htons(id);
531 }
532 EXPORT_SYMBOL(__ip_select_ident);
533
534 static void __build_flow_key(const struct net *net, struct flowi4 *fl4,
535 const struct sock *sk,
536 const struct iphdr *iph,
537 int oif, u8 tos,
538 u8 prot, u32 mark, int flow_flags)
539 {
540 if (sk) {
541 const struct inet_sock *inet = inet_sk(sk);
542
543 oif = sk->sk_bound_dev_if;
544 mark = sk->sk_mark;
545 tos = RT_CONN_FLAGS(sk);
546 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol;
547 }
548 flowi4_init_output(fl4, oif, mark, tos,
549 RT_SCOPE_UNIVERSE, prot,
550 flow_flags,
551 iph->daddr, iph->saddr, 0, 0,
552 sock_net_uid(net, sk));
553 }
554
555 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb,
556 const struct sock *sk)
557 {
558 const struct net *net = dev_net(skb->dev);
559 const struct iphdr *iph = ip_hdr(skb);
560 int oif = skb->dev->ifindex;
561 u8 tos = RT_TOS(iph->tos);
562 u8 prot = iph->protocol;
563 u32 mark = skb->mark;
564
565 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0);
566 }
567
568 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk)
569 {
570 const struct inet_sock *inet = inet_sk(sk);
571 const struct ip_options_rcu *inet_opt;
572 __be32 daddr = inet->inet_daddr;
573
574 rcu_read_lock();
575 inet_opt = rcu_dereference(inet->inet_opt);
576 if (inet_opt && inet_opt->opt.srr)
577 daddr = inet_opt->opt.faddr;
578 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
579 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
580 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
581 inet_sk_flowi_flags(sk),
582 daddr, inet->inet_saddr, 0, 0, sk->sk_uid);
583 rcu_read_unlock();
584 }
585
586 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk,
587 const struct sk_buff *skb)
588 {
589 if (skb)
590 build_skb_flow_key(fl4, skb, sk);
591 else
592 build_sk_flow_key(fl4, sk);
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 dst_dev_put(&rt->dst);
605 dst_release(&rt->dst);
606 }
607 rt = rcu_dereference(fnhe->fnhe_rth_output);
608 if (rt) {
609 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL);
610 dst_dev_put(&rt->dst);
611 dst_release(&rt->dst);
612 }
613 }
614
615 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash)
616 {
617 struct fib_nh_exception *fnhe, *oldest;
618
619 oldest = rcu_dereference(hash->chain);
620 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe;
621 fnhe = rcu_dereference(fnhe->fnhe_next)) {
622 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp))
623 oldest = fnhe;
624 }
625 fnhe_flush_routes(oldest);
626 return oldest;
627 }
628
629 static inline u32 fnhe_hashfun(__be32 daddr)
630 {
631 static u32 fnhe_hashrnd __read_mostly;
632 u32 hval;
633
634 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd));
635 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd);
636 return hash_32(hval, FNHE_HASH_SHIFT);
637 }
638
639 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe)
640 {
641 rt->rt_pmtu = fnhe->fnhe_pmtu;
642 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked;
643 rt->dst.expires = fnhe->fnhe_expires;
644
645 if (fnhe->fnhe_gw) {
646 rt->rt_flags |= RTCF_REDIRECTED;
647 rt->rt_gateway = fnhe->fnhe_gw;
648 rt->rt_uses_gateway = 1;
649 }
650 }
651
652 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw,
653 u32 pmtu, bool lock, unsigned long expires)
654 {
655 struct fnhe_hash_bucket *hash;
656 struct fib_nh_exception *fnhe;
657 struct rtable *rt;
658 u32 genid, hval;
659 unsigned int i;
660 int depth;
661
662 genid = fnhe_genid(dev_net(nh->nh_dev));
663 hval = fnhe_hashfun(daddr);
664
665 spin_lock_bh(&fnhe_lock);
666
667 hash = rcu_dereference(nh->nh_exceptions);
668 if (!hash) {
669 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC);
670 if (!hash)
671 goto out_unlock;
672 rcu_assign_pointer(nh->nh_exceptions, hash);
673 }
674
675 hash += hval;
676
677 depth = 0;
678 for (fnhe = rcu_dereference(hash->chain); fnhe;
679 fnhe = rcu_dereference(fnhe->fnhe_next)) {
680 if (fnhe->fnhe_daddr == daddr)
681 break;
682 depth++;
683 }
684
685 if (fnhe) {
686 if (fnhe->fnhe_genid != genid)
687 fnhe->fnhe_genid = genid;
688 if (gw)
689 fnhe->fnhe_gw = gw;
690 if (pmtu) {
691 fnhe->fnhe_pmtu = pmtu;
692 fnhe->fnhe_mtu_locked = lock;
693 }
694 fnhe->fnhe_expires = max(1UL, expires);
695 /* Update all cached dsts too */
696 rt = rcu_dereference(fnhe->fnhe_rth_input);
697 if (rt)
698 fill_route_from_fnhe(rt, fnhe);
699 rt = rcu_dereference(fnhe->fnhe_rth_output);
700 if (rt)
701 fill_route_from_fnhe(rt, fnhe);
702 } else {
703 if (depth > FNHE_RECLAIM_DEPTH)
704 fnhe = fnhe_oldest(hash);
705 else {
706 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC);
707 if (!fnhe)
708 goto out_unlock;
709
710 fnhe->fnhe_next = hash->chain;
711 rcu_assign_pointer(hash->chain, fnhe);
712 }
713 fnhe->fnhe_genid = genid;
714 fnhe->fnhe_daddr = daddr;
715 fnhe->fnhe_gw = gw;
716 fnhe->fnhe_pmtu = pmtu;
717 fnhe->fnhe_mtu_locked = lock;
718 fnhe->fnhe_expires = max(1UL, expires);
719
720 /* Exception created; mark the cached routes for the nexthop
721 * stale, so anyone caching it rechecks if this exception
722 * applies to them.
723 */
724 rt = rcu_dereference(nh->nh_rth_input);
725 if (rt)
726 rt->dst.obsolete = DST_OBSOLETE_KILL;
727
728 for_each_possible_cpu(i) {
729 struct rtable __rcu **prt;
730 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i);
731 rt = rcu_dereference(*prt);
732 if (rt)
733 rt->dst.obsolete = DST_OBSOLETE_KILL;
734 }
735 }
736
737 fnhe->fnhe_stamp = jiffies;
738
739 out_unlock:
740 spin_unlock_bh(&fnhe_lock);
741 }
742
743 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4,
744 bool kill_route)
745 {
746 __be32 new_gw = icmp_hdr(skb)->un.gateway;
747 __be32 old_gw = ip_hdr(skb)->saddr;
748 struct net_device *dev = skb->dev;
749 struct in_device *in_dev;
750 struct fib_result res;
751 struct neighbour *n;
752 struct net *net;
753
754 switch (icmp_hdr(skb)->code & 7) {
755 case ICMP_REDIR_NET:
756 case ICMP_REDIR_NETTOS:
757 case ICMP_REDIR_HOST:
758 case ICMP_REDIR_HOSTTOS:
759 break;
760
761 default:
762 return;
763 }
764
765 if (rt->rt_gateway != old_gw)
766 return;
767
768 in_dev = __in_dev_get_rcu(dev);
769 if (!in_dev)
770 return;
771
772 net = dev_net(dev);
773 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) ||
774 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) ||
775 ipv4_is_zeronet(new_gw))
776 goto reject_redirect;
777
778 if (!IN_DEV_SHARED_MEDIA(in_dev)) {
779 if (!inet_addr_onlink(in_dev, new_gw, old_gw))
780 goto reject_redirect;
781 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev))
782 goto reject_redirect;
783 } else {
784 if (inet_addr_type(net, new_gw) != RTN_UNICAST)
785 goto reject_redirect;
786 }
787
788 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw);
789 if (!n)
790 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev);
791 if (!IS_ERR(n)) {
792 if (!(n->nud_state & NUD_VALID)) {
793 neigh_event_send(n, NULL);
794 } else {
795 if (fib_lookup(net, fl4, &res, 0) == 0) {
796 struct fib_nh *nh = &FIB_RES_NH(res);
797
798 update_or_create_fnhe(nh, fl4->daddr, new_gw,
799 0, false,
800 jiffies + ip_rt_gc_timeout);
801 }
802 if (kill_route)
803 rt->dst.obsolete = DST_OBSOLETE_KILL;
804 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n);
805 }
806 neigh_release(n);
807 }
808 return;
809
810 reject_redirect:
811 #ifdef CONFIG_IP_ROUTE_VERBOSE
812 if (IN_DEV_LOG_MARTIANS(in_dev)) {
813 const struct iphdr *iph = (const struct iphdr *) skb->data;
814 __be32 daddr = iph->daddr;
815 __be32 saddr = iph->saddr;
816
817 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n"
818 " Advised path = %pI4 -> %pI4\n",
819 &old_gw, dev->name, &new_gw,
820 &saddr, &daddr);
821 }
822 #endif
823 ;
824 }
825
826 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
827 {
828 struct rtable *rt;
829 struct flowi4 fl4;
830 const struct iphdr *iph = (const struct iphdr *) skb->data;
831 struct net *net = dev_net(skb->dev);
832 int oif = skb->dev->ifindex;
833 u8 tos = RT_TOS(iph->tos);
834 u8 prot = iph->protocol;
835 u32 mark = skb->mark;
836
837 rt = (struct rtable *) dst;
838
839 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0);
840 __ip_do_redirect(rt, skb, &fl4, true);
841 }
842
843 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
844 {
845 struct rtable *rt = (struct rtable *)dst;
846 struct dst_entry *ret = dst;
847
848 if (rt) {
849 if (dst->obsolete > 0) {
850 ip_rt_put(rt);
851 ret = NULL;
852 } else if ((rt->rt_flags & RTCF_REDIRECTED) ||
853 rt->dst.expires) {
854 ip_rt_put(rt);
855 ret = NULL;
856 }
857 }
858 return ret;
859 }
860
861 /*
862 * Algorithm:
863 * 1. The first ip_rt_redirect_number redirects are sent
864 * with exponential backoff, then we stop sending them at all,
865 * assuming that the host ignores our redirects.
866 * 2. If we did not see packets requiring redirects
867 * during ip_rt_redirect_silence, we assume that the host
868 * forgot redirected route and start to send redirects again.
869 *
870 * This algorithm is much cheaper and more intelligent than dumb load limiting
871 * in icmp.c.
872 *
873 * NOTE. Do not forget to inhibit load limiting for redirects (redundant)
874 * and "frag. need" (breaks PMTU discovery) in icmp.c.
875 */
876
877 void ip_rt_send_redirect(struct sk_buff *skb)
878 {
879 struct rtable *rt = skb_rtable(skb);
880 struct in_device *in_dev;
881 struct inet_peer *peer;
882 struct net *net;
883 int log_martians;
884 int vif;
885
886 rcu_read_lock();
887 in_dev = __in_dev_get_rcu(rt->dst.dev);
888 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
889 rcu_read_unlock();
890 return;
891 }
892 log_martians = IN_DEV_LOG_MARTIANS(in_dev);
893 vif = l3mdev_master_ifindex_rcu(rt->dst.dev);
894 rcu_read_unlock();
895
896 net = dev_net(rt->dst.dev);
897 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1);
898 if (!peer) {
899 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST,
900 rt_nexthop(rt, ip_hdr(skb)->daddr));
901 return;
902 }
903
904 /* No redirected packets during ip_rt_redirect_silence;
905 * reset the algorithm.
906 */
907 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) {
908 peer->rate_tokens = 0;
909 peer->n_redirects = 0;
910 }
911
912 /* Too many ignored redirects; do not send anything
913 * set dst.rate_last to the last seen redirected packet.
914 */
915 if (peer->n_redirects >= ip_rt_redirect_number) {
916 peer->rate_last = jiffies;
917 goto out_put_peer;
918 }
919
920 /* Check for load limit; set rate_last to the latest sent
921 * redirect.
922 */
923 if (peer->rate_tokens == 0 ||
924 time_after(jiffies,
925 (peer->rate_last +
926 (ip_rt_redirect_load << peer->rate_tokens)))) {
927 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
928
929 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
930 peer->rate_last = jiffies;
931 ++peer->rate_tokens;
932 ++peer->n_redirects;
933 #ifdef CONFIG_IP_ROUTE_VERBOSE
934 if (log_martians &&
935 peer->rate_tokens == ip_rt_redirect_number)
936 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
937 &ip_hdr(skb)->saddr, inet_iif(skb),
938 &ip_hdr(skb)->daddr, &gw);
939 #endif
940 }
941 out_put_peer:
942 inet_putpeer(peer);
943 }
944
945 static int ip_error(struct sk_buff *skb)
946 {
947 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
948 struct rtable *rt = skb_rtable(skb);
949 struct inet_peer *peer;
950 unsigned long now;
951 struct net *net;
952 bool send;
953 int code;
954
955 /* IP on this device is disabled. */
956 if (!in_dev)
957 goto out;
958
959 net = dev_net(rt->dst.dev);
960 if (!IN_DEV_FORWARD(in_dev)) {
961 switch (rt->dst.error) {
962 case EHOSTUNREACH:
963 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
964 break;
965
966 case ENETUNREACH:
967 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
968 break;
969 }
970 goto out;
971 }
972
973 switch (rt->dst.error) {
974 case EINVAL:
975 default:
976 goto out;
977 case EHOSTUNREACH:
978 code = ICMP_HOST_UNREACH;
979 break;
980 case ENETUNREACH:
981 code = ICMP_NET_UNREACH;
982 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
983 break;
984 case EACCES:
985 code = ICMP_PKT_FILTERED;
986 break;
987 }
988
989 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
990 l3mdev_master_ifindex(skb->dev), 1);
991
992 send = true;
993 if (peer) {
994 now = jiffies;
995 peer->rate_tokens += now - peer->rate_last;
996 if (peer->rate_tokens > ip_rt_error_burst)
997 peer->rate_tokens = ip_rt_error_burst;
998 peer->rate_last = now;
999 if (peer->rate_tokens >= ip_rt_error_cost)
1000 peer->rate_tokens -= ip_rt_error_cost;
1001 else
1002 send = false;
1003 inet_putpeer(peer);
1004 }
1005 if (send)
1006 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1007
1008 out: kfree_skb(skb);
1009 return 0;
1010 }
1011
1012 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1013 {
1014 struct dst_entry *dst = &rt->dst;
1015 struct fib_result res;
1016 bool lock = false;
1017
1018 if (ip_mtu_locked(dst))
1019 return;
1020
1021 if (ipv4_mtu(dst) < mtu)
1022 return;
1023
1024 if (mtu < ip_rt_min_pmtu) {
1025 lock = true;
1026 mtu = ip_rt_min_pmtu;
1027 }
1028
1029 if (rt->rt_pmtu == mtu &&
1030 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1031 return;
1032
1033 rcu_read_lock();
1034 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1035 struct fib_nh *nh = &FIB_RES_NH(res);
1036
1037 update_or_create_fnhe(nh, fl4->daddr, 0, mtu, lock,
1038 jiffies + ip_rt_mtu_expires);
1039 }
1040 rcu_read_unlock();
1041 }
1042
1043 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1044 struct sk_buff *skb, u32 mtu)
1045 {
1046 struct rtable *rt = (struct rtable *) dst;
1047 struct flowi4 fl4;
1048
1049 ip_rt_build_flow_key(&fl4, sk, skb);
1050 __ip_rt_update_pmtu(rt, &fl4, mtu);
1051 }
1052
1053 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1054 int oif, u32 mark, u8 protocol, int flow_flags)
1055 {
1056 const struct iphdr *iph = (const struct iphdr *) skb->data;
1057 struct flowi4 fl4;
1058 struct rtable *rt;
1059
1060 if (!mark)
1061 mark = IP4_REPLY_MARK(net, skb->mark);
1062
1063 __build_flow_key(net, &fl4, NULL, iph, oif,
1064 RT_TOS(iph->tos), protocol, mark, flow_flags);
1065 rt = __ip_route_output_key(net, &fl4);
1066 if (!IS_ERR(rt)) {
1067 __ip_rt_update_pmtu(rt, &fl4, mtu);
1068 ip_rt_put(rt);
1069 }
1070 }
1071 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1072
1073 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1074 {
1075 const struct iphdr *iph = (const struct iphdr *) skb->data;
1076 struct flowi4 fl4;
1077 struct rtable *rt;
1078
1079 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1080
1081 if (!fl4.flowi4_mark)
1082 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1083
1084 rt = __ip_route_output_key(sock_net(sk), &fl4);
1085 if (!IS_ERR(rt)) {
1086 __ip_rt_update_pmtu(rt, &fl4, mtu);
1087 ip_rt_put(rt);
1088 }
1089 }
1090
1091 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1092 {
1093 const struct iphdr *iph = (const struct iphdr *) skb->data;
1094 struct flowi4 fl4;
1095 struct rtable *rt;
1096 struct dst_entry *odst = NULL;
1097 bool new = false;
1098 struct net *net = sock_net(sk);
1099
1100 bh_lock_sock(sk);
1101
1102 if (!ip_sk_accept_pmtu(sk))
1103 goto out;
1104
1105 odst = sk_dst_get(sk);
1106
1107 if (sock_owned_by_user(sk) || !odst) {
1108 __ipv4_sk_update_pmtu(skb, sk, mtu);
1109 goto out;
1110 }
1111
1112 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1113
1114 rt = (struct rtable *)odst;
1115 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1116 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1117 if (IS_ERR(rt))
1118 goto out;
1119
1120 new = true;
1121 }
1122
1123 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1124
1125 if (!dst_check(&rt->dst, 0)) {
1126 if (new)
1127 dst_release(&rt->dst);
1128
1129 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1130 if (IS_ERR(rt))
1131 goto out;
1132
1133 new = true;
1134 }
1135
1136 if (new)
1137 sk_dst_set(sk, &rt->dst);
1138
1139 out:
1140 bh_unlock_sock(sk);
1141 dst_release(odst);
1142 }
1143 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1144
1145 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1146 int oif, u32 mark, u8 protocol, int flow_flags)
1147 {
1148 const struct iphdr *iph = (const struct iphdr *) skb->data;
1149 struct flowi4 fl4;
1150 struct rtable *rt;
1151
1152 __build_flow_key(net, &fl4, NULL, iph, oif,
1153 RT_TOS(iph->tos), protocol, mark, flow_flags);
1154 rt = __ip_route_output_key(net, &fl4);
1155 if (!IS_ERR(rt)) {
1156 __ip_do_redirect(rt, skb, &fl4, false);
1157 ip_rt_put(rt);
1158 }
1159 }
1160 EXPORT_SYMBOL_GPL(ipv4_redirect);
1161
1162 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1163 {
1164 const struct iphdr *iph = (const struct iphdr *) skb->data;
1165 struct flowi4 fl4;
1166 struct rtable *rt;
1167 struct net *net = sock_net(sk);
1168
1169 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1170 rt = __ip_route_output_key(net, &fl4);
1171 if (!IS_ERR(rt)) {
1172 __ip_do_redirect(rt, skb, &fl4, false);
1173 ip_rt_put(rt);
1174 }
1175 }
1176 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1177
1178 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1179 {
1180 struct rtable *rt = (struct rtable *) dst;
1181
1182 /* All IPV4 dsts are created with ->obsolete set to the value
1183 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1184 * into this function always.
1185 *
1186 * When a PMTU/redirect information update invalidates a route,
1187 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1188 * DST_OBSOLETE_DEAD by dst_free().
1189 */
1190 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1191 return NULL;
1192 return dst;
1193 }
1194
1195 static void ipv4_send_dest_unreach(struct sk_buff *skb)
1196 {
1197 struct ip_options opt;
1198 int res;
1199
1200 /* Recompile ip options since IPCB may not be valid anymore.
1201 * Also check we have a reasonable ipv4 header.
1202 */
1203 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) ||
1204 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5)
1205 return;
1206
1207 memset(&opt, 0, sizeof(opt));
1208 if (ip_hdr(skb)->ihl > 5) {
1209 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4))
1210 return;
1211 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr);
1212
1213 rcu_read_lock();
1214 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL);
1215 rcu_read_unlock();
1216
1217 if (res)
1218 return;
1219 }
1220 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt);
1221 }
1222
1223 static void ipv4_link_failure(struct sk_buff *skb)
1224 {
1225 struct rtable *rt;
1226
1227 ipv4_send_dest_unreach(skb);
1228
1229 rt = skb_rtable(skb);
1230 if (rt)
1231 dst_set_expires(&rt->dst, 0);
1232 }
1233
1234 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1235 {
1236 pr_debug("%s: %pI4 -> %pI4, %s\n",
1237 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1238 skb->dev ? skb->dev->name : "?");
1239 kfree_skb(skb);
1240 WARN_ON(1);
1241 return 0;
1242 }
1243
1244 /*
1245 We do not cache source address of outgoing interface,
1246 because it is used only by IP RR, TS and SRR options,
1247 so that it out of fast path.
1248
1249 BTW remember: "addr" is allowed to be not aligned
1250 in IP options!
1251 */
1252
1253 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1254 {
1255 __be32 src;
1256
1257 if (rt_is_output_route(rt))
1258 src = ip_hdr(skb)->saddr;
1259 else {
1260 struct fib_result res;
1261 struct flowi4 fl4;
1262 struct iphdr *iph;
1263
1264 iph = ip_hdr(skb);
1265
1266 memset(&fl4, 0, sizeof(fl4));
1267 fl4.daddr = iph->daddr;
1268 fl4.saddr = iph->saddr;
1269 fl4.flowi4_tos = RT_TOS(iph->tos);
1270 fl4.flowi4_oif = rt->dst.dev->ifindex;
1271 fl4.flowi4_iif = skb->dev->ifindex;
1272 fl4.flowi4_mark = skb->mark;
1273
1274 rcu_read_lock();
1275 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1276 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1277 else
1278 src = inet_select_addr(rt->dst.dev,
1279 rt_nexthop(rt, iph->daddr),
1280 RT_SCOPE_UNIVERSE);
1281 rcu_read_unlock();
1282 }
1283 memcpy(addr, &src, 4);
1284 }
1285
1286 #ifdef CONFIG_IP_ROUTE_CLASSID
1287 static void set_class_tag(struct rtable *rt, u32 tag)
1288 {
1289 if (!(rt->dst.tclassid & 0xFFFF))
1290 rt->dst.tclassid |= tag & 0xFFFF;
1291 if (!(rt->dst.tclassid & 0xFFFF0000))
1292 rt->dst.tclassid |= tag & 0xFFFF0000;
1293 }
1294 #endif
1295
1296 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1297 {
1298 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1299 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1300 ip_rt_min_advmss);
1301
1302 return min(advmss, IPV4_MAX_PMTU - header_size);
1303 }
1304
1305 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1306 {
1307 const struct rtable *rt = (const struct rtable *) dst;
1308 unsigned int mtu = rt->rt_pmtu;
1309
1310 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1311 mtu = dst_metric_raw(dst, RTAX_MTU);
1312
1313 if (mtu)
1314 return mtu;
1315
1316 mtu = READ_ONCE(dst->dev->mtu);
1317
1318 if (unlikely(ip_mtu_locked(dst))) {
1319 if (rt->rt_uses_gateway && mtu > 576)
1320 mtu = 576;
1321 }
1322
1323 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1324
1325 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1326 }
1327
1328 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1329 {
1330 struct fnhe_hash_bucket *hash;
1331 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1332 u32 hval = fnhe_hashfun(daddr);
1333
1334 spin_lock_bh(&fnhe_lock);
1335
1336 hash = rcu_dereference_protected(nh->nh_exceptions,
1337 lockdep_is_held(&fnhe_lock));
1338 hash += hval;
1339
1340 fnhe_p = &hash->chain;
1341 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1342 while (fnhe) {
1343 if (fnhe->fnhe_daddr == daddr) {
1344 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1345 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1346 /* set fnhe_daddr to 0 to ensure it won't bind with
1347 * new dsts in rt_bind_exception().
1348 */
1349 fnhe->fnhe_daddr = 0;
1350 fnhe_flush_routes(fnhe);
1351 kfree_rcu(fnhe, rcu);
1352 break;
1353 }
1354 fnhe_p = &fnhe->fnhe_next;
1355 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1356 lockdep_is_held(&fnhe_lock));
1357 }
1358
1359 spin_unlock_bh(&fnhe_lock);
1360 }
1361
1362 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1363 {
1364 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1365 struct fib_nh_exception *fnhe;
1366 u32 hval;
1367
1368 if (!hash)
1369 return NULL;
1370
1371 hval = fnhe_hashfun(daddr);
1372
1373 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1374 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1375 if (fnhe->fnhe_daddr == daddr) {
1376 if (fnhe->fnhe_expires &&
1377 time_after(jiffies, fnhe->fnhe_expires)) {
1378 ip_del_fnhe(nh, daddr);
1379 break;
1380 }
1381 return fnhe;
1382 }
1383 }
1384 return NULL;
1385 }
1386
1387 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1388 __be32 daddr, const bool do_cache)
1389 {
1390 bool ret = false;
1391
1392 spin_lock_bh(&fnhe_lock);
1393
1394 if (daddr == fnhe->fnhe_daddr) {
1395 struct rtable __rcu **porig;
1396 struct rtable *orig;
1397 int genid = fnhe_genid(dev_net(rt->dst.dev));
1398
1399 if (rt_is_input_route(rt))
1400 porig = &fnhe->fnhe_rth_input;
1401 else
1402 porig = &fnhe->fnhe_rth_output;
1403 orig = rcu_dereference(*porig);
1404
1405 if (fnhe->fnhe_genid != genid) {
1406 fnhe->fnhe_genid = genid;
1407 fnhe->fnhe_gw = 0;
1408 fnhe->fnhe_pmtu = 0;
1409 fnhe->fnhe_expires = 0;
1410 fnhe_flush_routes(fnhe);
1411 orig = NULL;
1412 }
1413 fill_route_from_fnhe(rt, fnhe);
1414 if (!rt->rt_gateway)
1415 rt->rt_gateway = daddr;
1416
1417 if (do_cache) {
1418 dst_hold(&rt->dst);
1419 rcu_assign_pointer(*porig, rt);
1420 if (orig) {
1421 dst_dev_put(&orig->dst);
1422 dst_release(&orig->dst);
1423 }
1424 ret = true;
1425 }
1426
1427 fnhe->fnhe_stamp = jiffies;
1428 }
1429 spin_unlock_bh(&fnhe_lock);
1430
1431 return ret;
1432 }
1433
1434 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1435 {
1436 struct rtable *orig, *prev, **p;
1437 bool ret = true;
1438
1439 if (rt_is_input_route(rt)) {
1440 p = (struct rtable **)&nh->nh_rth_input;
1441 } else {
1442 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1443 }
1444 orig = *p;
1445
1446 /* hold dst before doing cmpxchg() to avoid race condition
1447 * on this dst
1448 */
1449 dst_hold(&rt->dst);
1450 prev = cmpxchg(p, orig, rt);
1451 if (prev == orig) {
1452 if (orig) {
1453 dst_dev_put(&orig->dst);
1454 dst_release(&orig->dst);
1455 }
1456 } else {
1457 dst_release(&rt->dst);
1458 ret = false;
1459 }
1460
1461 return ret;
1462 }
1463
1464 struct uncached_list {
1465 spinlock_t lock;
1466 struct list_head head;
1467 };
1468
1469 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1470
1471 static void rt_add_uncached_list(struct rtable *rt)
1472 {
1473 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1474
1475 rt->rt_uncached_list = ul;
1476
1477 spin_lock_bh(&ul->lock);
1478 list_add_tail(&rt->rt_uncached, &ul->head);
1479 spin_unlock_bh(&ul->lock);
1480 }
1481
1482 static void ipv4_dst_destroy(struct dst_entry *dst)
1483 {
1484 struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst);
1485 struct rtable *rt = (struct rtable *) dst;
1486
1487 if (p != &dst_default_metrics && refcount_dec_and_test(&p->refcnt))
1488 kfree(p);
1489
1490 if (!list_empty(&rt->rt_uncached)) {
1491 struct uncached_list *ul = rt->rt_uncached_list;
1492
1493 spin_lock_bh(&ul->lock);
1494 list_del(&rt->rt_uncached);
1495 spin_unlock_bh(&ul->lock);
1496 }
1497 }
1498
1499 void rt_flush_dev(struct net_device *dev)
1500 {
1501 struct net *net = dev_net(dev);
1502 struct rtable *rt;
1503 int cpu;
1504
1505 for_each_possible_cpu(cpu) {
1506 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1507
1508 spin_lock_bh(&ul->lock);
1509 list_for_each_entry(rt, &ul->head, rt_uncached) {
1510 if (rt->dst.dev != dev)
1511 continue;
1512 rt->dst.dev = net->loopback_dev;
1513 dev_hold(rt->dst.dev);
1514 dev_put(dev);
1515 }
1516 spin_unlock_bh(&ul->lock);
1517 }
1518 }
1519
1520 static bool rt_cache_valid(const struct rtable *rt)
1521 {
1522 return rt &&
1523 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1524 !rt_is_expired(rt);
1525 }
1526
1527 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1528 const struct fib_result *res,
1529 struct fib_nh_exception *fnhe,
1530 struct fib_info *fi, u16 type, u32 itag,
1531 const bool do_cache)
1532 {
1533 bool cached = false;
1534
1535 if (fi) {
1536 struct fib_nh *nh = &FIB_RES_NH(*res);
1537
1538 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1539 rt->rt_gateway = nh->nh_gw;
1540 rt->rt_uses_gateway = 1;
1541 }
1542 dst_init_metrics(&rt->dst, fi->fib_metrics->metrics, true);
1543 if (fi->fib_metrics != &dst_default_metrics) {
1544 rt->dst._metrics |= DST_METRICS_REFCOUNTED;
1545 refcount_inc(&fi->fib_metrics->refcnt);
1546 }
1547 #ifdef CONFIG_IP_ROUTE_CLASSID
1548 rt->dst.tclassid = nh->nh_tclassid;
1549 #endif
1550 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1551 if (unlikely(fnhe))
1552 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1553 else if (do_cache)
1554 cached = rt_cache_route(nh, rt);
1555 if (unlikely(!cached)) {
1556 /* Routes we intend to cache in nexthop exception or
1557 * FIB nexthop have the DST_NOCACHE bit clear.
1558 * However, if we are unsuccessful at storing this
1559 * route into the cache we really need to set it.
1560 */
1561 if (!rt->rt_gateway)
1562 rt->rt_gateway = daddr;
1563 rt_add_uncached_list(rt);
1564 }
1565 } else
1566 rt_add_uncached_list(rt);
1567
1568 #ifdef CONFIG_IP_ROUTE_CLASSID
1569 #ifdef CONFIG_IP_MULTIPLE_TABLES
1570 set_class_tag(rt, res->tclassid);
1571 #endif
1572 set_class_tag(rt, itag);
1573 #endif
1574 }
1575
1576 struct rtable *rt_dst_alloc(struct net_device *dev,
1577 unsigned int flags, u16 type,
1578 bool nopolicy, bool noxfrm, bool will_cache)
1579 {
1580 struct rtable *rt;
1581
1582 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1583 (will_cache ? 0 : DST_HOST) |
1584 (nopolicy ? DST_NOPOLICY : 0) |
1585 (noxfrm ? DST_NOXFRM : 0));
1586
1587 if (rt) {
1588 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1589 rt->rt_flags = flags;
1590 rt->rt_type = type;
1591 rt->rt_is_input = 0;
1592 rt->rt_iif = 0;
1593 rt->rt_pmtu = 0;
1594 rt->rt_mtu_locked = 0;
1595 rt->rt_gateway = 0;
1596 rt->rt_uses_gateway = 0;
1597 rt->rt_table_id = 0;
1598 INIT_LIST_HEAD(&rt->rt_uncached);
1599
1600 rt->dst.output = ip_output;
1601 if (flags & RTCF_LOCAL)
1602 rt->dst.input = ip_local_deliver;
1603 }
1604
1605 return rt;
1606 }
1607 EXPORT_SYMBOL(rt_dst_alloc);
1608
1609 /* called in rcu_read_lock() section */
1610 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1611 u8 tos, struct net_device *dev,
1612 struct in_device *in_dev, u32 *itag)
1613 {
1614 int err;
1615
1616 /* Primary sanity checks. */
1617 if (!in_dev)
1618 return -EINVAL;
1619
1620 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1621 skb->protocol != htons(ETH_P_IP))
1622 return -EINVAL;
1623
1624 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1625 return -EINVAL;
1626
1627 if (ipv4_is_zeronet(saddr)) {
1628 if (!ipv4_is_local_multicast(daddr))
1629 return -EINVAL;
1630 } else {
1631 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1632 in_dev, itag);
1633 if (err < 0)
1634 return err;
1635 }
1636 return 0;
1637 }
1638
1639 /* called in rcu_read_lock() section */
1640 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1641 u8 tos, struct net_device *dev, int our)
1642 {
1643 struct in_device *in_dev = __in_dev_get_rcu(dev);
1644 unsigned int flags = RTCF_MULTICAST;
1645 struct rtable *rth;
1646 u32 itag = 0;
1647 int err;
1648
1649 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1650 if (err)
1651 return err;
1652
1653 if (our)
1654 flags |= RTCF_LOCAL;
1655
1656 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1657 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1658 if (!rth)
1659 return -ENOBUFS;
1660
1661 #ifdef CONFIG_IP_ROUTE_CLASSID
1662 rth->dst.tclassid = itag;
1663 #endif
1664 rth->dst.output = ip_rt_bug;
1665 rth->rt_is_input= 1;
1666
1667 #ifdef CONFIG_IP_MROUTE
1668 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1669 rth->dst.input = ip_mr_input;
1670 #endif
1671 RT_CACHE_STAT_INC(in_slow_mc);
1672
1673 skb_dst_set(skb, &rth->dst);
1674 return 0;
1675 }
1676
1677
1678 static void ip_handle_martian_source(struct net_device *dev,
1679 struct in_device *in_dev,
1680 struct sk_buff *skb,
1681 __be32 daddr,
1682 __be32 saddr)
1683 {
1684 RT_CACHE_STAT_INC(in_martian_src);
1685 #ifdef CONFIG_IP_ROUTE_VERBOSE
1686 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1687 /*
1688 * RFC1812 recommendation, if source is martian,
1689 * the only hint is MAC header.
1690 */
1691 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1692 &daddr, &saddr, dev->name);
1693 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1694 print_hex_dump(KERN_WARNING, "ll header: ",
1695 DUMP_PREFIX_OFFSET, 16, 1,
1696 skb_mac_header(skb),
1697 dev->hard_header_len, true);
1698 }
1699 }
1700 #endif
1701 }
1702
1703 static void set_lwt_redirect(struct rtable *rth)
1704 {
1705 if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1706 rth->dst.lwtstate->orig_output = rth->dst.output;
1707 rth->dst.output = lwtunnel_output;
1708 }
1709
1710 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1711 rth->dst.lwtstate->orig_input = rth->dst.input;
1712 rth->dst.input = lwtunnel_input;
1713 }
1714 }
1715
1716 /* called in rcu_read_lock() section */
1717 static int __mkroute_input(struct sk_buff *skb,
1718 const struct fib_result *res,
1719 struct in_device *in_dev,
1720 __be32 daddr, __be32 saddr, u32 tos)
1721 {
1722 struct fib_nh_exception *fnhe;
1723 struct rtable *rth;
1724 int err;
1725 struct in_device *out_dev;
1726 bool do_cache;
1727 u32 itag = 0;
1728
1729 /* get a working reference to the output device */
1730 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1731 if (!out_dev) {
1732 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1733 return -EINVAL;
1734 }
1735
1736 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1737 in_dev->dev, in_dev, &itag);
1738 if (err < 0) {
1739 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1740 saddr);
1741
1742 goto cleanup;
1743 }
1744
1745 do_cache = res->fi && !itag;
1746 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1747 skb->protocol == htons(ETH_P_IP) &&
1748 (IN_DEV_SHARED_MEDIA(out_dev) ||
1749 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1750 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1751
1752 if (skb->protocol != htons(ETH_P_IP)) {
1753 /* Not IP (i.e. ARP). Do not create route, if it is
1754 * invalid for proxy arp. DNAT routes are always valid.
1755 *
1756 * Proxy arp feature have been extended to allow, ARP
1757 * replies back to the same interface, to support
1758 * Private VLAN switch technologies. See arp.c.
1759 */
1760 if (out_dev == in_dev &&
1761 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1762 err = -EINVAL;
1763 goto cleanup;
1764 }
1765 }
1766
1767 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1768 if (do_cache) {
1769 if (fnhe)
1770 rth = rcu_dereference(fnhe->fnhe_rth_input);
1771 else
1772 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1773 if (rt_cache_valid(rth)) {
1774 skb_dst_set_noref(skb, &rth->dst);
1775 goto out;
1776 }
1777 }
1778
1779 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1780 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1781 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1782 if (!rth) {
1783 err = -ENOBUFS;
1784 goto cleanup;
1785 }
1786
1787 rth->rt_is_input = 1;
1788 if (res->table)
1789 rth->rt_table_id = res->table->tb_id;
1790 RT_CACHE_STAT_INC(in_slow_tot);
1791
1792 rth->dst.input = ip_forward;
1793
1794 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1795 do_cache);
1796 set_lwt_redirect(rth);
1797 skb_dst_set(skb, &rth->dst);
1798 out:
1799 err = 0;
1800 cleanup:
1801 return err;
1802 }
1803
1804 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1805 /* To make ICMP packets follow the right flow, the multipath hash is
1806 * calculated from the inner IP addresses.
1807 */
1808 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1809 struct flow_keys *hash_keys)
1810 {
1811 const struct iphdr *outer_iph = ip_hdr(skb);
1812 const struct iphdr *inner_iph;
1813 const struct icmphdr *icmph;
1814 struct iphdr _inner_iph;
1815 struct icmphdr _icmph;
1816
1817 hash_keys->addrs.v4addrs.src = outer_iph->saddr;
1818 hash_keys->addrs.v4addrs.dst = outer_iph->daddr;
1819 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1820 return;
1821
1822 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1823 return;
1824
1825 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1826 &_icmph);
1827 if (!icmph)
1828 return;
1829
1830 if (icmph->type != ICMP_DEST_UNREACH &&
1831 icmph->type != ICMP_REDIRECT &&
1832 icmph->type != ICMP_TIME_EXCEEDED &&
1833 icmph->type != ICMP_PARAMETERPROB)
1834 return;
1835
1836 inner_iph = skb_header_pointer(skb,
1837 outer_iph->ihl * 4 + sizeof(_icmph),
1838 sizeof(_inner_iph), &_inner_iph);
1839 if (!inner_iph)
1840 return;
1841 hash_keys->addrs.v4addrs.src = inner_iph->saddr;
1842 hash_keys->addrs.v4addrs.dst = inner_iph->daddr;
1843 }
1844
1845 /* if skb is set it will be used and fl4 can be NULL */
1846 int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4,
1847 const struct sk_buff *skb)
1848 {
1849 struct net *net = fi->fib_net;
1850 struct flow_keys hash_keys;
1851 u32 mhash;
1852
1853 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1854 case 0:
1855 memset(&hash_keys, 0, sizeof(hash_keys));
1856 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1857 if (skb) {
1858 ip_multipath_l3_keys(skb, &hash_keys);
1859 } else {
1860 hash_keys.addrs.v4addrs.src = fl4->saddr;
1861 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1862 }
1863 break;
1864 case 1:
1865 /* skb is currently provided only when forwarding */
1866 if (skb) {
1867 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1868 struct flow_keys keys;
1869
1870 /* short-circuit if we already have L4 hash present */
1871 if (skb->l4_hash)
1872 return skb_get_hash_raw(skb) >> 1;
1873 memset(&hash_keys, 0, sizeof(hash_keys));
1874 skb_flow_dissect_flow_keys(skb, &keys, flag);
1875
1876 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1877 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1878 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1879 hash_keys.ports.src = keys.ports.src;
1880 hash_keys.ports.dst = keys.ports.dst;
1881 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1882 } else {
1883 memset(&hash_keys, 0, sizeof(hash_keys));
1884 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1885 hash_keys.addrs.v4addrs.src = fl4->saddr;
1886 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1887 hash_keys.ports.src = fl4->fl4_sport;
1888 hash_keys.ports.dst = fl4->fl4_dport;
1889 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1890 }
1891 break;
1892 }
1893 mhash = flow_hash_from_keys(&hash_keys);
1894
1895 return mhash >> 1;
1896 }
1897 EXPORT_SYMBOL_GPL(fib_multipath_hash);
1898 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1899
1900 static int ip_mkroute_input(struct sk_buff *skb,
1901 struct fib_result *res,
1902 struct in_device *in_dev,
1903 __be32 daddr, __be32 saddr, u32 tos)
1904 {
1905 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1906 if (res->fi && res->fi->fib_nhs > 1) {
1907 int h = fib_multipath_hash(res->fi, NULL, skb);
1908
1909 fib_select_multipath(res, h);
1910 }
1911 #endif
1912
1913 /* create a routing cache entry */
1914 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1915 }
1916
1917 /*
1918 * NOTE. We drop all the packets that has local source
1919 * addresses, because every properly looped back packet
1920 * must have correct destination already attached by output routine.
1921 *
1922 * Such approach solves two big problems:
1923 * 1. Not simplex devices are handled properly.
1924 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1925 * called with rcu_read_lock()
1926 */
1927
1928 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1929 u8 tos, struct net_device *dev,
1930 struct fib_result *res)
1931 {
1932 struct in_device *in_dev = __in_dev_get_rcu(dev);
1933 struct ip_tunnel_info *tun_info;
1934 struct flowi4 fl4;
1935 unsigned int flags = 0;
1936 u32 itag = 0;
1937 struct rtable *rth;
1938 int err = -EINVAL;
1939 struct net *net = dev_net(dev);
1940 bool do_cache;
1941
1942 /* IP on this device is disabled. */
1943
1944 if (!in_dev)
1945 goto out;
1946
1947 /* Check for the most weird martians, which can be not detected
1948 by fib_lookup.
1949 */
1950
1951 tun_info = skb_tunnel_info(skb);
1952 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1953 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1954 else
1955 fl4.flowi4_tun_key.tun_id = 0;
1956 skb_dst_drop(skb);
1957
1958 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1959 goto martian_source;
1960
1961 res->fi = NULL;
1962 res->table = NULL;
1963 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1964 goto brd_input;
1965
1966 /* Accept zero addresses only to limited broadcast;
1967 * I even do not know to fix it or not. Waiting for complains :-)
1968 */
1969 if (ipv4_is_zeronet(saddr))
1970 goto martian_source;
1971
1972 if (ipv4_is_zeronet(daddr))
1973 goto martian_destination;
1974
1975 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1976 * and call it once if daddr or/and saddr are loopback addresses
1977 */
1978 if (ipv4_is_loopback(daddr)) {
1979 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1980 goto martian_destination;
1981 } else if (ipv4_is_loopback(saddr)) {
1982 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1983 goto martian_source;
1984 }
1985
1986 /*
1987 * Now we are ready to route packet.
1988 */
1989 fl4.flowi4_oif = 0;
1990 fl4.flowi4_iif = dev->ifindex;
1991 fl4.flowi4_mark = skb->mark;
1992 fl4.flowi4_tos = tos;
1993 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1994 fl4.flowi4_flags = 0;
1995 fl4.daddr = daddr;
1996 fl4.saddr = saddr;
1997 fl4.flowi4_uid = sock_net_uid(net, NULL);
1998 err = fib_lookup(net, &fl4, res, 0);
1999 if (err != 0) {
2000 if (!IN_DEV_FORWARD(in_dev))
2001 err = -EHOSTUNREACH;
2002 goto no_route;
2003 }
2004
2005 if (res->type == RTN_BROADCAST)
2006 goto brd_input;
2007
2008 if (res->type == RTN_LOCAL) {
2009 err = fib_validate_source(skb, saddr, daddr, tos,
2010 0, dev, in_dev, &itag);
2011 if (err < 0)
2012 goto martian_source;
2013 goto local_input;
2014 }
2015
2016 if (!IN_DEV_FORWARD(in_dev)) {
2017 err = -EHOSTUNREACH;
2018 goto no_route;
2019 }
2020 if (res->type != RTN_UNICAST)
2021 goto martian_destination;
2022
2023 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos);
2024 out: return err;
2025
2026 brd_input:
2027 if (skb->protocol != htons(ETH_P_IP))
2028 goto e_inval;
2029
2030 if (!ipv4_is_zeronet(saddr)) {
2031 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2032 in_dev, &itag);
2033 if (err < 0)
2034 goto martian_source;
2035 }
2036 flags |= RTCF_BROADCAST;
2037 res->type = RTN_BROADCAST;
2038 RT_CACHE_STAT_INC(in_brd);
2039
2040 local_input:
2041 do_cache = false;
2042 if (res->fi) {
2043 if (!itag) {
2044 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
2045 if (rt_cache_valid(rth)) {
2046 skb_dst_set_noref(skb, &rth->dst);
2047 err = 0;
2048 goto out;
2049 }
2050 do_cache = true;
2051 }
2052 }
2053
2054 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2055 flags | RTCF_LOCAL, res->type,
2056 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2057 if (!rth)
2058 goto e_nobufs;
2059
2060 rth->dst.output= ip_rt_bug;
2061 #ifdef CONFIG_IP_ROUTE_CLASSID
2062 rth->dst.tclassid = itag;
2063 #endif
2064 rth->rt_is_input = 1;
2065 if (res->table)
2066 rth->rt_table_id = res->table->tb_id;
2067
2068 RT_CACHE_STAT_INC(in_slow_tot);
2069 if (res->type == RTN_UNREACHABLE) {
2070 rth->dst.input= ip_error;
2071 rth->dst.error= -err;
2072 rth->rt_flags &= ~RTCF_LOCAL;
2073 }
2074
2075 if (do_cache) {
2076 struct fib_nh *nh = &FIB_RES_NH(*res);
2077
2078 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
2079 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2080 WARN_ON(rth->dst.input == lwtunnel_input);
2081 rth->dst.lwtstate->orig_input = rth->dst.input;
2082 rth->dst.input = lwtunnel_input;
2083 }
2084
2085 if (unlikely(!rt_cache_route(nh, rth)))
2086 rt_add_uncached_list(rth);
2087 }
2088 skb_dst_set(skb, &rth->dst);
2089 err = 0;
2090 goto out;
2091
2092 no_route:
2093 RT_CACHE_STAT_INC(in_no_route);
2094 res->type = RTN_UNREACHABLE;
2095 res->fi = NULL;
2096 res->table = NULL;
2097 goto local_input;
2098
2099 /*
2100 * Do not cache martian addresses: they should be logged (RFC1812)
2101 */
2102 martian_destination:
2103 RT_CACHE_STAT_INC(in_martian_dst);
2104 #ifdef CONFIG_IP_ROUTE_VERBOSE
2105 if (IN_DEV_LOG_MARTIANS(in_dev))
2106 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2107 &daddr, &saddr, dev->name);
2108 #endif
2109
2110 e_inval:
2111 err = -EINVAL;
2112 goto out;
2113
2114 e_nobufs:
2115 err = -ENOBUFS;
2116 goto out;
2117
2118 martian_source:
2119 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2120 goto out;
2121 }
2122
2123 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2124 u8 tos, struct net_device *dev)
2125 {
2126 struct fib_result res;
2127 int err;
2128
2129 tos &= IPTOS_RT_MASK;
2130 rcu_read_lock();
2131 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2132 rcu_read_unlock();
2133
2134 return err;
2135 }
2136 EXPORT_SYMBOL(ip_route_input_noref);
2137
2138 /* called with rcu_read_lock held */
2139 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2140 u8 tos, struct net_device *dev, struct fib_result *res)
2141 {
2142 /* Multicast recognition logic is moved from route cache to here.
2143 The problem was that too many Ethernet cards have broken/missing
2144 hardware multicast filters :-( As result the host on multicasting
2145 network acquires a lot of useless route cache entries, sort of
2146 SDR messages from all the world. Now we try to get rid of them.
2147 Really, provided software IP multicast filter is organized
2148 reasonably (at least, hashed), it does not result in a slowdown
2149 comparing with route cache reject entries.
2150 Note, that multicast routers are not affected, because
2151 route cache entry is created eventually.
2152 */
2153 if (ipv4_is_multicast(daddr)) {
2154 struct in_device *in_dev = __in_dev_get_rcu(dev);
2155 int our = 0;
2156 int err = -EINVAL;
2157
2158 if (!in_dev)
2159 return err;
2160 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2161 ip_hdr(skb)->protocol);
2162
2163 /* check l3 master if no match yet */
2164 if (!our && netif_is_l3_slave(dev)) {
2165 struct in_device *l3_in_dev;
2166
2167 l3_in_dev = __in_dev_get_rcu(skb->dev);
2168 if (l3_in_dev)
2169 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2170 ip_hdr(skb)->protocol);
2171 }
2172
2173 if (our
2174 #ifdef CONFIG_IP_MROUTE
2175 ||
2176 (!ipv4_is_local_multicast(daddr) &&
2177 IN_DEV_MFORWARD(in_dev))
2178 #endif
2179 ) {
2180 err = ip_route_input_mc(skb, daddr, saddr,
2181 tos, dev, our);
2182 }
2183 return err;
2184 }
2185
2186 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2187 }
2188
2189 /* called with rcu_read_lock() */
2190 static struct rtable *__mkroute_output(const struct fib_result *res,
2191 const struct flowi4 *fl4, int orig_oif,
2192 struct net_device *dev_out,
2193 unsigned int flags)
2194 {
2195 struct fib_info *fi = res->fi;
2196 struct fib_nh_exception *fnhe;
2197 struct in_device *in_dev;
2198 u16 type = res->type;
2199 struct rtable *rth;
2200 bool do_cache;
2201
2202 in_dev = __in_dev_get_rcu(dev_out);
2203 if (!in_dev)
2204 return ERR_PTR(-EINVAL);
2205
2206 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2207 if (ipv4_is_loopback(fl4->saddr) &&
2208 !(dev_out->flags & IFF_LOOPBACK) &&
2209 !netif_is_l3_master(dev_out))
2210 return ERR_PTR(-EINVAL);
2211
2212 if (ipv4_is_lbcast(fl4->daddr))
2213 type = RTN_BROADCAST;
2214 else if (ipv4_is_multicast(fl4->daddr))
2215 type = RTN_MULTICAST;
2216 else if (ipv4_is_zeronet(fl4->daddr))
2217 return ERR_PTR(-EINVAL);
2218
2219 if (dev_out->flags & IFF_LOOPBACK)
2220 flags |= RTCF_LOCAL;
2221
2222 do_cache = true;
2223 if (type == RTN_BROADCAST) {
2224 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2225 fi = NULL;
2226 } else if (type == RTN_MULTICAST) {
2227 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2228 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2229 fl4->flowi4_proto))
2230 flags &= ~RTCF_LOCAL;
2231 else
2232 do_cache = false;
2233 /* If multicast route do not exist use
2234 * default one, but do not gateway in this case.
2235 * Yes, it is hack.
2236 */
2237 if (fi && res->prefixlen < 4)
2238 fi = NULL;
2239 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2240 (orig_oif != dev_out->ifindex)) {
2241 /* For local routes that require a particular output interface
2242 * we do not want to cache the result. Caching the result
2243 * causes incorrect behaviour when there are multiple source
2244 * addresses on the interface, the end result being that if the
2245 * intended recipient is waiting on that interface for the
2246 * packet he won't receive it because it will be delivered on
2247 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2248 * be set to the loopback interface as well.
2249 */
2250 do_cache = false;
2251 }
2252
2253 fnhe = NULL;
2254 do_cache &= fi != NULL;
2255 if (fi) {
2256 struct rtable __rcu **prth;
2257 struct fib_nh *nh = &FIB_RES_NH(*res);
2258
2259 fnhe = find_exception(nh, fl4->daddr);
2260 if (!do_cache)
2261 goto add;
2262 if (fnhe) {
2263 prth = &fnhe->fnhe_rth_output;
2264 } else {
2265 if (unlikely(fl4->flowi4_flags &
2266 FLOWI_FLAG_KNOWN_NH &&
2267 !(nh->nh_gw &&
2268 nh->nh_scope == RT_SCOPE_LINK))) {
2269 do_cache = false;
2270 goto add;
2271 }
2272 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2273 }
2274 rth = rcu_dereference(*prth);
2275 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2276 return rth;
2277 }
2278
2279 add:
2280 rth = rt_dst_alloc(dev_out, flags, type,
2281 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2282 IN_DEV_CONF_GET(in_dev, NOXFRM),
2283 do_cache);
2284 if (!rth)
2285 return ERR_PTR(-ENOBUFS);
2286
2287 rth->rt_iif = orig_oif;
2288 if (res->table)
2289 rth->rt_table_id = res->table->tb_id;
2290
2291 RT_CACHE_STAT_INC(out_slow_tot);
2292
2293 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2294 if (flags & RTCF_LOCAL &&
2295 !(dev_out->flags & IFF_LOOPBACK)) {
2296 rth->dst.output = ip_mc_output;
2297 RT_CACHE_STAT_INC(out_slow_mc);
2298 }
2299 #ifdef CONFIG_IP_MROUTE
2300 if (type == RTN_MULTICAST) {
2301 if (IN_DEV_MFORWARD(in_dev) &&
2302 !ipv4_is_local_multicast(fl4->daddr)) {
2303 rth->dst.input = ip_mr_input;
2304 rth->dst.output = ip_mc_output;
2305 }
2306 }
2307 #endif
2308 }
2309
2310 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2311 set_lwt_redirect(rth);
2312
2313 return rth;
2314 }
2315
2316 /*
2317 * Major route resolver routine.
2318 */
2319
2320 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2321 const struct sk_buff *skb)
2322 {
2323 __u8 tos = RT_FL_TOS(fl4);
2324 struct fib_result res = {
2325 .type = RTN_UNSPEC,
2326 .fi = NULL,
2327 .table = NULL,
2328 .tclassid = 0,
2329 };
2330 struct rtable *rth;
2331
2332 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2333 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2334 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2335 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2336
2337 rcu_read_lock();
2338 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2339 rcu_read_unlock();
2340
2341 return rth;
2342 }
2343 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2344
2345 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2346 struct fib_result *res,
2347 const struct sk_buff *skb)
2348 {
2349 struct net_device *dev_out = NULL;
2350 int orig_oif = fl4->flowi4_oif;
2351 unsigned int flags = 0;
2352 struct rtable *rth;
2353 int err = -ENETUNREACH;
2354
2355 if (fl4->saddr) {
2356 rth = ERR_PTR(-EINVAL);
2357 if (ipv4_is_multicast(fl4->saddr) ||
2358 ipv4_is_lbcast(fl4->saddr) ||
2359 ipv4_is_zeronet(fl4->saddr))
2360 goto out;
2361
2362 /* I removed check for oif == dev_out->oif here.
2363 It was wrong for two reasons:
2364 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2365 is assigned to multiple interfaces.
2366 2. Moreover, we are allowed to send packets with saddr
2367 of another iface. --ANK
2368 */
2369
2370 if (fl4->flowi4_oif == 0 &&
2371 (ipv4_is_multicast(fl4->daddr) ||
2372 ipv4_is_lbcast(fl4->daddr))) {
2373 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2374 dev_out = __ip_dev_find(net, fl4->saddr, false);
2375 if (!dev_out)
2376 goto out;
2377
2378 /* Special hack: user can direct multicasts
2379 and limited broadcast via necessary interface
2380 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2381 This hack is not just for fun, it allows
2382 vic,vat and friends to work.
2383 They bind socket to loopback, set ttl to zero
2384 and expect that it will work.
2385 From the viewpoint of routing cache they are broken,
2386 because we are not allowed to build multicast path
2387 with loopback source addr (look, routing cache
2388 cannot know, that ttl is zero, so that packet
2389 will not leave this host and route is valid).
2390 Luckily, this hack is good workaround.
2391 */
2392
2393 fl4->flowi4_oif = dev_out->ifindex;
2394 goto make_route;
2395 }
2396
2397 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2398 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2399 if (!__ip_dev_find(net, fl4->saddr, false))
2400 goto out;
2401 }
2402 }
2403
2404
2405 if (fl4->flowi4_oif) {
2406 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2407 rth = ERR_PTR(-ENODEV);
2408 if (!dev_out)
2409 goto out;
2410
2411 /* RACE: Check return value of inet_select_addr instead. */
2412 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2413 rth = ERR_PTR(-ENETUNREACH);
2414 goto out;
2415 }
2416 if (ipv4_is_local_multicast(fl4->daddr) ||
2417 ipv4_is_lbcast(fl4->daddr) ||
2418 fl4->flowi4_proto == IPPROTO_IGMP) {
2419 if (!fl4->saddr)
2420 fl4->saddr = inet_select_addr(dev_out, 0,
2421 RT_SCOPE_LINK);
2422 goto make_route;
2423 }
2424 if (!fl4->saddr) {
2425 if (ipv4_is_multicast(fl4->daddr))
2426 fl4->saddr = inet_select_addr(dev_out, 0,
2427 fl4->flowi4_scope);
2428 else if (!fl4->daddr)
2429 fl4->saddr = inet_select_addr(dev_out, 0,
2430 RT_SCOPE_HOST);
2431 }
2432 }
2433
2434 if (!fl4->daddr) {
2435 fl4->daddr = fl4->saddr;
2436 if (!fl4->daddr)
2437 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2438 dev_out = net->loopback_dev;
2439 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2440 res->type = RTN_LOCAL;
2441 flags |= RTCF_LOCAL;
2442 goto make_route;
2443 }
2444
2445 err = fib_lookup(net, fl4, res, 0);
2446 if (err) {
2447 res->fi = NULL;
2448 res->table = NULL;
2449 if (fl4->flowi4_oif &&
2450 (ipv4_is_multicast(fl4->daddr) ||
2451 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2452 /* Apparently, routing tables are wrong. Assume,
2453 that the destination is on link.
2454
2455 WHY? DW.
2456 Because we are allowed to send to iface
2457 even if it has NO routes and NO assigned
2458 addresses. When oif is specified, routing
2459 tables are looked up with only one purpose:
2460 to catch if destination is gatewayed, rather than
2461 direct. Moreover, if MSG_DONTROUTE is set,
2462 we send packet, ignoring both routing tables
2463 and ifaddr state. --ANK
2464
2465
2466 We could make it even if oif is unknown,
2467 likely IPv6, but we do not.
2468 */
2469
2470 if (fl4->saddr == 0)
2471 fl4->saddr = inet_select_addr(dev_out, 0,
2472 RT_SCOPE_LINK);
2473 res->type = RTN_UNICAST;
2474 goto make_route;
2475 }
2476 rth = ERR_PTR(err);
2477 goto out;
2478 }
2479
2480 if (res->type == RTN_LOCAL) {
2481 if (!fl4->saddr) {
2482 if (res->fi->fib_prefsrc)
2483 fl4->saddr = res->fi->fib_prefsrc;
2484 else
2485 fl4->saddr = fl4->daddr;
2486 }
2487
2488 /* L3 master device is the loopback for that domain */
2489 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2490 net->loopback_dev;
2491
2492 /* make sure orig_oif points to fib result device even
2493 * though packet rx/tx happens over loopback or l3mdev
2494 */
2495 orig_oif = FIB_RES_OIF(*res);
2496
2497 fl4->flowi4_oif = dev_out->ifindex;
2498 flags |= RTCF_LOCAL;
2499 goto make_route;
2500 }
2501
2502 fib_select_path(net, res, fl4, skb);
2503
2504 dev_out = FIB_RES_DEV(*res);
2505 fl4->flowi4_oif = dev_out->ifindex;
2506
2507
2508 make_route:
2509 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2510
2511 out:
2512 return rth;
2513 }
2514
2515 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2516 {
2517 return NULL;
2518 }
2519
2520 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2521 {
2522 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2523
2524 return mtu ? : dst->dev->mtu;
2525 }
2526
2527 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2528 struct sk_buff *skb, u32 mtu)
2529 {
2530 }
2531
2532 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2533 struct sk_buff *skb)
2534 {
2535 }
2536
2537 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2538 unsigned long old)
2539 {
2540 return NULL;
2541 }
2542
2543 static struct dst_ops ipv4_dst_blackhole_ops = {
2544 .family = AF_INET,
2545 .check = ipv4_blackhole_dst_check,
2546 .mtu = ipv4_blackhole_mtu,
2547 .default_advmss = ipv4_default_advmss,
2548 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2549 .redirect = ipv4_rt_blackhole_redirect,
2550 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2551 .neigh_lookup = ipv4_neigh_lookup,
2552 };
2553
2554 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2555 {
2556 struct rtable *ort = (struct rtable *) dst_orig;
2557 struct rtable *rt;
2558
2559 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2560 if (rt) {
2561 struct dst_entry *new = &rt->dst;
2562
2563 new->__use = 1;
2564 new->input = dst_discard;
2565 new->output = dst_discard_out;
2566
2567 new->dev = net->loopback_dev;
2568 if (new->dev)
2569 dev_hold(new->dev);
2570
2571 rt->rt_is_input = ort->rt_is_input;
2572 rt->rt_iif = ort->rt_iif;
2573 rt->rt_pmtu = ort->rt_pmtu;
2574 rt->rt_mtu_locked = ort->rt_mtu_locked;
2575
2576 rt->rt_genid = rt_genid_ipv4(net);
2577 rt->rt_flags = ort->rt_flags;
2578 rt->rt_type = ort->rt_type;
2579 rt->rt_gateway = ort->rt_gateway;
2580 rt->rt_uses_gateway = ort->rt_uses_gateway;
2581
2582 INIT_LIST_HEAD(&rt->rt_uncached);
2583 }
2584
2585 dst_release(dst_orig);
2586
2587 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2588 }
2589
2590 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2591 const struct sock *sk)
2592 {
2593 struct rtable *rt = __ip_route_output_key(net, flp4);
2594
2595 if (IS_ERR(rt))
2596 return rt;
2597
2598 if (flp4->flowi4_proto)
2599 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2600 flowi4_to_flowi(flp4),
2601 sk, 0);
2602
2603 return rt;
2604 }
2605 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2606
2607 /* called with rcu_read_lock held */
2608 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2609 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2610 u32 seq)
2611 {
2612 struct rtable *rt = skb_rtable(skb);
2613 struct rtmsg *r;
2614 struct nlmsghdr *nlh;
2615 unsigned long expires = 0;
2616 u32 error;
2617 u32 metrics[RTAX_MAX];
2618
2619 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0);
2620 if (!nlh)
2621 return -EMSGSIZE;
2622
2623 r = nlmsg_data(nlh);
2624 r->rtm_family = AF_INET;
2625 r->rtm_dst_len = 32;
2626 r->rtm_src_len = 0;
2627 r->rtm_tos = fl4->flowi4_tos;
2628 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2629 if (nla_put_u32(skb, RTA_TABLE, table_id))
2630 goto nla_put_failure;
2631 r->rtm_type = rt->rt_type;
2632 r->rtm_scope = RT_SCOPE_UNIVERSE;
2633 r->rtm_protocol = RTPROT_UNSPEC;
2634 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2635 if (rt->rt_flags & RTCF_NOTIFY)
2636 r->rtm_flags |= RTM_F_NOTIFY;
2637 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2638 r->rtm_flags |= RTCF_DOREDIRECT;
2639
2640 if (nla_put_in_addr(skb, RTA_DST, dst))
2641 goto nla_put_failure;
2642 if (src) {
2643 r->rtm_src_len = 32;
2644 if (nla_put_in_addr(skb, RTA_SRC, src))
2645 goto nla_put_failure;
2646 }
2647 if (rt->dst.dev &&
2648 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2649 goto nla_put_failure;
2650 #ifdef CONFIG_IP_ROUTE_CLASSID
2651 if (rt->dst.tclassid &&
2652 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2653 goto nla_put_failure;
2654 #endif
2655 if (!rt_is_input_route(rt) &&
2656 fl4->saddr != src) {
2657 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2658 goto nla_put_failure;
2659 }
2660 if (rt->rt_uses_gateway &&
2661 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2662 goto nla_put_failure;
2663
2664 expires = rt->dst.expires;
2665 if (expires) {
2666 unsigned long now = jiffies;
2667
2668 if (time_before(now, expires))
2669 expires -= now;
2670 else
2671 expires = 0;
2672 }
2673
2674 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2675 if (rt->rt_pmtu && expires)
2676 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2677 if (rt->rt_mtu_locked && expires)
2678 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2679 if (rtnetlink_put_metrics(skb, metrics) < 0)
2680 goto nla_put_failure;
2681
2682 if (fl4->flowi4_mark &&
2683 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2684 goto nla_put_failure;
2685
2686 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2687 nla_put_u32(skb, RTA_UID,
2688 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2689 goto nla_put_failure;
2690
2691 error = rt->dst.error;
2692
2693 if (rt_is_input_route(rt)) {
2694 #ifdef CONFIG_IP_MROUTE
2695 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2696 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2697 int err = ipmr_get_route(net, skb,
2698 fl4->saddr, fl4->daddr,
2699 r, portid);
2700
2701 if (err <= 0) {
2702 if (err == 0)
2703 return 0;
2704 goto nla_put_failure;
2705 }
2706 } else
2707 #endif
2708 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2709 goto nla_put_failure;
2710 }
2711
2712 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2713 goto nla_put_failure;
2714
2715 nlmsg_end(skb, nlh);
2716 return 0;
2717
2718 nla_put_failure:
2719 nlmsg_cancel(skb, nlh);
2720 return -EMSGSIZE;
2721 }
2722
2723 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2724 struct netlink_ext_ack *extack)
2725 {
2726 struct net *net = sock_net(in_skb->sk);
2727 struct rtmsg *rtm;
2728 struct nlattr *tb[RTA_MAX+1];
2729 struct fib_result res = {};
2730 struct rtable *rt = NULL;
2731 struct flowi4 fl4;
2732 __be32 dst = 0;
2733 __be32 src = 0;
2734 u32 iif;
2735 int err;
2736 int mark;
2737 struct sk_buff *skb;
2738 u32 table_id = RT_TABLE_MAIN;
2739 kuid_t uid;
2740
2741 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy,
2742 extack);
2743 if (err < 0)
2744 goto errout;
2745
2746 rtm = nlmsg_data(nlh);
2747
2748 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2749 if (!skb) {
2750 err = -ENOBUFS;
2751 goto errout;
2752 }
2753
2754 /* Reserve room for dummy headers, this skb can pass
2755 through good chunk of routing engine.
2756 */
2757 skb_reset_mac_header(skb);
2758 skb_reset_network_header(skb);
2759
2760 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2761 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2762 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2763 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2764 if (tb[RTA_UID])
2765 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2766 else
2767 uid = (iif ? INVALID_UID : current_uid());
2768
2769 /* Bugfix: need to give ip_route_input enough of an IP header to
2770 * not gag.
2771 */
2772 ip_hdr(skb)->protocol = IPPROTO_UDP;
2773 ip_hdr(skb)->saddr = src;
2774 ip_hdr(skb)->daddr = dst;
2775
2776 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2777
2778 memset(&fl4, 0, sizeof(fl4));
2779 fl4.daddr = dst;
2780 fl4.saddr = src;
2781 fl4.flowi4_tos = rtm->rtm_tos;
2782 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2783 fl4.flowi4_mark = mark;
2784 fl4.flowi4_uid = uid;
2785
2786 rcu_read_lock();
2787
2788 if (iif) {
2789 struct net_device *dev;
2790
2791 dev = dev_get_by_index_rcu(net, iif);
2792 if (!dev) {
2793 err = -ENODEV;
2794 goto errout_free;
2795 }
2796
2797 skb->protocol = htons(ETH_P_IP);
2798 skb->dev = dev;
2799 skb->mark = mark;
2800 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
2801 dev, &res);
2802
2803 rt = skb_rtable(skb);
2804 if (err == 0 && rt->dst.error)
2805 err = -rt->dst.error;
2806 } else {
2807 fl4.flowi4_iif = LOOPBACK_IFINDEX;
2808 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
2809 err = 0;
2810 if (IS_ERR(rt))
2811 err = PTR_ERR(rt);
2812 else
2813 skb_dst_set(skb, &rt->dst);
2814 }
2815
2816 if (err)
2817 goto errout_free;
2818
2819 if (rtm->rtm_flags & RTM_F_NOTIFY)
2820 rt->rt_flags |= RTCF_NOTIFY;
2821
2822 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2823 table_id = rt->rt_table_id;
2824
2825 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2826 if (!res.fi) {
2827 err = fib_props[res.type].error;
2828 if (!err)
2829 err = -EHOSTUNREACH;
2830 goto errout_free;
2831 }
2832 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
2833 nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
2834 rt->rt_type, res.prefix, res.prefixlen,
2835 fl4.flowi4_tos, res.fi, 0);
2836 } else {
2837 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2838 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq);
2839 }
2840 if (err < 0)
2841 goto errout_free;
2842
2843 rcu_read_unlock();
2844
2845 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2846 errout:
2847 return err;
2848
2849 errout_free:
2850 rcu_read_unlock();
2851 kfree_skb(skb);
2852 goto errout;
2853 }
2854
2855 void ip_rt_multicast_event(struct in_device *in_dev)
2856 {
2857 rt_cache_flush(dev_net(in_dev->dev));
2858 }
2859
2860 #ifdef CONFIG_SYSCTL
2861 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2862 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2863 static int ip_rt_gc_elasticity __read_mostly = 8;
2864
2865 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2866 void __user *buffer,
2867 size_t *lenp, loff_t *ppos)
2868 {
2869 struct net *net = (struct net *)__ctl->extra1;
2870
2871 if (write) {
2872 rt_cache_flush(net);
2873 fnhe_genid_bump(net);
2874 return 0;
2875 }
2876
2877 return -EINVAL;
2878 }
2879
2880 static struct ctl_table ipv4_route_table[] = {
2881 {
2882 .procname = "gc_thresh",
2883 .data = &ipv4_dst_ops.gc_thresh,
2884 .maxlen = sizeof(int),
2885 .mode = 0644,
2886 .proc_handler = proc_dointvec,
2887 },
2888 {
2889 .procname = "max_size",
2890 .data = &ip_rt_max_size,
2891 .maxlen = sizeof(int),
2892 .mode = 0644,
2893 .proc_handler = proc_dointvec,
2894 },
2895 {
2896 /* Deprecated. Use gc_min_interval_ms */
2897
2898 .procname = "gc_min_interval",
2899 .data = &ip_rt_gc_min_interval,
2900 .maxlen = sizeof(int),
2901 .mode = 0644,
2902 .proc_handler = proc_dointvec_jiffies,
2903 },
2904 {
2905 .procname = "gc_min_interval_ms",
2906 .data = &ip_rt_gc_min_interval,
2907 .maxlen = sizeof(int),
2908 .mode = 0644,
2909 .proc_handler = proc_dointvec_ms_jiffies,
2910 },
2911 {
2912 .procname = "gc_timeout",
2913 .data = &ip_rt_gc_timeout,
2914 .maxlen = sizeof(int),
2915 .mode = 0644,
2916 .proc_handler = proc_dointvec_jiffies,
2917 },
2918 {
2919 .procname = "gc_interval",
2920 .data = &ip_rt_gc_interval,
2921 .maxlen = sizeof(int),
2922 .mode = 0644,
2923 .proc_handler = proc_dointvec_jiffies,
2924 },
2925 {
2926 .procname = "redirect_load",
2927 .data = &ip_rt_redirect_load,
2928 .maxlen = sizeof(int),
2929 .mode = 0644,
2930 .proc_handler = proc_dointvec,
2931 },
2932 {
2933 .procname = "redirect_number",
2934 .data = &ip_rt_redirect_number,
2935 .maxlen = sizeof(int),
2936 .mode = 0644,
2937 .proc_handler = proc_dointvec,
2938 },
2939 {
2940 .procname = "redirect_silence",
2941 .data = &ip_rt_redirect_silence,
2942 .maxlen = sizeof(int),
2943 .mode = 0644,
2944 .proc_handler = proc_dointvec,
2945 },
2946 {
2947 .procname = "error_cost",
2948 .data = &ip_rt_error_cost,
2949 .maxlen = sizeof(int),
2950 .mode = 0644,
2951 .proc_handler = proc_dointvec,
2952 },
2953 {
2954 .procname = "error_burst",
2955 .data = &ip_rt_error_burst,
2956 .maxlen = sizeof(int),
2957 .mode = 0644,
2958 .proc_handler = proc_dointvec,
2959 },
2960 {
2961 .procname = "gc_elasticity",
2962 .data = &ip_rt_gc_elasticity,
2963 .maxlen = sizeof(int),
2964 .mode = 0644,
2965 .proc_handler = proc_dointvec,
2966 },
2967 {
2968 .procname = "mtu_expires",
2969 .data = &ip_rt_mtu_expires,
2970 .maxlen = sizeof(int),
2971 .mode = 0644,
2972 .proc_handler = proc_dointvec_jiffies,
2973 },
2974 {
2975 .procname = "min_pmtu",
2976 .data = &ip_rt_min_pmtu,
2977 .maxlen = sizeof(int),
2978 .mode = 0644,
2979 .proc_handler = proc_dointvec_minmax,
2980 .extra1 = &ip_min_valid_pmtu,
2981 },
2982 {
2983 .procname = "min_adv_mss",
2984 .data = &ip_rt_min_advmss,
2985 .maxlen = sizeof(int),
2986 .mode = 0644,
2987 .proc_handler = proc_dointvec,
2988 },
2989 { }
2990 };
2991
2992 static struct ctl_table ipv4_route_flush_table[] = {
2993 {
2994 .procname = "flush",
2995 .maxlen = sizeof(int),
2996 .mode = 0200,
2997 .proc_handler = ipv4_sysctl_rtcache_flush,
2998 },
2999 { },
3000 };
3001
3002 static __net_init int sysctl_route_net_init(struct net *net)
3003 {
3004 struct ctl_table *tbl;
3005
3006 tbl = ipv4_route_flush_table;
3007 if (!net_eq(net, &init_net)) {
3008 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
3009 if (!tbl)
3010 goto err_dup;
3011
3012 /* Don't export sysctls to unprivileged users */
3013 if (net->user_ns != &init_user_ns)
3014 tbl[0].procname = NULL;
3015 }
3016 tbl[0].extra1 = net;
3017
3018 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
3019 if (!net->ipv4.route_hdr)
3020 goto err_reg;
3021 return 0;
3022
3023 err_reg:
3024 if (tbl != ipv4_route_flush_table)
3025 kfree(tbl);
3026 err_dup:
3027 return -ENOMEM;
3028 }
3029
3030 static __net_exit void sysctl_route_net_exit(struct net *net)
3031 {
3032 struct ctl_table *tbl;
3033
3034 tbl = net->ipv4.route_hdr->ctl_table_arg;
3035 unregister_net_sysctl_table(net->ipv4.route_hdr);
3036 BUG_ON(tbl == ipv4_route_flush_table);
3037 kfree(tbl);
3038 }
3039
3040 static __net_initdata struct pernet_operations sysctl_route_ops = {
3041 .init = sysctl_route_net_init,
3042 .exit = sysctl_route_net_exit,
3043 };
3044 #endif
3045
3046 static __net_init int rt_genid_init(struct net *net)
3047 {
3048 atomic_set(&net->ipv4.rt_genid, 0);
3049 atomic_set(&net->fnhe_genid, 0);
3050 atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3051 return 0;
3052 }
3053
3054 static __net_initdata struct pernet_operations rt_genid_ops = {
3055 .init = rt_genid_init,
3056 };
3057
3058 static int __net_init ipv4_inetpeer_init(struct net *net)
3059 {
3060 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3061
3062 if (!bp)
3063 return -ENOMEM;
3064 inet_peer_base_init(bp);
3065 net->ipv4.peers = bp;
3066 return 0;
3067 }
3068
3069 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3070 {
3071 struct inet_peer_base *bp = net->ipv4.peers;
3072
3073 net->ipv4.peers = NULL;
3074 inetpeer_invalidate_tree(bp);
3075 kfree(bp);
3076 }
3077
3078 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3079 .init = ipv4_inetpeer_init,
3080 .exit = ipv4_inetpeer_exit,
3081 };
3082
3083 #ifdef CONFIG_IP_ROUTE_CLASSID
3084 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3085 #endif /* CONFIG_IP_ROUTE_CLASSID */
3086
3087 int __init ip_rt_init(void)
3088 {
3089 int rc = 0;
3090 int cpu;
3091
3092 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
3093 if (!ip_idents)
3094 panic("IP: failed to allocate ip_idents\n");
3095
3096 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3097
3098 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3099 if (!ip_tstamps)
3100 panic("IP: failed to allocate ip_tstamps\n");
3101
3102 for_each_possible_cpu(cpu) {
3103 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3104
3105 INIT_LIST_HEAD(&ul->head);
3106 spin_lock_init(&ul->lock);
3107 }
3108 #ifdef CONFIG_IP_ROUTE_CLASSID
3109 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3110 if (!ip_rt_acct)
3111 panic("IP: failed to allocate ip_rt_acct\n");
3112 #endif
3113
3114 ipv4_dst_ops.kmem_cachep =
3115 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3116 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3117
3118 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3119
3120 if (dst_entries_init(&ipv4_dst_ops) < 0)
3121 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3122
3123 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3124 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3125
3126 ipv4_dst_ops.gc_thresh = ~0;
3127 ip_rt_max_size = INT_MAX;
3128
3129 devinet_init();
3130 ip_fib_init();
3131
3132 if (ip_rt_proc_init())
3133 pr_err("Unable to create route proc files\n");
3134 #ifdef CONFIG_XFRM
3135 xfrm_init();
3136 xfrm4_init();
3137 #endif
3138 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3139 RTNL_FLAG_DOIT_UNLOCKED);
3140
3141 #ifdef CONFIG_SYSCTL
3142 register_pernet_subsys(&sysctl_route_ops);
3143 #endif
3144 register_pernet_subsys(&rt_genid_ops);
3145 register_pernet_subsys(&ipv4_inetpeer_ops);
3146 return rc;
3147 }
3148
3149 #ifdef CONFIG_SYSCTL
3150 /*
3151 * We really need to sanitize the damn ipv4 init order, then all
3152 * this nonsense will go away.
3153 */
3154 void __init ip_static_sysctl_init(void)
3155 {
3156 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3157 }
3158 #endif
Linux with added FPC-III support