]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - net/ipv4/route.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
MAINTAINERS: Update MAX77802 PMIC entry
[mirror_ubuntu-artful-kernel.git] / net / ipv4 / route.c
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * ROUTE - implementation of the IP router.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 * Linus Torvalds, <Linus.Torvalds@helsinki.fi>
12 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
13 *
14 * Fixes:
15 * Alan Cox : Verify area fixes.
16 * Alan Cox : cli() protects routing changes
17 * Rui Oliveira : ICMP routing table updates
18 * (rco@di.uminho.pt) Routing table insertion and update
19 * Linus Torvalds : Rewrote bits to be sensible
20 * Alan Cox : Added BSD route gw semantics
21 * Alan Cox : Super /proc >4K
22 * Alan Cox : MTU in route table
23 * Alan Cox : MSS actually. Also added the window
24 * clamper.
25 * Sam Lantinga : Fixed route matching in rt_del()
26 * Alan Cox : Routing cache support.
27 * Alan Cox : Removed compatibility cruft.
28 * Alan Cox : RTF_REJECT support.
29 * Alan Cox : TCP irtt support.
30 * Jonathan Naylor : Added Metric support.
31 * Miquel van Smoorenburg : BSD API fixes.
32 * Miquel van Smoorenburg : Metrics.
33 * Alan Cox : Use __u32 properly
34 * Alan Cox : Aligned routing errors more closely with BSD
35 * our system is still very different.
36 * Alan Cox : Faster /proc handling
37 * Alexey Kuznetsov : Massive rework to support tree based routing,
38 * routing caches and better behaviour.
39 *
40 * Olaf Erb : irtt wasn't being copied right.
41 * Bjorn Ekwall : Kerneld route support.
42 * Alan Cox : Multicast fixed (I hope)
43 * Pavel Krauz : Limited broadcast fixed
44 * Mike McLagan : Routing by source
45 * Alexey Kuznetsov : End of old history. Split to fib.c and
46 * route.c and rewritten from scratch.
47 * Andi Kleen : Load-limit warning messages.
48 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
49 * Vitaly E. Lavrov : Race condition in ip_route_input_slow.
50 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow.
51 * Vladimir V. Ivanov : IP rule info (flowid) is really useful.
52 * Marc Boucher : routing by fwmark
53 * Robert Olsson : Added rt_cache statistics
54 * Arnaldo C. Melo : Convert proc stuff to seq_file
55 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes.
56 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect
57 * Ilia Sotnikov : Removed TOS from hash calculations
58 *
59 * This program is free software; you can redistribute it and/or
60 * modify it under the terms of the GNU General Public License
61 * as published by the Free Software Foundation; either version
62 * 2 of the License, or (at your option) any later version.
63 */
64
65 #define pr_fmt(fmt) "IPv4: " fmt
66
67 #include <linux/module.h>
68 #include <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 header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1254 unsigned int advmss = max_t(unsigned int, dst->dev->mtu - header_size,
1255 ip_rt_min_advmss);
1256
1257 return min(advmss, IPV4_MAX_PMTU - header_size);
1258 }
1259
1260 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1261 {
1262 const struct rtable *rt = (const struct rtable *) dst;
1263 unsigned int mtu = rt->rt_pmtu;
1264
1265 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1266 mtu = dst_metric_raw(dst, RTAX_MTU);
1267
1268 if (mtu)
1269 return mtu;
1270
1271 mtu = dst->dev->mtu;
1272
1273 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) {
1274 if (rt->rt_uses_gateway && mtu > 576)
1275 mtu = 576;
1276 }
1277
1278 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1279
1280 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1281 }
1282
1283 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1284 {
1285 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1286 struct fib_nh_exception *fnhe;
1287 u32 hval;
1288
1289 if (!hash)
1290 return NULL;
1291
1292 hval = fnhe_hashfun(daddr);
1293
1294 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1295 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1296 if (fnhe->fnhe_daddr == daddr)
1297 return fnhe;
1298 }
1299 return NULL;
1300 }
1301
1302 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1303 __be32 daddr)
1304 {
1305 bool ret = false;
1306
1307 spin_lock_bh(&fnhe_lock);
1308
1309 if (daddr == fnhe->fnhe_daddr) {
1310 struct rtable __rcu **porig;
1311 struct rtable *orig;
1312 int genid = fnhe_genid(dev_net(rt->dst.dev));
1313
1314 if (rt_is_input_route(rt))
1315 porig = &fnhe->fnhe_rth_input;
1316 else
1317 porig = &fnhe->fnhe_rth_output;
1318 orig = rcu_dereference(*porig);
1319
1320 if (fnhe->fnhe_genid != genid) {
1321 fnhe->fnhe_genid = genid;
1322 fnhe->fnhe_gw = 0;
1323 fnhe->fnhe_pmtu = 0;
1324 fnhe->fnhe_expires = 0;
1325 fnhe_flush_routes(fnhe);
1326 orig = NULL;
1327 }
1328 fill_route_from_fnhe(rt, fnhe);
1329 if (!rt->rt_gateway)
1330 rt->rt_gateway = daddr;
1331
1332 if (!(rt->dst.flags & DST_NOCACHE)) {
1333 rcu_assign_pointer(*porig, rt);
1334 if (orig)
1335 rt_free(orig);
1336 ret = true;
1337 }
1338
1339 fnhe->fnhe_stamp = jiffies;
1340 }
1341 spin_unlock_bh(&fnhe_lock);
1342
1343 return ret;
1344 }
1345
1346 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1347 {
1348 struct rtable *orig, *prev, **p;
1349 bool ret = true;
1350
1351 if (rt_is_input_route(rt)) {
1352 p = (struct rtable **)&nh->nh_rth_input;
1353 } else {
1354 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1355 }
1356 orig = *p;
1357
1358 prev = cmpxchg(p, orig, rt);
1359 if (prev == orig) {
1360 if (orig)
1361 rt_free(orig);
1362 } else
1363 ret = false;
1364
1365 return ret;
1366 }
1367
1368 struct uncached_list {
1369 spinlock_t lock;
1370 struct list_head head;
1371 };
1372
1373 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1374
1375 static void rt_add_uncached_list(struct rtable *rt)
1376 {
1377 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1378
1379 rt->rt_uncached_list = ul;
1380
1381 spin_lock_bh(&ul->lock);
1382 list_add_tail(&rt->rt_uncached, &ul->head);
1383 spin_unlock_bh(&ul->lock);
1384 }
1385
1386 static void ipv4_dst_destroy(struct dst_entry *dst)
1387 {
1388 struct rtable *rt = (struct rtable *) dst;
1389
1390 if (!list_empty(&rt->rt_uncached)) {
1391 struct uncached_list *ul = rt->rt_uncached_list;
1392
1393 spin_lock_bh(&ul->lock);
1394 list_del(&rt->rt_uncached);
1395 spin_unlock_bh(&ul->lock);
1396 }
1397 }
1398
1399 void rt_flush_dev(struct net_device *dev)
1400 {
1401 struct net *net = dev_net(dev);
1402 struct rtable *rt;
1403 int cpu;
1404
1405 for_each_possible_cpu(cpu) {
1406 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1407
1408 spin_lock_bh(&ul->lock);
1409 list_for_each_entry(rt, &ul->head, rt_uncached) {
1410 if (rt->dst.dev != dev)
1411 continue;
1412 rt->dst.dev = net->loopback_dev;
1413 dev_hold(rt->dst.dev);
1414 dev_put(dev);
1415 }
1416 spin_unlock_bh(&ul->lock);
1417 }
1418 }
1419
1420 static bool rt_cache_valid(const struct rtable *rt)
1421 {
1422 return rt &&
1423 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1424 !rt_is_expired(rt);
1425 }
1426
1427 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1428 const struct fib_result *res,
1429 struct fib_nh_exception *fnhe,
1430 struct fib_info *fi, u16 type, u32 itag)
1431 {
1432 bool cached = false;
1433
1434 if (fi) {
1435 struct fib_nh *nh = &FIB_RES_NH(*res);
1436
1437 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1438 rt->rt_gateway = nh->nh_gw;
1439 rt->rt_uses_gateway = 1;
1440 }
1441 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
1442 #ifdef CONFIG_IP_ROUTE_CLASSID
1443 rt->dst.tclassid = nh->nh_tclassid;
1444 #endif
1445 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1446 if (unlikely(fnhe))
1447 cached = rt_bind_exception(rt, fnhe, daddr);
1448 else if (!(rt->dst.flags & DST_NOCACHE))
1449 cached = rt_cache_route(nh, rt);
1450 if (unlikely(!cached)) {
1451 /* Routes we intend to cache in nexthop exception or
1452 * FIB nexthop have the DST_NOCACHE bit clear.
1453 * However, if we are unsuccessful at storing this
1454 * route into the cache we really need to set it.
1455 */
1456 rt->dst.flags |= DST_NOCACHE;
1457 if (!rt->rt_gateway)
1458 rt->rt_gateway = daddr;
1459 rt_add_uncached_list(rt);
1460 }
1461 } else
1462 rt_add_uncached_list(rt);
1463
1464 #ifdef CONFIG_IP_ROUTE_CLASSID
1465 #ifdef CONFIG_IP_MULTIPLE_TABLES
1466 set_class_tag(rt, res->tclassid);
1467 #endif
1468 set_class_tag(rt, itag);
1469 #endif
1470 }
1471
1472 struct rtable *rt_dst_alloc(struct net_device *dev,
1473 unsigned int flags, u16 type,
1474 bool nopolicy, bool noxfrm, bool will_cache)
1475 {
1476 struct rtable *rt;
1477
1478 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1479 (will_cache ? 0 : (DST_HOST | DST_NOCACHE)) |
1480 (nopolicy ? DST_NOPOLICY : 0) |
1481 (noxfrm ? DST_NOXFRM : 0));
1482
1483 if (rt) {
1484 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1485 rt->rt_flags = flags;
1486 rt->rt_type = type;
1487 rt->rt_is_input = 0;
1488 rt->rt_iif = 0;
1489 rt->rt_pmtu = 0;
1490 rt->rt_gateway = 0;
1491 rt->rt_uses_gateway = 0;
1492 rt->rt_table_id = 0;
1493 INIT_LIST_HEAD(&rt->rt_uncached);
1494
1495 rt->dst.output = ip_output;
1496 if (flags & RTCF_LOCAL)
1497 rt->dst.input = ip_local_deliver;
1498 }
1499
1500 return rt;
1501 }
1502 EXPORT_SYMBOL(rt_dst_alloc);
1503
1504 /* called in rcu_read_lock() section */
1505 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1506 u8 tos, struct net_device *dev, int our)
1507 {
1508 struct rtable *rth;
1509 struct in_device *in_dev = __in_dev_get_rcu(dev);
1510 unsigned int flags = RTCF_MULTICAST;
1511 u32 itag = 0;
1512 int err;
1513
1514 /* Primary sanity checks. */
1515
1516 if (!in_dev)
1517 return -EINVAL;
1518
1519 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1520 skb->protocol != htons(ETH_P_IP))
1521 goto e_inval;
1522
1523 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1524 goto e_inval;
1525
1526 if (ipv4_is_zeronet(saddr)) {
1527 if (!ipv4_is_local_multicast(daddr))
1528 goto e_inval;
1529 } else {
1530 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1531 in_dev, &itag);
1532 if (err < 0)
1533 goto e_err;
1534 }
1535 if (our)
1536 flags |= RTCF_LOCAL;
1537
1538 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1539 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1540 if (!rth)
1541 goto e_nobufs;
1542
1543 #ifdef CONFIG_IP_ROUTE_CLASSID
1544 rth->dst.tclassid = itag;
1545 #endif
1546 rth->dst.output = ip_rt_bug;
1547 rth->rt_is_input= 1;
1548
1549 #ifdef CONFIG_IP_MROUTE
1550 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1551 rth->dst.input = ip_mr_input;
1552 #endif
1553 RT_CACHE_STAT_INC(in_slow_mc);
1554
1555 skb_dst_set(skb, &rth->dst);
1556 return 0;
1557
1558 e_nobufs:
1559 return -ENOBUFS;
1560 e_inval:
1561 return -EINVAL;
1562 e_err:
1563 return err;
1564 }
1565
1566
1567 static void ip_handle_martian_source(struct net_device *dev,
1568 struct in_device *in_dev,
1569 struct sk_buff *skb,
1570 __be32 daddr,
1571 __be32 saddr)
1572 {
1573 RT_CACHE_STAT_INC(in_martian_src);
1574 #ifdef CONFIG_IP_ROUTE_VERBOSE
1575 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1576 /*
1577 * RFC1812 recommendation, if source is martian,
1578 * the only hint is MAC header.
1579 */
1580 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1581 &daddr, &saddr, dev->name);
1582 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1583 print_hex_dump(KERN_WARNING, "ll header: ",
1584 DUMP_PREFIX_OFFSET, 16, 1,
1585 skb_mac_header(skb),
1586 dev->hard_header_len, true);
1587 }
1588 }
1589 #endif
1590 }
1591
1592 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1593 {
1594 struct fnhe_hash_bucket *hash;
1595 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1596 u32 hval = fnhe_hashfun(daddr);
1597
1598 spin_lock_bh(&fnhe_lock);
1599
1600 hash = rcu_dereference_protected(nh->nh_exceptions,
1601 lockdep_is_held(&fnhe_lock));
1602 hash += hval;
1603
1604 fnhe_p = &hash->chain;
1605 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1606 while (fnhe) {
1607 if (fnhe->fnhe_daddr == daddr) {
1608 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1609 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1610 fnhe_flush_routes(fnhe);
1611 kfree_rcu(fnhe, rcu);
1612 break;
1613 }
1614 fnhe_p = &fnhe->fnhe_next;
1615 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1616 lockdep_is_held(&fnhe_lock));
1617 }
1618
1619 spin_unlock_bh(&fnhe_lock);
1620 }
1621
1622 static void set_lwt_redirect(struct rtable *rth)
1623 {
1624 if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1625 rth->dst.lwtstate->orig_output = rth->dst.output;
1626 rth->dst.output = lwtunnel_output;
1627 }
1628
1629 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1630 rth->dst.lwtstate->orig_input = rth->dst.input;
1631 rth->dst.input = lwtunnel_input;
1632 }
1633 }
1634
1635 /* called in rcu_read_lock() section */
1636 static int __mkroute_input(struct sk_buff *skb,
1637 const struct fib_result *res,
1638 struct in_device *in_dev,
1639 __be32 daddr, __be32 saddr, u32 tos)
1640 {
1641 struct fib_nh_exception *fnhe;
1642 struct rtable *rth;
1643 int err;
1644 struct in_device *out_dev;
1645 bool do_cache;
1646 u32 itag = 0;
1647
1648 /* get a working reference to the output device */
1649 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1650 if (!out_dev) {
1651 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1652 return -EINVAL;
1653 }
1654
1655 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1656 in_dev->dev, in_dev, &itag);
1657 if (err < 0) {
1658 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1659 saddr);
1660
1661 goto cleanup;
1662 }
1663
1664 do_cache = res->fi && !itag;
1665 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1666 skb->protocol == htons(ETH_P_IP) &&
1667 (IN_DEV_SHARED_MEDIA(out_dev) ||
1668 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1669 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1670
1671 if (skb->protocol != htons(ETH_P_IP)) {
1672 /* Not IP (i.e. ARP). Do not create route, if it is
1673 * invalid for proxy arp. DNAT routes are always valid.
1674 *
1675 * Proxy arp feature have been extended to allow, ARP
1676 * replies back to the same interface, to support
1677 * Private VLAN switch technologies. See arp.c.
1678 */
1679 if (out_dev == in_dev &&
1680 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1681 err = -EINVAL;
1682 goto cleanup;
1683 }
1684 }
1685
1686 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1687 if (do_cache) {
1688 if (fnhe) {
1689 rth = rcu_dereference(fnhe->fnhe_rth_input);
1690 if (rth && rth->dst.expires &&
1691 time_after(jiffies, rth->dst.expires)) {
1692 ip_del_fnhe(&FIB_RES_NH(*res), daddr);
1693 fnhe = NULL;
1694 } else {
1695 goto rt_cache;
1696 }
1697 }
1698
1699 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1700
1701 rt_cache:
1702 if (rt_cache_valid(rth)) {
1703 skb_dst_set_noref(skb, &rth->dst);
1704 goto out;
1705 }
1706 }
1707
1708 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1709 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1710 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1711 if (!rth) {
1712 err = -ENOBUFS;
1713 goto cleanup;
1714 }
1715
1716 rth->rt_is_input = 1;
1717 if (res->table)
1718 rth->rt_table_id = res->table->tb_id;
1719 RT_CACHE_STAT_INC(in_slow_tot);
1720
1721 rth->dst.input = ip_forward;
1722
1723 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag);
1724 set_lwt_redirect(rth);
1725 skb_dst_set(skb, &rth->dst);
1726 out:
1727 err = 0;
1728 cleanup:
1729 return err;
1730 }
1731
1732 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1733 /* To make ICMP packets follow the right flow, the multipath hash is
1734 * calculated from the inner IP addresses.
1735 */
1736 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1737 struct flow_keys *hash_keys)
1738 {
1739 const struct iphdr *outer_iph = ip_hdr(skb);
1740 const struct iphdr *inner_iph;
1741 const struct icmphdr *icmph;
1742 struct iphdr _inner_iph;
1743 struct icmphdr _icmph;
1744
1745 hash_keys->addrs.v4addrs.src = outer_iph->saddr;
1746 hash_keys->addrs.v4addrs.dst = outer_iph->daddr;
1747 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1748 return;
1749
1750 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1751 return;
1752
1753 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1754 &_icmph);
1755 if (!icmph)
1756 return;
1757
1758 if (icmph->type != ICMP_DEST_UNREACH &&
1759 icmph->type != ICMP_REDIRECT &&
1760 icmph->type != ICMP_TIME_EXCEEDED &&
1761 icmph->type != ICMP_PARAMETERPROB)
1762 return;
1763
1764 inner_iph = skb_header_pointer(skb,
1765 outer_iph->ihl * 4 + sizeof(_icmph),
1766 sizeof(_inner_iph), &_inner_iph);
1767 if (!inner_iph)
1768 return;
1769 hash_keys->addrs.v4addrs.src = inner_iph->saddr;
1770 hash_keys->addrs.v4addrs.dst = inner_iph->daddr;
1771 }
1772
1773 /* if skb is set it will be used and fl4 can be NULL */
1774 int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4,
1775 const struct sk_buff *skb)
1776 {
1777 struct net *net = fi->fib_net;
1778 struct flow_keys hash_keys;
1779 u32 mhash;
1780
1781 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1782 case 0:
1783 memset(&hash_keys, 0, sizeof(hash_keys));
1784 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1785 if (skb) {
1786 ip_multipath_l3_keys(skb, &hash_keys);
1787 } else {
1788 hash_keys.addrs.v4addrs.src = fl4->saddr;
1789 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1790 }
1791 break;
1792 case 1:
1793 /* skb is currently provided only when forwarding */
1794 if (skb) {
1795 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1796 struct flow_keys keys;
1797
1798 /* short-circuit if we already have L4 hash present */
1799 if (skb->l4_hash)
1800 return skb_get_hash_raw(skb) >> 1;
1801 memset(&hash_keys, 0, sizeof(hash_keys));
1802 skb_flow_dissect_flow_keys(skb, &keys, flag);
1803 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1804 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1805 hash_keys.ports.src = keys.ports.src;
1806 hash_keys.ports.dst = keys.ports.dst;
1807 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1808 } else {
1809 memset(&hash_keys, 0, sizeof(hash_keys));
1810 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1811 hash_keys.addrs.v4addrs.src = fl4->saddr;
1812 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1813 hash_keys.ports.src = fl4->fl4_sport;
1814 hash_keys.ports.dst = fl4->fl4_dport;
1815 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1816 }
1817 break;
1818 }
1819 mhash = flow_hash_from_keys(&hash_keys);
1820
1821 return mhash >> 1;
1822 }
1823 EXPORT_SYMBOL_GPL(fib_multipath_hash);
1824 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1825
1826 static int ip_mkroute_input(struct sk_buff *skb,
1827 struct fib_result *res,
1828 struct in_device *in_dev,
1829 __be32 daddr, __be32 saddr, u32 tos)
1830 {
1831 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1832 if (res->fi && res->fi->fib_nhs > 1) {
1833 int h = fib_multipath_hash(res->fi, NULL, skb);
1834
1835 fib_select_multipath(res, h);
1836 }
1837 #endif
1838
1839 /* create a routing cache entry */
1840 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1841 }
1842
1843 /*
1844 * NOTE. We drop all the packets that has local source
1845 * addresses, because every properly looped back packet
1846 * must have correct destination already attached by output routine.
1847 *
1848 * Such approach solves two big problems:
1849 * 1. Not simplex devices are handled properly.
1850 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1851 * called with rcu_read_lock()
1852 */
1853
1854 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1855 u8 tos, struct net_device *dev)
1856 {
1857 struct fib_result res;
1858 struct in_device *in_dev = __in_dev_get_rcu(dev);
1859 struct ip_tunnel_info *tun_info;
1860 struct flowi4 fl4;
1861 unsigned int flags = 0;
1862 u32 itag = 0;
1863 struct rtable *rth;
1864 int err = -EINVAL;
1865 struct net *net = dev_net(dev);
1866 bool do_cache;
1867
1868 /* IP on this device is disabled. */
1869
1870 if (!in_dev)
1871 goto out;
1872
1873 /* Check for the most weird martians, which can be not detected
1874 by fib_lookup.
1875 */
1876
1877 tun_info = skb_tunnel_info(skb);
1878 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1879 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1880 else
1881 fl4.flowi4_tun_key.tun_id = 0;
1882 skb_dst_drop(skb);
1883
1884 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1885 goto martian_source;
1886
1887 res.fi = NULL;
1888 res.table = NULL;
1889 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1890 goto brd_input;
1891
1892 /* Accept zero addresses only to limited broadcast;
1893 * I even do not know to fix it or not. Waiting for complains :-)
1894 */
1895 if (ipv4_is_zeronet(saddr))
1896 goto martian_source;
1897
1898 if (ipv4_is_zeronet(daddr))
1899 goto martian_destination;
1900
1901 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1902 * and call it once if daddr or/and saddr are loopback addresses
1903 */
1904 if (ipv4_is_loopback(daddr)) {
1905 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1906 goto martian_destination;
1907 } else if (ipv4_is_loopback(saddr)) {
1908 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1909 goto martian_source;
1910 }
1911
1912 /*
1913 * Now we are ready to route packet.
1914 */
1915 fl4.flowi4_oif = 0;
1916 fl4.flowi4_iif = dev->ifindex;
1917 fl4.flowi4_mark = skb->mark;
1918 fl4.flowi4_tos = tos;
1919 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1920 fl4.flowi4_flags = 0;
1921 fl4.daddr = daddr;
1922 fl4.saddr = saddr;
1923 fl4.flowi4_uid = sock_net_uid(net, NULL);
1924 err = fib_lookup(net, &fl4, &res, 0);
1925 if (err != 0) {
1926 if (!IN_DEV_FORWARD(in_dev))
1927 err = -EHOSTUNREACH;
1928 goto no_route;
1929 }
1930
1931 if (res.type == RTN_BROADCAST)
1932 goto brd_input;
1933
1934 if (res.type == RTN_LOCAL) {
1935 err = fib_validate_source(skb, saddr, daddr, tos,
1936 0, dev, in_dev, &itag);
1937 if (err < 0)
1938 goto martian_source;
1939 goto local_input;
1940 }
1941
1942 if (!IN_DEV_FORWARD(in_dev)) {
1943 err = -EHOSTUNREACH;
1944 goto no_route;
1945 }
1946 if (res.type != RTN_UNICAST)
1947 goto martian_destination;
1948
1949 err = ip_mkroute_input(skb, &res, in_dev, daddr, saddr, tos);
1950 out: return err;
1951
1952 brd_input:
1953 if (skb->protocol != htons(ETH_P_IP))
1954 goto e_inval;
1955
1956 if (!ipv4_is_zeronet(saddr)) {
1957 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1958 in_dev, &itag);
1959 if (err < 0)
1960 goto martian_source;
1961 }
1962 flags |= RTCF_BROADCAST;
1963 res.type = RTN_BROADCAST;
1964 RT_CACHE_STAT_INC(in_brd);
1965
1966 local_input:
1967 do_cache = false;
1968 if (res.fi) {
1969 if (!itag) {
1970 rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
1971 if (rt_cache_valid(rth)) {
1972 skb_dst_set_noref(skb, &rth->dst);
1973 err = 0;
1974 goto out;
1975 }
1976 do_cache = true;
1977 }
1978 }
1979
1980 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
1981 flags | RTCF_LOCAL, res.type,
1982 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
1983 if (!rth)
1984 goto e_nobufs;
1985
1986 rth->dst.output= ip_rt_bug;
1987 #ifdef CONFIG_IP_ROUTE_CLASSID
1988 rth->dst.tclassid = itag;
1989 #endif
1990 rth->rt_is_input = 1;
1991 if (res.table)
1992 rth->rt_table_id = res.table->tb_id;
1993
1994 RT_CACHE_STAT_INC(in_slow_tot);
1995 if (res.type == RTN_UNREACHABLE) {
1996 rth->dst.input= ip_error;
1997 rth->dst.error= -err;
1998 rth->rt_flags &= ~RTCF_LOCAL;
1999 }
2000
2001 if (do_cache) {
2002 struct fib_nh *nh = &FIB_RES_NH(res);
2003
2004 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
2005 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2006 WARN_ON(rth->dst.input == lwtunnel_input);
2007 rth->dst.lwtstate->orig_input = rth->dst.input;
2008 rth->dst.input = lwtunnel_input;
2009 }
2010
2011 if (unlikely(!rt_cache_route(nh, rth))) {
2012 rth->dst.flags |= DST_NOCACHE;
2013 rt_add_uncached_list(rth);
2014 }
2015 }
2016 skb_dst_set(skb, &rth->dst);
2017 err = 0;
2018 goto out;
2019
2020 no_route:
2021 RT_CACHE_STAT_INC(in_no_route);
2022 res.type = RTN_UNREACHABLE;
2023 res.fi = NULL;
2024 res.table = NULL;
2025 goto local_input;
2026
2027 /*
2028 * Do not cache martian addresses: they should be logged (RFC1812)
2029 */
2030 martian_destination:
2031 RT_CACHE_STAT_INC(in_martian_dst);
2032 #ifdef CONFIG_IP_ROUTE_VERBOSE
2033 if (IN_DEV_LOG_MARTIANS(in_dev))
2034 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2035 &daddr, &saddr, dev->name);
2036 #endif
2037
2038 e_inval:
2039 err = -EINVAL;
2040 goto out;
2041
2042 e_nobufs:
2043 err = -ENOBUFS;
2044 goto out;
2045
2046 martian_source:
2047 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2048 goto out;
2049 }
2050
2051 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2052 u8 tos, struct net_device *dev)
2053 {
2054 int res;
2055
2056 tos &= IPTOS_RT_MASK;
2057 rcu_read_lock();
2058
2059 /* Multicast recognition logic is moved from route cache to here.
2060 The problem was that too many Ethernet cards have broken/missing
2061 hardware multicast filters :-( As result the host on multicasting
2062 network acquires a lot of useless route cache entries, sort of
2063 SDR messages from all the world. Now we try to get rid of them.
2064 Really, provided software IP multicast filter is organized
2065 reasonably (at least, hashed), it does not result in a slowdown
2066 comparing with route cache reject entries.
2067 Note, that multicast routers are not affected, because
2068 route cache entry is created eventually.
2069 */
2070 if (ipv4_is_multicast(daddr)) {
2071 struct in_device *in_dev = __in_dev_get_rcu(dev);
2072 int our = 0;
2073
2074 if (in_dev)
2075 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2076 ip_hdr(skb)->protocol);
2077
2078 /* check l3 master if no match yet */
2079 if ((!in_dev || !our) && netif_is_l3_slave(dev)) {
2080 struct in_device *l3_in_dev;
2081
2082 l3_in_dev = __in_dev_get_rcu(skb->dev);
2083 if (l3_in_dev)
2084 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2085 ip_hdr(skb)->protocol);
2086 }
2087
2088 res = -EINVAL;
2089 if (our
2090 #ifdef CONFIG_IP_MROUTE
2091 ||
2092 (!ipv4_is_local_multicast(daddr) &&
2093 IN_DEV_MFORWARD(in_dev))
2094 #endif
2095 ) {
2096 res = ip_route_input_mc(skb, daddr, saddr,
2097 tos, dev, our);
2098 }
2099 rcu_read_unlock();
2100 return res;
2101 }
2102 res = ip_route_input_slow(skb, daddr, saddr, tos, dev);
2103 rcu_read_unlock();
2104 return res;
2105 }
2106 EXPORT_SYMBOL(ip_route_input_noref);
2107
2108 /* called with rcu_read_lock() */
2109 static struct rtable *__mkroute_output(const struct fib_result *res,
2110 const struct flowi4 *fl4, int orig_oif,
2111 struct net_device *dev_out,
2112 unsigned int flags)
2113 {
2114 struct fib_info *fi = res->fi;
2115 struct fib_nh_exception *fnhe;
2116 struct in_device *in_dev;
2117 u16 type = res->type;
2118 struct rtable *rth;
2119 bool do_cache;
2120
2121 in_dev = __in_dev_get_rcu(dev_out);
2122 if (!in_dev)
2123 return ERR_PTR(-EINVAL);
2124
2125 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2126 if (ipv4_is_loopback(fl4->saddr) &&
2127 !(dev_out->flags & IFF_LOOPBACK) &&
2128 !netif_is_l3_master(dev_out))
2129 return ERR_PTR(-EINVAL);
2130
2131 if (ipv4_is_lbcast(fl4->daddr))
2132 type = RTN_BROADCAST;
2133 else if (ipv4_is_multicast(fl4->daddr))
2134 type = RTN_MULTICAST;
2135 else if (ipv4_is_zeronet(fl4->daddr))
2136 return ERR_PTR(-EINVAL);
2137
2138 if (dev_out->flags & IFF_LOOPBACK)
2139 flags |= RTCF_LOCAL;
2140
2141 do_cache = true;
2142 if (type == RTN_BROADCAST) {
2143 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2144 fi = NULL;
2145 } else if (type == RTN_MULTICAST) {
2146 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2147 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2148 fl4->flowi4_proto))
2149 flags &= ~RTCF_LOCAL;
2150 else
2151 do_cache = false;
2152 /* If multicast route do not exist use
2153 * default one, but do not gateway in this case.
2154 * Yes, it is hack.
2155 */
2156 if (fi && res->prefixlen < 4)
2157 fi = NULL;
2158 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2159 (orig_oif != dev_out->ifindex)) {
2160 /* For local routes that require a particular output interface
2161 * we do not want to cache the result. Caching the result
2162 * causes incorrect behaviour when there are multiple source
2163 * addresses on the interface, the end result being that if the
2164 * intended recipient is waiting on that interface for the
2165 * packet he won't receive it because it will be delivered on
2166 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2167 * be set to the loopback interface as well.
2168 */
2169 fi = NULL;
2170 }
2171
2172 fnhe = NULL;
2173 do_cache &= fi != NULL;
2174 if (do_cache) {
2175 struct rtable __rcu **prth;
2176 struct fib_nh *nh = &FIB_RES_NH(*res);
2177
2178 fnhe = find_exception(nh, fl4->daddr);
2179 if (fnhe) {
2180 prth = &fnhe->fnhe_rth_output;
2181 rth = rcu_dereference(*prth);
2182 if (rth && rth->dst.expires &&
2183 time_after(jiffies, rth->dst.expires)) {
2184 ip_del_fnhe(nh, fl4->daddr);
2185 fnhe = NULL;
2186 } else {
2187 goto rt_cache;
2188 }
2189 }
2190
2191 if (unlikely(fl4->flowi4_flags &
2192 FLOWI_FLAG_KNOWN_NH &&
2193 !(nh->nh_gw &&
2194 nh->nh_scope == RT_SCOPE_LINK))) {
2195 do_cache = false;
2196 goto add;
2197 }
2198 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2199 rth = rcu_dereference(*prth);
2200
2201 rt_cache:
2202 if (rt_cache_valid(rth)) {
2203 dst_hold(&rth->dst);
2204 return rth;
2205 }
2206 }
2207
2208 add:
2209 rth = rt_dst_alloc(dev_out, flags, type,
2210 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2211 IN_DEV_CONF_GET(in_dev, NOXFRM),
2212 do_cache);
2213 if (!rth)
2214 return ERR_PTR(-ENOBUFS);
2215
2216 rth->rt_iif = orig_oif ? : 0;
2217 if (res->table)
2218 rth->rt_table_id = res->table->tb_id;
2219
2220 RT_CACHE_STAT_INC(out_slow_tot);
2221
2222 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2223 if (flags & RTCF_LOCAL &&
2224 !(dev_out->flags & IFF_LOOPBACK)) {
2225 rth->dst.output = ip_mc_output;
2226 RT_CACHE_STAT_INC(out_slow_mc);
2227 }
2228 #ifdef CONFIG_IP_MROUTE
2229 if (type == RTN_MULTICAST) {
2230 if (IN_DEV_MFORWARD(in_dev) &&
2231 !ipv4_is_local_multicast(fl4->daddr)) {
2232 rth->dst.input = ip_mr_input;
2233 rth->dst.output = ip_mc_output;
2234 }
2235 }
2236 #endif
2237 }
2238
2239 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0);
2240 set_lwt_redirect(rth);
2241
2242 return rth;
2243 }
2244
2245 /*
2246 * Major route resolver routine.
2247 */
2248
2249 struct rtable *__ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2250 const struct sk_buff *skb)
2251 {
2252 struct net_device *dev_out = NULL;
2253 __u8 tos = RT_FL_TOS(fl4);
2254 unsigned int flags = 0;
2255 struct fib_result res;
2256 struct rtable *rth;
2257 int orig_oif;
2258 int err = -ENETUNREACH;
2259
2260 res.tclassid = 0;
2261 res.fi = NULL;
2262 res.table = NULL;
2263
2264 orig_oif = fl4->flowi4_oif;
2265
2266 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2267 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2268 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2269 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2270
2271 rcu_read_lock();
2272 if (fl4->saddr) {
2273 rth = ERR_PTR(-EINVAL);
2274 if (ipv4_is_multicast(fl4->saddr) ||
2275 ipv4_is_lbcast(fl4->saddr) ||
2276 ipv4_is_zeronet(fl4->saddr))
2277 goto out;
2278
2279 /* I removed check for oif == dev_out->oif here.
2280 It was wrong for two reasons:
2281 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2282 is assigned to multiple interfaces.
2283 2. Moreover, we are allowed to send packets with saddr
2284 of another iface. --ANK
2285 */
2286
2287 if (fl4->flowi4_oif == 0 &&
2288 (ipv4_is_multicast(fl4->daddr) ||
2289 ipv4_is_lbcast(fl4->daddr))) {
2290 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2291 dev_out = __ip_dev_find(net, fl4->saddr, false);
2292 if (!dev_out)
2293 goto out;
2294
2295 /* Special hack: user can direct multicasts
2296 and limited broadcast via necessary interface
2297 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2298 This hack is not just for fun, it allows
2299 vic,vat and friends to work.
2300 They bind socket to loopback, set ttl to zero
2301 and expect that it will work.
2302 From the viewpoint of routing cache they are broken,
2303 because we are not allowed to build multicast path
2304 with loopback source addr (look, routing cache
2305 cannot know, that ttl is zero, so that packet
2306 will not leave this host and route is valid).
2307 Luckily, this hack is good workaround.
2308 */
2309
2310 fl4->flowi4_oif = dev_out->ifindex;
2311 goto make_route;
2312 }
2313
2314 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2315 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2316 if (!__ip_dev_find(net, fl4->saddr, false))
2317 goto out;
2318 }
2319 }
2320
2321
2322 if (fl4->flowi4_oif) {
2323 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2324 rth = ERR_PTR(-ENODEV);
2325 if (!dev_out)
2326 goto out;
2327
2328 /* RACE: Check return value of inet_select_addr instead. */
2329 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2330 rth = ERR_PTR(-ENETUNREACH);
2331 goto out;
2332 }
2333 if (ipv4_is_local_multicast(fl4->daddr) ||
2334 ipv4_is_lbcast(fl4->daddr) ||
2335 fl4->flowi4_proto == IPPROTO_IGMP) {
2336 if (!fl4->saddr)
2337 fl4->saddr = inet_select_addr(dev_out, 0,
2338 RT_SCOPE_LINK);
2339 goto make_route;
2340 }
2341 if (!fl4->saddr) {
2342 if (ipv4_is_multicast(fl4->daddr))
2343 fl4->saddr = inet_select_addr(dev_out, 0,
2344 fl4->flowi4_scope);
2345 else if (!fl4->daddr)
2346 fl4->saddr = inet_select_addr(dev_out, 0,
2347 RT_SCOPE_HOST);
2348 }
2349 }
2350
2351 if (!fl4->daddr) {
2352 fl4->daddr = fl4->saddr;
2353 if (!fl4->daddr)
2354 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2355 dev_out = net->loopback_dev;
2356 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2357 res.type = RTN_LOCAL;
2358 flags |= RTCF_LOCAL;
2359 goto make_route;
2360 }
2361
2362 err = fib_lookup(net, fl4, &res, 0);
2363 if (err) {
2364 res.fi = NULL;
2365 res.table = NULL;
2366 if (fl4->flowi4_oif &&
2367 (ipv4_is_multicast(fl4->daddr) ||
2368 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2369 /* Apparently, routing tables are wrong. Assume,
2370 that the destination is on link.
2371
2372 WHY? DW.
2373 Because we are allowed to send to iface
2374 even if it has NO routes and NO assigned
2375 addresses. When oif is specified, routing
2376 tables are looked up with only one purpose:
2377 to catch if destination is gatewayed, rather than
2378 direct. Moreover, if MSG_DONTROUTE is set,
2379 we send packet, ignoring both routing tables
2380 and ifaddr state. --ANK
2381
2382
2383 We could make it even if oif is unknown,
2384 likely IPv6, but we do not.
2385 */
2386
2387 if (fl4->saddr == 0)
2388 fl4->saddr = inet_select_addr(dev_out, 0,
2389 RT_SCOPE_LINK);
2390 res.type = RTN_UNICAST;
2391 goto make_route;
2392 }
2393 rth = ERR_PTR(err);
2394 goto out;
2395 }
2396
2397 if (res.type == RTN_LOCAL) {
2398 if (!fl4->saddr) {
2399 if (res.fi->fib_prefsrc)
2400 fl4->saddr = res.fi->fib_prefsrc;
2401 else
2402 fl4->saddr = fl4->daddr;
2403 }
2404
2405 /* L3 master device is the loopback for that domain */
2406 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(res)) ? :
2407 net->loopback_dev;
2408 fl4->flowi4_oif = dev_out->ifindex;
2409 flags |= RTCF_LOCAL;
2410 goto make_route;
2411 }
2412
2413 fib_select_path(net, &res, fl4, skb);
2414
2415 dev_out = FIB_RES_DEV(res);
2416 fl4->flowi4_oif = dev_out->ifindex;
2417
2418
2419 make_route:
2420 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags);
2421
2422 out:
2423 rcu_read_unlock();
2424 return rth;
2425 }
2426 EXPORT_SYMBOL_GPL(__ip_route_output_key_hash);
2427
2428 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2429 {
2430 return NULL;
2431 }
2432
2433 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2434 {
2435 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2436
2437 return mtu ? : dst->dev->mtu;
2438 }
2439
2440 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2441 struct sk_buff *skb, u32 mtu)
2442 {
2443 }
2444
2445 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2446 struct sk_buff *skb)
2447 {
2448 }
2449
2450 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2451 unsigned long old)
2452 {
2453 return NULL;
2454 }
2455
2456 static struct dst_ops ipv4_dst_blackhole_ops = {
2457 .family = AF_INET,
2458 .check = ipv4_blackhole_dst_check,
2459 .mtu = ipv4_blackhole_mtu,
2460 .default_advmss = ipv4_default_advmss,
2461 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2462 .redirect = ipv4_rt_blackhole_redirect,
2463 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2464 .neigh_lookup = ipv4_neigh_lookup,
2465 };
2466
2467 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2468 {
2469 struct rtable *ort = (struct rtable *) dst_orig;
2470 struct rtable *rt;
2471
2472 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0);
2473 if (rt) {
2474 struct dst_entry *new = &rt->dst;
2475
2476 new->__use = 1;
2477 new->input = dst_discard;
2478 new->output = dst_discard_out;
2479
2480 new->dev = ort->dst.dev;
2481 if (new->dev)
2482 dev_hold(new->dev);
2483
2484 rt->rt_is_input = ort->rt_is_input;
2485 rt->rt_iif = ort->rt_iif;
2486 rt->rt_pmtu = ort->rt_pmtu;
2487
2488 rt->rt_genid = rt_genid_ipv4(net);
2489 rt->rt_flags = ort->rt_flags;
2490 rt->rt_type = ort->rt_type;
2491 rt->rt_gateway = ort->rt_gateway;
2492 rt->rt_uses_gateway = ort->rt_uses_gateway;
2493
2494 INIT_LIST_HEAD(&rt->rt_uncached);
2495 dst_free(new);
2496 }
2497
2498 dst_release(dst_orig);
2499
2500 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2501 }
2502
2503 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2504 const struct sock *sk)
2505 {
2506 struct rtable *rt = __ip_route_output_key(net, flp4);
2507
2508 if (IS_ERR(rt))
2509 return rt;
2510
2511 if (flp4->flowi4_proto)
2512 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2513 flowi4_to_flowi(flp4),
2514 sk, 0);
2515
2516 return rt;
2517 }
2518 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2519
2520 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2521 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2522 u32 seq, int event)
2523 {
2524 struct rtable *rt = skb_rtable(skb);
2525 struct rtmsg *r;
2526 struct nlmsghdr *nlh;
2527 unsigned long expires = 0;
2528 u32 error;
2529 u32 metrics[RTAX_MAX];
2530
2531 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), 0);
2532 if (!nlh)
2533 return -EMSGSIZE;
2534
2535 r = nlmsg_data(nlh);
2536 r->rtm_family = AF_INET;
2537 r->rtm_dst_len = 32;
2538 r->rtm_src_len = 0;
2539 r->rtm_tos = fl4->flowi4_tos;
2540 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2541 if (nla_put_u32(skb, RTA_TABLE, table_id))
2542 goto nla_put_failure;
2543 r->rtm_type = rt->rt_type;
2544 r->rtm_scope = RT_SCOPE_UNIVERSE;
2545 r->rtm_protocol = RTPROT_UNSPEC;
2546 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2547 if (rt->rt_flags & RTCF_NOTIFY)
2548 r->rtm_flags |= RTM_F_NOTIFY;
2549 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2550 r->rtm_flags |= RTCF_DOREDIRECT;
2551
2552 if (nla_put_in_addr(skb, RTA_DST, dst))
2553 goto nla_put_failure;
2554 if (src) {
2555 r->rtm_src_len = 32;
2556 if (nla_put_in_addr(skb, RTA_SRC, src))
2557 goto nla_put_failure;
2558 }
2559 if (rt->dst.dev &&
2560 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2561 goto nla_put_failure;
2562 #ifdef CONFIG_IP_ROUTE_CLASSID
2563 if (rt->dst.tclassid &&
2564 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2565 goto nla_put_failure;
2566 #endif
2567 if (!rt_is_input_route(rt) &&
2568 fl4->saddr != src) {
2569 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2570 goto nla_put_failure;
2571 }
2572 if (rt->rt_uses_gateway &&
2573 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2574 goto nla_put_failure;
2575
2576 expires = rt->dst.expires;
2577 if (expires) {
2578 unsigned long now = jiffies;
2579
2580 if (time_before(now, expires))
2581 expires -= now;
2582 else
2583 expires = 0;
2584 }
2585
2586 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2587 if (rt->rt_pmtu && expires)
2588 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2589 if (rtnetlink_put_metrics(skb, metrics) < 0)
2590 goto nla_put_failure;
2591
2592 if (fl4->flowi4_mark &&
2593 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2594 goto nla_put_failure;
2595
2596 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2597 nla_put_u32(skb, RTA_UID,
2598 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2599 goto nla_put_failure;
2600
2601 error = rt->dst.error;
2602
2603 if (rt_is_input_route(rt)) {
2604 #ifdef CONFIG_IP_MROUTE
2605 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2606 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2607 int err = ipmr_get_route(net, skb,
2608 fl4->saddr, fl4->daddr,
2609 r, portid);
2610
2611 if (err <= 0) {
2612 if (err == 0)
2613 return 0;
2614 goto nla_put_failure;
2615 }
2616 } else
2617 #endif
2618 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2619 goto nla_put_failure;
2620 }
2621
2622 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2623 goto nla_put_failure;
2624
2625 nlmsg_end(skb, nlh);
2626 return 0;
2627
2628 nla_put_failure:
2629 nlmsg_cancel(skb, nlh);
2630 return -EMSGSIZE;
2631 }
2632
2633 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2634 struct netlink_ext_ack *extack)
2635 {
2636 struct net *net = sock_net(in_skb->sk);
2637 struct rtmsg *rtm;
2638 struct nlattr *tb[RTA_MAX+1];
2639 struct rtable *rt = NULL;
2640 struct flowi4 fl4;
2641 __be32 dst = 0;
2642 __be32 src = 0;
2643 u32 iif;
2644 int err;
2645 int mark;
2646 struct sk_buff *skb;
2647 u32 table_id = RT_TABLE_MAIN;
2648 kuid_t uid;
2649
2650 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy,
2651 extack);
2652 if (err < 0)
2653 goto errout;
2654
2655 rtm = nlmsg_data(nlh);
2656
2657 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2658 if (!skb) {
2659 err = -ENOBUFS;
2660 goto errout;
2661 }
2662
2663 /* Reserve room for dummy headers, this skb can pass
2664 through good chunk of routing engine.
2665 */
2666 skb_reset_mac_header(skb);
2667 skb_reset_network_header(skb);
2668
2669 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2670 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2671 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2672 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2673 if (tb[RTA_UID])
2674 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2675 else
2676 uid = (iif ? INVALID_UID : current_uid());
2677
2678 /* Bugfix: need to give ip_route_input enough of an IP header to
2679 * not gag.
2680 */
2681 ip_hdr(skb)->protocol = IPPROTO_UDP;
2682 ip_hdr(skb)->saddr = src;
2683 ip_hdr(skb)->daddr = dst;
2684
2685 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2686
2687 memset(&fl4, 0, sizeof(fl4));
2688 fl4.daddr = dst;
2689 fl4.saddr = src;
2690 fl4.flowi4_tos = rtm->rtm_tos;
2691 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2692 fl4.flowi4_mark = mark;
2693 fl4.flowi4_uid = uid;
2694
2695 if (iif) {
2696 struct net_device *dev;
2697
2698 dev = __dev_get_by_index(net, iif);
2699 if (!dev) {
2700 err = -ENODEV;
2701 goto errout_free;
2702 }
2703
2704 skb->protocol = htons(ETH_P_IP);
2705 skb->dev = dev;
2706 skb->mark = mark;
2707 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev);
2708
2709 rt = skb_rtable(skb);
2710 if (err == 0 && rt->dst.error)
2711 err = -rt->dst.error;
2712 } else {
2713 rt = ip_route_output_key(net, &fl4);
2714
2715 err = 0;
2716 if (IS_ERR(rt))
2717 err = PTR_ERR(rt);
2718 }
2719
2720 if (err)
2721 goto errout_free;
2722
2723 skb_dst_set(skb, &rt->dst);
2724 if (rtm->rtm_flags & RTM_F_NOTIFY)
2725 rt->rt_flags |= RTCF_NOTIFY;
2726
2727 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2728 table_id = rt->rt_table_id;
2729
2730 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2731 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2732 RTM_NEWROUTE);
2733 if (err < 0)
2734 goto errout_free;
2735
2736 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2737 errout:
2738 return err;
2739
2740 errout_free:
2741 kfree_skb(skb);
2742 goto errout;
2743 }
2744
2745 void ip_rt_multicast_event(struct in_device *in_dev)
2746 {
2747 rt_cache_flush(dev_net(in_dev->dev));
2748 }
2749
2750 #ifdef CONFIG_SYSCTL
2751 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2752 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2753 static int ip_rt_gc_elasticity __read_mostly = 8;
2754
2755 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2756 void __user *buffer,
2757 size_t *lenp, loff_t *ppos)
2758 {
2759 struct net *net = (struct net *)__ctl->extra1;
2760
2761 if (write) {
2762 rt_cache_flush(net);
2763 fnhe_genid_bump(net);
2764 return 0;
2765 }
2766
2767 return -EINVAL;
2768 }
2769
2770 static struct ctl_table ipv4_route_table[] = {
2771 {
2772 .procname = "gc_thresh",
2773 .data = &ipv4_dst_ops.gc_thresh,
2774 .maxlen = sizeof(int),
2775 .mode = 0644,
2776 .proc_handler = proc_dointvec,
2777 },
2778 {
2779 .procname = "max_size",
2780 .data = &ip_rt_max_size,
2781 .maxlen = sizeof(int),
2782 .mode = 0644,
2783 .proc_handler = proc_dointvec,
2784 },
2785 {
2786 /* Deprecated. Use gc_min_interval_ms */
2787
2788 .procname = "gc_min_interval",
2789 .data = &ip_rt_gc_min_interval,
2790 .maxlen = sizeof(int),
2791 .mode = 0644,
2792 .proc_handler = proc_dointvec_jiffies,
2793 },
2794 {
2795 .procname = "gc_min_interval_ms",
2796 .data = &ip_rt_gc_min_interval,
2797 .maxlen = sizeof(int),
2798 .mode = 0644,
2799 .proc_handler = proc_dointvec_ms_jiffies,
2800 },
2801 {
2802 .procname = "gc_timeout",
2803 .data = &ip_rt_gc_timeout,
2804 .maxlen = sizeof(int),
2805 .mode = 0644,
2806 .proc_handler = proc_dointvec_jiffies,
2807 },
2808 {
2809 .procname = "gc_interval",
2810 .data = &ip_rt_gc_interval,
2811 .maxlen = sizeof(int),
2812 .mode = 0644,
2813 .proc_handler = proc_dointvec_jiffies,
2814 },
2815 {
2816 .procname = "redirect_load",
2817 .data = &ip_rt_redirect_load,
2818 .maxlen = sizeof(int),
2819 .mode = 0644,
2820 .proc_handler = proc_dointvec,
2821 },
2822 {
2823 .procname = "redirect_number",
2824 .data = &ip_rt_redirect_number,
2825 .maxlen = sizeof(int),
2826 .mode = 0644,
2827 .proc_handler = proc_dointvec,
2828 },
2829 {
2830 .procname = "redirect_silence",
2831 .data = &ip_rt_redirect_silence,
2832 .maxlen = sizeof(int),
2833 .mode = 0644,
2834 .proc_handler = proc_dointvec,
2835 },
2836 {
2837 .procname = "error_cost",
2838 .data = &ip_rt_error_cost,
2839 .maxlen = sizeof(int),
2840 .mode = 0644,
2841 .proc_handler = proc_dointvec,
2842 },
2843 {
2844 .procname = "error_burst",
2845 .data = &ip_rt_error_burst,
2846 .maxlen = sizeof(int),
2847 .mode = 0644,
2848 .proc_handler = proc_dointvec,
2849 },
2850 {
2851 .procname = "gc_elasticity",
2852 .data = &ip_rt_gc_elasticity,
2853 .maxlen = sizeof(int),
2854 .mode = 0644,
2855 .proc_handler = proc_dointvec,
2856 },
2857 {
2858 .procname = "mtu_expires",
2859 .data = &ip_rt_mtu_expires,
2860 .maxlen = sizeof(int),
2861 .mode = 0644,
2862 .proc_handler = proc_dointvec_jiffies,
2863 },
2864 {
2865 .procname = "min_pmtu",
2866 .data = &ip_rt_min_pmtu,
2867 .maxlen = sizeof(int),
2868 .mode = 0644,
2869 .proc_handler = proc_dointvec,
2870 },
2871 {
2872 .procname = "min_adv_mss",
2873 .data = &ip_rt_min_advmss,
2874 .maxlen = sizeof(int),
2875 .mode = 0644,
2876 .proc_handler = proc_dointvec,
2877 },
2878 { }
2879 };
2880
2881 static struct ctl_table ipv4_route_flush_table[] = {
2882 {
2883 .procname = "flush",
2884 .maxlen = sizeof(int),
2885 .mode = 0200,
2886 .proc_handler = ipv4_sysctl_rtcache_flush,
2887 },
2888 { },
2889 };
2890
2891 static __net_init int sysctl_route_net_init(struct net *net)
2892 {
2893 struct ctl_table *tbl;
2894
2895 tbl = ipv4_route_flush_table;
2896 if (!net_eq(net, &init_net)) {
2897 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2898 if (!tbl)
2899 goto err_dup;
2900
2901 /* Don't export sysctls to unprivileged users */
2902 if (net->user_ns != &init_user_ns)
2903 tbl[0].procname = NULL;
2904 }
2905 tbl[0].extra1 = net;
2906
2907 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2908 if (!net->ipv4.route_hdr)
2909 goto err_reg;
2910 return 0;
2911
2912 err_reg:
2913 if (tbl != ipv4_route_flush_table)
2914 kfree(tbl);
2915 err_dup:
2916 return -ENOMEM;
2917 }
2918
2919 static __net_exit void sysctl_route_net_exit(struct net *net)
2920 {
2921 struct ctl_table *tbl;
2922
2923 tbl = net->ipv4.route_hdr->ctl_table_arg;
2924 unregister_net_sysctl_table(net->ipv4.route_hdr);
2925 BUG_ON(tbl == ipv4_route_flush_table);
2926 kfree(tbl);
2927 }
2928
2929 static __net_initdata struct pernet_operations sysctl_route_ops = {
2930 .init = sysctl_route_net_init,
2931 .exit = sysctl_route_net_exit,
2932 };
2933 #endif
2934
2935 static __net_init int rt_genid_init(struct net *net)
2936 {
2937 atomic_set(&net->ipv4.rt_genid, 0);
2938 atomic_set(&net->fnhe_genid, 0);
2939 get_random_bytes(&net->ipv4.dev_addr_genid,
2940 sizeof(net->ipv4.dev_addr_genid));
2941 return 0;
2942 }
2943
2944 static __net_initdata struct pernet_operations rt_genid_ops = {
2945 .init = rt_genid_init,
2946 };
2947
2948 static int __net_init ipv4_inetpeer_init(struct net *net)
2949 {
2950 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
2951
2952 if (!bp)
2953 return -ENOMEM;
2954 inet_peer_base_init(bp);
2955 net->ipv4.peers = bp;
2956 return 0;
2957 }
2958
2959 static void __net_exit ipv4_inetpeer_exit(struct net *net)
2960 {
2961 struct inet_peer_base *bp = net->ipv4.peers;
2962
2963 net->ipv4.peers = NULL;
2964 inetpeer_invalidate_tree(bp);
2965 kfree(bp);
2966 }
2967
2968 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
2969 .init = ipv4_inetpeer_init,
2970 .exit = ipv4_inetpeer_exit,
2971 };
2972
2973 #ifdef CONFIG_IP_ROUTE_CLASSID
2974 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
2975 #endif /* CONFIG_IP_ROUTE_CLASSID */
2976
2977 int __init ip_rt_init(void)
2978 {
2979 int rc = 0;
2980 int cpu;
2981
2982 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
2983 if (!ip_idents)
2984 panic("IP: failed to allocate ip_idents\n");
2985
2986 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
2987
2988 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
2989 if (!ip_tstamps)
2990 panic("IP: failed to allocate ip_tstamps\n");
2991
2992 for_each_possible_cpu(cpu) {
2993 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
2994
2995 INIT_LIST_HEAD(&ul->head);
2996 spin_lock_init(&ul->lock);
2997 }
2998 #ifdef CONFIG_IP_ROUTE_CLASSID
2999 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3000 if (!ip_rt_acct)
3001 panic("IP: failed to allocate ip_rt_acct\n");
3002 #endif
3003
3004 ipv4_dst_ops.kmem_cachep =
3005 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3006 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3007
3008 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3009
3010 if (dst_entries_init(&ipv4_dst_ops) < 0)
3011 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3012
3013 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3014 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3015
3016 ipv4_dst_ops.gc_thresh = ~0;
3017 ip_rt_max_size = INT_MAX;
3018
3019 devinet_init();
3020 ip_fib_init();
3021
3022 if (ip_rt_proc_init())
3023 pr_err("Unable to create route proc files\n");
3024 #ifdef CONFIG_XFRM
3025 xfrm_init();
3026 xfrm4_init();
3027 #endif
3028 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL);
3029
3030 #ifdef CONFIG_SYSCTL
3031 register_pernet_subsys(&sysctl_route_ops);
3032 #endif
3033 register_pernet_subsys(&rt_genid_ops);
3034 register_pernet_subsys(&ipv4_inetpeer_ops);
3035 return rc;
3036 }
3037
3038 #ifdef CONFIG_SYSCTL
3039 /*
3040 * We really need to sanitize the damn ipv4 init order, then all
3041 * this nonsense will go away.
3042 */
3043 void __init ip_static_sysctl_init(void)
3044 {
3045 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3046 }
3047 #endif
Ubuntu Artful kernel mirror