]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - net/ipv4/route.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
UBUNTU: Ubuntu-4.15.0-96.97
[mirror_ubuntu-bionic-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 #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 = READ_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
910 /* Too many ignored redirects; do not send anything
911 * set dst.rate_last to the last seen redirected packet.
912 */
913 if (peer->rate_tokens >= ip_rt_redirect_number) {
914 peer->rate_last = jiffies;
915 goto out_put_peer;
916 }
917
918 /* Check for load limit; set rate_last to the latest sent
919 * redirect.
920 */
921 if (peer->rate_tokens == 0 ||
922 time_after(jiffies,
923 (peer->rate_last +
924 (ip_rt_redirect_load << peer->rate_tokens)))) {
925 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
926
927 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
928 peer->rate_last = jiffies;
929 ++peer->rate_tokens;
930 #ifdef CONFIG_IP_ROUTE_VERBOSE
931 if (log_martians &&
932 peer->rate_tokens == ip_rt_redirect_number)
933 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
934 &ip_hdr(skb)->saddr, inet_iif(skb),
935 &ip_hdr(skb)->daddr, &gw);
936 #endif
937 }
938 out_put_peer:
939 inet_putpeer(peer);
940 }
941
942 static int ip_error(struct sk_buff *skb)
943 {
944 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
945 struct rtable *rt = skb_rtable(skb);
946 struct inet_peer *peer;
947 unsigned long now;
948 struct net *net;
949 bool send;
950 int code;
951
952 /* IP on this device is disabled. */
953 if (!in_dev)
954 goto out;
955
956 net = dev_net(rt->dst.dev);
957 if (!IN_DEV_FORWARD(in_dev)) {
958 switch (rt->dst.error) {
959 case EHOSTUNREACH:
960 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
961 break;
962
963 case ENETUNREACH:
964 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
965 break;
966 }
967 goto out;
968 }
969
970 switch (rt->dst.error) {
971 case EINVAL:
972 default:
973 goto out;
974 case EHOSTUNREACH:
975 code = ICMP_HOST_UNREACH;
976 break;
977 case ENETUNREACH:
978 code = ICMP_NET_UNREACH;
979 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
980 break;
981 case EACCES:
982 code = ICMP_PKT_FILTERED;
983 break;
984 }
985
986 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
987 l3mdev_master_ifindex(skb->dev), 1);
988
989 send = true;
990 if (peer) {
991 now = jiffies;
992 peer->rate_tokens += now - peer->rate_last;
993 if (peer->rate_tokens > ip_rt_error_burst)
994 peer->rate_tokens = ip_rt_error_burst;
995 peer->rate_last = now;
996 if (peer->rate_tokens >= ip_rt_error_cost)
997 peer->rate_tokens -= ip_rt_error_cost;
998 else
999 send = false;
1000 inet_putpeer(peer);
1001 }
1002 if (send)
1003 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1004
1005 out: kfree_skb(skb);
1006 return 0;
1007 }
1008
1009 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1010 {
1011 struct dst_entry *dst = &rt->dst;
1012 struct fib_result res;
1013 bool lock = false;
1014
1015 if (ip_mtu_locked(dst))
1016 return;
1017
1018 if (ipv4_mtu(dst) < mtu)
1019 return;
1020
1021 if (mtu < ip_rt_min_pmtu) {
1022 lock = true;
1023 mtu = ip_rt_min_pmtu;
1024 }
1025
1026 if (rt->rt_pmtu == mtu &&
1027 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1028 return;
1029
1030 rcu_read_lock();
1031 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1032 struct fib_nh *nh = &FIB_RES_NH(res);
1033
1034 update_or_create_fnhe(nh, fl4->daddr, 0, mtu, lock,
1035 jiffies + ip_rt_mtu_expires);
1036 }
1037 rcu_read_unlock();
1038 }
1039
1040 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1041 struct sk_buff *skb, u32 mtu)
1042 {
1043 struct rtable *rt = (struct rtable *) dst;
1044 struct flowi4 fl4;
1045
1046 ip_rt_build_flow_key(&fl4, sk, skb);
1047 __ip_rt_update_pmtu(rt, &fl4, mtu);
1048 }
1049
1050 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1051 int oif, u32 mark, u8 protocol, int flow_flags)
1052 {
1053 const struct iphdr *iph = (const struct iphdr *) skb->data;
1054 struct flowi4 fl4;
1055 struct rtable *rt;
1056
1057 if (!mark)
1058 mark = IP4_REPLY_MARK(net, skb->mark);
1059
1060 __build_flow_key(net, &fl4, NULL, iph, oif,
1061 RT_TOS(iph->tos), protocol, mark, flow_flags);
1062 rt = __ip_route_output_key(net, &fl4);
1063 if (!IS_ERR(rt)) {
1064 __ip_rt_update_pmtu(rt, &fl4, mtu);
1065 ip_rt_put(rt);
1066 }
1067 }
1068 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1069
1070 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1071 {
1072 const struct iphdr *iph = (const struct iphdr *) skb->data;
1073 struct flowi4 fl4;
1074 struct rtable *rt;
1075
1076 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1077
1078 if (!fl4.flowi4_mark)
1079 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1080
1081 rt = __ip_route_output_key(sock_net(sk), &fl4);
1082 if (!IS_ERR(rt)) {
1083 __ip_rt_update_pmtu(rt, &fl4, mtu);
1084 ip_rt_put(rt);
1085 }
1086 }
1087
1088 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1089 {
1090 const struct iphdr *iph = (const struct iphdr *) skb->data;
1091 struct flowi4 fl4;
1092 struct rtable *rt;
1093 struct dst_entry *odst = NULL;
1094 bool new = false;
1095 struct net *net = sock_net(sk);
1096
1097 bh_lock_sock(sk);
1098
1099 if (!ip_sk_accept_pmtu(sk))
1100 goto out;
1101
1102 odst = sk_dst_get(sk);
1103
1104 if (sock_owned_by_user(sk) || !odst) {
1105 __ipv4_sk_update_pmtu(skb, sk, mtu);
1106 goto out;
1107 }
1108
1109 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1110
1111 rt = (struct rtable *)odst;
1112 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1113 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1114 if (IS_ERR(rt))
1115 goto out;
1116
1117 new = true;
1118 }
1119
1120 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1121
1122 if (!dst_check(&rt->dst, 0)) {
1123 if (new)
1124 dst_release(&rt->dst);
1125
1126 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1127 if (IS_ERR(rt))
1128 goto out;
1129
1130 new = true;
1131 }
1132
1133 if (new)
1134 sk_dst_set(sk, &rt->dst);
1135
1136 out:
1137 bh_unlock_sock(sk);
1138 dst_release(odst);
1139 }
1140 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1141
1142 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1143 int oif, u32 mark, u8 protocol, int flow_flags)
1144 {
1145 const struct iphdr *iph = (const struct iphdr *) skb->data;
1146 struct flowi4 fl4;
1147 struct rtable *rt;
1148
1149 __build_flow_key(net, &fl4, NULL, iph, oif,
1150 RT_TOS(iph->tos), protocol, mark, flow_flags);
1151 rt = __ip_route_output_key(net, &fl4);
1152 if (!IS_ERR(rt)) {
1153 __ip_do_redirect(rt, skb, &fl4, false);
1154 ip_rt_put(rt);
1155 }
1156 }
1157 EXPORT_SYMBOL_GPL(ipv4_redirect);
1158
1159 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1160 {
1161 const struct iphdr *iph = (const struct iphdr *) skb->data;
1162 struct flowi4 fl4;
1163 struct rtable *rt;
1164 struct net *net = sock_net(sk);
1165
1166 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1167 rt = __ip_route_output_key(net, &fl4);
1168 if (!IS_ERR(rt)) {
1169 __ip_do_redirect(rt, skb, &fl4, false);
1170 ip_rt_put(rt);
1171 }
1172 }
1173 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1174
1175 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1176 {
1177 struct rtable *rt = (struct rtable *) dst;
1178
1179 /* All IPV4 dsts are created with ->obsolete set to the value
1180 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1181 * into this function always.
1182 *
1183 * When a PMTU/redirect information update invalidates a route,
1184 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1185 * DST_OBSOLETE_DEAD by dst_free().
1186 */
1187 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1188 return NULL;
1189 return dst;
1190 }
1191
1192 static void ipv4_link_failure(struct sk_buff *skb)
1193 {
1194 struct rtable *rt;
1195
1196 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1197
1198 rt = skb_rtable(skb);
1199 if (rt)
1200 dst_set_expires(&rt->dst, 0);
1201 }
1202
1203 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1204 {
1205 pr_debug("%s: %pI4 -> %pI4, %s\n",
1206 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1207 skb->dev ? skb->dev->name : "?");
1208 kfree_skb(skb);
1209 WARN_ON(1);
1210 return 0;
1211 }
1212
1213 /*
1214 We do not cache source address of outgoing interface,
1215 because it is used only by IP RR, TS and SRR options,
1216 so that it out of fast path.
1217
1218 BTW remember: "addr" is allowed to be not aligned
1219 in IP options!
1220 */
1221
1222 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1223 {
1224 __be32 src;
1225
1226 if (rt_is_output_route(rt))
1227 src = ip_hdr(skb)->saddr;
1228 else {
1229 struct fib_result res;
1230 struct flowi4 fl4;
1231 struct iphdr *iph;
1232
1233 iph = ip_hdr(skb);
1234
1235 memset(&fl4, 0, sizeof(fl4));
1236 fl4.daddr = iph->daddr;
1237 fl4.saddr = iph->saddr;
1238 fl4.flowi4_tos = RT_TOS(iph->tos);
1239 fl4.flowi4_oif = rt->dst.dev->ifindex;
1240 fl4.flowi4_iif = skb->dev->ifindex;
1241 fl4.flowi4_mark = skb->mark;
1242
1243 rcu_read_lock();
1244 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1245 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1246 else
1247 src = inet_select_addr(rt->dst.dev,
1248 rt_nexthop(rt, iph->daddr),
1249 RT_SCOPE_UNIVERSE);
1250 rcu_read_unlock();
1251 }
1252 memcpy(addr, &src, 4);
1253 }
1254
1255 #ifdef CONFIG_IP_ROUTE_CLASSID
1256 static void set_class_tag(struct rtable *rt, u32 tag)
1257 {
1258 if (!(rt->dst.tclassid & 0xFFFF))
1259 rt->dst.tclassid |= tag & 0xFFFF;
1260 if (!(rt->dst.tclassid & 0xFFFF0000))
1261 rt->dst.tclassid |= tag & 0xFFFF0000;
1262 }
1263 #endif
1264
1265 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1266 {
1267 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1268 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1269 ip_rt_min_advmss);
1270
1271 return min(advmss, IPV4_MAX_PMTU - header_size);
1272 }
1273
1274 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1275 {
1276 const struct rtable *rt = (const struct rtable *) dst;
1277 unsigned int mtu = rt->rt_pmtu;
1278
1279 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1280 mtu = dst_metric_raw(dst, RTAX_MTU);
1281
1282 if (mtu)
1283 return mtu;
1284
1285 mtu = READ_ONCE(dst->dev->mtu);
1286
1287 if (unlikely(ip_mtu_locked(dst))) {
1288 if (rt->rt_uses_gateway && mtu > 576)
1289 mtu = 576;
1290 }
1291
1292 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1293
1294 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1295 }
1296
1297 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1298 {
1299 struct fnhe_hash_bucket *hash;
1300 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1301 u32 hval = fnhe_hashfun(daddr);
1302
1303 spin_lock_bh(&fnhe_lock);
1304
1305 hash = rcu_dereference_protected(nh->nh_exceptions,
1306 lockdep_is_held(&fnhe_lock));
1307 hash += hval;
1308
1309 fnhe_p = &hash->chain;
1310 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1311 while (fnhe) {
1312 if (fnhe->fnhe_daddr == daddr) {
1313 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1314 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1315 fnhe_flush_routes(fnhe);
1316 kfree_rcu(fnhe, rcu);
1317 break;
1318 }
1319 fnhe_p = &fnhe->fnhe_next;
1320 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1321 lockdep_is_held(&fnhe_lock));
1322 }
1323
1324 spin_unlock_bh(&fnhe_lock);
1325 }
1326
1327 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1328 {
1329 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1330 struct fib_nh_exception *fnhe;
1331 u32 hval;
1332
1333 if (!hash)
1334 return NULL;
1335
1336 hval = fnhe_hashfun(daddr);
1337
1338 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1339 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1340 if (fnhe->fnhe_daddr == daddr) {
1341 if (fnhe->fnhe_expires &&
1342 time_after(jiffies, fnhe->fnhe_expires)) {
1343 ip_del_fnhe(nh, daddr);
1344 break;
1345 }
1346 return fnhe;
1347 }
1348 }
1349 return NULL;
1350 }
1351
1352 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1353 __be32 daddr, const bool do_cache)
1354 {
1355 bool ret = false;
1356
1357 spin_lock_bh(&fnhe_lock);
1358
1359 if (daddr == fnhe->fnhe_daddr) {
1360 struct rtable __rcu **porig;
1361 struct rtable *orig;
1362 int genid = fnhe_genid(dev_net(rt->dst.dev));
1363
1364 if (rt_is_input_route(rt))
1365 porig = &fnhe->fnhe_rth_input;
1366 else
1367 porig = &fnhe->fnhe_rth_output;
1368 orig = rcu_dereference(*porig);
1369
1370 if (fnhe->fnhe_genid != genid) {
1371 fnhe->fnhe_genid = genid;
1372 fnhe->fnhe_gw = 0;
1373 fnhe->fnhe_pmtu = 0;
1374 fnhe->fnhe_expires = 0;
1375 fnhe_flush_routes(fnhe);
1376 orig = NULL;
1377 }
1378 fill_route_from_fnhe(rt, fnhe);
1379 if (!rt->rt_gateway)
1380 rt->rt_gateway = daddr;
1381
1382 if (do_cache) {
1383 dst_hold(&rt->dst);
1384 rcu_assign_pointer(*porig, rt);
1385 if (orig) {
1386 dst_dev_put(&orig->dst);
1387 dst_release(&orig->dst);
1388 }
1389 ret = true;
1390 }
1391
1392 fnhe->fnhe_stamp = jiffies;
1393 }
1394 spin_unlock_bh(&fnhe_lock);
1395
1396 return ret;
1397 }
1398
1399 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1400 {
1401 struct rtable *orig, *prev, **p;
1402 bool ret = true;
1403
1404 if (rt_is_input_route(rt)) {
1405 p = (struct rtable **)&nh->nh_rth_input;
1406 } else {
1407 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1408 }
1409 orig = *p;
1410
1411 /* hold dst before doing cmpxchg() to avoid race condition
1412 * on this dst
1413 */
1414 dst_hold(&rt->dst);
1415 prev = cmpxchg(p, orig, rt);
1416 if (prev == orig) {
1417 if (orig) {
1418 dst_dev_put(&orig->dst);
1419 dst_release(&orig->dst);
1420 }
1421 } else {
1422 dst_release(&rt->dst);
1423 ret = false;
1424 }
1425
1426 return ret;
1427 }
1428
1429 struct uncached_list {
1430 spinlock_t lock;
1431 struct list_head head;
1432 };
1433
1434 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1435
1436 void rt_add_uncached_list(struct rtable *rt)
1437 {
1438 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1439
1440 rt->rt_uncached_list = ul;
1441
1442 spin_lock_bh(&ul->lock);
1443 list_add_tail(&rt->rt_uncached, &ul->head);
1444 spin_unlock_bh(&ul->lock);
1445 }
1446
1447 void rt_del_uncached_list(struct rtable *rt)
1448 {
1449 if (!list_empty(&rt->rt_uncached)) {
1450 struct uncached_list *ul = rt->rt_uncached_list;
1451
1452 spin_lock_bh(&ul->lock);
1453 list_del(&rt->rt_uncached);
1454 spin_unlock_bh(&ul->lock);
1455 }
1456 }
1457
1458 static void ipv4_dst_destroy(struct dst_entry *dst)
1459 {
1460 struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst);
1461 struct rtable *rt = (struct rtable *)dst;
1462
1463 if (p != &dst_default_metrics && refcount_dec_and_test(&p->refcnt))
1464 kfree(p);
1465
1466 rt_del_uncached_list(rt);
1467 }
1468
1469 void rt_flush_dev(struct net_device *dev)
1470 {
1471 struct net *net = dev_net(dev);
1472 struct rtable *rt;
1473 int cpu;
1474
1475 for_each_possible_cpu(cpu) {
1476 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1477
1478 spin_lock_bh(&ul->lock);
1479 list_for_each_entry(rt, &ul->head, rt_uncached) {
1480 if (rt->dst.dev != dev)
1481 continue;
1482 rt->dst.dev = net->loopback_dev;
1483 dev_hold(rt->dst.dev);
1484 dev_put(dev);
1485 }
1486 spin_unlock_bh(&ul->lock);
1487 }
1488 }
1489
1490 static bool rt_cache_valid(const struct rtable *rt)
1491 {
1492 return rt &&
1493 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1494 !rt_is_expired(rt);
1495 }
1496
1497 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1498 const struct fib_result *res,
1499 struct fib_nh_exception *fnhe,
1500 struct fib_info *fi, u16 type, u32 itag,
1501 const bool do_cache)
1502 {
1503 bool cached = false;
1504
1505 if (fi) {
1506 struct fib_nh *nh = &FIB_RES_NH(*res);
1507
1508 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1509 rt->rt_gateway = nh->nh_gw;
1510 rt->rt_uses_gateway = 1;
1511 }
1512 dst_init_metrics(&rt->dst, fi->fib_metrics->metrics, true);
1513 if (fi->fib_metrics != &dst_default_metrics) {
1514 rt->dst._metrics |= DST_METRICS_REFCOUNTED;
1515 refcount_inc(&fi->fib_metrics->refcnt);
1516 }
1517 #ifdef CONFIG_IP_ROUTE_CLASSID
1518 rt->dst.tclassid = nh->nh_tclassid;
1519 #endif
1520 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1521 if (unlikely(fnhe))
1522 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1523 else if (do_cache)
1524 cached = rt_cache_route(nh, rt);
1525 if (unlikely(!cached)) {
1526 /* Routes we intend to cache in nexthop exception or
1527 * FIB nexthop have the DST_NOCACHE bit clear.
1528 * However, if we are unsuccessful at storing this
1529 * route into the cache we really need to set it.
1530 */
1531 if (!rt->rt_gateway)
1532 rt->rt_gateway = daddr;
1533 rt_add_uncached_list(rt);
1534 }
1535 } else
1536 rt_add_uncached_list(rt);
1537
1538 #ifdef CONFIG_IP_ROUTE_CLASSID
1539 #ifdef CONFIG_IP_MULTIPLE_TABLES
1540 set_class_tag(rt, res->tclassid);
1541 #endif
1542 set_class_tag(rt, itag);
1543 #endif
1544 }
1545
1546 struct rtable *rt_dst_alloc(struct net_device *dev,
1547 unsigned int flags, u16 type,
1548 bool nopolicy, bool noxfrm, bool will_cache)
1549 {
1550 struct rtable *rt;
1551
1552 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1553 (will_cache ? 0 : DST_HOST) |
1554 (nopolicy ? DST_NOPOLICY : 0) |
1555 (noxfrm ? DST_NOXFRM : 0));
1556
1557 if (rt) {
1558 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1559 rt->rt_flags = flags;
1560 rt->rt_type = type;
1561 rt->rt_is_input = 0;
1562 rt->rt_iif = 0;
1563 rt->rt_pmtu = 0;
1564 rt->rt_mtu_locked = 0;
1565 rt->rt_gateway = 0;
1566 rt->rt_uses_gateway = 0;
1567 rt->rt_table_id = 0;
1568 INIT_LIST_HEAD(&rt->rt_uncached);
1569
1570 rt->dst.output = ip_output;
1571 if (flags & RTCF_LOCAL)
1572 rt->dst.input = ip_local_deliver;
1573 }
1574
1575 return rt;
1576 }
1577 EXPORT_SYMBOL(rt_dst_alloc);
1578
1579 /* called in rcu_read_lock() section */
1580 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1581 u8 tos, struct net_device *dev,
1582 struct in_device *in_dev, u32 *itag)
1583 {
1584 int err;
1585
1586 /* Primary sanity checks. */
1587 if (!in_dev)
1588 return -EINVAL;
1589
1590 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1591 skb->protocol != htons(ETH_P_IP))
1592 return -EINVAL;
1593
1594 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1595 return -EINVAL;
1596
1597 if (ipv4_is_zeronet(saddr)) {
1598 if (!ipv4_is_local_multicast(daddr))
1599 return -EINVAL;
1600 } else {
1601 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1602 in_dev, itag);
1603 if (err < 0)
1604 return err;
1605 }
1606 return 0;
1607 }
1608
1609 /* called in rcu_read_lock() section */
1610 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1611 u8 tos, struct net_device *dev, int our)
1612 {
1613 struct in_device *in_dev = __in_dev_get_rcu(dev);
1614 unsigned int flags = RTCF_MULTICAST;
1615 struct rtable *rth;
1616 u32 itag = 0;
1617 int err;
1618
1619 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1620 if (err)
1621 return err;
1622
1623 if (our)
1624 flags |= RTCF_LOCAL;
1625
1626 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1627 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1628 if (!rth)
1629 return -ENOBUFS;
1630
1631 #ifdef CONFIG_IP_ROUTE_CLASSID
1632 rth->dst.tclassid = itag;
1633 #endif
1634 rth->dst.output = ip_rt_bug;
1635 rth->rt_is_input= 1;
1636
1637 #ifdef CONFIG_IP_MROUTE
1638 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1639 rth->dst.input = ip_mr_input;
1640 #endif
1641 RT_CACHE_STAT_INC(in_slow_mc);
1642
1643 skb_dst_set(skb, &rth->dst);
1644 return 0;
1645 }
1646
1647
1648 static void ip_handle_martian_source(struct net_device *dev,
1649 struct in_device *in_dev,
1650 struct sk_buff *skb,
1651 __be32 daddr,
1652 __be32 saddr)
1653 {
1654 RT_CACHE_STAT_INC(in_martian_src);
1655 #ifdef CONFIG_IP_ROUTE_VERBOSE
1656 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1657 /*
1658 * RFC1812 recommendation, if source is martian,
1659 * the only hint is MAC header.
1660 */
1661 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1662 &daddr, &saddr, dev->name);
1663 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1664 print_hex_dump(KERN_WARNING, "ll header: ",
1665 DUMP_PREFIX_OFFSET, 16, 1,
1666 skb_mac_header(skb),
1667 dev->hard_header_len, true);
1668 }
1669 }
1670 #endif
1671 }
1672
1673 static void set_lwt_redirect(struct rtable *rth)
1674 {
1675 if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1676 rth->dst.lwtstate->orig_output = rth->dst.output;
1677 rth->dst.output = lwtunnel_output;
1678 }
1679
1680 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1681 rth->dst.lwtstate->orig_input = rth->dst.input;
1682 rth->dst.input = lwtunnel_input;
1683 }
1684 }
1685
1686 /* called in rcu_read_lock() section */
1687 static int __mkroute_input(struct sk_buff *skb,
1688 const struct fib_result *res,
1689 struct in_device *in_dev,
1690 __be32 daddr, __be32 saddr, u32 tos)
1691 {
1692 struct fib_nh_exception *fnhe;
1693 struct rtable *rth;
1694 int err;
1695 struct in_device *out_dev;
1696 bool do_cache;
1697 u32 itag = 0;
1698
1699 /* get a working reference to the output device */
1700 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1701 if (!out_dev) {
1702 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1703 return -EINVAL;
1704 }
1705
1706 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1707 in_dev->dev, in_dev, &itag);
1708 if (err < 0) {
1709 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1710 saddr);
1711
1712 goto cleanup;
1713 }
1714
1715 do_cache = res->fi && !itag;
1716 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1717 skb->protocol == htons(ETH_P_IP) &&
1718 (IN_DEV_SHARED_MEDIA(out_dev) ||
1719 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1720 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1721
1722 if (skb->protocol != htons(ETH_P_IP)) {
1723 /* Not IP (i.e. ARP). Do not create route, if it is
1724 * invalid for proxy arp. DNAT routes are always valid.
1725 *
1726 * Proxy arp feature have been extended to allow, ARP
1727 * replies back to the same interface, to support
1728 * Private VLAN switch technologies. See arp.c.
1729 */
1730 if (out_dev == in_dev &&
1731 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1732 err = -EINVAL;
1733 goto cleanup;
1734 }
1735 }
1736
1737 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1738 if (do_cache) {
1739 if (fnhe)
1740 rth = rcu_dereference(fnhe->fnhe_rth_input);
1741 else
1742 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1743 if (rt_cache_valid(rth)) {
1744 skb_dst_set_noref(skb, &rth->dst);
1745 goto out;
1746 }
1747 }
1748
1749 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1750 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1751 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1752 if (!rth) {
1753 err = -ENOBUFS;
1754 goto cleanup;
1755 }
1756
1757 rth->rt_is_input = 1;
1758 if (res->table)
1759 rth->rt_table_id = res->table->tb_id;
1760 RT_CACHE_STAT_INC(in_slow_tot);
1761
1762 rth->dst.input = ip_forward;
1763
1764 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1765 do_cache);
1766 set_lwt_redirect(rth);
1767 skb_dst_set(skb, &rth->dst);
1768 out:
1769 err = 0;
1770 cleanup:
1771 return err;
1772 }
1773
1774 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1775 /* To make ICMP packets follow the right flow, the multipath hash is
1776 * calculated from the inner IP addresses.
1777 */
1778 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1779 struct flow_keys *hash_keys)
1780 {
1781 const struct iphdr *outer_iph = ip_hdr(skb);
1782 const struct iphdr *inner_iph;
1783 const struct icmphdr *icmph;
1784 struct iphdr _inner_iph;
1785 struct icmphdr _icmph;
1786
1787 hash_keys->addrs.v4addrs.src = outer_iph->saddr;
1788 hash_keys->addrs.v4addrs.dst = outer_iph->daddr;
1789 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1790 return;
1791
1792 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1793 return;
1794
1795 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1796 &_icmph);
1797 if (!icmph)
1798 return;
1799
1800 if (icmph->type != ICMP_DEST_UNREACH &&
1801 icmph->type != ICMP_REDIRECT &&
1802 icmph->type != ICMP_TIME_EXCEEDED &&
1803 icmph->type != ICMP_PARAMETERPROB)
1804 return;
1805
1806 inner_iph = skb_header_pointer(skb,
1807 outer_iph->ihl * 4 + sizeof(_icmph),
1808 sizeof(_inner_iph), &_inner_iph);
1809 if (!inner_iph)
1810 return;
1811 hash_keys->addrs.v4addrs.src = inner_iph->saddr;
1812 hash_keys->addrs.v4addrs.dst = inner_iph->daddr;
1813 }
1814
1815 /* if skb is set it will be used and fl4 can be NULL */
1816 int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4,
1817 const struct sk_buff *skb)
1818 {
1819 struct net *net = fi->fib_net;
1820 struct flow_keys hash_keys;
1821 u32 mhash;
1822
1823 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1824 case 0:
1825 memset(&hash_keys, 0, sizeof(hash_keys));
1826 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1827 if (skb) {
1828 ip_multipath_l3_keys(skb, &hash_keys);
1829 } else {
1830 hash_keys.addrs.v4addrs.src = fl4->saddr;
1831 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1832 }
1833 break;
1834 case 1:
1835 /* skb is currently provided only when forwarding */
1836 if (skb) {
1837 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1838 struct flow_keys keys;
1839
1840 /* short-circuit if we already have L4 hash present */
1841 if (skb->l4_hash)
1842 return skb_get_hash_raw(skb) >> 1;
1843 memset(&hash_keys, 0, sizeof(hash_keys));
1844 skb_flow_dissect_flow_keys(skb, &keys, flag);
1845
1846 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1847 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1848 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1849 hash_keys.ports.src = keys.ports.src;
1850 hash_keys.ports.dst = keys.ports.dst;
1851 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1852 } else {
1853 memset(&hash_keys, 0, sizeof(hash_keys));
1854 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1855 hash_keys.addrs.v4addrs.src = fl4->saddr;
1856 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1857 hash_keys.ports.src = fl4->fl4_sport;
1858 hash_keys.ports.dst = fl4->fl4_dport;
1859 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1860 }
1861 break;
1862 }
1863 mhash = flow_hash_from_keys(&hash_keys);
1864
1865 return mhash >> 1;
1866 }
1867 EXPORT_SYMBOL_GPL(fib_multipath_hash);
1868 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1869
1870 static int ip_mkroute_input(struct sk_buff *skb,
1871 struct fib_result *res,
1872 struct in_device *in_dev,
1873 __be32 daddr, __be32 saddr, u32 tos)
1874 {
1875 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1876 if (res->fi && res->fi->fib_nhs > 1) {
1877 int h = fib_multipath_hash(res->fi, NULL, skb);
1878
1879 fib_select_multipath(res, h);
1880 }
1881 #endif
1882
1883 /* create a routing cache entry */
1884 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1885 }
1886
1887 /*
1888 * NOTE. We drop all the packets that has local source
1889 * addresses, because every properly looped back packet
1890 * must have correct destination already attached by output routine.
1891 *
1892 * Such approach solves two big problems:
1893 * 1. Not simplex devices are handled properly.
1894 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1895 * called with rcu_read_lock()
1896 */
1897
1898 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1899 u8 tos, struct net_device *dev,
1900 struct fib_result *res)
1901 {
1902 struct in_device *in_dev = __in_dev_get_rcu(dev);
1903 struct ip_tunnel_info *tun_info;
1904 struct flowi4 fl4;
1905 unsigned int flags = 0;
1906 u32 itag = 0;
1907 struct rtable *rth;
1908 int err = -EINVAL;
1909 struct net *net = dev_net(dev);
1910 bool do_cache;
1911
1912 /* IP on this device is disabled. */
1913
1914 if (!in_dev)
1915 goto out;
1916
1917 /* Check for the most weird martians, which can be not detected
1918 by fib_lookup.
1919 */
1920
1921 tun_info = skb_tunnel_info(skb);
1922 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1923 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1924 else
1925 fl4.flowi4_tun_key.tun_id = 0;
1926 skb_dst_drop(skb);
1927
1928 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1929 goto martian_source;
1930
1931 res->fi = NULL;
1932 res->table = NULL;
1933 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1934 goto brd_input;
1935
1936 /* Accept zero addresses only to limited broadcast;
1937 * I even do not know to fix it or not. Waiting for complains :-)
1938 */
1939 if (ipv4_is_zeronet(saddr))
1940 goto martian_source;
1941
1942 if (ipv4_is_zeronet(daddr))
1943 goto martian_destination;
1944
1945 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1946 * and call it once if daddr or/and saddr are loopback addresses
1947 */
1948 if (ipv4_is_loopback(daddr)) {
1949 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1950 goto martian_destination;
1951 } else if (ipv4_is_loopback(saddr)) {
1952 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1953 goto martian_source;
1954 }
1955
1956 /*
1957 * Now we are ready to route packet.
1958 */
1959 fl4.flowi4_oif = 0;
1960 fl4.flowi4_iif = dev->ifindex;
1961 fl4.flowi4_mark = skb->mark;
1962 fl4.flowi4_tos = tos;
1963 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1964 fl4.flowi4_flags = 0;
1965 fl4.daddr = daddr;
1966 fl4.saddr = saddr;
1967 fl4.flowi4_uid = sock_net_uid(net, NULL);
1968 err = fib_lookup(net, &fl4, res, 0);
1969 if (err != 0) {
1970 if (!IN_DEV_FORWARD(in_dev))
1971 err = -EHOSTUNREACH;
1972 goto no_route;
1973 }
1974
1975 if (res->type == RTN_BROADCAST)
1976 goto brd_input;
1977
1978 if (res->type == RTN_LOCAL) {
1979 err = fib_validate_source(skb, saddr, daddr, tos,
1980 0, dev, in_dev, &itag);
1981 if (err < 0)
1982 goto martian_source;
1983 goto local_input;
1984 }
1985
1986 if (!IN_DEV_FORWARD(in_dev)) {
1987 err = -EHOSTUNREACH;
1988 goto no_route;
1989 }
1990 if (res->type != RTN_UNICAST)
1991 goto martian_destination;
1992
1993 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1994 out: return err;
1995
1996 brd_input:
1997 if (skb->protocol != htons(ETH_P_IP))
1998 goto e_inval;
1999
2000 if (!ipv4_is_zeronet(saddr)) {
2001 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2002 in_dev, &itag);
2003 if (err < 0)
2004 goto martian_source;
2005 }
2006 flags |= RTCF_BROADCAST;
2007 res->type = RTN_BROADCAST;
2008 RT_CACHE_STAT_INC(in_brd);
2009
2010 local_input:
2011 do_cache = false;
2012 if (res->fi) {
2013 if (!itag) {
2014 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
2015 if (rt_cache_valid(rth)) {
2016 skb_dst_set_noref(skb, &rth->dst);
2017 err = 0;
2018 goto out;
2019 }
2020 do_cache = true;
2021 }
2022 }
2023
2024 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2025 flags | RTCF_LOCAL, res->type,
2026 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2027 if (!rth)
2028 goto e_nobufs;
2029
2030 rth->dst.output= ip_rt_bug;
2031 #ifdef CONFIG_IP_ROUTE_CLASSID
2032 rth->dst.tclassid = itag;
2033 #endif
2034 rth->rt_is_input = 1;
2035 if (res->table)
2036 rth->rt_table_id = res->table->tb_id;
2037
2038 RT_CACHE_STAT_INC(in_slow_tot);
2039 if (res->type == RTN_UNREACHABLE) {
2040 rth->dst.input= ip_error;
2041 rth->dst.error= -err;
2042 rth->rt_flags &= ~RTCF_LOCAL;
2043 }
2044
2045 if (do_cache) {
2046 struct fib_nh *nh = &FIB_RES_NH(*res);
2047
2048 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
2049 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2050 WARN_ON(rth->dst.input == lwtunnel_input);
2051 rth->dst.lwtstate->orig_input = rth->dst.input;
2052 rth->dst.input = lwtunnel_input;
2053 }
2054
2055 if (unlikely(!rt_cache_route(nh, rth)))
2056 rt_add_uncached_list(rth);
2057 }
2058 skb_dst_set(skb, &rth->dst);
2059 err = 0;
2060 goto out;
2061
2062 no_route:
2063 RT_CACHE_STAT_INC(in_no_route);
2064 res->type = RTN_UNREACHABLE;
2065 res->fi = NULL;
2066 res->table = NULL;
2067 goto local_input;
2068
2069 /*
2070 * Do not cache martian addresses: they should be logged (RFC1812)
2071 */
2072 martian_destination:
2073 RT_CACHE_STAT_INC(in_martian_dst);
2074 #ifdef CONFIG_IP_ROUTE_VERBOSE
2075 if (IN_DEV_LOG_MARTIANS(in_dev))
2076 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2077 &daddr, &saddr, dev->name);
2078 #endif
2079
2080 e_inval:
2081 err = -EINVAL;
2082 goto out;
2083
2084 e_nobufs:
2085 err = -ENOBUFS;
2086 goto out;
2087
2088 martian_source:
2089 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2090 goto out;
2091 }
2092
2093 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2094 u8 tos, struct net_device *dev)
2095 {
2096 struct fib_result res;
2097 int err;
2098
2099 tos &= IPTOS_RT_MASK;
2100 rcu_read_lock();
2101 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2102 rcu_read_unlock();
2103
2104 return err;
2105 }
2106 EXPORT_SYMBOL(ip_route_input_noref);
2107
2108 /* called with rcu_read_lock held */
2109 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2110 u8 tos, struct net_device *dev, struct fib_result *res)
2111 {
2112 /* Multicast recognition logic is moved from route cache to here.
2113 The problem was that too many Ethernet cards have broken/missing
2114 hardware multicast filters :-( As result the host on multicasting
2115 network acquires a lot of useless route cache entries, sort of
2116 SDR messages from all the world. Now we try to get rid of them.
2117 Really, provided software IP multicast filter is organized
2118 reasonably (at least, hashed), it does not result in a slowdown
2119 comparing with route cache reject entries.
2120 Note, that multicast routers are not affected, because
2121 route cache entry is created eventually.
2122 */
2123 if (ipv4_is_multicast(daddr)) {
2124 struct in_device *in_dev = __in_dev_get_rcu(dev);
2125 int our = 0;
2126 int err = -EINVAL;
2127
2128 if (in_dev)
2129 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2130 ip_hdr(skb)->protocol);
2131
2132 /* check l3 master if no match yet */
2133 if ((!in_dev || !our) && netif_is_l3_slave(dev)) {
2134 struct in_device *l3_in_dev;
2135
2136 l3_in_dev = __in_dev_get_rcu(skb->dev);
2137 if (l3_in_dev)
2138 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2139 ip_hdr(skb)->protocol);
2140 }
2141
2142 if (our
2143 #ifdef CONFIG_IP_MROUTE
2144 ||
2145 (!ipv4_is_local_multicast(daddr) &&
2146 IN_DEV_MFORWARD(in_dev))
2147 #endif
2148 ) {
2149 err = ip_route_input_mc(skb, daddr, saddr,
2150 tos, dev, our);
2151 }
2152 return err;
2153 }
2154
2155 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2156 }
2157
2158 /* called with rcu_read_lock() */
2159 static struct rtable *__mkroute_output(const struct fib_result *res,
2160 const struct flowi4 *fl4, int orig_oif,
2161 struct net_device *dev_out,
2162 unsigned int flags)
2163 {
2164 struct fib_info *fi = res->fi;
2165 struct fib_nh_exception *fnhe;
2166 struct in_device *in_dev;
2167 u16 type = res->type;
2168 struct rtable *rth;
2169 bool do_cache;
2170
2171 in_dev = __in_dev_get_rcu(dev_out);
2172 if (!in_dev)
2173 return ERR_PTR(-EINVAL);
2174
2175 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2176 if (ipv4_is_loopback(fl4->saddr) &&
2177 !(dev_out->flags & IFF_LOOPBACK) &&
2178 !netif_is_l3_master(dev_out))
2179 return ERR_PTR(-EINVAL);
2180
2181 if (ipv4_is_lbcast(fl4->daddr))
2182 type = RTN_BROADCAST;
2183 else if (ipv4_is_multicast(fl4->daddr))
2184 type = RTN_MULTICAST;
2185 else if (ipv4_is_zeronet(fl4->daddr))
2186 return ERR_PTR(-EINVAL);
2187
2188 if (dev_out->flags & IFF_LOOPBACK)
2189 flags |= RTCF_LOCAL;
2190
2191 do_cache = true;
2192 if (type == RTN_BROADCAST) {
2193 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2194 fi = NULL;
2195 } else if (type == RTN_MULTICAST) {
2196 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2197 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2198 fl4->flowi4_proto))
2199 flags &= ~RTCF_LOCAL;
2200 else
2201 do_cache = false;
2202 /* If multicast route do not exist use
2203 * default one, but do not gateway in this case.
2204 * Yes, it is hack.
2205 */
2206 if (fi && res->prefixlen < 4)
2207 fi = NULL;
2208 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2209 (orig_oif != dev_out->ifindex)) {
2210 /* For local routes that require a particular output interface
2211 * we do not want to cache the result. Caching the result
2212 * causes incorrect behaviour when there are multiple source
2213 * addresses on the interface, the end result being that if the
2214 * intended recipient is waiting on that interface for the
2215 * packet he won't receive it because it will be delivered on
2216 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2217 * be set to the loopback interface as well.
2218 */
2219 do_cache = false;
2220 }
2221
2222 fnhe = NULL;
2223 do_cache &= fi != NULL;
2224 if (fi) {
2225 struct rtable __rcu **prth;
2226 struct fib_nh *nh = &FIB_RES_NH(*res);
2227
2228 fnhe = find_exception(nh, fl4->daddr);
2229 if (!do_cache)
2230 goto add;
2231 if (fnhe) {
2232 prth = &fnhe->fnhe_rth_output;
2233 } else {
2234 if (unlikely(fl4->flowi4_flags &
2235 FLOWI_FLAG_KNOWN_NH &&
2236 !(nh->nh_gw &&
2237 nh->nh_scope == RT_SCOPE_LINK))) {
2238 do_cache = false;
2239 goto add;
2240 }
2241 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2242 }
2243 rth = rcu_dereference(*prth);
2244 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2245 return rth;
2246 }
2247
2248 add:
2249 rth = rt_dst_alloc(dev_out, flags, type,
2250 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2251 IN_DEV_CONF_GET(in_dev, NOXFRM),
2252 do_cache);
2253 if (!rth)
2254 return ERR_PTR(-ENOBUFS);
2255
2256 rth->rt_iif = orig_oif;
2257 if (res->table)
2258 rth->rt_table_id = res->table->tb_id;
2259
2260 RT_CACHE_STAT_INC(out_slow_tot);
2261
2262 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2263 if (flags & RTCF_LOCAL &&
2264 !(dev_out->flags & IFF_LOOPBACK)) {
2265 rth->dst.output = ip_mc_output;
2266 RT_CACHE_STAT_INC(out_slow_mc);
2267 }
2268 #ifdef CONFIG_IP_MROUTE
2269 if (type == RTN_MULTICAST) {
2270 if (IN_DEV_MFORWARD(in_dev) &&
2271 !ipv4_is_local_multicast(fl4->daddr)) {
2272 rth->dst.input = ip_mr_input;
2273 rth->dst.output = ip_mc_output;
2274 }
2275 }
2276 #endif
2277 }
2278
2279 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2280 set_lwt_redirect(rth);
2281
2282 return rth;
2283 }
2284
2285 /*
2286 * Major route resolver routine.
2287 */
2288
2289 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2290 const struct sk_buff *skb)
2291 {
2292 __u8 tos = RT_FL_TOS(fl4);
2293 struct fib_result res = {
2294 .type = RTN_UNSPEC,
2295 .fi = NULL,
2296 .table = NULL,
2297 .tclassid = 0,
2298 };
2299 struct rtable *rth;
2300
2301 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2302 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2303 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2304 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2305
2306 rcu_read_lock();
2307 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2308 rcu_read_unlock();
2309
2310 return rth;
2311 }
2312 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2313
2314 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2315 struct fib_result *res,
2316 const struct sk_buff *skb)
2317 {
2318 struct net_device *dev_out = NULL;
2319 int orig_oif = fl4->flowi4_oif;
2320 unsigned int flags = 0;
2321 struct rtable *rth;
2322 int err = -ENETUNREACH;
2323
2324 if (fl4->saddr) {
2325 rth = ERR_PTR(-EINVAL);
2326 if (ipv4_is_multicast(fl4->saddr) ||
2327 ipv4_is_lbcast(fl4->saddr) ||
2328 ipv4_is_zeronet(fl4->saddr))
2329 goto out;
2330
2331 /* I removed check for oif == dev_out->oif here.
2332 It was wrong for two reasons:
2333 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2334 is assigned to multiple interfaces.
2335 2. Moreover, we are allowed to send packets with saddr
2336 of another iface. --ANK
2337 */
2338
2339 if (fl4->flowi4_oif == 0 &&
2340 (ipv4_is_multicast(fl4->daddr) ||
2341 ipv4_is_lbcast(fl4->daddr))) {
2342 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2343 dev_out = __ip_dev_find(net, fl4->saddr, false);
2344 if (!dev_out)
2345 goto out;
2346
2347 /* Special hack: user can direct multicasts
2348 and limited broadcast via necessary interface
2349 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2350 This hack is not just for fun, it allows
2351 vic,vat and friends to work.
2352 They bind socket to loopback, set ttl to zero
2353 and expect that it will work.
2354 From the viewpoint of routing cache they are broken,
2355 because we are not allowed to build multicast path
2356 with loopback source addr (look, routing cache
2357 cannot know, that ttl is zero, so that packet
2358 will not leave this host and route is valid).
2359 Luckily, this hack is good workaround.
2360 */
2361
2362 fl4->flowi4_oif = dev_out->ifindex;
2363 goto make_route;
2364 }
2365
2366 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2367 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2368 if (!__ip_dev_find(net, fl4->saddr, false))
2369 goto out;
2370 }
2371 }
2372
2373
2374 if (fl4->flowi4_oif) {
2375 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2376 rth = ERR_PTR(-ENODEV);
2377 if (!dev_out)
2378 goto out;
2379
2380 /* RACE: Check return value of inet_select_addr instead. */
2381 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2382 rth = ERR_PTR(-ENETUNREACH);
2383 goto out;
2384 }
2385 if (ipv4_is_local_multicast(fl4->daddr) ||
2386 ipv4_is_lbcast(fl4->daddr) ||
2387 fl4->flowi4_proto == IPPROTO_IGMP) {
2388 if (!fl4->saddr)
2389 fl4->saddr = inet_select_addr(dev_out, 0,
2390 RT_SCOPE_LINK);
2391 goto make_route;
2392 }
2393 if (!fl4->saddr) {
2394 if (ipv4_is_multicast(fl4->daddr))
2395 fl4->saddr = inet_select_addr(dev_out, 0,
2396 fl4->flowi4_scope);
2397 else if (!fl4->daddr)
2398 fl4->saddr = inet_select_addr(dev_out, 0,
2399 RT_SCOPE_HOST);
2400 }
2401 }
2402
2403 if (!fl4->daddr) {
2404 fl4->daddr = fl4->saddr;
2405 if (!fl4->daddr)
2406 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2407 dev_out = net->loopback_dev;
2408 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2409 res->type = RTN_LOCAL;
2410 flags |= RTCF_LOCAL;
2411 goto make_route;
2412 }
2413
2414 err = fib_lookup(net, fl4, res, 0);
2415 if (err) {
2416 res->fi = NULL;
2417 res->table = NULL;
2418 if (fl4->flowi4_oif &&
2419 (ipv4_is_multicast(fl4->daddr) ||
2420 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2421 /* Apparently, routing tables are wrong. Assume,
2422 that the destination is on link.
2423
2424 WHY? DW.
2425 Because we are allowed to send to iface
2426 even if it has NO routes and NO assigned
2427 addresses. When oif is specified, routing
2428 tables are looked up with only one purpose:
2429 to catch if destination is gatewayed, rather than
2430 direct. Moreover, if MSG_DONTROUTE is set,
2431 we send packet, ignoring both routing tables
2432 and ifaddr state. --ANK
2433
2434
2435 We could make it even if oif is unknown,
2436 likely IPv6, but we do not.
2437 */
2438
2439 if (fl4->saddr == 0)
2440 fl4->saddr = inet_select_addr(dev_out, 0,
2441 RT_SCOPE_LINK);
2442 res->type = RTN_UNICAST;
2443 goto make_route;
2444 }
2445 rth = ERR_PTR(err);
2446 goto out;
2447 }
2448
2449 if (res->type == RTN_LOCAL) {
2450 if (!fl4->saddr) {
2451 if (res->fi->fib_prefsrc)
2452 fl4->saddr = res->fi->fib_prefsrc;
2453 else
2454 fl4->saddr = fl4->daddr;
2455 }
2456
2457 /* L3 master device is the loopback for that domain */
2458 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2459 net->loopback_dev;
2460
2461 /* make sure orig_oif points to fib result device even
2462 * though packet rx/tx happens over loopback or l3mdev
2463 */
2464 orig_oif = FIB_RES_OIF(*res);
2465
2466 fl4->flowi4_oif = dev_out->ifindex;
2467 flags |= RTCF_LOCAL;
2468 goto make_route;
2469 }
2470
2471 fib_select_path(net, res, fl4, skb);
2472
2473 dev_out = FIB_RES_DEV(*res);
2474 fl4->flowi4_oif = dev_out->ifindex;
2475
2476
2477 make_route:
2478 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2479
2480 out:
2481 return rth;
2482 }
2483
2484 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2485 {
2486 return NULL;
2487 }
2488
2489 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2490 {
2491 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2492
2493 return mtu ? : dst->dev->mtu;
2494 }
2495
2496 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2497 struct sk_buff *skb, u32 mtu)
2498 {
2499 }
2500
2501 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2502 struct sk_buff *skb)
2503 {
2504 }
2505
2506 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2507 unsigned long old)
2508 {
2509 return NULL;
2510 }
2511
2512 static struct dst_ops ipv4_dst_blackhole_ops = {
2513 .family = AF_INET,
2514 .check = ipv4_blackhole_dst_check,
2515 .mtu = ipv4_blackhole_mtu,
2516 .default_advmss = ipv4_default_advmss,
2517 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2518 .redirect = ipv4_rt_blackhole_redirect,
2519 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2520 .neigh_lookup = ipv4_neigh_lookup,
2521 };
2522
2523 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2524 {
2525 struct rtable *ort = (struct rtable *) dst_orig;
2526 struct rtable *rt;
2527
2528 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2529 if (rt) {
2530 struct dst_entry *new = &rt->dst;
2531
2532 new->__use = 1;
2533 new->input = dst_discard;
2534 new->output = dst_discard_out;
2535
2536 new->dev = net->loopback_dev;
2537 if (new->dev)
2538 dev_hold(new->dev);
2539
2540 rt->rt_is_input = ort->rt_is_input;
2541 rt->rt_iif = ort->rt_iif;
2542 rt->rt_pmtu = ort->rt_pmtu;
2543 rt->rt_mtu_locked = ort->rt_mtu_locked;
2544
2545 rt->rt_genid = rt_genid_ipv4(net);
2546 rt->rt_flags = ort->rt_flags;
2547 rt->rt_type = ort->rt_type;
2548 rt->rt_gateway = ort->rt_gateway;
2549 rt->rt_uses_gateway = ort->rt_uses_gateway;
2550
2551 INIT_LIST_HEAD(&rt->rt_uncached);
2552 }
2553
2554 dst_release(dst_orig);
2555
2556 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2557 }
2558
2559 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2560 const struct sock *sk)
2561 {
2562 struct rtable *rt = __ip_route_output_key(net, flp4);
2563
2564 if (IS_ERR(rt))
2565 return rt;
2566
2567 if (flp4->flowi4_proto)
2568 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2569 flowi4_to_flowi(flp4),
2570 sk, 0);
2571
2572 return rt;
2573 }
2574 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2575
2576 /* called with rcu_read_lock held */
2577 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2578 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2579 u32 seq)
2580 {
2581 struct rtable *rt = skb_rtable(skb);
2582 struct rtmsg *r;
2583 struct nlmsghdr *nlh;
2584 unsigned long expires = 0;
2585 u32 error;
2586 u32 metrics[RTAX_MAX];
2587
2588 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0);
2589 if (!nlh)
2590 return -EMSGSIZE;
2591
2592 r = nlmsg_data(nlh);
2593 r->rtm_family = AF_INET;
2594 r->rtm_dst_len = 32;
2595 r->rtm_src_len = 0;
2596 r->rtm_tos = fl4->flowi4_tos;
2597 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2598 if (nla_put_u32(skb, RTA_TABLE, table_id))
2599 goto nla_put_failure;
2600 r->rtm_type = rt->rt_type;
2601 r->rtm_scope = RT_SCOPE_UNIVERSE;
2602 r->rtm_protocol = RTPROT_UNSPEC;
2603 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2604 if (rt->rt_flags & RTCF_NOTIFY)
2605 r->rtm_flags |= RTM_F_NOTIFY;
2606 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2607 r->rtm_flags |= RTCF_DOREDIRECT;
2608
2609 if (nla_put_in_addr(skb, RTA_DST, dst))
2610 goto nla_put_failure;
2611 if (src) {
2612 r->rtm_src_len = 32;
2613 if (nla_put_in_addr(skb, RTA_SRC, src))
2614 goto nla_put_failure;
2615 }
2616 if (rt->dst.dev &&
2617 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2618 goto nla_put_failure;
2619 #ifdef CONFIG_IP_ROUTE_CLASSID
2620 if (rt->dst.tclassid &&
2621 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2622 goto nla_put_failure;
2623 #endif
2624 if (!rt_is_input_route(rt) &&
2625 fl4->saddr != src) {
2626 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2627 goto nla_put_failure;
2628 }
2629 if (rt->rt_uses_gateway &&
2630 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2631 goto nla_put_failure;
2632
2633 expires = rt->dst.expires;
2634 if (expires) {
2635 unsigned long now = jiffies;
2636
2637 if (time_before(now, expires))
2638 expires -= now;
2639 else
2640 expires = 0;
2641 }
2642
2643 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2644 if (rt->rt_pmtu && expires)
2645 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2646 if (rt->rt_mtu_locked && expires)
2647 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2648 if (rtnetlink_put_metrics(skb, metrics) < 0)
2649 goto nla_put_failure;
2650
2651 if (fl4->flowi4_mark &&
2652 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2653 goto nla_put_failure;
2654
2655 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2656 nla_put_u32(skb, RTA_UID,
2657 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2658 goto nla_put_failure;
2659
2660 error = rt->dst.error;
2661
2662 if (rt_is_input_route(rt)) {
2663 #ifdef CONFIG_IP_MROUTE
2664 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2665 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2666 int err = ipmr_get_route(net, skb,
2667 fl4->saddr, fl4->daddr,
2668 r, portid);
2669
2670 if (err <= 0) {
2671 if (err == 0)
2672 return 0;
2673 goto nla_put_failure;
2674 }
2675 } else
2676 #endif
2677 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2678 goto nla_put_failure;
2679 }
2680
2681 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2682 goto nla_put_failure;
2683
2684 nlmsg_end(skb, nlh);
2685 return 0;
2686
2687 nla_put_failure:
2688 nlmsg_cancel(skb, nlh);
2689 return -EMSGSIZE;
2690 }
2691
2692 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2693 struct netlink_ext_ack *extack)
2694 {
2695 struct net *net = sock_net(in_skb->sk);
2696 struct rtmsg *rtm;
2697 struct nlattr *tb[RTA_MAX+1];
2698 struct fib_result res = {};
2699 struct rtable *rt = NULL;
2700 struct flowi4 fl4;
2701 __be32 dst = 0;
2702 __be32 src = 0;
2703 u32 iif;
2704 int err;
2705 int mark;
2706 struct sk_buff *skb;
2707 u32 table_id = RT_TABLE_MAIN;
2708 kuid_t uid;
2709
2710 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy,
2711 extack);
2712 if (err < 0)
2713 goto errout;
2714
2715 rtm = nlmsg_data(nlh);
2716
2717 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2718 if (!skb) {
2719 err = -ENOBUFS;
2720 goto errout;
2721 }
2722
2723 /* Reserve room for dummy headers, this skb can pass
2724 through good chunk of routing engine.
2725 */
2726 skb_reset_mac_header(skb);
2727 skb_reset_network_header(skb);
2728
2729 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2730 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2731 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2732 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2733 if (tb[RTA_UID])
2734 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2735 else
2736 uid = (iif ? INVALID_UID : current_uid());
2737
2738 /* Bugfix: need to give ip_route_input enough of an IP header to
2739 * not gag.
2740 */
2741 ip_hdr(skb)->protocol = IPPROTO_UDP;
2742 ip_hdr(skb)->saddr = src;
2743 ip_hdr(skb)->daddr = dst;
2744
2745 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2746
2747 memset(&fl4, 0, sizeof(fl4));
2748 fl4.daddr = dst;
2749 fl4.saddr = src;
2750 fl4.flowi4_tos = rtm->rtm_tos;
2751 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2752 fl4.flowi4_mark = mark;
2753 fl4.flowi4_uid = uid;
2754
2755 rcu_read_lock();
2756
2757 if (iif) {
2758 struct net_device *dev;
2759
2760 dev = dev_get_by_index_rcu(net, iif);
2761 if (!dev) {
2762 err = -ENODEV;
2763 goto errout_free;
2764 }
2765
2766 skb->protocol = htons(ETH_P_IP);
2767 skb->dev = dev;
2768 skb->mark = mark;
2769 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
2770 dev, &res);
2771
2772 rt = skb_rtable(skb);
2773 if (err == 0 && rt->dst.error)
2774 err = -rt->dst.error;
2775 } else {
2776 fl4.flowi4_iif = LOOPBACK_IFINDEX;
2777 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
2778 err = 0;
2779 if (IS_ERR(rt))
2780 err = PTR_ERR(rt);
2781 else
2782 skb_dst_set(skb, &rt->dst);
2783 }
2784
2785 if (err)
2786 goto errout_free;
2787
2788 if (rtm->rtm_flags & RTM_F_NOTIFY)
2789 rt->rt_flags |= RTCF_NOTIFY;
2790
2791 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2792 table_id = rt->rt_table_id;
2793
2794 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2795 if (!res.fi) {
2796 err = fib_props[res.type].error;
2797 if (!err)
2798 err = -EHOSTUNREACH;
2799 goto errout_free;
2800 }
2801 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
2802 nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
2803 rt->rt_type, res.prefix, res.prefixlen,
2804 fl4.flowi4_tos, res.fi, 0);
2805 } else {
2806 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2807 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq);
2808 }
2809 if (err < 0)
2810 goto errout_free;
2811
2812 rcu_read_unlock();
2813
2814 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2815 errout:
2816 return err;
2817
2818 errout_free:
2819 rcu_read_unlock();
2820 kfree_skb(skb);
2821 goto errout;
2822 }
2823
2824 void ip_rt_multicast_event(struct in_device *in_dev)
2825 {
2826 rt_cache_flush(dev_net(in_dev->dev));
2827 }
2828
2829 #ifdef CONFIG_SYSCTL
2830 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2831 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2832 static int ip_rt_gc_elasticity __read_mostly = 8;
2833
2834 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2835 void __user *buffer,
2836 size_t *lenp, loff_t *ppos)
2837 {
2838 struct net *net = (struct net *)__ctl->extra1;
2839
2840 if (write) {
2841 rt_cache_flush(net);
2842 fnhe_genid_bump(net);
2843 return 0;
2844 }
2845
2846 return -EINVAL;
2847 }
2848
2849 static struct ctl_table ipv4_route_table[] = {
2850 {
2851 .procname = "gc_thresh",
2852 .data = &ipv4_dst_ops.gc_thresh,
2853 .maxlen = sizeof(int),
2854 .mode = 0644,
2855 .proc_handler = proc_dointvec,
2856 },
2857 {
2858 .procname = "max_size",
2859 .data = &ip_rt_max_size,
2860 .maxlen = sizeof(int),
2861 .mode = 0644,
2862 .proc_handler = proc_dointvec,
2863 },
2864 {
2865 /* Deprecated. Use gc_min_interval_ms */
2866
2867 .procname = "gc_min_interval",
2868 .data = &ip_rt_gc_min_interval,
2869 .maxlen = sizeof(int),
2870 .mode = 0644,
2871 .proc_handler = proc_dointvec_jiffies,
2872 },
2873 {
2874 .procname = "gc_min_interval_ms",
2875 .data = &ip_rt_gc_min_interval,
2876 .maxlen = sizeof(int),
2877 .mode = 0644,
2878 .proc_handler = proc_dointvec_ms_jiffies,
2879 },
2880 {
2881 .procname = "gc_timeout",
2882 .data = &ip_rt_gc_timeout,
2883 .maxlen = sizeof(int),
2884 .mode = 0644,
2885 .proc_handler = proc_dointvec_jiffies,
2886 },
2887 {
2888 .procname = "gc_interval",
2889 .data = &ip_rt_gc_interval,
2890 .maxlen = sizeof(int),
2891 .mode = 0644,
2892 .proc_handler = proc_dointvec_jiffies,
2893 },
2894 {
2895 .procname = "redirect_load",
2896 .data = &ip_rt_redirect_load,
2897 .maxlen = sizeof(int),
2898 .mode = 0644,
2899 .proc_handler = proc_dointvec,
2900 },
2901 {
2902 .procname = "redirect_number",
2903 .data = &ip_rt_redirect_number,
2904 .maxlen = sizeof(int),
2905 .mode = 0644,
2906 .proc_handler = proc_dointvec,
2907 },
2908 {
2909 .procname = "redirect_silence",
2910 .data = &ip_rt_redirect_silence,
2911 .maxlen = sizeof(int),
2912 .mode = 0644,
2913 .proc_handler = proc_dointvec,
2914 },
2915 {
2916 .procname = "error_cost",
2917 .data = &ip_rt_error_cost,
2918 .maxlen = sizeof(int),
2919 .mode = 0644,
2920 .proc_handler = proc_dointvec,
2921 },
2922 {
2923 .procname = "error_burst",
2924 .data = &ip_rt_error_burst,
2925 .maxlen = sizeof(int),
2926 .mode = 0644,
2927 .proc_handler = proc_dointvec,
2928 },
2929 {
2930 .procname = "gc_elasticity",
2931 .data = &ip_rt_gc_elasticity,
2932 .maxlen = sizeof(int),
2933 .mode = 0644,
2934 .proc_handler = proc_dointvec,
2935 },
2936 {
2937 .procname = "mtu_expires",
2938 .data = &ip_rt_mtu_expires,
2939 .maxlen = sizeof(int),
2940 .mode = 0644,
2941 .proc_handler = proc_dointvec_jiffies,
2942 },
2943 {
2944 .procname = "min_pmtu",
2945 .data = &ip_rt_min_pmtu,
2946 .maxlen = sizeof(int),
2947 .mode = 0644,
2948 .proc_handler = proc_dointvec_minmax,
2949 .extra1 = &ip_min_valid_pmtu,
2950 },
2951 {
2952 .procname = "min_adv_mss",
2953 .data = &ip_rt_min_advmss,
2954 .maxlen = sizeof(int),
2955 .mode = 0644,
2956 .proc_handler = proc_dointvec,
2957 },
2958 { }
2959 };
2960
2961 static struct ctl_table ipv4_route_flush_table[] = {
2962 {
2963 .procname = "flush",
2964 .maxlen = sizeof(int),
2965 .mode = 0200,
2966 .proc_handler = ipv4_sysctl_rtcache_flush,
2967 },
2968 { },
2969 };
2970
2971 static __net_init int sysctl_route_net_init(struct net *net)
2972 {
2973 struct ctl_table *tbl;
2974
2975 tbl = ipv4_route_flush_table;
2976 if (!net_eq(net, &init_net)) {
2977 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2978 if (!tbl)
2979 goto err_dup;
2980
2981 /* Don't export sysctls to unprivileged users */
2982 if (net->user_ns != &init_user_ns)
2983 tbl[0].procname = NULL;
2984 }
2985 tbl[0].extra1 = net;
2986
2987 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2988 if (!net->ipv4.route_hdr)
2989 goto err_reg;
2990 return 0;
2991
2992 err_reg:
2993 if (tbl != ipv4_route_flush_table)
2994 kfree(tbl);
2995 err_dup:
2996 return -ENOMEM;
2997 }
2998
2999 static __net_exit void sysctl_route_net_exit(struct net *net)
3000 {
3001 struct ctl_table *tbl;
3002
3003 tbl = net->ipv4.route_hdr->ctl_table_arg;
3004 unregister_net_sysctl_table(net->ipv4.route_hdr);
3005 BUG_ON(tbl == ipv4_route_flush_table);
3006 kfree(tbl);
3007 }
3008
3009 static __net_initdata struct pernet_operations sysctl_route_ops = {
3010 .init = sysctl_route_net_init,
3011 .exit = sysctl_route_net_exit,
3012 };
3013 #endif
3014
3015 static __net_init int rt_genid_init(struct net *net)
3016 {
3017 atomic_set(&net->ipv4.rt_genid, 0);
3018 atomic_set(&net->fnhe_genid, 0);
3019 atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3020 return 0;
3021 }
3022
3023 static __net_initdata struct pernet_operations rt_genid_ops = {
3024 .init = rt_genid_init,
3025 };
3026
3027 static int __net_init ipv4_inetpeer_init(struct net *net)
3028 {
3029 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3030
3031 if (!bp)
3032 return -ENOMEM;
3033 inet_peer_base_init(bp);
3034 net->ipv4.peers = bp;
3035 return 0;
3036 }
3037
3038 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3039 {
3040 struct inet_peer_base *bp = net->ipv4.peers;
3041
3042 net->ipv4.peers = NULL;
3043 inetpeer_invalidate_tree(bp);
3044 kfree(bp);
3045 }
3046
3047 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3048 .init = ipv4_inetpeer_init,
3049 .exit = ipv4_inetpeer_exit,
3050 };
3051
3052 #ifdef CONFIG_IP_ROUTE_CLASSID
3053 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3054 #endif /* CONFIG_IP_ROUTE_CLASSID */
3055
3056 int __init ip_rt_init(void)
3057 {
3058 int cpu;
3059
3060 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
3061 if (!ip_idents)
3062 panic("IP: failed to allocate ip_idents\n");
3063
3064 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3065
3066 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3067 if (!ip_tstamps)
3068 panic("IP: failed to allocate ip_tstamps\n");
3069
3070 for_each_possible_cpu(cpu) {
3071 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3072
3073 INIT_LIST_HEAD(&ul->head);
3074 spin_lock_init(&ul->lock);
3075 }
3076 #ifdef CONFIG_IP_ROUTE_CLASSID
3077 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3078 if (!ip_rt_acct)
3079 panic("IP: failed to allocate ip_rt_acct\n");
3080 #endif
3081
3082 ipv4_dst_ops.kmem_cachep =
3083 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3084 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3085
3086 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3087
3088 if (dst_entries_init(&ipv4_dst_ops) < 0)
3089 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3090
3091 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3092 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3093
3094 ipv4_dst_ops.gc_thresh = ~0;
3095 ip_rt_max_size = INT_MAX;
3096
3097 devinet_init();
3098 ip_fib_init();
3099
3100 if (ip_rt_proc_init())
3101 pr_err("Unable to create route proc files\n");
3102 #ifdef CONFIG_XFRM
3103 xfrm_init();
3104 xfrm4_init();
3105 #endif
3106 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3107 RTNL_FLAG_DOIT_UNLOCKED);
3108
3109 #ifdef CONFIG_SYSCTL
3110 register_pernet_subsys(&sysctl_route_ops);
3111 #endif
3112 register_pernet_subsys(&rt_genid_ops);
3113 register_pernet_subsys(&ipv4_inetpeer_ops);
3114 return 0;
3115 }
3116
3117 #ifdef CONFIG_SYSCTL
3118 /*
3119 * We really need to sanitize the damn ipv4 init order, then all
3120 * this nonsense will go away.
3121 */
3122 void __init ip_static_sysctl_init(void)
3123 {
3124 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3125 }
3126 #endif
ubuntu-bionic-kernel mirror