]> 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
net: ipv4: use a dedicated counter for icmp_v4 redirect packets
[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 peer->n_redirects = 0;
910 }
911
912 /* Too many ignored redirects; do not send anything
913 * set dst.rate_last to the last seen redirected packet.
914 */
915 if (peer->n_redirects >= ip_rt_redirect_number) {
916 peer->rate_last = jiffies;
917 goto out_put_peer;
918 }
919
920 /* Check for load limit; set rate_last to the latest sent
921 * redirect.
922 */
923 if (peer->rate_tokens == 0 ||
924 time_after(jiffies,
925 (peer->rate_last +
926 (ip_rt_redirect_load << peer->rate_tokens)))) {
927 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr);
928
929 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw);
930 peer->rate_last = jiffies;
931 ++peer->rate_tokens;
932 ++peer->n_redirects;
933 #ifdef CONFIG_IP_ROUTE_VERBOSE
934 if (log_martians &&
935 peer->rate_tokens == ip_rt_redirect_number)
936 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n",
937 &ip_hdr(skb)->saddr, inet_iif(skb),
938 &ip_hdr(skb)->daddr, &gw);
939 #endif
940 }
941 out_put_peer:
942 inet_putpeer(peer);
943 }
944
945 static int ip_error(struct sk_buff *skb)
946 {
947 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
948 struct rtable *rt = skb_rtable(skb);
949 struct inet_peer *peer;
950 unsigned long now;
951 struct net *net;
952 bool send;
953 int code;
954
955 /* IP on this device is disabled. */
956 if (!in_dev)
957 goto out;
958
959 net = dev_net(rt->dst.dev);
960 if (!IN_DEV_FORWARD(in_dev)) {
961 switch (rt->dst.error) {
962 case EHOSTUNREACH:
963 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS);
964 break;
965
966 case ENETUNREACH:
967 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
968 break;
969 }
970 goto out;
971 }
972
973 switch (rt->dst.error) {
974 case EINVAL:
975 default:
976 goto out;
977 case EHOSTUNREACH:
978 code = ICMP_HOST_UNREACH;
979 break;
980 case ENETUNREACH:
981 code = ICMP_NET_UNREACH;
982 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES);
983 break;
984 case EACCES:
985 code = ICMP_PKT_FILTERED;
986 break;
987 }
988
989 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr,
990 l3mdev_master_ifindex(skb->dev), 1);
991
992 send = true;
993 if (peer) {
994 now = jiffies;
995 peer->rate_tokens += now - peer->rate_last;
996 if (peer->rate_tokens > ip_rt_error_burst)
997 peer->rate_tokens = ip_rt_error_burst;
998 peer->rate_last = now;
999 if (peer->rate_tokens >= ip_rt_error_cost)
1000 peer->rate_tokens -= ip_rt_error_cost;
1001 else
1002 send = false;
1003 inet_putpeer(peer);
1004 }
1005 if (send)
1006 icmp_send(skb, ICMP_DEST_UNREACH, code, 0);
1007
1008 out: kfree_skb(skb);
1009 return 0;
1010 }
1011
1012 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu)
1013 {
1014 struct dst_entry *dst = &rt->dst;
1015 u32 old_mtu = ipv4_mtu(dst);
1016 struct fib_result res;
1017 bool lock = false;
1018
1019 if (ip_mtu_locked(dst))
1020 return;
1021
1022 if (old_mtu < mtu)
1023 return;
1024
1025 if (mtu < ip_rt_min_pmtu) {
1026 lock = true;
1027 mtu = min(old_mtu, ip_rt_min_pmtu);
1028 }
1029
1030 if (rt->rt_pmtu == mtu && !lock &&
1031 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2))
1032 return;
1033
1034 rcu_read_lock();
1035 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) {
1036 struct fib_nh *nh = &FIB_RES_NH(res);
1037
1038 update_or_create_fnhe(nh, fl4->daddr, 0, mtu, lock,
1039 jiffies + ip_rt_mtu_expires);
1040 }
1041 rcu_read_unlock();
1042 }
1043
1044 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
1045 struct sk_buff *skb, u32 mtu)
1046 {
1047 struct rtable *rt = (struct rtable *) dst;
1048 struct flowi4 fl4;
1049
1050 ip_rt_build_flow_key(&fl4, sk, skb);
1051 __ip_rt_update_pmtu(rt, &fl4, mtu);
1052 }
1053
1054 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu,
1055 int oif, u32 mark, u8 protocol, int flow_flags)
1056 {
1057 const struct iphdr *iph = (const struct iphdr *) skb->data;
1058 struct flowi4 fl4;
1059 struct rtable *rt;
1060
1061 if (!mark)
1062 mark = IP4_REPLY_MARK(net, skb->mark);
1063
1064 __build_flow_key(net, &fl4, NULL, iph, oif,
1065 RT_TOS(iph->tos), protocol, mark, flow_flags);
1066 rt = __ip_route_output_key(net, &fl4);
1067 if (!IS_ERR(rt)) {
1068 __ip_rt_update_pmtu(rt, &fl4, mtu);
1069 ip_rt_put(rt);
1070 }
1071 }
1072 EXPORT_SYMBOL_GPL(ipv4_update_pmtu);
1073
1074 static 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
1080 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0);
1081
1082 if (!fl4.flowi4_mark)
1083 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark);
1084
1085 rt = __ip_route_output_key(sock_net(sk), &fl4);
1086 if (!IS_ERR(rt)) {
1087 __ip_rt_update_pmtu(rt, &fl4, mtu);
1088 ip_rt_put(rt);
1089 }
1090 }
1091
1092 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu)
1093 {
1094 const struct iphdr *iph = (const struct iphdr *) skb->data;
1095 struct flowi4 fl4;
1096 struct rtable *rt;
1097 struct dst_entry *odst = NULL;
1098 bool new = false;
1099 struct net *net = sock_net(sk);
1100
1101 bh_lock_sock(sk);
1102
1103 if (!ip_sk_accept_pmtu(sk))
1104 goto out;
1105
1106 odst = sk_dst_get(sk);
1107
1108 if (sock_owned_by_user(sk) || !odst) {
1109 __ipv4_sk_update_pmtu(skb, sk, mtu);
1110 goto out;
1111 }
1112
1113 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1114
1115 rt = (struct rtable *)odst;
1116 if (odst->obsolete && !odst->ops->check(odst, 0)) {
1117 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1118 if (IS_ERR(rt))
1119 goto out;
1120
1121 new = true;
1122 }
1123
1124 __ip_rt_update_pmtu((struct rtable *) rt->dst.path, &fl4, mtu);
1125
1126 if (!dst_check(&rt->dst, 0)) {
1127 if (new)
1128 dst_release(&rt->dst);
1129
1130 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
1131 if (IS_ERR(rt))
1132 goto out;
1133
1134 new = true;
1135 }
1136
1137 if (new)
1138 sk_dst_set(sk, &rt->dst);
1139
1140 out:
1141 bh_unlock_sock(sk);
1142 dst_release(odst);
1143 }
1144 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu);
1145
1146 void ipv4_redirect(struct sk_buff *skb, struct net *net,
1147 int oif, u32 mark, u8 protocol, int flow_flags)
1148 {
1149 const struct iphdr *iph = (const struct iphdr *) skb->data;
1150 struct flowi4 fl4;
1151 struct rtable *rt;
1152
1153 __build_flow_key(net, &fl4, NULL, iph, oif,
1154 RT_TOS(iph->tos), protocol, mark, flow_flags);
1155 rt = __ip_route_output_key(net, &fl4);
1156 if (!IS_ERR(rt)) {
1157 __ip_do_redirect(rt, skb, &fl4, false);
1158 ip_rt_put(rt);
1159 }
1160 }
1161 EXPORT_SYMBOL_GPL(ipv4_redirect);
1162
1163 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk)
1164 {
1165 const struct iphdr *iph = (const struct iphdr *) skb->data;
1166 struct flowi4 fl4;
1167 struct rtable *rt;
1168 struct net *net = sock_net(sk);
1169
1170 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0);
1171 rt = __ip_route_output_key(net, &fl4);
1172 if (!IS_ERR(rt)) {
1173 __ip_do_redirect(rt, skb, &fl4, false);
1174 ip_rt_put(rt);
1175 }
1176 }
1177 EXPORT_SYMBOL_GPL(ipv4_sk_redirect);
1178
1179 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie)
1180 {
1181 struct rtable *rt = (struct rtable *) dst;
1182
1183 /* All IPV4 dsts are created with ->obsolete set to the value
1184 * DST_OBSOLETE_FORCE_CHK which forces validation calls down
1185 * into this function always.
1186 *
1187 * When a PMTU/redirect information update invalidates a route,
1188 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or
1189 * DST_OBSOLETE_DEAD by dst_free().
1190 */
1191 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt))
1192 return NULL;
1193 return dst;
1194 }
1195
1196 static void ipv4_link_failure(struct sk_buff *skb)
1197 {
1198 struct rtable *rt;
1199
1200 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
1201
1202 rt = skb_rtable(skb);
1203 if (rt)
1204 dst_set_expires(&rt->dst, 0);
1205 }
1206
1207 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb)
1208 {
1209 pr_debug("%s: %pI4 -> %pI4, %s\n",
1210 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr,
1211 skb->dev ? skb->dev->name : "?");
1212 kfree_skb(skb);
1213 WARN_ON(1);
1214 return 0;
1215 }
1216
1217 /*
1218 We do not cache source address of outgoing interface,
1219 because it is used only by IP RR, TS and SRR options,
1220 so that it out of fast path.
1221
1222 BTW remember: "addr" is allowed to be not aligned
1223 in IP options!
1224 */
1225
1226 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt)
1227 {
1228 __be32 src;
1229
1230 if (rt_is_output_route(rt))
1231 src = ip_hdr(skb)->saddr;
1232 else {
1233 struct fib_result res;
1234 struct flowi4 fl4;
1235 struct iphdr *iph;
1236
1237 iph = ip_hdr(skb);
1238
1239 memset(&fl4, 0, sizeof(fl4));
1240 fl4.daddr = iph->daddr;
1241 fl4.saddr = iph->saddr;
1242 fl4.flowi4_tos = RT_TOS(iph->tos);
1243 fl4.flowi4_oif = rt->dst.dev->ifindex;
1244 fl4.flowi4_iif = skb->dev->ifindex;
1245 fl4.flowi4_mark = skb->mark;
1246
1247 rcu_read_lock();
1248 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0)
1249 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res);
1250 else
1251 src = inet_select_addr(rt->dst.dev,
1252 rt_nexthop(rt, iph->daddr),
1253 RT_SCOPE_UNIVERSE);
1254 rcu_read_unlock();
1255 }
1256 memcpy(addr, &src, 4);
1257 }
1258
1259 #ifdef CONFIG_IP_ROUTE_CLASSID
1260 static void set_class_tag(struct rtable *rt, u32 tag)
1261 {
1262 if (!(rt->dst.tclassid & 0xFFFF))
1263 rt->dst.tclassid |= tag & 0xFFFF;
1264 if (!(rt->dst.tclassid & 0xFFFF0000))
1265 rt->dst.tclassid |= tag & 0xFFFF0000;
1266 }
1267 #endif
1268
1269 static unsigned int ipv4_default_advmss(const struct dst_entry *dst)
1270 {
1271 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr);
1272 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size,
1273 ip_rt_min_advmss);
1274
1275 return min(advmss, IPV4_MAX_PMTU - header_size);
1276 }
1277
1278 static unsigned int ipv4_mtu(const struct dst_entry *dst)
1279 {
1280 const struct rtable *rt = (const struct rtable *) dst;
1281 unsigned int mtu = rt->rt_pmtu;
1282
1283 if (!mtu || time_after_eq(jiffies, rt->dst.expires))
1284 mtu = dst_metric_raw(dst, RTAX_MTU);
1285
1286 if (mtu)
1287 return mtu;
1288
1289 mtu = READ_ONCE(dst->dev->mtu);
1290
1291 if (unlikely(ip_mtu_locked(dst))) {
1292 if (rt->rt_uses_gateway && mtu > 576)
1293 mtu = 576;
1294 }
1295
1296 mtu = min_t(unsigned int, mtu, IP_MAX_MTU);
1297
1298 return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
1299 }
1300
1301 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr)
1302 {
1303 struct fnhe_hash_bucket *hash;
1304 struct fib_nh_exception *fnhe, __rcu **fnhe_p;
1305 u32 hval = fnhe_hashfun(daddr);
1306
1307 spin_lock_bh(&fnhe_lock);
1308
1309 hash = rcu_dereference_protected(nh->nh_exceptions,
1310 lockdep_is_held(&fnhe_lock));
1311 hash += hval;
1312
1313 fnhe_p = &hash->chain;
1314 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock));
1315 while (fnhe) {
1316 if (fnhe->fnhe_daddr == daddr) {
1317 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected(
1318 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock)));
1319 fnhe_flush_routes(fnhe);
1320 kfree_rcu(fnhe, rcu);
1321 break;
1322 }
1323 fnhe_p = &fnhe->fnhe_next;
1324 fnhe = rcu_dereference_protected(fnhe->fnhe_next,
1325 lockdep_is_held(&fnhe_lock));
1326 }
1327
1328 spin_unlock_bh(&fnhe_lock);
1329 }
1330
1331 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr)
1332 {
1333 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions);
1334 struct fib_nh_exception *fnhe;
1335 u32 hval;
1336
1337 if (!hash)
1338 return NULL;
1339
1340 hval = fnhe_hashfun(daddr);
1341
1342 for (fnhe = rcu_dereference(hash[hval].chain); fnhe;
1343 fnhe = rcu_dereference(fnhe->fnhe_next)) {
1344 if (fnhe->fnhe_daddr == daddr) {
1345 if (fnhe->fnhe_expires &&
1346 time_after(jiffies, fnhe->fnhe_expires)) {
1347 ip_del_fnhe(nh, daddr);
1348 break;
1349 }
1350 return fnhe;
1351 }
1352 }
1353 return NULL;
1354 }
1355
1356 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe,
1357 __be32 daddr, const bool do_cache)
1358 {
1359 bool ret = false;
1360
1361 spin_lock_bh(&fnhe_lock);
1362
1363 if (daddr == fnhe->fnhe_daddr) {
1364 struct rtable __rcu **porig;
1365 struct rtable *orig;
1366 int genid = fnhe_genid(dev_net(rt->dst.dev));
1367
1368 if (rt_is_input_route(rt))
1369 porig = &fnhe->fnhe_rth_input;
1370 else
1371 porig = &fnhe->fnhe_rth_output;
1372 orig = rcu_dereference(*porig);
1373
1374 if (fnhe->fnhe_genid != genid) {
1375 fnhe->fnhe_genid = genid;
1376 fnhe->fnhe_gw = 0;
1377 fnhe->fnhe_pmtu = 0;
1378 fnhe->fnhe_expires = 0;
1379 fnhe_flush_routes(fnhe);
1380 orig = NULL;
1381 }
1382 fill_route_from_fnhe(rt, fnhe);
1383 if (!rt->rt_gateway)
1384 rt->rt_gateway = daddr;
1385
1386 if (do_cache) {
1387 dst_hold(&rt->dst);
1388 rcu_assign_pointer(*porig, rt);
1389 if (orig) {
1390 dst_dev_put(&orig->dst);
1391 dst_release(&orig->dst);
1392 }
1393 ret = true;
1394 }
1395
1396 fnhe->fnhe_stamp = jiffies;
1397 }
1398 spin_unlock_bh(&fnhe_lock);
1399
1400 return ret;
1401 }
1402
1403 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt)
1404 {
1405 struct rtable *orig, *prev, **p;
1406 bool ret = true;
1407
1408 if (rt_is_input_route(rt)) {
1409 p = (struct rtable **)&nh->nh_rth_input;
1410 } else {
1411 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output);
1412 }
1413 orig = *p;
1414
1415 /* hold dst before doing cmpxchg() to avoid race condition
1416 * on this dst
1417 */
1418 dst_hold(&rt->dst);
1419 prev = cmpxchg(p, orig, rt);
1420 if (prev == orig) {
1421 if (orig) {
1422 dst_dev_put(&orig->dst);
1423 dst_release(&orig->dst);
1424 }
1425 } else {
1426 dst_release(&rt->dst);
1427 ret = false;
1428 }
1429
1430 return ret;
1431 }
1432
1433 struct uncached_list {
1434 spinlock_t lock;
1435 struct list_head head;
1436 };
1437
1438 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list);
1439
1440 void rt_add_uncached_list(struct rtable *rt)
1441 {
1442 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list);
1443
1444 rt->rt_uncached_list = ul;
1445
1446 spin_lock_bh(&ul->lock);
1447 list_add_tail(&rt->rt_uncached, &ul->head);
1448 spin_unlock_bh(&ul->lock);
1449 }
1450
1451 void rt_del_uncached_list(struct rtable *rt)
1452 {
1453 if (!list_empty(&rt->rt_uncached)) {
1454 struct uncached_list *ul = rt->rt_uncached_list;
1455
1456 spin_lock_bh(&ul->lock);
1457 list_del(&rt->rt_uncached);
1458 spin_unlock_bh(&ul->lock);
1459 }
1460 }
1461
1462 static void ipv4_dst_destroy(struct dst_entry *dst)
1463 {
1464 struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst);
1465 struct rtable *rt = (struct rtable *)dst;
1466
1467 if (p != &dst_default_metrics && refcount_dec_and_test(&p->refcnt))
1468 kfree(p);
1469
1470 rt_del_uncached_list(rt);
1471 }
1472
1473 void rt_flush_dev(struct net_device *dev)
1474 {
1475 struct net *net = dev_net(dev);
1476 struct rtable *rt;
1477 int cpu;
1478
1479 for_each_possible_cpu(cpu) {
1480 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
1481
1482 spin_lock_bh(&ul->lock);
1483 list_for_each_entry(rt, &ul->head, rt_uncached) {
1484 if (rt->dst.dev != dev)
1485 continue;
1486 rt->dst.dev = net->loopback_dev;
1487 dev_hold(rt->dst.dev);
1488 dev_put(dev);
1489 }
1490 spin_unlock_bh(&ul->lock);
1491 }
1492 }
1493
1494 static bool rt_cache_valid(const struct rtable *rt)
1495 {
1496 return rt &&
1497 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
1498 !rt_is_expired(rt);
1499 }
1500
1501 static void rt_set_nexthop(struct rtable *rt, __be32 daddr,
1502 const struct fib_result *res,
1503 struct fib_nh_exception *fnhe,
1504 struct fib_info *fi, u16 type, u32 itag,
1505 const bool do_cache)
1506 {
1507 bool cached = false;
1508
1509 if (fi) {
1510 struct fib_nh *nh = &FIB_RES_NH(*res);
1511
1512 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) {
1513 rt->rt_gateway = nh->nh_gw;
1514 rt->rt_uses_gateway = 1;
1515 }
1516 dst_init_metrics(&rt->dst, fi->fib_metrics->metrics, true);
1517 if (fi->fib_metrics != &dst_default_metrics) {
1518 rt->dst._metrics |= DST_METRICS_REFCOUNTED;
1519 refcount_inc(&fi->fib_metrics->refcnt);
1520 }
1521 #ifdef CONFIG_IP_ROUTE_CLASSID
1522 rt->dst.tclassid = nh->nh_tclassid;
1523 #endif
1524 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
1525 if (unlikely(fnhe))
1526 cached = rt_bind_exception(rt, fnhe, daddr, do_cache);
1527 else if (do_cache)
1528 cached = rt_cache_route(nh, rt);
1529 if (unlikely(!cached)) {
1530 /* Routes we intend to cache in nexthop exception or
1531 * FIB nexthop have the DST_NOCACHE bit clear.
1532 * However, if we are unsuccessful at storing this
1533 * route into the cache we really need to set it.
1534 */
1535 if (!rt->rt_gateway)
1536 rt->rt_gateway = daddr;
1537 rt_add_uncached_list(rt);
1538 }
1539 } else
1540 rt_add_uncached_list(rt);
1541
1542 #ifdef CONFIG_IP_ROUTE_CLASSID
1543 #ifdef CONFIG_IP_MULTIPLE_TABLES
1544 set_class_tag(rt, res->tclassid);
1545 #endif
1546 set_class_tag(rt, itag);
1547 #endif
1548 }
1549
1550 struct rtable *rt_dst_alloc(struct net_device *dev,
1551 unsigned int flags, u16 type,
1552 bool nopolicy, bool noxfrm, bool will_cache)
1553 {
1554 struct rtable *rt;
1555
1556 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK,
1557 (will_cache ? 0 : DST_HOST) |
1558 (nopolicy ? DST_NOPOLICY : 0) |
1559 (noxfrm ? DST_NOXFRM : 0));
1560
1561 if (rt) {
1562 rt->rt_genid = rt_genid_ipv4(dev_net(dev));
1563 rt->rt_flags = flags;
1564 rt->rt_type = type;
1565 rt->rt_is_input = 0;
1566 rt->rt_iif = 0;
1567 rt->rt_pmtu = 0;
1568 rt->rt_mtu_locked = 0;
1569 rt->rt_gateway = 0;
1570 rt->rt_uses_gateway = 0;
1571 rt->rt_table_id = 0;
1572 INIT_LIST_HEAD(&rt->rt_uncached);
1573
1574 rt->dst.output = ip_output;
1575 if (flags & RTCF_LOCAL)
1576 rt->dst.input = ip_local_deliver;
1577 }
1578
1579 return rt;
1580 }
1581 EXPORT_SYMBOL(rt_dst_alloc);
1582
1583 /* called in rcu_read_lock() section */
1584 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1585 u8 tos, struct net_device *dev,
1586 struct in_device *in_dev, u32 *itag)
1587 {
1588 int err;
1589
1590 /* Primary sanity checks. */
1591 if (!in_dev)
1592 return -EINVAL;
1593
1594 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||
1595 skb->protocol != htons(ETH_P_IP))
1596 return -EINVAL;
1597
1598 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev))
1599 return -EINVAL;
1600
1601 if (ipv4_is_zeronet(saddr)) {
1602 if (!ipv4_is_local_multicast(daddr))
1603 return -EINVAL;
1604 } else {
1605 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
1606 in_dev, itag);
1607 if (err < 0)
1608 return err;
1609 }
1610 return 0;
1611 }
1612
1613 /* called in rcu_read_lock() section */
1614 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1615 u8 tos, struct net_device *dev, int our)
1616 {
1617 struct in_device *in_dev = __in_dev_get_rcu(dev);
1618 unsigned int flags = RTCF_MULTICAST;
1619 struct rtable *rth;
1620 u32 itag = 0;
1621 int err;
1622
1623 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag);
1624 if (err)
1625 return err;
1626
1627 if (our)
1628 flags |= RTCF_LOCAL;
1629
1630 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST,
1631 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false);
1632 if (!rth)
1633 return -ENOBUFS;
1634
1635 #ifdef CONFIG_IP_ROUTE_CLASSID
1636 rth->dst.tclassid = itag;
1637 #endif
1638 rth->dst.output = ip_rt_bug;
1639 rth->rt_is_input= 1;
1640
1641 #ifdef CONFIG_IP_MROUTE
1642 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev))
1643 rth->dst.input = ip_mr_input;
1644 #endif
1645 RT_CACHE_STAT_INC(in_slow_mc);
1646
1647 skb_dst_set(skb, &rth->dst);
1648 return 0;
1649 }
1650
1651
1652 static void ip_handle_martian_source(struct net_device *dev,
1653 struct in_device *in_dev,
1654 struct sk_buff *skb,
1655 __be32 daddr,
1656 __be32 saddr)
1657 {
1658 RT_CACHE_STAT_INC(in_martian_src);
1659 #ifdef CONFIG_IP_ROUTE_VERBOSE
1660 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) {
1661 /*
1662 * RFC1812 recommendation, if source is martian,
1663 * the only hint is MAC header.
1664 */
1665 pr_warn("martian source %pI4 from %pI4, on dev %s\n",
1666 &daddr, &saddr, dev->name);
1667 if (dev->hard_header_len && skb_mac_header_was_set(skb)) {
1668 print_hex_dump(KERN_WARNING, "ll header: ",
1669 DUMP_PREFIX_OFFSET, 16, 1,
1670 skb_mac_header(skb),
1671 dev->hard_header_len, true);
1672 }
1673 }
1674 #endif
1675 }
1676
1677 static void set_lwt_redirect(struct rtable *rth)
1678 {
1679 if (lwtunnel_output_redirect(rth->dst.lwtstate)) {
1680 rth->dst.lwtstate->orig_output = rth->dst.output;
1681 rth->dst.output = lwtunnel_output;
1682 }
1683
1684 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
1685 rth->dst.lwtstate->orig_input = rth->dst.input;
1686 rth->dst.input = lwtunnel_input;
1687 }
1688 }
1689
1690 /* called in rcu_read_lock() section */
1691 static int __mkroute_input(struct sk_buff *skb,
1692 const struct fib_result *res,
1693 struct in_device *in_dev,
1694 __be32 daddr, __be32 saddr, u32 tos)
1695 {
1696 struct fib_nh_exception *fnhe;
1697 struct rtable *rth;
1698 int err;
1699 struct in_device *out_dev;
1700 bool do_cache;
1701 u32 itag = 0;
1702
1703 /* get a working reference to the output device */
1704 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res));
1705 if (!out_dev) {
1706 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n");
1707 return -EINVAL;
1708 }
1709
1710 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res),
1711 in_dev->dev, in_dev, &itag);
1712 if (err < 0) {
1713 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
1714 saddr);
1715
1716 goto cleanup;
1717 }
1718
1719 do_cache = res->fi && !itag;
1720 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) &&
1721 skb->protocol == htons(ETH_P_IP) &&
1722 (IN_DEV_SHARED_MEDIA(out_dev) ||
1723 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1724 IPCB(skb)->flags |= IPSKB_DOREDIRECT;
1725
1726 if (skb->protocol != htons(ETH_P_IP)) {
1727 /* Not IP (i.e. ARP). Do not create route, if it is
1728 * invalid for proxy arp. DNAT routes are always valid.
1729 *
1730 * Proxy arp feature have been extended to allow, ARP
1731 * replies back to the same interface, to support
1732 * Private VLAN switch technologies. See arp.c.
1733 */
1734 if (out_dev == in_dev &&
1735 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) {
1736 err = -EINVAL;
1737 goto cleanup;
1738 }
1739 }
1740
1741 fnhe = find_exception(&FIB_RES_NH(*res), daddr);
1742 if (do_cache) {
1743 if (fnhe)
1744 rth = rcu_dereference(fnhe->fnhe_rth_input);
1745 else
1746 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
1747 if (rt_cache_valid(rth)) {
1748 skb_dst_set_noref(skb, &rth->dst);
1749 goto out;
1750 }
1751 }
1752
1753 rth = rt_dst_alloc(out_dev->dev, 0, res->type,
1754 IN_DEV_CONF_GET(in_dev, NOPOLICY),
1755 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache);
1756 if (!rth) {
1757 err = -ENOBUFS;
1758 goto cleanup;
1759 }
1760
1761 rth->rt_is_input = 1;
1762 if (res->table)
1763 rth->rt_table_id = res->table->tb_id;
1764 RT_CACHE_STAT_INC(in_slow_tot);
1765
1766 rth->dst.input = ip_forward;
1767
1768 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag,
1769 do_cache);
1770 set_lwt_redirect(rth);
1771 skb_dst_set(skb, &rth->dst);
1772 out:
1773 err = 0;
1774 cleanup:
1775 return err;
1776 }
1777
1778 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1779 /* To make ICMP packets follow the right flow, the multipath hash is
1780 * calculated from the inner IP addresses.
1781 */
1782 static void ip_multipath_l3_keys(const struct sk_buff *skb,
1783 struct flow_keys *hash_keys)
1784 {
1785 const struct iphdr *outer_iph = ip_hdr(skb);
1786 const struct iphdr *inner_iph;
1787 const struct icmphdr *icmph;
1788 struct iphdr _inner_iph;
1789 struct icmphdr _icmph;
1790
1791 hash_keys->addrs.v4addrs.src = outer_iph->saddr;
1792 hash_keys->addrs.v4addrs.dst = outer_iph->daddr;
1793 if (likely(outer_iph->protocol != IPPROTO_ICMP))
1794 return;
1795
1796 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0))
1797 return;
1798
1799 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph),
1800 &_icmph);
1801 if (!icmph)
1802 return;
1803
1804 if (icmph->type != ICMP_DEST_UNREACH &&
1805 icmph->type != ICMP_REDIRECT &&
1806 icmph->type != ICMP_TIME_EXCEEDED &&
1807 icmph->type != ICMP_PARAMETERPROB)
1808 return;
1809
1810 inner_iph = skb_header_pointer(skb,
1811 outer_iph->ihl * 4 + sizeof(_icmph),
1812 sizeof(_inner_iph), &_inner_iph);
1813 if (!inner_iph)
1814 return;
1815 hash_keys->addrs.v4addrs.src = inner_iph->saddr;
1816 hash_keys->addrs.v4addrs.dst = inner_iph->daddr;
1817 }
1818
1819 /* if skb is set it will be used and fl4 can be NULL */
1820 int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4,
1821 const struct sk_buff *skb)
1822 {
1823 struct net *net = fi->fib_net;
1824 struct flow_keys hash_keys;
1825 u32 mhash;
1826
1827 switch (net->ipv4.sysctl_fib_multipath_hash_policy) {
1828 case 0:
1829 memset(&hash_keys, 0, sizeof(hash_keys));
1830 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1831 if (skb) {
1832 ip_multipath_l3_keys(skb, &hash_keys);
1833 } else {
1834 hash_keys.addrs.v4addrs.src = fl4->saddr;
1835 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1836 }
1837 break;
1838 case 1:
1839 /* skb is currently provided only when forwarding */
1840 if (skb) {
1841 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1842 struct flow_keys keys;
1843
1844 /* short-circuit if we already have L4 hash present */
1845 if (skb->l4_hash)
1846 return skb_get_hash_raw(skb) >> 1;
1847 memset(&hash_keys, 0, sizeof(hash_keys));
1848 skb_flow_dissect_flow_keys(skb, &keys, flag);
1849
1850 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1851 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
1852 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
1853 hash_keys.ports.src = keys.ports.src;
1854 hash_keys.ports.dst = keys.ports.dst;
1855 hash_keys.basic.ip_proto = keys.basic.ip_proto;
1856 } else {
1857 memset(&hash_keys, 0, sizeof(hash_keys));
1858 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
1859 hash_keys.addrs.v4addrs.src = fl4->saddr;
1860 hash_keys.addrs.v4addrs.dst = fl4->daddr;
1861 hash_keys.ports.src = fl4->fl4_sport;
1862 hash_keys.ports.dst = fl4->fl4_dport;
1863 hash_keys.basic.ip_proto = fl4->flowi4_proto;
1864 }
1865 break;
1866 }
1867 mhash = flow_hash_from_keys(&hash_keys);
1868
1869 return mhash >> 1;
1870 }
1871 EXPORT_SYMBOL_GPL(fib_multipath_hash);
1872 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
1873
1874 static int ip_mkroute_input(struct sk_buff *skb,
1875 struct fib_result *res,
1876 struct in_device *in_dev,
1877 __be32 daddr, __be32 saddr, u32 tos)
1878 {
1879 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1880 if (res->fi && res->fi->fib_nhs > 1) {
1881 int h = fib_multipath_hash(res->fi, NULL, skb);
1882
1883 fib_select_multipath(res, h);
1884 }
1885 #endif
1886
1887 /* create a routing cache entry */
1888 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1889 }
1890
1891 /*
1892 * NOTE. We drop all the packets that has local source
1893 * addresses, because every properly looped back packet
1894 * must have correct destination already attached by output routine.
1895 *
1896 * Such approach solves two big problems:
1897 * 1. Not simplex devices are handled properly.
1898 * 2. IP spoofing attempts are filtered with 100% of guarantee.
1899 * called with rcu_read_lock()
1900 */
1901
1902 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1903 u8 tos, struct net_device *dev,
1904 struct fib_result *res)
1905 {
1906 struct in_device *in_dev = __in_dev_get_rcu(dev);
1907 struct ip_tunnel_info *tun_info;
1908 struct flowi4 fl4;
1909 unsigned int flags = 0;
1910 u32 itag = 0;
1911 struct rtable *rth;
1912 int err = -EINVAL;
1913 struct net *net = dev_net(dev);
1914 bool do_cache;
1915
1916 /* IP on this device is disabled. */
1917
1918 if (!in_dev)
1919 goto out;
1920
1921 /* Check for the most weird martians, which can be not detected
1922 by fib_lookup.
1923 */
1924
1925 tun_info = skb_tunnel_info(skb);
1926 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
1927 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id;
1928 else
1929 fl4.flowi4_tun_key.tun_id = 0;
1930 skb_dst_drop(skb);
1931
1932 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr))
1933 goto martian_source;
1934
1935 res->fi = NULL;
1936 res->table = NULL;
1937 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0))
1938 goto brd_input;
1939
1940 /* Accept zero addresses only to limited broadcast;
1941 * I even do not know to fix it or not. Waiting for complains :-)
1942 */
1943 if (ipv4_is_zeronet(saddr))
1944 goto martian_source;
1945
1946 if (ipv4_is_zeronet(daddr))
1947 goto martian_destination;
1948
1949 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(),
1950 * and call it once if daddr or/and saddr are loopback addresses
1951 */
1952 if (ipv4_is_loopback(daddr)) {
1953 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1954 goto martian_destination;
1955 } else if (ipv4_is_loopback(saddr)) {
1956 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net))
1957 goto martian_source;
1958 }
1959
1960 /*
1961 * Now we are ready to route packet.
1962 */
1963 fl4.flowi4_oif = 0;
1964 fl4.flowi4_iif = dev->ifindex;
1965 fl4.flowi4_mark = skb->mark;
1966 fl4.flowi4_tos = tos;
1967 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
1968 fl4.flowi4_flags = 0;
1969 fl4.daddr = daddr;
1970 fl4.saddr = saddr;
1971 fl4.flowi4_uid = sock_net_uid(net, NULL);
1972 err = fib_lookup(net, &fl4, res, 0);
1973 if (err != 0) {
1974 if (!IN_DEV_FORWARD(in_dev))
1975 err = -EHOSTUNREACH;
1976 goto no_route;
1977 }
1978
1979 if (res->type == RTN_BROADCAST)
1980 goto brd_input;
1981
1982 if (res->type == RTN_LOCAL) {
1983 err = fib_validate_source(skb, saddr, daddr, tos,
1984 0, dev, in_dev, &itag);
1985 if (err < 0)
1986 goto martian_source;
1987 goto local_input;
1988 }
1989
1990 if (!IN_DEV_FORWARD(in_dev)) {
1991 err = -EHOSTUNREACH;
1992 goto no_route;
1993 }
1994 if (res->type != RTN_UNICAST)
1995 goto martian_destination;
1996
1997 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos);
1998 out: return err;
1999
2000 brd_input:
2001 if (skb->protocol != htons(ETH_P_IP))
2002 goto e_inval;
2003
2004 if (!ipv4_is_zeronet(saddr)) {
2005 err = fib_validate_source(skb, saddr, 0, tos, 0, dev,
2006 in_dev, &itag);
2007 if (err < 0)
2008 goto martian_source;
2009 }
2010 flags |= RTCF_BROADCAST;
2011 res->type = RTN_BROADCAST;
2012 RT_CACHE_STAT_INC(in_brd);
2013
2014 local_input:
2015 do_cache = false;
2016 if (res->fi) {
2017 if (!itag) {
2018 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
2019 if (rt_cache_valid(rth)) {
2020 skb_dst_set_noref(skb, &rth->dst);
2021 err = 0;
2022 goto out;
2023 }
2024 do_cache = true;
2025 }
2026 }
2027
2028 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev,
2029 flags | RTCF_LOCAL, res->type,
2030 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache);
2031 if (!rth)
2032 goto e_nobufs;
2033
2034 rth->dst.output= ip_rt_bug;
2035 #ifdef CONFIG_IP_ROUTE_CLASSID
2036 rth->dst.tclassid = itag;
2037 #endif
2038 rth->rt_is_input = 1;
2039 if (res->table)
2040 rth->rt_table_id = res->table->tb_id;
2041
2042 RT_CACHE_STAT_INC(in_slow_tot);
2043 if (res->type == RTN_UNREACHABLE) {
2044 rth->dst.input= ip_error;
2045 rth->dst.error= -err;
2046 rth->rt_flags &= ~RTCF_LOCAL;
2047 }
2048
2049 if (do_cache) {
2050 struct fib_nh *nh = &FIB_RES_NH(*res);
2051
2052 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate);
2053 if (lwtunnel_input_redirect(rth->dst.lwtstate)) {
2054 WARN_ON(rth->dst.input == lwtunnel_input);
2055 rth->dst.lwtstate->orig_input = rth->dst.input;
2056 rth->dst.input = lwtunnel_input;
2057 }
2058
2059 if (unlikely(!rt_cache_route(nh, rth)))
2060 rt_add_uncached_list(rth);
2061 }
2062 skb_dst_set(skb, &rth->dst);
2063 err = 0;
2064 goto out;
2065
2066 no_route:
2067 RT_CACHE_STAT_INC(in_no_route);
2068 res->type = RTN_UNREACHABLE;
2069 res->fi = NULL;
2070 res->table = NULL;
2071 goto local_input;
2072
2073 /*
2074 * Do not cache martian addresses: they should be logged (RFC1812)
2075 */
2076 martian_destination:
2077 RT_CACHE_STAT_INC(in_martian_dst);
2078 #ifdef CONFIG_IP_ROUTE_VERBOSE
2079 if (IN_DEV_LOG_MARTIANS(in_dev))
2080 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n",
2081 &daddr, &saddr, dev->name);
2082 #endif
2083
2084 e_inval:
2085 err = -EINVAL;
2086 goto out;
2087
2088 e_nobufs:
2089 err = -ENOBUFS;
2090 goto out;
2091
2092 martian_source:
2093 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
2094 goto out;
2095 }
2096
2097 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2098 u8 tos, struct net_device *dev)
2099 {
2100 struct fib_result res;
2101 int err;
2102
2103 tos &= IPTOS_RT_MASK;
2104 rcu_read_lock();
2105 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res);
2106 rcu_read_unlock();
2107
2108 return err;
2109 }
2110 EXPORT_SYMBOL(ip_route_input_noref);
2111
2112 /* called with rcu_read_lock held */
2113 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr,
2114 u8 tos, struct net_device *dev, struct fib_result *res)
2115 {
2116 /* Multicast recognition logic is moved from route cache to here.
2117 The problem was that too many Ethernet cards have broken/missing
2118 hardware multicast filters :-( As result the host on multicasting
2119 network acquires a lot of useless route cache entries, sort of
2120 SDR messages from all the world. Now we try to get rid of them.
2121 Really, provided software IP multicast filter is organized
2122 reasonably (at least, hashed), it does not result in a slowdown
2123 comparing with route cache reject entries.
2124 Note, that multicast routers are not affected, because
2125 route cache entry is created eventually.
2126 */
2127 if (ipv4_is_multicast(daddr)) {
2128 struct in_device *in_dev = __in_dev_get_rcu(dev);
2129 int our = 0;
2130 int err = -EINVAL;
2131
2132 if (in_dev)
2133 our = ip_check_mc_rcu(in_dev, daddr, saddr,
2134 ip_hdr(skb)->protocol);
2135
2136 /* check l3 master if no match yet */
2137 if ((!in_dev || !our) && netif_is_l3_slave(dev)) {
2138 struct in_device *l3_in_dev;
2139
2140 l3_in_dev = __in_dev_get_rcu(skb->dev);
2141 if (l3_in_dev)
2142 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr,
2143 ip_hdr(skb)->protocol);
2144 }
2145
2146 if (our
2147 #ifdef CONFIG_IP_MROUTE
2148 ||
2149 (!ipv4_is_local_multicast(daddr) &&
2150 IN_DEV_MFORWARD(in_dev))
2151 #endif
2152 ) {
2153 err = ip_route_input_mc(skb, daddr, saddr,
2154 tos, dev, our);
2155 }
2156 return err;
2157 }
2158
2159 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res);
2160 }
2161
2162 /* called with rcu_read_lock() */
2163 static struct rtable *__mkroute_output(const struct fib_result *res,
2164 const struct flowi4 *fl4, int orig_oif,
2165 struct net_device *dev_out,
2166 unsigned int flags)
2167 {
2168 struct fib_info *fi = res->fi;
2169 struct fib_nh_exception *fnhe;
2170 struct in_device *in_dev;
2171 u16 type = res->type;
2172 struct rtable *rth;
2173 bool do_cache;
2174
2175 in_dev = __in_dev_get_rcu(dev_out);
2176 if (!in_dev)
2177 return ERR_PTR(-EINVAL);
2178
2179 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev)))
2180 if (ipv4_is_loopback(fl4->saddr) &&
2181 !(dev_out->flags & IFF_LOOPBACK) &&
2182 !netif_is_l3_master(dev_out))
2183 return ERR_PTR(-EINVAL);
2184
2185 if (ipv4_is_lbcast(fl4->daddr))
2186 type = RTN_BROADCAST;
2187 else if (ipv4_is_multicast(fl4->daddr))
2188 type = RTN_MULTICAST;
2189 else if (ipv4_is_zeronet(fl4->daddr))
2190 return ERR_PTR(-EINVAL);
2191
2192 if (dev_out->flags & IFF_LOOPBACK)
2193 flags |= RTCF_LOCAL;
2194
2195 do_cache = true;
2196 if (type == RTN_BROADCAST) {
2197 flags |= RTCF_BROADCAST | RTCF_LOCAL;
2198 fi = NULL;
2199 } else if (type == RTN_MULTICAST) {
2200 flags |= RTCF_MULTICAST | RTCF_LOCAL;
2201 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr,
2202 fl4->flowi4_proto))
2203 flags &= ~RTCF_LOCAL;
2204 else
2205 do_cache = false;
2206 /* If multicast route do not exist use
2207 * default one, but do not gateway in this case.
2208 * Yes, it is hack.
2209 */
2210 if (fi && res->prefixlen < 4)
2211 fi = NULL;
2212 } else if ((type == RTN_LOCAL) && (orig_oif != 0) &&
2213 (orig_oif != dev_out->ifindex)) {
2214 /* For local routes that require a particular output interface
2215 * we do not want to cache the result. Caching the result
2216 * causes incorrect behaviour when there are multiple source
2217 * addresses on the interface, the end result being that if the
2218 * intended recipient is waiting on that interface for the
2219 * packet he won't receive it because it will be delivered on
2220 * the loopback interface and the IP_PKTINFO ipi_ifindex will
2221 * be set to the loopback interface as well.
2222 */
2223 do_cache = false;
2224 }
2225
2226 fnhe = NULL;
2227 do_cache &= fi != NULL;
2228 if (fi) {
2229 struct rtable __rcu **prth;
2230 struct fib_nh *nh = &FIB_RES_NH(*res);
2231
2232 fnhe = find_exception(nh, fl4->daddr);
2233 if (!do_cache)
2234 goto add;
2235 if (fnhe) {
2236 prth = &fnhe->fnhe_rth_output;
2237 } else {
2238 if (unlikely(fl4->flowi4_flags &
2239 FLOWI_FLAG_KNOWN_NH &&
2240 !(nh->nh_gw &&
2241 nh->nh_scope == RT_SCOPE_LINK))) {
2242 do_cache = false;
2243 goto add;
2244 }
2245 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output);
2246 }
2247 rth = rcu_dereference(*prth);
2248 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst))
2249 return rth;
2250 }
2251
2252 add:
2253 rth = rt_dst_alloc(dev_out, flags, type,
2254 IN_DEV_CONF_GET(in_dev, NOPOLICY),
2255 IN_DEV_CONF_GET(in_dev, NOXFRM),
2256 do_cache);
2257 if (!rth)
2258 return ERR_PTR(-ENOBUFS);
2259
2260 rth->rt_iif = orig_oif;
2261 if (res->table)
2262 rth->rt_table_id = res->table->tb_id;
2263
2264 RT_CACHE_STAT_INC(out_slow_tot);
2265
2266 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
2267 if (flags & RTCF_LOCAL &&
2268 !(dev_out->flags & IFF_LOOPBACK)) {
2269 rth->dst.output = ip_mc_output;
2270 RT_CACHE_STAT_INC(out_slow_mc);
2271 }
2272 #ifdef CONFIG_IP_MROUTE
2273 if (type == RTN_MULTICAST) {
2274 if (IN_DEV_MFORWARD(in_dev) &&
2275 !ipv4_is_local_multicast(fl4->daddr)) {
2276 rth->dst.input = ip_mr_input;
2277 rth->dst.output = ip_mc_output;
2278 }
2279 }
2280 #endif
2281 }
2282
2283 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache);
2284 set_lwt_redirect(rth);
2285
2286 return rth;
2287 }
2288
2289 /*
2290 * Major route resolver routine.
2291 */
2292
2293 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4,
2294 const struct sk_buff *skb)
2295 {
2296 __u8 tos = RT_FL_TOS(fl4);
2297 struct fib_result res = {
2298 .type = RTN_UNSPEC,
2299 .fi = NULL,
2300 .table = NULL,
2301 .tclassid = 0,
2302 };
2303 struct rtable *rth;
2304
2305 fl4->flowi4_iif = LOOPBACK_IFINDEX;
2306 fl4->flowi4_tos = tos & IPTOS_RT_MASK;
2307 fl4->flowi4_scope = ((tos & RTO_ONLINK) ?
2308 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE);
2309
2310 rcu_read_lock();
2311 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb);
2312 rcu_read_unlock();
2313
2314 return rth;
2315 }
2316 EXPORT_SYMBOL_GPL(ip_route_output_key_hash);
2317
2318 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4,
2319 struct fib_result *res,
2320 const struct sk_buff *skb)
2321 {
2322 struct net_device *dev_out = NULL;
2323 int orig_oif = fl4->flowi4_oif;
2324 unsigned int flags = 0;
2325 struct rtable *rth;
2326 int err = -ENETUNREACH;
2327
2328 if (fl4->saddr) {
2329 rth = ERR_PTR(-EINVAL);
2330 if (ipv4_is_multicast(fl4->saddr) ||
2331 ipv4_is_lbcast(fl4->saddr) ||
2332 ipv4_is_zeronet(fl4->saddr))
2333 goto out;
2334
2335 /* I removed check for oif == dev_out->oif here.
2336 It was wrong for two reasons:
2337 1. ip_dev_find(net, saddr) can return wrong iface, if saddr
2338 is assigned to multiple interfaces.
2339 2. Moreover, we are allowed to send packets with saddr
2340 of another iface. --ANK
2341 */
2342
2343 if (fl4->flowi4_oif == 0 &&
2344 (ipv4_is_multicast(fl4->daddr) ||
2345 ipv4_is_lbcast(fl4->daddr))) {
2346 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2347 dev_out = __ip_dev_find(net, fl4->saddr, false);
2348 if (!dev_out)
2349 goto out;
2350
2351 /* Special hack: user can direct multicasts
2352 and limited broadcast via necessary interface
2353 without fiddling with IP_MULTICAST_IF or IP_PKTINFO.
2354 This hack is not just for fun, it allows
2355 vic,vat and friends to work.
2356 They bind socket to loopback, set ttl to zero
2357 and expect that it will work.
2358 From the viewpoint of routing cache they are broken,
2359 because we are not allowed to build multicast path
2360 with loopback source addr (look, routing cache
2361 cannot know, that ttl is zero, so that packet
2362 will not leave this host and route is valid).
2363 Luckily, this hack is good workaround.
2364 */
2365
2366 fl4->flowi4_oif = dev_out->ifindex;
2367 goto make_route;
2368 }
2369
2370 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) {
2371 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */
2372 if (!__ip_dev_find(net, fl4->saddr, false))
2373 goto out;
2374 }
2375 }
2376
2377
2378 if (fl4->flowi4_oif) {
2379 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif);
2380 rth = ERR_PTR(-ENODEV);
2381 if (!dev_out)
2382 goto out;
2383
2384 /* RACE: Check return value of inet_select_addr instead. */
2385 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) {
2386 rth = ERR_PTR(-ENETUNREACH);
2387 goto out;
2388 }
2389 if (ipv4_is_local_multicast(fl4->daddr) ||
2390 ipv4_is_lbcast(fl4->daddr) ||
2391 fl4->flowi4_proto == IPPROTO_IGMP) {
2392 if (!fl4->saddr)
2393 fl4->saddr = inet_select_addr(dev_out, 0,
2394 RT_SCOPE_LINK);
2395 goto make_route;
2396 }
2397 if (!fl4->saddr) {
2398 if (ipv4_is_multicast(fl4->daddr))
2399 fl4->saddr = inet_select_addr(dev_out, 0,
2400 fl4->flowi4_scope);
2401 else if (!fl4->daddr)
2402 fl4->saddr = inet_select_addr(dev_out, 0,
2403 RT_SCOPE_HOST);
2404 }
2405 }
2406
2407 if (!fl4->daddr) {
2408 fl4->daddr = fl4->saddr;
2409 if (!fl4->daddr)
2410 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK);
2411 dev_out = net->loopback_dev;
2412 fl4->flowi4_oif = LOOPBACK_IFINDEX;
2413 res->type = RTN_LOCAL;
2414 flags |= RTCF_LOCAL;
2415 goto make_route;
2416 }
2417
2418 err = fib_lookup(net, fl4, res, 0);
2419 if (err) {
2420 res->fi = NULL;
2421 res->table = NULL;
2422 if (fl4->flowi4_oif &&
2423 (ipv4_is_multicast(fl4->daddr) ||
2424 !netif_index_is_l3_master(net, fl4->flowi4_oif))) {
2425 /* Apparently, routing tables are wrong. Assume,
2426 that the destination is on link.
2427
2428 WHY? DW.
2429 Because we are allowed to send to iface
2430 even if it has NO routes and NO assigned
2431 addresses. When oif is specified, routing
2432 tables are looked up with only one purpose:
2433 to catch if destination is gatewayed, rather than
2434 direct. Moreover, if MSG_DONTROUTE is set,
2435 we send packet, ignoring both routing tables
2436 and ifaddr state. --ANK
2437
2438
2439 We could make it even if oif is unknown,
2440 likely IPv6, but we do not.
2441 */
2442
2443 if (fl4->saddr == 0)
2444 fl4->saddr = inet_select_addr(dev_out, 0,
2445 RT_SCOPE_LINK);
2446 res->type = RTN_UNICAST;
2447 goto make_route;
2448 }
2449 rth = ERR_PTR(err);
2450 goto out;
2451 }
2452
2453 if (res->type == RTN_LOCAL) {
2454 if (!fl4->saddr) {
2455 if (res->fi->fib_prefsrc)
2456 fl4->saddr = res->fi->fib_prefsrc;
2457 else
2458 fl4->saddr = fl4->daddr;
2459 }
2460
2461 /* L3 master device is the loopback for that domain */
2462 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? :
2463 net->loopback_dev;
2464
2465 /* make sure orig_oif points to fib result device even
2466 * though packet rx/tx happens over loopback or l3mdev
2467 */
2468 orig_oif = FIB_RES_OIF(*res);
2469
2470 fl4->flowi4_oif = dev_out->ifindex;
2471 flags |= RTCF_LOCAL;
2472 goto make_route;
2473 }
2474
2475 fib_select_path(net, res, fl4, skb);
2476
2477 dev_out = FIB_RES_DEV(*res);
2478 fl4->flowi4_oif = dev_out->ifindex;
2479
2480
2481 make_route:
2482 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags);
2483
2484 out:
2485 return rth;
2486 }
2487
2488 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie)
2489 {
2490 return NULL;
2491 }
2492
2493 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst)
2494 {
2495 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
2496
2497 return mtu ? : dst->dev->mtu;
2498 }
2499
2500 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
2501 struct sk_buff *skb, u32 mtu)
2502 {
2503 }
2504
2505 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
2506 struct sk_buff *skb)
2507 {
2508 }
2509
2510 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst,
2511 unsigned long old)
2512 {
2513 return NULL;
2514 }
2515
2516 static struct dst_ops ipv4_dst_blackhole_ops = {
2517 .family = AF_INET,
2518 .check = ipv4_blackhole_dst_check,
2519 .mtu = ipv4_blackhole_mtu,
2520 .default_advmss = ipv4_default_advmss,
2521 .update_pmtu = ipv4_rt_blackhole_update_pmtu,
2522 .redirect = ipv4_rt_blackhole_redirect,
2523 .cow_metrics = ipv4_rt_blackhole_cow_metrics,
2524 .neigh_lookup = ipv4_neigh_lookup,
2525 };
2526
2527 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2528 {
2529 struct rtable *ort = (struct rtable *) dst_orig;
2530 struct rtable *rt;
2531
2532 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0);
2533 if (rt) {
2534 struct dst_entry *new = &rt->dst;
2535
2536 new->__use = 1;
2537 new->input = dst_discard;
2538 new->output = dst_discard_out;
2539
2540 new->dev = net->loopback_dev;
2541 if (new->dev)
2542 dev_hold(new->dev);
2543
2544 rt->rt_is_input = ort->rt_is_input;
2545 rt->rt_iif = ort->rt_iif;
2546 rt->rt_pmtu = ort->rt_pmtu;
2547 rt->rt_mtu_locked = ort->rt_mtu_locked;
2548
2549 rt->rt_genid = rt_genid_ipv4(net);
2550 rt->rt_flags = ort->rt_flags;
2551 rt->rt_type = ort->rt_type;
2552 rt->rt_gateway = ort->rt_gateway;
2553 rt->rt_uses_gateway = ort->rt_uses_gateway;
2554
2555 INIT_LIST_HEAD(&rt->rt_uncached);
2556 }
2557
2558 dst_release(dst_orig);
2559
2560 return rt ? &rt->dst : ERR_PTR(-ENOMEM);
2561 }
2562
2563 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4,
2564 const struct sock *sk)
2565 {
2566 struct rtable *rt = __ip_route_output_key(net, flp4);
2567
2568 if (IS_ERR(rt))
2569 return rt;
2570
2571 if (flp4->flowi4_proto)
2572 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst,
2573 flowi4_to_flowi(flp4),
2574 sk, 0);
2575
2576 return rt;
2577 }
2578 EXPORT_SYMBOL_GPL(ip_route_output_flow);
2579
2580 /* called with rcu_read_lock held */
2581 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id,
2582 struct flowi4 *fl4, struct sk_buff *skb, u32 portid,
2583 u32 seq)
2584 {
2585 struct rtable *rt = skb_rtable(skb);
2586 struct rtmsg *r;
2587 struct nlmsghdr *nlh;
2588 unsigned long expires = 0;
2589 u32 error;
2590 u32 metrics[RTAX_MAX];
2591
2592 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0);
2593 if (!nlh)
2594 return -EMSGSIZE;
2595
2596 r = nlmsg_data(nlh);
2597 r->rtm_family = AF_INET;
2598 r->rtm_dst_len = 32;
2599 r->rtm_src_len = 0;
2600 r->rtm_tos = fl4->flowi4_tos;
2601 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT;
2602 if (nla_put_u32(skb, RTA_TABLE, table_id))
2603 goto nla_put_failure;
2604 r->rtm_type = rt->rt_type;
2605 r->rtm_scope = RT_SCOPE_UNIVERSE;
2606 r->rtm_protocol = RTPROT_UNSPEC;
2607 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
2608 if (rt->rt_flags & RTCF_NOTIFY)
2609 r->rtm_flags |= RTM_F_NOTIFY;
2610 if (IPCB(skb)->flags & IPSKB_DOREDIRECT)
2611 r->rtm_flags |= RTCF_DOREDIRECT;
2612
2613 if (nla_put_in_addr(skb, RTA_DST, dst))
2614 goto nla_put_failure;
2615 if (src) {
2616 r->rtm_src_len = 32;
2617 if (nla_put_in_addr(skb, RTA_SRC, src))
2618 goto nla_put_failure;
2619 }
2620 if (rt->dst.dev &&
2621 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex))
2622 goto nla_put_failure;
2623 #ifdef CONFIG_IP_ROUTE_CLASSID
2624 if (rt->dst.tclassid &&
2625 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid))
2626 goto nla_put_failure;
2627 #endif
2628 if (!rt_is_input_route(rt) &&
2629 fl4->saddr != src) {
2630 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr))
2631 goto nla_put_failure;
2632 }
2633 if (rt->rt_uses_gateway &&
2634 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway))
2635 goto nla_put_failure;
2636
2637 expires = rt->dst.expires;
2638 if (expires) {
2639 unsigned long now = jiffies;
2640
2641 if (time_before(now, expires))
2642 expires -= now;
2643 else
2644 expires = 0;
2645 }
2646
2647 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics));
2648 if (rt->rt_pmtu && expires)
2649 metrics[RTAX_MTU - 1] = rt->rt_pmtu;
2650 if (rt->rt_mtu_locked && expires)
2651 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU);
2652 if (rtnetlink_put_metrics(skb, metrics) < 0)
2653 goto nla_put_failure;
2654
2655 if (fl4->flowi4_mark &&
2656 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark))
2657 goto nla_put_failure;
2658
2659 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) &&
2660 nla_put_u32(skb, RTA_UID,
2661 from_kuid_munged(current_user_ns(), fl4->flowi4_uid)))
2662 goto nla_put_failure;
2663
2664 error = rt->dst.error;
2665
2666 if (rt_is_input_route(rt)) {
2667 #ifdef CONFIG_IP_MROUTE
2668 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) &&
2669 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) {
2670 int err = ipmr_get_route(net, skb,
2671 fl4->saddr, fl4->daddr,
2672 r, portid);
2673
2674 if (err <= 0) {
2675 if (err == 0)
2676 return 0;
2677 goto nla_put_failure;
2678 }
2679 } else
2680 #endif
2681 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex))
2682 goto nla_put_failure;
2683 }
2684
2685 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0)
2686 goto nla_put_failure;
2687
2688 nlmsg_end(skb, nlh);
2689 return 0;
2690
2691 nla_put_failure:
2692 nlmsg_cancel(skb, nlh);
2693 return -EMSGSIZE;
2694 }
2695
2696 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
2697 struct netlink_ext_ack *extack)
2698 {
2699 struct net *net = sock_net(in_skb->sk);
2700 struct rtmsg *rtm;
2701 struct nlattr *tb[RTA_MAX+1];
2702 struct fib_result res = {};
2703 struct rtable *rt = NULL;
2704 struct flowi4 fl4;
2705 __be32 dst = 0;
2706 __be32 src = 0;
2707 u32 iif;
2708 int err;
2709 int mark;
2710 struct sk_buff *skb;
2711 u32 table_id = RT_TABLE_MAIN;
2712 kuid_t uid;
2713
2714 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy,
2715 extack);
2716 if (err < 0)
2717 goto errout;
2718
2719 rtm = nlmsg_data(nlh);
2720
2721 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2722 if (!skb) {
2723 err = -ENOBUFS;
2724 goto errout;
2725 }
2726
2727 /* Reserve room for dummy headers, this skb can pass
2728 through good chunk of routing engine.
2729 */
2730 skb_reset_mac_header(skb);
2731 skb_reset_network_header(skb);
2732
2733 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0;
2734 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0;
2735 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0;
2736 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0;
2737 if (tb[RTA_UID])
2738 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID]));
2739 else
2740 uid = (iif ? INVALID_UID : current_uid());
2741
2742 /* Bugfix: need to give ip_route_input enough of an IP header to
2743 * not gag.
2744 */
2745 ip_hdr(skb)->protocol = IPPROTO_UDP;
2746 ip_hdr(skb)->saddr = src;
2747 ip_hdr(skb)->daddr = dst;
2748
2749 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr));
2750
2751 memset(&fl4, 0, sizeof(fl4));
2752 fl4.daddr = dst;
2753 fl4.saddr = src;
2754 fl4.flowi4_tos = rtm->rtm_tos;
2755 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0;
2756 fl4.flowi4_mark = mark;
2757 fl4.flowi4_uid = uid;
2758
2759 rcu_read_lock();
2760
2761 if (iif) {
2762 struct net_device *dev;
2763
2764 dev = dev_get_by_index_rcu(net, iif);
2765 if (!dev) {
2766 err = -ENODEV;
2767 goto errout_free;
2768 }
2769
2770 skb->protocol = htons(ETH_P_IP);
2771 skb->dev = dev;
2772 skb->mark = mark;
2773 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos,
2774 dev, &res);
2775
2776 rt = skb_rtable(skb);
2777 if (err == 0 && rt->dst.error)
2778 err = -rt->dst.error;
2779 } else {
2780 fl4.flowi4_iif = LOOPBACK_IFINDEX;
2781 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb);
2782 err = 0;
2783 if (IS_ERR(rt))
2784 err = PTR_ERR(rt);
2785 else
2786 skb_dst_set(skb, &rt->dst);
2787 }
2788
2789 if (err)
2790 goto errout_free;
2791
2792 if (rtm->rtm_flags & RTM_F_NOTIFY)
2793 rt->rt_flags |= RTCF_NOTIFY;
2794
2795 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE)
2796 table_id = rt->rt_table_id;
2797
2798 if (rtm->rtm_flags & RTM_F_FIB_MATCH) {
2799 if (!res.fi) {
2800 err = fib_props[res.type].error;
2801 if (!err)
2802 err = -EHOSTUNREACH;
2803 goto errout_free;
2804 }
2805 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid,
2806 nlh->nlmsg_seq, RTM_NEWROUTE, table_id,
2807 rt->rt_type, res.prefix, res.prefixlen,
2808 fl4.flowi4_tos, res.fi, 0);
2809 } else {
2810 err = rt_fill_info(net, dst, src, table_id, &fl4, skb,
2811 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq);
2812 }
2813 if (err < 0)
2814 goto errout_free;
2815
2816 rcu_read_unlock();
2817
2818 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
2819 errout:
2820 return err;
2821
2822 errout_free:
2823 rcu_read_unlock();
2824 kfree_skb(skb);
2825 goto errout;
2826 }
2827
2828 void ip_rt_multicast_event(struct in_device *in_dev)
2829 {
2830 rt_cache_flush(dev_net(in_dev->dev));
2831 }
2832
2833 #ifdef CONFIG_SYSCTL
2834 static int ip_rt_gc_interval __read_mostly = 60 * HZ;
2835 static int ip_rt_gc_min_interval __read_mostly = HZ / 2;
2836 static int ip_rt_gc_elasticity __read_mostly = 8;
2837
2838 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write,
2839 void __user *buffer,
2840 size_t *lenp, loff_t *ppos)
2841 {
2842 struct net *net = (struct net *)__ctl->extra1;
2843
2844 if (write) {
2845 rt_cache_flush(net);
2846 fnhe_genid_bump(net);
2847 return 0;
2848 }
2849
2850 return -EINVAL;
2851 }
2852
2853 static struct ctl_table ipv4_route_table[] = {
2854 {
2855 .procname = "gc_thresh",
2856 .data = &ipv4_dst_ops.gc_thresh,
2857 .maxlen = sizeof(int),
2858 .mode = 0644,
2859 .proc_handler = proc_dointvec,
2860 },
2861 {
2862 .procname = "max_size",
2863 .data = &ip_rt_max_size,
2864 .maxlen = sizeof(int),
2865 .mode = 0644,
2866 .proc_handler = proc_dointvec,
2867 },
2868 {
2869 /* Deprecated. Use gc_min_interval_ms */
2870
2871 .procname = "gc_min_interval",
2872 .data = &ip_rt_gc_min_interval,
2873 .maxlen = sizeof(int),
2874 .mode = 0644,
2875 .proc_handler = proc_dointvec_jiffies,
2876 },
2877 {
2878 .procname = "gc_min_interval_ms",
2879 .data = &ip_rt_gc_min_interval,
2880 .maxlen = sizeof(int),
2881 .mode = 0644,
2882 .proc_handler = proc_dointvec_ms_jiffies,
2883 },
2884 {
2885 .procname = "gc_timeout",
2886 .data = &ip_rt_gc_timeout,
2887 .maxlen = sizeof(int),
2888 .mode = 0644,
2889 .proc_handler = proc_dointvec_jiffies,
2890 },
2891 {
2892 .procname = "gc_interval",
2893 .data = &ip_rt_gc_interval,
2894 .maxlen = sizeof(int),
2895 .mode = 0644,
2896 .proc_handler = proc_dointvec_jiffies,
2897 },
2898 {
2899 .procname = "redirect_load",
2900 .data = &ip_rt_redirect_load,
2901 .maxlen = sizeof(int),
2902 .mode = 0644,
2903 .proc_handler = proc_dointvec,
2904 },
2905 {
2906 .procname = "redirect_number",
2907 .data = &ip_rt_redirect_number,
2908 .maxlen = sizeof(int),
2909 .mode = 0644,
2910 .proc_handler = proc_dointvec,
2911 },
2912 {
2913 .procname = "redirect_silence",
2914 .data = &ip_rt_redirect_silence,
2915 .maxlen = sizeof(int),
2916 .mode = 0644,
2917 .proc_handler = proc_dointvec,
2918 },
2919 {
2920 .procname = "error_cost",
2921 .data = &ip_rt_error_cost,
2922 .maxlen = sizeof(int),
2923 .mode = 0644,
2924 .proc_handler = proc_dointvec,
2925 },
2926 {
2927 .procname = "error_burst",
2928 .data = &ip_rt_error_burst,
2929 .maxlen = sizeof(int),
2930 .mode = 0644,
2931 .proc_handler = proc_dointvec,
2932 },
2933 {
2934 .procname = "gc_elasticity",
2935 .data = &ip_rt_gc_elasticity,
2936 .maxlen = sizeof(int),
2937 .mode = 0644,
2938 .proc_handler = proc_dointvec,
2939 },
2940 {
2941 .procname = "mtu_expires",
2942 .data = &ip_rt_mtu_expires,
2943 .maxlen = sizeof(int),
2944 .mode = 0644,
2945 .proc_handler = proc_dointvec_jiffies,
2946 },
2947 {
2948 .procname = "min_pmtu",
2949 .data = &ip_rt_min_pmtu,
2950 .maxlen = sizeof(int),
2951 .mode = 0644,
2952 .proc_handler = proc_dointvec_minmax,
2953 .extra1 = &ip_min_valid_pmtu,
2954 },
2955 {
2956 .procname = "min_adv_mss",
2957 .data = &ip_rt_min_advmss,
2958 .maxlen = sizeof(int),
2959 .mode = 0644,
2960 .proc_handler = proc_dointvec,
2961 },
2962 { }
2963 };
2964
2965 static struct ctl_table ipv4_route_flush_table[] = {
2966 {
2967 .procname = "flush",
2968 .maxlen = sizeof(int),
2969 .mode = 0200,
2970 .proc_handler = ipv4_sysctl_rtcache_flush,
2971 },
2972 { },
2973 };
2974
2975 static __net_init int sysctl_route_net_init(struct net *net)
2976 {
2977 struct ctl_table *tbl;
2978
2979 tbl = ipv4_route_flush_table;
2980 if (!net_eq(net, &init_net)) {
2981 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL);
2982 if (!tbl)
2983 goto err_dup;
2984
2985 /* Don't export sysctls to unprivileged users */
2986 if (net->user_ns != &init_user_ns)
2987 tbl[0].procname = NULL;
2988 }
2989 tbl[0].extra1 = net;
2990
2991 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl);
2992 if (!net->ipv4.route_hdr)
2993 goto err_reg;
2994 return 0;
2995
2996 err_reg:
2997 if (tbl != ipv4_route_flush_table)
2998 kfree(tbl);
2999 err_dup:
3000 return -ENOMEM;
3001 }
3002
3003 static __net_exit void sysctl_route_net_exit(struct net *net)
3004 {
3005 struct ctl_table *tbl;
3006
3007 tbl = net->ipv4.route_hdr->ctl_table_arg;
3008 unregister_net_sysctl_table(net->ipv4.route_hdr);
3009 BUG_ON(tbl == ipv4_route_flush_table);
3010 kfree(tbl);
3011 }
3012
3013 static __net_initdata struct pernet_operations sysctl_route_ops = {
3014 .init = sysctl_route_net_init,
3015 .exit = sysctl_route_net_exit,
3016 };
3017 #endif
3018
3019 static __net_init int rt_genid_init(struct net *net)
3020 {
3021 atomic_set(&net->ipv4.rt_genid, 0);
3022 atomic_set(&net->fnhe_genid, 0);
3023 atomic_set(&net->ipv4.dev_addr_genid, get_random_int());
3024 return 0;
3025 }
3026
3027 static __net_initdata struct pernet_operations rt_genid_ops = {
3028 .init = rt_genid_init,
3029 };
3030
3031 static int __net_init ipv4_inetpeer_init(struct net *net)
3032 {
3033 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
3034
3035 if (!bp)
3036 return -ENOMEM;
3037 inet_peer_base_init(bp);
3038 net->ipv4.peers = bp;
3039 return 0;
3040 }
3041
3042 static void __net_exit ipv4_inetpeer_exit(struct net *net)
3043 {
3044 struct inet_peer_base *bp = net->ipv4.peers;
3045
3046 net->ipv4.peers = NULL;
3047 inetpeer_invalidate_tree(bp);
3048 kfree(bp);
3049 }
3050
3051 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = {
3052 .init = ipv4_inetpeer_init,
3053 .exit = ipv4_inetpeer_exit,
3054 };
3055
3056 #ifdef CONFIG_IP_ROUTE_CLASSID
3057 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
3058 #endif /* CONFIG_IP_ROUTE_CLASSID */
3059
3060 int __init ip_rt_init(void)
3061 {
3062 int cpu;
3063
3064 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
3065 if (!ip_idents)
3066 panic("IP: failed to allocate ip_idents\n");
3067
3068 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
3069
3070 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
3071 if (!ip_tstamps)
3072 panic("IP: failed to allocate ip_tstamps\n");
3073
3074 for_each_possible_cpu(cpu) {
3075 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
3076
3077 INIT_LIST_HEAD(&ul->head);
3078 spin_lock_init(&ul->lock);
3079 }
3080 #ifdef CONFIG_IP_ROUTE_CLASSID
3081 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct));
3082 if (!ip_rt_acct)
3083 panic("IP: failed to allocate ip_rt_acct\n");
3084 #endif
3085
3086 ipv4_dst_ops.kmem_cachep =
3087 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0,
3088 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3089
3090 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep;
3091
3092 if (dst_entries_init(&ipv4_dst_ops) < 0)
3093 panic("IP: failed to allocate ipv4_dst_ops counter\n");
3094
3095 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0)
3096 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n");
3097
3098 ipv4_dst_ops.gc_thresh = ~0;
3099 ip_rt_max_size = INT_MAX;
3100
3101 devinet_init();
3102 ip_fib_init();
3103
3104 if (ip_rt_proc_init())
3105 pr_err("Unable to create route proc files\n");
3106 #ifdef CONFIG_XFRM
3107 xfrm_init();
3108 xfrm4_init();
3109 #endif
3110 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL,
3111 RTNL_FLAG_DOIT_UNLOCKED);
3112
3113 #ifdef CONFIG_SYSCTL
3114 register_pernet_subsys(&sysctl_route_ops);
3115 #endif
3116 register_pernet_subsys(&rt_genid_ops);
3117 register_pernet_subsys(&ipv4_inetpeer_ops);
3118 return 0;
3119 }
3120
3121 #ifdef CONFIG_SYSCTL
3122 /*
3123 * We really need to sanitize the damn ipv4 init order, then all
3124 * this nonsense will go away.
3125 */
3126 void __init ip_static_sysctl_init(void)
3127 {
3128 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table);
3129 }
3130 #endif
ubuntu-bionic-kernel mirror