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 <asm/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/bootmem.h> 74 #include <linux/string.h> 75 #include <linux/socket.h> 76 #include <linux/sockios.h> 77 #include <linux/errno.h> 78 #include <linux/in.h> 79 #include <linux/inet.h> 80 #include <linux/netdevice.h> 81 #include <linux/proc_fs.h> 82 #include <linux/init.h> 83 #include <linux/workqueue.h> 84 #include <linux/skbuff.h> 85 #include <linux/inetdevice.h> 86 #include <linux/igmp.h> 87 #include <linux/pkt_sched.h> 88 #include <linux/mroute.h> 89 #include <linux/netfilter_ipv4.h> 90 #include <linux/random.h> 91 #include <linux/jhash.h> 92 #include <linux/rcupdate.h> 93 #include <linux/times.h> 94 #include <linux/slab.h> 95 #include <linux/prefetch.h> 96 #include <net/dst.h> 97 #include <net/net_namespace.h> 98 #include <net/protocol.h> 99 #include <net/ip.h> 100 #include <net/route.h> 101 #include <net/inetpeer.h> 102 #include <net/sock.h> 103 #include <net/ip_fib.h> 104 #include <net/arp.h> 105 #include <net/tcp.h> 106 #include <net/icmp.h> 107 #include <net/xfrm.h> 108 #include <net/netevent.h> 109 #include <net/rtnetlink.h> 110 #ifdef CONFIG_SYSCTL 111 #include <linux/sysctl.h> 112 #include <linux/kmemleak.h> 113 #endif 114 #include <net/secure_seq.h> 115 116 #define RT_FL_TOS(oldflp4) \ 117 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)) 118 119 #define IP_MAX_MTU 0xFFF0 120 121 #define RT_GC_TIMEOUT (300*HZ) 122 123 static int ip_rt_max_size; 124 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT; 125 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 126 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 127 static int ip_rt_redirect_number __read_mostly = 9; 128 static int ip_rt_redirect_load __read_mostly = HZ / 50; 129 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1)); 130 static int ip_rt_error_cost __read_mostly = HZ; 131 static int ip_rt_error_burst __read_mostly = 5 * HZ; 132 static int ip_rt_gc_elasticity __read_mostly = 8; 133 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ; 134 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20; 135 static int ip_rt_min_advmss __read_mostly = 256; 136 137 /* 138 * Interface to generic destination cache. 139 */ 140 141 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie); 142 static unsigned int ipv4_default_advmss(const struct dst_entry *dst); 143 static unsigned int ipv4_mtu(const struct dst_entry *dst); 144 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst); 145 static void ipv4_link_failure(struct sk_buff *skb); 146 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 147 struct sk_buff *skb, u32 mtu); 148 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, 149 struct sk_buff *skb); 150 151 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev, 152 int how) 153 { 154 } 155 156 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old) 157 { 158 WARN_ON(1); 159 return NULL; 160 } 161 162 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 163 struct sk_buff *skb, 164 const void *daddr); 165 166 static struct dst_ops ipv4_dst_ops = { 167 .family = AF_INET, 168 .protocol = cpu_to_be16(ETH_P_IP), 169 .check = ipv4_dst_check, 170 .default_advmss = ipv4_default_advmss, 171 .mtu = ipv4_mtu, 172 .cow_metrics = ipv4_cow_metrics, 173 .ifdown = ipv4_dst_ifdown, 174 .negative_advice = ipv4_negative_advice, 175 .link_failure = ipv4_link_failure, 176 .update_pmtu = ip_rt_update_pmtu, 177 .redirect = ip_do_redirect, 178 .local_out = __ip_local_out, 179 .neigh_lookup = ipv4_neigh_lookup, 180 }; 181 182 #define ECN_OR_COST(class) TC_PRIO_##class 183 184 const __u8 ip_tos2prio[16] = { 185 TC_PRIO_BESTEFFORT, 186 ECN_OR_COST(BESTEFFORT), 187 TC_PRIO_BESTEFFORT, 188 ECN_OR_COST(BESTEFFORT), 189 TC_PRIO_BULK, 190 ECN_OR_COST(BULK), 191 TC_PRIO_BULK, 192 ECN_OR_COST(BULK), 193 TC_PRIO_INTERACTIVE, 194 ECN_OR_COST(INTERACTIVE), 195 TC_PRIO_INTERACTIVE, 196 ECN_OR_COST(INTERACTIVE), 197 TC_PRIO_INTERACTIVE_BULK, 198 ECN_OR_COST(INTERACTIVE_BULK), 199 TC_PRIO_INTERACTIVE_BULK, 200 ECN_OR_COST(INTERACTIVE_BULK) 201 }; 202 EXPORT_SYMBOL(ip_tos2prio); 203 204 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat); 205 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field) 206 207 static inline int rt_genid(struct net *net) 208 { 209 return atomic_read(&net->ipv4.rt_genid); 210 } 211 212 #ifdef CONFIG_PROC_FS 213 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos) 214 { 215 if (*pos) 216 return NULL; 217 return SEQ_START_TOKEN; 218 } 219 220 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos) 221 { 222 ++*pos; 223 return NULL; 224 } 225 226 static void rt_cache_seq_stop(struct seq_file *seq, void *v) 227 { 228 } 229 230 static int rt_cache_seq_show(struct seq_file *seq, void *v) 231 { 232 if (v == SEQ_START_TOKEN) 233 seq_printf(seq, "%-127s\n", 234 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t" 235 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t" 236 "HHUptod\tSpecDst"); 237 return 0; 238 } 239 240 static const struct seq_operations rt_cache_seq_ops = { 241 .start = rt_cache_seq_start, 242 .next = rt_cache_seq_next, 243 .stop = rt_cache_seq_stop, 244 .show = rt_cache_seq_show, 245 }; 246 247 static int rt_cache_seq_open(struct inode *inode, struct file *file) 248 { 249 return seq_open(file, &rt_cache_seq_ops); 250 } 251 252 static const struct file_operations rt_cache_seq_fops = { 253 .owner = THIS_MODULE, 254 .open = rt_cache_seq_open, 255 .read = seq_read, 256 .llseek = seq_lseek, 257 .release = seq_release, 258 }; 259 260 261 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos) 262 { 263 int cpu; 264 265 if (*pos == 0) 266 return SEQ_START_TOKEN; 267 268 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) { 269 if (!cpu_possible(cpu)) 270 continue; 271 *pos = cpu+1; 272 return &per_cpu(rt_cache_stat, cpu); 273 } 274 return NULL; 275 } 276 277 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos) 278 { 279 int cpu; 280 281 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) { 282 if (!cpu_possible(cpu)) 283 continue; 284 *pos = cpu+1; 285 return &per_cpu(rt_cache_stat, cpu); 286 } 287 return NULL; 288 289 } 290 291 static void rt_cpu_seq_stop(struct seq_file *seq, void *v) 292 { 293 294 } 295 296 static int rt_cpu_seq_show(struct seq_file *seq, void *v) 297 { 298 struct rt_cache_stat *st = v; 299 300 if (v == SEQ_START_TOKEN) { 301 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"); 302 return 0; 303 } 304 305 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x " 306 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n", 307 dst_entries_get_slow(&ipv4_dst_ops), 308 st->in_hit, 309 st->in_slow_tot, 310 st->in_slow_mc, 311 st->in_no_route, 312 st->in_brd, 313 st->in_martian_dst, 314 st->in_martian_src, 315 316 st->out_hit, 317 st->out_slow_tot, 318 st->out_slow_mc, 319 320 st->gc_total, 321 st->gc_ignored, 322 st->gc_goal_miss, 323 st->gc_dst_overflow, 324 st->in_hlist_search, 325 st->out_hlist_search 326 ); 327 return 0; 328 } 329 330 static const struct seq_operations rt_cpu_seq_ops = { 331 .start = rt_cpu_seq_start, 332 .next = rt_cpu_seq_next, 333 .stop = rt_cpu_seq_stop, 334 .show = rt_cpu_seq_show, 335 }; 336 337 338 static int rt_cpu_seq_open(struct inode *inode, struct file *file) 339 { 340 return seq_open(file, &rt_cpu_seq_ops); 341 } 342 343 static const struct file_operations rt_cpu_seq_fops = { 344 .owner = THIS_MODULE, 345 .open = rt_cpu_seq_open, 346 .read = seq_read, 347 .llseek = seq_lseek, 348 .release = seq_release, 349 }; 350 351 #ifdef CONFIG_IP_ROUTE_CLASSID 352 static int rt_acct_proc_show(struct seq_file *m, void *v) 353 { 354 struct ip_rt_acct *dst, *src; 355 unsigned int i, j; 356 357 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL); 358 if (!dst) 359 return -ENOMEM; 360 361 for_each_possible_cpu(i) { 362 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i); 363 for (j = 0; j < 256; j++) { 364 dst[j].o_bytes += src[j].o_bytes; 365 dst[j].o_packets += src[j].o_packets; 366 dst[j].i_bytes += src[j].i_bytes; 367 dst[j].i_packets += src[j].i_packets; 368 } 369 } 370 371 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct)); 372 kfree(dst); 373 return 0; 374 } 375 376 static int rt_acct_proc_open(struct inode *inode, struct file *file) 377 { 378 return single_open(file, rt_acct_proc_show, NULL); 379 } 380 381 static const struct file_operations rt_acct_proc_fops = { 382 .owner = THIS_MODULE, 383 .open = rt_acct_proc_open, 384 .read = seq_read, 385 .llseek = seq_lseek, 386 .release = single_release, 387 }; 388 #endif 389 390 static int __net_init ip_rt_do_proc_init(struct net *net) 391 { 392 struct proc_dir_entry *pde; 393 394 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO, 395 &rt_cache_seq_fops); 396 if (!pde) 397 goto err1; 398 399 pde = proc_create("rt_cache", S_IRUGO, 400 net->proc_net_stat, &rt_cpu_seq_fops); 401 if (!pde) 402 goto err2; 403 404 #ifdef CONFIG_IP_ROUTE_CLASSID 405 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops); 406 if (!pde) 407 goto err3; 408 #endif 409 return 0; 410 411 #ifdef CONFIG_IP_ROUTE_CLASSID 412 err3: 413 remove_proc_entry("rt_cache", net->proc_net_stat); 414 #endif 415 err2: 416 remove_proc_entry("rt_cache", net->proc_net); 417 err1: 418 return -ENOMEM; 419 } 420 421 static void __net_exit ip_rt_do_proc_exit(struct net *net) 422 { 423 remove_proc_entry("rt_cache", net->proc_net_stat); 424 remove_proc_entry("rt_cache", net->proc_net); 425 #ifdef CONFIG_IP_ROUTE_CLASSID 426 remove_proc_entry("rt_acct", net->proc_net); 427 #endif 428 } 429 430 static struct pernet_operations ip_rt_proc_ops __net_initdata = { 431 .init = ip_rt_do_proc_init, 432 .exit = ip_rt_do_proc_exit, 433 }; 434 435 static int __init ip_rt_proc_init(void) 436 { 437 return register_pernet_subsys(&ip_rt_proc_ops); 438 } 439 440 #else 441 static inline int ip_rt_proc_init(void) 442 { 443 return 0; 444 } 445 #endif /* CONFIG_PROC_FS */ 446 447 static inline int rt_is_expired(struct rtable *rth) 448 { 449 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev)); 450 } 451 452 /* 453 * Perturbation of rt_genid by a small quantity [1..256] 454 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate() 455 * many times (2^24) without giving recent rt_genid. 456 * Jenkins hash is strong enough that litle changes of rt_genid are OK. 457 */ 458 static void rt_cache_invalidate(struct net *net) 459 { 460 unsigned char shuffle; 461 462 get_random_bytes(&shuffle, sizeof(shuffle)); 463 atomic_add(shuffle + 1U, &net->ipv4.rt_genid); 464 } 465 466 /* 467 * delay < 0 : invalidate cache (fast : entries will be deleted later) 468 * delay >= 0 : invalidate & flush cache (can be long) 469 */ 470 void rt_cache_flush(struct net *net, int delay) 471 { 472 rt_cache_invalidate(net); 473 } 474 475 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 476 struct sk_buff *skb, 477 const void *daddr) 478 { 479 struct net_device *dev = dst->dev; 480 const __be32 *pkey = daddr; 481 const struct rtable *rt; 482 struct neighbour *n; 483 484 rt = (const struct rtable *) dst; 485 if (rt->rt_gateway) 486 pkey = (const __be32 *) &rt->rt_gateway; 487 else if (skb) 488 pkey = &ip_hdr(skb)->daddr; 489 490 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey); 491 if (n) 492 return n; 493 return neigh_create(&arp_tbl, pkey, dev); 494 } 495 496 /* 497 * Peer allocation may fail only in serious out-of-memory conditions. However 498 * we still can generate some output. 499 * Random ID selection looks a bit dangerous because we have no chances to 500 * select ID being unique in a reasonable period of time. 501 * But broken packet identifier may be better than no packet at all. 502 */ 503 static void ip_select_fb_ident(struct iphdr *iph) 504 { 505 static DEFINE_SPINLOCK(ip_fb_id_lock); 506 static u32 ip_fallback_id; 507 u32 salt; 508 509 spin_lock_bh(&ip_fb_id_lock); 510 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr); 511 iph->id = htons(salt & 0xFFFF); 512 ip_fallback_id = salt; 513 spin_unlock_bh(&ip_fb_id_lock); 514 } 515 516 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more) 517 { 518 struct net *net = dev_net(dst->dev); 519 struct inet_peer *peer; 520 521 peer = inet_getpeer_v4(net->ipv4.peers, iph->daddr, 1); 522 if (peer) { 523 iph->id = htons(inet_getid(peer, more)); 524 inet_putpeer(peer); 525 return; 526 } 527 528 ip_select_fb_ident(iph); 529 } 530 EXPORT_SYMBOL(__ip_select_ident); 531 532 static void __build_flow_key(struct flowi4 *fl4, const struct sock *sk, 533 const struct iphdr *iph, 534 int oif, u8 tos, 535 u8 prot, u32 mark, int flow_flags) 536 { 537 if (sk) { 538 const struct inet_sock *inet = inet_sk(sk); 539 540 oif = sk->sk_bound_dev_if; 541 mark = sk->sk_mark; 542 tos = RT_CONN_FLAGS(sk); 543 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol; 544 } 545 flowi4_init_output(fl4, oif, mark, tos, 546 RT_SCOPE_UNIVERSE, prot, 547 flow_flags, 548 iph->daddr, iph->saddr, 0, 0); 549 } 550 551 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb, 552 const struct sock *sk) 553 { 554 const struct iphdr *iph = ip_hdr(skb); 555 int oif = skb->dev->ifindex; 556 u8 tos = RT_TOS(iph->tos); 557 u8 prot = iph->protocol; 558 u32 mark = skb->mark; 559 560 __build_flow_key(fl4, sk, iph, oif, tos, prot, mark, 0); 561 } 562 563 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk) 564 { 565 const struct inet_sock *inet = inet_sk(sk); 566 const struct ip_options_rcu *inet_opt; 567 __be32 daddr = inet->inet_daddr; 568 569 rcu_read_lock(); 570 inet_opt = rcu_dereference(inet->inet_opt); 571 if (inet_opt && inet_opt->opt.srr) 572 daddr = inet_opt->opt.faddr; 573 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark, 574 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 575 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol, 576 inet_sk_flowi_flags(sk), 577 daddr, inet->inet_saddr, 0, 0); 578 rcu_read_unlock(); 579 } 580 581 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk, 582 const struct sk_buff *skb) 583 { 584 if (skb) 585 build_skb_flow_key(fl4, skb, sk); 586 else 587 build_sk_flow_key(fl4, sk); 588 } 589 590 static DEFINE_SEQLOCK(fnhe_seqlock); 591 592 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash) 593 { 594 struct fib_nh_exception *fnhe, *oldest; 595 596 oldest = rcu_dereference(hash->chain); 597 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe; 598 fnhe = rcu_dereference(fnhe->fnhe_next)) { 599 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) 600 oldest = fnhe; 601 } 602 return oldest; 603 } 604 605 static inline u32 fnhe_hashfun(__be32 daddr) 606 { 607 u32 hval; 608 609 hval = (__force u32) daddr; 610 hval ^= (hval >> 11) ^ (hval >> 22); 611 612 return hval & (FNHE_HASH_SIZE - 1); 613 } 614 615 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw, 616 u32 pmtu, unsigned long expires) 617 { 618 struct fnhe_hash_bucket *hash; 619 struct fib_nh_exception *fnhe; 620 int depth; 621 u32 hval = fnhe_hashfun(daddr); 622 623 write_seqlock_bh(&fnhe_seqlock); 624 625 hash = nh->nh_exceptions; 626 if (!hash) { 627 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC); 628 if (!hash) 629 goto out_unlock; 630 nh->nh_exceptions = hash; 631 } 632 633 hash += hval; 634 635 depth = 0; 636 for (fnhe = rcu_dereference(hash->chain); fnhe; 637 fnhe = rcu_dereference(fnhe->fnhe_next)) { 638 if (fnhe->fnhe_daddr == daddr) 639 break; 640 depth++; 641 } 642 643 if (fnhe) { 644 if (gw) 645 fnhe->fnhe_gw = gw; 646 if (pmtu) { 647 fnhe->fnhe_pmtu = pmtu; 648 fnhe->fnhe_expires = expires; 649 } 650 } else { 651 if (depth > FNHE_RECLAIM_DEPTH) 652 fnhe = fnhe_oldest(hash); 653 else { 654 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC); 655 if (!fnhe) 656 goto out_unlock; 657 658 fnhe->fnhe_next = hash->chain; 659 rcu_assign_pointer(hash->chain, fnhe); 660 } 661 fnhe->fnhe_daddr = daddr; 662 fnhe->fnhe_gw = gw; 663 fnhe->fnhe_pmtu = pmtu; 664 fnhe->fnhe_expires = expires; 665 } 666 667 fnhe->fnhe_stamp = jiffies; 668 669 out_unlock: 670 write_sequnlock_bh(&fnhe_seqlock); 671 return; 672 } 673 674 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4, 675 bool kill_route) 676 { 677 __be32 new_gw = icmp_hdr(skb)->un.gateway; 678 __be32 old_gw = ip_hdr(skb)->saddr; 679 struct net_device *dev = skb->dev; 680 struct in_device *in_dev; 681 struct fib_result res; 682 struct neighbour *n; 683 struct net *net; 684 685 switch (icmp_hdr(skb)->code & 7) { 686 case ICMP_REDIR_NET: 687 case ICMP_REDIR_NETTOS: 688 case ICMP_REDIR_HOST: 689 case ICMP_REDIR_HOSTTOS: 690 break; 691 692 default: 693 return; 694 } 695 696 if (rt->rt_gateway != old_gw) 697 return; 698 699 in_dev = __in_dev_get_rcu(dev); 700 if (!in_dev) 701 return; 702 703 net = dev_net(dev); 704 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) || 705 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) || 706 ipv4_is_zeronet(new_gw)) 707 goto reject_redirect; 708 709 if (!IN_DEV_SHARED_MEDIA(in_dev)) { 710 if (!inet_addr_onlink(in_dev, new_gw, old_gw)) 711 goto reject_redirect; 712 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) 713 goto reject_redirect; 714 } else { 715 if (inet_addr_type(net, new_gw) != RTN_UNICAST) 716 goto reject_redirect; 717 } 718 719 n = ipv4_neigh_lookup(&rt->dst, NULL, &new_gw); 720 if (n) { 721 if (!(n->nud_state & NUD_VALID)) { 722 neigh_event_send(n, NULL); 723 } else { 724 if (fib_lookup(net, fl4, &res) == 0) { 725 struct fib_nh *nh = &FIB_RES_NH(res); 726 727 update_or_create_fnhe(nh, fl4->daddr, new_gw, 728 0, 0); 729 } 730 if (kill_route) 731 rt->dst.obsolete = DST_OBSOLETE_KILL; 732 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n); 733 } 734 neigh_release(n); 735 } 736 return; 737 738 reject_redirect: 739 #ifdef CONFIG_IP_ROUTE_VERBOSE 740 if (IN_DEV_LOG_MARTIANS(in_dev)) { 741 const struct iphdr *iph = (const struct iphdr *) skb->data; 742 __be32 daddr = iph->daddr; 743 __be32 saddr = iph->saddr; 744 745 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n" 746 " Advised path = %pI4 -> %pI4\n", 747 &old_gw, dev->name, &new_gw, 748 &saddr, &daddr); 749 } 750 #endif 751 ; 752 } 753 754 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb) 755 { 756 struct rtable *rt; 757 struct flowi4 fl4; 758 759 rt = (struct rtable *) dst; 760 761 ip_rt_build_flow_key(&fl4, sk, skb); 762 __ip_do_redirect(rt, skb, &fl4, true); 763 } 764 765 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst) 766 { 767 struct rtable *rt = (struct rtable *)dst; 768 struct dst_entry *ret = dst; 769 770 if (rt) { 771 if (dst->obsolete > 0) { 772 ip_rt_put(rt); 773 ret = NULL; 774 } else if ((rt->rt_flags & RTCF_REDIRECTED) || 775 rt->dst.expires) { 776 ip_rt_put(rt); 777 ret = NULL; 778 } 779 } 780 return ret; 781 } 782 783 /* 784 * Algorithm: 785 * 1. The first ip_rt_redirect_number redirects are sent 786 * with exponential backoff, then we stop sending them at all, 787 * assuming that the host ignores our redirects. 788 * 2. If we did not see packets requiring redirects 789 * during ip_rt_redirect_silence, we assume that the host 790 * forgot redirected route and start to send redirects again. 791 * 792 * This algorithm is much cheaper and more intelligent than dumb load limiting 793 * in icmp.c. 794 * 795 * NOTE. Do not forget to inhibit load limiting for redirects (redundant) 796 * and "frag. need" (breaks PMTU discovery) in icmp.c. 797 */ 798 799 void ip_rt_send_redirect(struct sk_buff *skb) 800 { 801 struct rtable *rt = skb_rtable(skb); 802 struct in_device *in_dev; 803 struct inet_peer *peer; 804 struct net *net; 805 int log_martians; 806 807 rcu_read_lock(); 808 in_dev = __in_dev_get_rcu(rt->dst.dev); 809 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) { 810 rcu_read_unlock(); 811 return; 812 } 813 log_martians = IN_DEV_LOG_MARTIANS(in_dev); 814 rcu_read_unlock(); 815 816 net = dev_net(rt->dst.dev); 817 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1); 818 if (!peer) { 819 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway); 820 return; 821 } 822 823 /* No redirected packets during ip_rt_redirect_silence; 824 * reset the algorithm. 825 */ 826 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) 827 peer->rate_tokens = 0; 828 829 /* Too many ignored redirects; do not send anything 830 * set dst.rate_last to the last seen redirected packet. 831 */ 832 if (peer->rate_tokens >= ip_rt_redirect_number) { 833 peer->rate_last = jiffies; 834 goto out_put_peer; 835 } 836 837 /* Check for load limit; set rate_last to the latest sent 838 * redirect. 839 */ 840 if (peer->rate_tokens == 0 || 841 time_after(jiffies, 842 (peer->rate_last + 843 (ip_rt_redirect_load << peer->rate_tokens)))) { 844 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway); 845 peer->rate_last = jiffies; 846 ++peer->rate_tokens; 847 #ifdef CONFIG_IP_ROUTE_VERBOSE 848 if (log_martians && 849 peer->rate_tokens == ip_rt_redirect_number) 850 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n", 851 &ip_hdr(skb)->saddr, inet_iif(skb), 852 &ip_hdr(skb)->daddr, &rt->rt_gateway); 853 #endif 854 } 855 out_put_peer: 856 inet_putpeer(peer); 857 } 858 859 static int ip_error(struct sk_buff *skb) 860 { 861 struct in_device *in_dev = __in_dev_get_rcu(skb->dev); 862 struct rtable *rt = skb_rtable(skb); 863 struct inet_peer *peer; 864 unsigned long now; 865 struct net *net; 866 bool send; 867 int code; 868 869 net = dev_net(rt->dst.dev); 870 if (!IN_DEV_FORWARD(in_dev)) { 871 switch (rt->dst.error) { 872 case EHOSTUNREACH: 873 IP_INC_STATS_BH(net, IPSTATS_MIB_INADDRERRORS); 874 break; 875 876 case ENETUNREACH: 877 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES); 878 break; 879 } 880 goto out; 881 } 882 883 switch (rt->dst.error) { 884 case EINVAL: 885 default: 886 goto out; 887 case EHOSTUNREACH: 888 code = ICMP_HOST_UNREACH; 889 break; 890 case ENETUNREACH: 891 code = ICMP_NET_UNREACH; 892 IP_INC_STATS_BH(net, IPSTATS_MIB_INNOROUTES); 893 break; 894 case EACCES: 895 code = ICMP_PKT_FILTERED; 896 break; 897 } 898 899 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1); 900 901 send = true; 902 if (peer) { 903 now = jiffies; 904 peer->rate_tokens += now - peer->rate_last; 905 if (peer->rate_tokens > ip_rt_error_burst) 906 peer->rate_tokens = ip_rt_error_burst; 907 peer->rate_last = now; 908 if (peer->rate_tokens >= ip_rt_error_cost) 909 peer->rate_tokens -= ip_rt_error_cost; 910 else 911 send = false; 912 inet_putpeer(peer); 913 } 914 if (send) 915 icmp_send(skb, ICMP_DEST_UNREACH, code, 0); 916 917 out: kfree_skb(skb); 918 return 0; 919 } 920 921 static u32 __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu) 922 { 923 struct fib_result res; 924 925 if (mtu < ip_rt_min_pmtu) 926 mtu = ip_rt_min_pmtu; 927 928 if (fib_lookup(dev_net(rt->dst.dev), fl4, &res) == 0) { 929 struct fib_nh *nh = &FIB_RES_NH(res); 930 931 update_or_create_fnhe(nh, fl4->daddr, 0, mtu, 932 jiffies + ip_rt_mtu_expires); 933 } 934 return mtu; 935 } 936 937 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 938 struct sk_buff *skb, u32 mtu) 939 { 940 struct rtable *rt = (struct rtable *) dst; 941 struct flowi4 fl4; 942 943 ip_rt_build_flow_key(&fl4, sk, skb); 944 mtu = __ip_rt_update_pmtu(rt, &fl4, mtu); 945 946 if (!rt->rt_pmtu) { 947 dst->obsolete = DST_OBSOLETE_KILL; 948 } else { 949 rt->rt_pmtu = mtu; 950 dst_set_expires(&rt->dst, ip_rt_mtu_expires); 951 } 952 } 953 954 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, 955 int oif, u32 mark, u8 protocol, int flow_flags) 956 { 957 const struct iphdr *iph = (const struct iphdr *) skb->data; 958 struct flowi4 fl4; 959 struct rtable *rt; 960 961 __build_flow_key(&fl4, NULL, iph, oif, 962 RT_TOS(iph->tos), protocol, mark, flow_flags); 963 rt = __ip_route_output_key(net, &fl4); 964 if (!IS_ERR(rt)) { 965 __ip_rt_update_pmtu(rt, &fl4, mtu); 966 ip_rt_put(rt); 967 } 968 } 969 EXPORT_SYMBOL_GPL(ipv4_update_pmtu); 970 971 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 972 { 973 const struct iphdr *iph = (const struct iphdr *) skb->data; 974 struct flowi4 fl4; 975 struct rtable *rt; 976 977 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0); 978 rt = __ip_route_output_key(sock_net(sk), &fl4); 979 if (!IS_ERR(rt)) { 980 __ip_rt_update_pmtu(rt, &fl4, mtu); 981 ip_rt_put(rt); 982 } 983 } 984 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu); 985 986 void ipv4_redirect(struct sk_buff *skb, struct net *net, 987 int oif, u32 mark, u8 protocol, int flow_flags) 988 { 989 const struct iphdr *iph = (const struct iphdr *) skb->data; 990 struct flowi4 fl4; 991 struct rtable *rt; 992 993 __build_flow_key(&fl4, NULL, iph, oif, 994 RT_TOS(iph->tos), protocol, mark, flow_flags); 995 rt = __ip_route_output_key(net, &fl4); 996 if (!IS_ERR(rt)) { 997 __ip_do_redirect(rt, skb, &fl4, false); 998 ip_rt_put(rt); 999 } 1000 } 1001 EXPORT_SYMBOL_GPL(ipv4_redirect); 1002 1003 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk) 1004 { 1005 const struct iphdr *iph = (const struct iphdr *) skb->data; 1006 struct flowi4 fl4; 1007 struct rtable *rt; 1008 1009 __build_flow_key(&fl4, sk, iph, 0, 0, 0, 0, 0); 1010 rt = __ip_route_output_key(sock_net(sk), &fl4); 1011 if (!IS_ERR(rt)) { 1012 __ip_do_redirect(rt, skb, &fl4, false); 1013 ip_rt_put(rt); 1014 } 1015 } 1016 EXPORT_SYMBOL_GPL(ipv4_sk_redirect); 1017 1018 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie) 1019 { 1020 struct rtable *rt = (struct rtable *) dst; 1021 1022 /* All IPV4 dsts are created with ->obsolete set to the value 1023 * DST_OBSOLETE_FORCE_CHK which forces validation calls down 1024 * into this function always. 1025 * 1026 * When a PMTU/redirect information update invalidates a 1027 * route, this is indicated by setting obsolete to 1028 * DST_OBSOLETE_KILL. 1029 */ 1030 if (dst->obsolete == DST_OBSOLETE_KILL || rt_is_expired(rt)) 1031 return NULL; 1032 return dst; 1033 } 1034 1035 static void ipv4_link_failure(struct sk_buff *skb) 1036 { 1037 struct rtable *rt; 1038 1039 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0); 1040 1041 rt = skb_rtable(skb); 1042 if (rt) 1043 dst_set_expires(&rt->dst, 0); 1044 } 1045 1046 static int ip_rt_bug(struct sk_buff *skb) 1047 { 1048 pr_debug("%s: %pI4 -> %pI4, %s\n", 1049 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1050 skb->dev ? skb->dev->name : "?"); 1051 kfree_skb(skb); 1052 WARN_ON(1); 1053 return 0; 1054 } 1055 1056 /* 1057 We do not cache source address of outgoing interface, 1058 because it is used only by IP RR, TS and SRR options, 1059 so that it out of fast path. 1060 1061 BTW remember: "addr" is allowed to be not aligned 1062 in IP options! 1063 */ 1064 1065 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt) 1066 { 1067 __be32 src; 1068 1069 if (rt_is_output_route(rt)) 1070 src = ip_hdr(skb)->saddr; 1071 else { 1072 struct fib_result res; 1073 struct flowi4 fl4; 1074 struct iphdr *iph; 1075 1076 iph = ip_hdr(skb); 1077 1078 memset(&fl4, 0, sizeof(fl4)); 1079 fl4.daddr = iph->daddr; 1080 fl4.saddr = iph->saddr; 1081 fl4.flowi4_tos = RT_TOS(iph->tos); 1082 fl4.flowi4_oif = rt->dst.dev->ifindex; 1083 fl4.flowi4_iif = skb->dev->ifindex; 1084 fl4.flowi4_mark = skb->mark; 1085 1086 rcu_read_lock(); 1087 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0) 1088 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res); 1089 else 1090 src = inet_select_addr(rt->dst.dev, 1091 rt_nexthop(rt, iph->daddr), 1092 RT_SCOPE_UNIVERSE); 1093 rcu_read_unlock(); 1094 } 1095 memcpy(addr, &src, 4); 1096 } 1097 1098 #ifdef CONFIG_IP_ROUTE_CLASSID 1099 static void set_class_tag(struct rtable *rt, u32 tag) 1100 { 1101 if (!(rt->dst.tclassid & 0xFFFF)) 1102 rt->dst.tclassid |= tag & 0xFFFF; 1103 if (!(rt->dst.tclassid & 0xFFFF0000)) 1104 rt->dst.tclassid |= tag & 0xFFFF0000; 1105 } 1106 #endif 1107 1108 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1109 { 1110 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS); 1111 1112 if (advmss == 0) { 1113 advmss = max_t(unsigned int, dst->dev->mtu - 40, 1114 ip_rt_min_advmss); 1115 if (advmss > 65535 - 40) 1116 advmss = 65535 - 40; 1117 } 1118 return advmss; 1119 } 1120 1121 static unsigned int ipv4_mtu(const struct dst_entry *dst) 1122 { 1123 const struct rtable *rt = (const struct rtable *) dst; 1124 unsigned int mtu = rt->rt_pmtu; 1125 1126 if (mtu && time_after_eq(jiffies, rt->dst.expires)) 1127 mtu = 0; 1128 1129 if (!mtu) 1130 mtu = dst_metric_raw(dst, RTAX_MTU); 1131 1132 if (mtu && rt_is_output_route(rt)) 1133 return mtu; 1134 1135 mtu = dst->dev->mtu; 1136 1137 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) { 1138 if (rt->rt_gateway && mtu > 576) 1139 mtu = 576; 1140 } 1141 1142 if (mtu > IP_MAX_MTU) 1143 mtu = IP_MAX_MTU; 1144 1145 return mtu; 1146 } 1147 1148 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr) 1149 { 1150 struct fnhe_hash_bucket *hash = nh->nh_exceptions; 1151 struct fib_nh_exception *fnhe; 1152 u32 hval; 1153 1154 if (!hash) 1155 return NULL; 1156 1157 hval = fnhe_hashfun(daddr); 1158 1159 for (fnhe = rcu_dereference(hash[hval].chain); fnhe; 1160 fnhe = rcu_dereference(fnhe->fnhe_next)) { 1161 if (fnhe->fnhe_daddr == daddr) 1162 return fnhe; 1163 } 1164 return NULL; 1165 } 1166 1167 static void rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, 1168 __be32 daddr) 1169 { 1170 __be32 fnhe_daddr, gw; 1171 unsigned long expires; 1172 unsigned int seq; 1173 u32 pmtu; 1174 1175 restart: 1176 seq = read_seqbegin(&fnhe_seqlock); 1177 fnhe_daddr = fnhe->fnhe_daddr; 1178 gw = fnhe->fnhe_gw; 1179 pmtu = fnhe->fnhe_pmtu; 1180 expires = fnhe->fnhe_expires; 1181 if (read_seqretry(&fnhe_seqlock, seq)) 1182 goto restart; 1183 1184 if (daddr != fnhe_daddr) 1185 return; 1186 1187 if (pmtu) { 1188 unsigned long diff = expires - jiffies; 1189 1190 if (time_before(jiffies, expires)) { 1191 rt->rt_pmtu = pmtu; 1192 dst_set_expires(&rt->dst, diff); 1193 } 1194 } 1195 if (gw) { 1196 rt->rt_flags |= RTCF_REDIRECTED; 1197 rt->rt_gateway = gw; 1198 } 1199 fnhe->fnhe_stamp = jiffies; 1200 } 1201 1202 static inline void rt_release_rcu(struct rcu_head *head) 1203 { 1204 struct dst_entry *dst = container_of(head, struct dst_entry, rcu_head); 1205 dst_release(dst); 1206 } 1207 1208 static void rt_cache_route(struct fib_nh *nh, struct rtable *rt) 1209 { 1210 struct rtable *orig, *prev, **p = &nh->nh_rth_output; 1211 1212 if (rt_is_input_route(rt)) 1213 p = &nh->nh_rth_input; 1214 1215 orig = *p; 1216 1217 prev = cmpxchg(p, orig, rt); 1218 if (prev == orig) { 1219 dst_clone(&rt->dst); 1220 if (orig) 1221 call_rcu_bh(&orig->dst.rcu_head, rt_release_rcu); 1222 } 1223 } 1224 1225 static bool rt_cache_valid(struct rtable *rt) 1226 { 1227 return (rt && rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK); 1228 } 1229 1230 static void rt_set_nexthop(struct rtable *rt, __be32 daddr, 1231 const struct fib_result *res, 1232 struct fib_nh_exception *fnhe, 1233 struct fib_info *fi, u16 type, u32 itag) 1234 { 1235 if (fi) { 1236 struct fib_nh *nh = &FIB_RES_NH(*res); 1237 1238 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) 1239 rt->rt_gateway = nh->nh_gw; 1240 if (unlikely(fnhe)) 1241 rt_bind_exception(rt, fnhe, daddr); 1242 dst_init_metrics(&rt->dst, fi->fib_metrics, true); 1243 #ifdef CONFIG_IP_ROUTE_CLASSID 1244 rt->dst.tclassid = nh->nh_tclassid; 1245 #endif 1246 if (!(rt->dst.flags & DST_HOST)) 1247 rt_cache_route(nh, rt); 1248 } 1249 1250 #ifdef CONFIG_IP_ROUTE_CLASSID 1251 #ifdef CONFIG_IP_MULTIPLE_TABLES 1252 set_class_tag(rt, res->tclassid); 1253 #endif 1254 set_class_tag(rt, itag); 1255 #endif 1256 } 1257 1258 static struct rtable *rt_dst_alloc(struct net_device *dev, 1259 bool nopolicy, bool noxfrm, bool will_cache) 1260 { 1261 return dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1262 (will_cache ? 0 : DST_HOST) | DST_NOCACHE | 1263 (nopolicy ? DST_NOPOLICY : 0) | 1264 (noxfrm ? DST_NOXFRM : 0)); 1265 } 1266 1267 /* called in rcu_read_lock() section */ 1268 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1269 u8 tos, struct net_device *dev, int our) 1270 { 1271 struct rtable *rth; 1272 struct in_device *in_dev = __in_dev_get_rcu(dev); 1273 u32 itag = 0; 1274 int err; 1275 1276 /* Primary sanity checks. */ 1277 1278 if (in_dev == NULL) 1279 return -EINVAL; 1280 1281 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) || 1282 skb->protocol != htons(ETH_P_IP)) 1283 goto e_inval; 1284 1285 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 1286 if (ipv4_is_loopback(saddr)) 1287 goto e_inval; 1288 1289 if (ipv4_is_zeronet(saddr)) { 1290 if (!ipv4_is_local_multicast(daddr)) 1291 goto e_inval; 1292 } else { 1293 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 1294 in_dev, &itag); 1295 if (err < 0) 1296 goto e_err; 1297 } 1298 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, 1299 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false); 1300 if (!rth) 1301 goto e_nobufs; 1302 1303 #ifdef CONFIG_IP_ROUTE_CLASSID 1304 rth->dst.tclassid = itag; 1305 #endif 1306 rth->dst.output = ip_rt_bug; 1307 1308 rth->rt_genid = rt_genid(dev_net(dev)); 1309 rth->rt_flags = RTCF_MULTICAST; 1310 rth->rt_type = RTN_MULTICAST; 1311 rth->rt_is_input= 1; 1312 rth->rt_iif = 0; 1313 rth->rt_pmtu = 0; 1314 rth->rt_gateway = 0; 1315 if (our) { 1316 rth->dst.input= ip_local_deliver; 1317 rth->rt_flags |= RTCF_LOCAL; 1318 } 1319 1320 #ifdef CONFIG_IP_MROUTE 1321 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1322 rth->dst.input = ip_mr_input; 1323 #endif 1324 RT_CACHE_STAT_INC(in_slow_mc); 1325 1326 skb_dst_set(skb, &rth->dst); 1327 return 0; 1328 1329 e_nobufs: 1330 return -ENOBUFS; 1331 e_inval: 1332 return -EINVAL; 1333 e_err: 1334 return err; 1335 } 1336 1337 1338 static void ip_handle_martian_source(struct net_device *dev, 1339 struct in_device *in_dev, 1340 struct sk_buff *skb, 1341 __be32 daddr, 1342 __be32 saddr) 1343 { 1344 RT_CACHE_STAT_INC(in_martian_src); 1345 #ifdef CONFIG_IP_ROUTE_VERBOSE 1346 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1347 /* 1348 * RFC1812 recommendation, if source is martian, 1349 * the only hint is MAC header. 1350 */ 1351 pr_warn("martian source %pI4 from %pI4, on dev %s\n", 1352 &daddr, &saddr, dev->name); 1353 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1354 print_hex_dump(KERN_WARNING, "ll header: ", 1355 DUMP_PREFIX_OFFSET, 16, 1, 1356 skb_mac_header(skb), 1357 dev->hard_header_len, true); 1358 } 1359 } 1360 #endif 1361 } 1362 1363 /* called in rcu_read_lock() section */ 1364 static int __mkroute_input(struct sk_buff *skb, 1365 const struct fib_result *res, 1366 struct in_device *in_dev, 1367 __be32 daddr, __be32 saddr, u32 tos, 1368 struct rtable **result) 1369 { 1370 struct rtable *rth; 1371 int err; 1372 struct in_device *out_dev; 1373 unsigned int flags = 0; 1374 bool do_cache; 1375 u32 itag; 1376 1377 /* get a working reference to the output device */ 1378 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res)); 1379 if (out_dev == NULL) { 1380 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n"); 1381 return -EINVAL; 1382 } 1383 1384 1385 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res), 1386 in_dev->dev, in_dev, &itag); 1387 if (err < 0) { 1388 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1389 saddr); 1390 1391 goto cleanup; 1392 } 1393 1394 if (out_dev == in_dev && err && 1395 (IN_DEV_SHARED_MEDIA(out_dev) || 1396 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) 1397 flags |= RTCF_DOREDIRECT; 1398 1399 if (skb->protocol != htons(ETH_P_IP)) { 1400 /* Not IP (i.e. ARP). Do not create route, if it is 1401 * invalid for proxy arp. DNAT routes are always valid. 1402 * 1403 * Proxy arp feature have been extended to allow, ARP 1404 * replies back to the same interface, to support 1405 * Private VLAN switch technologies. See arp.c. 1406 */ 1407 if (out_dev == in_dev && 1408 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 1409 err = -EINVAL; 1410 goto cleanup; 1411 } 1412 } 1413 1414 do_cache = false; 1415 if (res->fi) { 1416 if (!itag) { 1417 rth = FIB_RES_NH(*res).nh_rth_input; 1418 if (rt_cache_valid(rth)) { 1419 dst_hold(&rth->dst); 1420 goto out; 1421 } 1422 do_cache = true; 1423 } 1424 } 1425 1426 rth = rt_dst_alloc(out_dev->dev, 1427 IN_DEV_CONF_GET(in_dev, NOPOLICY), 1428 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache); 1429 if (!rth) { 1430 err = -ENOBUFS; 1431 goto cleanup; 1432 } 1433 1434 rth->rt_genid = rt_genid(dev_net(rth->dst.dev)); 1435 rth->rt_flags = flags; 1436 rth->rt_type = res->type; 1437 rth->rt_is_input = 1; 1438 rth->rt_iif = 0; 1439 rth->rt_pmtu = 0; 1440 rth->rt_gateway = 0; 1441 1442 rth->dst.input = ip_forward; 1443 rth->dst.output = ip_output; 1444 1445 rt_set_nexthop(rth, daddr, res, NULL, res->fi, res->type, itag); 1446 out: 1447 *result = rth; 1448 err = 0; 1449 cleanup: 1450 return err; 1451 } 1452 1453 static int ip_mkroute_input(struct sk_buff *skb, 1454 struct fib_result *res, 1455 const struct flowi4 *fl4, 1456 struct in_device *in_dev, 1457 __be32 daddr, __be32 saddr, u32 tos) 1458 { 1459 struct rtable *rth = NULL; 1460 int err; 1461 1462 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1463 if (res->fi && res->fi->fib_nhs > 1) 1464 fib_select_multipath(res); 1465 #endif 1466 1467 /* create a routing cache entry */ 1468 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth); 1469 if (err) 1470 return err; 1471 1472 skb_dst_set(skb, &rth->dst); 1473 return 0; 1474 } 1475 1476 /* 1477 * NOTE. We drop all the packets that has local source 1478 * addresses, because every properly looped back packet 1479 * must have correct destination already attached by output routine. 1480 * 1481 * Such approach solves two big problems: 1482 * 1. Not simplex devices are handled properly. 1483 * 2. IP spoofing attempts are filtered with 100% of guarantee. 1484 * called with rcu_read_lock() 1485 */ 1486 1487 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1488 u8 tos, struct net_device *dev) 1489 { 1490 struct fib_result res; 1491 struct in_device *in_dev = __in_dev_get_rcu(dev); 1492 struct flowi4 fl4; 1493 unsigned int flags = 0; 1494 u32 itag = 0; 1495 struct rtable *rth; 1496 int err = -EINVAL; 1497 struct net *net = dev_net(dev); 1498 bool do_cache; 1499 1500 /* IP on this device is disabled. */ 1501 1502 if (!in_dev) 1503 goto out; 1504 1505 /* Check for the most weird martians, which can be not detected 1506 by fib_lookup. 1507 */ 1508 1509 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 1510 goto martian_source; 1511 1512 res.fi = NULL; 1513 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 1514 goto brd_input; 1515 1516 /* Accept zero addresses only to limited broadcast; 1517 * I even do not know to fix it or not. Waiting for complains :-) 1518 */ 1519 if (ipv4_is_zeronet(saddr)) 1520 goto martian_source; 1521 1522 if (ipv4_is_zeronet(daddr)) 1523 goto martian_destination; 1524 1525 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) { 1526 if (ipv4_is_loopback(daddr)) 1527 goto martian_destination; 1528 1529 if (ipv4_is_loopback(saddr)) 1530 goto martian_source; 1531 } 1532 1533 /* 1534 * Now we are ready to route packet. 1535 */ 1536 fl4.flowi4_oif = 0; 1537 fl4.flowi4_iif = dev->ifindex; 1538 fl4.flowi4_mark = skb->mark; 1539 fl4.flowi4_tos = tos; 1540 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 1541 fl4.daddr = daddr; 1542 fl4.saddr = saddr; 1543 err = fib_lookup(net, &fl4, &res); 1544 if (err != 0) 1545 goto no_route; 1546 1547 RT_CACHE_STAT_INC(in_slow_tot); 1548 1549 if (res.type == RTN_BROADCAST) 1550 goto brd_input; 1551 1552 if (res.type == RTN_LOCAL) { 1553 err = fib_validate_source(skb, saddr, daddr, tos, 1554 net->loopback_dev->ifindex, 1555 dev, in_dev, &itag); 1556 if (err < 0) 1557 goto martian_source_keep_err; 1558 goto local_input; 1559 } 1560 1561 if (!IN_DEV_FORWARD(in_dev)) 1562 goto no_route; 1563 if (res.type != RTN_UNICAST) 1564 goto martian_destination; 1565 1566 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos); 1567 out: return err; 1568 1569 brd_input: 1570 if (skb->protocol != htons(ETH_P_IP)) 1571 goto e_inval; 1572 1573 if (!ipv4_is_zeronet(saddr)) { 1574 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 1575 in_dev, &itag); 1576 if (err < 0) 1577 goto martian_source_keep_err; 1578 } 1579 flags |= RTCF_BROADCAST; 1580 res.type = RTN_BROADCAST; 1581 RT_CACHE_STAT_INC(in_brd); 1582 1583 local_input: 1584 do_cache = false; 1585 if (res.fi) { 1586 if (!itag) { 1587 rth = FIB_RES_NH(res).nh_rth_input; 1588 if (rt_cache_valid(rth)) { 1589 dst_hold(&rth->dst); 1590 goto set_and_out; 1591 } 1592 do_cache = true; 1593 } 1594 } 1595 1596 rth = rt_dst_alloc(net->loopback_dev, 1597 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache); 1598 if (!rth) 1599 goto e_nobufs; 1600 1601 rth->dst.input= ip_local_deliver; 1602 rth->dst.output= ip_rt_bug; 1603 #ifdef CONFIG_IP_ROUTE_CLASSID 1604 rth->dst.tclassid = itag; 1605 #endif 1606 1607 rth->rt_genid = rt_genid(net); 1608 rth->rt_flags = flags|RTCF_LOCAL; 1609 rth->rt_type = res.type; 1610 rth->rt_is_input = 1; 1611 rth->rt_iif = 0; 1612 rth->rt_pmtu = 0; 1613 rth->rt_gateway = 0; 1614 if (res.type == RTN_UNREACHABLE) { 1615 rth->dst.input= ip_error; 1616 rth->dst.error= -err; 1617 rth->rt_flags &= ~RTCF_LOCAL; 1618 } 1619 if (do_cache) 1620 rt_cache_route(&FIB_RES_NH(res), rth); 1621 set_and_out: 1622 skb_dst_set(skb, &rth->dst); 1623 err = 0; 1624 goto out; 1625 1626 no_route: 1627 RT_CACHE_STAT_INC(in_no_route); 1628 res.type = RTN_UNREACHABLE; 1629 if (err == -ESRCH) 1630 err = -ENETUNREACH; 1631 goto local_input; 1632 1633 /* 1634 * Do not cache martian addresses: they should be logged (RFC1812) 1635 */ 1636 martian_destination: 1637 RT_CACHE_STAT_INC(in_martian_dst); 1638 #ifdef CONFIG_IP_ROUTE_VERBOSE 1639 if (IN_DEV_LOG_MARTIANS(in_dev)) 1640 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n", 1641 &daddr, &saddr, dev->name); 1642 #endif 1643 1644 e_inval: 1645 err = -EINVAL; 1646 goto out; 1647 1648 e_nobufs: 1649 err = -ENOBUFS; 1650 goto out; 1651 1652 martian_source: 1653 err = -EINVAL; 1654 martian_source_keep_err: 1655 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 1656 goto out; 1657 } 1658 1659 int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1660 u8 tos, struct net_device *dev) 1661 { 1662 int res; 1663 1664 rcu_read_lock(); 1665 1666 /* Multicast recognition logic is moved from route cache to here. 1667 The problem was that too many Ethernet cards have broken/missing 1668 hardware multicast filters :-( As result the host on multicasting 1669 network acquires a lot of useless route cache entries, sort of 1670 SDR messages from all the world. Now we try to get rid of them. 1671 Really, provided software IP multicast filter is organized 1672 reasonably (at least, hashed), it does not result in a slowdown 1673 comparing with route cache reject entries. 1674 Note, that multicast routers are not affected, because 1675 route cache entry is created eventually. 1676 */ 1677 if (ipv4_is_multicast(daddr)) { 1678 struct in_device *in_dev = __in_dev_get_rcu(dev); 1679 1680 if (in_dev) { 1681 int our = ip_check_mc_rcu(in_dev, daddr, saddr, 1682 ip_hdr(skb)->protocol); 1683 if (our 1684 #ifdef CONFIG_IP_MROUTE 1685 || 1686 (!ipv4_is_local_multicast(daddr) && 1687 IN_DEV_MFORWARD(in_dev)) 1688 #endif 1689 ) { 1690 int res = ip_route_input_mc(skb, daddr, saddr, 1691 tos, dev, our); 1692 rcu_read_unlock(); 1693 return res; 1694 } 1695 } 1696 rcu_read_unlock(); 1697 return -EINVAL; 1698 } 1699 res = ip_route_input_slow(skb, daddr, saddr, tos, dev); 1700 rcu_read_unlock(); 1701 return res; 1702 } 1703 EXPORT_SYMBOL(ip_route_input); 1704 1705 /* called with rcu_read_lock() */ 1706 static struct rtable *__mkroute_output(const struct fib_result *res, 1707 const struct flowi4 *fl4, int orig_oif, 1708 struct net_device *dev_out, 1709 unsigned int flags) 1710 { 1711 struct fib_info *fi = res->fi; 1712 struct fib_nh_exception *fnhe; 1713 struct in_device *in_dev; 1714 u16 type = res->type; 1715 struct rtable *rth; 1716 1717 in_dev = __in_dev_get_rcu(dev_out); 1718 if (!in_dev) 1719 return ERR_PTR(-EINVAL); 1720 1721 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 1722 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK)) 1723 return ERR_PTR(-EINVAL); 1724 1725 if (ipv4_is_lbcast(fl4->daddr)) 1726 type = RTN_BROADCAST; 1727 else if (ipv4_is_multicast(fl4->daddr)) 1728 type = RTN_MULTICAST; 1729 else if (ipv4_is_zeronet(fl4->daddr)) 1730 return ERR_PTR(-EINVAL); 1731 1732 if (dev_out->flags & IFF_LOOPBACK) 1733 flags |= RTCF_LOCAL; 1734 1735 if (type == RTN_BROADCAST) { 1736 flags |= RTCF_BROADCAST | RTCF_LOCAL; 1737 fi = NULL; 1738 } else if (type == RTN_MULTICAST) { 1739 flags |= RTCF_MULTICAST | RTCF_LOCAL; 1740 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr, 1741 fl4->flowi4_proto)) 1742 flags &= ~RTCF_LOCAL; 1743 /* If multicast route do not exist use 1744 * default one, but do not gateway in this case. 1745 * Yes, it is hack. 1746 */ 1747 if (fi && res->prefixlen < 4) 1748 fi = NULL; 1749 } 1750 1751 fnhe = NULL; 1752 if (fi) { 1753 fnhe = find_exception(&FIB_RES_NH(*res), fl4->daddr); 1754 if (!fnhe) { 1755 rth = FIB_RES_NH(*res).nh_rth_output; 1756 if (rt_cache_valid(rth)) { 1757 dst_hold(&rth->dst); 1758 return rth; 1759 } 1760 } 1761 } 1762 rth = rt_dst_alloc(dev_out, 1763 IN_DEV_CONF_GET(in_dev, NOPOLICY), 1764 IN_DEV_CONF_GET(in_dev, NOXFRM), 1765 fi && !fnhe); 1766 if (!rth) 1767 return ERR_PTR(-ENOBUFS); 1768 1769 rth->dst.output = ip_output; 1770 1771 rth->rt_genid = rt_genid(dev_net(dev_out)); 1772 rth->rt_flags = flags; 1773 rth->rt_type = type; 1774 rth->rt_is_input = 0; 1775 rth->rt_iif = orig_oif ? : 0; 1776 rth->rt_pmtu = 0; 1777 rth->rt_gateway = 0; 1778 1779 RT_CACHE_STAT_INC(out_slow_tot); 1780 1781 if (flags & RTCF_LOCAL) 1782 rth->dst.input = ip_local_deliver; 1783 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 1784 if (flags & RTCF_LOCAL && 1785 !(dev_out->flags & IFF_LOOPBACK)) { 1786 rth->dst.output = ip_mc_output; 1787 RT_CACHE_STAT_INC(out_slow_mc); 1788 } 1789 #ifdef CONFIG_IP_MROUTE 1790 if (type == RTN_MULTICAST) { 1791 if (IN_DEV_MFORWARD(in_dev) && 1792 !ipv4_is_local_multicast(fl4->daddr)) { 1793 rth->dst.input = ip_mr_input; 1794 rth->dst.output = ip_mc_output; 1795 } 1796 } 1797 #endif 1798 } 1799 1800 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0); 1801 1802 return rth; 1803 } 1804 1805 /* 1806 * Major route resolver routine. 1807 */ 1808 1809 struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4) 1810 { 1811 struct net_device *dev_out = NULL; 1812 __u8 tos = RT_FL_TOS(fl4); 1813 unsigned int flags = 0; 1814 struct fib_result res; 1815 struct rtable *rth; 1816 int orig_oif; 1817 1818 res.tclassid = 0; 1819 res.fi = NULL; 1820 res.table = NULL; 1821 1822 orig_oif = fl4->flowi4_oif; 1823 1824 fl4->flowi4_iif = net->loopback_dev->ifindex; 1825 fl4->flowi4_tos = tos & IPTOS_RT_MASK; 1826 fl4->flowi4_scope = ((tos & RTO_ONLINK) ? 1827 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE); 1828 1829 rcu_read_lock(); 1830 if (fl4->saddr) { 1831 rth = ERR_PTR(-EINVAL); 1832 if (ipv4_is_multicast(fl4->saddr) || 1833 ipv4_is_lbcast(fl4->saddr) || 1834 ipv4_is_zeronet(fl4->saddr)) 1835 goto out; 1836 1837 /* I removed check for oif == dev_out->oif here. 1838 It was wrong for two reasons: 1839 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 1840 is assigned to multiple interfaces. 1841 2. Moreover, we are allowed to send packets with saddr 1842 of another iface. --ANK 1843 */ 1844 1845 if (fl4->flowi4_oif == 0 && 1846 (ipv4_is_multicast(fl4->daddr) || 1847 ipv4_is_lbcast(fl4->daddr))) { 1848 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 1849 dev_out = __ip_dev_find(net, fl4->saddr, false); 1850 if (dev_out == NULL) 1851 goto out; 1852 1853 /* Special hack: user can direct multicasts 1854 and limited broadcast via necessary interface 1855 without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 1856 This hack is not just for fun, it allows 1857 vic,vat and friends to work. 1858 They bind socket to loopback, set ttl to zero 1859 and expect that it will work. 1860 From the viewpoint of routing cache they are broken, 1861 because we are not allowed to build multicast path 1862 with loopback source addr (look, routing cache 1863 cannot know, that ttl is zero, so that packet 1864 will not leave this host and route is valid). 1865 Luckily, this hack is good workaround. 1866 */ 1867 1868 fl4->flowi4_oif = dev_out->ifindex; 1869 goto make_route; 1870 } 1871 1872 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 1873 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 1874 if (!__ip_dev_find(net, fl4->saddr, false)) 1875 goto out; 1876 } 1877 } 1878 1879 1880 if (fl4->flowi4_oif) { 1881 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif); 1882 rth = ERR_PTR(-ENODEV); 1883 if (dev_out == NULL) 1884 goto out; 1885 1886 /* RACE: Check return value of inet_select_addr instead. */ 1887 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 1888 rth = ERR_PTR(-ENETUNREACH); 1889 goto out; 1890 } 1891 if (ipv4_is_local_multicast(fl4->daddr) || 1892 ipv4_is_lbcast(fl4->daddr)) { 1893 if (!fl4->saddr) 1894 fl4->saddr = inet_select_addr(dev_out, 0, 1895 RT_SCOPE_LINK); 1896 goto make_route; 1897 } 1898 if (fl4->saddr) { 1899 if (ipv4_is_multicast(fl4->daddr)) 1900 fl4->saddr = inet_select_addr(dev_out, 0, 1901 fl4->flowi4_scope); 1902 else if (!fl4->daddr) 1903 fl4->saddr = inet_select_addr(dev_out, 0, 1904 RT_SCOPE_HOST); 1905 } 1906 } 1907 1908 if (!fl4->daddr) { 1909 fl4->daddr = fl4->saddr; 1910 if (!fl4->daddr) 1911 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK); 1912 dev_out = net->loopback_dev; 1913 fl4->flowi4_oif = net->loopback_dev->ifindex; 1914 res.type = RTN_LOCAL; 1915 flags |= RTCF_LOCAL; 1916 goto make_route; 1917 } 1918 1919 if (fib_lookup(net, fl4, &res)) { 1920 res.fi = NULL; 1921 res.table = NULL; 1922 if (fl4->flowi4_oif) { 1923 /* Apparently, routing tables are wrong. Assume, 1924 that the destination is on link. 1925 1926 WHY? DW. 1927 Because we are allowed to send to iface 1928 even if it has NO routes and NO assigned 1929 addresses. When oif is specified, routing 1930 tables are looked up with only one purpose: 1931 to catch if destination is gatewayed, rather than 1932 direct. Moreover, if MSG_DONTROUTE is set, 1933 we send packet, ignoring both routing tables 1934 and ifaddr state. --ANK 1935 1936 1937 We could make it even if oif is unknown, 1938 likely IPv6, but we do not. 1939 */ 1940 1941 if (fl4->saddr == 0) 1942 fl4->saddr = inet_select_addr(dev_out, 0, 1943 RT_SCOPE_LINK); 1944 res.type = RTN_UNICAST; 1945 goto make_route; 1946 } 1947 rth = ERR_PTR(-ENETUNREACH); 1948 goto out; 1949 } 1950 1951 if (res.type == RTN_LOCAL) { 1952 if (!fl4->saddr) { 1953 if (res.fi->fib_prefsrc) 1954 fl4->saddr = res.fi->fib_prefsrc; 1955 else 1956 fl4->saddr = fl4->daddr; 1957 } 1958 dev_out = net->loopback_dev; 1959 fl4->flowi4_oif = dev_out->ifindex; 1960 res.fi = NULL; 1961 flags |= RTCF_LOCAL; 1962 goto make_route; 1963 } 1964 1965 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1966 if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0) 1967 fib_select_multipath(&res); 1968 else 1969 #endif 1970 if (!res.prefixlen && 1971 res.table->tb_num_default > 1 && 1972 res.type == RTN_UNICAST && !fl4->flowi4_oif) 1973 fib_select_default(&res); 1974 1975 if (!fl4->saddr) 1976 fl4->saddr = FIB_RES_PREFSRC(net, res); 1977 1978 dev_out = FIB_RES_DEV(res); 1979 fl4->flowi4_oif = dev_out->ifindex; 1980 1981 1982 make_route: 1983 rth = __mkroute_output(&res, fl4, orig_oif, dev_out, flags); 1984 1985 out: 1986 rcu_read_unlock(); 1987 return rth; 1988 } 1989 EXPORT_SYMBOL_GPL(__ip_route_output_key); 1990 1991 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie) 1992 { 1993 return NULL; 1994 } 1995 1996 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst) 1997 { 1998 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU); 1999 2000 return mtu ? : dst->dev->mtu; 2001 } 2002 2003 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk, 2004 struct sk_buff *skb, u32 mtu) 2005 { 2006 } 2007 2008 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk, 2009 struct sk_buff *skb) 2010 { 2011 } 2012 2013 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst, 2014 unsigned long old) 2015 { 2016 return NULL; 2017 } 2018 2019 static struct dst_ops ipv4_dst_blackhole_ops = { 2020 .family = AF_INET, 2021 .protocol = cpu_to_be16(ETH_P_IP), 2022 .check = ipv4_blackhole_dst_check, 2023 .mtu = ipv4_blackhole_mtu, 2024 .default_advmss = ipv4_default_advmss, 2025 .update_pmtu = ipv4_rt_blackhole_update_pmtu, 2026 .redirect = ipv4_rt_blackhole_redirect, 2027 .cow_metrics = ipv4_rt_blackhole_cow_metrics, 2028 .neigh_lookup = ipv4_neigh_lookup, 2029 }; 2030 2031 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2032 { 2033 struct rtable *ort = (struct rtable *) dst_orig; 2034 struct rtable *rt; 2035 2036 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_NONE, 0); 2037 if (rt) { 2038 struct dst_entry *new = &rt->dst; 2039 2040 new->__use = 1; 2041 new->input = dst_discard; 2042 new->output = dst_discard; 2043 2044 new->dev = ort->dst.dev; 2045 if (new->dev) 2046 dev_hold(new->dev); 2047 2048 rt->rt_is_input = ort->rt_is_input; 2049 rt->rt_iif = ort->rt_iif; 2050 rt->rt_pmtu = ort->rt_pmtu; 2051 2052 rt->rt_genid = rt_genid(net); 2053 rt->rt_flags = ort->rt_flags; 2054 rt->rt_type = ort->rt_type; 2055 rt->rt_gateway = ort->rt_gateway; 2056 2057 dst_free(new); 2058 } 2059 2060 dst_release(dst_orig); 2061 2062 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2063 } 2064 2065 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2066 struct sock *sk) 2067 { 2068 struct rtable *rt = __ip_route_output_key(net, flp4); 2069 2070 if (IS_ERR(rt)) 2071 return rt; 2072 2073 if (flp4->flowi4_proto) 2074 rt = (struct rtable *) xfrm_lookup(net, &rt->dst, 2075 flowi4_to_flowi(flp4), 2076 sk, 0); 2077 2078 return rt; 2079 } 2080 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2081 2082 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, 2083 struct flowi4 *fl4, struct sk_buff *skb, u32 pid, 2084 u32 seq, int event, int nowait, unsigned int flags) 2085 { 2086 struct rtable *rt = skb_rtable(skb); 2087 struct rtmsg *r; 2088 struct nlmsghdr *nlh; 2089 unsigned long expires = 0; 2090 u32 error; 2091 u32 metrics[RTAX_MAX]; 2092 2093 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags); 2094 if (nlh == NULL) 2095 return -EMSGSIZE; 2096 2097 r = nlmsg_data(nlh); 2098 r->rtm_family = AF_INET; 2099 r->rtm_dst_len = 32; 2100 r->rtm_src_len = 0; 2101 r->rtm_tos = fl4->flowi4_tos; 2102 r->rtm_table = RT_TABLE_MAIN; 2103 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN)) 2104 goto nla_put_failure; 2105 r->rtm_type = rt->rt_type; 2106 r->rtm_scope = RT_SCOPE_UNIVERSE; 2107 r->rtm_protocol = RTPROT_UNSPEC; 2108 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 2109 if (rt->rt_flags & RTCF_NOTIFY) 2110 r->rtm_flags |= RTM_F_NOTIFY; 2111 2112 if (nla_put_be32(skb, RTA_DST, dst)) 2113 goto nla_put_failure; 2114 if (src) { 2115 r->rtm_src_len = 32; 2116 if (nla_put_be32(skb, RTA_SRC, src)) 2117 goto nla_put_failure; 2118 } 2119 if (rt->dst.dev && 2120 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex)) 2121 goto nla_put_failure; 2122 #ifdef CONFIG_IP_ROUTE_CLASSID 2123 if (rt->dst.tclassid && 2124 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid)) 2125 goto nla_put_failure; 2126 #endif 2127 if (!rt_is_input_route(rt) && 2128 fl4->saddr != src) { 2129 if (nla_put_be32(skb, RTA_PREFSRC, fl4->saddr)) 2130 goto nla_put_failure; 2131 } 2132 if (rt->rt_gateway && 2133 nla_put_be32(skb, RTA_GATEWAY, rt->rt_gateway)) 2134 goto nla_put_failure; 2135 2136 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics)); 2137 if (rt->rt_pmtu) 2138 metrics[RTAX_MTU - 1] = rt->rt_pmtu; 2139 if (rtnetlink_put_metrics(skb, metrics) < 0) 2140 goto nla_put_failure; 2141 2142 if (fl4->flowi4_mark && 2143 nla_put_be32(skb, RTA_MARK, fl4->flowi4_mark)) 2144 goto nla_put_failure; 2145 2146 error = rt->dst.error; 2147 expires = rt->dst.expires; 2148 if (expires) { 2149 if (time_before(jiffies, expires)) 2150 expires -= jiffies; 2151 else 2152 expires = 0; 2153 } 2154 2155 if (rt_is_input_route(rt)) { 2156 if (nla_put_u32(skb, RTA_IIF, rt->rt_iif)) 2157 goto nla_put_failure; 2158 } 2159 2160 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0) 2161 goto nla_put_failure; 2162 2163 return nlmsg_end(skb, nlh); 2164 2165 nla_put_failure: 2166 nlmsg_cancel(skb, nlh); 2167 return -EMSGSIZE; 2168 } 2169 2170 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg) 2171 { 2172 struct net *net = sock_net(in_skb->sk); 2173 struct rtmsg *rtm; 2174 struct nlattr *tb[RTA_MAX+1]; 2175 struct rtable *rt = NULL; 2176 struct flowi4 fl4; 2177 __be32 dst = 0; 2178 __be32 src = 0; 2179 u32 iif; 2180 int err; 2181 int mark; 2182 struct sk_buff *skb; 2183 2184 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy); 2185 if (err < 0) 2186 goto errout; 2187 2188 rtm = nlmsg_data(nlh); 2189 2190 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 2191 if (skb == NULL) { 2192 err = -ENOBUFS; 2193 goto errout; 2194 } 2195 2196 /* Reserve room for dummy headers, this skb can pass 2197 through good chunk of routing engine. 2198 */ 2199 skb_reset_mac_header(skb); 2200 skb_reset_network_header(skb); 2201 2202 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */ 2203 ip_hdr(skb)->protocol = IPPROTO_ICMP; 2204 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr)); 2205 2206 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0; 2207 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0; 2208 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0; 2209 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0; 2210 2211 memset(&fl4, 0, sizeof(fl4)); 2212 fl4.daddr = dst; 2213 fl4.saddr = src; 2214 fl4.flowi4_tos = rtm->rtm_tos; 2215 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0; 2216 fl4.flowi4_mark = mark; 2217 2218 if (iif) { 2219 struct net_device *dev; 2220 2221 dev = __dev_get_by_index(net, iif); 2222 if (dev == NULL) { 2223 err = -ENODEV; 2224 goto errout_free; 2225 } 2226 2227 skb->protocol = htons(ETH_P_IP); 2228 skb->dev = dev; 2229 skb->mark = mark; 2230 local_bh_disable(); 2231 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev); 2232 local_bh_enable(); 2233 2234 rt = skb_rtable(skb); 2235 if (err == 0 && rt->dst.error) 2236 err = -rt->dst.error; 2237 } else { 2238 rt = ip_route_output_key(net, &fl4); 2239 2240 err = 0; 2241 if (IS_ERR(rt)) 2242 err = PTR_ERR(rt); 2243 } 2244 2245 if (err) 2246 goto errout_free; 2247 2248 skb_dst_set(skb, &rt->dst); 2249 if (rtm->rtm_flags & RTM_F_NOTIFY) 2250 rt->rt_flags |= RTCF_NOTIFY; 2251 2252 err = rt_fill_info(net, dst, src, &fl4, skb, 2253 NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, 2254 RTM_NEWROUTE, 0, 0); 2255 if (err <= 0) 2256 goto errout_free; 2257 2258 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid); 2259 errout: 2260 return err; 2261 2262 errout_free: 2263 kfree_skb(skb); 2264 goto errout; 2265 } 2266 2267 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb) 2268 { 2269 return skb->len; 2270 } 2271 2272 void ip_rt_multicast_event(struct in_device *in_dev) 2273 { 2274 rt_cache_flush(dev_net(in_dev->dev), 0); 2275 } 2276 2277 #ifdef CONFIG_SYSCTL 2278 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write, 2279 void __user *buffer, 2280 size_t *lenp, loff_t *ppos) 2281 { 2282 if (write) { 2283 int flush_delay; 2284 ctl_table ctl; 2285 struct net *net; 2286 2287 memcpy(&ctl, __ctl, sizeof(ctl)); 2288 ctl.data = &flush_delay; 2289 proc_dointvec(&ctl, write, buffer, lenp, ppos); 2290 2291 net = (struct net *)__ctl->extra1; 2292 rt_cache_flush(net, flush_delay); 2293 return 0; 2294 } 2295 2296 return -EINVAL; 2297 } 2298 2299 static ctl_table ipv4_route_table[] = { 2300 { 2301 .procname = "gc_thresh", 2302 .data = &ipv4_dst_ops.gc_thresh, 2303 .maxlen = sizeof(int), 2304 .mode = 0644, 2305 .proc_handler = proc_dointvec, 2306 }, 2307 { 2308 .procname = "max_size", 2309 .data = &ip_rt_max_size, 2310 .maxlen = sizeof(int), 2311 .mode = 0644, 2312 .proc_handler = proc_dointvec, 2313 }, 2314 { 2315 /* Deprecated. Use gc_min_interval_ms */ 2316 2317 .procname = "gc_min_interval", 2318 .data = &ip_rt_gc_min_interval, 2319 .maxlen = sizeof(int), 2320 .mode = 0644, 2321 .proc_handler = proc_dointvec_jiffies, 2322 }, 2323 { 2324 .procname = "gc_min_interval_ms", 2325 .data = &ip_rt_gc_min_interval, 2326 .maxlen = sizeof(int), 2327 .mode = 0644, 2328 .proc_handler = proc_dointvec_ms_jiffies, 2329 }, 2330 { 2331 .procname = "gc_timeout", 2332 .data = &ip_rt_gc_timeout, 2333 .maxlen = sizeof(int), 2334 .mode = 0644, 2335 .proc_handler = proc_dointvec_jiffies, 2336 }, 2337 { 2338 .procname = "gc_interval", 2339 .data = &ip_rt_gc_interval, 2340 .maxlen = sizeof(int), 2341 .mode = 0644, 2342 .proc_handler = proc_dointvec_jiffies, 2343 }, 2344 { 2345 .procname = "redirect_load", 2346 .data = &ip_rt_redirect_load, 2347 .maxlen = sizeof(int), 2348 .mode = 0644, 2349 .proc_handler = proc_dointvec, 2350 }, 2351 { 2352 .procname = "redirect_number", 2353 .data = &ip_rt_redirect_number, 2354 .maxlen = sizeof(int), 2355 .mode = 0644, 2356 .proc_handler = proc_dointvec, 2357 }, 2358 { 2359 .procname = "redirect_silence", 2360 .data = &ip_rt_redirect_silence, 2361 .maxlen = sizeof(int), 2362 .mode = 0644, 2363 .proc_handler = proc_dointvec, 2364 }, 2365 { 2366 .procname = "error_cost", 2367 .data = &ip_rt_error_cost, 2368 .maxlen = sizeof(int), 2369 .mode = 0644, 2370 .proc_handler = proc_dointvec, 2371 }, 2372 { 2373 .procname = "error_burst", 2374 .data = &ip_rt_error_burst, 2375 .maxlen = sizeof(int), 2376 .mode = 0644, 2377 .proc_handler = proc_dointvec, 2378 }, 2379 { 2380 .procname = "gc_elasticity", 2381 .data = &ip_rt_gc_elasticity, 2382 .maxlen = sizeof(int), 2383 .mode = 0644, 2384 .proc_handler = proc_dointvec, 2385 }, 2386 { 2387 .procname = "mtu_expires", 2388 .data = &ip_rt_mtu_expires, 2389 .maxlen = sizeof(int), 2390 .mode = 0644, 2391 .proc_handler = proc_dointvec_jiffies, 2392 }, 2393 { 2394 .procname = "min_pmtu", 2395 .data = &ip_rt_min_pmtu, 2396 .maxlen = sizeof(int), 2397 .mode = 0644, 2398 .proc_handler = proc_dointvec, 2399 }, 2400 { 2401 .procname = "min_adv_mss", 2402 .data = &ip_rt_min_advmss, 2403 .maxlen = sizeof(int), 2404 .mode = 0644, 2405 .proc_handler = proc_dointvec, 2406 }, 2407 { } 2408 }; 2409 2410 static struct ctl_table ipv4_route_flush_table[] = { 2411 { 2412 .procname = "flush", 2413 .maxlen = sizeof(int), 2414 .mode = 0200, 2415 .proc_handler = ipv4_sysctl_rtcache_flush, 2416 }, 2417 { }, 2418 }; 2419 2420 static __net_init int sysctl_route_net_init(struct net *net) 2421 { 2422 struct ctl_table *tbl; 2423 2424 tbl = ipv4_route_flush_table; 2425 if (!net_eq(net, &init_net)) { 2426 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL); 2427 if (tbl == NULL) 2428 goto err_dup; 2429 } 2430 tbl[0].extra1 = net; 2431 2432 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl); 2433 if (net->ipv4.route_hdr == NULL) 2434 goto err_reg; 2435 return 0; 2436 2437 err_reg: 2438 if (tbl != ipv4_route_flush_table) 2439 kfree(tbl); 2440 err_dup: 2441 return -ENOMEM; 2442 } 2443 2444 static __net_exit void sysctl_route_net_exit(struct net *net) 2445 { 2446 struct ctl_table *tbl; 2447 2448 tbl = net->ipv4.route_hdr->ctl_table_arg; 2449 unregister_net_sysctl_table(net->ipv4.route_hdr); 2450 BUG_ON(tbl == ipv4_route_flush_table); 2451 kfree(tbl); 2452 } 2453 2454 static __net_initdata struct pernet_operations sysctl_route_ops = { 2455 .init = sysctl_route_net_init, 2456 .exit = sysctl_route_net_exit, 2457 }; 2458 #endif 2459 2460 static __net_init int rt_genid_init(struct net *net) 2461 { 2462 get_random_bytes(&net->ipv4.rt_genid, 2463 sizeof(net->ipv4.rt_genid)); 2464 get_random_bytes(&net->ipv4.dev_addr_genid, 2465 sizeof(net->ipv4.dev_addr_genid)); 2466 return 0; 2467 } 2468 2469 static __net_initdata struct pernet_operations rt_genid_ops = { 2470 .init = rt_genid_init, 2471 }; 2472 2473 static int __net_init ipv4_inetpeer_init(struct net *net) 2474 { 2475 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL); 2476 2477 if (!bp) 2478 return -ENOMEM; 2479 inet_peer_base_init(bp); 2480 net->ipv4.peers = bp; 2481 return 0; 2482 } 2483 2484 static void __net_exit ipv4_inetpeer_exit(struct net *net) 2485 { 2486 struct inet_peer_base *bp = net->ipv4.peers; 2487 2488 net->ipv4.peers = NULL; 2489 inetpeer_invalidate_tree(bp); 2490 kfree(bp); 2491 } 2492 2493 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = { 2494 .init = ipv4_inetpeer_init, 2495 .exit = ipv4_inetpeer_exit, 2496 }; 2497 2498 #ifdef CONFIG_IP_ROUTE_CLASSID 2499 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 2500 #endif /* CONFIG_IP_ROUTE_CLASSID */ 2501 2502 int __init ip_rt_init(void) 2503 { 2504 int rc = 0; 2505 2506 #ifdef CONFIG_IP_ROUTE_CLASSID 2507 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 2508 if (!ip_rt_acct) 2509 panic("IP: failed to allocate ip_rt_acct\n"); 2510 #endif 2511 2512 ipv4_dst_ops.kmem_cachep = 2513 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0, 2514 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 2515 2516 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 2517 2518 if (dst_entries_init(&ipv4_dst_ops) < 0) 2519 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 2520 2521 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 2522 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 2523 2524 ipv4_dst_ops.gc_thresh = ~0; 2525 ip_rt_max_size = INT_MAX; 2526 2527 devinet_init(); 2528 ip_fib_init(); 2529 2530 if (ip_rt_proc_init()) 2531 pr_err("Unable to create route proc files\n"); 2532 #ifdef CONFIG_XFRM 2533 xfrm_init(); 2534 xfrm4_init(ip_rt_max_size); 2535 #endif 2536 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, NULL); 2537 2538 #ifdef CONFIG_SYSCTL 2539 register_pernet_subsys(&sysctl_route_ops); 2540 #endif 2541 register_pernet_subsys(&rt_genid_ops); 2542 register_pernet_subsys(&ipv4_inetpeer_ops); 2543 return rc; 2544 } 2545 2546 #ifdef CONFIG_SYSCTL 2547 /* 2548 * We really need to sanitize the damn ipv4 init order, then all 2549 * this nonsense will go away. 2550 */ 2551 void __init ip_static_sysctl_init(void) 2552 { 2553 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table); 2554 } 2555 #endif 2556