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 int 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 * Interface to generic destination cache. 137 */ 138 139 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie); 140 static unsigned int ipv4_default_advmss(const struct dst_entry *dst); 141 static unsigned int ipv4_mtu(const struct dst_entry *dst); 142 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst); 143 static void ipv4_link_failure(struct sk_buff *skb); 144 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 145 struct sk_buff *skb, u32 mtu); 146 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, 147 struct sk_buff *skb); 148 static void ipv4_dst_destroy(struct dst_entry *dst); 149 150 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old) 151 { 152 WARN_ON(1); 153 return NULL; 154 } 155 156 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 157 struct sk_buff *skb, 158 const void *daddr); 159 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr); 160 161 static struct dst_ops ipv4_dst_ops = { 162 .family = AF_INET, 163 .check = ipv4_dst_check, 164 .default_advmss = ipv4_default_advmss, 165 .mtu = ipv4_mtu, 166 .cow_metrics = ipv4_cow_metrics, 167 .destroy = ipv4_dst_destroy, 168 .negative_advice = ipv4_negative_advice, 169 .link_failure = ipv4_link_failure, 170 .update_pmtu = ip_rt_update_pmtu, 171 .redirect = ip_do_redirect, 172 .local_out = __ip_local_out, 173 .neigh_lookup = ipv4_neigh_lookup, 174 .confirm_neigh = ipv4_confirm_neigh, 175 }; 176 177 #define ECN_OR_COST(class) TC_PRIO_##class 178 179 const __u8 ip_tos2prio[16] = { 180 TC_PRIO_BESTEFFORT, 181 ECN_OR_COST(BESTEFFORT), 182 TC_PRIO_BESTEFFORT, 183 ECN_OR_COST(BESTEFFORT), 184 TC_PRIO_BULK, 185 ECN_OR_COST(BULK), 186 TC_PRIO_BULK, 187 ECN_OR_COST(BULK), 188 TC_PRIO_INTERACTIVE, 189 ECN_OR_COST(INTERACTIVE), 190 TC_PRIO_INTERACTIVE, 191 ECN_OR_COST(INTERACTIVE), 192 TC_PRIO_INTERACTIVE_BULK, 193 ECN_OR_COST(INTERACTIVE_BULK), 194 TC_PRIO_INTERACTIVE_BULK, 195 ECN_OR_COST(INTERACTIVE_BULK) 196 }; 197 EXPORT_SYMBOL(ip_tos2prio); 198 199 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat); 200 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field) 201 202 #ifdef CONFIG_PROC_FS 203 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos) 204 { 205 if (*pos) 206 return NULL; 207 return SEQ_START_TOKEN; 208 } 209 210 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos) 211 { 212 ++*pos; 213 return NULL; 214 } 215 216 static void rt_cache_seq_stop(struct seq_file *seq, void *v) 217 { 218 } 219 220 static int rt_cache_seq_show(struct seq_file *seq, void *v) 221 { 222 if (v == SEQ_START_TOKEN) 223 seq_printf(seq, "%-127s\n", 224 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t" 225 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t" 226 "HHUptod\tSpecDst"); 227 return 0; 228 } 229 230 static const struct seq_operations rt_cache_seq_ops = { 231 .start = rt_cache_seq_start, 232 .next = rt_cache_seq_next, 233 .stop = rt_cache_seq_stop, 234 .show = rt_cache_seq_show, 235 }; 236 237 static int rt_cache_seq_open(struct inode *inode, struct file *file) 238 { 239 return seq_open(file, &rt_cache_seq_ops); 240 } 241 242 static const struct file_operations rt_cache_seq_fops = { 243 .open = rt_cache_seq_open, 244 .read = seq_read, 245 .llseek = seq_lseek, 246 .release = seq_release, 247 }; 248 249 250 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos) 251 { 252 int cpu; 253 254 if (*pos == 0) 255 return SEQ_START_TOKEN; 256 257 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) { 258 if (!cpu_possible(cpu)) 259 continue; 260 *pos = cpu+1; 261 return &per_cpu(rt_cache_stat, cpu); 262 } 263 return NULL; 264 } 265 266 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos) 267 { 268 int cpu; 269 270 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) { 271 if (!cpu_possible(cpu)) 272 continue; 273 *pos = cpu+1; 274 return &per_cpu(rt_cache_stat, cpu); 275 } 276 return NULL; 277 278 } 279 280 static void rt_cpu_seq_stop(struct seq_file *seq, void *v) 281 { 282 283 } 284 285 static int rt_cpu_seq_show(struct seq_file *seq, void *v) 286 { 287 struct rt_cache_stat *st = v; 288 289 if (v == SEQ_START_TOKEN) { 290 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"); 291 return 0; 292 } 293 294 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x " 295 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n", 296 dst_entries_get_slow(&ipv4_dst_ops), 297 0, /* st->in_hit */ 298 st->in_slow_tot, 299 st->in_slow_mc, 300 st->in_no_route, 301 st->in_brd, 302 st->in_martian_dst, 303 st->in_martian_src, 304 305 0, /* st->out_hit */ 306 st->out_slow_tot, 307 st->out_slow_mc, 308 309 0, /* st->gc_total */ 310 0, /* st->gc_ignored */ 311 0, /* st->gc_goal_miss */ 312 0, /* st->gc_dst_overflow */ 313 0, /* st->in_hlist_search */ 314 0 /* st->out_hlist_search */ 315 ); 316 return 0; 317 } 318 319 static const struct seq_operations rt_cpu_seq_ops = { 320 .start = rt_cpu_seq_start, 321 .next = rt_cpu_seq_next, 322 .stop = rt_cpu_seq_stop, 323 .show = rt_cpu_seq_show, 324 }; 325 326 327 static int rt_cpu_seq_open(struct inode *inode, struct file *file) 328 { 329 return seq_open(file, &rt_cpu_seq_ops); 330 } 331 332 static const struct file_operations rt_cpu_seq_fops = { 333 .open = rt_cpu_seq_open, 334 .read = seq_read, 335 .llseek = seq_lseek, 336 .release = seq_release, 337 }; 338 339 #ifdef CONFIG_IP_ROUTE_CLASSID 340 static int rt_acct_proc_show(struct seq_file *m, void *v) 341 { 342 struct ip_rt_acct *dst, *src; 343 unsigned int i, j; 344 345 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL); 346 if (!dst) 347 return -ENOMEM; 348 349 for_each_possible_cpu(i) { 350 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i); 351 for (j = 0; j < 256; j++) { 352 dst[j].o_bytes += src[j].o_bytes; 353 dst[j].o_packets += src[j].o_packets; 354 dst[j].i_bytes += src[j].i_bytes; 355 dst[j].i_packets += src[j].i_packets; 356 } 357 } 358 359 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct)); 360 kfree(dst); 361 return 0; 362 } 363 364 static int rt_acct_proc_open(struct inode *inode, struct file *file) 365 { 366 return single_open(file, rt_acct_proc_show, NULL); 367 } 368 369 static const struct file_operations rt_acct_proc_fops = { 370 .open = rt_acct_proc_open, 371 .read = seq_read, 372 .llseek = seq_lseek, 373 .release = single_release, 374 }; 375 #endif 376 377 static int __net_init ip_rt_do_proc_init(struct net *net) 378 { 379 struct proc_dir_entry *pde; 380 381 pde = proc_create("rt_cache", S_IRUGO, net->proc_net, 382 &rt_cache_seq_fops); 383 if (!pde) 384 goto err1; 385 386 pde = proc_create("rt_cache", S_IRUGO, 387 net->proc_net_stat, &rt_cpu_seq_fops); 388 if (!pde) 389 goto err2; 390 391 #ifdef CONFIG_IP_ROUTE_CLASSID 392 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops); 393 if (!pde) 394 goto err3; 395 #endif 396 return 0; 397 398 #ifdef CONFIG_IP_ROUTE_CLASSID 399 err3: 400 remove_proc_entry("rt_cache", net->proc_net_stat); 401 #endif 402 err2: 403 remove_proc_entry("rt_cache", net->proc_net); 404 err1: 405 return -ENOMEM; 406 } 407 408 static void __net_exit ip_rt_do_proc_exit(struct net *net) 409 { 410 remove_proc_entry("rt_cache", net->proc_net_stat); 411 remove_proc_entry("rt_cache", net->proc_net); 412 #ifdef CONFIG_IP_ROUTE_CLASSID 413 remove_proc_entry("rt_acct", net->proc_net); 414 #endif 415 } 416 417 static struct pernet_operations ip_rt_proc_ops __net_initdata = { 418 .init = ip_rt_do_proc_init, 419 .exit = ip_rt_do_proc_exit, 420 }; 421 422 static int __init ip_rt_proc_init(void) 423 { 424 return register_pernet_subsys(&ip_rt_proc_ops); 425 } 426 427 #else 428 static inline int ip_rt_proc_init(void) 429 { 430 return 0; 431 } 432 #endif /* CONFIG_PROC_FS */ 433 434 static inline bool rt_is_expired(const struct rtable *rth) 435 { 436 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev)); 437 } 438 439 void rt_cache_flush(struct net *net) 440 { 441 rt_genid_bump_ipv4(net); 442 } 443 444 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 445 struct sk_buff *skb, 446 const void *daddr) 447 { 448 struct net_device *dev = dst->dev; 449 const __be32 *pkey = daddr; 450 const struct rtable *rt; 451 struct neighbour *n; 452 453 rt = (const struct rtable *) dst; 454 if (rt->rt_gateway) 455 pkey = (const __be32 *) &rt->rt_gateway; 456 else if (skb) 457 pkey = &ip_hdr(skb)->daddr; 458 459 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey); 460 if (n) 461 return n; 462 return neigh_create(&arp_tbl, pkey, dev); 463 } 464 465 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr) 466 { 467 struct net_device *dev = dst->dev; 468 const __be32 *pkey = daddr; 469 const struct rtable *rt; 470 471 rt = (const struct rtable *)dst; 472 if (rt->rt_gateway) 473 pkey = (const __be32 *)&rt->rt_gateway; 474 else if (!daddr || 475 (rt->rt_flags & 476 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) 477 return; 478 479 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey); 480 } 481 482 #define IP_IDENTS_SZ 2048u 483 484 static atomic_t *ip_idents __read_mostly; 485 static u32 *ip_tstamps __read_mostly; 486 487 /* In order to protect privacy, we add a perturbation to identifiers 488 * if one generator is seldom used. This makes hard for an attacker 489 * to infer how many packets were sent between two points in time. 490 */ 491 u32 ip_idents_reserve(u32 hash, int segs) 492 { 493 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ; 494 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ; 495 u32 old = READ_ONCE(*p_tstamp); 496 u32 now = (u32)jiffies; 497 u32 new, delta = 0; 498 499 if (old != now && cmpxchg(p_tstamp, old, now) == old) 500 delta = prandom_u32_max(now - old); 501 502 /* Do not use atomic_add_return() as it makes UBSAN unhappy */ 503 do { 504 old = (u32)atomic_read(p_id); 505 new = old + delta + segs; 506 } while (atomic_cmpxchg(p_id, old, new) != old); 507 508 return new - segs; 509 } 510 EXPORT_SYMBOL(ip_idents_reserve); 511 512 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs) 513 { 514 static u32 ip_idents_hashrnd __read_mostly; 515 u32 hash, id; 516 517 net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd)); 518 519 hash = jhash_3words((__force u32)iph->daddr, 520 (__force u32)iph->saddr, 521 iph->protocol ^ net_hash_mix(net), 522 ip_idents_hashrnd); 523 id = ip_idents_reserve(hash, segs); 524 iph->id = htons(id); 525 } 526 EXPORT_SYMBOL(__ip_select_ident); 527 528 static void __build_flow_key(const struct net *net, struct flowi4 *fl4, 529 const struct sock *sk, 530 const struct iphdr *iph, 531 int oif, u8 tos, 532 u8 prot, u32 mark, int flow_flags) 533 { 534 if (sk) { 535 const struct inet_sock *inet = inet_sk(sk); 536 537 oif = sk->sk_bound_dev_if; 538 mark = sk->sk_mark; 539 tos = RT_CONN_FLAGS(sk); 540 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol; 541 } 542 flowi4_init_output(fl4, oif, mark, tos, 543 RT_SCOPE_UNIVERSE, prot, 544 flow_flags, 545 iph->daddr, iph->saddr, 0, 0, 546 sock_net_uid(net, sk)); 547 } 548 549 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb, 550 const struct sock *sk) 551 { 552 const struct net *net = dev_net(skb->dev); 553 const struct iphdr *iph = ip_hdr(skb); 554 int oif = skb->dev->ifindex; 555 u8 tos = RT_TOS(iph->tos); 556 u8 prot = iph->protocol; 557 u32 mark = skb->mark; 558 559 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0); 560 } 561 562 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk) 563 { 564 const struct inet_sock *inet = inet_sk(sk); 565 const struct ip_options_rcu *inet_opt; 566 __be32 daddr = inet->inet_daddr; 567 568 rcu_read_lock(); 569 inet_opt = rcu_dereference(inet->inet_opt); 570 if (inet_opt && inet_opt->opt.srr) 571 daddr = inet_opt->opt.faddr; 572 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark, 573 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 574 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol, 575 inet_sk_flowi_flags(sk), 576 daddr, inet->inet_saddr, 0, 0, sk->sk_uid); 577 rcu_read_unlock(); 578 } 579 580 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk, 581 const struct sk_buff *skb) 582 { 583 if (skb) 584 build_skb_flow_key(fl4, skb, sk); 585 else 586 build_sk_flow_key(fl4, sk); 587 } 588 589 static DEFINE_SPINLOCK(fnhe_lock); 590 591 static void fnhe_flush_routes(struct fib_nh_exception *fnhe) 592 { 593 struct rtable *rt; 594 595 rt = rcu_dereference(fnhe->fnhe_rth_input); 596 if (rt) { 597 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL); 598 dst_dev_put(&rt->dst); 599 dst_release(&rt->dst); 600 } 601 rt = rcu_dereference(fnhe->fnhe_rth_output); 602 if (rt) { 603 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL); 604 dst_dev_put(&rt->dst); 605 dst_release(&rt->dst); 606 } 607 } 608 609 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash) 610 { 611 struct fib_nh_exception *fnhe, *oldest; 612 613 oldest = rcu_dereference(hash->chain); 614 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe; 615 fnhe = rcu_dereference(fnhe->fnhe_next)) { 616 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) 617 oldest = fnhe; 618 } 619 fnhe_flush_routes(oldest); 620 return oldest; 621 } 622 623 static inline u32 fnhe_hashfun(__be32 daddr) 624 { 625 static u32 fnhe_hashrnd __read_mostly; 626 u32 hval; 627 628 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd)); 629 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd); 630 return hash_32(hval, FNHE_HASH_SHIFT); 631 } 632 633 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe) 634 { 635 rt->rt_pmtu = fnhe->fnhe_pmtu; 636 rt->dst.expires = fnhe->fnhe_expires; 637 638 if (fnhe->fnhe_gw) { 639 rt->rt_flags |= RTCF_REDIRECTED; 640 rt->rt_gateway = fnhe->fnhe_gw; 641 rt->rt_uses_gateway = 1; 642 } 643 } 644 645 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw, 646 u32 pmtu, unsigned long expires) 647 { 648 struct fnhe_hash_bucket *hash; 649 struct fib_nh_exception *fnhe; 650 struct rtable *rt; 651 u32 genid, hval; 652 unsigned int i; 653 int depth; 654 655 genid = fnhe_genid(dev_net(nh->nh_dev)); 656 hval = fnhe_hashfun(daddr); 657 658 spin_lock_bh(&fnhe_lock); 659 660 hash = rcu_dereference(nh->nh_exceptions); 661 if (!hash) { 662 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC); 663 if (!hash) 664 goto out_unlock; 665 rcu_assign_pointer(nh->nh_exceptions, hash); 666 } 667 668 hash += hval; 669 670 depth = 0; 671 for (fnhe = rcu_dereference(hash->chain); fnhe; 672 fnhe = rcu_dereference(fnhe->fnhe_next)) { 673 if (fnhe->fnhe_daddr == daddr) 674 break; 675 depth++; 676 } 677 678 if (fnhe) { 679 if (fnhe->fnhe_genid != genid) 680 fnhe->fnhe_genid = genid; 681 if (gw) 682 fnhe->fnhe_gw = gw; 683 if (pmtu) 684 fnhe->fnhe_pmtu = pmtu; 685 fnhe->fnhe_expires = max(1UL, expires); 686 /* Update all cached dsts too */ 687 rt = rcu_dereference(fnhe->fnhe_rth_input); 688 if (rt) 689 fill_route_from_fnhe(rt, fnhe); 690 rt = rcu_dereference(fnhe->fnhe_rth_output); 691 if (rt) 692 fill_route_from_fnhe(rt, fnhe); 693 } else { 694 if (depth > FNHE_RECLAIM_DEPTH) 695 fnhe = fnhe_oldest(hash); 696 else { 697 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC); 698 if (!fnhe) 699 goto out_unlock; 700 701 fnhe->fnhe_next = hash->chain; 702 rcu_assign_pointer(hash->chain, fnhe); 703 } 704 fnhe->fnhe_genid = genid; 705 fnhe->fnhe_daddr = daddr; 706 fnhe->fnhe_gw = gw; 707 fnhe->fnhe_pmtu = pmtu; 708 fnhe->fnhe_expires = expires; 709 710 /* Exception created; mark the cached routes for the nexthop 711 * stale, so anyone caching it rechecks if this exception 712 * applies to them. 713 */ 714 rt = rcu_dereference(nh->nh_rth_input); 715 if (rt) 716 rt->dst.obsolete = DST_OBSOLETE_KILL; 717 718 for_each_possible_cpu(i) { 719 struct rtable __rcu **prt; 720 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i); 721 rt = rcu_dereference(*prt); 722 if (rt) 723 rt->dst.obsolete = DST_OBSOLETE_KILL; 724 } 725 } 726 727 fnhe->fnhe_stamp = jiffies; 728 729 out_unlock: 730 spin_unlock_bh(&fnhe_lock); 731 } 732 733 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4, 734 bool kill_route) 735 { 736 __be32 new_gw = icmp_hdr(skb)->un.gateway; 737 __be32 old_gw = ip_hdr(skb)->saddr; 738 struct net_device *dev = skb->dev; 739 struct in_device *in_dev; 740 struct fib_result res; 741 struct neighbour *n; 742 struct net *net; 743 744 switch (icmp_hdr(skb)->code & 7) { 745 case ICMP_REDIR_NET: 746 case ICMP_REDIR_NETTOS: 747 case ICMP_REDIR_HOST: 748 case ICMP_REDIR_HOSTTOS: 749 break; 750 751 default: 752 return; 753 } 754 755 if (rt->rt_gateway != old_gw) 756 return; 757 758 in_dev = __in_dev_get_rcu(dev); 759 if (!in_dev) 760 return; 761 762 net = dev_net(dev); 763 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) || 764 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) || 765 ipv4_is_zeronet(new_gw)) 766 goto reject_redirect; 767 768 if (!IN_DEV_SHARED_MEDIA(in_dev)) { 769 if (!inet_addr_onlink(in_dev, new_gw, old_gw)) 770 goto reject_redirect; 771 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) 772 goto reject_redirect; 773 } else { 774 if (inet_addr_type(net, new_gw) != RTN_UNICAST) 775 goto reject_redirect; 776 } 777 778 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw); 779 if (!n) 780 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev); 781 if (!IS_ERR(n)) { 782 if (!(n->nud_state & NUD_VALID)) { 783 neigh_event_send(n, NULL); 784 } else { 785 if (fib_lookup(net, fl4, &res, 0) == 0) { 786 struct fib_nh *nh = &FIB_RES_NH(res); 787 788 update_or_create_fnhe(nh, fl4->daddr, new_gw, 789 0, jiffies + ip_rt_gc_timeout); 790 } 791 if (kill_route) 792 rt->dst.obsolete = DST_OBSOLETE_KILL; 793 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n); 794 } 795 neigh_release(n); 796 } 797 return; 798 799 reject_redirect: 800 #ifdef CONFIG_IP_ROUTE_VERBOSE 801 if (IN_DEV_LOG_MARTIANS(in_dev)) { 802 const struct iphdr *iph = (const struct iphdr *) skb->data; 803 __be32 daddr = iph->daddr; 804 __be32 saddr = iph->saddr; 805 806 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n" 807 " Advised path = %pI4 -> %pI4\n", 808 &old_gw, dev->name, &new_gw, 809 &saddr, &daddr); 810 } 811 #endif 812 ; 813 } 814 815 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb) 816 { 817 struct rtable *rt; 818 struct flowi4 fl4; 819 const struct iphdr *iph = (const struct iphdr *) skb->data; 820 struct net *net = dev_net(skb->dev); 821 int oif = skb->dev->ifindex; 822 u8 tos = RT_TOS(iph->tos); 823 u8 prot = iph->protocol; 824 u32 mark = skb->mark; 825 826 rt = (struct rtable *) dst; 827 828 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0); 829 __ip_do_redirect(rt, skb, &fl4, true); 830 } 831 832 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst) 833 { 834 struct rtable *rt = (struct rtable *)dst; 835 struct dst_entry *ret = dst; 836 837 if (rt) { 838 if (dst->obsolete > 0) { 839 ip_rt_put(rt); 840 ret = NULL; 841 } else if ((rt->rt_flags & RTCF_REDIRECTED) || 842 rt->dst.expires) { 843 ip_rt_put(rt); 844 ret = NULL; 845 } 846 } 847 return ret; 848 } 849 850 /* 851 * Algorithm: 852 * 1. The first ip_rt_redirect_number redirects are sent 853 * with exponential backoff, then we stop sending them at all, 854 * assuming that the host ignores our redirects. 855 * 2. If we did not see packets requiring redirects 856 * during ip_rt_redirect_silence, we assume that the host 857 * forgot redirected route and start to send redirects again. 858 * 859 * This algorithm is much cheaper and more intelligent than dumb load limiting 860 * in icmp.c. 861 * 862 * NOTE. Do not forget to inhibit load limiting for redirects (redundant) 863 * and "frag. need" (breaks PMTU discovery) in icmp.c. 864 */ 865 866 void ip_rt_send_redirect(struct sk_buff *skb) 867 { 868 struct rtable *rt = skb_rtable(skb); 869 struct in_device *in_dev; 870 struct inet_peer *peer; 871 struct net *net; 872 int log_martians; 873 int vif; 874 875 rcu_read_lock(); 876 in_dev = __in_dev_get_rcu(rt->dst.dev); 877 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) { 878 rcu_read_unlock(); 879 return; 880 } 881 log_martians = IN_DEV_LOG_MARTIANS(in_dev); 882 vif = l3mdev_master_ifindex_rcu(rt->dst.dev); 883 rcu_read_unlock(); 884 885 net = dev_net(rt->dst.dev); 886 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1); 887 if (!peer) { 888 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, 889 rt_nexthop(rt, ip_hdr(skb)->daddr)); 890 return; 891 } 892 893 /* No redirected packets during ip_rt_redirect_silence; 894 * reset the algorithm. 895 */ 896 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) 897 peer->rate_tokens = 0; 898 899 /* Too many ignored redirects; do not send anything 900 * set dst.rate_last to the last seen redirected packet. 901 */ 902 if (peer->rate_tokens >= ip_rt_redirect_number) { 903 peer->rate_last = jiffies; 904 goto out_put_peer; 905 } 906 907 /* Check for load limit; set rate_last to the latest sent 908 * redirect. 909 */ 910 if (peer->rate_tokens == 0 || 911 time_after(jiffies, 912 (peer->rate_last + 913 (ip_rt_redirect_load << peer->rate_tokens)))) { 914 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr); 915 916 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw); 917 peer->rate_last = jiffies; 918 ++peer->rate_tokens; 919 #ifdef CONFIG_IP_ROUTE_VERBOSE 920 if (log_martians && 921 peer->rate_tokens == ip_rt_redirect_number) 922 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n", 923 &ip_hdr(skb)->saddr, inet_iif(skb), 924 &ip_hdr(skb)->daddr, &gw); 925 #endif 926 } 927 out_put_peer: 928 inet_putpeer(peer); 929 } 930 931 static int ip_error(struct sk_buff *skb) 932 { 933 struct in_device *in_dev = __in_dev_get_rcu(skb->dev); 934 struct rtable *rt = skb_rtable(skb); 935 struct inet_peer *peer; 936 unsigned long now; 937 struct net *net; 938 bool send; 939 int code; 940 941 /* IP on this device is disabled. */ 942 if (!in_dev) 943 goto out; 944 945 net = dev_net(rt->dst.dev); 946 if (!IN_DEV_FORWARD(in_dev)) { 947 switch (rt->dst.error) { 948 case EHOSTUNREACH: 949 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS); 950 break; 951 952 case ENETUNREACH: 953 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 954 break; 955 } 956 goto out; 957 } 958 959 switch (rt->dst.error) { 960 case EINVAL: 961 default: 962 goto out; 963 case EHOSTUNREACH: 964 code = ICMP_HOST_UNREACH; 965 break; 966 case ENETUNREACH: 967 code = ICMP_NET_UNREACH; 968 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 969 break; 970 case EACCES: 971 code = ICMP_PKT_FILTERED; 972 break; 973 } 974 975 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 976 l3mdev_master_ifindex(skb->dev), 1); 977 978 send = true; 979 if (peer) { 980 now = jiffies; 981 peer->rate_tokens += now - peer->rate_last; 982 if (peer->rate_tokens > ip_rt_error_burst) 983 peer->rate_tokens = ip_rt_error_burst; 984 peer->rate_last = now; 985 if (peer->rate_tokens >= ip_rt_error_cost) 986 peer->rate_tokens -= ip_rt_error_cost; 987 else 988 send = false; 989 inet_putpeer(peer); 990 } 991 if (send) 992 icmp_send(skb, ICMP_DEST_UNREACH, code, 0); 993 994 out: kfree_skb(skb); 995 return 0; 996 } 997 998 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu) 999 { 1000 struct dst_entry *dst = &rt->dst; 1001 struct fib_result res; 1002 1003 if (dst_metric_locked(dst, RTAX_MTU)) 1004 return; 1005 1006 if (ipv4_mtu(dst) < mtu) 1007 return; 1008 1009 if (mtu < ip_rt_min_pmtu) 1010 mtu = ip_rt_min_pmtu; 1011 1012 if (rt->rt_pmtu == mtu && 1013 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2)) 1014 return; 1015 1016 rcu_read_lock(); 1017 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) { 1018 struct fib_nh *nh = &FIB_RES_NH(res); 1019 1020 update_or_create_fnhe(nh, fl4->daddr, 0, mtu, 1021 jiffies + ip_rt_mtu_expires); 1022 } 1023 rcu_read_unlock(); 1024 } 1025 1026 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 1027 struct sk_buff *skb, u32 mtu) 1028 { 1029 struct rtable *rt = (struct rtable *) dst; 1030 struct flowi4 fl4; 1031 1032 ip_rt_build_flow_key(&fl4, sk, skb); 1033 __ip_rt_update_pmtu(rt, &fl4, mtu); 1034 } 1035 1036 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, 1037 int oif, u32 mark, u8 protocol, int flow_flags) 1038 { 1039 const struct iphdr *iph = (const struct iphdr *) skb->data; 1040 struct flowi4 fl4; 1041 struct rtable *rt; 1042 1043 if (!mark) 1044 mark = IP4_REPLY_MARK(net, skb->mark); 1045 1046 __build_flow_key(net, &fl4, NULL, iph, oif, 1047 RT_TOS(iph->tos), protocol, mark, flow_flags); 1048 rt = __ip_route_output_key(net, &fl4); 1049 if (!IS_ERR(rt)) { 1050 __ip_rt_update_pmtu(rt, &fl4, mtu); 1051 ip_rt_put(rt); 1052 } 1053 } 1054 EXPORT_SYMBOL_GPL(ipv4_update_pmtu); 1055 1056 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1057 { 1058 const struct iphdr *iph = (const struct iphdr *) skb->data; 1059 struct flowi4 fl4; 1060 struct rtable *rt; 1061 1062 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0); 1063 1064 if (!fl4.flowi4_mark) 1065 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark); 1066 1067 rt = __ip_route_output_key(sock_net(sk), &fl4); 1068 if (!IS_ERR(rt)) { 1069 __ip_rt_update_pmtu(rt, &fl4, mtu); 1070 ip_rt_put(rt); 1071 } 1072 } 1073 1074 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1075 { 1076 const struct iphdr *iph = (const struct iphdr *) skb->data; 1077 struct flowi4 fl4; 1078 struct rtable *rt; 1079 struct dst_entry *odst = NULL; 1080 bool new = false; 1081 struct net *net = sock_net(sk); 1082 1083 bh_lock_sock(sk); 1084 1085 if (!ip_sk_accept_pmtu(sk)) 1086 goto out; 1087 1088 odst = sk_dst_get(sk); 1089 1090 if (sock_owned_by_user(sk) || !odst) { 1091 __ipv4_sk_update_pmtu(skb, sk, mtu); 1092 goto out; 1093 } 1094 1095 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1096 1097 rt = (struct rtable *)odst; 1098 if (odst->obsolete && !odst->ops->check(odst, 0)) { 1099 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1100 if (IS_ERR(rt)) 1101 goto out; 1102 1103 new = true; 1104 } 1105 1106 __ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu); 1107 1108 if (!dst_check(&rt->dst, 0)) { 1109 if (new) 1110 dst_release(&rt->dst); 1111 1112 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1113 if (IS_ERR(rt)) 1114 goto out; 1115 1116 new = true; 1117 } 1118 1119 if (new) 1120 sk_dst_set(sk, &rt->dst); 1121 1122 out: 1123 bh_unlock_sock(sk); 1124 dst_release(odst); 1125 } 1126 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu); 1127 1128 void ipv4_redirect(struct sk_buff *skb, struct net *net, 1129 int oif, u32 mark, u8 protocol, int flow_flags) 1130 { 1131 const struct iphdr *iph = (const struct iphdr *) skb->data; 1132 struct flowi4 fl4; 1133 struct rtable *rt; 1134 1135 __build_flow_key(net, &fl4, NULL, iph, oif, 1136 RT_TOS(iph->tos), protocol, mark, flow_flags); 1137 rt = __ip_route_output_key(net, &fl4); 1138 if (!IS_ERR(rt)) { 1139 __ip_do_redirect(rt, skb, &fl4, false); 1140 ip_rt_put(rt); 1141 } 1142 } 1143 EXPORT_SYMBOL_GPL(ipv4_redirect); 1144 1145 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk) 1146 { 1147 const struct iphdr *iph = (const struct iphdr *) skb->data; 1148 struct flowi4 fl4; 1149 struct rtable *rt; 1150 struct net *net = sock_net(sk); 1151 1152 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1153 rt = __ip_route_output_key(net, &fl4); 1154 if (!IS_ERR(rt)) { 1155 __ip_do_redirect(rt, skb, &fl4, false); 1156 ip_rt_put(rt); 1157 } 1158 } 1159 EXPORT_SYMBOL_GPL(ipv4_sk_redirect); 1160 1161 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie) 1162 { 1163 struct rtable *rt = (struct rtable *) dst; 1164 1165 /* All IPV4 dsts are created with ->obsolete set to the value 1166 * DST_OBSOLETE_FORCE_CHK which forces validation calls down 1167 * into this function always. 1168 * 1169 * When a PMTU/redirect information update invalidates a route, 1170 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or 1171 * DST_OBSOLETE_DEAD by dst_free(). 1172 */ 1173 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt)) 1174 return NULL; 1175 return dst; 1176 } 1177 1178 static void ipv4_link_failure(struct sk_buff *skb) 1179 { 1180 struct rtable *rt; 1181 1182 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0); 1183 1184 rt = skb_rtable(skb); 1185 if (rt) 1186 dst_set_expires(&rt->dst, 0); 1187 } 1188 1189 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb) 1190 { 1191 pr_debug("%s: %pI4 -> %pI4, %s\n", 1192 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1193 skb->dev ? skb->dev->name : "?"); 1194 kfree_skb(skb); 1195 WARN_ON(1); 1196 return 0; 1197 } 1198 1199 /* 1200 We do not cache source address of outgoing interface, 1201 because it is used only by IP RR, TS and SRR options, 1202 so that it out of fast path. 1203 1204 BTW remember: "addr" is allowed to be not aligned 1205 in IP options! 1206 */ 1207 1208 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt) 1209 { 1210 __be32 src; 1211 1212 if (rt_is_output_route(rt)) 1213 src = ip_hdr(skb)->saddr; 1214 else { 1215 struct fib_result res; 1216 struct flowi4 fl4; 1217 struct iphdr *iph; 1218 1219 iph = ip_hdr(skb); 1220 1221 memset(&fl4, 0, sizeof(fl4)); 1222 fl4.daddr = iph->daddr; 1223 fl4.saddr = iph->saddr; 1224 fl4.flowi4_tos = RT_TOS(iph->tos); 1225 fl4.flowi4_oif = rt->dst.dev->ifindex; 1226 fl4.flowi4_iif = skb->dev->ifindex; 1227 fl4.flowi4_mark = skb->mark; 1228 1229 rcu_read_lock(); 1230 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0) 1231 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res); 1232 else 1233 src = inet_select_addr(rt->dst.dev, 1234 rt_nexthop(rt, iph->daddr), 1235 RT_SCOPE_UNIVERSE); 1236 rcu_read_unlock(); 1237 } 1238 memcpy(addr, &src, 4); 1239 } 1240 1241 #ifdef CONFIG_IP_ROUTE_CLASSID 1242 static void set_class_tag(struct rtable *rt, u32 tag) 1243 { 1244 if (!(rt->dst.tclassid & 0xFFFF)) 1245 rt->dst.tclassid |= tag & 0xFFFF; 1246 if (!(rt->dst.tclassid & 0xFFFF0000)) 1247 rt->dst.tclassid |= tag & 0xFFFF0000; 1248 } 1249 #endif 1250 1251 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1252 { 1253 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr); 1254 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size, 1255 ip_rt_min_advmss); 1256 1257 return min(advmss, IPV4_MAX_PMTU - header_size); 1258 } 1259 1260 static unsigned int ipv4_mtu(const struct dst_entry *dst) 1261 { 1262 const struct rtable *rt = (const struct rtable *) dst; 1263 unsigned int mtu = rt->rt_pmtu; 1264 1265 if (!mtu || time_after_eq(jiffies, rt->dst.expires)) 1266 mtu = dst_metric_raw(dst, RTAX_MTU); 1267 1268 if (mtu) 1269 return mtu; 1270 1271 mtu = READ_ONCE(dst->dev->mtu); 1272 1273 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) { 1274 if (rt->rt_uses_gateway && mtu > 576) 1275 mtu = 576; 1276 } 1277 1278 mtu = min_t(unsigned int, mtu, IP_MAX_MTU); 1279 1280 return mtu - lwtunnel_headroom(dst->lwtstate, mtu); 1281 } 1282 1283 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr) 1284 { 1285 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions); 1286 struct fib_nh_exception *fnhe; 1287 u32 hval; 1288 1289 if (!hash) 1290 return NULL; 1291 1292 hval = fnhe_hashfun(daddr); 1293 1294 for (fnhe = rcu_dereference(hash[hval].chain); fnhe; 1295 fnhe = rcu_dereference(fnhe->fnhe_next)) { 1296 if (fnhe->fnhe_daddr == daddr) 1297 return fnhe; 1298 } 1299 return NULL; 1300 } 1301 1302 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, 1303 __be32 daddr, const bool do_cache) 1304 { 1305 bool ret = false; 1306 1307 spin_lock_bh(&fnhe_lock); 1308 1309 if (daddr == fnhe->fnhe_daddr) { 1310 struct rtable __rcu **porig; 1311 struct rtable *orig; 1312 int genid = fnhe_genid(dev_net(rt->dst.dev)); 1313 1314 if (rt_is_input_route(rt)) 1315 porig = &fnhe->fnhe_rth_input; 1316 else 1317 porig = &fnhe->fnhe_rth_output; 1318 orig = rcu_dereference(*porig); 1319 1320 if (fnhe->fnhe_genid != genid) { 1321 fnhe->fnhe_genid = genid; 1322 fnhe->fnhe_gw = 0; 1323 fnhe->fnhe_pmtu = 0; 1324 fnhe->fnhe_expires = 0; 1325 fnhe_flush_routes(fnhe); 1326 orig = NULL; 1327 } 1328 fill_route_from_fnhe(rt, fnhe); 1329 if (!rt->rt_gateway) 1330 rt->rt_gateway = daddr; 1331 1332 if (do_cache) { 1333 dst_hold(&rt->dst); 1334 rcu_assign_pointer(*porig, rt); 1335 if (orig) { 1336 dst_dev_put(&orig->dst); 1337 dst_release(&orig->dst); 1338 } 1339 ret = true; 1340 } 1341 1342 fnhe->fnhe_stamp = jiffies; 1343 } 1344 spin_unlock_bh(&fnhe_lock); 1345 1346 return ret; 1347 } 1348 1349 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt) 1350 { 1351 struct rtable *orig, *prev, **p; 1352 bool ret = true; 1353 1354 if (rt_is_input_route(rt)) { 1355 p = (struct rtable **)&nh->nh_rth_input; 1356 } else { 1357 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output); 1358 } 1359 orig = *p; 1360 1361 /* hold dst before doing cmpxchg() to avoid race condition 1362 * on this dst 1363 */ 1364 dst_hold(&rt->dst); 1365 prev = cmpxchg(p, orig, rt); 1366 if (prev == orig) { 1367 if (orig) { 1368 dst_dev_put(&orig->dst); 1369 dst_release(&orig->dst); 1370 } 1371 } else { 1372 dst_release(&rt->dst); 1373 ret = false; 1374 } 1375 1376 return ret; 1377 } 1378 1379 struct uncached_list { 1380 spinlock_t lock; 1381 struct list_head head; 1382 }; 1383 1384 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list); 1385 1386 static void rt_add_uncached_list(struct rtable *rt) 1387 { 1388 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list); 1389 1390 rt->rt_uncached_list = ul; 1391 1392 spin_lock_bh(&ul->lock); 1393 list_add_tail(&rt->rt_uncached, &ul->head); 1394 spin_unlock_bh(&ul->lock); 1395 } 1396 1397 static void ipv4_dst_destroy(struct dst_entry *dst) 1398 { 1399 struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst); 1400 struct rtable *rt = (struct rtable *) dst; 1401 1402 if (p != &dst_default_metrics && refcount_dec_and_test(&p->refcnt)) 1403 kfree(p); 1404 1405 if (!list_empty(&rt->rt_uncached)) { 1406 struct uncached_list *ul = rt->rt_uncached_list; 1407 1408 spin_lock_bh(&ul->lock); 1409 list_del(&rt->rt_uncached); 1410 spin_unlock_bh(&ul->lock); 1411 } 1412 } 1413 1414 void rt_flush_dev(struct net_device *dev) 1415 { 1416 struct net *net = dev_net(dev); 1417 struct rtable *rt; 1418 int cpu; 1419 1420 for_each_possible_cpu(cpu) { 1421 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 1422 1423 spin_lock_bh(&ul->lock); 1424 list_for_each_entry(rt, &ul->head, rt_uncached) { 1425 if (rt->dst.dev != dev) 1426 continue; 1427 rt->dst.dev = net->loopback_dev; 1428 dev_hold(rt->dst.dev); 1429 dev_put(dev); 1430 } 1431 spin_unlock_bh(&ul->lock); 1432 } 1433 } 1434 1435 static bool rt_cache_valid(const struct rtable *rt) 1436 { 1437 return rt && 1438 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK && 1439 !rt_is_expired(rt); 1440 } 1441 1442 static void rt_set_nexthop(struct rtable *rt, __be32 daddr, 1443 const struct fib_result *res, 1444 struct fib_nh_exception *fnhe, 1445 struct fib_info *fi, u16 type, u32 itag, 1446 const bool do_cache) 1447 { 1448 bool cached = false; 1449 1450 if (fi) { 1451 struct fib_nh *nh = &FIB_RES_NH(*res); 1452 1453 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) { 1454 rt->rt_gateway = nh->nh_gw; 1455 rt->rt_uses_gateway = 1; 1456 } 1457 dst_init_metrics(&rt->dst, fi->fib_metrics->metrics, true); 1458 if (fi->fib_metrics != &dst_default_metrics) { 1459 rt->dst._metrics |= DST_METRICS_REFCOUNTED; 1460 refcount_inc(&fi->fib_metrics->refcnt); 1461 } 1462 #ifdef CONFIG_IP_ROUTE_CLASSID 1463 rt->dst.tclassid = nh->nh_tclassid; 1464 #endif 1465 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate); 1466 if (unlikely(fnhe)) 1467 cached = rt_bind_exception(rt, fnhe, daddr, do_cache); 1468 else if (do_cache) 1469 cached = rt_cache_route(nh, rt); 1470 if (unlikely(!cached)) { 1471 /* Routes we intend to cache in nexthop exception or 1472 * FIB nexthop have the DST_NOCACHE bit clear. 1473 * However, if we are unsuccessful at storing this 1474 * route into the cache we really need to set it. 1475 */ 1476 if (!rt->rt_gateway) 1477 rt->rt_gateway = daddr; 1478 rt_add_uncached_list(rt); 1479 } 1480 } else 1481 rt_add_uncached_list(rt); 1482 1483 #ifdef CONFIG_IP_ROUTE_CLASSID 1484 #ifdef CONFIG_IP_MULTIPLE_TABLES 1485 set_class_tag(rt, res->tclassid); 1486 #endif 1487 set_class_tag(rt, itag); 1488 #endif 1489 } 1490 1491 struct rtable *rt_dst_alloc(struct net_device *dev, 1492 unsigned int flags, u16 type, 1493 bool nopolicy, bool noxfrm, bool will_cache) 1494 { 1495 struct rtable *rt; 1496 1497 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1498 (will_cache ? 0 : DST_HOST) | 1499 (nopolicy ? DST_NOPOLICY : 0) | 1500 (noxfrm ? DST_NOXFRM : 0)); 1501 1502 if (rt) { 1503 rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1504 rt->rt_flags = flags; 1505 rt->rt_type = type; 1506 rt->rt_is_input = 0; 1507 rt->rt_iif = 0; 1508 rt->rt_pmtu = 0; 1509 rt->rt_gateway = 0; 1510 rt->rt_uses_gateway = 0; 1511 rt->rt_table_id = 0; 1512 INIT_LIST_HEAD(&rt->rt_uncached); 1513 1514 rt->dst.output = ip_output; 1515 if (flags & RTCF_LOCAL) 1516 rt->dst.input = ip_local_deliver; 1517 } 1518 1519 return rt; 1520 } 1521 EXPORT_SYMBOL(rt_dst_alloc); 1522 1523 /* called in rcu_read_lock() section */ 1524 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1525 u8 tos, struct net_device *dev, 1526 struct in_device *in_dev, u32 *itag) 1527 { 1528 int err; 1529 1530 /* Primary sanity checks. */ 1531 if (!in_dev) 1532 return -EINVAL; 1533 1534 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) || 1535 skb->protocol != htons(ETH_P_IP)) 1536 return -EINVAL; 1537 1538 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev)) 1539 return -EINVAL; 1540 1541 if (ipv4_is_zeronet(saddr)) { 1542 if (!ipv4_is_local_multicast(daddr)) 1543 return -EINVAL; 1544 } else { 1545 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 1546 in_dev, itag); 1547 if (err < 0) 1548 return err; 1549 } 1550 return 0; 1551 } 1552 1553 /* called in rcu_read_lock() section */ 1554 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1555 u8 tos, struct net_device *dev, int our) 1556 { 1557 struct in_device *in_dev = __in_dev_get_rcu(dev); 1558 unsigned int flags = RTCF_MULTICAST; 1559 struct rtable *rth; 1560 u32 itag = 0; 1561 int err; 1562 1563 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag); 1564 if (err) 1565 return err; 1566 1567 if (our) 1568 flags |= RTCF_LOCAL; 1569 1570 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST, 1571 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false); 1572 if (!rth) 1573 return -ENOBUFS; 1574 1575 #ifdef CONFIG_IP_ROUTE_CLASSID 1576 rth->dst.tclassid = itag; 1577 #endif 1578 rth->dst.output = ip_rt_bug; 1579 rth->rt_is_input= 1; 1580 1581 #ifdef CONFIG_IP_MROUTE 1582 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1583 rth->dst.input = ip_mr_input; 1584 #endif 1585 RT_CACHE_STAT_INC(in_slow_mc); 1586 1587 skb_dst_set(skb, &rth->dst); 1588 return 0; 1589 } 1590 1591 1592 static void ip_handle_martian_source(struct net_device *dev, 1593 struct in_device *in_dev, 1594 struct sk_buff *skb, 1595 __be32 daddr, 1596 __be32 saddr) 1597 { 1598 RT_CACHE_STAT_INC(in_martian_src); 1599 #ifdef CONFIG_IP_ROUTE_VERBOSE 1600 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1601 /* 1602 * RFC1812 recommendation, if source is martian, 1603 * the only hint is MAC header. 1604 */ 1605 pr_warn("martian source %pI4 from %pI4, on dev %s\n", 1606 &daddr, &saddr, dev->name); 1607 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1608 print_hex_dump(KERN_WARNING, "ll header: ", 1609 DUMP_PREFIX_OFFSET, 16, 1, 1610 skb_mac_header(skb), 1611 dev->hard_header_len, true); 1612 } 1613 } 1614 #endif 1615 } 1616 1617 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr) 1618 { 1619 struct fnhe_hash_bucket *hash; 1620 struct fib_nh_exception *fnhe, __rcu **fnhe_p; 1621 u32 hval = fnhe_hashfun(daddr); 1622 1623 spin_lock_bh(&fnhe_lock); 1624 1625 hash = rcu_dereference_protected(nh->nh_exceptions, 1626 lockdep_is_held(&fnhe_lock)); 1627 hash += hval; 1628 1629 fnhe_p = &hash->chain; 1630 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock)); 1631 while (fnhe) { 1632 if (fnhe->fnhe_daddr == daddr) { 1633 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected( 1634 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock))); 1635 fnhe_flush_routes(fnhe); 1636 kfree_rcu(fnhe, rcu); 1637 break; 1638 } 1639 fnhe_p = &fnhe->fnhe_next; 1640 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1641 lockdep_is_held(&fnhe_lock)); 1642 } 1643 1644 spin_unlock_bh(&fnhe_lock); 1645 } 1646 1647 static void set_lwt_redirect(struct rtable *rth) 1648 { 1649 if (lwtunnel_output_redirect(rth->dst.lwtstate)) { 1650 rth->dst.lwtstate->orig_output = rth->dst.output; 1651 rth->dst.output = lwtunnel_output; 1652 } 1653 1654 if (lwtunnel_input_redirect(rth->dst.lwtstate)) { 1655 rth->dst.lwtstate->orig_input = rth->dst.input; 1656 rth->dst.input = lwtunnel_input; 1657 } 1658 } 1659 1660 /* called in rcu_read_lock() section */ 1661 static int __mkroute_input(struct sk_buff *skb, 1662 const struct fib_result *res, 1663 struct in_device *in_dev, 1664 __be32 daddr, __be32 saddr, u32 tos) 1665 { 1666 struct fib_nh_exception *fnhe; 1667 struct rtable *rth; 1668 int err; 1669 struct in_device *out_dev; 1670 bool do_cache; 1671 u32 itag = 0; 1672 1673 /* get a working reference to the output device */ 1674 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res)); 1675 if (!out_dev) { 1676 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n"); 1677 return -EINVAL; 1678 } 1679 1680 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res), 1681 in_dev->dev, in_dev, &itag); 1682 if (err < 0) { 1683 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1684 saddr); 1685 1686 goto cleanup; 1687 } 1688 1689 do_cache = res->fi && !itag; 1690 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) && 1691 skb->protocol == htons(ETH_P_IP) && 1692 (IN_DEV_SHARED_MEDIA(out_dev) || 1693 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) 1694 IPCB(skb)->flags |= IPSKB_DOREDIRECT; 1695 1696 if (skb->protocol != htons(ETH_P_IP)) { 1697 /* Not IP (i.e. ARP). Do not create route, if it is 1698 * invalid for proxy arp. DNAT routes are always valid. 1699 * 1700 * Proxy arp feature have been extended to allow, ARP 1701 * replies back to the same interface, to support 1702 * Private VLAN switch technologies. See arp.c. 1703 */ 1704 if (out_dev == in_dev && 1705 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 1706 err = -EINVAL; 1707 goto cleanup; 1708 } 1709 } 1710 1711 fnhe = find_exception(&FIB_RES_NH(*res), daddr); 1712 if (do_cache) { 1713 if (fnhe) { 1714 rth = rcu_dereference(fnhe->fnhe_rth_input); 1715 if (rth && rth->dst.expires && 1716 time_after(jiffies, rth->dst.expires)) { 1717 ip_del_fnhe(&FIB_RES_NH(*res), daddr); 1718 fnhe = NULL; 1719 } else { 1720 goto rt_cache; 1721 } 1722 } 1723 1724 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input); 1725 1726 rt_cache: 1727 if (rt_cache_valid(rth)) { 1728 skb_dst_set_noref(skb, &rth->dst); 1729 goto out; 1730 } 1731 } 1732 1733 rth = rt_dst_alloc(out_dev->dev, 0, res->type, 1734 IN_DEV_CONF_GET(in_dev, NOPOLICY), 1735 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache); 1736 if (!rth) { 1737 err = -ENOBUFS; 1738 goto cleanup; 1739 } 1740 1741 rth->rt_is_input = 1; 1742 if (res->table) 1743 rth->rt_table_id = res->table->tb_id; 1744 RT_CACHE_STAT_INC(in_slow_tot); 1745 1746 rth->dst.input = ip_forward; 1747 1748 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag, 1749 do_cache); 1750 set_lwt_redirect(rth); 1751 skb_dst_set(skb, &rth->dst); 1752 out: 1753 err = 0; 1754 cleanup: 1755 return err; 1756 } 1757 1758 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1759 /* To make ICMP packets follow the right flow, the multipath hash is 1760 * calculated from the inner IP addresses. 1761 */ 1762 static void ip_multipath_l3_keys(const struct sk_buff *skb, 1763 struct flow_keys *hash_keys) 1764 { 1765 const struct iphdr *outer_iph = ip_hdr(skb); 1766 const struct iphdr *inner_iph; 1767 const struct icmphdr *icmph; 1768 struct iphdr _inner_iph; 1769 struct icmphdr _icmph; 1770 1771 hash_keys->addrs.v4addrs.src = outer_iph->saddr; 1772 hash_keys->addrs.v4addrs.dst = outer_iph->daddr; 1773 if (likely(outer_iph->protocol != IPPROTO_ICMP)) 1774 return; 1775 1776 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0)) 1777 return; 1778 1779 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph), 1780 &_icmph); 1781 if (!icmph) 1782 return; 1783 1784 if (icmph->type != ICMP_DEST_UNREACH && 1785 icmph->type != ICMP_REDIRECT && 1786 icmph->type != ICMP_TIME_EXCEEDED && 1787 icmph->type != ICMP_PARAMETERPROB) 1788 return; 1789 1790 inner_iph = skb_header_pointer(skb, 1791 outer_iph->ihl * 4 + sizeof(_icmph), 1792 sizeof(_inner_iph), &_inner_iph); 1793 if (!inner_iph) 1794 return; 1795 hash_keys->addrs.v4addrs.src = inner_iph->saddr; 1796 hash_keys->addrs.v4addrs.dst = inner_iph->daddr; 1797 } 1798 1799 /* if skb is set it will be used and fl4 can be NULL */ 1800 int fib_multipath_hash(const struct fib_info *fi, const struct flowi4 *fl4, 1801 const struct sk_buff *skb) 1802 { 1803 struct net *net = fi->fib_net; 1804 struct flow_keys hash_keys; 1805 u32 mhash; 1806 1807 switch (net->ipv4.sysctl_fib_multipath_hash_policy) { 1808 case 0: 1809 memset(&hash_keys, 0, sizeof(hash_keys)); 1810 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1811 if (skb) { 1812 ip_multipath_l3_keys(skb, &hash_keys); 1813 } else { 1814 hash_keys.addrs.v4addrs.src = fl4->saddr; 1815 hash_keys.addrs.v4addrs.dst = fl4->daddr; 1816 } 1817 break; 1818 case 1: 1819 /* skb is currently provided only when forwarding */ 1820 if (skb) { 1821 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 1822 struct flow_keys keys; 1823 1824 /* short-circuit if we already have L4 hash present */ 1825 if (skb->l4_hash) 1826 return skb_get_hash_raw(skb) >> 1; 1827 memset(&hash_keys, 0, sizeof(hash_keys)); 1828 skb_flow_dissect_flow_keys(skb, &keys, flag); 1829 1830 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1831 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 1832 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 1833 hash_keys.ports.src = keys.ports.src; 1834 hash_keys.ports.dst = keys.ports.dst; 1835 hash_keys.basic.ip_proto = keys.basic.ip_proto; 1836 } else { 1837 memset(&hash_keys, 0, sizeof(hash_keys)); 1838 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1839 hash_keys.addrs.v4addrs.src = fl4->saddr; 1840 hash_keys.addrs.v4addrs.dst = fl4->daddr; 1841 hash_keys.ports.src = fl4->fl4_sport; 1842 hash_keys.ports.dst = fl4->fl4_dport; 1843 hash_keys.basic.ip_proto = fl4->flowi4_proto; 1844 } 1845 break; 1846 } 1847 mhash = flow_hash_from_keys(&hash_keys); 1848 1849 return mhash >> 1; 1850 } 1851 EXPORT_SYMBOL_GPL(fib_multipath_hash); 1852 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 1853 1854 static int ip_mkroute_input(struct sk_buff *skb, 1855 struct fib_result *res, 1856 struct in_device *in_dev, 1857 __be32 daddr, __be32 saddr, u32 tos) 1858 { 1859 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1860 if (res->fi && res->fi->fib_nhs > 1) { 1861 int h = fib_multipath_hash(res->fi, NULL, skb); 1862 1863 fib_select_multipath(res, h); 1864 } 1865 #endif 1866 1867 /* create a routing cache entry */ 1868 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos); 1869 } 1870 1871 /* 1872 * NOTE. We drop all the packets that has local source 1873 * addresses, because every properly looped back packet 1874 * must have correct destination already attached by output routine. 1875 * 1876 * Such approach solves two big problems: 1877 * 1. Not simplex devices are handled properly. 1878 * 2. IP spoofing attempts are filtered with 100% of guarantee. 1879 * called with rcu_read_lock() 1880 */ 1881 1882 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1883 u8 tos, struct net_device *dev, 1884 struct fib_result *res) 1885 { 1886 struct in_device *in_dev = __in_dev_get_rcu(dev); 1887 struct ip_tunnel_info *tun_info; 1888 struct flowi4 fl4; 1889 unsigned int flags = 0; 1890 u32 itag = 0; 1891 struct rtable *rth; 1892 int err = -EINVAL; 1893 struct net *net = dev_net(dev); 1894 bool do_cache; 1895 1896 /* IP on this device is disabled. */ 1897 1898 if (!in_dev) 1899 goto out; 1900 1901 /* Check for the most weird martians, which can be not detected 1902 by fib_lookup. 1903 */ 1904 1905 tun_info = skb_tunnel_info(skb); 1906 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX)) 1907 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id; 1908 else 1909 fl4.flowi4_tun_key.tun_id = 0; 1910 skb_dst_drop(skb); 1911 1912 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 1913 goto martian_source; 1914 1915 res->fi = NULL; 1916 res->table = NULL; 1917 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 1918 goto brd_input; 1919 1920 /* Accept zero addresses only to limited broadcast; 1921 * I even do not know to fix it or not. Waiting for complains :-) 1922 */ 1923 if (ipv4_is_zeronet(saddr)) 1924 goto martian_source; 1925 1926 if (ipv4_is_zeronet(daddr)) 1927 goto martian_destination; 1928 1929 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(), 1930 * and call it once if daddr or/and saddr are loopback addresses 1931 */ 1932 if (ipv4_is_loopback(daddr)) { 1933 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 1934 goto martian_destination; 1935 } else if (ipv4_is_loopback(saddr)) { 1936 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 1937 goto martian_source; 1938 } 1939 1940 /* 1941 * Now we are ready to route packet. 1942 */ 1943 fl4.flowi4_oif = 0; 1944 fl4.flowi4_iif = dev->ifindex; 1945 fl4.flowi4_mark = skb->mark; 1946 fl4.flowi4_tos = tos; 1947 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 1948 fl4.flowi4_flags = 0; 1949 fl4.daddr = daddr; 1950 fl4.saddr = saddr; 1951 fl4.flowi4_uid = sock_net_uid(net, NULL); 1952 err = fib_lookup(net, &fl4, res, 0); 1953 if (err != 0) { 1954 if (!IN_DEV_FORWARD(in_dev)) 1955 err = -EHOSTUNREACH; 1956 goto no_route; 1957 } 1958 1959 if (res->type == RTN_BROADCAST) 1960 goto brd_input; 1961 1962 if (res->type == RTN_LOCAL) { 1963 err = fib_validate_source(skb, saddr, daddr, tos, 1964 0, dev, in_dev, &itag); 1965 if (err < 0) 1966 goto martian_source; 1967 goto local_input; 1968 } 1969 1970 if (!IN_DEV_FORWARD(in_dev)) { 1971 err = -EHOSTUNREACH; 1972 goto no_route; 1973 } 1974 if (res->type != RTN_UNICAST) 1975 goto martian_destination; 1976 1977 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos); 1978 out: return err; 1979 1980 brd_input: 1981 if (skb->protocol != htons(ETH_P_IP)) 1982 goto e_inval; 1983 1984 if (!ipv4_is_zeronet(saddr)) { 1985 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 1986 in_dev, &itag); 1987 if (err < 0) 1988 goto martian_source; 1989 } 1990 flags |= RTCF_BROADCAST; 1991 res->type = RTN_BROADCAST; 1992 RT_CACHE_STAT_INC(in_brd); 1993 1994 local_input: 1995 do_cache = false; 1996 if (res->fi) { 1997 if (!itag) { 1998 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input); 1999 if (rt_cache_valid(rth)) { 2000 skb_dst_set_noref(skb, &rth->dst); 2001 err = 0; 2002 goto out; 2003 } 2004 do_cache = true; 2005 } 2006 } 2007 2008 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev, 2009 flags | RTCF_LOCAL, res->type, 2010 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache); 2011 if (!rth) 2012 goto e_nobufs; 2013 2014 rth->dst.output= ip_rt_bug; 2015 #ifdef CONFIG_IP_ROUTE_CLASSID 2016 rth->dst.tclassid = itag; 2017 #endif 2018 rth->rt_is_input = 1; 2019 if (res->table) 2020 rth->rt_table_id = res->table->tb_id; 2021 2022 RT_CACHE_STAT_INC(in_slow_tot); 2023 if (res->type == RTN_UNREACHABLE) { 2024 rth->dst.input= ip_error; 2025 rth->dst.error= -err; 2026 rth->rt_flags &= ~RTCF_LOCAL; 2027 } 2028 2029 if (do_cache) { 2030 struct fib_nh *nh = &FIB_RES_NH(*res); 2031 2032 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate); 2033 if (lwtunnel_input_redirect(rth->dst.lwtstate)) { 2034 WARN_ON(rth->dst.input == lwtunnel_input); 2035 rth->dst.lwtstate->orig_input = rth->dst.input; 2036 rth->dst.input = lwtunnel_input; 2037 } 2038 2039 if (unlikely(!rt_cache_route(nh, rth))) 2040 rt_add_uncached_list(rth); 2041 } 2042 skb_dst_set(skb, &rth->dst); 2043 err = 0; 2044 goto out; 2045 2046 no_route: 2047 RT_CACHE_STAT_INC(in_no_route); 2048 res->type = RTN_UNREACHABLE; 2049 res->fi = NULL; 2050 res->table = NULL; 2051 goto local_input; 2052 2053 /* 2054 * Do not cache martian addresses: they should be logged (RFC1812) 2055 */ 2056 martian_destination: 2057 RT_CACHE_STAT_INC(in_martian_dst); 2058 #ifdef CONFIG_IP_ROUTE_VERBOSE 2059 if (IN_DEV_LOG_MARTIANS(in_dev)) 2060 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n", 2061 &daddr, &saddr, dev->name); 2062 #endif 2063 2064 e_inval: 2065 err = -EINVAL; 2066 goto out; 2067 2068 e_nobufs: 2069 err = -ENOBUFS; 2070 goto out; 2071 2072 martian_source: 2073 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2074 goto out; 2075 } 2076 2077 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2078 u8 tos, struct net_device *dev) 2079 { 2080 struct fib_result res; 2081 int err; 2082 2083 tos &= IPTOS_RT_MASK; 2084 rcu_read_lock(); 2085 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res); 2086 rcu_read_unlock(); 2087 2088 return err; 2089 } 2090 EXPORT_SYMBOL(ip_route_input_noref); 2091 2092 /* called with rcu_read_lock held */ 2093 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2094 u8 tos, struct net_device *dev, struct fib_result *res) 2095 { 2096 /* Multicast recognition logic is moved from route cache to here. 2097 The problem was that too many Ethernet cards have broken/missing 2098 hardware multicast filters :-( As result the host on multicasting 2099 network acquires a lot of useless route cache entries, sort of 2100 SDR messages from all the world. Now we try to get rid of them. 2101 Really, provided software IP multicast filter is organized 2102 reasonably (at least, hashed), it does not result in a slowdown 2103 comparing with route cache reject entries. 2104 Note, that multicast routers are not affected, because 2105 route cache entry is created eventually. 2106 */ 2107 if (ipv4_is_multicast(daddr)) { 2108 struct in_device *in_dev = __in_dev_get_rcu(dev); 2109 int our = 0; 2110 int err = -EINVAL; 2111 2112 if (in_dev) 2113 our = ip_check_mc_rcu(in_dev, daddr, saddr, 2114 ip_hdr(skb)->protocol); 2115 2116 /* check l3 master if no match yet */ 2117 if ((!in_dev || !our) && netif_is_l3_slave(dev)) { 2118 struct in_device *l3_in_dev; 2119 2120 l3_in_dev = __in_dev_get_rcu(skb->dev); 2121 if (l3_in_dev) 2122 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr, 2123 ip_hdr(skb)->protocol); 2124 } 2125 2126 if (our 2127 #ifdef CONFIG_IP_MROUTE 2128 || 2129 (!ipv4_is_local_multicast(daddr) && 2130 IN_DEV_MFORWARD(in_dev)) 2131 #endif 2132 ) { 2133 err = ip_route_input_mc(skb, daddr, saddr, 2134 tos, dev, our); 2135 } 2136 return err; 2137 } 2138 2139 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res); 2140 } 2141 2142 /* called with rcu_read_lock() */ 2143 static struct rtable *__mkroute_output(const struct fib_result *res, 2144 const struct flowi4 *fl4, int orig_oif, 2145 struct net_device *dev_out, 2146 unsigned int flags) 2147 { 2148 struct fib_info *fi = res->fi; 2149 struct fib_nh_exception *fnhe; 2150 struct in_device *in_dev; 2151 u16 type = res->type; 2152 struct rtable *rth; 2153 bool do_cache; 2154 2155 in_dev = __in_dev_get_rcu(dev_out); 2156 if (!in_dev) 2157 return ERR_PTR(-EINVAL); 2158 2159 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 2160 if (ipv4_is_loopback(fl4->saddr) && 2161 !(dev_out->flags & IFF_LOOPBACK) && 2162 !netif_is_l3_master(dev_out)) 2163 return ERR_PTR(-EINVAL); 2164 2165 if (ipv4_is_lbcast(fl4->daddr)) 2166 type = RTN_BROADCAST; 2167 else if (ipv4_is_multicast(fl4->daddr)) 2168 type = RTN_MULTICAST; 2169 else if (ipv4_is_zeronet(fl4->daddr)) 2170 return ERR_PTR(-EINVAL); 2171 2172 if (dev_out->flags & IFF_LOOPBACK) 2173 flags |= RTCF_LOCAL; 2174 2175 do_cache = true; 2176 if (type == RTN_BROADCAST) { 2177 flags |= RTCF_BROADCAST | RTCF_LOCAL; 2178 fi = NULL; 2179 } else if (type == RTN_MULTICAST) { 2180 flags |= RTCF_MULTICAST | RTCF_LOCAL; 2181 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr, 2182 fl4->flowi4_proto)) 2183 flags &= ~RTCF_LOCAL; 2184 else 2185 do_cache = false; 2186 /* If multicast route do not exist use 2187 * default one, but do not gateway in this case. 2188 * Yes, it is hack. 2189 */ 2190 if (fi && res->prefixlen < 4) 2191 fi = NULL; 2192 } else if ((type == RTN_LOCAL) && (orig_oif != 0) && 2193 (orig_oif != dev_out->ifindex)) { 2194 /* For local routes that require a particular output interface 2195 * we do not want to cache the result. Caching the result 2196 * causes incorrect behaviour when there are multiple source 2197 * addresses on the interface, the end result being that if the 2198 * intended recipient is waiting on that interface for the 2199 * packet he won't receive it because it will be delivered on 2200 * the loopback interface and the IP_PKTINFO ipi_ifindex will 2201 * be set to the loopback interface as well. 2202 */ 2203 fi = NULL; 2204 } 2205 2206 fnhe = NULL; 2207 do_cache &= fi != NULL; 2208 if (do_cache) { 2209 struct rtable __rcu **prth; 2210 struct fib_nh *nh = &FIB_RES_NH(*res); 2211 2212 fnhe = find_exception(nh, fl4->daddr); 2213 if (fnhe) { 2214 prth = &fnhe->fnhe_rth_output; 2215 rth = rcu_dereference(*prth); 2216 if (rth && rth->dst.expires && 2217 time_after(jiffies, rth->dst.expires)) { 2218 ip_del_fnhe(nh, fl4->daddr); 2219 fnhe = NULL; 2220 } else { 2221 goto rt_cache; 2222 } 2223 } 2224 2225 if (unlikely(fl4->flowi4_flags & 2226 FLOWI_FLAG_KNOWN_NH && 2227 !(nh->nh_gw && 2228 nh->nh_scope == RT_SCOPE_LINK))) { 2229 do_cache = false; 2230 goto add; 2231 } 2232 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output); 2233 rth = rcu_dereference(*prth); 2234 2235 rt_cache: 2236 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst)) 2237 return rth; 2238 } 2239 2240 add: 2241 rth = rt_dst_alloc(dev_out, flags, type, 2242 IN_DEV_CONF_GET(in_dev, NOPOLICY), 2243 IN_DEV_CONF_GET(in_dev, NOXFRM), 2244 do_cache); 2245 if (!rth) 2246 return ERR_PTR(-ENOBUFS); 2247 2248 rth->rt_iif = orig_oif; 2249 if (res->table) 2250 rth->rt_table_id = res->table->tb_id; 2251 2252 RT_CACHE_STAT_INC(out_slow_tot); 2253 2254 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 2255 if (flags & RTCF_LOCAL && 2256 !(dev_out->flags & IFF_LOOPBACK)) { 2257 rth->dst.output = ip_mc_output; 2258 RT_CACHE_STAT_INC(out_slow_mc); 2259 } 2260 #ifdef CONFIG_IP_MROUTE 2261 if (type == RTN_MULTICAST) { 2262 if (IN_DEV_MFORWARD(in_dev) && 2263 !ipv4_is_local_multicast(fl4->daddr)) { 2264 rth->dst.input = ip_mr_input; 2265 rth->dst.output = ip_mc_output; 2266 } 2267 } 2268 #endif 2269 } 2270 2271 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache); 2272 set_lwt_redirect(rth); 2273 2274 return rth; 2275 } 2276 2277 /* 2278 * Major route resolver routine. 2279 */ 2280 2281 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4, 2282 const struct sk_buff *skb) 2283 { 2284 __u8 tos = RT_FL_TOS(fl4); 2285 struct fib_result res; 2286 struct rtable *rth; 2287 2288 res.tclassid = 0; 2289 res.fi = NULL; 2290 res.table = NULL; 2291 2292 fl4->flowi4_iif = LOOPBACK_IFINDEX; 2293 fl4->flowi4_tos = tos & IPTOS_RT_MASK; 2294 fl4->flowi4_scope = ((tos & RTO_ONLINK) ? 2295 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE); 2296 2297 rcu_read_lock(); 2298 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb); 2299 rcu_read_unlock(); 2300 2301 return rth; 2302 } 2303 EXPORT_SYMBOL_GPL(ip_route_output_key_hash); 2304 2305 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4, 2306 struct fib_result *res, 2307 const struct sk_buff *skb) 2308 { 2309 struct net_device *dev_out = NULL; 2310 int orig_oif = fl4->flowi4_oif; 2311 unsigned int flags = 0; 2312 struct rtable *rth; 2313 int err = -ENETUNREACH; 2314 2315 if (fl4->saddr) { 2316 rth = ERR_PTR(-EINVAL); 2317 if (ipv4_is_multicast(fl4->saddr) || 2318 ipv4_is_lbcast(fl4->saddr) || 2319 ipv4_is_zeronet(fl4->saddr)) 2320 goto out; 2321 2322 /* I removed check for oif == dev_out->oif here. 2323 It was wrong for two reasons: 2324 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 2325 is assigned to multiple interfaces. 2326 2. Moreover, we are allowed to send packets with saddr 2327 of another iface. --ANK 2328 */ 2329 2330 if (fl4->flowi4_oif == 0 && 2331 (ipv4_is_multicast(fl4->daddr) || 2332 ipv4_is_lbcast(fl4->daddr))) { 2333 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2334 dev_out = __ip_dev_find(net, fl4->saddr, false); 2335 if (!dev_out) 2336 goto out; 2337 2338 /* Special hack: user can direct multicasts 2339 and limited broadcast via necessary interface 2340 without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 2341 This hack is not just for fun, it allows 2342 vic,vat and friends to work. 2343 They bind socket to loopback, set ttl to zero 2344 and expect that it will work. 2345 From the viewpoint of routing cache they are broken, 2346 because we are not allowed to build multicast path 2347 with loopback source addr (look, routing cache 2348 cannot know, that ttl is zero, so that packet 2349 will not leave this host and route is valid). 2350 Luckily, this hack is good workaround. 2351 */ 2352 2353 fl4->flowi4_oif = dev_out->ifindex; 2354 goto make_route; 2355 } 2356 2357 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 2358 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2359 if (!__ip_dev_find(net, fl4->saddr, false)) 2360 goto out; 2361 } 2362 } 2363 2364 2365 if (fl4->flowi4_oif) { 2366 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif); 2367 rth = ERR_PTR(-ENODEV); 2368 if (!dev_out) 2369 goto out; 2370 2371 /* RACE: Check return value of inet_select_addr instead. */ 2372 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 2373 rth = ERR_PTR(-ENETUNREACH); 2374 goto out; 2375 } 2376 if (ipv4_is_local_multicast(fl4->daddr) || 2377 ipv4_is_lbcast(fl4->daddr) || 2378 fl4->flowi4_proto == IPPROTO_IGMP) { 2379 if (!fl4->saddr) 2380 fl4->saddr = inet_select_addr(dev_out, 0, 2381 RT_SCOPE_LINK); 2382 goto make_route; 2383 } 2384 if (!fl4->saddr) { 2385 if (ipv4_is_multicast(fl4->daddr)) 2386 fl4->saddr = inet_select_addr(dev_out, 0, 2387 fl4->flowi4_scope); 2388 else if (!fl4->daddr) 2389 fl4->saddr = inet_select_addr(dev_out, 0, 2390 RT_SCOPE_HOST); 2391 } 2392 } 2393 2394 if (!fl4->daddr) { 2395 fl4->daddr = fl4->saddr; 2396 if (!fl4->daddr) 2397 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK); 2398 dev_out = net->loopback_dev; 2399 fl4->flowi4_oif = LOOPBACK_IFINDEX; 2400 res->type = RTN_LOCAL; 2401 flags |= RTCF_LOCAL; 2402 goto make_route; 2403 } 2404 2405 err = fib_lookup(net, fl4, res, 0); 2406 if (err) { 2407 res->fi = NULL; 2408 res->table = NULL; 2409 if (fl4->flowi4_oif && 2410 (ipv4_is_multicast(fl4->daddr) || 2411 !netif_index_is_l3_master(net, fl4->flowi4_oif))) { 2412 /* Apparently, routing tables are wrong. Assume, 2413 that the destination is on link. 2414 2415 WHY? DW. 2416 Because we are allowed to send to iface 2417 even if it has NO routes and NO assigned 2418 addresses. When oif is specified, routing 2419 tables are looked up with only one purpose: 2420 to catch if destination is gatewayed, rather than 2421 direct. Moreover, if MSG_DONTROUTE is set, 2422 we send packet, ignoring both routing tables 2423 and ifaddr state. --ANK 2424 2425 2426 We could make it even if oif is unknown, 2427 likely IPv6, but we do not. 2428 */ 2429 2430 if (fl4->saddr == 0) 2431 fl4->saddr = inet_select_addr(dev_out, 0, 2432 RT_SCOPE_LINK); 2433 res->type = RTN_UNICAST; 2434 goto make_route; 2435 } 2436 rth = ERR_PTR(err); 2437 goto out; 2438 } 2439 2440 if (res->type == RTN_LOCAL) { 2441 if (!fl4->saddr) { 2442 if (res->fi->fib_prefsrc) 2443 fl4->saddr = res->fi->fib_prefsrc; 2444 else 2445 fl4->saddr = fl4->daddr; 2446 } 2447 2448 /* L3 master device is the loopback for that domain */ 2449 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? : 2450 net->loopback_dev; 2451 2452 /* make sure orig_oif points to fib result device even 2453 * though packet rx/tx happens over loopback or l3mdev 2454 */ 2455 orig_oif = FIB_RES_OIF(*res); 2456 2457 fl4->flowi4_oif = dev_out->ifindex; 2458 flags |= RTCF_LOCAL; 2459 goto make_route; 2460 } 2461 2462 fib_select_path(net, res, fl4, skb); 2463 2464 dev_out = FIB_RES_DEV(*res); 2465 fl4->flowi4_oif = dev_out->ifindex; 2466 2467 2468 make_route: 2469 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags); 2470 2471 out: 2472 return rth; 2473 } 2474 2475 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie) 2476 { 2477 return NULL; 2478 } 2479 2480 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst) 2481 { 2482 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU); 2483 2484 return mtu ? : dst->dev->mtu; 2485 } 2486 2487 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk, 2488 struct sk_buff *skb, u32 mtu) 2489 { 2490 } 2491 2492 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk, 2493 struct sk_buff *skb) 2494 { 2495 } 2496 2497 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst, 2498 unsigned long old) 2499 { 2500 return NULL; 2501 } 2502 2503 static struct dst_ops ipv4_dst_blackhole_ops = { 2504 .family = AF_INET, 2505 .check = ipv4_blackhole_dst_check, 2506 .mtu = ipv4_blackhole_mtu, 2507 .default_advmss = ipv4_default_advmss, 2508 .update_pmtu = ipv4_rt_blackhole_update_pmtu, 2509 .redirect = ipv4_rt_blackhole_redirect, 2510 .cow_metrics = ipv4_rt_blackhole_cow_metrics, 2511 .neigh_lookup = ipv4_neigh_lookup, 2512 }; 2513 2514 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2515 { 2516 struct rtable *ort = (struct rtable *) dst_orig; 2517 struct rtable *rt; 2518 2519 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0); 2520 if (rt) { 2521 struct dst_entry *new = &rt->dst; 2522 2523 new->__use = 1; 2524 new->input = dst_discard; 2525 new->output = dst_discard_out; 2526 2527 new->dev = net->loopback_dev; 2528 if (new->dev) 2529 dev_hold(new->dev); 2530 2531 rt->rt_is_input = ort->rt_is_input; 2532 rt->rt_iif = ort->rt_iif; 2533 rt->rt_pmtu = ort->rt_pmtu; 2534 2535 rt->rt_genid = rt_genid_ipv4(net); 2536 rt->rt_flags = ort->rt_flags; 2537 rt->rt_type = ort->rt_type; 2538 rt->rt_gateway = ort->rt_gateway; 2539 rt->rt_uses_gateway = ort->rt_uses_gateway; 2540 2541 INIT_LIST_HEAD(&rt->rt_uncached); 2542 } 2543 2544 dst_release(dst_orig); 2545 2546 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2547 } 2548 2549 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2550 const struct sock *sk) 2551 { 2552 struct rtable *rt = __ip_route_output_key(net, flp4); 2553 2554 if (IS_ERR(rt)) 2555 return rt; 2556 2557 if (flp4->flowi4_proto) 2558 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst, 2559 flowi4_to_flowi(flp4), 2560 sk, 0); 2561 2562 return rt; 2563 } 2564 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2565 2566 /* called with rcu_read_lock held */ 2567 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id, 2568 struct flowi4 *fl4, struct sk_buff *skb, u32 portid, 2569 u32 seq) 2570 { 2571 struct rtable *rt = skb_rtable(skb); 2572 struct rtmsg *r; 2573 struct nlmsghdr *nlh; 2574 unsigned long expires = 0; 2575 u32 error; 2576 u32 metrics[RTAX_MAX]; 2577 2578 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0); 2579 if (!nlh) 2580 return -EMSGSIZE; 2581 2582 r = nlmsg_data(nlh); 2583 r->rtm_family = AF_INET; 2584 r->rtm_dst_len = 32; 2585 r->rtm_src_len = 0; 2586 r->rtm_tos = fl4->flowi4_tos; 2587 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT; 2588 if (nla_put_u32(skb, RTA_TABLE, table_id)) 2589 goto nla_put_failure; 2590 r->rtm_type = rt->rt_type; 2591 r->rtm_scope = RT_SCOPE_UNIVERSE; 2592 r->rtm_protocol = RTPROT_UNSPEC; 2593 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 2594 if (rt->rt_flags & RTCF_NOTIFY) 2595 r->rtm_flags |= RTM_F_NOTIFY; 2596 if (IPCB(skb)->flags & IPSKB_DOREDIRECT) 2597 r->rtm_flags |= RTCF_DOREDIRECT; 2598 2599 if (nla_put_in_addr(skb, RTA_DST, dst)) 2600 goto nla_put_failure; 2601 if (src) { 2602 r->rtm_src_len = 32; 2603 if (nla_put_in_addr(skb, RTA_SRC, src)) 2604 goto nla_put_failure; 2605 } 2606 if (rt->dst.dev && 2607 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex)) 2608 goto nla_put_failure; 2609 #ifdef CONFIG_IP_ROUTE_CLASSID 2610 if (rt->dst.tclassid && 2611 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid)) 2612 goto nla_put_failure; 2613 #endif 2614 if (!rt_is_input_route(rt) && 2615 fl4->saddr != src) { 2616 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr)) 2617 goto nla_put_failure; 2618 } 2619 if (rt->rt_uses_gateway && 2620 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway)) 2621 goto nla_put_failure; 2622 2623 expires = rt->dst.expires; 2624 if (expires) { 2625 unsigned long now = jiffies; 2626 2627 if (time_before(now, expires)) 2628 expires -= now; 2629 else 2630 expires = 0; 2631 } 2632 2633 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics)); 2634 if (rt->rt_pmtu && expires) 2635 metrics[RTAX_MTU - 1] = rt->rt_pmtu; 2636 if (rtnetlink_put_metrics(skb, metrics) < 0) 2637 goto nla_put_failure; 2638 2639 if (fl4->flowi4_mark && 2640 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark)) 2641 goto nla_put_failure; 2642 2643 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) && 2644 nla_put_u32(skb, RTA_UID, 2645 from_kuid_munged(current_user_ns(), fl4->flowi4_uid))) 2646 goto nla_put_failure; 2647 2648 error = rt->dst.error; 2649 2650 if (rt_is_input_route(rt)) { 2651 #ifdef CONFIG_IP_MROUTE 2652 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) && 2653 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) { 2654 int err = ipmr_get_route(net, skb, 2655 fl4->saddr, fl4->daddr, 2656 r, portid); 2657 2658 if (err <= 0) { 2659 if (err == 0) 2660 return 0; 2661 goto nla_put_failure; 2662 } 2663 } else 2664 #endif 2665 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex)) 2666 goto nla_put_failure; 2667 } 2668 2669 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0) 2670 goto nla_put_failure; 2671 2672 nlmsg_end(skb, nlh); 2673 return 0; 2674 2675 nla_put_failure: 2676 nlmsg_cancel(skb, nlh); 2677 return -EMSGSIZE; 2678 } 2679 2680 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, 2681 struct netlink_ext_ack *extack) 2682 { 2683 struct net *net = sock_net(in_skb->sk); 2684 struct rtmsg *rtm; 2685 struct nlattr *tb[RTA_MAX+1]; 2686 struct fib_result res = {}; 2687 struct rtable *rt = NULL; 2688 struct flowi4 fl4; 2689 __be32 dst = 0; 2690 __be32 src = 0; 2691 u32 iif; 2692 int err; 2693 int mark; 2694 struct sk_buff *skb; 2695 u32 table_id = RT_TABLE_MAIN; 2696 kuid_t uid; 2697 2698 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy, 2699 extack); 2700 if (err < 0) 2701 goto errout; 2702 2703 rtm = nlmsg_data(nlh); 2704 2705 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 2706 if (!skb) { 2707 err = -ENOBUFS; 2708 goto errout; 2709 } 2710 2711 /* Reserve room for dummy headers, this skb can pass 2712 through good chunk of routing engine. 2713 */ 2714 skb_reset_mac_header(skb); 2715 skb_reset_network_header(skb); 2716 2717 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0; 2718 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0; 2719 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0; 2720 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0; 2721 if (tb[RTA_UID]) 2722 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID])); 2723 else 2724 uid = (iif ? INVALID_UID : current_uid()); 2725 2726 /* Bugfix: need to give ip_route_input enough of an IP header to 2727 * not gag. 2728 */ 2729 ip_hdr(skb)->protocol = IPPROTO_UDP; 2730 ip_hdr(skb)->saddr = src; 2731 ip_hdr(skb)->daddr = dst; 2732 2733 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr)); 2734 2735 memset(&fl4, 0, sizeof(fl4)); 2736 fl4.daddr = dst; 2737 fl4.saddr = src; 2738 fl4.flowi4_tos = rtm->rtm_tos; 2739 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0; 2740 fl4.flowi4_mark = mark; 2741 fl4.flowi4_uid = uid; 2742 2743 rcu_read_lock(); 2744 2745 if (iif) { 2746 struct net_device *dev; 2747 2748 dev = dev_get_by_index_rcu(net, iif); 2749 if (!dev) { 2750 err = -ENODEV; 2751 goto errout_free; 2752 } 2753 2754 skb->protocol = htons(ETH_P_IP); 2755 skb->dev = dev; 2756 skb->mark = mark; 2757 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos, 2758 dev, &res); 2759 2760 rt = skb_rtable(skb); 2761 if (err == 0 && rt->dst.error) 2762 err = -rt->dst.error; 2763 } else { 2764 fl4.flowi4_iif = LOOPBACK_IFINDEX; 2765 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb); 2766 err = 0; 2767 if (IS_ERR(rt)) 2768 err = PTR_ERR(rt); 2769 else 2770 skb_dst_set(skb, &rt->dst); 2771 } 2772 2773 if (err) 2774 goto errout_free; 2775 2776 if (rtm->rtm_flags & RTM_F_NOTIFY) 2777 rt->rt_flags |= RTCF_NOTIFY; 2778 2779 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE) 2780 table_id = rt->rt_table_id; 2781 2782 if (rtm->rtm_flags & RTM_F_FIB_MATCH) { 2783 if (!res.fi) { 2784 err = fib_props[res.type].error; 2785 if (!err) 2786 err = -EHOSTUNREACH; 2787 goto errout_free; 2788 } 2789 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid, 2790 nlh->nlmsg_seq, RTM_NEWROUTE, table_id, 2791 rt->rt_type, res.prefix, res.prefixlen, 2792 fl4.flowi4_tos, res.fi, 0); 2793 } else { 2794 err = rt_fill_info(net, dst, src, table_id, &fl4, skb, 2795 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq); 2796 } 2797 if (err < 0) 2798 goto errout_free; 2799 2800 rcu_read_unlock(); 2801 2802 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 2803 errout: 2804 return err; 2805 2806 errout_free: 2807 rcu_read_unlock(); 2808 kfree_skb(skb); 2809 goto errout; 2810 } 2811 2812 void ip_rt_multicast_event(struct in_device *in_dev) 2813 { 2814 rt_cache_flush(dev_net(in_dev->dev)); 2815 } 2816 2817 #ifdef CONFIG_SYSCTL 2818 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 2819 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 2820 static int ip_rt_gc_elasticity __read_mostly = 8; 2821 2822 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write, 2823 void __user *buffer, 2824 size_t *lenp, loff_t *ppos) 2825 { 2826 struct net *net = (struct net *)__ctl->extra1; 2827 2828 if (write) { 2829 rt_cache_flush(net); 2830 fnhe_genid_bump(net); 2831 return 0; 2832 } 2833 2834 return -EINVAL; 2835 } 2836 2837 static struct ctl_table ipv4_route_table[] = { 2838 { 2839 .procname = "gc_thresh", 2840 .data = &ipv4_dst_ops.gc_thresh, 2841 .maxlen = sizeof(int), 2842 .mode = 0644, 2843 .proc_handler = proc_dointvec, 2844 }, 2845 { 2846 .procname = "max_size", 2847 .data = &ip_rt_max_size, 2848 .maxlen = sizeof(int), 2849 .mode = 0644, 2850 .proc_handler = proc_dointvec, 2851 }, 2852 { 2853 /* Deprecated. Use gc_min_interval_ms */ 2854 2855 .procname = "gc_min_interval", 2856 .data = &ip_rt_gc_min_interval, 2857 .maxlen = sizeof(int), 2858 .mode = 0644, 2859 .proc_handler = proc_dointvec_jiffies, 2860 }, 2861 { 2862 .procname = "gc_min_interval_ms", 2863 .data = &ip_rt_gc_min_interval, 2864 .maxlen = sizeof(int), 2865 .mode = 0644, 2866 .proc_handler = proc_dointvec_ms_jiffies, 2867 }, 2868 { 2869 .procname = "gc_timeout", 2870 .data = &ip_rt_gc_timeout, 2871 .maxlen = sizeof(int), 2872 .mode = 0644, 2873 .proc_handler = proc_dointvec_jiffies, 2874 }, 2875 { 2876 .procname = "gc_interval", 2877 .data = &ip_rt_gc_interval, 2878 .maxlen = sizeof(int), 2879 .mode = 0644, 2880 .proc_handler = proc_dointvec_jiffies, 2881 }, 2882 { 2883 .procname = "redirect_load", 2884 .data = &ip_rt_redirect_load, 2885 .maxlen = sizeof(int), 2886 .mode = 0644, 2887 .proc_handler = proc_dointvec, 2888 }, 2889 { 2890 .procname = "redirect_number", 2891 .data = &ip_rt_redirect_number, 2892 .maxlen = sizeof(int), 2893 .mode = 0644, 2894 .proc_handler = proc_dointvec, 2895 }, 2896 { 2897 .procname = "redirect_silence", 2898 .data = &ip_rt_redirect_silence, 2899 .maxlen = sizeof(int), 2900 .mode = 0644, 2901 .proc_handler = proc_dointvec, 2902 }, 2903 { 2904 .procname = "error_cost", 2905 .data = &ip_rt_error_cost, 2906 .maxlen = sizeof(int), 2907 .mode = 0644, 2908 .proc_handler = proc_dointvec, 2909 }, 2910 { 2911 .procname = "error_burst", 2912 .data = &ip_rt_error_burst, 2913 .maxlen = sizeof(int), 2914 .mode = 0644, 2915 .proc_handler = proc_dointvec, 2916 }, 2917 { 2918 .procname = "gc_elasticity", 2919 .data = &ip_rt_gc_elasticity, 2920 .maxlen = sizeof(int), 2921 .mode = 0644, 2922 .proc_handler = proc_dointvec, 2923 }, 2924 { 2925 .procname = "mtu_expires", 2926 .data = &ip_rt_mtu_expires, 2927 .maxlen = sizeof(int), 2928 .mode = 0644, 2929 .proc_handler = proc_dointvec_jiffies, 2930 }, 2931 { 2932 .procname = "min_pmtu", 2933 .data = &ip_rt_min_pmtu, 2934 .maxlen = sizeof(int), 2935 .mode = 0644, 2936 .proc_handler = proc_dointvec, 2937 }, 2938 { 2939 .procname = "min_adv_mss", 2940 .data = &ip_rt_min_advmss, 2941 .maxlen = sizeof(int), 2942 .mode = 0644, 2943 .proc_handler = proc_dointvec, 2944 }, 2945 { } 2946 }; 2947 2948 static struct ctl_table ipv4_route_flush_table[] = { 2949 { 2950 .procname = "flush", 2951 .maxlen = sizeof(int), 2952 .mode = 0200, 2953 .proc_handler = ipv4_sysctl_rtcache_flush, 2954 }, 2955 { }, 2956 }; 2957 2958 static __net_init int sysctl_route_net_init(struct net *net) 2959 { 2960 struct ctl_table *tbl; 2961 2962 tbl = ipv4_route_flush_table; 2963 if (!net_eq(net, &init_net)) { 2964 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL); 2965 if (!tbl) 2966 goto err_dup; 2967 2968 /* Don't export sysctls to unprivileged users */ 2969 if (net->user_ns != &init_user_ns) 2970 tbl[0].procname = NULL; 2971 } 2972 tbl[0].extra1 = net; 2973 2974 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl); 2975 if (!net->ipv4.route_hdr) 2976 goto err_reg; 2977 return 0; 2978 2979 err_reg: 2980 if (tbl != ipv4_route_flush_table) 2981 kfree(tbl); 2982 err_dup: 2983 return -ENOMEM; 2984 } 2985 2986 static __net_exit void sysctl_route_net_exit(struct net *net) 2987 { 2988 struct ctl_table *tbl; 2989 2990 tbl = net->ipv4.route_hdr->ctl_table_arg; 2991 unregister_net_sysctl_table(net->ipv4.route_hdr); 2992 BUG_ON(tbl == ipv4_route_flush_table); 2993 kfree(tbl); 2994 } 2995 2996 static __net_initdata struct pernet_operations sysctl_route_ops = { 2997 .init = sysctl_route_net_init, 2998 .exit = sysctl_route_net_exit, 2999 }; 3000 #endif 3001 3002 static __net_init int rt_genid_init(struct net *net) 3003 { 3004 atomic_set(&net->ipv4.rt_genid, 0); 3005 atomic_set(&net->fnhe_genid, 0); 3006 atomic_set(&net->ipv4.dev_addr_genid, get_random_int()); 3007 return 0; 3008 } 3009 3010 static __net_initdata struct pernet_operations rt_genid_ops = { 3011 .init = rt_genid_init, 3012 }; 3013 3014 static int __net_init ipv4_inetpeer_init(struct net *net) 3015 { 3016 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL); 3017 3018 if (!bp) 3019 return -ENOMEM; 3020 inet_peer_base_init(bp); 3021 net->ipv4.peers = bp; 3022 return 0; 3023 } 3024 3025 static void __net_exit ipv4_inetpeer_exit(struct net *net) 3026 { 3027 struct inet_peer_base *bp = net->ipv4.peers; 3028 3029 net->ipv4.peers = NULL; 3030 inetpeer_invalidate_tree(bp); 3031 kfree(bp); 3032 } 3033 3034 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = { 3035 .init = ipv4_inetpeer_init, 3036 .exit = ipv4_inetpeer_exit, 3037 }; 3038 3039 #ifdef CONFIG_IP_ROUTE_CLASSID 3040 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 3041 #endif /* CONFIG_IP_ROUTE_CLASSID */ 3042 3043 int __init ip_rt_init(void) 3044 { 3045 int cpu; 3046 3047 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL); 3048 if (!ip_idents) 3049 panic("IP: failed to allocate ip_idents\n"); 3050 3051 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents)); 3052 3053 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL); 3054 if (!ip_tstamps) 3055 panic("IP: failed to allocate ip_tstamps\n"); 3056 3057 for_each_possible_cpu(cpu) { 3058 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 3059 3060 INIT_LIST_HEAD(&ul->head); 3061 spin_lock_init(&ul->lock); 3062 } 3063 #ifdef CONFIG_IP_ROUTE_CLASSID 3064 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 3065 if (!ip_rt_acct) 3066 panic("IP: failed to allocate ip_rt_acct\n"); 3067 #endif 3068 3069 ipv4_dst_ops.kmem_cachep = 3070 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0, 3071 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 3072 3073 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 3074 3075 if (dst_entries_init(&ipv4_dst_ops) < 0) 3076 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 3077 3078 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 3079 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 3080 3081 ipv4_dst_ops.gc_thresh = ~0; 3082 ip_rt_max_size = INT_MAX; 3083 3084 devinet_init(); 3085 ip_fib_init(); 3086 3087 if (ip_rt_proc_init()) 3088 pr_err("Unable to create route proc files\n"); 3089 #ifdef CONFIG_XFRM 3090 xfrm_init(); 3091 xfrm4_init(); 3092 #endif 3093 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, 3094 RTNL_FLAG_DOIT_UNLOCKED); 3095 3096 #ifdef CONFIG_SYSCTL 3097 register_pernet_subsys(&sysctl_route_ops); 3098 #endif 3099 register_pernet_subsys(&rt_genid_ops); 3100 register_pernet_subsys(&ipv4_inetpeer_ops); 3101 return 0; 3102 } 3103 3104 #ifdef CONFIG_SYSCTL 3105 /* 3106 * We really need to sanitize the damn ipv4 init order, then all 3107 * this nonsense will go away. 3108 */ 3109 void __init ip_static_sysctl_init(void) 3110 { 3111 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table); 3112 } 3113 #endif 3114