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 #endif 112 #include <net/secure_seq.h> 113 #include <net/ip_tunnels.h> 114 #include <net/l3mdev.h> 115 116 #include "fib_lookup.h" 117 118 #define RT_FL_TOS(oldflp4) \ 119 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)) 120 121 #define RT_GC_TIMEOUT (300*HZ) 122 123 static int ip_rt_max_size; 124 static int ip_rt_redirect_number __read_mostly = 9; 125 static int ip_rt_redirect_load __read_mostly = HZ / 50; 126 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1)); 127 static int ip_rt_error_cost __read_mostly = HZ; 128 static int ip_rt_error_burst __read_mostly = 5 * HZ; 129 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ; 130 static u32 ip_rt_min_pmtu __read_mostly = 512 + 20 + 20; 131 static int ip_rt_min_advmss __read_mostly = 256; 132 133 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT; 134 135 /* 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", 0444, net->proc_net, 382 &rt_cache_seq_fops); 383 if (!pde) 384 goto err1; 385 386 pde = proc_create("rt_cache", 0444, 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->rt_mtu_locked = fnhe->fnhe_mtu_locked; 637 rt->dst.expires = fnhe->fnhe_expires; 638 639 if (fnhe->fnhe_gw) { 640 rt->rt_flags |= RTCF_REDIRECTED; 641 rt->rt_gateway = fnhe->fnhe_gw; 642 rt->rt_uses_gateway = 1; 643 } 644 } 645 646 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw, 647 u32 pmtu, bool lock, unsigned long expires) 648 { 649 struct fnhe_hash_bucket *hash; 650 struct fib_nh_exception *fnhe; 651 struct rtable *rt; 652 u32 genid, hval; 653 unsigned int i; 654 int depth; 655 656 genid = fnhe_genid(dev_net(nh->nh_dev)); 657 hval = fnhe_hashfun(daddr); 658 659 spin_lock_bh(&fnhe_lock); 660 661 hash = rcu_dereference(nh->nh_exceptions); 662 if (!hash) { 663 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC); 664 if (!hash) 665 goto out_unlock; 666 rcu_assign_pointer(nh->nh_exceptions, hash); 667 } 668 669 hash += hval; 670 671 depth = 0; 672 for (fnhe = rcu_dereference(hash->chain); fnhe; 673 fnhe = rcu_dereference(fnhe->fnhe_next)) { 674 if (fnhe->fnhe_daddr == daddr) 675 break; 676 depth++; 677 } 678 679 if (fnhe) { 680 if (fnhe->fnhe_genid != genid) 681 fnhe->fnhe_genid = genid; 682 if (gw) 683 fnhe->fnhe_gw = gw; 684 if (pmtu) { 685 fnhe->fnhe_pmtu = pmtu; 686 fnhe->fnhe_mtu_locked = lock; 687 } 688 fnhe->fnhe_expires = max(1UL, expires); 689 /* Update all cached dsts too */ 690 rt = rcu_dereference(fnhe->fnhe_rth_input); 691 if (rt) 692 fill_route_from_fnhe(rt, fnhe); 693 rt = rcu_dereference(fnhe->fnhe_rth_output); 694 if (rt) 695 fill_route_from_fnhe(rt, fnhe); 696 } else { 697 if (depth > FNHE_RECLAIM_DEPTH) 698 fnhe = fnhe_oldest(hash); 699 else { 700 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC); 701 if (!fnhe) 702 goto out_unlock; 703 704 fnhe->fnhe_next = hash->chain; 705 rcu_assign_pointer(hash->chain, fnhe); 706 } 707 fnhe->fnhe_genid = genid; 708 fnhe->fnhe_daddr = daddr; 709 fnhe->fnhe_gw = gw; 710 fnhe->fnhe_pmtu = pmtu; 711 fnhe->fnhe_mtu_locked = lock; 712 fnhe->fnhe_expires = max(1UL, expires); 713 714 /* Exception created; mark the cached routes for the nexthop 715 * stale, so anyone caching it rechecks if this exception 716 * applies to them. 717 */ 718 rt = rcu_dereference(nh->nh_rth_input); 719 if (rt) 720 rt->dst.obsolete = DST_OBSOLETE_KILL; 721 722 for_each_possible_cpu(i) { 723 struct rtable __rcu **prt; 724 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i); 725 rt = rcu_dereference(*prt); 726 if (rt) 727 rt->dst.obsolete = DST_OBSOLETE_KILL; 728 } 729 } 730 731 fnhe->fnhe_stamp = jiffies; 732 733 out_unlock: 734 spin_unlock_bh(&fnhe_lock); 735 } 736 737 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4, 738 bool kill_route) 739 { 740 __be32 new_gw = icmp_hdr(skb)->un.gateway; 741 __be32 old_gw = ip_hdr(skb)->saddr; 742 struct net_device *dev = skb->dev; 743 struct in_device *in_dev; 744 struct fib_result res; 745 struct neighbour *n; 746 struct net *net; 747 748 switch (icmp_hdr(skb)->code & 7) { 749 case ICMP_REDIR_NET: 750 case ICMP_REDIR_NETTOS: 751 case ICMP_REDIR_HOST: 752 case ICMP_REDIR_HOSTTOS: 753 break; 754 755 default: 756 return; 757 } 758 759 if (rt->rt_gateway != old_gw) 760 return; 761 762 in_dev = __in_dev_get_rcu(dev); 763 if (!in_dev) 764 return; 765 766 net = dev_net(dev); 767 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) || 768 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) || 769 ipv4_is_zeronet(new_gw)) 770 goto reject_redirect; 771 772 if (!IN_DEV_SHARED_MEDIA(in_dev)) { 773 if (!inet_addr_onlink(in_dev, new_gw, old_gw)) 774 goto reject_redirect; 775 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) 776 goto reject_redirect; 777 } else { 778 if (inet_addr_type(net, new_gw) != RTN_UNICAST) 779 goto reject_redirect; 780 } 781 782 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw); 783 if (!n) 784 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev); 785 if (!IS_ERR(n)) { 786 if (!(n->nud_state & NUD_VALID)) { 787 neigh_event_send(n, NULL); 788 } else { 789 if (fib_lookup(net, fl4, &res, 0) == 0) { 790 struct fib_nh *nh = &FIB_RES_NH(res); 791 792 update_or_create_fnhe(nh, fl4->daddr, new_gw, 793 0, false, 794 jiffies + ip_rt_gc_timeout); 795 } 796 if (kill_route) 797 rt->dst.obsolete = DST_OBSOLETE_KILL; 798 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n); 799 } 800 neigh_release(n); 801 } 802 return; 803 804 reject_redirect: 805 #ifdef CONFIG_IP_ROUTE_VERBOSE 806 if (IN_DEV_LOG_MARTIANS(in_dev)) { 807 const struct iphdr *iph = (const struct iphdr *) skb->data; 808 __be32 daddr = iph->daddr; 809 __be32 saddr = iph->saddr; 810 811 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n" 812 " Advised path = %pI4 -> %pI4\n", 813 &old_gw, dev->name, &new_gw, 814 &saddr, &daddr); 815 } 816 #endif 817 ; 818 } 819 820 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb) 821 { 822 struct rtable *rt; 823 struct flowi4 fl4; 824 const struct iphdr *iph = (const struct iphdr *) skb->data; 825 struct net *net = dev_net(skb->dev); 826 int oif = skb->dev->ifindex; 827 u8 tos = RT_TOS(iph->tos); 828 u8 prot = iph->protocol; 829 u32 mark = skb->mark; 830 831 rt = (struct rtable *) dst; 832 833 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0); 834 __ip_do_redirect(rt, skb, &fl4, true); 835 } 836 837 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst) 838 { 839 struct rtable *rt = (struct rtable *)dst; 840 struct dst_entry *ret = dst; 841 842 if (rt) { 843 if (dst->obsolete > 0) { 844 ip_rt_put(rt); 845 ret = NULL; 846 } else if ((rt->rt_flags & RTCF_REDIRECTED) || 847 rt->dst.expires) { 848 ip_rt_put(rt); 849 ret = NULL; 850 } 851 } 852 return ret; 853 } 854 855 /* 856 * Algorithm: 857 * 1. The first ip_rt_redirect_number redirects are sent 858 * with exponential backoff, then we stop sending them at all, 859 * assuming that the host ignores our redirects. 860 * 2. If we did not see packets requiring redirects 861 * during ip_rt_redirect_silence, we assume that the host 862 * forgot redirected route and start to send redirects again. 863 * 864 * This algorithm is much cheaper and more intelligent than dumb load limiting 865 * in icmp.c. 866 * 867 * NOTE. Do not forget to inhibit load limiting for redirects (redundant) 868 * and "frag. need" (breaks PMTU discovery) in icmp.c. 869 */ 870 871 void ip_rt_send_redirect(struct sk_buff *skb) 872 { 873 struct rtable *rt = skb_rtable(skb); 874 struct in_device *in_dev; 875 struct inet_peer *peer; 876 struct net *net; 877 int log_martians; 878 int vif; 879 880 rcu_read_lock(); 881 in_dev = __in_dev_get_rcu(rt->dst.dev); 882 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) { 883 rcu_read_unlock(); 884 return; 885 } 886 log_martians = IN_DEV_LOG_MARTIANS(in_dev); 887 vif = l3mdev_master_ifindex_rcu(rt->dst.dev); 888 rcu_read_unlock(); 889 890 net = dev_net(rt->dst.dev); 891 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1); 892 if (!peer) { 893 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, 894 rt_nexthop(rt, ip_hdr(skb)->daddr)); 895 return; 896 } 897 898 /* No redirected packets during ip_rt_redirect_silence; 899 * reset the algorithm. 900 */ 901 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) 902 peer->rate_tokens = 0; 903 904 /* Too many ignored redirects; do not send anything 905 * set dst.rate_last to the last seen redirected packet. 906 */ 907 if (peer->rate_tokens >= ip_rt_redirect_number) { 908 peer->rate_last = jiffies; 909 goto out_put_peer; 910 } 911 912 /* Check for load limit; set rate_last to the latest sent 913 * redirect. 914 */ 915 if (peer->rate_tokens == 0 || 916 time_after(jiffies, 917 (peer->rate_last + 918 (ip_rt_redirect_load << peer->rate_tokens)))) { 919 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr); 920 921 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw); 922 peer->rate_last = jiffies; 923 ++peer->rate_tokens; 924 #ifdef CONFIG_IP_ROUTE_VERBOSE 925 if (log_martians && 926 peer->rate_tokens == ip_rt_redirect_number) 927 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n", 928 &ip_hdr(skb)->saddr, inet_iif(skb), 929 &ip_hdr(skb)->daddr, &gw); 930 #endif 931 } 932 out_put_peer: 933 inet_putpeer(peer); 934 } 935 936 static int ip_error(struct sk_buff *skb) 937 { 938 struct rtable *rt = skb_rtable(skb); 939 struct net_device *dev = skb->dev; 940 struct in_device *in_dev; 941 struct inet_peer *peer; 942 unsigned long now; 943 struct net *net; 944 bool send; 945 int code; 946 947 if (netif_is_l3_master(skb->dev)) { 948 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif); 949 if (!dev) 950 goto out; 951 } 952 953 in_dev = __in_dev_get_rcu(dev); 954 955 /* IP on this device is disabled. */ 956 if (!in_dev) 957 goto out; 958 959 net = dev_net(rt->dst.dev); 960 if (!IN_DEV_FORWARD(in_dev)) { 961 switch (rt->dst.error) { 962 case EHOSTUNREACH: 963 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS); 964 break; 965 966 case ENETUNREACH: 967 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 968 break; 969 } 970 goto out; 971 } 972 973 switch (rt->dst.error) { 974 case EINVAL: 975 default: 976 goto out; 977 case EHOSTUNREACH: 978 code = ICMP_HOST_UNREACH; 979 break; 980 case ENETUNREACH: 981 code = ICMP_NET_UNREACH; 982 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 983 break; 984 case EACCES: 985 code = ICMP_PKT_FILTERED; 986 break; 987 } 988 989 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 990 l3mdev_master_ifindex(skb->dev), 1); 991 992 send = true; 993 if (peer) { 994 now = jiffies; 995 peer->rate_tokens += now - peer->rate_last; 996 if (peer->rate_tokens > ip_rt_error_burst) 997 peer->rate_tokens = ip_rt_error_burst; 998 peer->rate_last = now; 999 if (peer->rate_tokens >= ip_rt_error_cost) 1000 peer->rate_tokens -= ip_rt_error_cost; 1001 else 1002 send = false; 1003 inet_putpeer(peer); 1004 } 1005 if (send) 1006 icmp_send(skb, ICMP_DEST_UNREACH, code, 0); 1007 1008 out: kfree_skb(skb); 1009 return 0; 1010 } 1011 1012 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu) 1013 { 1014 struct dst_entry *dst = &rt->dst; 1015 struct fib_result res; 1016 bool lock = false; 1017 1018 if (ip_mtu_locked(dst)) 1019 return; 1020 1021 if (ipv4_mtu(dst) < mtu) 1022 return; 1023 1024 if (mtu < ip_rt_min_pmtu) { 1025 lock = true; 1026 mtu = ip_rt_min_pmtu; 1027 } 1028 1029 if (rt->rt_pmtu == mtu && 1030 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2)) 1031 return; 1032 1033 rcu_read_lock(); 1034 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) { 1035 struct fib_nh *nh = &FIB_RES_NH(res); 1036 1037 update_or_create_fnhe(nh, fl4->daddr, 0, mtu, lock, 1038 jiffies + ip_rt_mtu_expires); 1039 } 1040 rcu_read_unlock(); 1041 } 1042 1043 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 1044 struct sk_buff *skb, u32 mtu) 1045 { 1046 struct rtable *rt = (struct rtable *) dst; 1047 struct flowi4 fl4; 1048 1049 ip_rt_build_flow_key(&fl4, sk, skb); 1050 __ip_rt_update_pmtu(rt, &fl4, mtu); 1051 } 1052 1053 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, 1054 int oif, u32 mark, u8 protocol, int flow_flags) 1055 { 1056 const struct iphdr *iph = (const struct iphdr *) skb->data; 1057 struct flowi4 fl4; 1058 struct rtable *rt; 1059 1060 if (!mark) 1061 mark = IP4_REPLY_MARK(net, skb->mark); 1062 1063 __build_flow_key(net, &fl4, NULL, iph, oif, 1064 RT_TOS(iph->tos), protocol, mark, flow_flags); 1065 rt = __ip_route_output_key(net, &fl4); 1066 if (!IS_ERR(rt)) { 1067 __ip_rt_update_pmtu(rt, &fl4, mtu); 1068 ip_rt_put(rt); 1069 } 1070 } 1071 EXPORT_SYMBOL_GPL(ipv4_update_pmtu); 1072 1073 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1074 { 1075 const struct iphdr *iph = (const struct iphdr *) skb->data; 1076 struct flowi4 fl4; 1077 struct rtable *rt; 1078 1079 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0); 1080 1081 if (!fl4.flowi4_mark) 1082 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark); 1083 1084 rt = __ip_route_output_key(sock_net(sk), &fl4); 1085 if (!IS_ERR(rt)) { 1086 __ip_rt_update_pmtu(rt, &fl4, mtu); 1087 ip_rt_put(rt); 1088 } 1089 } 1090 1091 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1092 { 1093 const struct iphdr *iph = (const struct iphdr *) skb->data; 1094 struct flowi4 fl4; 1095 struct rtable *rt; 1096 struct dst_entry *odst = NULL; 1097 bool new = false; 1098 struct net *net = sock_net(sk); 1099 1100 bh_lock_sock(sk); 1101 1102 if (!ip_sk_accept_pmtu(sk)) 1103 goto out; 1104 1105 odst = sk_dst_get(sk); 1106 1107 if (sock_owned_by_user(sk) || !odst) { 1108 __ipv4_sk_update_pmtu(skb, sk, mtu); 1109 goto out; 1110 } 1111 1112 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1113 1114 rt = (struct rtable *)odst; 1115 if (odst->obsolete && !odst->ops->check(odst, 0)) { 1116 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1117 if (IS_ERR(rt)) 1118 goto out; 1119 1120 new = true; 1121 } 1122 1123 __ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu); 1124 1125 if (!dst_check(&rt->dst, 0)) { 1126 if (new) 1127 dst_release(&rt->dst); 1128 1129 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1130 if (IS_ERR(rt)) 1131 goto out; 1132 1133 new = true; 1134 } 1135 1136 if (new) 1137 sk_dst_set(sk, &rt->dst); 1138 1139 out: 1140 bh_unlock_sock(sk); 1141 dst_release(odst); 1142 } 1143 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu); 1144 1145 void ipv4_redirect(struct sk_buff *skb, struct net *net, 1146 int oif, u32 mark, u8 protocol, int flow_flags) 1147 { 1148 const struct iphdr *iph = (const struct iphdr *) skb->data; 1149 struct flowi4 fl4; 1150 struct rtable *rt; 1151 1152 __build_flow_key(net, &fl4, NULL, iph, oif, 1153 RT_TOS(iph->tos), protocol, mark, flow_flags); 1154 rt = __ip_route_output_key(net, &fl4); 1155 if (!IS_ERR(rt)) { 1156 __ip_do_redirect(rt, skb, &fl4, false); 1157 ip_rt_put(rt); 1158 } 1159 } 1160 EXPORT_SYMBOL_GPL(ipv4_redirect); 1161 1162 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk) 1163 { 1164 const struct iphdr *iph = (const struct iphdr *) skb->data; 1165 struct flowi4 fl4; 1166 struct rtable *rt; 1167 struct net *net = sock_net(sk); 1168 1169 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1170 rt = __ip_route_output_key(net, &fl4); 1171 if (!IS_ERR(rt)) { 1172 __ip_do_redirect(rt, skb, &fl4, false); 1173 ip_rt_put(rt); 1174 } 1175 } 1176 EXPORT_SYMBOL_GPL(ipv4_sk_redirect); 1177 1178 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie) 1179 { 1180 struct rtable *rt = (struct rtable *) dst; 1181 1182 /* All IPV4 dsts are created with ->obsolete set to the value 1183 * DST_OBSOLETE_FORCE_CHK which forces validation calls down 1184 * into this function always. 1185 * 1186 * When a PMTU/redirect information update invalidates a route, 1187 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or 1188 * DST_OBSOLETE_DEAD by dst_free(). 1189 */ 1190 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt)) 1191 return NULL; 1192 return dst; 1193 } 1194 1195 static void ipv4_link_failure(struct sk_buff *skb) 1196 { 1197 struct rtable *rt; 1198 1199 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0); 1200 1201 rt = skb_rtable(skb); 1202 if (rt) 1203 dst_set_expires(&rt->dst, 0); 1204 } 1205 1206 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb) 1207 { 1208 pr_debug("%s: %pI4 -> %pI4, %s\n", 1209 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1210 skb->dev ? skb->dev->name : "?"); 1211 kfree_skb(skb); 1212 WARN_ON(1); 1213 return 0; 1214 } 1215 1216 /* 1217 We do not cache source address of outgoing interface, 1218 because it is used only by IP RR, TS and SRR options, 1219 so that it out of fast path. 1220 1221 BTW remember: "addr" is allowed to be not aligned 1222 in IP options! 1223 */ 1224 1225 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt) 1226 { 1227 __be32 src; 1228 1229 if (rt_is_output_route(rt)) 1230 src = ip_hdr(skb)->saddr; 1231 else { 1232 struct fib_result res; 1233 struct flowi4 fl4; 1234 struct iphdr *iph; 1235 1236 iph = ip_hdr(skb); 1237 1238 memset(&fl4, 0, sizeof(fl4)); 1239 fl4.daddr = iph->daddr; 1240 fl4.saddr = iph->saddr; 1241 fl4.flowi4_tos = RT_TOS(iph->tos); 1242 fl4.flowi4_oif = rt->dst.dev->ifindex; 1243 fl4.flowi4_iif = skb->dev->ifindex; 1244 fl4.flowi4_mark = skb->mark; 1245 1246 rcu_read_lock(); 1247 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0) 1248 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res); 1249 else 1250 src = inet_select_addr(rt->dst.dev, 1251 rt_nexthop(rt, iph->daddr), 1252 RT_SCOPE_UNIVERSE); 1253 rcu_read_unlock(); 1254 } 1255 memcpy(addr, &src, 4); 1256 } 1257 1258 #ifdef CONFIG_IP_ROUTE_CLASSID 1259 static void set_class_tag(struct rtable *rt, u32 tag) 1260 { 1261 if (!(rt->dst.tclassid & 0xFFFF)) 1262 rt->dst.tclassid |= tag & 0xFFFF; 1263 if (!(rt->dst.tclassid & 0xFFFF0000)) 1264 rt->dst.tclassid |= tag & 0xFFFF0000; 1265 } 1266 #endif 1267 1268 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1269 { 1270 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr); 1271 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size, 1272 ip_rt_min_advmss); 1273 1274 return min(advmss, IPV4_MAX_PMTU - header_size); 1275 } 1276 1277 static unsigned int ipv4_mtu(const struct dst_entry *dst) 1278 { 1279 const struct rtable *rt = (const struct rtable *) dst; 1280 unsigned int mtu = rt->rt_pmtu; 1281 1282 if (!mtu || time_after_eq(jiffies, rt->dst.expires)) 1283 mtu = dst_metric_raw(dst, RTAX_MTU); 1284 1285 if (mtu) 1286 return mtu; 1287 1288 mtu = READ_ONCE(dst->dev->mtu); 1289 1290 if (unlikely(ip_mtu_locked(dst))) { 1291 if (rt->rt_uses_gateway && mtu > 576) 1292 mtu = 576; 1293 } 1294 1295 mtu = min_t(unsigned int, mtu, IP_MAX_MTU); 1296 1297 return mtu - lwtunnel_headroom(dst->lwtstate, mtu); 1298 } 1299 1300 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr) 1301 { 1302 struct fnhe_hash_bucket *hash; 1303 struct fib_nh_exception *fnhe, __rcu **fnhe_p; 1304 u32 hval = fnhe_hashfun(daddr); 1305 1306 spin_lock_bh(&fnhe_lock); 1307 1308 hash = rcu_dereference_protected(nh->nh_exceptions, 1309 lockdep_is_held(&fnhe_lock)); 1310 hash += hval; 1311 1312 fnhe_p = &hash->chain; 1313 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock)); 1314 while (fnhe) { 1315 if (fnhe->fnhe_daddr == daddr) { 1316 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected( 1317 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock))); 1318 fnhe_flush_routes(fnhe); 1319 kfree_rcu(fnhe, rcu); 1320 break; 1321 } 1322 fnhe_p = &fnhe->fnhe_next; 1323 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1324 lockdep_is_held(&fnhe_lock)); 1325 } 1326 1327 spin_unlock_bh(&fnhe_lock); 1328 } 1329 1330 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr) 1331 { 1332 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions); 1333 struct fib_nh_exception *fnhe; 1334 u32 hval; 1335 1336 if (!hash) 1337 return NULL; 1338 1339 hval = fnhe_hashfun(daddr); 1340 1341 for (fnhe = rcu_dereference(hash[hval].chain); fnhe; 1342 fnhe = rcu_dereference(fnhe->fnhe_next)) { 1343 if (fnhe->fnhe_daddr == daddr) { 1344 if (fnhe->fnhe_expires && 1345 time_after(jiffies, fnhe->fnhe_expires)) { 1346 ip_del_fnhe(nh, daddr); 1347 break; 1348 } 1349 return fnhe; 1350 } 1351 } 1352 return NULL; 1353 } 1354 1355 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, 1356 __be32 daddr, const bool do_cache) 1357 { 1358 bool ret = false; 1359 1360 spin_lock_bh(&fnhe_lock); 1361 1362 if (daddr == fnhe->fnhe_daddr) { 1363 struct rtable __rcu **porig; 1364 struct rtable *orig; 1365 int genid = fnhe_genid(dev_net(rt->dst.dev)); 1366 1367 if (rt_is_input_route(rt)) 1368 porig = &fnhe->fnhe_rth_input; 1369 else 1370 porig = &fnhe->fnhe_rth_output; 1371 orig = rcu_dereference(*porig); 1372 1373 if (fnhe->fnhe_genid != genid) { 1374 fnhe->fnhe_genid = genid; 1375 fnhe->fnhe_gw = 0; 1376 fnhe->fnhe_pmtu = 0; 1377 fnhe->fnhe_expires = 0; 1378 fnhe->fnhe_mtu_locked = false; 1379 fnhe_flush_routes(fnhe); 1380 orig = NULL; 1381 } 1382 fill_route_from_fnhe(rt, fnhe); 1383 if (!rt->rt_gateway) 1384 rt->rt_gateway = daddr; 1385 1386 if (do_cache) { 1387 dst_hold(&rt->dst); 1388 rcu_assign_pointer(*porig, rt); 1389 if (orig) { 1390 dst_dev_put(&orig->dst); 1391 dst_release(&orig->dst); 1392 } 1393 ret = true; 1394 } 1395 1396 fnhe->fnhe_stamp = jiffies; 1397 } 1398 spin_unlock_bh(&fnhe_lock); 1399 1400 return ret; 1401 } 1402 1403 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt) 1404 { 1405 struct rtable *orig, *prev, **p; 1406 bool ret = true; 1407 1408 if (rt_is_input_route(rt)) { 1409 p = (struct rtable **)&nh->nh_rth_input; 1410 } else { 1411 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output); 1412 } 1413 orig = *p; 1414 1415 /* hold dst before doing cmpxchg() to avoid race condition 1416 * on this dst 1417 */ 1418 dst_hold(&rt->dst); 1419 prev = cmpxchg(p, orig, rt); 1420 if (prev == orig) { 1421 if (orig) { 1422 dst_dev_put(&orig->dst); 1423 dst_release(&orig->dst); 1424 } 1425 } else { 1426 dst_release(&rt->dst); 1427 ret = false; 1428 } 1429 1430 return ret; 1431 } 1432 1433 struct uncached_list { 1434 spinlock_t lock; 1435 struct list_head head; 1436 }; 1437 1438 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list); 1439 1440 void rt_add_uncached_list(struct rtable *rt) 1441 { 1442 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list); 1443 1444 rt->rt_uncached_list = ul; 1445 1446 spin_lock_bh(&ul->lock); 1447 list_add_tail(&rt->rt_uncached, &ul->head); 1448 spin_unlock_bh(&ul->lock); 1449 } 1450 1451 void rt_del_uncached_list(struct rtable *rt) 1452 { 1453 if (!list_empty(&rt->rt_uncached)) { 1454 struct uncached_list *ul = rt->rt_uncached_list; 1455 1456 spin_lock_bh(&ul->lock); 1457 list_del(&rt->rt_uncached); 1458 spin_unlock_bh(&ul->lock); 1459 } 1460 } 1461 1462 static void ipv4_dst_destroy(struct dst_entry *dst) 1463 { 1464 struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst); 1465 struct rtable *rt = (struct rtable *)dst; 1466 1467 if (p != &dst_default_metrics && refcount_dec_and_test(&p->refcnt)) 1468 kfree(p); 1469 1470 rt_del_uncached_list(rt); 1471 } 1472 1473 void rt_flush_dev(struct net_device *dev) 1474 { 1475 struct net *net = dev_net(dev); 1476 struct rtable *rt; 1477 int cpu; 1478 1479 for_each_possible_cpu(cpu) { 1480 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 1481 1482 spin_lock_bh(&ul->lock); 1483 list_for_each_entry(rt, &ul->head, rt_uncached) { 1484 if (rt->dst.dev != dev) 1485 continue; 1486 rt->dst.dev = net->loopback_dev; 1487 dev_hold(rt->dst.dev); 1488 dev_put(dev); 1489 } 1490 spin_unlock_bh(&ul->lock); 1491 } 1492 } 1493 1494 static bool rt_cache_valid(const struct rtable *rt) 1495 { 1496 return rt && 1497 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK && 1498 !rt_is_expired(rt); 1499 } 1500 1501 static void rt_set_nexthop(struct rtable *rt, __be32 daddr, 1502 const struct fib_result *res, 1503 struct fib_nh_exception *fnhe, 1504 struct fib_info *fi, u16 type, u32 itag, 1505 const bool do_cache) 1506 { 1507 bool cached = false; 1508 1509 if (fi) { 1510 struct fib_nh *nh = &FIB_RES_NH(*res); 1511 1512 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) { 1513 rt->rt_gateway = nh->nh_gw; 1514 rt->rt_uses_gateway = 1; 1515 } 1516 dst_init_metrics(&rt->dst, fi->fib_metrics->metrics, true); 1517 if (fi->fib_metrics != &dst_default_metrics) { 1518 rt->dst._metrics |= DST_METRICS_REFCOUNTED; 1519 refcount_inc(&fi->fib_metrics->refcnt); 1520 } 1521 #ifdef CONFIG_IP_ROUTE_CLASSID 1522 rt->dst.tclassid = nh->nh_tclassid; 1523 #endif 1524 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate); 1525 if (unlikely(fnhe)) 1526 cached = rt_bind_exception(rt, fnhe, daddr, do_cache); 1527 else if (do_cache) 1528 cached = rt_cache_route(nh, rt); 1529 if (unlikely(!cached)) { 1530 /* Routes we intend to cache in nexthop exception or 1531 * FIB nexthop have the DST_NOCACHE bit clear. 1532 * However, if we are unsuccessful at storing this 1533 * route into the cache we really need to set it. 1534 */ 1535 if (!rt->rt_gateway) 1536 rt->rt_gateway = daddr; 1537 rt_add_uncached_list(rt); 1538 } 1539 } else 1540 rt_add_uncached_list(rt); 1541 1542 #ifdef CONFIG_IP_ROUTE_CLASSID 1543 #ifdef CONFIG_IP_MULTIPLE_TABLES 1544 set_class_tag(rt, res->tclassid); 1545 #endif 1546 set_class_tag(rt, itag); 1547 #endif 1548 } 1549 1550 struct rtable *rt_dst_alloc(struct net_device *dev, 1551 unsigned int flags, u16 type, 1552 bool nopolicy, bool noxfrm, bool will_cache) 1553 { 1554 struct rtable *rt; 1555 1556 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1557 (will_cache ? 0 : DST_HOST) | 1558 (nopolicy ? DST_NOPOLICY : 0) | 1559 (noxfrm ? DST_NOXFRM : 0)); 1560 1561 if (rt) { 1562 rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1563 rt->rt_flags = flags; 1564 rt->rt_type = type; 1565 rt->rt_is_input = 0; 1566 rt->rt_iif = 0; 1567 rt->rt_pmtu = 0; 1568 rt->rt_mtu_locked = 0; 1569 rt->rt_gateway = 0; 1570 rt->rt_uses_gateway = 0; 1571 INIT_LIST_HEAD(&rt->rt_uncached); 1572 1573 rt->dst.output = ip_output; 1574 if (flags & RTCF_LOCAL) 1575 rt->dst.input = ip_local_deliver; 1576 } 1577 1578 return rt; 1579 } 1580 EXPORT_SYMBOL(rt_dst_alloc); 1581 1582 /* called in rcu_read_lock() section */ 1583 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1584 u8 tos, struct net_device *dev, 1585 struct in_device *in_dev, u32 *itag) 1586 { 1587 int err; 1588 1589 /* Primary sanity checks. */ 1590 if (!in_dev) 1591 return -EINVAL; 1592 1593 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) || 1594 skb->protocol != htons(ETH_P_IP)) 1595 return -EINVAL; 1596 1597 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev)) 1598 return -EINVAL; 1599 1600 if (ipv4_is_zeronet(saddr)) { 1601 if (!ipv4_is_local_multicast(daddr)) 1602 return -EINVAL; 1603 } else { 1604 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 1605 in_dev, itag); 1606 if (err < 0) 1607 return err; 1608 } 1609 return 0; 1610 } 1611 1612 /* called in rcu_read_lock() section */ 1613 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1614 u8 tos, struct net_device *dev, int our) 1615 { 1616 struct in_device *in_dev = __in_dev_get_rcu(dev); 1617 unsigned int flags = RTCF_MULTICAST; 1618 struct rtable *rth; 1619 u32 itag = 0; 1620 int err; 1621 1622 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag); 1623 if (err) 1624 return err; 1625 1626 if (our) 1627 flags |= RTCF_LOCAL; 1628 1629 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST, 1630 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false); 1631 if (!rth) 1632 return -ENOBUFS; 1633 1634 #ifdef CONFIG_IP_ROUTE_CLASSID 1635 rth->dst.tclassid = itag; 1636 #endif 1637 rth->dst.output = ip_rt_bug; 1638 rth->rt_is_input= 1; 1639 1640 #ifdef CONFIG_IP_MROUTE 1641 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1642 rth->dst.input = ip_mr_input; 1643 #endif 1644 RT_CACHE_STAT_INC(in_slow_mc); 1645 1646 skb_dst_set(skb, &rth->dst); 1647 return 0; 1648 } 1649 1650 1651 static void ip_handle_martian_source(struct net_device *dev, 1652 struct in_device *in_dev, 1653 struct sk_buff *skb, 1654 __be32 daddr, 1655 __be32 saddr) 1656 { 1657 RT_CACHE_STAT_INC(in_martian_src); 1658 #ifdef CONFIG_IP_ROUTE_VERBOSE 1659 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1660 /* 1661 * RFC1812 recommendation, if source is martian, 1662 * the only hint is MAC header. 1663 */ 1664 pr_warn("martian source %pI4 from %pI4, on dev %s\n", 1665 &daddr, &saddr, dev->name); 1666 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1667 print_hex_dump(KERN_WARNING, "ll header: ", 1668 DUMP_PREFIX_OFFSET, 16, 1, 1669 skb_mac_header(skb), 1670 dev->hard_header_len, true); 1671 } 1672 } 1673 #endif 1674 } 1675 1676 /* called in rcu_read_lock() section */ 1677 static int __mkroute_input(struct sk_buff *skb, 1678 const struct fib_result *res, 1679 struct in_device *in_dev, 1680 __be32 daddr, __be32 saddr, u32 tos) 1681 { 1682 struct fib_nh_exception *fnhe; 1683 struct rtable *rth; 1684 int err; 1685 struct in_device *out_dev; 1686 bool do_cache; 1687 u32 itag = 0; 1688 1689 /* get a working reference to the output device */ 1690 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res)); 1691 if (!out_dev) { 1692 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n"); 1693 return -EINVAL; 1694 } 1695 1696 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res), 1697 in_dev->dev, in_dev, &itag); 1698 if (err < 0) { 1699 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1700 saddr); 1701 1702 goto cleanup; 1703 } 1704 1705 do_cache = res->fi && !itag; 1706 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) && 1707 skb->protocol == htons(ETH_P_IP) && 1708 (IN_DEV_SHARED_MEDIA(out_dev) || 1709 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) 1710 IPCB(skb)->flags |= IPSKB_DOREDIRECT; 1711 1712 if (skb->protocol != htons(ETH_P_IP)) { 1713 /* Not IP (i.e. ARP). Do not create route, if it is 1714 * invalid for proxy arp. DNAT routes are always valid. 1715 * 1716 * Proxy arp feature have been extended to allow, ARP 1717 * replies back to the same interface, to support 1718 * Private VLAN switch technologies. See arp.c. 1719 */ 1720 if (out_dev == in_dev && 1721 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 1722 err = -EINVAL; 1723 goto cleanup; 1724 } 1725 } 1726 1727 fnhe = find_exception(&FIB_RES_NH(*res), daddr); 1728 if (do_cache) { 1729 if (fnhe) 1730 rth = rcu_dereference(fnhe->fnhe_rth_input); 1731 else 1732 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input); 1733 if (rt_cache_valid(rth)) { 1734 skb_dst_set_noref(skb, &rth->dst); 1735 goto out; 1736 } 1737 } 1738 1739 rth = rt_dst_alloc(out_dev->dev, 0, res->type, 1740 IN_DEV_CONF_GET(in_dev, NOPOLICY), 1741 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache); 1742 if (!rth) { 1743 err = -ENOBUFS; 1744 goto cleanup; 1745 } 1746 1747 rth->rt_is_input = 1; 1748 RT_CACHE_STAT_INC(in_slow_tot); 1749 1750 rth->dst.input = ip_forward; 1751 1752 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag, 1753 do_cache); 1754 lwtunnel_set_redirect(&rth->dst); 1755 skb_dst_set(skb, &rth->dst); 1756 out: 1757 err = 0; 1758 cleanup: 1759 return err; 1760 } 1761 1762 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1763 /* To make ICMP packets follow the right flow, the multipath hash is 1764 * calculated from the inner IP addresses. 1765 */ 1766 static void ip_multipath_l3_keys(const struct sk_buff *skb, 1767 struct flow_keys *hash_keys) 1768 { 1769 const struct iphdr *outer_iph = ip_hdr(skb); 1770 const struct iphdr *key_iph = outer_iph; 1771 const struct iphdr *inner_iph; 1772 const struct icmphdr *icmph; 1773 struct iphdr _inner_iph; 1774 struct icmphdr _icmph; 1775 1776 if (likely(outer_iph->protocol != IPPROTO_ICMP)) 1777 goto out; 1778 1779 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0)) 1780 goto out; 1781 1782 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph), 1783 &_icmph); 1784 if (!icmph) 1785 goto out; 1786 1787 if (icmph->type != ICMP_DEST_UNREACH && 1788 icmph->type != ICMP_REDIRECT && 1789 icmph->type != ICMP_TIME_EXCEEDED && 1790 icmph->type != ICMP_PARAMETERPROB) 1791 goto out; 1792 1793 inner_iph = skb_header_pointer(skb, 1794 outer_iph->ihl * 4 + sizeof(_icmph), 1795 sizeof(_inner_iph), &_inner_iph); 1796 if (!inner_iph) 1797 goto out; 1798 1799 key_iph = inner_iph; 1800 out: 1801 hash_keys->addrs.v4addrs.src = key_iph->saddr; 1802 hash_keys->addrs.v4addrs.dst = key_iph->daddr; 1803 } 1804 1805 /* if skb is set it will be used and fl4 can be NULL */ 1806 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4, 1807 const struct sk_buff *skb, struct flow_keys *flkeys) 1808 { 1809 struct flow_keys hash_keys; 1810 u32 mhash; 1811 1812 switch (net->ipv4.sysctl_fib_multipath_hash_policy) { 1813 case 0: 1814 memset(&hash_keys, 0, sizeof(hash_keys)); 1815 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1816 if (skb) { 1817 ip_multipath_l3_keys(skb, &hash_keys); 1818 } else { 1819 hash_keys.addrs.v4addrs.src = fl4->saddr; 1820 hash_keys.addrs.v4addrs.dst = fl4->daddr; 1821 } 1822 break; 1823 case 1: 1824 /* skb is currently provided only when forwarding */ 1825 if (skb) { 1826 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 1827 struct flow_keys keys; 1828 1829 /* short-circuit if we already have L4 hash present */ 1830 if (skb->l4_hash) 1831 return skb_get_hash_raw(skb) >> 1; 1832 1833 memset(&hash_keys, 0, sizeof(hash_keys)); 1834 1835 if (!flkeys) { 1836 skb_flow_dissect_flow_keys(skb, &keys, flag); 1837 flkeys = &keys; 1838 } 1839 1840 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1841 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src; 1842 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst; 1843 hash_keys.ports.src = flkeys->ports.src; 1844 hash_keys.ports.dst = flkeys->ports.dst; 1845 hash_keys.basic.ip_proto = flkeys->basic.ip_proto; 1846 } else { 1847 memset(&hash_keys, 0, sizeof(hash_keys)); 1848 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1849 hash_keys.addrs.v4addrs.src = fl4->saddr; 1850 hash_keys.addrs.v4addrs.dst = fl4->daddr; 1851 hash_keys.ports.src = fl4->fl4_sport; 1852 hash_keys.ports.dst = fl4->fl4_dport; 1853 hash_keys.basic.ip_proto = fl4->flowi4_proto; 1854 } 1855 break; 1856 } 1857 mhash = flow_hash_from_keys(&hash_keys); 1858 1859 return mhash >> 1; 1860 } 1861 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 1862 1863 static int ip_mkroute_input(struct sk_buff *skb, 1864 struct fib_result *res, 1865 struct in_device *in_dev, 1866 __be32 daddr, __be32 saddr, u32 tos, 1867 struct flow_keys *hkeys) 1868 { 1869 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1870 if (res->fi && res->fi->fib_nhs > 1) { 1871 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys); 1872 1873 fib_select_multipath(res, h); 1874 } 1875 #endif 1876 1877 /* create a routing cache entry */ 1878 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos); 1879 } 1880 1881 /* 1882 * NOTE. We drop all the packets that has local source 1883 * addresses, because every properly looped back packet 1884 * must have correct destination already attached by output routine. 1885 * 1886 * Such approach solves two big problems: 1887 * 1. Not simplex devices are handled properly. 1888 * 2. IP spoofing attempts are filtered with 100% of guarantee. 1889 * called with rcu_read_lock() 1890 */ 1891 1892 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1893 u8 tos, struct net_device *dev, 1894 struct fib_result *res) 1895 { 1896 struct in_device *in_dev = __in_dev_get_rcu(dev); 1897 struct flow_keys *flkeys = NULL, _flkeys; 1898 struct net *net = dev_net(dev); 1899 struct ip_tunnel_info *tun_info; 1900 int err = -EINVAL; 1901 unsigned int flags = 0; 1902 u32 itag = 0; 1903 struct rtable *rth; 1904 struct flowi4 fl4; 1905 bool do_cache; 1906 1907 /* IP on this device is disabled. */ 1908 1909 if (!in_dev) 1910 goto out; 1911 1912 /* Check for the most weird martians, which can be not detected 1913 by fib_lookup. 1914 */ 1915 1916 tun_info = skb_tunnel_info(skb); 1917 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX)) 1918 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id; 1919 else 1920 fl4.flowi4_tun_key.tun_id = 0; 1921 skb_dst_drop(skb); 1922 1923 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 1924 goto martian_source; 1925 1926 res->fi = NULL; 1927 res->table = NULL; 1928 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 1929 goto brd_input; 1930 1931 /* Accept zero addresses only to limited broadcast; 1932 * I even do not know to fix it or not. Waiting for complains :-) 1933 */ 1934 if (ipv4_is_zeronet(saddr)) 1935 goto martian_source; 1936 1937 if (ipv4_is_zeronet(daddr)) 1938 goto martian_destination; 1939 1940 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(), 1941 * and call it once if daddr or/and saddr are loopback addresses 1942 */ 1943 if (ipv4_is_loopback(daddr)) { 1944 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 1945 goto martian_destination; 1946 } else if (ipv4_is_loopback(saddr)) { 1947 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 1948 goto martian_source; 1949 } 1950 1951 /* 1952 * Now we are ready to route packet. 1953 */ 1954 fl4.flowi4_oif = 0; 1955 fl4.flowi4_iif = dev->ifindex; 1956 fl4.flowi4_mark = skb->mark; 1957 fl4.flowi4_tos = tos; 1958 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 1959 fl4.flowi4_flags = 0; 1960 fl4.daddr = daddr; 1961 fl4.saddr = saddr; 1962 fl4.flowi4_uid = sock_net_uid(net, NULL); 1963 1964 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) { 1965 flkeys = &_flkeys; 1966 } else { 1967 fl4.flowi4_proto = 0; 1968 fl4.fl4_sport = 0; 1969 fl4.fl4_dport = 0; 1970 } 1971 1972 err = fib_lookup(net, &fl4, res, 0); 1973 if (err != 0) { 1974 if (!IN_DEV_FORWARD(in_dev)) 1975 err = -EHOSTUNREACH; 1976 goto no_route; 1977 } 1978 1979 if (res->type == RTN_BROADCAST) 1980 goto brd_input; 1981 1982 if (res->type == RTN_LOCAL) { 1983 err = fib_validate_source(skb, saddr, daddr, tos, 1984 0, dev, in_dev, &itag); 1985 if (err < 0) 1986 goto martian_source; 1987 goto local_input; 1988 } 1989 1990 if (!IN_DEV_FORWARD(in_dev)) { 1991 err = -EHOSTUNREACH; 1992 goto no_route; 1993 } 1994 if (res->type != RTN_UNICAST) 1995 goto martian_destination; 1996 1997 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys); 1998 out: return err; 1999 2000 brd_input: 2001 if (skb->protocol != htons(ETH_P_IP)) 2002 goto e_inval; 2003 2004 if (!ipv4_is_zeronet(saddr)) { 2005 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 2006 in_dev, &itag); 2007 if (err < 0) 2008 goto martian_source; 2009 } 2010 flags |= RTCF_BROADCAST; 2011 res->type = RTN_BROADCAST; 2012 RT_CACHE_STAT_INC(in_brd); 2013 2014 local_input: 2015 do_cache = false; 2016 if (res->fi) { 2017 if (!itag) { 2018 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input); 2019 if (rt_cache_valid(rth)) { 2020 skb_dst_set_noref(skb, &rth->dst); 2021 err = 0; 2022 goto out; 2023 } 2024 do_cache = true; 2025 } 2026 } 2027 2028 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev, 2029 flags | RTCF_LOCAL, res->type, 2030 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache); 2031 if (!rth) 2032 goto e_nobufs; 2033 2034 rth->dst.output= ip_rt_bug; 2035 #ifdef CONFIG_IP_ROUTE_CLASSID 2036 rth->dst.tclassid = itag; 2037 #endif 2038 rth->rt_is_input = 1; 2039 2040 RT_CACHE_STAT_INC(in_slow_tot); 2041 if (res->type == RTN_UNREACHABLE) { 2042 rth->dst.input= ip_error; 2043 rth->dst.error= -err; 2044 rth->rt_flags &= ~RTCF_LOCAL; 2045 } 2046 2047 if (do_cache) { 2048 struct fib_nh *nh = &FIB_RES_NH(*res); 2049 2050 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate); 2051 if (lwtunnel_input_redirect(rth->dst.lwtstate)) { 2052 WARN_ON(rth->dst.input == lwtunnel_input); 2053 rth->dst.lwtstate->orig_input = rth->dst.input; 2054 rth->dst.input = lwtunnel_input; 2055 } 2056 2057 if (unlikely(!rt_cache_route(nh, rth))) 2058 rt_add_uncached_list(rth); 2059 } 2060 skb_dst_set(skb, &rth->dst); 2061 err = 0; 2062 goto out; 2063 2064 no_route: 2065 RT_CACHE_STAT_INC(in_no_route); 2066 res->type = RTN_UNREACHABLE; 2067 res->fi = NULL; 2068 res->table = NULL; 2069 goto local_input; 2070 2071 /* 2072 * Do not cache martian addresses: they should be logged (RFC1812) 2073 */ 2074 martian_destination: 2075 RT_CACHE_STAT_INC(in_martian_dst); 2076 #ifdef CONFIG_IP_ROUTE_VERBOSE 2077 if (IN_DEV_LOG_MARTIANS(in_dev)) 2078 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n", 2079 &daddr, &saddr, dev->name); 2080 #endif 2081 2082 e_inval: 2083 err = -EINVAL; 2084 goto out; 2085 2086 e_nobufs: 2087 err = -ENOBUFS; 2088 goto out; 2089 2090 martian_source: 2091 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2092 goto out; 2093 } 2094 2095 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2096 u8 tos, struct net_device *dev) 2097 { 2098 struct fib_result res; 2099 int err; 2100 2101 tos &= IPTOS_RT_MASK; 2102 rcu_read_lock(); 2103 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res); 2104 rcu_read_unlock(); 2105 2106 return err; 2107 } 2108 EXPORT_SYMBOL(ip_route_input_noref); 2109 2110 /* called with rcu_read_lock held */ 2111 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2112 u8 tos, struct net_device *dev, struct fib_result *res) 2113 { 2114 /* Multicast recognition logic is moved from route cache to here. 2115 The problem was that too many Ethernet cards have broken/missing 2116 hardware multicast filters :-( As result the host on multicasting 2117 network acquires a lot of useless route cache entries, sort of 2118 SDR messages from all the world. Now we try to get rid of them. 2119 Really, provided software IP multicast filter is organized 2120 reasonably (at least, hashed), it does not result in a slowdown 2121 comparing with route cache reject entries. 2122 Note, that multicast routers are not affected, because 2123 route cache entry is created eventually. 2124 */ 2125 if (ipv4_is_multicast(daddr)) { 2126 struct in_device *in_dev = __in_dev_get_rcu(dev); 2127 int our = 0; 2128 int err = -EINVAL; 2129 2130 if (in_dev) 2131 our = ip_check_mc_rcu(in_dev, daddr, saddr, 2132 ip_hdr(skb)->protocol); 2133 2134 /* check l3 master if no match yet */ 2135 if ((!in_dev || !our) && netif_is_l3_slave(dev)) { 2136 struct in_device *l3_in_dev; 2137 2138 l3_in_dev = __in_dev_get_rcu(skb->dev); 2139 if (l3_in_dev) 2140 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr, 2141 ip_hdr(skb)->protocol); 2142 } 2143 2144 if (our 2145 #ifdef CONFIG_IP_MROUTE 2146 || 2147 (!ipv4_is_local_multicast(daddr) && 2148 IN_DEV_MFORWARD(in_dev)) 2149 #endif 2150 ) { 2151 err = ip_route_input_mc(skb, daddr, saddr, 2152 tos, dev, our); 2153 } 2154 return err; 2155 } 2156 2157 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res); 2158 } 2159 2160 /* called with rcu_read_lock() */ 2161 static struct rtable *__mkroute_output(const struct fib_result *res, 2162 const struct flowi4 *fl4, int orig_oif, 2163 struct net_device *dev_out, 2164 unsigned int flags) 2165 { 2166 struct fib_info *fi = res->fi; 2167 struct fib_nh_exception *fnhe; 2168 struct in_device *in_dev; 2169 u16 type = res->type; 2170 struct rtable *rth; 2171 bool do_cache; 2172 2173 in_dev = __in_dev_get_rcu(dev_out); 2174 if (!in_dev) 2175 return ERR_PTR(-EINVAL); 2176 2177 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 2178 if (ipv4_is_loopback(fl4->saddr) && 2179 !(dev_out->flags & IFF_LOOPBACK) && 2180 !netif_is_l3_master(dev_out)) 2181 return ERR_PTR(-EINVAL); 2182 2183 if (ipv4_is_lbcast(fl4->daddr)) 2184 type = RTN_BROADCAST; 2185 else if (ipv4_is_multicast(fl4->daddr)) 2186 type = RTN_MULTICAST; 2187 else if (ipv4_is_zeronet(fl4->daddr)) 2188 return ERR_PTR(-EINVAL); 2189 2190 if (dev_out->flags & IFF_LOOPBACK) 2191 flags |= RTCF_LOCAL; 2192 2193 do_cache = true; 2194 if (type == RTN_BROADCAST) { 2195 flags |= RTCF_BROADCAST | RTCF_LOCAL; 2196 fi = NULL; 2197 } else if (type == RTN_MULTICAST) { 2198 flags |= RTCF_MULTICAST | RTCF_LOCAL; 2199 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr, 2200 fl4->flowi4_proto)) 2201 flags &= ~RTCF_LOCAL; 2202 else 2203 do_cache = false; 2204 /* If multicast route do not exist use 2205 * default one, but do not gateway in this case. 2206 * Yes, it is hack. 2207 */ 2208 if (fi && res->prefixlen < 4) 2209 fi = NULL; 2210 } else if ((type == RTN_LOCAL) && (orig_oif != 0) && 2211 (orig_oif != dev_out->ifindex)) { 2212 /* For local routes that require a particular output interface 2213 * we do not want to cache the result. Caching the result 2214 * causes incorrect behaviour when there are multiple source 2215 * addresses on the interface, the end result being that if the 2216 * intended recipient is waiting on that interface for the 2217 * packet he won't receive it because it will be delivered on 2218 * the loopback interface and the IP_PKTINFO ipi_ifindex will 2219 * be set to the loopback interface as well. 2220 */ 2221 do_cache = false; 2222 } 2223 2224 fnhe = NULL; 2225 do_cache &= fi != NULL; 2226 if (fi) { 2227 struct rtable __rcu **prth; 2228 struct fib_nh *nh = &FIB_RES_NH(*res); 2229 2230 fnhe = find_exception(nh, fl4->daddr); 2231 if (!do_cache) 2232 goto add; 2233 if (fnhe) { 2234 prth = &fnhe->fnhe_rth_output; 2235 } else { 2236 if (unlikely(fl4->flowi4_flags & 2237 FLOWI_FLAG_KNOWN_NH && 2238 !(nh->nh_gw && 2239 nh->nh_scope == RT_SCOPE_LINK))) { 2240 do_cache = false; 2241 goto add; 2242 } 2243 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output); 2244 } 2245 rth = rcu_dereference(*prth); 2246 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst)) 2247 return rth; 2248 } 2249 2250 add: 2251 rth = rt_dst_alloc(dev_out, flags, type, 2252 IN_DEV_CONF_GET(in_dev, NOPOLICY), 2253 IN_DEV_CONF_GET(in_dev, NOXFRM), 2254 do_cache); 2255 if (!rth) 2256 return ERR_PTR(-ENOBUFS); 2257 2258 rth->rt_iif = orig_oif; 2259 2260 RT_CACHE_STAT_INC(out_slow_tot); 2261 2262 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 2263 if (flags & RTCF_LOCAL && 2264 !(dev_out->flags & IFF_LOOPBACK)) { 2265 rth->dst.output = ip_mc_output; 2266 RT_CACHE_STAT_INC(out_slow_mc); 2267 } 2268 #ifdef CONFIG_IP_MROUTE 2269 if (type == RTN_MULTICAST) { 2270 if (IN_DEV_MFORWARD(in_dev) && 2271 !ipv4_is_local_multicast(fl4->daddr)) { 2272 rth->dst.input = ip_mr_input; 2273 rth->dst.output = ip_mc_output; 2274 } 2275 } 2276 #endif 2277 } 2278 2279 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache); 2280 lwtunnel_set_redirect(&rth->dst); 2281 2282 return rth; 2283 } 2284 2285 /* 2286 * Major route resolver routine. 2287 */ 2288 2289 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4, 2290 const struct sk_buff *skb) 2291 { 2292 __u8 tos = RT_FL_TOS(fl4); 2293 struct fib_result res = { 2294 .type = RTN_UNSPEC, 2295 .fi = NULL, 2296 .table = NULL, 2297 .tclassid = 0, 2298 }; 2299 struct rtable *rth; 2300 2301 fl4->flowi4_iif = LOOPBACK_IFINDEX; 2302 fl4->flowi4_tos = tos & IPTOS_RT_MASK; 2303 fl4->flowi4_scope = ((tos & RTO_ONLINK) ? 2304 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE); 2305 2306 rcu_read_lock(); 2307 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb); 2308 rcu_read_unlock(); 2309 2310 return rth; 2311 } 2312 EXPORT_SYMBOL_GPL(ip_route_output_key_hash); 2313 2314 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4, 2315 struct fib_result *res, 2316 const struct sk_buff *skb) 2317 { 2318 struct net_device *dev_out = NULL; 2319 int orig_oif = fl4->flowi4_oif; 2320 unsigned int flags = 0; 2321 struct rtable *rth; 2322 int err = -ENETUNREACH; 2323 2324 if (fl4->saddr) { 2325 rth = ERR_PTR(-EINVAL); 2326 if (ipv4_is_multicast(fl4->saddr) || 2327 ipv4_is_lbcast(fl4->saddr) || 2328 ipv4_is_zeronet(fl4->saddr)) 2329 goto out; 2330 2331 /* I removed check for oif == dev_out->oif here. 2332 It was wrong for two reasons: 2333 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 2334 is assigned to multiple interfaces. 2335 2. Moreover, we are allowed to send packets with saddr 2336 of another iface. --ANK 2337 */ 2338 2339 if (fl4->flowi4_oif == 0 && 2340 (ipv4_is_multicast(fl4->daddr) || 2341 ipv4_is_lbcast(fl4->daddr))) { 2342 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2343 dev_out = __ip_dev_find(net, fl4->saddr, false); 2344 if (!dev_out) 2345 goto out; 2346 2347 /* Special hack: user can direct multicasts 2348 and limited broadcast via necessary interface 2349 without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 2350 This hack is not just for fun, it allows 2351 vic,vat and friends to work. 2352 They bind socket to loopback, set ttl to zero 2353 and expect that it will work. 2354 From the viewpoint of routing cache they are broken, 2355 because we are not allowed to build multicast path 2356 with loopback source addr (look, routing cache 2357 cannot know, that ttl is zero, so that packet 2358 will not leave this host and route is valid). 2359 Luckily, this hack is good workaround. 2360 */ 2361 2362 fl4->flowi4_oif = dev_out->ifindex; 2363 goto make_route; 2364 } 2365 2366 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 2367 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2368 if (!__ip_dev_find(net, fl4->saddr, false)) 2369 goto out; 2370 } 2371 } 2372 2373 2374 if (fl4->flowi4_oif) { 2375 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif); 2376 rth = ERR_PTR(-ENODEV); 2377 if (!dev_out) 2378 goto out; 2379 2380 /* RACE: Check return value of inet_select_addr instead. */ 2381 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 2382 rth = ERR_PTR(-ENETUNREACH); 2383 goto out; 2384 } 2385 if (ipv4_is_local_multicast(fl4->daddr) || 2386 ipv4_is_lbcast(fl4->daddr) || 2387 fl4->flowi4_proto == IPPROTO_IGMP) { 2388 if (!fl4->saddr) 2389 fl4->saddr = inet_select_addr(dev_out, 0, 2390 RT_SCOPE_LINK); 2391 goto make_route; 2392 } 2393 if (!fl4->saddr) { 2394 if (ipv4_is_multicast(fl4->daddr)) 2395 fl4->saddr = inet_select_addr(dev_out, 0, 2396 fl4->flowi4_scope); 2397 else if (!fl4->daddr) 2398 fl4->saddr = inet_select_addr(dev_out, 0, 2399 RT_SCOPE_HOST); 2400 } 2401 } 2402 2403 if (!fl4->daddr) { 2404 fl4->daddr = fl4->saddr; 2405 if (!fl4->daddr) 2406 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK); 2407 dev_out = net->loopback_dev; 2408 fl4->flowi4_oif = LOOPBACK_IFINDEX; 2409 res->type = RTN_LOCAL; 2410 flags |= RTCF_LOCAL; 2411 goto make_route; 2412 } 2413 2414 err = fib_lookup(net, fl4, res, 0); 2415 if (err) { 2416 res->fi = NULL; 2417 res->table = NULL; 2418 if (fl4->flowi4_oif && 2419 (ipv4_is_multicast(fl4->daddr) || 2420 !netif_index_is_l3_master(net, fl4->flowi4_oif))) { 2421 /* Apparently, routing tables are wrong. Assume, 2422 that the destination is on link. 2423 2424 WHY? DW. 2425 Because we are allowed to send to iface 2426 even if it has NO routes and NO assigned 2427 addresses. When oif is specified, routing 2428 tables are looked up with only one purpose: 2429 to catch if destination is gatewayed, rather than 2430 direct. Moreover, if MSG_DONTROUTE is set, 2431 we send packet, ignoring both routing tables 2432 and ifaddr state. --ANK 2433 2434 2435 We could make it even if oif is unknown, 2436 likely IPv6, but we do not. 2437 */ 2438 2439 if (fl4->saddr == 0) 2440 fl4->saddr = inet_select_addr(dev_out, 0, 2441 RT_SCOPE_LINK); 2442 res->type = RTN_UNICAST; 2443 goto make_route; 2444 } 2445 rth = ERR_PTR(err); 2446 goto out; 2447 } 2448 2449 if (res->type == RTN_LOCAL) { 2450 if (!fl4->saddr) { 2451 if (res->fi->fib_prefsrc) 2452 fl4->saddr = res->fi->fib_prefsrc; 2453 else 2454 fl4->saddr = fl4->daddr; 2455 } 2456 2457 /* L3 master device is the loopback for that domain */ 2458 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? : 2459 net->loopback_dev; 2460 2461 /* make sure orig_oif points to fib result device even 2462 * though packet rx/tx happens over loopback or l3mdev 2463 */ 2464 orig_oif = FIB_RES_OIF(*res); 2465 2466 fl4->flowi4_oif = dev_out->ifindex; 2467 flags |= RTCF_LOCAL; 2468 goto make_route; 2469 } 2470 2471 fib_select_path(net, res, fl4, skb); 2472 2473 dev_out = FIB_RES_DEV(*res); 2474 fl4->flowi4_oif = dev_out->ifindex; 2475 2476 2477 make_route: 2478 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags); 2479 2480 out: 2481 return rth; 2482 } 2483 2484 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie) 2485 { 2486 return NULL; 2487 } 2488 2489 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst) 2490 { 2491 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU); 2492 2493 return mtu ? : dst->dev->mtu; 2494 } 2495 2496 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk, 2497 struct sk_buff *skb, u32 mtu) 2498 { 2499 } 2500 2501 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk, 2502 struct sk_buff *skb) 2503 { 2504 } 2505 2506 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst, 2507 unsigned long old) 2508 { 2509 return NULL; 2510 } 2511 2512 static struct dst_ops ipv4_dst_blackhole_ops = { 2513 .family = AF_INET, 2514 .check = ipv4_blackhole_dst_check, 2515 .mtu = ipv4_blackhole_mtu, 2516 .default_advmss = ipv4_default_advmss, 2517 .update_pmtu = ipv4_rt_blackhole_update_pmtu, 2518 .redirect = ipv4_rt_blackhole_redirect, 2519 .cow_metrics = ipv4_rt_blackhole_cow_metrics, 2520 .neigh_lookup = ipv4_neigh_lookup, 2521 }; 2522 2523 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2524 { 2525 struct rtable *ort = (struct rtable *) dst_orig; 2526 struct rtable *rt; 2527 2528 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0); 2529 if (rt) { 2530 struct dst_entry *new = &rt->dst; 2531 2532 new->__use = 1; 2533 new->input = dst_discard; 2534 new->output = dst_discard_out; 2535 2536 new->dev = net->loopback_dev; 2537 if (new->dev) 2538 dev_hold(new->dev); 2539 2540 rt->rt_is_input = ort->rt_is_input; 2541 rt->rt_iif = ort->rt_iif; 2542 rt->rt_pmtu = ort->rt_pmtu; 2543 rt->rt_mtu_locked = ort->rt_mtu_locked; 2544 2545 rt->rt_genid = rt_genid_ipv4(net); 2546 rt->rt_flags = ort->rt_flags; 2547 rt->rt_type = ort->rt_type; 2548 rt->rt_gateway = ort->rt_gateway; 2549 rt->rt_uses_gateway = ort->rt_uses_gateway; 2550 2551 INIT_LIST_HEAD(&rt->rt_uncached); 2552 } 2553 2554 dst_release(dst_orig); 2555 2556 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2557 } 2558 2559 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2560 const struct sock *sk) 2561 { 2562 struct rtable *rt = __ip_route_output_key(net, flp4); 2563 2564 if (IS_ERR(rt)) 2565 return rt; 2566 2567 if (flp4->flowi4_proto) 2568 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst, 2569 flowi4_to_flowi(flp4), 2570 sk, 0); 2571 2572 return rt; 2573 } 2574 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2575 2576 /* called with rcu_read_lock held */ 2577 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id, 2578 struct flowi4 *fl4, struct sk_buff *skb, u32 portid, 2579 u32 seq) 2580 { 2581 struct rtable *rt = skb_rtable(skb); 2582 struct rtmsg *r; 2583 struct nlmsghdr *nlh; 2584 unsigned long expires = 0; 2585 u32 error; 2586 u32 metrics[RTAX_MAX]; 2587 2588 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0); 2589 if (!nlh) 2590 return -EMSGSIZE; 2591 2592 r = nlmsg_data(nlh); 2593 r->rtm_family = AF_INET; 2594 r->rtm_dst_len = 32; 2595 r->rtm_src_len = 0; 2596 r->rtm_tos = fl4->flowi4_tos; 2597 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT; 2598 if (nla_put_u32(skb, RTA_TABLE, table_id)) 2599 goto nla_put_failure; 2600 r->rtm_type = rt->rt_type; 2601 r->rtm_scope = RT_SCOPE_UNIVERSE; 2602 r->rtm_protocol = RTPROT_UNSPEC; 2603 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 2604 if (rt->rt_flags & RTCF_NOTIFY) 2605 r->rtm_flags |= RTM_F_NOTIFY; 2606 if (IPCB(skb)->flags & IPSKB_DOREDIRECT) 2607 r->rtm_flags |= RTCF_DOREDIRECT; 2608 2609 if (nla_put_in_addr(skb, RTA_DST, dst)) 2610 goto nla_put_failure; 2611 if (src) { 2612 r->rtm_src_len = 32; 2613 if (nla_put_in_addr(skb, RTA_SRC, src)) 2614 goto nla_put_failure; 2615 } 2616 if (rt->dst.dev && 2617 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex)) 2618 goto nla_put_failure; 2619 #ifdef CONFIG_IP_ROUTE_CLASSID 2620 if (rt->dst.tclassid && 2621 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid)) 2622 goto nla_put_failure; 2623 #endif 2624 if (!rt_is_input_route(rt) && 2625 fl4->saddr != src) { 2626 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr)) 2627 goto nla_put_failure; 2628 } 2629 if (rt->rt_uses_gateway && 2630 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway)) 2631 goto nla_put_failure; 2632 2633 expires = rt->dst.expires; 2634 if (expires) { 2635 unsigned long now = jiffies; 2636 2637 if (time_before(now, expires)) 2638 expires -= now; 2639 else 2640 expires = 0; 2641 } 2642 2643 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics)); 2644 if (rt->rt_pmtu && expires) 2645 metrics[RTAX_MTU - 1] = rt->rt_pmtu; 2646 if (rt->rt_mtu_locked && expires) 2647 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU); 2648 if (rtnetlink_put_metrics(skb, metrics) < 0) 2649 goto nla_put_failure; 2650 2651 if (fl4->flowi4_mark && 2652 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark)) 2653 goto nla_put_failure; 2654 2655 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) && 2656 nla_put_u32(skb, RTA_UID, 2657 from_kuid_munged(current_user_ns(), fl4->flowi4_uid))) 2658 goto nla_put_failure; 2659 2660 error = rt->dst.error; 2661 2662 if (rt_is_input_route(rt)) { 2663 #ifdef CONFIG_IP_MROUTE 2664 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) && 2665 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) { 2666 int err = ipmr_get_route(net, skb, 2667 fl4->saddr, fl4->daddr, 2668 r, portid); 2669 2670 if (err <= 0) { 2671 if (err == 0) 2672 return 0; 2673 goto nla_put_failure; 2674 } 2675 } else 2676 #endif 2677 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex)) 2678 goto nla_put_failure; 2679 } 2680 2681 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0) 2682 goto nla_put_failure; 2683 2684 nlmsg_end(skb, nlh); 2685 return 0; 2686 2687 nla_put_failure: 2688 nlmsg_cancel(skb, nlh); 2689 return -EMSGSIZE; 2690 } 2691 2692 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, 2693 struct netlink_ext_ack *extack) 2694 { 2695 struct net *net = sock_net(in_skb->sk); 2696 struct rtmsg *rtm; 2697 struct nlattr *tb[RTA_MAX+1]; 2698 struct fib_result res = {}; 2699 struct rtable *rt = NULL; 2700 struct flowi4 fl4; 2701 __be32 dst = 0; 2702 __be32 src = 0; 2703 u32 iif; 2704 int err; 2705 int mark; 2706 struct sk_buff *skb; 2707 u32 table_id = RT_TABLE_MAIN; 2708 kuid_t uid; 2709 2710 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy, 2711 extack); 2712 if (err < 0) 2713 goto errout; 2714 2715 rtm = nlmsg_data(nlh); 2716 2717 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 2718 if (!skb) { 2719 err = -ENOBUFS; 2720 goto errout; 2721 } 2722 2723 /* Reserve room for dummy headers, this skb can pass 2724 through good chunk of routing engine. 2725 */ 2726 skb_reset_mac_header(skb); 2727 skb_reset_network_header(skb); 2728 2729 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0; 2730 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0; 2731 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0; 2732 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0; 2733 if (tb[RTA_UID]) 2734 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID])); 2735 else 2736 uid = (iif ? INVALID_UID : current_uid()); 2737 2738 /* Bugfix: need to give ip_route_input enough of an IP header to 2739 * not gag. 2740 */ 2741 ip_hdr(skb)->protocol = IPPROTO_UDP; 2742 ip_hdr(skb)->saddr = src; 2743 ip_hdr(skb)->daddr = dst; 2744 2745 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr)); 2746 2747 memset(&fl4, 0, sizeof(fl4)); 2748 fl4.daddr = dst; 2749 fl4.saddr = src; 2750 fl4.flowi4_tos = rtm->rtm_tos; 2751 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0; 2752 fl4.flowi4_mark = mark; 2753 fl4.flowi4_uid = uid; 2754 2755 rcu_read_lock(); 2756 2757 if (iif) { 2758 struct net_device *dev; 2759 2760 dev = dev_get_by_index_rcu(net, iif); 2761 if (!dev) { 2762 err = -ENODEV; 2763 goto errout_free; 2764 } 2765 2766 skb->protocol = htons(ETH_P_IP); 2767 skb->dev = dev; 2768 skb->mark = mark; 2769 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos, 2770 dev, &res); 2771 2772 rt = skb_rtable(skb); 2773 if (err == 0 && rt->dst.error) 2774 err = -rt->dst.error; 2775 } else { 2776 fl4.flowi4_iif = LOOPBACK_IFINDEX; 2777 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb); 2778 err = 0; 2779 if (IS_ERR(rt)) 2780 err = PTR_ERR(rt); 2781 else 2782 skb_dst_set(skb, &rt->dst); 2783 } 2784 2785 if (err) 2786 goto errout_free; 2787 2788 if (rtm->rtm_flags & RTM_F_NOTIFY) 2789 rt->rt_flags |= RTCF_NOTIFY; 2790 2791 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE) 2792 table_id = res.table ? res.table->tb_id : 0; 2793 2794 if (rtm->rtm_flags & RTM_F_FIB_MATCH) { 2795 if (!res.fi) { 2796 err = fib_props[res.type].error; 2797 if (!err) 2798 err = -EHOSTUNREACH; 2799 goto errout_free; 2800 } 2801 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid, 2802 nlh->nlmsg_seq, RTM_NEWROUTE, table_id, 2803 rt->rt_type, res.prefix, res.prefixlen, 2804 fl4.flowi4_tos, res.fi, 0); 2805 } else { 2806 err = rt_fill_info(net, dst, src, table_id, &fl4, skb, 2807 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq); 2808 } 2809 if (err < 0) 2810 goto errout_free; 2811 2812 rcu_read_unlock(); 2813 2814 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 2815 errout: 2816 return err; 2817 2818 errout_free: 2819 rcu_read_unlock(); 2820 kfree_skb(skb); 2821 goto errout; 2822 } 2823 2824 void ip_rt_multicast_event(struct in_device *in_dev) 2825 { 2826 rt_cache_flush(dev_net(in_dev->dev)); 2827 } 2828 2829 #ifdef CONFIG_SYSCTL 2830 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 2831 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 2832 static int ip_rt_gc_elasticity __read_mostly = 8; 2833 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU; 2834 2835 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write, 2836 void __user *buffer, 2837 size_t *lenp, loff_t *ppos) 2838 { 2839 struct net *net = (struct net *)__ctl->extra1; 2840 2841 if (write) { 2842 rt_cache_flush(net); 2843 fnhe_genid_bump(net); 2844 return 0; 2845 } 2846 2847 return -EINVAL; 2848 } 2849 2850 static struct ctl_table ipv4_route_table[] = { 2851 { 2852 .procname = "gc_thresh", 2853 .data = &ipv4_dst_ops.gc_thresh, 2854 .maxlen = sizeof(int), 2855 .mode = 0644, 2856 .proc_handler = proc_dointvec, 2857 }, 2858 { 2859 .procname = "max_size", 2860 .data = &ip_rt_max_size, 2861 .maxlen = sizeof(int), 2862 .mode = 0644, 2863 .proc_handler = proc_dointvec, 2864 }, 2865 { 2866 /* Deprecated. Use gc_min_interval_ms */ 2867 2868 .procname = "gc_min_interval", 2869 .data = &ip_rt_gc_min_interval, 2870 .maxlen = sizeof(int), 2871 .mode = 0644, 2872 .proc_handler = proc_dointvec_jiffies, 2873 }, 2874 { 2875 .procname = "gc_min_interval_ms", 2876 .data = &ip_rt_gc_min_interval, 2877 .maxlen = sizeof(int), 2878 .mode = 0644, 2879 .proc_handler = proc_dointvec_ms_jiffies, 2880 }, 2881 { 2882 .procname = "gc_timeout", 2883 .data = &ip_rt_gc_timeout, 2884 .maxlen = sizeof(int), 2885 .mode = 0644, 2886 .proc_handler = proc_dointvec_jiffies, 2887 }, 2888 { 2889 .procname = "gc_interval", 2890 .data = &ip_rt_gc_interval, 2891 .maxlen = sizeof(int), 2892 .mode = 0644, 2893 .proc_handler = proc_dointvec_jiffies, 2894 }, 2895 { 2896 .procname = "redirect_load", 2897 .data = &ip_rt_redirect_load, 2898 .maxlen = sizeof(int), 2899 .mode = 0644, 2900 .proc_handler = proc_dointvec, 2901 }, 2902 { 2903 .procname = "redirect_number", 2904 .data = &ip_rt_redirect_number, 2905 .maxlen = sizeof(int), 2906 .mode = 0644, 2907 .proc_handler = proc_dointvec, 2908 }, 2909 { 2910 .procname = "redirect_silence", 2911 .data = &ip_rt_redirect_silence, 2912 .maxlen = sizeof(int), 2913 .mode = 0644, 2914 .proc_handler = proc_dointvec, 2915 }, 2916 { 2917 .procname = "error_cost", 2918 .data = &ip_rt_error_cost, 2919 .maxlen = sizeof(int), 2920 .mode = 0644, 2921 .proc_handler = proc_dointvec, 2922 }, 2923 { 2924 .procname = "error_burst", 2925 .data = &ip_rt_error_burst, 2926 .maxlen = sizeof(int), 2927 .mode = 0644, 2928 .proc_handler = proc_dointvec, 2929 }, 2930 { 2931 .procname = "gc_elasticity", 2932 .data = &ip_rt_gc_elasticity, 2933 .maxlen = sizeof(int), 2934 .mode = 0644, 2935 .proc_handler = proc_dointvec, 2936 }, 2937 { 2938 .procname = "mtu_expires", 2939 .data = &ip_rt_mtu_expires, 2940 .maxlen = sizeof(int), 2941 .mode = 0644, 2942 .proc_handler = proc_dointvec_jiffies, 2943 }, 2944 { 2945 .procname = "min_pmtu", 2946 .data = &ip_rt_min_pmtu, 2947 .maxlen = sizeof(int), 2948 .mode = 0644, 2949 .proc_handler = proc_dointvec_minmax, 2950 .extra1 = &ip_min_valid_pmtu, 2951 }, 2952 { 2953 .procname = "min_adv_mss", 2954 .data = &ip_rt_min_advmss, 2955 .maxlen = sizeof(int), 2956 .mode = 0644, 2957 .proc_handler = proc_dointvec, 2958 }, 2959 { } 2960 }; 2961 2962 static struct ctl_table ipv4_route_flush_table[] = { 2963 { 2964 .procname = "flush", 2965 .maxlen = sizeof(int), 2966 .mode = 0200, 2967 .proc_handler = ipv4_sysctl_rtcache_flush, 2968 }, 2969 { }, 2970 }; 2971 2972 static __net_init int sysctl_route_net_init(struct net *net) 2973 { 2974 struct ctl_table *tbl; 2975 2976 tbl = ipv4_route_flush_table; 2977 if (!net_eq(net, &init_net)) { 2978 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL); 2979 if (!tbl) 2980 goto err_dup; 2981 2982 /* Don't export sysctls to unprivileged users */ 2983 if (net->user_ns != &init_user_ns) 2984 tbl[0].procname = NULL; 2985 } 2986 tbl[0].extra1 = net; 2987 2988 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl); 2989 if (!net->ipv4.route_hdr) 2990 goto err_reg; 2991 return 0; 2992 2993 err_reg: 2994 if (tbl != ipv4_route_flush_table) 2995 kfree(tbl); 2996 err_dup: 2997 return -ENOMEM; 2998 } 2999 3000 static __net_exit void sysctl_route_net_exit(struct net *net) 3001 { 3002 struct ctl_table *tbl; 3003 3004 tbl = net->ipv4.route_hdr->ctl_table_arg; 3005 unregister_net_sysctl_table(net->ipv4.route_hdr); 3006 BUG_ON(tbl == ipv4_route_flush_table); 3007 kfree(tbl); 3008 } 3009 3010 static __net_initdata struct pernet_operations sysctl_route_ops = { 3011 .init = sysctl_route_net_init, 3012 .exit = sysctl_route_net_exit, 3013 }; 3014 #endif 3015 3016 static __net_init int rt_genid_init(struct net *net) 3017 { 3018 atomic_set(&net->ipv4.rt_genid, 0); 3019 atomic_set(&net->fnhe_genid, 0); 3020 atomic_set(&net->ipv4.dev_addr_genid, get_random_int()); 3021 return 0; 3022 } 3023 3024 static __net_initdata struct pernet_operations rt_genid_ops = { 3025 .init = rt_genid_init, 3026 }; 3027 3028 static int __net_init ipv4_inetpeer_init(struct net *net) 3029 { 3030 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL); 3031 3032 if (!bp) 3033 return -ENOMEM; 3034 inet_peer_base_init(bp); 3035 net->ipv4.peers = bp; 3036 return 0; 3037 } 3038 3039 static void __net_exit ipv4_inetpeer_exit(struct net *net) 3040 { 3041 struct inet_peer_base *bp = net->ipv4.peers; 3042 3043 net->ipv4.peers = NULL; 3044 inetpeer_invalidate_tree(bp); 3045 kfree(bp); 3046 } 3047 3048 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = { 3049 .init = ipv4_inetpeer_init, 3050 .exit = ipv4_inetpeer_exit, 3051 }; 3052 3053 #ifdef CONFIG_IP_ROUTE_CLASSID 3054 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 3055 #endif /* CONFIG_IP_ROUTE_CLASSID */ 3056 3057 int __init ip_rt_init(void) 3058 { 3059 int cpu; 3060 3061 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL); 3062 if (!ip_idents) 3063 panic("IP: failed to allocate ip_idents\n"); 3064 3065 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents)); 3066 3067 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL); 3068 if (!ip_tstamps) 3069 panic("IP: failed to allocate ip_tstamps\n"); 3070 3071 for_each_possible_cpu(cpu) { 3072 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 3073 3074 INIT_LIST_HEAD(&ul->head); 3075 spin_lock_init(&ul->lock); 3076 } 3077 #ifdef CONFIG_IP_ROUTE_CLASSID 3078 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 3079 if (!ip_rt_acct) 3080 panic("IP: failed to allocate ip_rt_acct\n"); 3081 #endif 3082 3083 ipv4_dst_ops.kmem_cachep = 3084 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0, 3085 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 3086 3087 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 3088 3089 if (dst_entries_init(&ipv4_dst_ops) < 0) 3090 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 3091 3092 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 3093 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 3094 3095 ipv4_dst_ops.gc_thresh = ~0; 3096 ip_rt_max_size = INT_MAX; 3097 3098 devinet_init(); 3099 ip_fib_init(); 3100 3101 if (ip_rt_proc_init()) 3102 pr_err("Unable to create route proc files\n"); 3103 #ifdef CONFIG_XFRM 3104 xfrm_init(); 3105 xfrm4_init(); 3106 #endif 3107 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, 3108 RTNL_FLAG_DOIT_UNLOCKED); 3109 3110 #ifdef CONFIG_SYSCTL 3111 register_pernet_subsys(&sysctl_route_ops); 3112 #endif 3113 register_pernet_subsys(&rt_genid_ops); 3114 register_pernet_subsys(&ipv4_inetpeer_ops); 3115 return 0; 3116 } 3117 3118 #ifdef CONFIG_SYSCTL 3119 /* 3120 * We really need to sanitize the damn ipv4 init order, then all 3121 * this nonsense will go away. 3122 */ 3123 void __init ip_static_sysctl_init(void) 3124 { 3125 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table); 3126 } 3127 #endif 3128