1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * ROUTE - implementation of the IP router. 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Alan Cox, <gw4pts@gw4pts.ampr.org> 11 * Linus Torvalds, <Linus.Torvalds@helsinki.fi> 12 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 13 * 14 * Fixes: 15 * Alan Cox : Verify area fixes. 16 * Alan Cox : cli() protects routing changes 17 * Rui Oliveira : ICMP routing table updates 18 * (rco@di.uminho.pt) Routing table insertion and update 19 * Linus Torvalds : Rewrote bits to be sensible 20 * Alan Cox : Added BSD route gw semantics 21 * Alan Cox : Super /proc >4K 22 * Alan Cox : MTU in route table 23 * Alan Cox : MSS actually. Also added the window 24 * clamper. 25 * Sam Lantinga : Fixed route matching in rt_del() 26 * Alan Cox : Routing cache support. 27 * Alan Cox : Removed compatibility cruft. 28 * Alan Cox : RTF_REJECT support. 29 * Alan Cox : TCP irtt support. 30 * Jonathan Naylor : Added Metric support. 31 * Miquel van Smoorenburg : BSD API fixes. 32 * Miquel van Smoorenburg : Metrics. 33 * Alan Cox : Use __u32 properly 34 * Alan Cox : Aligned routing errors more closely with BSD 35 * our system is still very different. 36 * Alan Cox : Faster /proc handling 37 * Alexey Kuznetsov : Massive rework to support tree based routing, 38 * routing caches and better behaviour. 39 * 40 * Olaf Erb : irtt wasn't being copied right. 41 * Bjorn Ekwall : Kerneld route support. 42 * Alan Cox : Multicast fixed (I hope) 43 * Pavel Krauz : Limited broadcast fixed 44 * Mike McLagan : Routing by source 45 * Alexey Kuznetsov : End of old history. Split to fib.c and 46 * route.c and rewritten from scratch. 47 * Andi Kleen : Load-limit warning messages. 48 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 49 * Vitaly E. Lavrov : Race condition in ip_route_input_slow. 50 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow. 51 * Vladimir V. Ivanov : IP rule info (flowid) is really useful. 52 * Marc Boucher : routing by fwmark 53 * Robert Olsson : Added rt_cache statistics 54 * Arnaldo C. Melo : Convert proc stuff to seq_file 55 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes. 56 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect 57 * Ilia Sotnikov : Removed TOS from hash calculations 58 * 59 * This program is free software; you can redistribute it and/or 60 * modify it under the terms of the GNU General Public License 61 * as published by the Free Software Foundation; either version 62 * 2 of the License, or (at your option) any later version. 63 */ 64 65 #define pr_fmt(fmt) "IPv4: " fmt 66 67 #include <linux/module.h> 68 #include <linux/uaccess.h> 69 #include <linux/bitops.h> 70 #include <linux/types.h> 71 #include <linux/kernel.h> 72 #include <linux/mm.h> 73 #include <linux/string.h> 74 #include <linux/socket.h> 75 #include <linux/sockios.h> 76 #include <linux/errno.h> 77 #include <linux/in.h> 78 #include <linux/inet.h> 79 #include <linux/netdevice.h> 80 #include <linux/proc_fs.h> 81 #include <linux/init.h> 82 #include <linux/skbuff.h> 83 #include <linux/inetdevice.h> 84 #include <linux/igmp.h> 85 #include <linux/pkt_sched.h> 86 #include <linux/mroute.h> 87 #include <linux/netfilter_ipv4.h> 88 #include <linux/random.h> 89 #include <linux/rcupdate.h> 90 #include <linux/times.h> 91 #include <linux/slab.h> 92 #include <linux/jhash.h> 93 #include <net/dst.h> 94 #include <net/dst_metadata.h> 95 #include <net/net_namespace.h> 96 #include <net/protocol.h> 97 #include <net/ip.h> 98 #include <net/route.h> 99 #include <net/inetpeer.h> 100 #include <net/sock.h> 101 #include <net/ip_fib.h> 102 #include <net/arp.h> 103 #include <net/tcp.h> 104 #include <net/icmp.h> 105 #include <net/xfrm.h> 106 #include <net/lwtunnel.h> 107 #include <net/netevent.h> 108 #include <net/rtnetlink.h> 109 #ifdef CONFIG_SYSCTL 110 #include <linux/sysctl.h> 111 #include <linux/kmemleak.h> 112 #endif 113 #include <net/secure_seq.h> 114 #include <net/ip_tunnels.h> 115 #include <net/l3mdev.h> 116 117 #include "fib_lookup.h" 118 119 #define RT_FL_TOS(oldflp4) \ 120 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)) 121 122 #define RT_GC_TIMEOUT (300*HZ) 123 124 static int ip_rt_max_size; 125 static int ip_rt_redirect_number __read_mostly = 9; 126 static int ip_rt_redirect_load __read_mostly = HZ / 50; 127 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1)); 128 static int ip_rt_error_cost __read_mostly = HZ; 129 static int ip_rt_error_burst __read_mostly = 5 * HZ; 130 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ; 131 static u32 ip_rt_min_pmtu __read_mostly = 512 + 20 + 20; 132 static int ip_rt_min_advmss __read_mostly = 256; 133 134 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT; 135 136 /* 137 * Interface to generic destination cache. 138 */ 139 140 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie); 141 static unsigned int ipv4_default_advmss(const struct dst_entry *dst); 142 static unsigned int ipv4_mtu(const struct dst_entry *dst); 143 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst); 144 static void ipv4_link_failure(struct sk_buff *skb); 145 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 146 struct sk_buff *skb, u32 mtu); 147 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, 148 struct sk_buff *skb); 149 static void ipv4_dst_destroy(struct dst_entry *dst); 150 151 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old) 152 { 153 WARN_ON(1); 154 return NULL; 155 } 156 157 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 158 struct sk_buff *skb, 159 const void *daddr); 160 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr); 161 162 static struct dst_ops ipv4_dst_ops = { 163 .family = AF_INET, 164 .check = ipv4_dst_check, 165 .default_advmss = ipv4_default_advmss, 166 .mtu = ipv4_mtu, 167 .cow_metrics = ipv4_cow_metrics, 168 .destroy = ipv4_dst_destroy, 169 .negative_advice = ipv4_negative_advice, 170 .link_failure = ipv4_link_failure, 171 .update_pmtu = ip_rt_update_pmtu, 172 .redirect = ip_do_redirect, 173 .local_out = __ip_local_out, 174 .neigh_lookup = ipv4_neigh_lookup, 175 .confirm_neigh = ipv4_confirm_neigh, 176 }; 177 178 #define ECN_OR_COST(class) TC_PRIO_##class 179 180 const __u8 ip_tos2prio[16] = { 181 TC_PRIO_BESTEFFORT, 182 ECN_OR_COST(BESTEFFORT), 183 TC_PRIO_BESTEFFORT, 184 ECN_OR_COST(BESTEFFORT), 185 TC_PRIO_BULK, 186 ECN_OR_COST(BULK), 187 TC_PRIO_BULK, 188 ECN_OR_COST(BULK), 189 TC_PRIO_INTERACTIVE, 190 ECN_OR_COST(INTERACTIVE), 191 TC_PRIO_INTERACTIVE, 192 ECN_OR_COST(INTERACTIVE), 193 TC_PRIO_INTERACTIVE_BULK, 194 ECN_OR_COST(INTERACTIVE_BULK), 195 TC_PRIO_INTERACTIVE_BULK, 196 ECN_OR_COST(INTERACTIVE_BULK) 197 }; 198 EXPORT_SYMBOL(ip_tos2prio); 199 200 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat); 201 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field) 202 203 #ifdef CONFIG_PROC_FS 204 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos) 205 { 206 if (*pos) 207 return NULL; 208 return SEQ_START_TOKEN; 209 } 210 211 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos) 212 { 213 ++*pos; 214 return NULL; 215 } 216 217 static void rt_cache_seq_stop(struct seq_file *seq, void *v) 218 { 219 } 220 221 static int rt_cache_seq_show(struct seq_file *seq, void *v) 222 { 223 if (v == SEQ_START_TOKEN) 224 seq_printf(seq, "%-127s\n", 225 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t" 226 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t" 227 "HHUptod\tSpecDst"); 228 return 0; 229 } 230 231 static const struct seq_operations rt_cache_seq_ops = { 232 .start = rt_cache_seq_start, 233 .next = rt_cache_seq_next, 234 .stop = rt_cache_seq_stop, 235 .show = rt_cache_seq_show, 236 }; 237 238 static int rt_cache_seq_open(struct inode *inode, struct file *file) 239 { 240 return seq_open(file, &rt_cache_seq_ops); 241 } 242 243 static const struct file_operations rt_cache_seq_fops = { 244 .open = rt_cache_seq_open, 245 .read = seq_read, 246 .llseek = seq_lseek, 247 .release = seq_release, 248 }; 249 250 251 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos) 252 { 253 int cpu; 254 255 if (*pos == 0) 256 return SEQ_START_TOKEN; 257 258 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) { 259 if (!cpu_possible(cpu)) 260 continue; 261 *pos = cpu+1; 262 return &per_cpu(rt_cache_stat, cpu); 263 } 264 return NULL; 265 } 266 267 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos) 268 { 269 int cpu; 270 271 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) { 272 if (!cpu_possible(cpu)) 273 continue; 274 *pos = cpu+1; 275 return &per_cpu(rt_cache_stat, cpu); 276 } 277 return NULL; 278 279 } 280 281 static void rt_cpu_seq_stop(struct seq_file *seq, void *v) 282 { 283 284 } 285 286 static int rt_cpu_seq_show(struct seq_file *seq, void *v) 287 { 288 struct rt_cache_stat *st = v; 289 290 if (v == SEQ_START_TOKEN) { 291 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"); 292 return 0; 293 } 294 295 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x " 296 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n", 297 dst_entries_get_slow(&ipv4_dst_ops), 298 0, /* st->in_hit */ 299 st->in_slow_tot, 300 st->in_slow_mc, 301 st->in_no_route, 302 st->in_brd, 303 st->in_martian_dst, 304 st->in_martian_src, 305 306 0, /* st->out_hit */ 307 st->out_slow_tot, 308 st->out_slow_mc, 309 310 0, /* st->gc_total */ 311 0, /* st->gc_ignored */ 312 0, /* st->gc_goal_miss */ 313 0, /* st->gc_dst_overflow */ 314 0, /* st->in_hlist_search */ 315 0 /* st->out_hlist_search */ 316 ); 317 return 0; 318 } 319 320 static const struct seq_operations rt_cpu_seq_ops = { 321 .start = rt_cpu_seq_start, 322 .next = rt_cpu_seq_next, 323 .stop = rt_cpu_seq_stop, 324 .show = rt_cpu_seq_show, 325 }; 326 327 328 static int rt_cpu_seq_open(struct inode *inode, struct file *file) 329 { 330 return seq_open(file, &rt_cpu_seq_ops); 331 } 332 333 static const struct file_operations rt_cpu_seq_fops = { 334 .open = rt_cpu_seq_open, 335 .read = seq_read, 336 .llseek = seq_lseek, 337 .release = seq_release, 338 }; 339 340 #ifdef CONFIG_IP_ROUTE_CLASSID 341 static int rt_acct_proc_show(struct seq_file *m, void *v) 342 { 343 struct ip_rt_acct *dst, *src; 344 unsigned int i, j; 345 346 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL); 347 if (!dst) 348 return -ENOMEM; 349 350 for_each_possible_cpu(i) { 351 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i); 352 for (j = 0; j < 256; j++) { 353 dst[j].o_bytes += src[j].o_bytes; 354 dst[j].o_packets += src[j].o_packets; 355 dst[j].i_bytes += src[j].i_bytes; 356 dst[j].i_packets += src[j].i_packets; 357 } 358 } 359 360 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct)); 361 kfree(dst); 362 return 0; 363 } 364 365 static int rt_acct_proc_open(struct inode *inode, struct file *file) 366 { 367 return single_open(file, rt_acct_proc_show, NULL); 368 } 369 370 static const struct file_operations rt_acct_proc_fops = { 371 .open = rt_acct_proc_open, 372 .read = seq_read, 373 .llseek = seq_lseek, 374 .release = single_release, 375 }; 376 #endif 377 378 static int __net_init ip_rt_do_proc_init(struct net *net) 379 { 380 struct proc_dir_entry *pde; 381 382 pde = proc_create("rt_cache", 0444, net->proc_net, 383 &rt_cache_seq_fops); 384 if (!pde) 385 goto err1; 386 387 pde = proc_create("rt_cache", 0444, 388 net->proc_net_stat, &rt_cpu_seq_fops); 389 if (!pde) 390 goto err2; 391 392 #ifdef CONFIG_IP_ROUTE_CLASSID 393 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops); 394 if (!pde) 395 goto err3; 396 #endif 397 return 0; 398 399 #ifdef CONFIG_IP_ROUTE_CLASSID 400 err3: 401 remove_proc_entry("rt_cache", net->proc_net_stat); 402 #endif 403 err2: 404 remove_proc_entry("rt_cache", net->proc_net); 405 err1: 406 return -ENOMEM; 407 } 408 409 static void __net_exit ip_rt_do_proc_exit(struct net *net) 410 { 411 remove_proc_entry("rt_cache", net->proc_net_stat); 412 remove_proc_entry("rt_cache", net->proc_net); 413 #ifdef CONFIG_IP_ROUTE_CLASSID 414 remove_proc_entry("rt_acct", net->proc_net); 415 #endif 416 } 417 418 static struct pernet_operations ip_rt_proc_ops __net_initdata = { 419 .init = ip_rt_do_proc_init, 420 .exit = ip_rt_do_proc_exit, 421 }; 422 423 static int __init ip_rt_proc_init(void) 424 { 425 return register_pernet_subsys(&ip_rt_proc_ops); 426 } 427 428 #else 429 static inline int ip_rt_proc_init(void) 430 { 431 return 0; 432 } 433 #endif /* CONFIG_PROC_FS */ 434 435 static inline bool rt_is_expired(const struct rtable *rth) 436 { 437 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev)); 438 } 439 440 void rt_cache_flush(struct net *net) 441 { 442 rt_genid_bump_ipv4(net); 443 } 444 445 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 446 struct sk_buff *skb, 447 const void *daddr) 448 { 449 struct net_device *dev = dst->dev; 450 const __be32 *pkey = daddr; 451 const struct rtable *rt; 452 struct neighbour *n; 453 454 rt = (const struct rtable *) dst; 455 if (rt->rt_gateway) 456 pkey = (const __be32 *) &rt->rt_gateway; 457 else if (skb) 458 pkey = &ip_hdr(skb)->daddr; 459 460 n = __ipv4_neigh_lookup(dev, *(__force u32 *)pkey); 461 if (n) 462 return n; 463 return neigh_create(&arp_tbl, pkey, dev); 464 } 465 466 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr) 467 { 468 struct net_device *dev = dst->dev; 469 const __be32 *pkey = daddr; 470 const struct rtable *rt; 471 472 rt = (const struct rtable *)dst; 473 if (rt->rt_gateway) 474 pkey = (const __be32 *)&rt->rt_gateway; 475 else if (!daddr || 476 (rt->rt_flags & 477 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) 478 return; 479 480 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey); 481 } 482 483 #define IP_IDENTS_SZ 2048u 484 485 static atomic_t *ip_idents __read_mostly; 486 static u32 *ip_tstamps __read_mostly; 487 488 /* In order to protect privacy, we add a perturbation to identifiers 489 * if one generator is seldom used. This makes hard for an attacker 490 * to infer how many packets were sent between two points in time. 491 */ 492 u32 ip_idents_reserve(u32 hash, int segs) 493 { 494 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ; 495 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ; 496 u32 old = READ_ONCE(*p_tstamp); 497 u32 now = (u32)jiffies; 498 u32 new, delta = 0; 499 500 if (old != now && cmpxchg(p_tstamp, old, now) == old) 501 delta = prandom_u32_max(now - old); 502 503 /* Do not use atomic_add_return() as it makes UBSAN unhappy */ 504 do { 505 old = (u32)atomic_read(p_id); 506 new = old + delta + segs; 507 } while (atomic_cmpxchg(p_id, old, new) != old); 508 509 return new - segs; 510 } 511 EXPORT_SYMBOL(ip_idents_reserve); 512 513 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs) 514 { 515 static u32 ip_idents_hashrnd __read_mostly; 516 u32 hash, id; 517 518 net_get_random_once(&ip_idents_hashrnd, sizeof(ip_idents_hashrnd)); 519 520 hash = jhash_3words((__force u32)iph->daddr, 521 (__force u32)iph->saddr, 522 iph->protocol ^ net_hash_mix(net), 523 ip_idents_hashrnd); 524 id = ip_idents_reserve(hash, segs); 525 iph->id = htons(id); 526 } 527 EXPORT_SYMBOL(__ip_select_ident); 528 529 static void __build_flow_key(const struct net *net, struct flowi4 *fl4, 530 const struct sock *sk, 531 const struct iphdr *iph, 532 int oif, u8 tos, 533 u8 prot, u32 mark, int flow_flags) 534 { 535 if (sk) { 536 const struct inet_sock *inet = inet_sk(sk); 537 538 oif = sk->sk_bound_dev_if; 539 mark = sk->sk_mark; 540 tos = RT_CONN_FLAGS(sk); 541 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol; 542 } 543 flowi4_init_output(fl4, oif, mark, tos, 544 RT_SCOPE_UNIVERSE, prot, 545 flow_flags, 546 iph->daddr, iph->saddr, 0, 0, 547 sock_net_uid(net, sk)); 548 } 549 550 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb, 551 const struct sock *sk) 552 { 553 const struct net *net = dev_net(skb->dev); 554 const struct iphdr *iph = ip_hdr(skb); 555 int oif = skb->dev->ifindex; 556 u8 tos = RT_TOS(iph->tos); 557 u8 prot = iph->protocol; 558 u32 mark = skb->mark; 559 560 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0); 561 } 562 563 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk) 564 { 565 const struct inet_sock *inet = inet_sk(sk); 566 const struct ip_options_rcu *inet_opt; 567 __be32 daddr = inet->inet_daddr; 568 569 rcu_read_lock(); 570 inet_opt = rcu_dereference(inet->inet_opt); 571 if (inet_opt && inet_opt->opt.srr) 572 daddr = inet_opt->opt.faddr; 573 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark, 574 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 575 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol, 576 inet_sk_flowi_flags(sk), 577 daddr, inet->inet_saddr, 0, 0, sk->sk_uid); 578 rcu_read_unlock(); 579 } 580 581 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk, 582 const struct sk_buff *skb) 583 { 584 if (skb) 585 build_skb_flow_key(fl4, skb, sk); 586 else 587 build_sk_flow_key(fl4, sk); 588 } 589 590 static DEFINE_SPINLOCK(fnhe_lock); 591 592 static void fnhe_flush_routes(struct fib_nh_exception *fnhe) 593 { 594 struct rtable *rt; 595 596 rt = rcu_dereference(fnhe->fnhe_rth_input); 597 if (rt) { 598 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL); 599 dst_dev_put(&rt->dst); 600 dst_release(&rt->dst); 601 } 602 rt = rcu_dereference(fnhe->fnhe_rth_output); 603 if (rt) { 604 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL); 605 dst_dev_put(&rt->dst); 606 dst_release(&rt->dst); 607 } 608 } 609 610 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash) 611 { 612 struct fib_nh_exception *fnhe, *oldest; 613 614 oldest = rcu_dereference(hash->chain); 615 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe; 616 fnhe = rcu_dereference(fnhe->fnhe_next)) { 617 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) 618 oldest = fnhe; 619 } 620 fnhe_flush_routes(oldest); 621 return oldest; 622 } 623 624 static inline u32 fnhe_hashfun(__be32 daddr) 625 { 626 static u32 fnhe_hashrnd __read_mostly; 627 u32 hval; 628 629 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd)); 630 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd); 631 return hash_32(hval, FNHE_HASH_SHIFT); 632 } 633 634 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe) 635 { 636 rt->rt_pmtu = fnhe->fnhe_pmtu; 637 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked; 638 rt->dst.expires = fnhe->fnhe_expires; 639 640 if (fnhe->fnhe_gw) { 641 rt->rt_flags |= RTCF_REDIRECTED; 642 rt->rt_gateway = fnhe->fnhe_gw; 643 rt->rt_uses_gateway = 1; 644 } 645 } 646 647 static void update_or_create_fnhe(struct fib_nh *nh, __be32 daddr, __be32 gw, 648 u32 pmtu, bool lock, unsigned long expires) 649 { 650 struct fnhe_hash_bucket *hash; 651 struct fib_nh_exception *fnhe; 652 struct rtable *rt; 653 u32 genid, hval; 654 unsigned int i; 655 int depth; 656 657 genid = fnhe_genid(dev_net(nh->nh_dev)); 658 hval = fnhe_hashfun(daddr); 659 660 spin_lock_bh(&fnhe_lock); 661 662 hash = rcu_dereference(nh->nh_exceptions); 663 if (!hash) { 664 hash = kzalloc(FNHE_HASH_SIZE * sizeof(*hash), GFP_ATOMIC); 665 if (!hash) 666 goto out_unlock; 667 rcu_assign_pointer(nh->nh_exceptions, hash); 668 } 669 670 hash += hval; 671 672 depth = 0; 673 for (fnhe = rcu_dereference(hash->chain); fnhe; 674 fnhe = rcu_dereference(fnhe->fnhe_next)) { 675 if (fnhe->fnhe_daddr == daddr) 676 break; 677 depth++; 678 } 679 680 if (fnhe) { 681 if (fnhe->fnhe_genid != genid) 682 fnhe->fnhe_genid = genid; 683 if (gw) 684 fnhe->fnhe_gw = gw; 685 if (pmtu) { 686 fnhe->fnhe_pmtu = pmtu; 687 fnhe->fnhe_mtu_locked = lock; 688 } 689 fnhe->fnhe_expires = max(1UL, expires); 690 /* Update all cached dsts too */ 691 rt = rcu_dereference(fnhe->fnhe_rth_input); 692 if (rt) 693 fill_route_from_fnhe(rt, fnhe); 694 rt = rcu_dereference(fnhe->fnhe_rth_output); 695 if (rt) 696 fill_route_from_fnhe(rt, fnhe); 697 } else { 698 if (depth > FNHE_RECLAIM_DEPTH) 699 fnhe = fnhe_oldest(hash); 700 else { 701 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC); 702 if (!fnhe) 703 goto out_unlock; 704 705 fnhe->fnhe_next = hash->chain; 706 rcu_assign_pointer(hash->chain, fnhe); 707 } 708 fnhe->fnhe_genid = genid; 709 fnhe->fnhe_daddr = daddr; 710 fnhe->fnhe_gw = gw; 711 fnhe->fnhe_pmtu = pmtu; 712 fnhe->fnhe_mtu_locked = lock; 713 fnhe->fnhe_expires = expires; 714 715 /* Exception created; mark the cached routes for the nexthop 716 * stale, so anyone caching it rechecks if this exception 717 * applies to them. 718 */ 719 rt = rcu_dereference(nh->nh_rth_input); 720 if (rt) 721 rt->dst.obsolete = DST_OBSOLETE_KILL; 722 723 for_each_possible_cpu(i) { 724 struct rtable __rcu **prt; 725 prt = per_cpu_ptr(nh->nh_pcpu_rth_output, i); 726 rt = rcu_dereference(*prt); 727 if (rt) 728 rt->dst.obsolete = DST_OBSOLETE_KILL; 729 } 730 } 731 732 fnhe->fnhe_stamp = jiffies; 733 734 out_unlock: 735 spin_unlock_bh(&fnhe_lock); 736 } 737 738 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4, 739 bool kill_route) 740 { 741 __be32 new_gw = icmp_hdr(skb)->un.gateway; 742 __be32 old_gw = ip_hdr(skb)->saddr; 743 struct net_device *dev = skb->dev; 744 struct in_device *in_dev; 745 struct fib_result res; 746 struct neighbour *n; 747 struct net *net; 748 749 switch (icmp_hdr(skb)->code & 7) { 750 case ICMP_REDIR_NET: 751 case ICMP_REDIR_NETTOS: 752 case ICMP_REDIR_HOST: 753 case ICMP_REDIR_HOSTTOS: 754 break; 755 756 default: 757 return; 758 } 759 760 if (rt->rt_gateway != old_gw) 761 return; 762 763 in_dev = __in_dev_get_rcu(dev); 764 if (!in_dev) 765 return; 766 767 net = dev_net(dev); 768 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) || 769 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) || 770 ipv4_is_zeronet(new_gw)) 771 goto reject_redirect; 772 773 if (!IN_DEV_SHARED_MEDIA(in_dev)) { 774 if (!inet_addr_onlink(in_dev, new_gw, old_gw)) 775 goto reject_redirect; 776 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) 777 goto reject_redirect; 778 } else { 779 if (inet_addr_type(net, new_gw) != RTN_UNICAST) 780 goto reject_redirect; 781 } 782 783 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw); 784 if (!n) 785 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev); 786 if (!IS_ERR(n)) { 787 if (!(n->nud_state & NUD_VALID)) { 788 neigh_event_send(n, NULL); 789 } else { 790 if (fib_lookup(net, fl4, &res, 0) == 0) { 791 struct fib_nh *nh = &FIB_RES_NH(res); 792 793 update_or_create_fnhe(nh, fl4->daddr, new_gw, 794 0, false, 795 jiffies + ip_rt_gc_timeout); 796 } 797 if (kill_route) 798 rt->dst.obsolete = DST_OBSOLETE_KILL; 799 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n); 800 } 801 neigh_release(n); 802 } 803 return; 804 805 reject_redirect: 806 #ifdef CONFIG_IP_ROUTE_VERBOSE 807 if (IN_DEV_LOG_MARTIANS(in_dev)) { 808 const struct iphdr *iph = (const struct iphdr *) skb->data; 809 __be32 daddr = iph->daddr; 810 __be32 saddr = iph->saddr; 811 812 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n" 813 " Advised path = %pI4 -> %pI4\n", 814 &old_gw, dev->name, &new_gw, 815 &saddr, &daddr); 816 } 817 #endif 818 ; 819 } 820 821 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb) 822 { 823 struct rtable *rt; 824 struct flowi4 fl4; 825 const struct iphdr *iph = (const struct iphdr *) skb->data; 826 struct net *net = dev_net(skb->dev); 827 int oif = skb->dev->ifindex; 828 u8 tos = RT_TOS(iph->tos); 829 u8 prot = iph->protocol; 830 u32 mark = skb->mark; 831 832 rt = (struct rtable *) dst; 833 834 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0); 835 __ip_do_redirect(rt, skb, &fl4, true); 836 } 837 838 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst) 839 { 840 struct rtable *rt = (struct rtable *)dst; 841 struct dst_entry *ret = dst; 842 843 if (rt) { 844 if (dst->obsolete > 0) { 845 ip_rt_put(rt); 846 ret = NULL; 847 } else if ((rt->rt_flags & RTCF_REDIRECTED) || 848 rt->dst.expires) { 849 ip_rt_put(rt); 850 ret = NULL; 851 } 852 } 853 return ret; 854 } 855 856 /* 857 * Algorithm: 858 * 1. The first ip_rt_redirect_number redirects are sent 859 * with exponential backoff, then we stop sending them at all, 860 * assuming that the host ignores our redirects. 861 * 2. If we did not see packets requiring redirects 862 * during ip_rt_redirect_silence, we assume that the host 863 * forgot redirected route and start to send redirects again. 864 * 865 * This algorithm is much cheaper and more intelligent than dumb load limiting 866 * in icmp.c. 867 * 868 * NOTE. Do not forget to inhibit load limiting for redirects (redundant) 869 * and "frag. need" (breaks PMTU discovery) in icmp.c. 870 */ 871 872 void ip_rt_send_redirect(struct sk_buff *skb) 873 { 874 struct rtable *rt = skb_rtable(skb); 875 struct in_device *in_dev; 876 struct inet_peer *peer; 877 struct net *net; 878 int log_martians; 879 int vif; 880 881 rcu_read_lock(); 882 in_dev = __in_dev_get_rcu(rt->dst.dev); 883 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) { 884 rcu_read_unlock(); 885 return; 886 } 887 log_martians = IN_DEV_LOG_MARTIANS(in_dev); 888 vif = l3mdev_master_ifindex_rcu(rt->dst.dev); 889 rcu_read_unlock(); 890 891 net = dev_net(rt->dst.dev); 892 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1); 893 if (!peer) { 894 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, 895 rt_nexthop(rt, ip_hdr(skb)->daddr)); 896 return; 897 } 898 899 /* No redirected packets during ip_rt_redirect_silence; 900 * reset the algorithm. 901 */ 902 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) 903 peer->rate_tokens = 0; 904 905 /* Too many ignored redirects; do not send anything 906 * set dst.rate_last to the last seen redirected packet. 907 */ 908 if (peer->rate_tokens >= ip_rt_redirect_number) { 909 peer->rate_last = jiffies; 910 goto out_put_peer; 911 } 912 913 /* Check for load limit; set rate_last to the latest sent 914 * redirect. 915 */ 916 if (peer->rate_tokens == 0 || 917 time_after(jiffies, 918 (peer->rate_last + 919 (ip_rt_redirect_load << peer->rate_tokens)))) { 920 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr); 921 922 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw); 923 peer->rate_last = jiffies; 924 ++peer->rate_tokens; 925 #ifdef CONFIG_IP_ROUTE_VERBOSE 926 if (log_martians && 927 peer->rate_tokens == ip_rt_redirect_number) 928 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n", 929 &ip_hdr(skb)->saddr, inet_iif(skb), 930 &ip_hdr(skb)->daddr, &gw); 931 #endif 932 } 933 out_put_peer: 934 inet_putpeer(peer); 935 } 936 937 static int ip_error(struct sk_buff *skb) 938 { 939 struct rtable *rt = skb_rtable(skb); 940 struct net_device *dev = skb->dev; 941 struct in_device *in_dev; 942 struct inet_peer *peer; 943 unsigned long now; 944 struct net *net; 945 bool send; 946 int code; 947 948 if (netif_is_l3_master(skb->dev)) { 949 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif); 950 if (!dev) 951 goto out; 952 } 953 954 in_dev = __in_dev_get_rcu(dev); 955 956 /* IP on this device is disabled. */ 957 if (!in_dev) 958 goto out; 959 960 net = dev_net(rt->dst.dev); 961 if (!IN_DEV_FORWARD(in_dev)) { 962 switch (rt->dst.error) { 963 case EHOSTUNREACH: 964 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS); 965 break; 966 967 case ENETUNREACH: 968 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 969 break; 970 } 971 goto out; 972 } 973 974 switch (rt->dst.error) { 975 case EINVAL: 976 default: 977 goto out; 978 case EHOSTUNREACH: 979 code = ICMP_HOST_UNREACH; 980 break; 981 case ENETUNREACH: 982 code = ICMP_NET_UNREACH; 983 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 984 break; 985 case EACCES: 986 code = ICMP_PKT_FILTERED; 987 break; 988 } 989 990 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 991 l3mdev_master_ifindex(skb->dev), 1); 992 993 send = true; 994 if (peer) { 995 now = jiffies; 996 peer->rate_tokens += now - peer->rate_last; 997 if (peer->rate_tokens > ip_rt_error_burst) 998 peer->rate_tokens = ip_rt_error_burst; 999 peer->rate_last = now; 1000 if (peer->rate_tokens >= ip_rt_error_cost) 1001 peer->rate_tokens -= ip_rt_error_cost; 1002 else 1003 send = false; 1004 inet_putpeer(peer); 1005 } 1006 if (send) 1007 icmp_send(skb, ICMP_DEST_UNREACH, code, 0); 1008 1009 out: kfree_skb(skb); 1010 return 0; 1011 } 1012 1013 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu) 1014 { 1015 struct dst_entry *dst = &rt->dst; 1016 struct fib_result res; 1017 bool lock = false; 1018 1019 if (ip_mtu_locked(dst)) 1020 return; 1021 1022 if (ipv4_mtu(dst) < mtu) 1023 return; 1024 1025 if (mtu < ip_rt_min_pmtu) { 1026 lock = true; 1027 mtu = ip_rt_min_pmtu; 1028 } 1029 1030 if (rt->rt_pmtu == mtu && 1031 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2)) 1032 return; 1033 1034 rcu_read_lock(); 1035 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) { 1036 struct fib_nh *nh = &FIB_RES_NH(res); 1037 1038 update_or_create_fnhe(nh, fl4->daddr, 0, mtu, lock, 1039 jiffies + ip_rt_mtu_expires); 1040 } 1041 rcu_read_unlock(); 1042 } 1043 1044 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 1045 struct sk_buff *skb, u32 mtu) 1046 { 1047 struct rtable *rt = (struct rtable *) dst; 1048 struct flowi4 fl4; 1049 1050 ip_rt_build_flow_key(&fl4, sk, skb); 1051 __ip_rt_update_pmtu(rt, &fl4, mtu); 1052 } 1053 1054 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, 1055 int oif, u32 mark, u8 protocol, int flow_flags) 1056 { 1057 const struct iphdr *iph = (const struct iphdr *) skb->data; 1058 struct flowi4 fl4; 1059 struct rtable *rt; 1060 1061 if (!mark) 1062 mark = IP4_REPLY_MARK(net, skb->mark); 1063 1064 __build_flow_key(net, &fl4, NULL, iph, oif, 1065 RT_TOS(iph->tos), protocol, mark, flow_flags); 1066 rt = __ip_route_output_key(net, &fl4); 1067 if (!IS_ERR(rt)) { 1068 __ip_rt_update_pmtu(rt, &fl4, mtu); 1069 ip_rt_put(rt); 1070 } 1071 } 1072 EXPORT_SYMBOL_GPL(ipv4_update_pmtu); 1073 1074 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1075 { 1076 const struct iphdr *iph = (const struct iphdr *) skb->data; 1077 struct flowi4 fl4; 1078 struct rtable *rt; 1079 1080 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0); 1081 1082 if (!fl4.flowi4_mark) 1083 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark); 1084 1085 rt = __ip_route_output_key(sock_net(sk), &fl4); 1086 if (!IS_ERR(rt)) { 1087 __ip_rt_update_pmtu(rt, &fl4, mtu); 1088 ip_rt_put(rt); 1089 } 1090 } 1091 1092 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1093 { 1094 const struct iphdr *iph = (const struct iphdr *) skb->data; 1095 struct flowi4 fl4; 1096 struct rtable *rt; 1097 struct dst_entry *odst = NULL; 1098 bool new = false; 1099 struct net *net = sock_net(sk); 1100 1101 bh_lock_sock(sk); 1102 1103 if (!ip_sk_accept_pmtu(sk)) 1104 goto out; 1105 1106 odst = sk_dst_get(sk); 1107 1108 if (sock_owned_by_user(sk) || !odst) { 1109 __ipv4_sk_update_pmtu(skb, sk, mtu); 1110 goto out; 1111 } 1112 1113 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1114 1115 rt = (struct rtable *)odst; 1116 if (odst->obsolete && !odst->ops->check(odst, 0)) { 1117 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1118 if (IS_ERR(rt)) 1119 goto out; 1120 1121 new = true; 1122 } 1123 1124 __ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu); 1125 1126 if (!dst_check(&rt->dst, 0)) { 1127 if (new) 1128 dst_release(&rt->dst); 1129 1130 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1131 if (IS_ERR(rt)) 1132 goto out; 1133 1134 new = true; 1135 } 1136 1137 if (new) 1138 sk_dst_set(sk, &rt->dst); 1139 1140 out: 1141 bh_unlock_sock(sk); 1142 dst_release(odst); 1143 } 1144 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu); 1145 1146 void ipv4_redirect(struct sk_buff *skb, struct net *net, 1147 int oif, u32 mark, u8 protocol, int flow_flags) 1148 { 1149 const struct iphdr *iph = (const struct iphdr *) skb->data; 1150 struct flowi4 fl4; 1151 struct rtable *rt; 1152 1153 __build_flow_key(net, &fl4, NULL, iph, oif, 1154 RT_TOS(iph->tos), protocol, mark, flow_flags); 1155 rt = __ip_route_output_key(net, &fl4); 1156 if (!IS_ERR(rt)) { 1157 __ip_do_redirect(rt, skb, &fl4, false); 1158 ip_rt_put(rt); 1159 } 1160 } 1161 EXPORT_SYMBOL_GPL(ipv4_redirect); 1162 1163 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk) 1164 { 1165 const struct iphdr *iph = (const struct iphdr *) skb->data; 1166 struct flowi4 fl4; 1167 struct rtable *rt; 1168 struct net *net = sock_net(sk); 1169 1170 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1171 rt = __ip_route_output_key(net, &fl4); 1172 if (!IS_ERR(rt)) { 1173 __ip_do_redirect(rt, skb, &fl4, false); 1174 ip_rt_put(rt); 1175 } 1176 } 1177 EXPORT_SYMBOL_GPL(ipv4_sk_redirect); 1178 1179 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie) 1180 { 1181 struct rtable *rt = (struct rtable *) dst; 1182 1183 /* All IPV4 dsts are created with ->obsolete set to the value 1184 * DST_OBSOLETE_FORCE_CHK which forces validation calls down 1185 * into this function always. 1186 * 1187 * When a PMTU/redirect information update invalidates a route, 1188 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or 1189 * DST_OBSOLETE_DEAD by dst_free(). 1190 */ 1191 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt)) 1192 return NULL; 1193 return dst; 1194 } 1195 1196 static void ipv4_link_failure(struct sk_buff *skb) 1197 { 1198 struct rtable *rt; 1199 1200 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0); 1201 1202 rt = skb_rtable(skb); 1203 if (rt) 1204 dst_set_expires(&rt->dst, 0); 1205 } 1206 1207 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb) 1208 { 1209 pr_debug("%s: %pI4 -> %pI4, %s\n", 1210 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1211 skb->dev ? skb->dev->name : "?"); 1212 kfree_skb(skb); 1213 WARN_ON(1); 1214 return 0; 1215 } 1216 1217 /* 1218 We do not cache source address of outgoing interface, 1219 because it is used only by IP RR, TS and SRR options, 1220 so that it out of fast path. 1221 1222 BTW remember: "addr" is allowed to be not aligned 1223 in IP options! 1224 */ 1225 1226 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt) 1227 { 1228 __be32 src; 1229 1230 if (rt_is_output_route(rt)) 1231 src = ip_hdr(skb)->saddr; 1232 else { 1233 struct fib_result res; 1234 struct flowi4 fl4; 1235 struct iphdr *iph; 1236 1237 iph = ip_hdr(skb); 1238 1239 memset(&fl4, 0, sizeof(fl4)); 1240 fl4.daddr = iph->daddr; 1241 fl4.saddr = iph->saddr; 1242 fl4.flowi4_tos = RT_TOS(iph->tos); 1243 fl4.flowi4_oif = rt->dst.dev->ifindex; 1244 fl4.flowi4_iif = skb->dev->ifindex; 1245 fl4.flowi4_mark = skb->mark; 1246 1247 rcu_read_lock(); 1248 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0) 1249 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res); 1250 else 1251 src = inet_select_addr(rt->dst.dev, 1252 rt_nexthop(rt, iph->daddr), 1253 RT_SCOPE_UNIVERSE); 1254 rcu_read_unlock(); 1255 } 1256 memcpy(addr, &src, 4); 1257 } 1258 1259 #ifdef CONFIG_IP_ROUTE_CLASSID 1260 static void set_class_tag(struct rtable *rt, u32 tag) 1261 { 1262 if (!(rt->dst.tclassid & 0xFFFF)) 1263 rt->dst.tclassid |= tag & 0xFFFF; 1264 if (!(rt->dst.tclassid & 0xFFFF0000)) 1265 rt->dst.tclassid |= tag & 0xFFFF0000; 1266 } 1267 #endif 1268 1269 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1270 { 1271 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr); 1272 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size, 1273 ip_rt_min_advmss); 1274 1275 return min(advmss, IPV4_MAX_PMTU - header_size); 1276 } 1277 1278 static unsigned int ipv4_mtu(const struct dst_entry *dst) 1279 { 1280 const struct rtable *rt = (const struct rtable *) dst; 1281 unsigned int mtu = rt->rt_pmtu; 1282 1283 if (!mtu || time_after_eq(jiffies, rt->dst.expires)) 1284 mtu = dst_metric_raw(dst, RTAX_MTU); 1285 1286 if (mtu) 1287 return mtu; 1288 1289 mtu = READ_ONCE(dst->dev->mtu); 1290 1291 if (unlikely(ip_mtu_locked(dst))) { 1292 if (rt->rt_uses_gateway && mtu > 576) 1293 mtu = 576; 1294 } 1295 1296 mtu = min_t(unsigned int, mtu, IP_MAX_MTU); 1297 1298 return mtu - lwtunnel_headroom(dst->lwtstate, mtu); 1299 } 1300 1301 static struct fib_nh_exception *find_exception(struct fib_nh *nh, __be32 daddr) 1302 { 1303 struct fnhe_hash_bucket *hash = rcu_dereference(nh->nh_exceptions); 1304 struct fib_nh_exception *fnhe; 1305 u32 hval; 1306 1307 if (!hash) 1308 return NULL; 1309 1310 hval = fnhe_hashfun(daddr); 1311 1312 for (fnhe = rcu_dereference(hash[hval].chain); fnhe; 1313 fnhe = rcu_dereference(fnhe->fnhe_next)) { 1314 if (fnhe->fnhe_daddr == daddr) 1315 return fnhe; 1316 } 1317 return NULL; 1318 } 1319 1320 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, 1321 __be32 daddr, const bool do_cache) 1322 { 1323 bool ret = false; 1324 1325 spin_lock_bh(&fnhe_lock); 1326 1327 if (daddr == fnhe->fnhe_daddr) { 1328 struct rtable __rcu **porig; 1329 struct rtable *orig; 1330 int genid = fnhe_genid(dev_net(rt->dst.dev)); 1331 1332 if (rt_is_input_route(rt)) 1333 porig = &fnhe->fnhe_rth_input; 1334 else 1335 porig = &fnhe->fnhe_rth_output; 1336 orig = rcu_dereference(*porig); 1337 1338 if (fnhe->fnhe_genid != genid) { 1339 fnhe->fnhe_genid = genid; 1340 fnhe->fnhe_gw = 0; 1341 fnhe->fnhe_pmtu = 0; 1342 fnhe->fnhe_expires = 0; 1343 fnhe_flush_routes(fnhe); 1344 orig = NULL; 1345 } 1346 fill_route_from_fnhe(rt, fnhe); 1347 if (!rt->rt_gateway) 1348 rt->rt_gateway = daddr; 1349 1350 if (do_cache) { 1351 dst_hold(&rt->dst); 1352 rcu_assign_pointer(*porig, rt); 1353 if (orig) { 1354 dst_dev_put(&orig->dst); 1355 dst_release(&orig->dst); 1356 } 1357 ret = true; 1358 } 1359 1360 fnhe->fnhe_stamp = jiffies; 1361 } 1362 spin_unlock_bh(&fnhe_lock); 1363 1364 return ret; 1365 } 1366 1367 static bool rt_cache_route(struct fib_nh *nh, struct rtable *rt) 1368 { 1369 struct rtable *orig, *prev, **p; 1370 bool ret = true; 1371 1372 if (rt_is_input_route(rt)) { 1373 p = (struct rtable **)&nh->nh_rth_input; 1374 } else { 1375 p = (struct rtable **)raw_cpu_ptr(nh->nh_pcpu_rth_output); 1376 } 1377 orig = *p; 1378 1379 /* hold dst before doing cmpxchg() to avoid race condition 1380 * on this dst 1381 */ 1382 dst_hold(&rt->dst); 1383 prev = cmpxchg(p, orig, rt); 1384 if (prev == orig) { 1385 if (orig) { 1386 dst_dev_put(&orig->dst); 1387 dst_release(&orig->dst); 1388 } 1389 } else { 1390 dst_release(&rt->dst); 1391 ret = false; 1392 } 1393 1394 return ret; 1395 } 1396 1397 struct uncached_list { 1398 spinlock_t lock; 1399 struct list_head head; 1400 }; 1401 1402 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list); 1403 1404 void rt_add_uncached_list(struct rtable *rt) 1405 { 1406 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list); 1407 1408 rt->rt_uncached_list = ul; 1409 1410 spin_lock_bh(&ul->lock); 1411 list_add_tail(&rt->rt_uncached, &ul->head); 1412 spin_unlock_bh(&ul->lock); 1413 } 1414 1415 void rt_del_uncached_list(struct rtable *rt) 1416 { 1417 if (!list_empty(&rt->rt_uncached)) { 1418 struct uncached_list *ul = rt->rt_uncached_list; 1419 1420 spin_lock_bh(&ul->lock); 1421 list_del(&rt->rt_uncached); 1422 spin_unlock_bh(&ul->lock); 1423 } 1424 } 1425 1426 static void ipv4_dst_destroy(struct dst_entry *dst) 1427 { 1428 struct dst_metrics *p = (struct dst_metrics *)DST_METRICS_PTR(dst); 1429 struct rtable *rt = (struct rtable *)dst; 1430 1431 if (p != &dst_default_metrics && refcount_dec_and_test(&p->refcnt)) 1432 kfree(p); 1433 1434 rt_del_uncached_list(rt); 1435 } 1436 1437 void rt_flush_dev(struct net_device *dev) 1438 { 1439 struct net *net = dev_net(dev); 1440 struct rtable *rt; 1441 int cpu; 1442 1443 for_each_possible_cpu(cpu) { 1444 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 1445 1446 spin_lock_bh(&ul->lock); 1447 list_for_each_entry(rt, &ul->head, rt_uncached) { 1448 if (rt->dst.dev != dev) 1449 continue; 1450 rt->dst.dev = net->loopback_dev; 1451 dev_hold(rt->dst.dev); 1452 dev_put(dev); 1453 } 1454 spin_unlock_bh(&ul->lock); 1455 } 1456 } 1457 1458 static bool rt_cache_valid(const struct rtable *rt) 1459 { 1460 return rt && 1461 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK && 1462 !rt_is_expired(rt); 1463 } 1464 1465 static void rt_set_nexthop(struct rtable *rt, __be32 daddr, 1466 const struct fib_result *res, 1467 struct fib_nh_exception *fnhe, 1468 struct fib_info *fi, u16 type, u32 itag, 1469 const bool do_cache) 1470 { 1471 bool cached = false; 1472 1473 if (fi) { 1474 struct fib_nh *nh = &FIB_RES_NH(*res); 1475 1476 if (nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK) { 1477 rt->rt_gateway = nh->nh_gw; 1478 rt->rt_uses_gateway = 1; 1479 } 1480 dst_init_metrics(&rt->dst, fi->fib_metrics->metrics, true); 1481 if (fi->fib_metrics != &dst_default_metrics) { 1482 rt->dst._metrics |= DST_METRICS_REFCOUNTED; 1483 refcount_inc(&fi->fib_metrics->refcnt); 1484 } 1485 #ifdef CONFIG_IP_ROUTE_CLASSID 1486 rt->dst.tclassid = nh->nh_tclassid; 1487 #endif 1488 rt->dst.lwtstate = lwtstate_get(nh->nh_lwtstate); 1489 if (unlikely(fnhe)) 1490 cached = rt_bind_exception(rt, fnhe, daddr, do_cache); 1491 else if (do_cache) 1492 cached = rt_cache_route(nh, rt); 1493 if (unlikely(!cached)) { 1494 /* Routes we intend to cache in nexthop exception or 1495 * FIB nexthop have the DST_NOCACHE bit clear. 1496 * However, if we are unsuccessful at storing this 1497 * route into the cache we really need to set it. 1498 */ 1499 if (!rt->rt_gateway) 1500 rt->rt_gateway = daddr; 1501 rt_add_uncached_list(rt); 1502 } 1503 } else 1504 rt_add_uncached_list(rt); 1505 1506 #ifdef CONFIG_IP_ROUTE_CLASSID 1507 #ifdef CONFIG_IP_MULTIPLE_TABLES 1508 set_class_tag(rt, res->tclassid); 1509 #endif 1510 set_class_tag(rt, itag); 1511 #endif 1512 } 1513 1514 struct rtable *rt_dst_alloc(struct net_device *dev, 1515 unsigned int flags, u16 type, 1516 bool nopolicy, bool noxfrm, bool will_cache) 1517 { 1518 struct rtable *rt; 1519 1520 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1521 (will_cache ? 0 : DST_HOST) | 1522 (nopolicy ? DST_NOPOLICY : 0) | 1523 (noxfrm ? DST_NOXFRM : 0)); 1524 1525 if (rt) { 1526 rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1527 rt->rt_flags = flags; 1528 rt->rt_type = type; 1529 rt->rt_is_input = 0; 1530 rt->rt_iif = 0; 1531 rt->rt_pmtu = 0; 1532 rt->rt_mtu_locked = 0; 1533 rt->rt_gateway = 0; 1534 rt->rt_uses_gateway = 0; 1535 INIT_LIST_HEAD(&rt->rt_uncached); 1536 1537 rt->dst.output = ip_output; 1538 if (flags & RTCF_LOCAL) 1539 rt->dst.input = ip_local_deliver; 1540 } 1541 1542 return rt; 1543 } 1544 EXPORT_SYMBOL(rt_dst_alloc); 1545 1546 /* called in rcu_read_lock() section */ 1547 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1548 u8 tos, struct net_device *dev, 1549 struct in_device *in_dev, u32 *itag) 1550 { 1551 int err; 1552 1553 /* Primary sanity checks. */ 1554 if (!in_dev) 1555 return -EINVAL; 1556 1557 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) || 1558 skb->protocol != htons(ETH_P_IP)) 1559 return -EINVAL; 1560 1561 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev)) 1562 return -EINVAL; 1563 1564 if (ipv4_is_zeronet(saddr)) { 1565 if (!ipv4_is_local_multicast(daddr)) 1566 return -EINVAL; 1567 } else { 1568 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 1569 in_dev, itag); 1570 if (err < 0) 1571 return err; 1572 } 1573 return 0; 1574 } 1575 1576 /* called in rcu_read_lock() section */ 1577 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1578 u8 tos, struct net_device *dev, int our) 1579 { 1580 struct in_device *in_dev = __in_dev_get_rcu(dev); 1581 unsigned int flags = RTCF_MULTICAST; 1582 struct rtable *rth; 1583 u32 itag = 0; 1584 int err; 1585 1586 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag); 1587 if (err) 1588 return err; 1589 1590 if (our) 1591 flags |= RTCF_LOCAL; 1592 1593 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST, 1594 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false); 1595 if (!rth) 1596 return -ENOBUFS; 1597 1598 #ifdef CONFIG_IP_ROUTE_CLASSID 1599 rth->dst.tclassid = itag; 1600 #endif 1601 rth->dst.output = ip_rt_bug; 1602 rth->rt_is_input= 1; 1603 1604 #ifdef CONFIG_IP_MROUTE 1605 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1606 rth->dst.input = ip_mr_input; 1607 #endif 1608 RT_CACHE_STAT_INC(in_slow_mc); 1609 1610 skb_dst_set(skb, &rth->dst); 1611 return 0; 1612 } 1613 1614 1615 static void ip_handle_martian_source(struct net_device *dev, 1616 struct in_device *in_dev, 1617 struct sk_buff *skb, 1618 __be32 daddr, 1619 __be32 saddr) 1620 { 1621 RT_CACHE_STAT_INC(in_martian_src); 1622 #ifdef CONFIG_IP_ROUTE_VERBOSE 1623 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1624 /* 1625 * RFC1812 recommendation, if source is martian, 1626 * the only hint is MAC header. 1627 */ 1628 pr_warn("martian source %pI4 from %pI4, on dev %s\n", 1629 &daddr, &saddr, dev->name); 1630 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1631 print_hex_dump(KERN_WARNING, "ll header: ", 1632 DUMP_PREFIX_OFFSET, 16, 1, 1633 skb_mac_header(skb), 1634 dev->hard_header_len, true); 1635 } 1636 } 1637 #endif 1638 } 1639 1640 static void ip_del_fnhe(struct fib_nh *nh, __be32 daddr) 1641 { 1642 struct fnhe_hash_bucket *hash; 1643 struct fib_nh_exception *fnhe, __rcu **fnhe_p; 1644 u32 hval = fnhe_hashfun(daddr); 1645 1646 spin_lock_bh(&fnhe_lock); 1647 1648 hash = rcu_dereference_protected(nh->nh_exceptions, 1649 lockdep_is_held(&fnhe_lock)); 1650 hash += hval; 1651 1652 fnhe_p = &hash->chain; 1653 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock)); 1654 while (fnhe) { 1655 if (fnhe->fnhe_daddr == daddr) { 1656 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected( 1657 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock))); 1658 fnhe_flush_routes(fnhe); 1659 kfree_rcu(fnhe, rcu); 1660 break; 1661 } 1662 fnhe_p = &fnhe->fnhe_next; 1663 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1664 lockdep_is_held(&fnhe_lock)); 1665 } 1666 1667 spin_unlock_bh(&fnhe_lock); 1668 } 1669 1670 /* called in rcu_read_lock() section */ 1671 static int __mkroute_input(struct sk_buff *skb, 1672 const struct fib_result *res, 1673 struct in_device *in_dev, 1674 __be32 daddr, __be32 saddr, u32 tos) 1675 { 1676 struct fib_nh_exception *fnhe; 1677 struct rtable *rth; 1678 int err; 1679 struct in_device *out_dev; 1680 bool do_cache; 1681 u32 itag = 0; 1682 1683 /* get a working reference to the output device */ 1684 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res)); 1685 if (!out_dev) { 1686 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n"); 1687 return -EINVAL; 1688 } 1689 1690 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res), 1691 in_dev->dev, in_dev, &itag); 1692 if (err < 0) { 1693 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1694 saddr); 1695 1696 goto cleanup; 1697 } 1698 1699 do_cache = res->fi && !itag; 1700 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) && 1701 skb->protocol == htons(ETH_P_IP) && 1702 (IN_DEV_SHARED_MEDIA(out_dev) || 1703 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) 1704 IPCB(skb)->flags |= IPSKB_DOREDIRECT; 1705 1706 if (skb->protocol != htons(ETH_P_IP)) { 1707 /* Not IP (i.e. ARP). Do not create route, if it is 1708 * invalid for proxy arp. DNAT routes are always valid. 1709 * 1710 * Proxy arp feature have been extended to allow, ARP 1711 * replies back to the same interface, to support 1712 * Private VLAN switch technologies. See arp.c. 1713 */ 1714 if (out_dev == in_dev && 1715 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 1716 err = -EINVAL; 1717 goto cleanup; 1718 } 1719 } 1720 1721 fnhe = find_exception(&FIB_RES_NH(*res), daddr); 1722 if (do_cache) { 1723 if (fnhe) { 1724 rth = rcu_dereference(fnhe->fnhe_rth_input); 1725 if (rth && rth->dst.expires && 1726 time_after(jiffies, rth->dst.expires)) { 1727 ip_del_fnhe(&FIB_RES_NH(*res), daddr); 1728 fnhe = NULL; 1729 } else { 1730 goto rt_cache; 1731 } 1732 } 1733 1734 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input); 1735 1736 rt_cache: 1737 if (rt_cache_valid(rth)) { 1738 skb_dst_set_noref(skb, &rth->dst); 1739 goto out; 1740 } 1741 } 1742 1743 rth = rt_dst_alloc(out_dev->dev, 0, res->type, 1744 IN_DEV_CONF_GET(in_dev, NOPOLICY), 1745 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache); 1746 if (!rth) { 1747 err = -ENOBUFS; 1748 goto cleanup; 1749 } 1750 1751 rth->rt_is_input = 1; 1752 RT_CACHE_STAT_INC(in_slow_tot); 1753 1754 rth->dst.input = ip_forward; 1755 1756 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag, 1757 do_cache); 1758 lwtunnel_set_redirect(&rth->dst); 1759 skb_dst_set(skb, &rth->dst); 1760 out: 1761 err = 0; 1762 cleanup: 1763 return err; 1764 } 1765 1766 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1767 /* To make ICMP packets follow the right flow, the multipath hash is 1768 * calculated from the inner IP addresses. 1769 */ 1770 static void ip_multipath_l3_keys(const struct sk_buff *skb, 1771 struct flow_keys *hash_keys) 1772 { 1773 const struct iphdr *outer_iph = ip_hdr(skb); 1774 const struct iphdr *key_iph = outer_iph; 1775 const struct iphdr *inner_iph; 1776 const struct icmphdr *icmph; 1777 struct iphdr _inner_iph; 1778 struct icmphdr _icmph; 1779 1780 if (likely(outer_iph->protocol != IPPROTO_ICMP)) 1781 goto out; 1782 1783 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0)) 1784 goto out; 1785 1786 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph), 1787 &_icmph); 1788 if (!icmph) 1789 goto out; 1790 1791 if (icmph->type != ICMP_DEST_UNREACH && 1792 icmph->type != ICMP_REDIRECT && 1793 icmph->type != ICMP_TIME_EXCEEDED && 1794 icmph->type != ICMP_PARAMETERPROB) 1795 goto out; 1796 1797 inner_iph = skb_header_pointer(skb, 1798 outer_iph->ihl * 4 + sizeof(_icmph), 1799 sizeof(_inner_iph), &_inner_iph); 1800 if (!inner_iph) 1801 goto out; 1802 1803 key_iph = inner_iph; 1804 out: 1805 hash_keys->addrs.v4addrs.src = key_iph->saddr; 1806 hash_keys->addrs.v4addrs.dst = key_iph->daddr; 1807 } 1808 1809 /* if skb is set it will be used and fl4 can be NULL */ 1810 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4, 1811 const struct sk_buff *skb, struct flow_keys *flkeys) 1812 { 1813 struct flow_keys hash_keys; 1814 u32 mhash; 1815 1816 switch (net->ipv4.sysctl_fib_multipath_hash_policy) { 1817 case 0: 1818 memset(&hash_keys, 0, sizeof(hash_keys)); 1819 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1820 if (skb) { 1821 ip_multipath_l3_keys(skb, &hash_keys); 1822 } else { 1823 hash_keys.addrs.v4addrs.src = fl4->saddr; 1824 hash_keys.addrs.v4addrs.dst = fl4->daddr; 1825 } 1826 break; 1827 case 1: 1828 /* skb is currently provided only when forwarding */ 1829 if (skb) { 1830 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 1831 struct flow_keys keys; 1832 1833 /* short-circuit if we already have L4 hash present */ 1834 if (skb->l4_hash) 1835 return skb_get_hash_raw(skb) >> 1; 1836 1837 memset(&hash_keys, 0, sizeof(hash_keys)); 1838 1839 if (!flkeys) { 1840 skb_flow_dissect_flow_keys(skb, &keys, flag); 1841 flkeys = &keys; 1842 } 1843 1844 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1845 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src; 1846 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst; 1847 hash_keys.ports.src = flkeys->ports.src; 1848 hash_keys.ports.dst = flkeys->ports.dst; 1849 hash_keys.basic.ip_proto = flkeys->basic.ip_proto; 1850 } else { 1851 memset(&hash_keys, 0, sizeof(hash_keys)); 1852 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1853 hash_keys.addrs.v4addrs.src = fl4->saddr; 1854 hash_keys.addrs.v4addrs.dst = fl4->daddr; 1855 hash_keys.ports.src = fl4->fl4_sport; 1856 hash_keys.ports.dst = fl4->fl4_dport; 1857 hash_keys.basic.ip_proto = fl4->flowi4_proto; 1858 } 1859 break; 1860 } 1861 mhash = flow_hash_from_keys(&hash_keys); 1862 1863 return mhash >> 1; 1864 } 1865 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 1866 1867 static int ip_mkroute_input(struct sk_buff *skb, 1868 struct fib_result *res, 1869 struct in_device *in_dev, 1870 __be32 daddr, __be32 saddr, u32 tos, 1871 struct flow_keys *hkeys) 1872 { 1873 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1874 if (res->fi && res->fi->fib_nhs > 1) { 1875 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys); 1876 1877 fib_select_multipath(res, h); 1878 } 1879 #endif 1880 1881 /* create a routing cache entry */ 1882 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos); 1883 } 1884 1885 /* 1886 * NOTE. We drop all the packets that has local source 1887 * addresses, because every properly looped back packet 1888 * must have correct destination already attached by output routine. 1889 * 1890 * Such approach solves two big problems: 1891 * 1. Not simplex devices are handled properly. 1892 * 2. IP spoofing attempts are filtered with 100% of guarantee. 1893 * called with rcu_read_lock() 1894 */ 1895 1896 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1897 u8 tos, struct net_device *dev, 1898 struct fib_result *res) 1899 { 1900 struct in_device *in_dev = __in_dev_get_rcu(dev); 1901 struct flow_keys *flkeys = NULL, _flkeys; 1902 struct net *net = dev_net(dev); 1903 struct ip_tunnel_info *tun_info; 1904 int err = -EINVAL; 1905 unsigned int flags = 0; 1906 u32 itag = 0; 1907 struct rtable *rth; 1908 struct flowi4 fl4; 1909 bool do_cache; 1910 1911 /* IP on this device is disabled. */ 1912 1913 if (!in_dev) 1914 goto out; 1915 1916 /* Check for the most weird martians, which can be not detected 1917 by fib_lookup. 1918 */ 1919 1920 tun_info = skb_tunnel_info(skb); 1921 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX)) 1922 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id; 1923 else 1924 fl4.flowi4_tun_key.tun_id = 0; 1925 skb_dst_drop(skb); 1926 1927 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 1928 goto martian_source; 1929 1930 res->fi = NULL; 1931 res->table = NULL; 1932 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 1933 goto brd_input; 1934 1935 /* Accept zero addresses only to limited broadcast; 1936 * I even do not know to fix it or not. Waiting for complains :-) 1937 */ 1938 if (ipv4_is_zeronet(saddr)) 1939 goto martian_source; 1940 1941 if (ipv4_is_zeronet(daddr)) 1942 goto martian_destination; 1943 1944 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(), 1945 * and call it once if daddr or/and saddr are loopback addresses 1946 */ 1947 if (ipv4_is_loopback(daddr)) { 1948 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 1949 goto martian_destination; 1950 } else if (ipv4_is_loopback(saddr)) { 1951 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 1952 goto martian_source; 1953 } 1954 1955 /* 1956 * Now we are ready to route packet. 1957 */ 1958 fl4.flowi4_oif = 0; 1959 fl4.flowi4_iif = dev->ifindex; 1960 fl4.flowi4_mark = skb->mark; 1961 fl4.flowi4_tos = tos; 1962 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 1963 fl4.flowi4_flags = 0; 1964 fl4.daddr = daddr; 1965 fl4.saddr = saddr; 1966 fl4.flowi4_uid = sock_net_uid(net, NULL); 1967 1968 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) 1969 flkeys = &_flkeys; 1970 1971 err = fib_lookup(net, &fl4, res, 0); 1972 if (err != 0) { 1973 if (!IN_DEV_FORWARD(in_dev)) 1974 err = -EHOSTUNREACH; 1975 goto no_route; 1976 } 1977 1978 if (res->type == RTN_BROADCAST) 1979 goto brd_input; 1980 1981 if (res->type == RTN_LOCAL) { 1982 err = fib_validate_source(skb, saddr, daddr, tos, 1983 0, dev, in_dev, &itag); 1984 if (err < 0) 1985 goto martian_source; 1986 goto local_input; 1987 } 1988 1989 if (!IN_DEV_FORWARD(in_dev)) { 1990 err = -EHOSTUNREACH; 1991 goto no_route; 1992 } 1993 if (res->type != RTN_UNICAST) 1994 goto martian_destination; 1995 1996 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys); 1997 out: return err; 1998 1999 brd_input: 2000 if (skb->protocol != htons(ETH_P_IP)) 2001 goto e_inval; 2002 2003 if (!ipv4_is_zeronet(saddr)) { 2004 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 2005 in_dev, &itag); 2006 if (err < 0) 2007 goto martian_source; 2008 } 2009 flags |= RTCF_BROADCAST; 2010 res->type = RTN_BROADCAST; 2011 RT_CACHE_STAT_INC(in_brd); 2012 2013 local_input: 2014 do_cache = false; 2015 if (res->fi) { 2016 if (!itag) { 2017 rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input); 2018 if (rt_cache_valid(rth)) { 2019 skb_dst_set_noref(skb, &rth->dst); 2020 err = 0; 2021 goto out; 2022 } 2023 do_cache = true; 2024 } 2025 } 2026 2027 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev, 2028 flags | RTCF_LOCAL, res->type, 2029 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache); 2030 if (!rth) 2031 goto e_nobufs; 2032 2033 rth->dst.output= ip_rt_bug; 2034 #ifdef CONFIG_IP_ROUTE_CLASSID 2035 rth->dst.tclassid = itag; 2036 #endif 2037 rth->rt_is_input = 1; 2038 2039 RT_CACHE_STAT_INC(in_slow_tot); 2040 if (res->type == RTN_UNREACHABLE) { 2041 rth->dst.input= ip_error; 2042 rth->dst.error= -err; 2043 rth->rt_flags &= ~RTCF_LOCAL; 2044 } 2045 2046 if (do_cache) { 2047 struct fib_nh *nh = &FIB_RES_NH(*res); 2048 2049 rth->dst.lwtstate = lwtstate_get(nh->nh_lwtstate); 2050 if (lwtunnel_input_redirect(rth->dst.lwtstate)) { 2051 WARN_ON(rth->dst.input == lwtunnel_input); 2052 rth->dst.lwtstate->orig_input = rth->dst.input; 2053 rth->dst.input = lwtunnel_input; 2054 } 2055 2056 if (unlikely(!rt_cache_route(nh, rth))) 2057 rt_add_uncached_list(rth); 2058 } 2059 skb_dst_set(skb, &rth->dst); 2060 err = 0; 2061 goto out; 2062 2063 no_route: 2064 RT_CACHE_STAT_INC(in_no_route); 2065 res->type = RTN_UNREACHABLE; 2066 res->fi = NULL; 2067 res->table = NULL; 2068 goto local_input; 2069 2070 /* 2071 * Do not cache martian addresses: they should be logged (RFC1812) 2072 */ 2073 martian_destination: 2074 RT_CACHE_STAT_INC(in_martian_dst); 2075 #ifdef CONFIG_IP_ROUTE_VERBOSE 2076 if (IN_DEV_LOG_MARTIANS(in_dev)) 2077 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n", 2078 &daddr, &saddr, dev->name); 2079 #endif 2080 2081 e_inval: 2082 err = -EINVAL; 2083 goto out; 2084 2085 e_nobufs: 2086 err = -ENOBUFS; 2087 goto out; 2088 2089 martian_source: 2090 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2091 goto out; 2092 } 2093 2094 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2095 u8 tos, struct net_device *dev) 2096 { 2097 struct fib_result res; 2098 int err; 2099 2100 tos &= IPTOS_RT_MASK; 2101 rcu_read_lock(); 2102 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res); 2103 rcu_read_unlock(); 2104 2105 return err; 2106 } 2107 EXPORT_SYMBOL(ip_route_input_noref); 2108 2109 /* called with rcu_read_lock held */ 2110 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2111 u8 tos, struct net_device *dev, struct fib_result *res) 2112 { 2113 /* Multicast recognition logic is moved from route cache to here. 2114 The problem was that too many Ethernet cards have broken/missing 2115 hardware multicast filters :-( As result the host on multicasting 2116 network acquires a lot of useless route cache entries, sort of 2117 SDR messages from all the world. Now we try to get rid of them. 2118 Really, provided software IP multicast filter is organized 2119 reasonably (at least, hashed), it does not result in a slowdown 2120 comparing with route cache reject entries. 2121 Note, that multicast routers are not affected, because 2122 route cache entry is created eventually. 2123 */ 2124 if (ipv4_is_multicast(daddr)) { 2125 struct in_device *in_dev = __in_dev_get_rcu(dev); 2126 int our = 0; 2127 int err = -EINVAL; 2128 2129 if (in_dev) 2130 our = ip_check_mc_rcu(in_dev, daddr, saddr, 2131 ip_hdr(skb)->protocol); 2132 2133 /* check l3 master if no match yet */ 2134 if ((!in_dev || !our) && netif_is_l3_slave(dev)) { 2135 struct in_device *l3_in_dev; 2136 2137 l3_in_dev = __in_dev_get_rcu(skb->dev); 2138 if (l3_in_dev) 2139 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr, 2140 ip_hdr(skb)->protocol); 2141 } 2142 2143 if (our 2144 #ifdef CONFIG_IP_MROUTE 2145 || 2146 (!ipv4_is_local_multicast(daddr) && 2147 IN_DEV_MFORWARD(in_dev)) 2148 #endif 2149 ) { 2150 err = ip_route_input_mc(skb, daddr, saddr, 2151 tos, dev, our); 2152 } 2153 return err; 2154 } 2155 2156 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res); 2157 } 2158 2159 /* called with rcu_read_lock() */ 2160 static struct rtable *__mkroute_output(const struct fib_result *res, 2161 const struct flowi4 *fl4, int orig_oif, 2162 struct net_device *dev_out, 2163 unsigned int flags) 2164 { 2165 struct fib_info *fi = res->fi; 2166 struct fib_nh_exception *fnhe; 2167 struct in_device *in_dev; 2168 u16 type = res->type; 2169 struct rtable *rth; 2170 bool do_cache; 2171 2172 in_dev = __in_dev_get_rcu(dev_out); 2173 if (!in_dev) 2174 return ERR_PTR(-EINVAL); 2175 2176 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 2177 if (ipv4_is_loopback(fl4->saddr) && 2178 !(dev_out->flags & IFF_LOOPBACK) && 2179 !netif_is_l3_master(dev_out)) 2180 return ERR_PTR(-EINVAL); 2181 2182 if (ipv4_is_lbcast(fl4->daddr)) 2183 type = RTN_BROADCAST; 2184 else if (ipv4_is_multicast(fl4->daddr)) 2185 type = RTN_MULTICAST; 2186 else if (ipv4_is_zeronet(fl4->daddr)) 2187 return ERR_PTR(-EINVAL); 2188 2189 if (dev_out->flags & IFF_LOOPBACK) 2190 flags |= RTCF_LOCAL; 2191 2192 do_cache = true; 2193 if (type == RTN_BROADCAST) { 2194 flags |= RTCF_BROADCAST | RTCF_LOCAL; 2195 fi = NULL; 2196 } else if (type == RTN_MULTICAST) { 2197 flags |= RTCF_MULTICAST | RTCF_LOCAL; 2198 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr, 2199 fl4->flowi4_proto)) 2200 flags &= ~RTCF_LOCAL; 2201 else 2202 do_cache = false; 2203 /* If multicast route do not exist use 2204 * default one, but do not gateway in this case. 2205 * Yes, it is hack. 2206 */ 2207 if (fi && res->prefixlen < 4) 2208 fi = NULL; 2209 } else if ((type == RTN_LOCAL) && (orig_oif != 0) && 2210 (orig_oif != dev_out->ifindex)) { 2211 /* For local routes that require a particular output interface 2212 * we do not want to cache the result. Caching the result 2213 * causes incorrect behaviour when there are multiple source 2214 * addresses on the interface, the end result being that if the 2215 * intended recipient is waiting on that interface for the 2216 * packet he won't receive it because it will be delivered on 2217 * the loopback interface and the IP_PKTINFO ipi_ifindex will 2218 * be set to the loopback interface as well. 2219 */ 2220 fi = NULL; 2221 } 2222 2223 fnhe = NULL; 2224 do_cache &= fi != NULL; 2225 if (do_cache) { 2226 struct rtable __rcu **prth; 2227 struct fib_nh *nh = &FIB_RES_NH(*res); 2228 2229 fnhe = find_exception(nh, fl4->daddr); 2230 if (fnhe) { 2231 prth = &fnhe->fnhe_rth_output; 2232 rth = rcu_dereference(*prth); 2233 if (rth && rth->dst.expires && 2234 time_after(jiffies, rth->dst.expires)) { 2235 ip_del_fnhe(nh, fl4->daddr); 2236 fnhe = NULL; 2237 } else { 2238 goto rt_cache; 2239 } 2240 } 2241 2242 if (unlikely(fl4->flowi4_flags & 2243 FLOWI_FLAG_KNOWN_NH && 2244 !(nh->nh_gw && 2245 nh->nh_scope == RT_SCOPE_LINK))) { 2246 do_cache = false; 2247 goto add; 2248 } 2249 prth = raw_cpu_ptr(nh->nh_pcpu_rth_output); 2250 rth = rcu_dereference(*prth); 2251 2252 rt_cache: 2253 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst)) 2254 return rth; 2255 } 2256 2257 add: 2258 rth = rt_dst_alloc(dev_out, flags, type, 2259 IN_DEV_CONF_GET(in_dev, NOPOLICY), 2260 IN_DEV_CONF_GET(in_dev, NOXFRM), 2261 do_cache); 2262 if (!rth) 2263 return ERR_PTR(-ENOBUFS); 2264 2265 rth->rt_iif = orig_oif; 2266 2267 RT_CACHE_STAT_INC(out_slow_tot); 2268 2269 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 2270 if (flags & RTCF_LOCAL && 2271 !(dev_out->flags & IFF_LOOPBACK)) { 2272 rth->dst.output = ip_mc_output; 2273 RT_CACHE_STAT_INC(out_slow_mc); 2274 } 2275 #ifdef CONFIG_IP_MROUTE 2276 if (type == RTN_MULTICAST) { 2277 if (IN_DEV_MFORWARD(in_dev) && 2278 !ipv4_is_local_multicast(fl4->daddr)) { 2279 rth->dst.input = ip_mr_input; 2280 rth->dst.output = ip_mc_output; 2281 } 2282 } 2283 #endif 2284 } 2285 2286 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache); 2287 lwtunnel_set_redirect(&rth->dst); 2288 2289 return rth; 2290 } 2291 2292 /* 2293 * Major route resolver routine. 2294 */ 2295 2296 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4, 2297 const struct sk_buff *skb) 2298 { 2299 __u8 tos = RT_FL_TOS(fl4); 2300 struct fib_result res; 2301 struct rtable *rth; 2302 2303 res.tclassid = 0; 2304 res.fi = NULL; 2305 res.table = NULL; 2306 2307 fl4->flowi4_iif = LOOPBACK_IFINDEX; 2308 fl4->flowi4_tos = tos & IPTOS_RT_MASK; 2309 fl4->flowi4_scope = ((tos & RTO_ONLINK) ? 2310 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE); 2311 2312 rcu_read_lock(); 2313 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb); 2314 rcu_read_unlock(); 2315 2316 return rth; 2317 } 2318 EXPORT_SYMBOL_GPL(ip_route_output_key_hash); 2319 2320 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4, 2321 struct fib_result *res, 2322 const struct sk_buff *skb) 2323 { 2324 struct net_device *dev_out = NULL; 2325 int orig_oif = fl4->flowi4_oif; 2326 unsigned int flags = 0; 2327 struct rtable *rth; 2328 int err = -ENETUNREACH; 2329 2330 if (fl4->saddr) { 2331 rth = ERR_PTR(-EINVAL); 2332 if (ipv4_is_multicast(fl4->saddr) || 2333 ipv4_is_lbcast(fl4->saddr) || 2334 ipv4_is_zeronet(fl4->saddr)) 2335 goto out; 2336 2337 /* I removed check for oif == dev_out->oif here. 2338 It was wrong for two reasons: 2339 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 2340 is assigned to multiple interfaces. 2341 2. Moreover, we are allowed to send packets with saddr 2342 of another iface. --ANK 2343 */ 2344 2345 if (fl4->flowi4_oif == 0 && 2346 (ipv4_is_multicast(fl4->daddr) || 2347 ipv4_is_lbcast(fl4->daddr))) { 2348 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2349 dev_out = __ip_dev_find(net, fl4->saddr, false); 2350 if (!dev_out) 2351 goto out; 2352 2353 /* Special hack: user can direct multicasts 2354 and limited broadcast via necessary interface 2355 without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 2356 This hack is not just for fun, it allows 2357 vic,vat and friends to work. 2358 They bind socket to loopback, set ttl to zero 2359 and expect that it will work. 2360 From the viewpoint of routing cache they are broken, 2361 because we are not allowed to build multicast path 2362 with loopback source addr (look, routing cache 2363 cannot know, that ttl is zero, so that packet 2364 will not leave this host and route is valid). 2365 Luckily, this hack is good workaround. 2366 */ 2367 2368 fl4->flowi4_oif = dev_out->ifindex; 2369 goto make_route; 2370 } 2371 2372 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 2373 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2374 if (!__ip_dev_find(net, fl4->saddr, false)) 2375 goto out; 2376 } 2377 } 2378 2379 2380 if (fl4->flowi4_oif) { 2381 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif); 2382 rth = ERR_PTR(-ENODEV); 2383 if (!dev_out) 2384 goto out; 2385 2386 /* RACE: Check return value of inet_select_addr instead. */ 2387 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 2388 rth = ERR_PTR(-ENETUNREACH); 2389 goto out; 2390 } 2391 if (ipv4_is_local_multicast(fl4->daddr) || 2392 ipv4_is_lbcast(fl4->daddr) || 2393 fl4->flowi4_proto == IPPROTO_IGMP) { 2394 if (!fl4->saddr) 2395 fl4->saddr = inet_select_addr(dev_out, 0, 2396 RT_SCOPE_LINK); 2397 goto make_route; 2398 } 2399 if (!fl4->saddr) { 2400 if (ipv4_is_multicast(fl4->daddr)) 2401 fl4->saddr = inet_select_addr(dev_out, 0, 2402 fl4->flowi4_scope); 2403 else if (!fl4->daddr) 2404 fl4->saddr = inet_select_addr(dev_out, 0, 2405 RT_SCOPE_HOST); 2406 } 2407 } 2408 2409 if (!fl4->daddr) { 2410 fl4->daddr = fl4->saddr; 2411 if (!fl4->daddr) 2412 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK); 2413 dev_out = net->loopback_dev; 2414 fl4->flowi4_oif = LOOPBACK_IFINDEX; 2415 res->type = RTN_LOCAL; 2416 flags |= RTCF_LOCAL; 2417 goto make_route; 2418 } 2419 2420 err = fib_lookup(net, fl4, res, 0); 2421 if (err) { 2422 res->fi = NULL; 2423 res->table = NULL; 2424 if (fl4->flowi4_oif && 2425 (ipv4_is_multicast(fl4->daddr) || 2426 !netif_index_is_l3_master(net, fl4->flowi4_oif))) { 2427 /* Apparently, routing tables are wrong. Assume, 2428 that the destination is on link. 2429 2430 WHY? DW. 2431 Because we are allowed to send to iface 2432 even if it has NO routes and NO assigned 2433 addresses. When oif is specified, routing 2434 tables are looked up with only one purpose: 2435 to catch if destination is gatewayed, rather than 2436 direct. Moreover, if MSG_DONTROUTE is set, 2437 we send packet, ignoring both routing tables 2438 and ifaddr state. --ANK 2439 2440 2441 We could make it even if oif is unknown, 2442 likely IPv6, but we do not. 2443 */ 2444 2445 if (fl4->saddr == 0) 2446 fl4->saddr = inet_select_addr(dev_out, 0, 2447 RT_SCOPE_LINK); 2448 res->type = RTN_UNICAST; 2449 goto make_route; 2450 } 2451 rth = ERR_PTR(err); 2452 goto out; 2453 } 2454 2455 if (res->type == RTN_LOCAL) { 2456 if (!fl4->saddr) { 2457 if (res->fi->fib_prefsrc) 2458 fl4->saddr = res->fi->fib_prefsrc; 2459 else 2460 fl4->saddr = fl4->daddr; 2461 } 2462 2463 /* L3 master device is the loopback for that domain */ 2464 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? : 2465 net->loopback_dev; 2466 2467 /* make sure orig_oif points to fib result device even 2468 * though packet rx/tx happens over loopback or l3mdev 2469 */ 2470 orig_oif = FIB_RES_OIF(*res); 2471 2472 fl4->flowi4_oif = dev_out->ifindex; 2473 flags |= RTCF_LOCAL; 2474 goto make_route; 2475 } 2476 2477 fib_select_path(net, res, fl4, skb); 2478 2479 dev_out = FIB_RES_DEV(*res); 2480 fl4->flowi4_oif = dev_out->ifindex; 2481 2482 2483 make_route: 2484 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags); 2485 2486 out: 2487 return rth; 2488 } 2489 2490 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie) 2491 { 2492 return NULL; 2493 } 2494 2495 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst) 2496 { 2497 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU); 2498 2499 return mtu ? : dst->dev->mtu; 2500 } 2501 2502 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk, 2503 struct sk_buff *skb, u32 mtu) 2504 { 2505 } 2506 2507 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk, 2508 struct sk_buff *skb) 2509 { 2510 } 2511 2512 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst, 2513 unsigned long old) 2514 { 2515 return NULL; 2516 } 2517 2518 static struct dst_ops ipv4_dst_blackhole_ops = { 2519 .family = AF_INET, 2520 .check = ipv4_blackhole_dst_check, 2521 .mtu = ipv4_blackhole_mtu, 2522 .default_advmss = ipv4_default_advmss, 2523 .update_pmtu = ipv4_rt_blackhole_update_pmtu, 2524 .redirect = ipv4_rt_blackhole_redirect, 2525 .cow_metrics = ipv4_rt_blackhole_cow_metrics, 2526 .neigh_lookup = ipv4_neigh_lookup, 2527 }; 2528 2529 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2530 { 2531 struct rtable *ort = (struct rtable *) dst_orig; 2532 struct rtable *rt; 2533 2534 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0); 2535 if (rt) { 2536 struct dst_entry *new = &rt->dst; 2537 2538 new->__use = 1; 2539 new->input = dst_discard; 2540 new->output = dst_discard_out; 2541 2542 new->dev = net->loopback_dev; 2543 if (new->dev) 2544 dev_hold(new->dev); 2545 2546 rt->rt_is_input = ort->rt_is_input; 2547 rt->rt_iif = ort->rt_iif; 2548 rt->rt_pmtu = ort->rt_pmtu; 2549 rt->rt_mtu_locked = ort->rt_mtu_locked; 2550 2551 rt->rt_genid = rt_genid_ipv4(net); 2552 rt->rt_flags = ort->rt_flags; 2553 rt->rt_type = ort->rt_type; 2554 rt->rt_gateway = ort->rt_gateway; 2555 rt->rt_uses_gateway = ort->rt_uses_gateway; 2556 2557 INIT_LIST_HEAD(&rt->rt_uncached); 2558 } 2559 2560 dst_release(dst_orig); 2561 2562 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2563 } 2564 2565 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2566 const struct sock *sk) 2567 { 2568 struct rtable *rt = __ip_route_output_key(net, flp4); 2569 2570 if (IS_ERR(rt)) 2571 return rt; 2572 2573 if (flp4->flowi4_proto) 2574 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst, 2575 flowi4_to_flowi(flp4), 2576 sk, 0); 2577 2578 return rt; 2579 } 2580 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2581 2582 /* called with rcu_read_lock held */ 2583 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, u32 table_id, 2584 struct flowi4 *fl4, struct sk_buff *skb, u32 portid, 2585 u32 seq) 2586 { 2587 struct rtable *rt = skb_rtable(skb); 2588 struct rtmsg *r; 2589 struct nlmsghdr *nlh; 2590 unsigned long expires = 0; 2591 u32 error; 2592 u32 metrics[RTAX_MAX]; 2593 2594 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0); 2595 if (!nlh) 2596 return -EMSGSIZE; 2597 2598 r = nlmsg_data(nlh); 2599 r->rtm_family = AF_INET; 2600 r->rtm_dst_len = 32; 2601 r->rtm_src_len = 0; 2602 r->rtm_tos = fl4->flowi4_tos; 2603 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT; 2604 if (nla_put_u32(skb, RTA_TABLE, table_id)) 2605 goto nla_put_failure; 2606 r->rtm_type = rt->rt_type; 2607 r->rtm_scope = RT_SCOPE_UNIVERSE; 2608 r->rtm_protocol = RTPROT_UNSPEC; 2609 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 2610 if (rt->rt_flags & RTCF_NOTIFY) 2611 r->rtm_flags |= RTM_F_NOTIFY; 2612 if (IPCB(skb)->flags & IPSKB_DOREDIRECT) 2613 r->rtm_flags |= RTCF_DOREDIRECT; 2614 2615 if (nla_put_in_addr(skb, RTA_DST, dst)) 2616 goto nla_put_failure; 2617 if (src) { 2618 r->rtm_src_len = 32; 2619 if (nla_put_in_addr(skb, RTA_SRC, src)) 2620 goto nla_put_failure; 2621 } 2622 if (rt->dst.dev && 2623 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex)) 2624 goto nla_put_failure; 2625 #ifdef CONFIG_IP_ROUTE_CLASSID 2626 if (rt->dst.tclassid && 2627 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid)) 2628 goto nla_put_failure; 2629 #endif 2630 if (!rt_is_input_route(rt) && 2631 fl4->saddr != src) { 2632 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr)) 2633 goto nla_put_failure; 2634 } 2635 if (rt->rt_uses_gateway && 2636 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gateway)) 2637 goto nla_put_failure; 2638 2639 expires = rt->dst.expires; 2640 if (expires) { 2641 unsigned long now = jiffies; 2642 2643 if (time_before(now, expires)) 2644 expires -= now; 2645 else 2646 expires = 0; 2647 } 2648 2649 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics)); 2650 if (rt->rt_pmtu && expires) 2651 metrics[RTAX_MTU - 1] = rt->rt_pmtu; 2652 if (rt->rt_mtu_locked && expires) 2653 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU); 2654 if (rtnetlink_put_metrics(skb, metrics) < 0) 2655 goto nla_put_failure; 2656 2657 if (fl4->flowi4_mark && 2658 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark)) 2659 goto nla_put_failure; 2660 2661 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) && 2662 nla_put_u32(skb, RTA_UID, 2663 from_kuid_munged(current_user_ns(), fl4->flowi4_uid))) 2664 goto nla_put_failure; 2665 2666 error = rt->dst.error; 2667 2668 if (rt_is_input_route(rt)) { 2669 #ifdef CONFIG_IP_MROUTE 2670 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) && 2671 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) { 2672 int err = ipmr_get_route(net, skb, 2673 fl4->saddr, fl4->daddr, 2674 r, portid); 2675 2676 if (err <= 0) { 2677 if (err == 0) 2678 return 0; 2679 goto nla_put_failure; 2680 } 2681 } else 2682 #endif 2683 if (nla_put_u32(skb, RTA_IIF, skb->dev->ifindex)) 2684 goto nla_put_failure; 2685 } 2686 2687 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0) 2688 goto nla_put_failure; 2689 2690 nlmsg_end(skb, nlh); 2691 return 0; 2692 2693 nla_put_failure: 2694 nlmsg_cancel(skb, nlh); 2695 return -EMSGSIZE; 2696 } 2697 2698 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, 2699 struct netlink_ext_ack *extack) 2700 { 2701 struct net *net = sock_net(in_skb->sk); 2702 struct rtmsg *rtm; 2703 struct nlattr *tb[RTA_MAX+1]; 2704 struct fib_result res = {}; 2705 struct rtable *rt = NULL; 2706 struct flowi4 fl4; 2707 __be32 dst = 0; 2708 __be32 src = 0; 2709 u32 iif; 2710 int err; 2711 int mark; 2712 struct sk_buff *skb; 2713 u32 table_id = RT_TABLE_MAIN; 2714 kuid_t uid; 2715 2716 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy, 2717 extack); 2718 if (err < 0) 2719 goto errout; 2720 2721 rtm = nlmsg_data(nlh); 2722 2723 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 2724 if (!skb) { 2725 err = -ENOBUFS; 2726 goto errout; 2727 } 2728 2729 /* Reserve room for dummy headers, this skb can pass 2730 through good chunk of routing engine. 2731 */ 2732 skb_reset_mac_header(skb); 2733 skb_reset_network_header(skb); 2734 2735 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0; 2736 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0; 2737 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0; 2738 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0; 2739 if (tb[RTA_UID]) 2740 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID])); 2741 else 2742 uid = (iif ? INVALID_UID : current_uid()); 2743 2744 /* Bugfix: need to give ip_route_input enough of an IP header to 2745 * not gag. 2746 */ 2747 ip_hdr(skb)->protocol = IPPROTO_UDP; 2748 ip_hdr(skb)->saddr = src; 2749 ip_hdr(skb)->daddr = dst; 2750 2751 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr)); 2752 2753 memset(&fl4, 0, sizeof(fl4)); 2754 fl4.daddr = dst; 2755 fl4.saddr = src; 2756 fl4.flowi4_tos = rtm->rtm_tos; 2757 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0; 2758 fl4.flowi4_mark = mark; 2759 fl4.flowi4_uid = uid; 2760 2761 rcu_read_lock(); 2762 2763 if (iif) { 2764 struct net_device *dev; 2765 2766 dev = dev_get_by_index_rcu(net, iif); 2767 if (!dev) { 2768 err = -ENODEV; 2769 goto errout_free; 2770 } 2771 2772 skb->protocol = htons(ETH_P_IP); 2773 skb->dev = dev; 2774 skb->mark = mark; 2775 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos, 2776 dev, &res); 2777 2778 rt = skb_rtable(skb); 2779 if (err == 0 && rt->dst.error) 2780 err = -rt->dst.error; 2781 } else { 2782 fl4.flowi4_iif = LOOPBACK_IFINDEX; 2783 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb); 2784 err = 0; 2785 if (IS_ERR(rt)) 2786 err = PTR_ERR(rt); 2787 else 2788 skb_dst_set(skb, &rt->dst); 2789 } 2790 2791 if (err) 2792 goto errout_free; 2793 2794 if (rtm->rtm_flags & RTM_F_NOTIFY) 2795 rt->rt_flags |= RTCF_NOTIFY; 2796 2797 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE) 2798 table_id = res.table ? res.table->tb_id : 0; 2799 2800 if (rtm->rtm_flags & RTM_F_FIB_MATCH) { 2801 if (!res.fi) { 2802 err = fib_props[res.type].error; 2803 if (!err) 2804 err = -EHOSTUNREACH; 2805 goto errout_free; 2806 } 2807 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid, 2808 nlh->nlmsg_seq, RTM_NEWROUTE, table_id, 2809 rt->rt_type, res.prefix, res.prefixlen, 2810 fl4.flowi4_tos, res.fi, 0); 2811 } else { 2812 err = rt_fill_info(net, dst, src, table_id, &fl4, skb, 2813 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq); 2814 } 2815 if (err < 0) 2816 goto errout_free; 2817 2818 rcu_read_unlock(); 2819 2820 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 2821 errout: 2822 return err; 2823 2824 errout_free: 2825 rcu_read_unlock(); 2826 kfree_skb(skb); 2827 goto errout; 2828 } 2829 2830 void ip_rt_multicast_event(struct in_device *in_dev) 2831 { 2832 rt_cache_flush(dev_net(in_dev->dev)); 2833 } 2834 2835 #ifdef CONFIG_SYSCTL 2836 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 2837 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 2838 static int ip_rt_gc_elasticity __read_mostly = 8; 2839 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU; 2840 2841 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write, 2842 void __user *buffer, 2843 size_t *lenp, loff_t *ppos) 2844 { 2845 struct net *net = (struct net *)__ctl->extra1; 2846 2847 if (write) { 2848 rt_cache_flush(net); 2849 fnhe_genid_bump(net); 2850 return 0; 2851 } 2852 2853 return -EINVAL; 2854 } 2855 2856 static struct ctl_table ipv4_route_table[] = { 2857 { 2858 .procname = "gc_thresh", 2859 .data = &ipv4_dst_ops.gc_thresh, 2860 .maxlen = sizeof(int), 2861 .mode = 0644, 2862 .proc_handler = proc_dointvec, 2863 }, 2864 { 2865 .procname = "max_size", 2866 .data = &ip_rt_max_size, 2867 .maxlen = sizeof(int), 2868 .mode = 0644, 2869 .proc_handler = proc_dointvec, 2870 }, 2871 { 2872 /* Deprecated. Use gc_min_interval_ms */ 2873 2874 .procname = "gc_min_interval", 2875 .data = &ip_rt_gc_min_interval, 2876 .maxlen = sizeof(int), 2877 .mode = 0644, 2878 .proc_handler = proc_dointvec_jiffies, 2879 }, 2880 { 2881 .procname = "gc_min_interval_ms", 2882 .data = &ip_rt_gc_min_interval, 2883 .maxlen = sizeof(int), 2884 .mode = 0644, 2885 .proc_handler = proc_dointvec_ms_jiffies, 2886 }, 2887 { 2888 .procname = "gc_timeout", 2889 .data = &ip_rt_gc_timeout, 2890 .maxlen = sizeof(int), 2891 .mode = 0644, 2892 .proc_handler = proc_dointvec_jiffies, 2893 }, 2894 { 2895 .procname = "gc_interval", 2896 .data = &ip_rt_gc_interval, 2897 .maxlen = sizeof(int), 2898 .mode = 0644, 2899 .proc_handler = proc_dointvec_jiffies, 2900 }, 2901 { 2902 .procname = "redirect_load", 2903 .data = &ip_rt_redirect_load, 2904 .maxlen = sizeof(int), 2905 .mode = 0644, 2906 .proc_handler = proc_dointvec, 2907 }, 2908 { 2909 .procname = "redirect_number", 2910 .data = &ip_rt_redirect_number, 2911 .maxlen = sizeof(int), 2912 .mode = 0644, 2913 .proc_handler = proc_dointvec, 2914 }, 2915 { 2916 .procname = "redirect_silence", 2917 .data = &ip_rt_redirect_silence, 2918 .maxlen = sizeof(int), 2919 .mode = 0644, 2920 .proc_handler = proc_dointvec, 2921 }, 2922 { 2923 .procname = "error_cost", 2924 .data = &ip_rt_error_cost, 2925 .maxlen = sizeof(int), 2926 .mode = 0644, 2927 .proc_handler = proc_dointvec, 2928 }, 2929 { 2930 .procname = "error_burst", 2931 .data = &ip_rt_error_burst, 2932 .maxlen = sizeof(int), 2933 .mode = 0644, 2934 .proc_handler = proc_dointvec, 2935 }, 2936 { 2937 .procname = "gc_elasticity", 2938 .data = &ip_rt_gc_elasticity, 2939 .maxlen = sizeof(int), 2940 .mode = 0644, 2941 .proc_handler = proc_dointvec, 2942 }, 2943 { 2944 .procname = "mtu_expires", 2945 .data = &ip_rt_mtu_expires, 2946 .maxlen = sizeof(int), 2947 .mode = 0644, 2948 .proc_handler = proc_dointvec_jiffies, 2949 }, 2950 { 2951 .procname = "min_pmtu", 2952 .data = &ip_rt_min_pmtu, 2953 .maxlen = sizeof(int), 2954 .mode = 0644, 2955 .proc_handler = proc_dointvec_minmax, 2956 .extra1 = &ip_min_valid_pmtu, 2957 }, 2958 { 2959 .procname = "min_adv_mss", 2960 .data = &ip_rt_min_advmss, 2961 .maxlen = sizeof(int), 2962 .mode = 0644, 2963 .proc_handler = proc_dointvec, 2964 }, 2965 { } 2966 }; 2967 2968 static struct ctl_table ipv4_route_flush_table[] = { 2969 { 2970 .procname = "flush", 2971 .maxlen = sizeof(int), 2972 .mode = 0200, 2973 .proc_handler = ipv4_sysctl_rtcache_flush, 2974 }, 2975 { }, 2976 }; 2977 2978 static __net_init int sysctl_route_net_init(struct net *net) 2979 { 2980 struct ctl_table *tbl; 2981 2982 tbl = ipv4_route_flush_table; 2983 if (!net_eq(net, &init_net)) { 2984 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL); 2985 if (!tbl) 2986 goto err_dup; 2987 2988 /* Don't export sysctls to unprivileged users */ 2989 if (net->user_ns != &init_user_ns) 2990 tbl[0].procname = NULL; 2991 } 2992 tbl[0].extra1 = net; 2993 2994 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl); 2995 if (!net->ipv4.route_hdr) 2996 goto err_reg; 2997 return 0; 2998 2999 err_reg: 3000 if (tbl != ipv4_route_flush_table) 3001 kfree(tbl); 3002 err_dup: 3003 return -ENOMEM; 3004 } 3005 3006 static __net_exit void sysctl_route_net_exit(struct net *net) 3007 { 3008 struct ctl_table *tbl; 3009 3010 tbl = net->ipv4.route_hdr->ctl_table_arg; 3011 unregister_net_sysctl_table(net->ipv4.route_hdr); 3012 BUG_ON(tbl == ipv4_route_flush_table); 3013 kfree(tbl); 3014 } 3015 3016 static __net_initdata struct pernet_operations sysctl_route_ops = { 3017 .init = sysctl_route_net_init, 3018 .exit = sysctl_route_net_exit, 3019 }; 3020 #endif 3021 3022 static __net_init int rt_genid_init(struct net *net) 3023 { 3024 atomic_set(&net->ipv4.rt_genid, 0); 3025 atomic_set(&net->fnhe_genid, 0); 3026 atomic_set(&net->ipv4.dev_addr_genid, get_random_int()); 3027 return 0; 3028 } 3029 3030 static __net_initdata struct pernet_operations rt_genid_ops = { 3031 .init = rt_genid_init, 3032 }; 3033 3034 static int __net_init ipv4_inetpeer_init(struct net *net) 3035 { 3036 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL); 3037 3038 if (!bp) 3039 return -ENOMEM; 3040 inet_peer_base_init(bp); 3041 net->ipv4.peers = bp; 3042 return 0; 3043 } 3044 3045 static void __net_exit ipv4_inetpeer_exit(struct net *net) 3046 { 3047 struct inet_peer_base *bp = net->ipv4.peers; 3048 3049 net->ipv4.peers = NULL; 3050 inetpeer_invalidate_tree(bp); 3051 kfree(bp); 3052 } 3053 3054 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = { 3055 .init = ipv4_inetpeer_init, 3056 .exit = ipv4_inetpeer_exit, 3057 }; 3058 3059 #ifdef CONFIG_IP_ROUTE_CLASSID 3060 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 3061 #endif /* CONFIG_IP_ROUTE_CLASSID */ 3062 3063 int __init ip_rt_init(void) 3064 { 3065 int cpu; 3066 3067 ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL); 3068 if (!ip_idents) 3069 panic("IP: failed to allocate ip_idents\n"); 3070 3071 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents)); 3072 3073 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL); 3074 if (!ip_tstamps) 3075 panic("IP: failed to allocate ip_tstamps\n"); 3076 3077 for_each_possible_cpu(cpu) { 3078 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 3079 3080 INIT_LIST_HEAD(&ul->head); 3081 spin_lock_init(&ul->lock); 3082 } 3083 #ifdef CONFIG_IP_ROUTE_CLASSID 3084 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 3085 if (!ip_rt_acct) 3086 panic("IP: failed to allocate ip_rt_acct\n"); 3087 #endif 3088 3089 ipv4_dst_ops.kmem_cachep = 3090 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0, 3091 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 3092 3093 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 3094 3095 if (dst_entries_init(&ipv4_dst_ops) < 0) 3096 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 3097 3098 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 3099 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 3100 3101 ipv4_dst_ops.gc_thresh = ~0; 3102 ip_rt_max_size = INT_MAX; 3103 3104 devinet_init(); 3105 ip_fib_init(); 3106 3107 if (ip_rt_proc_init()) 3108 pr_err("Unable to create route proc files\n"); 3109 #ifdef CONFIG_XFRM 3110 xfrm_init(); 3111 xfrm4_init(); 3112 #endif 3113 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, 3114 RTNL_FLAG_DOIT_UNLOCKED); 3115 3116 #ifdef CONFIG_SYSCTL 3117 register_pernet_subsys(&sysctl_route_ops); 3118 #endif 3119 register_pernet_subsys(&rt_genid_ops); 3120 register_pernet_subsys(&ipv4_inetpeer_ops); 3121 return 0; 3122 } 3123 3124 #ifdef CONFIG_SYSCTL 3125 /* 3126 * We really need to sanitize the damn ipv4 init order, then all 3127 * this nonsense will go away. 3128 */ 3129 void __init ip_static_sysctl_init(void) 3130 { 3131 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table); 3132 } 3133 #endif 3134