1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * ROUTE - implementation of the IP router. 8 * 9 * Authors: Ross Biro 10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 11 * Alan Cox, <gw4pts@gw4pts.ampr.org> 12 * Linus Torvalds, <Linus.Torvalds@helsinki.fi> 13 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 14 * 15 * Fixes: 16 * Alan Cox : Verify area fixes. 17 * Alan Cox : cli() protects routing changes 18 * Rui Oliveira : ICMP routing table updates 19 * (rco@di.uminho.pt) Routing table insertion and update 20 * Linus Torvalds : Rewrote bits to be sensible 21 * Alan Cox : Added BSD route gw semantics 22 * Alan Cox : Super /proc >4K 23 * Alan Cox : MTU in route table 24 * Alan Cox : MSS actually. Also added the window 25 * clamper. 26 * Sam Lantinga : Fixed route matching in rt_del() 27 * Alan Cox : Routing cache support. 28 * Alan Cox : Removed compatibility cruft. 29 * Alan Cox : RTF_REJECT support. 30 * Alan Cox : TCP irtt support. 31 * Jonathan Naylor : Added Metric support. 32 * Miquel van Smoorenburg : BSD API fixes. 33 * Miquel van Smoorenburg : Metrics. 34 * Alan Cox : Use __u32 properly 35 * Alan Cox : Aligned routing errors more closely with BSD 36 * our system is still very different. 37 * Alan Cox : Faster /proc handling 38 * Alexey Kuznetsov : Massive rework to support tree based routing, 39 * routing caches and better behaviour. 40 * 41 * Olaf Erb : irtt wasn't being copied right. 42 * Bjorn Ekwall : Kerneld route support. 43 * Alan Cox : Multicast fixed (I hope) 44 * Pavel Krauz : Limited broadcast fixed 45 * Mike McLagan : Routing by source 46 * Alexey Kuznetsov : End of old history. Split to fib.c and 47 * route.c and rewritten from scratch. 48 * Andi Kleen : Load-limit warning messages. 49 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 50 * Vitaly E. Lavrov : Race condition in ip_route_input_slow. 51 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow. 52 * Vladimir V. Ivanov : IP rule info (flowid) is really useful. 53 * Marc Boucher : routing by fwmark 54 * Robert Olsson : Added rt_cache statistics 55 * Arnaldo C. Melo : Convert proc stuff to seq_file 56 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes. 57 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect 58 * Ilia Sotnikov : Removed TOS from hash calculations 59 */ 60 61 #define pr_fmt(fmt) "IPv4: " fmt 62 63 #include <linux/module.h> 64 #include <linux/uaccess.h> 65 #include <linux/bitops.h> 66 #include <linux/types.h> 67 #include <linux/kernel.h> 68 #include <linux/mm.h> 69 #include <linux/memblock.h> 70 #include <linux/string.h> 71 #include <linux/socket.h> 72 #include <linux/sockios.h> 73 #include <linux/errno.h> 74 #include <linux/in.h> 75 #include <linux/inet.h> 76 #include <linux/netdevice.h> 77 #include <linux/proc_fs.h> 78 #include <linux/init.h> 79 #include <linux/skbuff.h> 80 #include <linux/inetdevice.h> 81 #include <linux/igmp.h> 82 #include <linux/pkt_sched.h> 83 #include <linux/mroute.h> 84 #include <linux/netfilter_ipv4.h> 85 #include <linux/random.h> 86 #include <linux/rcupdate.h> 87 #include <linux/times.h> 88 #include <linux/slab.h> 89 #include <linux/jhash.h> 90 #include <net/dst.h> 91 #include <net/dst_metadata.h> 92 #include <net/net_namespace.h> 93 #include <net/protocol.h> 94 #include <net/ip.h> 95 #include <net/route.h> 96 #include <net/inetpeer.h> 97 #include <net/sock.h> 98 #include <net/ip_fib.h> 99 #include <net/nexthop.h> 100 #include <net/arp.h> 101 #include <net/tcp.h> 102 #include <net/icmp.h> 103 #include <net/xfrm.h> 104 #include <net/lwtunnel.h> 105 #include <net/netevent.h> 106 #include <net/rtnetlink.h> 107 #ifdef CONFIG_SYSCTL 108 #include <linux/sysctl.h> 109 #endif 110 #include <net/secure_seq.h> 111 #include <net/ip_tunnels.h> 112 #include <net/l3mdev.h> 113 114 #include "fib_lookup.h" 115 116 #define RT_FL_TOS(oldflp4) \ 117 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)) 118 119 #define RT_GC_TIMEOUT (300*HZ) 120 121 static int ip_rt_max_size; 122 static int ip_rt_redirect_number __read_mostly = 9; 123 static int ip_rt_redirect_load __read_mostly = HZ / 50; 124 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1)); 125 static int ip_rt_error_cost __read_mostly = HZ; 126 static int ip_rt_error_burst __read_mostly = 5 * HZ; 127 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ; 128 static u32 ip_rt_min_pmtu __read_mostly = 512 + 20 + 20; 129 static int ip_rt_min_advmss __read_mostly = 256; 130 131 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT; 132 133 /* 134 * Interface to generic destination cache. 135 */ 136 137 INDIRECT_CALLABLE_SCOPE 138 struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie); 139 static unsigned int ipv4_default_advmss(const struct dst_entry *dst); 140 INDIRECT_CALLABLE_SCOPE 141 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 bool confirm_neigh); 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 void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos) 239 { 240 int cpu; 241 242 if (*pos == 0) 243 return SEQ_START_TOKEN; 244 245 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) { 246 if (!cpu_possible(cpu)) 247 continue; 248 *pos = cpu+1; 249 return &per_cpu(rt_cache_stat, cpu); 250 } 251 return NULL; 252 } 253 254 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos) 255 { 256 int cpu; 257 258 for (cpu = *pos; 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 (*pos)++; 265 return NULL; 266 267 } 268 269 static void rt_cpu_seq_stop(struct seq_file *seq, void *v) 270 { 271 272 } 273 274 static int rt_cpu_seq_show(struct seq_file *seq, void *v) 275 { 276 struct rt_cache_stat *st = v; 277 278 if (v == SEQ_START_TOKEN) { 279 seq_puts(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"); 280 return 0; 281 } 282 283 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x " 284 "%08x %08x %08x %08x %08x %08x " 285 "%08x %08x %08x %08x\n", 286 dst_entries_get_slow(&ipv4_dst_ops), 287 0, /* st->in_hit */ 288 st->in_slow_tot, 289 st->in_slow_mc, 290 st->in_no_route, 291 st->in_brd, 292 st->in_martian_dst, 293 st->in_martian_src, 294 295 0, /* st->out_hit */ 296 st->out_slow_tot, 297 st->out_slow_mc, 298 299 0, /* st->gc_total */ 300 0, /* st->gc_ignored */ 301 0, /* st->gc_goal_miss */ 302 0, /* st->gc_dst_overflow */ 303 0, /* st->in_hlist_search */ 304 0 /* st->out_hlist_search */ 305 ); 306 return 0; 307 } 308 309 static const struct seq_operations rt_cpu_seq_ops = { 310 .start = rt_cpu_seq_start, 311 .next = rt_cpu_seq_next, 312 .stop = rt_cpu_seq_stop, 313 .show = rt_cpu_seq_show, 314 }; 315 316 #ifdef CONFIG_IP_ROUTE_CLASSID 317 static int rt_acct_proc_show(struct seq_file *m, void *v) 318 { 319 struct ip_rt_acct *dst, *src; 320 unsigned int i, j; 321 322 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL); 323 if (!dst) 324 return -ENOMEM; 325 326 for_each_possible_cpu(i) { 327 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i); 328 for (j = 0; j < 256; j++) { 329 dst[j].o_bytes += src[j].o_bytes; 330 dst[j].o_packets += src[j].o_packets; 331 dst[j].i_bytes += src[j].i_bytes; 332 dst[j].i_packets += src[j].i_packets; 333 } 334 } 335 336 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct)); 337 kfree(dst); 338 return 0; 339 } 340 #endif 341 342 static int __net_init ip_rt_do_proc_init(struct net *net) 343 { 344 struct proc_dir_entry *pde; 345 346 pde = proc_create_seq("rt_cache", 0444, net->proc_net, 347 &rt_cache_seq_ops); 348 if (!pde) 349 goto err1; 350 351 pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat, 352 &rt_cpu_seq_ops); 353 if (!pde) 354 goto err2; 355 356 #ifdef CONFIG_IP_ROUTE_CLASSID 357 pde = proc_create_single("rt_acct", 0, net->proc_net, 358 rt_acct_proc_show); 359 if (!pde) 360 goto err3; 361 #endif 362 return 0; 363 364 #ifdef CONFIG_IP_ROUTE_CLASSID 365 err3: 366 remove_proc_entry("rt_cache", net->proc_net_stat); 367 #endif 368 err2: 369 remove_proc_entry("rt_cache", net->proc_net); 370 err1: 371 return -ENOMEM; 372 } 373 374 static void __net_exit ip_rt_do_proc_exit(struct net *net) 375 { 376 remove_proc_entry("rt_cache", net->proc_net_stat); 377 remove_proc_entry("rt_cache", net->proc_net); 378 #ifdef CONFIG_IP_ROUTE_CLASSID 379 remove_proc_entry("rt_acct", net->proc_net); 380 #endif 381 } 382 383 static struct pernet_operations ip_rt_proc_ops __net_initdata = { 384 .init = ip_rt_do_proc_init, 385 .exit = ip_rt_do_proc_exit, 386 }; 387 388 static int __init ip_rt_proc_init(void) 389 { 390 return register_pernet_subsys(&ip_rt_proc_ops); 391 } 392 393 #else 394 static inline int ip_rt_proc_init(void) 395 { 396 return 0; 397 } 398 #endif /* CONFIG_PROC_FS */ 399 400 static inline bool rt_is_expired(const struct rtable *rth) 401 { 402 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev)); 403 } 404 405 void rt_cache_flush(struct net *net) 406 { 407 rt_genid_bump_ipv4(net); 408 } 409 410 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 411 struct sk_buff *skb, 412 const void *daddr) 413 { 414 const struct rtable *rt = container_of(dst, struct rtable, dst); 415 struct net_device *dev = dst->dev; 416 struct neighbour *n; 417 418 rcu_read_lock_bh(); 419 420 if (likely(rt->rt_gw_family == AF_INET)) { 421 n = ip_neigh_gw4(dev, rt->rt_gw4); 422 } else if (rt->rt_gw_family == AF_INET6) { 423 n = ip_neigh_gw6(dev, &rt->rt_gw6); 424 } else { 425 __be32 pkey; 426 427 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr); 428 n = ip_neigh_gw4(dev, pkey); 429 } 430 431 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt)) 432 n = NULL; 433 434 rcu_read_unlock_bh(); 435 436 return n; 437 } 438 439 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr) 440 { 441 const struct rtable *rt = container_of(dst, struct rtable, dst); 442 struct net_device *dev = dst->dev; 443 const __be32 *pkey = daddr; 444 445 if (rt->rt_gw_family == AF_INET) { 446 pkey = (const __be32 *)&rt->rt_gw4; 447 } else if (rt->rt_gw_family == AF_INET6) { 448 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6); 449 } else if (!daddr || 450 (rt->rt_flags & 451 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) { 452 return; 453 } 454 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey); 455 } 456 457 /* Hash tables of size 2048..262144 depending on RAM size. 458 * Each bucket uses 8 bytes. 459 */ 460 static u32 ip_idents_mask __read_mostly; 461 static atomic_t *ip_idents __read_mostly; 462 static u32 *ip_tstamps __read_mostly; 463 464 /* In order to protect privacy, we add a perturbation to identifiers 465 * if one generator is seldom used. This makes hard for an attacker 466 * to infer how many packets were sent between two points in time. 467 */ 468 u32 ip_idents_reserve(u32 hash, int segs) 469 { 470 u32 bucket, old, now = (u32)jiffies; 471 atomic_t *p_id; 472 u32 *p_tstamp; 473 u32 delta = 0; 474 475 bucket = hash & ip_idents_mask; 476 p_tstamp = ip_tstamps + bucket; 477 p_id = ip_idents + bucket; 478 old = READ_ONCE(*p_tstamp); 479 480 if (old != now && cmpxchg(p_tstamp, old, now) == old) 481 delta = prandom_u32_max(now - old); 482 483 /* If UBSAN reports an error there, please make sure your compiler 484 * supports -fno-strict-overflow before reporting it that was a bug 485 * in UBSAN, and it has been fixed in GCC-8. 486 */ 487 return atomic_add_return(segs + delta, p_id) - segs; 488 } 489 EXPORT_SYMBOL(ip_idents_reserve); 490 491 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs) 492 { 493 u32 hash, id; 494 495 /* Note the following code is not safe, but this is okay. */ 496 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key))) 497 get_random_bytes(&net->ipv4.ip_id_key, 498 sizeof(net->ipv4.ip_id_key)); 499 500 hash = siphash_3u32((__force u32)iph->daddr, 501 (__force u32)iph->saddr, 502 iph->protocol, 503 &net->ipv4.ip_id_key); 504 id = ip_idents_reserve(hash, segs); 505 iph->id = htons(id); 506 } 507 EXPORT_SYMBOL(__ip_select_ident); 508 509 static void __build_flow_key(const struct net *net, struct flowi4 *fl4, 510 const struct sock *sk, 511 const struct iphdr *iph, 512 int oif, u8 tos, 513 u8 prot, u32 mark, int flow_flags) 514 { 515 if (sk) { 516 const struct inet_sock *inet = inet_sk(sk); 517 518 oif = sk->sk_bound_dev_if; 519 mark = sk->sk_mark; 520 tos = RT_CONN_FLAGS(sk); 521 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol; 522 } 523 flowi4_init_output(fl4, oif, mark, tos, 524 RT_SCOPE_UNIVERSE, prot, 525 flow_flags, 526 iph->daddr, iph->saddr, 0, 0, 527 sock_net_uid(net, sk)); 528 } 529 530 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb, 531 const struct sock *sk) 532 { 533 const struct net *net = dev_net(skb->dev); 534 const struct iphdr *iph = ip_hdr(skb); 535 int oif = skb->dev->ifindex; 536 u8 tos = RT_TOS(iph->tos); 537 u8 prot = iph->protocol; 538 u32 mark = skb->mark; 539 540 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0); 541 } 542 543 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk) 544 { 545 const struct inet_sock *inet = inet_sk(sk); 546 const struct ip_options_rcu *inet_opt; 547 __be32 daddr = inet->inet_daddr; 548 549 rcu_read_lock(); 550 inet_opt = rcu_dereference(inet->inet_opt); 551 if (inet_opt && inet_opt->opt.srr) 552 daddr = inet_opt->opt.faddr; 553 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark, 554 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 555 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol, 556 inet_sk_flowi_flags(sk), 557 daddr, inet->inet_saddr, 0, 0, sk->sk_uid); 558 rcu_read_unlock(); 559 } 560 561 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk, 562 const struct sk_buff *skb) 563 { 564 if (skb) 565 build_skb_flow_key(fl4, skb, sk); 566 else 567 build_sk_flow_key(fl4, sk); 568 } 569 570 static DEFINE_SPINLOCK(fnhe_lock); 571 572 static void fnhe_flush_routes(struct fib_nh_exception *fnhe) 573 { 574 struct rtable *rt; 575 576 rt = rcu_dereference(fnhe->fnhe_rth_input); 577 if (rt) { 578 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL); 579 dst_dev_put(&rt->dst); 580 dst_release(&rt->dst); 581 } 582 rt = rcu_dereference(fnhe->fnhe_rth_output); 583 if (rt) { 584 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL); 585 dst_dev_put(&rt->dst); 586 dst_release(&rt->dst); 587 } 588 } 589 590 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash) 591 { 592 struct fib_nh_exception *fnhe, *oldest; 593 594 oldest = rcu_dereference(hash->chain); 595 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe; 596 fnhe = rcu_dereference(fnhe->fnhe_next)) { 597 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) 598 oldest = fnhe; 599 } 600 fnhe_flush_routes(oldest); 601 return oldest; 602 } 603 604 static inline u32 fnhe_hashfun(__be32 daddr) 605 { 606 static u32 fnhe_hashrnd __read_mostly; 607 u32 hval; 608 609 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd)); 610 hval = jhash_1word((__force u32)daddr, fnhe_hashrnd); 611 return hash_32(hval, FNHE_HASH_SHIFT); 612 } 613 614 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe) 615 { 616 rt->rt_pmtu = fnhe->fnhe_pmtu; 617 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked; 618 rt->dst.expires = fnhe->fnhe_expires; 619 620 if (fnhe->fnhe_gw) { 621 rt->rt_flags |= RTCF_REDIRECTED; 622 rt->rt_uses_gateway = 1; 623 rt->rt_gw_family = AF_INET; 624 rt->rt_gw4 = fnhe->fnhe_gw; 625 } 626 } 627 628 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr, 629 __be32 gw, u32 pmtu, bool lock, 630 unsigned long expires) 631 { 632 struct fnhe_hash_bucket *hash; 633 struct fib_nh_exception *fnhe; 634 struct rtable *rt; 635 u32 genid, hval; 636 unsigned int i; 637 int depth; 638 639 genid = fnhe_genid(dev_net(nhc->nhc_dev)); 640 hval = fnhe_hashfun(daddr); 641 642 spin_lock_bh(&fnhe_lock); 643 644 hash = rcu_dereference(nhc->nhc_exceptions); 645 if (!hash) { 646 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC); 647 if (!hash) 648 goto out_unlock; 649 rcu_assign_pointer(nhc->nhc_exceptions, hash); 650 } 651 652 hash += hval; 653 654 depth = 0; 655 for (fnhe = rcu_dereference(hash->chain); fnhe; 656 fnhe = rcu_dereference(fnhe->fnhe_next)) { 657 if (fnhe->fnhe_daddr == daddr) 658 break; 659 depth++; 660 } 661 662 if (fnhe) { 663 if (fnhe->fnhe_genid != genid) 664 fnhe->fnhe_genid = genid; 665 if (gw) 666 fnhe->fnhe_gw = gw; 667 if (pmtu) { 668 fnhe->fnhe_pmtu = pmtu; 669 fnhe->fnhe_mtu_locked = lock; 670 } 671 fnhe->fnhe_expires = max(1UL, expires); 672 /* Update all cached dsts too */ 673 rt = rcu_dereference(fnhe->fnhe_rth_input); 674 if (rt) 675 fill_route_from_fnhe(rt, fnhe); 676 rt = rcu_dereference(fnhe->fnhe_rth_output); 677 if (rt) 678 fill_route_from_fnhe(rt, fnhe); 679 } else { 680 if (depth > FNHE_RECLAIM_DEPTH) 681 fnhe = fnhe_oldest(hash); 682 else { 683 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC); 684 if (!fnhe) 685 goto out_unlock; 686 687 fnhe->fnhe_next = hash->chain; 688 rcu_assign_pointer(hash->chain, fnhe); 689 } 690 fnhe->fnhe_genid = genid; 691 fnhe->fnhe_daddr = daddr; 692 fnhe->fnhe_gw = gw; 693 fnhe->fnhe_pmtu = pmtu; 694 fnhe->fnhe_mtu_locked = lock; 695 fnhe->fnhe_expires = max(1UL, expires); 696 697 /* Exception created; mark the cached routes for the nexthop 698 * stale, so anyone caching it rechecks if this exception 699 * applies to them. 700 */ 701 rt = rcu_dereference(nhc->nhc_rth_input); 702 if (rt) 703 rt->dst.obsolete = DST_OBSOLETE_KILL; 704 705 for_each_possible_cpu(i) { 706 struct rtable __rcu **prt; 707 708 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i); 709 rt = rcu_dereference(*prt); 710 if (rt) 711 rt->dst.obsolete = DST_OBSOLETE_KILL; 712 } 713 } 714 715 fnhe->fnhe_stamp = jiffies; 716 717 out_unlock: 718 spin_unlock_bh(&fnhe_lock); 719 } 720 721 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4, 722 bool kill_route) 723 { 724 __be32 new_gw = icmp_hdr(skb)->un.gateway; 725 __be32 old_gw = ip_hdr(skb)->saddr; 726 struct net_device *dev = skb->dev; 727 struct in_device *in_dev; 728 struct fib_result res; 729 struct neighbour *n; 730 struct net *net; 731 732 switch (icmp_hdr(skb)->code & 7) { 733 case ICMP_REDIR_NET: 734 case ICMP_REDIR_NETTOS: 735 case ICMP_REDIR_HOST: 736 case ICMP_REDIR_HOSTTOS: 737 break; 738 739 default: 740 return; 741 } 742 743 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw) 744 return; 745 746 in_dev = __in_dev_get_rcu(dev); 747 if (!in_dev) 748 return; 749 750 net = dev_net(dev); 751 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) || 752 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) || 753 ipv4_is_zeronet(new_gw)) 754 goto reject_redirect; 755 756 if (!IN_DEV_SHARED_MEDIA(in_dev)) { 757 if (!inet_addr_onlink(in_dev, new_gw, old_gw)) 758 goto reject_redirect; 759 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) 760 goto reject_redirect; 761 } else { 762 if (inet_addr_type(net, new_gw) != RTN_UNICAST) 763 goto reject_redirect; 764 } 765 766 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw); 767 if (!n) 768 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev); 769 if (!IS_ERR(n)) { 770 if (!(n->nud_state & NUD_VALID)) { 771 neigh_event_send(n, NULL); 772 } else { 773 if (fib_lookup(net, fl4, &res, 0) == 0) { 774 struct fib_nh_common *nhc; 775 776 fib_select_path(net, &res, fl4, skb); 777 nhc = FIB_RES_NHC(res); 778 update_or_create_fnhe(nhc, fl4->daddr, new_gw, 779 0, false, 780 jiffies + ip_rt_gc_timeout); 781 } 782 if (kill_route) 783 rt->dst.obsolete = DST_OBSOLETE_KILL; 784 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n); 785 } 786 neigh_release(n); 787 } 788 return; 789 790 reject_redirect: 791 #ifdef CONFIG_IP_ROUTE_VERBOSE 792 if (IN_DEV_LOG_MARTIANS(in_dev)) { 793 const struct iphdr *iph = (const struct iphdr *) skb->data; 794 __be32 daddr = iph->daddr; 795 __be32 saddr = iph->saddr; 796 797 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n" 798 " Advised path = %pI4 -> %pI4\n", 799 &old_gw, dev->name, &new_gw, 800 &saddr, &daddr); 801 } 802 #endif 803 ; 804 } 805 806 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb) 807 { 808 struct rtable *rt; 809 struct flowi4 fl4; 810 const struct iphdr *iph = (const struct iphdr *) skb->data; 811 struct net *net = dev_net(skb->dev); 812 int oif = skb->dev->ifindex; 813 u8 tos = RT_TOS(iph->tos); 814 u8 prot = iph->protocol; 815 u32 mark = skb->mark; 816 817 rt = (struct rtable *) dst; 818 819 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0); 820 __ip_do_redirect(rt, skb, &fl4, true); 821 } 822 823 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst) 824 { 825 struct rtable *rt = (struct rtable *)dst; 826 struct dst_entry *ret = dst; 827 828 if (rt) { 829 if (dst->obsolete > 0) { 830 ip_rt_put(rt); 831 ret = NULL; 832 } else if ((rt->rt_flags & RTCF_REDIRECTED) || 833 rt->dst.expires) { 834 ip_rt_put(rt); 835 ret = NULL; 836 } 837 } 838 return ret; 839 } 840 841 /* 842 * Algorithm: 843 * 1. The first ip_rt_redirect_number redirects are sent 844 * with exponential backoff, then we stop sending them at all, 845 * assuming that the host ignores our redirects. 846 * 2. If we did not see packets requiring redirects 847 * during ip_rt_redirect_silence, we assume that the host 848 * forgot redirected route and start to send redirects again. 849 * 850 * This algorithm is much cheaper and more intelligent than dumb load limiting 851 * in icmp.c. 852 * 853 * NOTE. Do not forget to inhibit load limiting for redirects (redundant) 854 * and "frag. need" (breaks PMTU discovery) in icmp.c. 855 */ 856 857 void ip_rt_send_redirect(struct sk_buff *skb) 858 { 859 struct rtable *rt = skb_rtable(skb); 860 struct in_device *in_dev; 861 struct inet_peer *peer; 862 struct net *net; 863 int log_martians; 864 int vif; 865 866 rcu_read_lock(); 867 in_dev = __in_dev_get_rcu(rt->dst.dev); 868 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) { 869 rcu_read_unlock(); 870 return; 871 } 872 log_martians = IN_DEV_LOG_MARTIANS(in_dev); 873 vif = l3mdev_master_ifindex_rcu(rt->dst.dev); 874 rcu_read_unlock(); 875 876 net = dev_net(rt->dst.dev); 877 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1); 878 if (!peer) { 879 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, 880 rt_nexthop(rt, ip_hdr(skb)->daddr)); 881 return; 882 } 883 884 /* No redirected packets during ip_rt_redirect_silence; 885 * reset the algorithm. 886 */ 887 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) { 888 peer->rate_tokens = 0; 889 peer->n_redirects = 0; 890 } 891 892 /* Too many ignored redirects; do not send anything 893 * set dst.rate_last to the last seen redirected packet. 894 */ 895 if (peer->n_redirects >= ip_rt_redirect_number) { 896 peer->rate_last = jiffies; 897 goto out_put_peer; 898 } 899 900 /* Check for load limit; set rate_last to the latest sent 901 * redirect. 902 */ 903 if (peer->n_redirects == 0 || 904 time_after(jiffies, 905 (peer->rate_last + 906 (ip_rt_redirect_load << peer->n_redirects)))) { 907 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr); 908 909 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw); 910 peer->rate_last = jiffies; 911 ++peer->n_redirects; 912 #ifdef CONFIG_IP_ROUTE_VERBOSE 913 if (log_martians && 914 peer->n_redirects == ip_rt_redirect_number) 915 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n", 916 &ip_hdr(skb)->saddr, inet_iif(skb), 917 &ip_hdr(skb)->daddr, &gw); 918 #endif 919 } 920 out_put_peer: 921 inet_putpeer(peer); 922 } 923 924 static int ip_error(struct sk_buff *skb) 925 { 926 struct rtable *rt = skb_rtable(skb); 927 struct net_device *dev = skb->dev; 928 struct in_device *in_dev; 929 struct inet_peer *peer; 930 unsigned long now; 931 struct net *net; 932 bool send; 933 int code; 934 935 if (netif_is_l3_master(skb->dev)) { 936 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif); 937 if (!dev) 938 goto out; 939 } 940 941 in_dev = __in_dev_get_rcu(dev); 942 943 /* IP on this device is disabled. */ 944 if (!in_dev) 945 goto out; 946 947 net = dev_net(rt->dst.dev); 948 if (!IN_DEV_FORWARD(in_dev)) { 949 switch (rt->dst.error) { 950 case EHOSTUNREACH: 951 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS); 952 break; 953 954 case ENETUNREACH: 955 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 956 break; 957 } 958 goto out; 959 } 960 961 switch (rt->dst.error) { 962 case EINVAL: 963 default: 964 goto out; 965 case EHOSTUNREACH: 966 code = ICMP_HOST_UNREACH; 967 break; 968 case ENETUNREACH: 969 code = ICMP_NET_UNREACH; 970 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 971 break; 972 case EACCES: 973 code = ICMP_PKT_FILTERED; 974 break; 975 } 976 977 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 978 l3mdev_master_ifindex(skb->dev), 1); 979 980 send = true; 981 if (peer) { 982 now = jiffies; 983 peer->rate_tokens += now - peer->rate_last; 984 if (peer->rate_tokens > ip_rt_error_burst) 985 peer->rate_tokens = ip_rt_error_burst; 986 peer->rate_last = now; 987 if (peer->rate_tokens >= ip_rt_error_cost) 988 peer->rate_tokens -= ip_rt_error_cost; 989 else 990 send = false; 991 inet_putpeer(peer); 992 } 993 if (send) 994 icmp_send(skb, ICMP_DEST_UNREACH, code, 0); 995 996 out: kfree_skb(skb); 997 return 0; 998 } 999 1000 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu) 1001 { 1002 struct dst_entry *dst = &rt->dst; 1003 struct net *net = dev_net(dst->dev); 1004 struct fib_result res; 1005 bool lock = false; 1006 u32 old_mtu; 1007 1008 if (ip_mtu_locked(dst)) 1009 return; 1010 1011 old_mtu = ipv4_mtu(dst); 1012 if (old_mtu < mtu) 1013 return; 1014 1015 if (mtu < ip_rt_min_pmtu) { 1016 lock = true; 1017 mtu = min(old_mtu, ip_rt_min_pmtu); 1018 } 1019 1020 if (rt->rt_pmtu == mtu && !lock && 1021 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2)) 1022 return; 1023 1024 rcu_read_lock(); 1025 if (fib_lookup(net, fl4, &res, 0) == 0) { 1026 struct fib_nh_common *nhc; 1027 1028 fib_select_path(net, &res, fl4, NULL); 1029 nhc = FIB_RES_NHC(res); 1030 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1031 jiffies + ip_rt_mtu_expires); 1032 } 1033 rcu_read_unlock(); 1034 } 1035 1036 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 1037 struct sk_buff *skb, u32 mtu, 1038 bool confirm_neigh) 1039 { 1040 struct rtable *rt = (struct rtable *) dst; 1041 struct flowi4 fl4; 1042 1043 ip_rt_build_flow_key(&fl4, sk, skb); 1044 1045 /* Don't make lookup fail for bridged encapsulations */ 1046 if (skb && netif_is_any_bridge_port(skb->dev)) 1047 fl4.flowi4_oif = 0; 1048 1049 __ip_rt_update_pmtu(rt, &fl4, mtu); 1050 } 1051 1052 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, 1053 int oif, u8 protocol) 1054 { 1055 const struct iphdr *iph = (const struct iphdr *)skb->data; 1056 struct flowi4 fl4; 1057 struct rtable *rt; 1058 u32 mark = IP4_REPLY_MARK(net, skb->mark); 1059 1060 __build_flow_key(net, &fl4, NULL, iph, oif, 1061 RT_TOS(iph->tos), protocol, mark, 0); 1062 rt = __ip_route_output_key(net, &fl4); 1063 if (!IS_ERR(rt)) { 1064 __ip_rt_update_pmtu(rt, &fl4, mtu); 1065 ip_rt_put(rt); 1066 } 1067 } 1068 EXPORT_SYMBOL_GPL(ipv4_update_pmtu); 1069 1070 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1071 { 1072 const struct iphdr *iph = (const struct iphdr *)skb->data; 1073 struct flowi4 fl4; 1074 struct rtable *rt; 1075 1076 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0); 1077 1078 if (!fl4.flowi4_mark) 1079 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark); 1080 1081 rt = __ip_route_output_key(sock_net(sk), &fl4); 1082 if (!IS_ERR(rt)) { 1083 __ip_rt_update_pmtu(rt, &fl4, mtu); 1084 ip_rt_put(rt); 1085 } 1086 } 1087 1088 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1089 { 1090 const struct iphdr *iph = (const struct iphdr *)skb->data; 1091 struct flowi4 fl4; 1092 struct rtable *rt; 1093 struct dst_entry *odst = NULL; 1094 bool new = false; 1095 struct net *net = sock_net(sk); 1096 1097 bh_lock_sock(sk); 1098 1099 if (!ip_sk_accept_pmtu(sk)) 1100 goto out; 1101 1102 odst = sk_dst_get(sk); 1103 1104 if (sock_owned_by_user(sk) || !odst) { 1105 __ipv4_sk_update_pmtu(skb, sk, mtu); 1106 goto out; 1107 } 1108 1109 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1110 1111 rt = (struct rtable *)odst; 1112 if (odst->obsolete && !odst->ops->check(odst, 0)) { 1113 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1114 if (IS_ERR(rt)) 1115 goto out; 1116 1117 new = true; 1118 } 1119 1120 __ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu); 1121 1122 if (!dst_check(&rt->dst, 0)) { 1123 if (new) 1124 dst_release(&rt->dst); 1125 1126 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1127 if (IS_ERR(rt)) 1128 goto out; 1129 1130 new = true; 1131 } 1132 1133 if (new) 1134 sk_dst_set(sk, &rt->dst); 1135 1136 out: 1137 bh_unlock_sock(sk); 1138 dst_release(odst); 1139 } 1140 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu); 1141 1142 void ipv4_redirect(struct sk_buff *skb, struct net *net, 1143 int oif, u8 protocol) 1144 { 1145 const struct iphdr *iph = (const struct iphdr *)skb->data; 1146 struct flowi4 fl4; 1147 struct rtable *rt; 1148 1149 __build_flow_key(net, &fl4, NULL, iph, oif, 1150 RT_TOS(iph->tos), protocol, 0, 0); 1151 rt = __ip_route_output_key(net, &fl4); 1152 if (!IS_ERR(rt)) { 1153 __ip_do_redirect(rt, skb, &fl4, false); 1154 ip_rt_put(rt); 1155 } 1156 } 1157 EXPORT_SYMBOL_GPL(ipv4_redirect); 1158 1159 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk) 1160 { 1161 const struct iphdr *iph = (const struct iphdr *)skb->data; 1162 struct flowi4 fl4; 1163 struct rtable *rt; 1164 struct net *net = sock_net(sk); 1165 1166 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1167 rt = __ip_route_output_key(net, &fl4); 1168 if (!IS_ERR(rt)) { 1169 __ip_do_redirect(rt, skb, &fl4, false); 1170 ip_rt_put(rt); 1171 } 1172 } 1173 EXPORT_SYMBOL_GPL(ipv4_sk_redirect); 1174 1175 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst, 1176 u32 cookie) 1177 { 1178 struct rtable *rt = (struct rtable *) dst; 1179 1180 /* All IPV4 dsts are created with ->obsolete set to the value 1181 * DST_OBSOLETE_FORCE_CHK which forces validation calls down 1182 * into this function always. 1183 * 1184 * When a PMTU/redirect information update invalidates a route, 1185 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or 1186 * DST_OBSOLETE_DEAD. 1187 */ 1188 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt)) 1189 return NULL; 1190 return dst; 1191 } 1192 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check); 1193 1194 static void ipv4_send_dest_unreach(struct sk_buff *skb) 1195 { 1196 struct ip_options opt; 1197 int res; 1198 1199 /* Recompile ip options since IPCB may not be valid anymore. 1200 * Also check we have a reasonable ipv4 header. 1201 */ 1202 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) || 1203 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5) 1204 return; 1205 1206 memset(&opt, 0, sizeof(opt)); 1207 if (ip_hdr(skb)->ihl > 5) { 1208 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4)) 1209 return; 1210 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr); 1211 1212 rcu_read_lock(); 1213 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL); 1214 rcu_read_unlock(); 1215 1216 if (res) 1217 return; 1218 } 1219 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt); 1220 } 1221 1222 static void ipv4_link_failure(struct sk_buff *skb) 1223 { 1224 struct rtable *rt; 1225 1226 ipv4_send_dest_unreach(skb); 1227 1228 rt = skb_rtable(skb); 1229 if (rt) 1230 dst_set_expires(&rt->dst, 0); 1231 } 1232 1233 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb) 1234 { 1235 pr_debug("%s: %pI4 -> %pI4, %s\n", 1236 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1237 skb->dev ? skb->dev->name : "?"); 1238 kfree_skb(skb); 1239 WARN_ON(1); 1240 return 0; 1241 } 1242 1243 /* 1244 * We do not cache source address of outgoing interface, 1245 * because it is used only by IP RR, TS and SRR options, 1246 * so that it out of fast path. 1247 * 1248 * BTW remember: "addr" is allowed to be not aligned 1249 * in IP options! 1250 */ 1251 1252 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt) 1253 { 1254 __be32 src; 1255 1256 if (rt_is_output_route(rt)) 1257 src = ip_hdr(skb)->saddr; 1258 else { 1259 struct fib_result res; 1260 struct iphdr *iph = ip_hdr(skb); 1261 struct flowi4 fl4 = { 1262 .daddr = iph->daddr, 1263 .saddr = iph->saddr, 1264 .flowi4_tos = RT_TOS(iph->tos), 1265 .flowi4_oif = rt->dst.dev->ifindex, 1266 .flowi4_iif = skb->dev->ifindex, 1267 .flowi4_mark = skb->mark, 1268 }; 1269 1270 rcu_read_lock(); 1271 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0) 1272 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res); 1273 else 1274 src = inet_select_addr(rt->dst.dev, 1275 rt_nexthop(rt, iph->daddr), 1276 RT_SCOPE_UNIVERSE); 1277 rcu_read_unlock(); 1278 } 1279 memcpy(addr, &src, 4); 1280 } 1281 1282 #ifdef CONFIG_IP_ROUTE_CLASSID 1283 static void set_class_tag(struct rtable *rt, u32 tag) 1284 { 1285 if (!(rt->dst.tclassid & 0xFFFF)) 1286 rt->dst.tclassid |= tag & 0xFFFF; 1287 if (!(rt->dst.tclassid & 0xFFFF0000)) 1288 rt->dst.tclassid |= tag & 0xFFFF0000; 1289 } 1290 #endif 1291 1292 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1293 { 1294 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr); 1295 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size, 1296 ip_rt_min_advmss); 1297 1298 return min(advmss, IPV4_MAX_PMTU - header_size); 1299 } 1300 1301 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst) 1302 { 1303 return ip_dst_mtu_maybe_forward(dst, false); 1304 } 1305 EXPORT_INDIRECT_CALLABLE(ipv4_mtu); 1306 1307 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr) 1308 { 1309 struct fnhe_hash_bucket *hash; 1310 struct fib_nh_exception *fnhe, __rcu **fnhe_p; 1311 u32 hval = fnhe_hashfun(daddr); 1312 1313 spin_lock_bh(&fnhe_lock); 1314 1315 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1316 lockdep_is_held(&fnhe_lock)); 1317 hash += hval; 1318 1319 fnhe_p = &hash->chain; 1320 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock)); 1321 while (fnhe) { 1322 if (fnhe->fnhe_daddr == daddr) { 1323 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected( 1324 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock))); 1325 /* set fnhe_daddr to 0 to ensure it won't bind with 1326 * new dsts in rt_bind_exception(). 1327 */ 1328 fnhe->fnhe_daddr = 0; 1329 fnhe_flush_routes(fnhe); 1330 kfree_rcu(fnhe, rcu); 1331 break; 1332 } 1333 fnhe_p = &fnhe->fnhe_next; 1334 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1335 lockdep_is_held(&fnhe_lock)); 1336 } 1337 1338 spin_unlock_bh(&fnhe_lock); 1339 } 1340 1341 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc, 1342 __be32 daddr) 1343 { 1344 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions); 1345 struct fib_nh_exception *fnhe; 1346 u32 hval; 1347 1348 if (!hash) 1349 return NULL; 1350 1351 hval = fnhe_hashfun(daddr); 1352 1353 for (fnhe = rcu_dereference(hash[hval].chain); fnhe; 1354 fnhe = rcu_dereference(fnhe->fnhe_next)) { 1355 if (fnhe->fnhe_daddr == daddr) { 1356 if (fnhe->fnhe_expires && 1357 time_after(jiffies, fnhe->fnhe_expires)) { 1358 ip_del_fnhe(nhc, daddr); 1359 break; 1360 } 1361 return fnhe; 1362 } 1363 } 1364 return NULL; 1365 } 1366 1367 /* MTU selection: 1368 * 1. mtu on route is locked - use it 1369 * 2. mtu from nexthop exception 1370 * 3. mtu from egress device 1371 */ 1372 1373 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr) 1374 { 1375 struct fib_nh_common *nhc = res->nhc; 1376 struct net_device *dev = nhc->nhc_dev; 1377 struct fib_info *fi = res->fi; 1378 u32 mtu = 0; 1379 1380 if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu || 1381 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU)) 1382 mtu = fi->fib_mtu; 1383 1384 if (likely(!mtu)) { 1385 struct fib_nh_exception *fnhe; 1386 1387 fnhe = find_exception(nhc, daddr); 1388 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires)) 1389 mtu = fnhe->fnhe_pmtu; 1390 } 1391 1392 if (likely(!mtu)) 1393 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU); 1394 1395 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu); 1396 } 1397 1398 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, 1399 __be32 daddr, const bool do_cache) 1400 { 1401 bool ret = false; 1402 1403 spin_lock_bh(&fnhe_lock); 1404 1405 if (daddr == fnhe->fnhe_daddr) { 1406 struct rtable __rcu **porig; 1407 struct rtable *orig; 1408 int genid = fnhe_genid(dev_net(rt->dst.dev)); 1409 1410 if (rt_is_input_route(rt)) 1411 porig = &fnhe->fnhe_rth_input; 1412 else 1413 porig = &fnhe->fnhe_rth_output; 1414 orig = rcu_dereference(*porig); 1415 1416 if (fnhe->fnhe_genid != genid) { 1417 fnhe->fnhe_genid = genid; 1418 fnhe->fnhe_gw = 0; 1419 fnhe->fnhe_pmtu = 0; 1420 fnhe->fnhe_expires = 0; 1421 fnhe->fnhe_mtu_locked = false; 1422 fnhe_flush_routes(fnhe); 1423 orig = NULL; 1424 } 1425 fill_route_from_fnhe(rt, fnhe); 1426 if (!rt->rt_gw4) { 1427 rt->rt_gw4 = daddr; 1428 rt->rt_gw_family = AF_INET; 1429 } 1430 1431 if (do_cache) { 1432 dst_hold(&rt->dst); 1433 rcu_assign_pointer(*porig, rt); 1434 if (orig) { 1435 dst_dev_put(&orig->dst); 1436 dst_release(&orig->dst); 1437 } 1438 ret = true; 1439 } 1440 1441 fnhe->fnhe_stamp = jiffies; 1442 } 1443 spin_unlock_bh(&fnhe_lock); 1444 1445 return ret; 1446 } 1447 1448 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt) 1449 { 1450 struct rtable *orig, *prev, **p; 1451 bool ret = true; 1452 1453 if (rt_is_input_route(rt)) { 1454 p = (struct rtable **)&nhc->nhc_rth_input; 1455 } else { 1456 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 1457 } 1458 orig = *p; 1459 1460 /* hold dst before doing cmpxchg() to avoid race condition 1461 * on this dst 1462 */ 1463 dst_hold(&rt->dst); 1464 prev = cmpxchg(p, orig, rt); 1465 if (prev == orig) { 1466 if (orig) { 1467 rt_add_uncached_list(orig); 1468 dst_release(&orig->dst); 1469 } 1470 } else { 1471 dst_release(&rt->dst); 1472 ret = false; 1473 } 1474 1475 return ret; 1476 } 1477 1478 struct uncached_list { 1479 spinlock_t lock; 1480 struct list_head head; 1481 }; 1482 1483 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list); 1484 1485 void rt_add_uncached_list(struct rtable *rt) 1486 { 1487 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list); 1488 1489 rt->rt_uncached_list = ul; 1490 1491 spin_lock_bh(&ul->lock); 1492 list_add_tail(&rt->rt_uncached, &ul->head); 1493 spin_unlock_bh(&ul->lock); 1494 } 1495 1496 void rt_del_uncached_list(struct rtable *rt) 1497 { 1498 if (!list_empty(&rt->rt_uncached)) { 1499 struct uncached_list *ul = rt->rt_uncached_list; 1500 1501 spin_lock_bh(&ul->lock); 1502 list_del(&rt->rt_uncached); 1503 spin_unlock_bh(&ul->lock); 1504 } 1505 } 1506 1507 static void ipv4_dst_destroy(struct dst_entry *dst) 1508 { 1509 struct rtable *rt = (struct rtable *)dst; 1510 1511 ip_dst_metrics_put(dst); 1512 rt_del_uncached_list(rt); 1513 } 1514 1515 void rt_flush_dev(struct net_device *dev) 1516 { 1517 struct rtable *rt; 1518 int cpu; 1519 1520 for_each_possible_cpu(cpu) { 1521 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 1522 1523 spin_lock_bh(&ul->lock); 1524 list_for_each_entry(rt, &ul->head, rt_uncached) { 1525 if (rt->dst.dev != dev) 1526 continue; 1527 rt->dst.dev = blackhole_netdev; 1528 dev_hold(rt->dst.dev); 1529 dev_put(dev); 1530 } 1531 spin_unlock_bh(&ul->lock); 1532 } 1533 } 1534 1535 static bool rt_cache_valid(const struct rtable *rt) 1536 { 1537 return rt && 1538 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK && 1539 !rt_is_expired(rt); 1540 } 1541 1542 static void rt_set_nexthop(struct rtable *rt, __be32 daddr, 1543 const struct fib_result *res, 1544 struct fib_nh_exception *fnhe, 1545 struct fib_info *fi, u16 type, u32 itag, 1546 const bool do_cache) 1547 { 1548 bool cached = false; 1549 1550 if (fi) { 1551 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1552 1553 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) { 1554 rt->rt_uses_gateway = 1; 1555 rt->rt_gw_family = nhc->nhc_gw_family; 1556 /* only INET and INET6 are supported */ 1557 if (likely(nhc->nhc_gw_family == AF_INET)) 1558 rt->rt_gw4 = nhc->nhc_gw.ipv4; 1559 else 1560 rt->rt_gw6 = nhc->nhc_gw.ipv6; 1561 } 1562 1563 ip_dst_init_metrics(&rt->dst, fi->fib_metrics); 1564 1565 #ifdef CONFIG_IP_ROUTE_CLASSID 1566 if (nhc->nhc_family == AF_INET) { 1567 struct fib_nh *nh; 1568 1569 nh = container_of(nhc, struct fib_nh, nh_common); 1570 rt->dst.tclassid = nh->nh_tclassid; 1571 } 1572 #endif 1573 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 1574 if (unlikely(fnhe)) 1575 cached = rt_bind_exception(rt, fnhe, daddr, do_cache); 1576 else if (do_cache) 1577 cached = rt_cache_route(nhc, rt); 1578 if (unlikely(!cached)) { 1579 /* Routes we intend to cache in nexthop exception or 1580 * FIB nexthop have the DST_NOCACHE bit clear. 1581 * However, if we are unsuccessful at storing this 1582 * route into the cache we really need to set it. 1583 */ 1584 if (!rt->rt_gw4) { 1585 rt->rt_gw_family = AF_INET; 1586 rt->rt_gw4 = daddr; 1587 } 1588 rt_add_uncached_list(rt); 1589 } 1590 } else 1591 rt_add_uncached_list(rt); 1592 1593 #ifdef CONFIG_IP_ROUTE_CLASSID 1594 #ifdef CONFIG_IP_MULTIPLE_TABLES 1595 set_class_tag(rt, res->tclassid); 1596 #endif 1597 set_class_tag(rt, itag); 1598 #endif 1599 } 1600 1601 struct rtable *rt_dst_alloc(struct net_device *dev, 1602 unsigned int flags, u16 type, 1603 bool nopolicy, bool noxfrm) 1604 { 1605 struct rtable *rt; 1606 1607 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1608 (nopolicy ? DST_NOPOLICY : 0) | 1609 (noxfrm ? DST_NOXFRM : 0)); 1610 1611 if (rt) { 1612 rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1613 rt->rt_flags = flags; 1614 rt->rt_type = type; 1615 rt->rt_is_input = 0; 1616 rt->rt_iif = 0; 1617 rt->rt_pmtu = 0; 1618 rt->rt_mtu_locked = 0; 1619 rt->rt_uses_gateway = 0; 1620 rt->rt_gw_family = 0; 1621 rt->rt_gw4 = 0; 1622 INIT_LIST_HEAD(&rt->rt_uncached); 1623 1624 rt->dst.output = ip_output; 1625 if (flags & RTCF_LOCAL) 1626 rt->dst.input = ip_local_deliver; 1627 } 1628 1629 return rt; 1630 } 1631 EXPORT_SYMBOL(rt_dst_alloc); 1632 1633 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt) 1634 { 1635 struct rtable *new_rt; 1636 1637 new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1638 rt->dst.flags); 1639 1640 if (new_rt) { 1641 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1642 new_rt->rt_flags = rt->rt_flags; 1643 new_rt->rt_type = rt->rt_type; 1644 new_rt->rt_is_input = rt->rt_is_input; 1645 new_rt->rt_iif = rt->rt_iif; 1646 new_rt->rt_pmtu = rt->rt_pmtu; 1647 new_rt->rt_mtu_locked = rt->rt_mtu_locked; 1648 new_rt->rt_gw_family = rt->rt_gw_family; 1649 if (rt->rt_gw_family == AF_INET) 1650 new_rt->rt_gw4 = rt->rt_gw4; 1651 else if (rt->rt_gw_family == AF_INET6) 1652 new_rt->rt_gw6 = rt->rt_gw6; 1653 INIT_LIST_HEAD(&new_rt->rt_uncached); 1654 1655 new_rt->dst.input = rt->dst.input; 1656 new_rt->dst.output = rt->dst.output; 1657 new_rt->dst.error = rt->dst.error; 1658 new_rt->dst.lastuse = jiffies; 1659 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate); 1660 } 1661 return new_rt; 1662 } 1663 EXPORT_SYMBOL(rt_dst_clone); 1664 1665 /* called in rcu_read_lock() section */ 1666 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1667 u8 tos, struct net_device *dev, 1668 struct in_device *in_dev, u32 *itag) 1669 { 1670 int err; 1671 1672 /* Primary sanity checks. */ 1673 if (!in_dev) 1674 return -EINVAL; 1675 1676 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) || 1677 skb->protocol != htons(ETH_P_IP)) 1678 return -EINVAL; 1679 1680 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev)) 1681 return -EINVAL; 1682 1683 if (ipv4_is_zeronet(saddr)) { 1684 if (!ipv4_is_local_multicast(daddr) && 1685 ip_hdr(skb)->protocol != IPPROTO_IGMP) 1686 return -EINVAL; 1687 } else { 1688 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 1689 in_dev, itag); 1690 if (err < 0) 1691 return err; 1692 } 1693 return 0; 1694 } 1695 1696 /* called in rcu_read_lock() section */ 1697 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1698 u8 tos, struct net_device *dev, int our) 1699 { 1700 struct in_device *in_dev = __in_dev_get_rcu(dev); 1701 unsigned int flags = RTCF_MULTICAST; 1702 struct rtable *rth; 1703 u32 itag = 0; 1704 int err; 1705 1706 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag); 1707 if (err) 1708 return err; 1709 1710 if (our) 1711 flags |= RTCF_LOCAL; 1712 1713 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST, 1714 IN_DEV_ORCONF(in_dev, NOPOLICY), false); 1715 if (!rth) 1716 return -ENOBUFS; 1717 1718 #ifdef CONFIG_IP_ROUTE_CLASSID 1719 rth->dst.tclassid = itag; 1720 #endif 1721 rth->dst.output = ip_rt_bug; 1722 rth->rt_is_input= 1; 1723 1724 #ifdef CONFIG_IP_MROUTE 1725 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1726 rth->dst.input = ip_mr_input; 1727 #endif 1728 RT_CACHE_STAT_INC(in_slow_mc); 1729 1730 skb_dst_set(skb, &rth->dst); 1731 return 0; 1732 } 1733 1734 1735 static void ip_handle_martian_source(struct net_device *dev, 1736 struct in_device *in_dev, 1737 struct sk_buff *skb, 1738 __be32 daddr, 1739 __be32 saddr) 1740 { 1741 RT_CACHE_STAT_INC(in_martian_src); 1742 #ifdef CONFIG_IP_ROUTE_VERBOSE 1743 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1744 /* 1745 * RFC1812 recommendation, if source is martian, 1746 * the only hint is MAC header. 1747 */ 1748 pr_warn("martian source %pI4 from %pI4, on dev %s\n", 1749 &daddr, &saddr, dev->name); 1750 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1751 print_hex_dump(KERN_WARNING, "ll header: ", 1752 DUMP_PREFIX_OFFSET, 16, 1, 1753 skb_mac_header(skb), 1754 dev->hard_header_len, false); 1755 } 1756 } 1757 #endif 1758 } 1759 1760 /* called in rcu_read_lock() section */ 1761 static int __mkroute_input(struct sk_buff *skb, 1762 const struct fib_result *res, 1763 struct in_device *in_dev, 1764 __be32 daddr, __be32 saddr, u32 tos) 1765 { 1766 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1767 struct net_device *dev = nhc->nhc_dev; 1768 struct fib_nh_exception *fnhe; 1769 struct rtable *rth; 1770 int err; 1771 struct in_device *out_dev; 1772 bool do_cache; 1773 u32 itag = 0; 1774 1775 /* get a working reference to the output device */ 1776 out_dev = __in_dev_get_rcu(dev); 1777 if (!out_dev) { 1778 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n"); 1779 return -EINVAL; 1780 } 1781 1782 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res), 1783 in_dev->dev, in_dev, &itag); 1784 if (err < 0) { 1785 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1786 saddr); 1787 1788 goto cleanup; 1789 } 1790 1791 do_cache = res->fi && !itag; 1792 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) && 1793 skb->protocol == htons(ETH_P_IP)) { 1794 __be32 gw; 1795 1796 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0; 1797 if (IN_DEV_SHARED_MEDIA(out_dev) || 1798 inet_addr_onlink(out_dev, saddr, gw)) 1799 IPCB(skb)->flags |= IPSKB_DOREDIRECT; 1800 } 1801 1802 if (skb->protocol != htons(ETH_P_IP)) { 1803 /* Not IP (i.e. ARP). Do not create route, if it is 1804 * invalid for proxy arp. DNAT routes are always valid. 1805 * 1806 * Proxy arp feature have been extended to allow, ARP 1807 * replies back to the same interface, to support 1808 * Private VLAN switch technologies. See arp.c. 1809 */ 1810 if (out_dev == in_dev && 1811 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 1812 err = -EINVAL; 1813 goto cleanup; 1814 } 1815 } 1816 1817 fnhe = find_exception(nhc, daddr); 1818 if (do_cache) { 1819 if (fnhe) 1820 rth = rcu_dereference(fnhe->fnhe_rth_input); 1821 else 1822 rth = rcu_dereference(nhc->nhc_rth_input); 1823 if (rt_cache_valid(rth)) { 1824 skb_dst_set_noref(skb, &rth->dst); 1825 goto out; 1826 } 1827 } 1828 1829 rth = rt_dst_alloc(out_dev->dev, 0, res->type, 1830 IN_DEV_ORCONF(in_dev, NOPOLICY), 1831 IN_DEV_ORCONF(out_dev, NOXFRM)); 1832 if (!rth) { 1833 err = -ENOBUFS; 1834 goto cleanup; 1835 } 1836 1837 rth->rt_is_input = 1; 1838 RT_CACHE_STAT_INC(in_slow_tot); 1839 1840 rth->dst.input = ip_forward; 1841 1842 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag, 1843 do_cache); 1844 lwtunnel_set_redirect(&rth->dst); 1845 skb_dst_set(skb, &rth->dst); 1846 out: 1847 err = 0; 1848 cleanup: 1849 return err; 1850 } 1851 1852 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1853 /* To make ICMP packets follow the right flow, the multipath hash is 1854 * calculated from the inner IP addresses. 1855 */ 1856 static void ip_multipath_l3_keys(const struct sk_buff *skb, 1857 struct flow_keys *hash_keys) 1858 { 1859 const struct iphdr *outer_iph = ip_hdr(skb); 1860 const struct iphdr *key_iph = outer_iph; 1861 const struct iphdr *inner_iph; 1862 const struct icmphdr *icmph; 1863 struct iphdr _inner_iph; 1864 struct icmphdr _icmph; 1865 1866 if (likely(outer_iph->protocol != IPPROTO_ICMP)) 1867 goto out; 1868 1869 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0)) 1870 goto out; 1871 1872 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph), 1873 &_icmph); 1874 if (!icmph) 1875 goto out; 1876 1877 if (!icmp_is_err(icmph->type)) 1878 goto out; 1879 1880 inner_iph = skb_header_pointer(skb, 1881 outer_iph->ihl * 4 + sizeof(_icmph), 1882 sizeof(_inner_iph), &_inner_iph); 1883 if (!inner_iph) 1884 goto out; 1885 1886 key_iph = inner_iph; 1887 out: 1888 hash_keys->addrs.v4addrs.src = key_iph->saddr; 1889 hash_keys->addrs.v4addrs.dst = key_iph->daddr; 1890 } 1891 1892 static u32 fib_multipath_custom_hash_outer(const struct net *net, 1893 const struct sk_buff *skb, 1894 bool *p_has_inner) 1895 { 1896 u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields; 1897 struct flow_keys keys, hash_keys; 1898 1899 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 1900 return 0; 1901 1902 memset(&hash_keys, 0, sizeof(hash_keys)); 1903 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP); 1904 1905 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1906 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 1907 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 1908 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 1909 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 1910 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 1911 hash_keys.basic.ip_proto = keys.basic.ip_proto; 1912 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) 1913 hash_keys.ports.src = keys.ports.src; 1914 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 1915 hash_keys.ports.dst = keys.ports.dst; 1916 1917 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION); 1918 return flow_hash_from_keys(&hash_keys); 1919 } 1920 1921 static u32 fib_multipath_custom_hash_inner(const struct net *net, 1922 const struct sk_buff *skb, 1923 bool has_inner) 1924 { 1925 u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields; 1926 struct flow_keys keys, hash_keys; 1927 1928 /* We assume the packet carries an encapsulation, but if none was 1929 * encountered during dissection of the outer flow, then there is no 1930 * point in calling the flow dissector again. 1931 */ 1932 if (!has_inner) 1933 return 0; 1934 1935 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK)) 1936 return 0; 1937 1938 memset(&hash_keys, 0, sizeof(hash_keys)); 1939 skb_flow_dissect_flow_keys(skb, &keys, 0); 1940 1941 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION)) 1942 return 0; 1943 1944 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 1945 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1946 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 1947 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 1948 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 1949 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 1950 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 1951 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 1952 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 1953 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 1954 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 1955 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 1956 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL) 1957 hash_keys.tags.flow_label = keys.tags.flow_label; 1958 } 1959 1960 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO) 1961 hash_keys.basic.ip_proto = keys.basic.ip_proto; 1962 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT) 1963 hash_keys.ports.src = keys.ports.src; 1964 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT) 1965 hash_keys.ports.dst = keys.ports.dst; 1966 1967 return flow_hash_from_keys(&hash_keys); 1968 } 1969 1970 static u32 fib_multipath_custom_hash_skb(const struct net *net, 1971 const struct sk_buff *skb) 1972 { 1973 u32 mhash, mhash_inner; 1974 bool has_inner = true; 1975 1976 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner); 1977 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner); 1978 1979 return jhash_2words(mhash, mhash_inner, 0); 1980 } 1981 1982 static u32 fib_multipath_custom_hash_fl4(const struct net *net, 1983 const struct flowi4 *fl4) 1984 { 1985 u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields; 1986 struct flow_keys hash_keys; 1987 1988 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 1989 return 0; 1990 1991 memset(&hash_keys, 0, sizeof(hash_keys)); 1992 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1993 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 1994 hash_keys.addrs.v4addrs.src = fl4->saddr; 1995 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 1996 hash_keys.addrs.v4addrs.dst = fl4->daddr; 1997 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 1998 hash_keys.basic.ip_proto = fl4->flowi4_proto; 1999 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) 2000 hash_keys.ports.src = fl4->fl4_sport; 2001 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 2002 hash_keys.ports.dst = fl4->fl4_dport; 2003 2004 return flow_hash_from_keys(&hash_keys); 2005 } 2006 2007 /* if skb is set it will be used and fl4 can be NULL */ 2008 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4, 2009 const struct sk_buff *skb, struct flow_keys *flkeys) 2010 { 2011 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0; 2012 struct flow_keys hash_keys; 2013 u32 mhash = 0; 2014 2015 switch (net->ipv4.sysctl_fib_multipath_hash_policy) { 2016 case 0: 2017 memset(&hash_keys, 0, sizeof(hash_keys)); 2018 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2019 if (skb) { 2020 ip_multipath_l3_keys(skb, &hash_keys); 2021 } else { 2022 hash_keys.addrs.v4addrs.src = fl4->saddr; 2023 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2024 } 2025 mhash = flow_hash_from_keys(&hash_keys); 2026 break; 2027 case 1: 2028 /* skb is currently provided only when forwarding */ 2029 if (skb) { 2030 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 2031 struct flow_keys keys; 2032 2033 /* short-circuit if we already have L4 hash present */ 2034 if (skb->l4_hash) 2035 return skb_get_hash_raw(skb) >> 1; 2036 2037 memset(&hash_keys, 0, sizeof(hash_keys)); 2038 2039 if (!flkeys) { 2040 skb_flow_dissect_flow_keys(skb, &keys, flag); 2041 flkeys = &keys; 2042 } 2043 2044 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2045 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src; 2046 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst; 2047 hash_keys.ports.src = flkeys->ports.src; 2048 hash_keys.ports.dst = flkeys->ports.dst; 2049 hash_keys.basic.ip_proto = flkeys->basic.ip_proto; 2050 } else { 2051 memset(&hash_keys, 0, sizeof(hash_keys)); 2052 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2053 hash_keys.addrs.v4addrs.src = fl4->saddr; 2054 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2055 hash_keys.ports.src = fl4->fl4_sport; 2056 hash_keys.ports.dst = fl4->fl4_dport; 2057 hash_keys.basic.ip_proto = fl4->flowi4_proto; 2058 } 2059 mhash = flow_hash_from_keys(&hash_keys); 2060 break; 2061 case 2: 2062 memset(&hash_keys, 0, sizeof(hash_keys)); 2063 /* skb is currently provided only when forwarding */ 2064 if (skb) { 2065 struct flow_keys keys; 2066 2067 skb_flow_dissect_flow_keys(skb, &keys, 0); 2068 /* Inner can be v4 or v6 */ 2069 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 2070 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2071 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 2072 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 2073 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 2074 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 2075 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 2076 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 2077 hash_keys.tags.flow_label = keys.tags.flow_label; 2078 hash_keys.basic.ip_proto = keys.basic.ip_proto; 2079 } else { 2080 /* Same as case 0 */ 2081 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2082 ip_multipath_l3_keys(skb, &hash_keys); 2083 } 2084 } else { 2085 /* Same as case 0 */ 2086 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2087 hash_keys.addrs.v4addrs.src = fl4->saddr; 2088 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2089 } 2090 mhash = flow_hash_from_keys(&hash_keys); 2091 break; 2092 case 3: 2093 if (skb) 2094 mhash = fib_multipath_custom_hash_skb(net, skb); 2095 else 2096 mhash = fib_multipath_custom_hash_fl4(net, fl4); 2097 break; 2098 } 2099 2100 if (multipath_hash) 2101 mhash = jhash_2words(mhash, multipath_hash, 0); 2102 2103 return mhash >> 1; 2104 } 2105 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 2106 2107 static int ip_mkroute_input(struct sk_buff *skb, 2108 struct fib_result *res, 2109 struct in_device *in_dev, 2110 __be32 daddr, __be32 saddr, u32 tos, 2111 struct flow_keys *hkeys) 2112 { 2113 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2114 if (res->fi && fib_info_num_path(res->fi) > 1) { 2115 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys); 2116 2117 fib_select_multipath(res, h); 2118 } 2119 #endif 2120 2121 /* create a routing cache entry */ 2122 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos); 2123 } 2124 2125 /* Implements all the saddr-related checks as ip_route_input_slow(), 2126 * assuming daddr is valid and the destination is not a local broadcast one. 2127 * Uses the provided hint instead of performing a route lookup. 2128 */ 2129 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2130 u8 tos, struct net_device *dev, 2131 const struct sk_buff *hint) 2132 { 2133 struct in_device *in_dev = __in_dev_get_rcu(dev); 2134 struct rtable *rt = skb_rtable(hint); 2135 struct net *net = dev_net(dev); 2136 int err = -EINVAL; 2137 u32 tag = 0; 2138 2139 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 2140 goto martian_source; 2141 2142 if (ipv4_is_zeronet(saddr)) 2143 goto martian_source; 2144 2145 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2146 goto martian_source; 2147 2148 if (rt->rt_type != RTN_LOCAL) 2149 goto skip_validate_source; 2150 2151 tos &= IPTOS_RT_MASK; 2152 err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag); 2153 if (err < 0) 2154 goto martian_source; 2155 2156 skip_validate_source: 2157 skb_dst_copy(skb, hint); 2158 return 0; 2159 2160 martian_source: 2161 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2162 return err; 2163 } 2164 2165 /* get device for dst_alloc with local routes */ 2166 static struct net_device *ip_rt_get_dev(struct net *net, 2167 const struct fib_result *res) 2168 { 2169 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL; 2170 struct net_device *dev = NULL; 2171 2172 if (nhc) 2173 dev = l3mdev_master_dev_rcu(nhc->nhc_dev); 2174 2175 return dev ? : net->loopback_dev; 2176 } 2177 2178 /* 2179 * NOTE. We drop all the packets that has local source 2180 * addresses, because every properly looped back packet 2181 * must have correct destination already attached by output routine. 2182 * Changes in the enforced policies must be applied also to 2183 * ip_route_use_hint(). 2184 * 2185 * Such approach solves two big problems: 2186 * 1. Not simplex devices are handled properly. 2187 * 2. IP spoofing attempts are filtered with 100% of guarantee. 2188 * called with rcu_read_lock() 2189 */ 2190 2191 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2192 u8 tos, struct net_device *dev, 2193 struct fib_result *res) 2194 { 2195 struct in_device *in_dev = __in_dev_get_rcu(dev); 2196 struct flow_keys *flkeys = NULL, _flkeys; 2197 struct net *net = dev_net(dev); 2198 struct ip_tunnel_info *tun_info; 2199 int err = -EINVAL; 2200 unsigned int flags = 0; 2201 u32 itag = 0; 2202 struct rtable *rth; 2203 struct flowi4 fl4; 2204 bool do_cache = true; 2205 2206 /* IP on this device is disabled. */ 2207 2208 if (!in_dev) 2209 goto out; 2210 2211 /* Check for the most weird martians, which can be not detected 2212 * by fib_lookup. 2213 */ 2214 2215 tun_info = skb_tunnel_info(skb); 2216 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX)) 2217 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id; 2218 else 2219 fl4.flowi4_tun_key.tun_id = 0; 2220 skb_dst_drop(skb); 2221 2222 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 2223 goto martian_source; 2224 2225 res->fi = NULL; 2226 res->table = NULL; 2227 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 2228 goto brd_input; 2229 2230 /* Accept zero addresses only to limited broadcast; 2231 * I even do not know to fix it or not. Waiting for complains :-) 2232 */ 2233 if (ipv4_is_zeronet(saddr)) 2234 goto martian_source; 2235 2236 if (ipv4_is_zeronet(daddr)) 2237 goto martian_destination; 2238 2239 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(), 2240 * and call it once if daddr or/and saddr are loopback addresses 2241 */ 2242 if (ipv4_is_loopback(daddr)) { 2243 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2244 goto martian_destination; 2245 } else if (ipv4_is_loopback(saddr)) { 2246 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2247 goto martian_source; 2248 } 2249 2250 /* 2251 * Now we are ready to route packet. 2252 */ 2253 fl4.flowi4_oif = 0; 2254 fl4.flowi4_iif = dev->ifindex; 2255 fl4.flowi4_mark = skb->mark; 2256 fl4.flowi4_tos = tos; 2257 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 2258 fl4.flowi4_flags = 0; 2259 fl4.daddr = daddr; 2260 fl4.saddr = saddr; 2261 fl4.flowi4_uid = sock_net_uid(net, NULL); 2262 fl4.flowi4_multipath_hash = 0; 2263 2264 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) { 2265 flkeys = &_flkeys; 2266 } else { 2267 fl4.flowi4_proto = 0; 2268 fl4.fl4_sport = 0; 2269 fl4.fl4_dport = 0; 2270 } 2271 2272 err = fib_lookup(net, &fl4, res, 0); 2273 if (err != 0) { 2274 if (!IN_DEV_FORWARD(in_dev)) 2275 err = -EHOSTUNREACH; 2276 goto no_route; 2277 } 2278 2279 if (res->type == RTN_BROADCAST) { 2280 if (IN_DEV_BFORWARD(in_dev)) 2281 goto make_route; 2282 /* not do cache if bc_forwarding is enabled */ 2283 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING)) 2284 do_cache = false; 2285 goto brd_input; 2286 } 2287 2288 if (res->type == RTN_LOCAL) { 2289 err = fib_validate_source(skb, saddr, daddr, tos, 2290 0, dev, in_dev, &itag); 2291 if (err < 0) 2292 goto martian_source; 2293 goto local_input; 2294 } 2295 2296 if (!IN_DEV_FORWARD(in_dev)) { 2297 err = -EHOSTUNREACH; 2298 goto no_route; 2299 } 2300 if (res->type != RTN_UNICAST) 2301 goto martian_destination; 2302 2303 make_route: 2304 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys); 2305 out: return err; 2306 2307 brd_input: 2308 if (skb->protocol != htons(ETH_P_IP)) 2309 goto e_inval; 2310 2311 if (!ipv4_is_zeronet(saddr)) { 2312 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 2313 in_dev, &itag); 2314 if (err < 0) 2315 goto martian_source; 2316 } 2317 flags |= RTCF_BROADCAST; 2318 res->type = RTN_BROADCAST; 2319 RT_CACHE_STAT_INC(in_brd); 2320 2321 local_input: 2322 do_cache &= res->fi && !itag; 2323 if (do_cache) { 2324 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2325 2326 rth = rcu_dereference(nhc->nhc_rth_input); 2327 if (rt_cache_valid(rth)) { 2328 skb_dst_set_noref(skb, &rth->dst); 2329 err = 0; 2330 goto out; 2331 } 2332 } 2333 2334 rth = rt_dst_alloc(ip_rt_get_dev(net, res), 2335 flags | RTCF_LOCAL, res->type, 2336 IN_DEV_ORCONF(in_dev, NOPOLICY), false); 2337 if (!rth) 2338 goto e_nobufs; 2339 2340 rth->dst.output= ip_rt_bug; 2341 #ifdef CONFIG_IP_ROUTE_CLASSID 2342 rth->dst.tclassid = itag; 2343 #endif 2344 rth->rt_is_input = 1; 2345 2346 RT_CACHE_STAT_INC(in_slow_tot); 2347 if (res->type == RTN_UNREACHABLE) { 2348 rth->dst.input= ip_error; 2349 rth->dst.error= -err; 2350 rth->rt_flags &= ~RTCF_LOCAL; 2351 } 2352 2353 if (do_cache) { 2354 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2355 2356 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 2357 if (lwtunnel_input_redirect(rth->dst.lwtstate)) { 2358 WARN_ON(rth->dst.input == lwtunnel_input); 2359 rth->dst.lwtstate->orig_input = rth->dst.input; 2360 rth->dst.input = lwtunnel_input; 2361 } 2362 2363 if (unlikely(!rt_cache_route(nhc, rth))) 2364 rt_add_uncached_list(rth); 2365 } 2366 skb_dst_set(skb, &rth->dst); 2367 err = 0; 2368 goto out; 2369 2370 no_route: 2371 RT_CACHE_STAT_INC(in_no_route); 2372 res->type = RTN_UNREACHABLE; 2373 res->fi = NULL; 2374 res->table = NULL; 2375 goto local_input; 2376 2377 /* 2378 * Do not cache martian addresses: they should be logged (RFC1812) 2379 */ 2380 martian_destination: 2381 RT_CACHE_STAT_INC(in_martian_dst); 2382 #ifdef CONFIG_IP_ROUTE_VERBOSE 2383 if (IN_DEV_LOG_MARTIANS(in_dev)) 2384 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n", 2385 &daddr, &saddr, dev->name); 2386 #endif 2387 2388 e_inval: 2389 err = -EINVAL; 2390 goto out; 2391 2392 e_nobufs: 2393 err = -ENOBUFS; 2394 goto out; 2395 2396 martian_source: 2397 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2398 goto out; 2399 } 2400 2401 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2402 u8 tos, struct net_device *dev) 2403 { 2404 struct fib_result res; 2405 int err; 2406 2407 tos &= IPTOS_RT_MASK; 2408 rcu_read_lock(); 2409 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res); 2410 rcu_read_unlock(); 2411 2412 return err; 2413 } 2414 EXPORT_SYMBOL(ip_route_input_noref); 2415 2416 /* called with rcu_read_lock held */ 2417 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2418 u8 tos, struct net_device *dev, struct fib_result *res) 2419 { 2420 /* Multicast recognition logic is moved from route cache to here. 2421 * The problem was that too many Ethernet cards have broken/missing 2422 * hardware multicast filters :-( As result the host on multicasting 2423 * network acquires a lot of useless route cache entries, sort of 2424 * SDR messages from all the world. Now we try to get rid of them. 2425 * Really, provided software IP multicast filter is organized 2426 * reasonably (at least, hashed), it does not result in a slowdown 2427 * comparing with route cache reject entries. 2428 * Note, that multicast routers are not affected, because 2429 * route cache entry is created eventually. 2430 */ 2431 if (ipv4_is_multicast(daddr)) { 2432 struct in_device *in_dev = __in_dev_get_rcu(dev); 2433 int our = 0; 2434 int err = -EINVAL; 2435 2436 if (!in_dev) 2437 return err; 2438 our = ip_check_mc_rcu(in_dev, daddr, saddr, 2439 ip_hdr(skb)->protocol); 2440 2441 /* check l3 master if no match yet */ 2442 if (!our && netif_is_l3_slave(dev)) { 2443 struct in_device *l3_in_dev; 2444 2445 l3_in_dev = __in_dev_get_rcu(skb->dev); 2446 if (l3_in_dev) 2447 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr, 2448 ip_hdr(skb)->protocol); 2449 } 2450 2451 if (our 2452 #ifdef CONFIG_IP_MROUTE 2453 || 2454 (!ipv4_is_local_multicast(daddr) && 2455 IN_DEV_MFORWARD(in_dev)) 2456 #endif 2457 ) { 2458 err = ip_route_input_mc(skb, daddr, saddr, 2459 tos, dev, our); 2460 } 2461 return err; 2462 } 2463 2464 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res); 2465 } 2466 2467 /* called with rcu_read_lock() */ 2468 static struct rtable *__mkroute_output(const struct fib_result *res, 2469 const struct flowi4 *fl4, int orig_oif, 2470 struct net_device *dev_out, 2471 unsigned int flags) 2472 { 2473 struct fib_info *fi = res->fi; 2474 struct fib_nh_exception *fnhe; 2475 struct in_device *in_dev; 2476 u16 type = res->type; 2477 struct rtable *rth; 2478 bool do_cache; 2479 2480 in_dev = __in_dev_get_rcu(dev_out); 2481 if (!in_dev) 2482 return ERR_PTR(-EINVAL); 2483 2484 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 2485 if (ipv4_is_loopback(fl4->saddr) && 2486 !(dev_out->flags & IFF_LOOPBACK) && 2487 !netif_is_l3_master(dev_out)) 2488 return ERR_PTR(-EINVAL); 2489 2490 if (ipv4_is_lbcast(fl4->daddr)) 2491 type = RTN_BROADCAST; 2492 else if (ipv4_is_multicast(fl4->daddr)) 2493 type = RTN_MULTICAST; 2494 else if (ipv4_is_zeronet(fl4->daddr)) 2495 return ERR_PTR(-EINVAL); 2496 2497 if (dev_out->flags & IFF_LOOPBACK) 2498 flags |= RTCF_LOCAL; 2499 2500 do_cache = true; 2501 if (type == RTN_BROADCAST) { 2502 flags |= RTCF_BROADCAST | RTCF_LOCAL; 2503 fi = NULL; 2504 } else if (type == RTN_MULTICAST) { 2505 flags |= RTCF_MULTICAST | RTCF_LOCAL; 2506 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr, 2507 fl4->flowi4_proto)) 2508 flags &= ~RTCF_LOCAL; 2509 else 2510 do_cache = false; 2511 /* If multicast route do not exist use 2512 * default one, but do not gateway in this case. 2513 * Yes, it is hack. 2514 */ 2515 if (fi && res->prefixlen < 4) 2516 fi = NULL; 2517 } else if ((type == RTN_LOCAL) && (orig_oif != 0) && 2518 (orig_oif != dev_out->ifindex)) { 2519 /* For local routes that require a particular output interface 2520 * we do not want to cache the result. Caching the result 2521 * causes incorrect behaviour when there are multiple source 2522 * addresses on the interface, the end result being that if the 2523 * intended recipient is waiting on that interface for the 2524 * packet he won't receive it because it will be delivered on 2525 * the loopback interface and the IP_PKTINFO ipi_ifindex will 2526 * be set to the loopback interface as well. 2527 */ 2528 do_cache = false; 2529 } 2530 2531 fnhe = NULL; 2532 do_cache &= fi != NULL; 2533 if (fi) { 2534 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2535 struct rtable __rcu **prth; 2536 2537 fnhe = find_exception(nhc, fl4->daddr); 2538 if (!do_cache) 2539 goto add; 2540 if (fnhe) { 2541 prth = &fnhe->fnhe_rth_output; 2542 } else { 2543 if (unlikely(fl4->flowi4_flags & 2544 FLOWI_FLAG_KNOWN_NH && 2545 !(nhc->nhc_gw_family && 2546 nhc->nhc_scope == RT_SCOPE_LINK))) { 2547 do_cache = false; 2548 goto add; 2549 } 2550 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 2551 } 2552 rth = rcu_dereference(*prth); 2553 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst)) 2554 return rth; 2555 } 2556 2557 add: 2558 rth = rt_dst_alloc(dev_out, flags, type, 2559 IN_DEV_ORCONF(in_dev, NOPOLICY), 2560 IN_DEV_ORCONF(in_dev, NOXFRM)); 2561 if (!rth) 2562 return ERR_PTR(-ENOBUFS); 2563 2564 rth->rt_iif = orig_oif; 2565 2566 RT_CACHE_STAT_INC(out_slow_tot); 2567 2568 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 2569 if (flags & RTCF_LOCAL && 2570 !(dev_out->flags & IFF_LOOPBACK)) { 2571 rth->dst.output = ip_mc_output; 2572 RT_CACHE_STAT_INC(out_slow_mc); 2573 } 2574 #ifdef CONFIG_IP_MROUTE 2575 if (type == RTN_MULTICAST) { 2576 if (IN_DEV_MFORWARD(in_dev) && 2577 !ipv4_is_local_multicast(fl4->daddr)) { 2578 rth->dst.input = ip_mr_input; 2579 rth->dst.output = ip_mc_output; 2580 } 2581 } 2582 #endif 2583 } 2584 2585 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache); 2586 lwtunnel_set_redirect(&rth->dst); 2587 2588 return rth; 2589 } 2590 2591 /* 2592 * Major route resolver routine. 2593 */ 2594 2595 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4, 2596 const struct sk_buff *skb) 2597 { 2598 __u8 tos = RT_FL_TOS(fl4); 2599 struct fib_result res = { 2600 .type = RTN_UNSPEC, 2601 .fi = NULL, 2602 .table = NULL, 2603 .tclassid = 0, 2604 }; 2605 struct rtable *rth; 2606 2607 fl4->flowi4_iif = LOOPBACK_IFINDEX; 2608 fl4->flowi4_tos = tos & IPTOS_RT_MASK; 2609 fl4->flowi4_scope = ((tos & RTO_ONLINK) ? 2610 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE); 2611 2612 rcu_read_lock(); 2613 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb); 2614 rcu_read_unlock(); 2615 2616 return rth; 2617 } 2618 EXPORT_SYMBOL_GPL(ip_route_output_key_hash); 2619 2620 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4, 2621 struct fib_result *res, 2622 const struct sk_buff *skb) 2623 { 2624 struct net_device *dev_out = NULL; 2625 int orig_oif = fl4->flowi4_oif; 2626 unsigned int flags = 0; 2627 struct rtable *rth; 2628 int err; 2629 2630 if (fl4->saddr) { 2631 if (ipv4_is_multicast(fl4->saddr) || 2632 ipv4_is_lbcast(fl4->saddr) || 2633 ipv4_is_zeronet(fl4->saddr)) { 2634 rth = ERR_PTR(-EINVAL); 2635 goto out; 2636 } 2637 2638 rth = ERR_PTR(-ENETUNREACH); 2639 2640 /* I removed check for oif == dev_out->oif here. 2641 * It was wrong for two reasons: 2642 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 2643 * is assigned to multiple interfaces. 2644 * 2. Moreover, we are allowed to send packets with saddr 2645 * of another iface. --ANK 2646 */ 2647 2648 if (fl4->flowi4_oif == 0 && 2649 (ipv4_is_multicast(fl4->daddr) || 2650 ipv4_is_lbcast(fl4->daddr))) { 2651 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2652 dev_out = __ip_dev_find(net, fl4->saddr, false); 2653 if (!dev_out) 2654 goto out; 2655 2656 /* Special hack: user can direct multicasts 2657 * and limited broadcast via necessary interface 2658 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 2659 * This hack is not just for fun, it allows 2660 * vic,vat and friends to work. 2661 * They bind socket to loopback, set ttl to zero 2662 * and expect that it will work. 2663 * From the viewpoint of routing cache they are broken, 2664 * because we are not allowed to build multicast path 2665 * with loopback source addr (look, routing cache 2666 * cannot know, that ttl is zero, so that packet 2667 * will not leave this host and route is valid). 2668 * Luckily, this hack is good workaround. 2669 */ 2670 2671 fl4->flowi4_oif = dev_out->ifindex; 2672 goto make_route; 2673 } 2674 2675 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 2676 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2677 if (!__ip_dev_find(net, fl4->saddr, false)) 2678 goto out; 2679 } 2680 } 2681 2682 2683 if (fl4->flowi4_oif) { 2684 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif); 2685 rth = ERR_PTR(-ENODEV); 2686 if (!dev_out) 2687 goto out; 2688 2689 /* RACE: Check return value of inet_select_addr instead. */ 2690 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 2691 rth = ERR_PTR(-ENETUNREACH); 2692 goto out; 2693 } 2694 if (ipv4_is_local_multicast(fl4->daddr) || 2695 ipv4_is_lbcast(fl4->daddr) || 2696 fl4->flowi4_proto == IPPROTO_IGMP) { 2697 if (!fl4->saddr) 2698 fl4->saddr = inet_select_addr(dev_out, 0, 2699 RT_SCOPE_LINK); 2700 goto make_route; 2701 } 2702 if (!fl4->saddr) { 2703 if (ipv4_is_multicast(fl4->daddr)) 2704 fl4->saddr = inet_select_addr(dev_out, 0, 2705 fl4->flowi4_scope); 2706 else if (!fl4->daddr) 2707 fl4->saddr = inet_select_addr(dev_out, 0, 2708 RT_SCOPE_HOST); 2709 } 2710 } 2711 2712 if (!fl4->daddr) { 2713 fl4->daddr = fl4->saddr; 2714 if (!fl4->daddr) 2715 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK); 2716 dev_out = net->loopback_dev; 2717 fl4->flowi4_oif = LOOPBACK_IFINDEX; 2718 res->type = RTN_LOCAL; 2719 flags |= RTCF_LOCAL; 2720 goto make_route; 2721 } 2722 2723 err = fib_lookup(net, fl4, res, 0); 2724 if (err) { 2725 res->fi = NULL; 2726 res->table = NULL; 2727 if (fl4->flowi4_oif && 2728 (ipv4_is_multicast(fl4->daddr) || 2729 !netif_index_is_l3_master(net, fl4->flowi4_oif))) { 2730 /* Apparently, routing tables are wrong. Assume, 2731 * that the destination is on link. 2732 * 2733 * WHY? DW. 2734 * Because we are allowed to send to iface 2735 * even if it has NO routes and NO assigned 2736 * addresses. When oif is specified, routing 2737 * tables are looked up with only one purpose: 2738 * to catch if destination is gatewayed, rather than 2739 * direct. Moreover, if MSG_DONTROUTE is set, 2740 * we send packet, ignoring both routing tables 2741 * and ifaddr state. --ANK 2742 * 2743 * 2744 * We could make it even if oif is unknown, 2745 * likely IPv6, but we do not. 2746 */ 2747 2748 if (fl4->saddr == 0) 2749 fl4->saddr = inet_select_addr(dev_out, 0, 2750 RT_SCOPE_LINK); 2751 res->type = RTN_UNICAST; 2752 goto make_route; 2753 } 2754 rth = ERR_PTR(err); 2755 goto out; 2756 } 2757 2758 if (res->type == RTN_LOCAL) { 2759 if (!fl4->saddr) { 2760 if (res->fi->fib_prefsrc) 2761 fl4->saddr = res->fi->fib_prefsrc; 2762 else 2763 fl4->saddr = fl4->daddr; 2764 } 2765 2766 /* L3 master device is the loopback for that domain */ 2767 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? : 2768 net->loopback_dev; 2769 2770 /* make sure orig_oif points to fib result device even 2771 * though packet rx/tx happens over loopback or l3mdev 2772 */ 2773 orig_oif = FIB_RES_OIF(*res); 2774 2775 fl4->flowi4_oif = dev_out->ifindex; 2776 flags |= RTCF_LOCAL; 2777 goto make_route; 2778 } 2779 2780 fib_select_path(net, res, fl4, skb); 2781 2782 dev_out = FIB_RES_DEV(*res); 2783 2784 make_route: 2785 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags); 2786 2787 out: 2788 return rth; 2789 } 2790 2791 static struct dst_ops ipv4_dst_blackhole_ops = { 2792 .family = AF_INET, 2793 .default_advmss = ipv4_default_advmss, 2794 .neigh_lookup = ipv4_neigh_lookup, 2795 .check = dst_blackhole_check, 2796 .cow_metrics = dst_blackhole_cow_metrics, 2797 .update_pmtu = dst_blackhole_update_pmtu, 2798 .redirect = dst_blackhole_redirect, 2799 .mtu = dst_blackhole_mtu, 2800 }; 2801 2802 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2803 { 2804 struct rtable *ort = (struct rtable *) dst_orig; 2805 struct rtable *rt; 2806 2807 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0); 2808 if (rt) { 2809 struct dst_entry *new = &rt->dst; 2810 2811 new->__use = 1; 2812 new->input = dst_discard; 2813 new->output = dst_discard_out; 2814 2815 new->dev = net->loopback_dev; 2816 dev_hold(new->dev); 2817 2818 rt->rt_is_input = ort->rt_is_input; 2819 rt->rt_iif = ort->rt_iif; 2820 rt->rt_pmtu = ort->rt_pmtu; 2821 rt->rt_mtu_locked = ort->rt_mtu_locked; 2822 2823 rt->rt_genid = rt_genid_ipv4(net); 2824 rt->rt_flags = ort->rt_flags; 2825 rt->rt_type = ort->rt_type; 2826 rt->rt_uses_gateway = ort->rt_uses_gateway; 2827 rt->rt_gw_family = ort->rt_gw_family; 2828 if (rt->rt_gw_family == AF_INET) 2829 rt->rt_gw4 = ort->rt_gw4; 2830 else if (rt->rt_gw_family == AF_INET6) 2831 rt->rt_gw6 = ort->rt_gw6; 2832 2833 INIT_LIST_HEAD(&rt->rt_uncached); 2834 } 2835 2836 dst_release(dst_orig); 2837 2838 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2839 } 2840 2841 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2842 const struct sock *sk) 2843 { 2844 struct rtable *rt = __ip_route_output_key(net, flp4); 2845 2846 if (IS_ERR(rt)) 2847 return rt; 2848 2849 if (flp4->flowi4_proto) { 2850 flp4->flowi4_oif = rt->dst.dev->ifindex; 2851 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst, 2852 flowi4_to_flowi(flp4), 2853 sk, 0); 2854 } 2855 2856 return rt; 2857 } 2858 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2859 2860 struct rtable *ip_route_output_tunnel(struct sk_buff *skb, 2861 struct net_device *dev, 2862 struct net *net, __be32 *saddr, 2863 const struct ip_tunnel_info *info, 2864 u8 protocol, bool use_cache) 2865 { 2866 #ifdef CONFIG_DST_CACHE 2867 struct dst_cache *dst_cache; 2868 #endif 2869 struct rtable *rt = NULL; 2870 struct flowi4 fl4; 2871 __u8 tos; 2872 2873 #ifdef CONFIG_DST_CACHE 2874 dst_cache = (struct dst_cache *)&info->dst_cache; 2875 if (use_cache) { 2876 rt = dst_cache_get_ip4(dst_cache, saddr); 2877 if (rt) 2878 return rt; 2879 } 2880 #endif 2881 memset(&fl4, 0, sizeof(fl4)); 2882 fl4.flowi4_mark = skb->mark; 2883 fl4.flowi4_proto = protocol; 2884 fl4.daddr = info->key.u.ipv4.dst; 2885 fl4.saddr = info->key.u.ipv4.src; 2886 tos = info->key.tos; 2887 fl4.flowi4_tos = RT_TOS(tos); 2888 2889 rt = ip_route_output_key(net, &fl4); 2890 if (IS_ERR(rt)) { 2891 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr); 2892 return ERR_PTR(-ENETUNREACH); 2893 } 2894 if (rt->dst.dev == dev) { /* is this necessary? */ 2895 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr); 2896 ip_rt_put(rt); 2897 return ERR_PTR(-ELOOP); 2898 } 2899 #ifdef CONFIG_DST_CACHE 2900 if (use_cache) 2901 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr); 2902 #endif 2903 *saddr = fl4.saddr; 2904 return rt; 2905 } 2906 EXPORT_SYMBOL_GPL(ip_route_output_tunnel); 2907 2908 /* called with rcu_read_lock held */ 2909 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, 2910 struct rtable *rt, u32 table_id, struct flowi4 *fl4, 2911 struct sk_buff *skb, u32 portid, u32 seq, 2912 unsigned int flags) 2913 { 2914 struct rtmsg *r; 2915 struct nlmsghdr *nlh; 2916 unsigned long expires = 0; 2917 u32 error; 2918 u32 metrics[RTAX_MAX]; 2919 2920 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags); 2921 if (!nlh) 2922 return -EMSGSIZE; 2923 2924 r = nlmsg_data(nlh); 2925 r->rtm_family = AF_INET; 2926 r->rtm_dst_len = 32; 2927 r->rtm_src_len = 0; 2928 r->rtm_tos = fl4 ? fl4->flowi4_tos : 0; 2929 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT; 2930 if (nla_put_u32(skb, RTA_TABLE, table_id)) 2931 goto nla_put_failure; 2932 r->rtm_type = rt->rt_type; 2933 r->rtm_scope = RT_SCOPE_UNIVERSE; 2934 r->rtm_protocol = RTPROT_UNSPEC; 2935 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 2936 if (rt->rt_flags & RTCF_NOTIFY) 2937 r->rtm_flags |= RTM_F_NOTIFY; 2938 if (IPCB(skb)->flags & IPSKB_DOREDIRECT) 2939 r->rtm_flags |= RTCF_DOREDIRECT; 2940 2941 if (nla_put_in_addr(skb, RTA_DST, dst)) 2942 goto nla_put_failure; 2943 if (src) { 2944 r->rtm_src_len = 32; 2945 if (nla_put_in_addr(skb, RTA_SRC, src)) 2946 goto nla_put_failure; 2947 } 2948 if (rt->dst.dev && 2949 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex)) 2950 goto nla_put_failure; 2951 if (rt->dst.lwtstate && 2952 lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0) 2953 goto nla_put_failure; 2954 #ifdef CONFIG_IP_ROUTE_CLASSID 2955 if (rt->dst.tclassid && 2956 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid)) 2957 goto nla_put_failure; 2958 #endif 2959 if (fl4 && !rt_is_input_route(rt) && 2960 fl4->saddr != src) { 2961 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr)) 2962 goto nla_put_failure; 2963 } 2964 if (rt->rt_uses_gateway) { 2965 if (rt->rt_gw_family == AF_INET && 2966 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) { 2967 goto nla_put_failure; 2968 } else if (rt->rt_gw_family == AF_INET6) { 2969 int alen = sizeof(struct in6_addr); 2970 struct nlattr *nla; 2971 struct rtvia *via; 2972 2973 nla = nla_reserve(skb, RTA_VIA, alen + 2); 2974 if (!nla) 2975 goto nla_put_failure; 2976 2977 via = nla_data(nla); 2978 via->rtvia_family = AF_INET6; 2979 memcpy(via->rtvia_addr, &rt->rt_gw6, alen); 2980 } 2981 } 2982 2983 expires = rt->dst.expires; 2984 if (expires) { 2985 unsigned long now = jiffies; 2986 2987 if (time_before(now, expires)) 2988 expires -= now; 2989 else 2990 expires = 0; 2991 } 2992 2993 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics)); 2994 if (rt->rt_pmtu && expires) 2995 metrics[RTAX_MTU - 1] = rt->rt_pmtu; 2996 if (rt->rt_mtu_locked && expires) 2997 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU); 2998 if (rtnetlink_put_metrics(skb, metrics) < 0) 2999 goto nla_put_failure; 3000 3001 if (fl4) { 3002 if (fl4->flowi4_mark && 3003 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark)) 3004 goto nla_put_failure; 3005 3006 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) && 3007 nla_put_u32(skb, RTA_UID, 3008 from_kuid_munged(current_user_ns(), 3009 fl4->flowi4_uid))) 3010 goto nla_put_failure; 3011 3012 if (rt_is_input_route(rt)) { 3013 #ifdef CONFIG_IP_MROUTE 3014 if (ipv4_is_multicast(dst) && 3015 !ipv4_is_local_multicast(dst) && 3016 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) { 3017 int err = ipmr_get_route(net, skb, 3018 fl4->saddr, fl4->daddr, 3019 r, portid); 3020 3021 if (err <= 0) { 3022 if (err == 0) 3023 return 0; 3024 goto nla_put_failure; 3025 } 3026 } else 3027 #endif 3028 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif)) 3029 goto nla_put_failure; 3030 } 3031 } 3032 3033 error = rt->dst.error; 3034 3035 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0) 3036 goto nla_put_failure; 3037 3038 nlmsg_end(skb, nlh); 3039 return 0; 3040 3041 nla_put_failure: 3042 nlmsg_cancel(skb, nlh); 3043 return -EMSGSIZE; 3044 } 3045 3046 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb, 3047 struct netlink_callback *cb, u32 table_id, 3048 struct fnhe_hash_bucket *bucket, int genid, 3049 int *fa_index, int fa_start, unsigned int flags) 3050 { 3051 int i; 3052 3053 for (i = 0; i < FNHE_HASH_SIZE; i++) { 3054 struct fib_nh_exception *fnhe; 3055 3056 for (fnhe = rcu_dereference(bucket[i].chain); fnhe; 3057 fnhe = rcu_dereference(fnhe->fnhe_next)) { 3058 struct rtable *rt; 3059 int err; 3060 3061 if (*fa_index < fa_start) 3062 goto next; 3063 3064 if (fnhe->fnhe_genid != genid) 3065 goto next; 3066 3067 if (fnhe->fnhe_expires && 3068 time_after(jiffies, fnhe->fnhe_expires)) 3069 goto next; 3070 3071 rt = rcu_dereference(fnhe->fnhe_rth_input); 3072 if (!rt) 3073 rt = rcu_dereference(fnhe->fnhe_rth_output); 3074 if (!rt) 3075 goto next; 3076 3077 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt, 3078 table_id, NULL, skb, 3079 NETLINK_CB(cb->skb).portid, 3080 cb->nlh->nlmsg_seq, flags); 3081 if (err) 3082 return err; 3083 next: 3084 (*fa_index)++; 3085 } 3086 } 3087 3088 return 0; 3089 } 3090 3091 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb, 3092 u32 table_id, struct fib_info *fi, 3093 int *fa_index, int fa_start, unsigned int flags) 3094 { 3095 struct net *net = sock_net(cb->skb->sk); 3096 int nhsel, genid = fnhe_genid(net); 3097 3098 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) { 3099 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel); 3100 struct fnhe_hash_bucket *bucket; 3101 int err; 3102 3103 if (nhc->nhc_flags & RTNH_F_DEAD) 3104 continue; 3105 3106 rcu_read_lock(); 3107 bucket = rcu_dereference(nhc->nhc_exceptions); 3108 err = 0; 3109 if (bucket) 3110 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket, 3111 genid, fa_index, fa_start, 3112 flags); 3113 rcu_read_unlock(); 3114 if (err) 3115 return err; 3116 } 3117 3118 return 0; 3119 } 3120 3121 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst, 3122 u8 ip_proto, __be16 sport, 3123 __be16 dport) 3124 { 3125 struct sk_buff *skb; 3126 struct iphdr *iph; 3127 3128 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 3129 if (!skb) 3130 return NULL; 3131 3132 /* Reserve room for dummy headers, this skb can pass 3133 * through good chunk of routing engine. 3134 */ 3135 skb_reset_mac_header(skb); 3136 skb_reset_network_header(skb); 3137 skb->protocol = htons(ETH_P_IP); 3138 iph = skb_put(skb, sizeof(struct iphdr)); 3139 iph->protocol = ip_proto; 3140 iph->saddr = src; 3141 iph->daddr = dst; 3142 iph->version = 0x4; 3143 iph->frag_off = 0; 3144 iph->ihl = 0x5; 3145 skb_set_transport_header(skb, skb->len); 3146 3147 switch (iph->protocol) { 3148 case IPPROTO_UDP: { 3149 struct udphdr *udph; 3150 3151 udph = skb_put_zero(skb, sizeof(struct udphdr)); 3152 udph->source = sport; 3153 udph->dest = dport; 3154 udph->len = sizeof(struct udphdr); 3155 udph->check = 0; 3156 break; 3157 } 3158 case IPPROTO_TCP: { 3159 struct tcphdr *tcph; 3160 3161 tcph = skb_put_zero(skb, sizeof(struct tcphdr)); 3162 tcph->source = sport; 3163 tcph->dest = dport; 3164 tcph->doff = sizeof(struct tcphdr) / 4; 3165 tcph->rst = 1; 3166 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr), 3167 src, dst, 0); 3168 break; 3169 } 3170 case IPPROTO_ICMP: { 3171 struct icmphdr *icmph; 3172 3173 icmph = skb_put_zero(skb, sizeof(struct icmphdr)); 3174 icmph->type = ICMP_ECHO; 3175 icmph->code = 0; 3176 } 3177 } 3178 3179 return skb; 3180 } 3181 3182 static int inet_rtm_valid_getroute_req(struct sk_buff *skb, 3183 const struct nlmsghdr *nlh, 3184 struct nlattr **tb, 3185 struct netlink_ext_ack *extack) 3186 { 3187 struct rtmsg *rtm; 3188 int i, err; 3189 3190 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) { 3191 NL_SET_ERR_MSG(extack, 3192 "ipv4: Invalid header for route get request"); 3193 return -EINVAL; 3194 } 3195 3196 if (!netlink_strict_get_check(skb)) 3197 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX, 3198 rtm_ipv4_policy, extack); 3199 3200 rtm = nlmsg_data(nlh); 3201 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) || 3202 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) || 3203 rtm->rtm_table || rtm->rtm_protocol || 3204 rtm->rtm_scope || rtm->rtm_type) { 3205 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request"); 3206 return -EINVAL; 3207 } 3208 3209 if (rtm->rtm_flags & ~(RTM_F_NOTIFY | 3210 RTM_F_LOOKUP_TABLE | 3211 RTM_F_FIB_MATCH)) { 3212 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request"); 3213 return -EINVAL; 3214 } 3215 3216 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX, 3217 rtm_ipv4_policy, extack); 3218 if (err) 3219 return err; 3220 3221 if ((tb[RTA_SRC] && !rtm->rtm_src_len) || 3222 (tb[RTA_DST] && !rtm->rtm_dst_len)) { 3223 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4"); 3224 return -EINVAL; 3225 } 3226 3227 for (i = 0; i <= RTA_MAX; i++) { 3228 if (!tb[i]) 3229 continue; 3230 3231 switch (i) { 3232 case RTA_IIF: 3233 case RTA_OIF: 3234 case RTA_SRC: 3235 case RTA_DST: 3236 case RTA_IP_PROTO: 3237 case RTA_SPORT: 3238 case RTA_DPORT: 3239 case RTA_MARK: 3240 case RTA_UID: 3241 break; 3242 default: 3243 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request"); 3244 return -EINVAL; 3245 } 3246 } 3247 3248 return 0; 3249 } 3250 3251 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, 3252 struct netlink_ext_ack *extack) 3253 { 3254 struct net *net = sock_net(in_skb->sk); 3255 struct nlattr *tb[RTA_MAX+1]; 3256 u32 table_id = RT_TABLE_MAIN; 3257 __be16 sport = 0, dport = 0; 3258 struct fib_result res = {}; 3259 u8 ip_proto = IPPROTO_UDP; 3260 struct rtable *rt = NULL; 3261 struct sk_buff *skb; 3262 struct rtmsg *rtm; 3263 struct flowi4 fl4 = {}; 3264 __be32 dst = 0; 3265 __be32 src = 0; 3266 kuid_t uid; 3267 u32 iif; 3268 int err; 3269 int mark; 3270 3271 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack); 3272 if (err < 0) 3273 return err; 3274 3275 rtm = nlmsg_data(nlh); 3276 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0; 3277 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0; 3278 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0; 3279 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0; 3280 if (tb[RTA_UID]) 3281 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID])); 3282 else 3283 uid = (iif ? INVALID_UID : current_uid()); 3284 3285 if (tb[RTA_IP_PROTO]) { 3286 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO], 3287 &ip_proto, AF_INET, extack); 3288 if (err) 3289 return err; 3290 } 3291 3292 if (tb[RTA_SPORT]) 3293 sport = nla_get_be16(tb[RTA_SPORT]); 3294 3295 if (tb[RTA_DPORT]) 3296 dport = nla_get_be16(tb[RTA_DPORT]); 3297 3298 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport); 3299 if (!skb) 3300 return -ENOBUFS; 3301 3302 fl4.daddr = dst; 3303 fl4.saddr = src; 3304 fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK; 3305 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0; 3306 fl4.flowi4_mark = mark; 3307 fl4.flowi4_uid = uid; 3308 if (sport) 3309 fl4.fl4_sport = sport; 3310 if (dport) 3311 fl4.fl4_dport = dport; 3312 fl4.flowi4_proto = ip_proto; 3313 3314 rcu_read_lock(); 3315 3316 if (iif) { 3317 struct net_device *dev; 3318 3319 dev = dev_get_by_index_rcu(net, iif); 3320 if (!dev) { 3321 err = -ENODEV; 3322 goto errout_rcu; 3323 } 3324 3325 fl4.flowi4_iif = iif; /* for rt_fill_info */ 3326 skb->dev = dev; 3327 skb->mark = mark; 3328 err = ip_route_input_rcu(skb, dst, src, 3329 rtm->rtm_tos & IPTOS_RT_MASK, dev, 3330 &res); 3331 3332 rt = skb_rtable(skb); 3333 if (err == 0 && rt->dst.error) 3334 err = -rt->dst.error; 3335 } else { 3336 fl4.flowi4_iif = LOOPBACK_IFINDEX; 3337 skb->dev = net->loopback_dev; 3338 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb); 3339 err = 0; 3340 if (IS_ERR(rt)) 3341 err = PTR_ERR(rt); 3342 else 3343 skb_dst_set(skb, &rt->dst); 3344 } 3345 3346 if (err) 3347 goto errout_rcu; 3348 3349 if (rtm->rtm_flags & RTM_F_NOTIFY) 3350 rt->rt_flags |= RTCF_NOTIFY; 3351 3352 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE) 3353 table_id = res.table ? res.table->tb_id : 0; 3354 3355 /* reset skb for netlink reply msg */ 3356 skb_trim(skb, 0); 3357 skb_reset_network_header(skb); 3358 skb_reset_transport_header(skb); 3359 skb_reset_mac_header(skb); 3360 3361 if (rtm->rtm_flags & RTM_F_FIB_MATCH) { 3362 struct fib_rt_info fri; 3363 3364 if (!res.fi) { 3365 err = fib_props[res.type].error; 3366 if (!err) 3367 err = -EHOSTUNREACH; 3368 goto errout_rcu; 3369 } 3370 fri.fi = res.fi; 3371 fri.tb_id = table_id; 3372 fri.dst = res.prefix; 3373 fri.dst_len = res.prefixlen; 3374 fri.tos = fl4.flowi4_tos; 3375 fri.type = rt->rt_type; 3376 fri.offload = 0; 3377 fri.trap = 0; 3378 fri.offload_failed = 0; 3379 if (res.fa_head) { 3380 struct fib_alias *fa; 3381 3382 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) { 3383 u8 slen = 32 - fri.dst_len; 3384 3385 if (fa->fa_slen == slen && 3386 fa->tb_id == fri.tb_id && 3387 fa->fa_tos == fri.tos && 3388 fa->fa_info == res.fi && 3389 fa->fa_type == fri.type) { 3390 fri.offload = fa->offload; 3391 fri.trap = fa->trap; 3392 break; 3393 } 3394 } 3395 } 3396 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid, 3397 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0); 3398 } else { 3399 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb, 3400 NETLINK_CB(in_skb).portid, 3401 nlh->nlmsg_seq, 0); 3402 } 3403 if (err < 0) 3404 goto errout_rcu; 3405 3406 rcu_read_unlock(); 3407 3408 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 3409 3410 errout_free: 3411 return err; 3412 errout_rcu: 3413 rcu_read_unlock(); 3414 kfree_skb(skb); 3415 goto errout_free; 3416 } 3417 3418 void ip_rt_multicast_event(struct in_device *in_dev) 3419 { 3420 rt_cache_flush(dev_net(in_dev->dev)); 3421 } 3422 3423 #ifdef CONFIG_SYSCTL 3424 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 3425 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 3426 static int ip_rt_gc_elasticity __read_mostly = 8; 3427 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU; 3428 3429 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write, 3430 void *buffer, size_t *lenp, loff_t *ppos) 3431 { 3432 struct net *net = (struct net *)__ctl->extra1; 3433 3434 if (write) { 3435 rt_cache_flush(net); 3436 fnhe_genid_bump(net); 3437 return 0; 3438 } 3439 3440 return -EINVAL; 3441 } 3442 3443 static struct ctl_table ipv4_route_table[] = { 3444 { 3445 .procname = "gc_thresh", 3446 .data = &ipv4_dst_ops.gc_thresh, 3447 .maxlen = sizeof(int), 3448 .mode = 0644, 3449 .proc_handler = proc_dointvec, 3450 }, 3451 { 3452 .procname = "max_size", 3453 .data = &ip_rt_max_size, 3454 .maxlen = sizeof(int), 3455 .mode = 0644, 3456 .proc_handler = proc_dointvec, 3457 }, 3458 { 3459 /* Deprecated. Use gc_min_interval_ms */ 3460 3461 .procname = "gc_min_interval", 3462 .data = &ip_rt_gc_min_interval, 3463 .maxlen = sizeof(int), 3464 .mode = 0644, 3465 .proc_handler = proc_dointvec_jiffies, 3466 }, 3467 { 3468 .procname = "gc_min_interval_ms", 3469 .data = &ip_rt_gc_min_interval, 3470 .maxlen = sizeof(int), 3471 .mode = 0644, 3472 .proc_handler = proc_dointvec_ms_jiffies, 3473 }, 3474 { 3475 .procname = "gc_timeout", 3476 .data = &ip_rt_gc_timeout, 3477 .maxlen = sizeof(int), 3478 .mode = 0644, 3479 .proc_handler = proc_dointvec_jiffies, 3480 }, 3481 { 3482 .procname = "gc_interval", 3483 .data = &ip_rt_gc_interval, 3484 .maxlen = sizeof(int), 3485 .mode = 0644, 3486 .proc_handler = proc_dointvec_jiffies, 3487 }, 3488 { 3489 .procname = "redirect_load", 3490 .data = &ip_rt_redirect_load, 3491 .maxlen = sizeof(int), 3492 .mode = 0644, 3493 .proc_handler = proc_dointvec, 3494 }, 3495 { 3496 .procname = "redirect_number", 3497 .data = &ip_rt_redirect_number, 3498 .maxlen = sizeof(int), 3499 .mode = 0644, 3500 .proc_handler = proc_dointvec, 3501 }, 3502 { 3503 .procname = "redirect_silence", 3504 .data = &ip_rt_redirect_silence, 3505 .maxlen = sizeof(int), 3506 .mode = 0644, 3507 .proc_handler = proc_dointvec, 3508 }, 3509 { 3510 .procname = "error_cost", 3511 .data = &ip_rt_error_cost, 3512 .maxlen = sizeof(int), 3513 .mode = 0644, 3514 .proc_handler = proc_dointvec, 3515 }, 3516 { 3517 .procname = "error_burst", 3518 .data = &ip_rt_error_burst, 3519 .maxlen = sizeof(int), 3520 .mode = 0644, 3521 .proc_handler = proc_dointvec, 3522 }, 3523 { 3524 .procname = "gc_elasticity", 3525 .data = &ip_rt_gc_elasticity, 3526 .maxlen = sizeof(int), 3527 .mode = 0644, 3528 .proc_handler = proc_dointvec, 3529 }, 3530 { 3531 .procname = "mtu_expires", 3532 .data = &ip_rt_mtu_expires, 3533 .maxlen = sizeof(int), 3534 .mode = 0644, 3535 .proc_handler = proc_dointvec_jiffies, 3536 }, 3537 { 3538 .procname = "min_pmtu", 3539 .data = &ip_rt_min_pmtu, 3540 .maxlen = sizeof(int), 3541 .mode = 0644, 3542 .proc_handler = proc_dointvec_minmax, 3543 .extra1 = &ip_min_valid_pmtu, 3544 }, 3545 { 3546 .procname = "min_adv_mss", 3547 .data = &ip_rt_min_advmss, 3548 .maxlen = sizeof(int), 3549 .mode = 0644, 3550 .proc_handler = proc_dointvec, 3551 }, 3552 { } 3553 }; 3554 3555 static const char ipv4_route_flush_procname[] = "flush"; 3556 3557 static struct ctl_table ipv4_route_flush_table[] = { 3558 { 3559 .procname = ipv4_route_flush_procname, 3560 .maxlen = sizeof(int), 3561 .mode = 0200, 3562 .proc_handler = ipv4_sysctl_rtcache_flush, 3563 }, 3564 { }, 3565 }; 3566 3567 static __net_init int sysctl_route_net_init(struct net *net) 3568 { 3569 struct ctl_table *tbl; 3570 3571 tbl = ipv4_route_flush_table; 3572 if (!net_eq(net, &init_net)) { 3573 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL); 3574 if (!tbl) 3575 goto err_dup; 3576 3577 /* Don't export non-whitelisted sysctls to unprivileged users */ 3578 if (net->user_ns != &init_user_ns) { 3579 if (tbl[0].procname != ipv4_route_flush_procname) 3580 tbl[0].procname = NULL; 3581 } 3582 } 3583 tbl[0].extra1 = net; 3584 3585 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl); 3586 if (!net->ipv4.route_hdr) 3587 goto err_reg; 3588 return 0; 3589 3590 err_reg: 3591 if (tbl != ipv4_route_flush_table) 3592 kfree(tbl); 3593 err_dup: 3594 return -ENOMEM; 3595 } 3596 3597 static __net_exit void sysctl_route_net_exit(struct net *net) 3598 { 3599 struct ctl_table *tbl; 3600 3601 tbl = net->ipv4.route_hdr->ctl_table_arg; 3602 unregister_net_sysctl_table(net->ipv4.route_hdr); 3603 BUG_ON(tbl == ipv4_route_flush_table); 3604 kfree(tbl); 3605 } 3606 3607 static __net_initdata struct pernet_operations sysctl_route_ops = { 3608 .init = sysctl_route_net_init, 3609 .exit = sysctl_route_net_exit, 3610 }; 3611 #endif 3612 3613 static __net_init int rt_genid_init(struct net *net) 3614 { 3615 atomic_set(&net->ipv4.rt_genid, 0); 3616 atomic_set(&net->fnhe_genid, 0); 3617 atomic_set(&net->ipv4.dev_addr_genid, get_random_int()); 3618 return 0; 3619 } 3620 3621 static __net_initdata struct pernet_operations rt_genid_ops = { 3622 .init = rt_genid_init, 3623 }; 3624 3625 static int __net_init ipv4_inetpeer_init(struct net *net) 3626 { 3627 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL); 3628 3629 if (!bp) 3630 return -ENOMEM; 3631 inet_peer_base_init(bp); 3632 net->ipv4.peers = bp; 3633 return 0; 3634 } 3635 3636 static void __net_exit ipv4_inetpeer_exit(struct net *net) 3637 { 3638 struct inet_peer_base *bp = net->ipv4.peers; 3639 3640 net->ipv4.peers = NULL; 3641 inetpeer_invalidate_tree(bp); 3642 kfree(bp); 3643 } 3644 3645 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = { 3646 .init = ipv4_inetpeer_init, 3647 .exit = ipv4_inetpeer_exit, 3648 }; 3649 3650 #ifdef CONFIG_IP_ROUTE_CLASSID 3651 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 3652 #endif /* CONFIG_IP_ROUTE_CLASSID */ 3653 3654 int __init ip_rt_init(void) 3655 { 3656 void *idents_hash; 3657 int cpu; 3658 3659 /* For modern hosts, this will use 2 MB of memory */ 3660 idents_hash = alloc_large_system_hash("IP idents", 3661 sizeof(*ip_idents) + sizeof(*ip_tstamps), 3662 0, 3663 16, /* one bucket per 64 KB */ 3664 HASH_ZERO, 3665 NULL, 3666 &ip_idents_mask, 3667 2048, 3668 256*1024); 3669 3670 ip_idents = idents_hash; 3671 3672 prandom_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents)); 3673 3674 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents); 3675 3676 for_each_possible_cpu(cpu) { 3677 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 3678 3679 INIT_LIST_HEAD(&ul->head); 3680 spin_lock_init(&ul->lock); 3681 } 3682 #ifdef CONFIG_IP_ROUTE_CLASSID 3683 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 3684 if (!ip_rt_acct) 3685 panic("IP: failed to allocate ip_rt_acct\n"); 3686 #endif 3687 3688 ipv4_dst_ops.kmem_cachep = 3689 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0, 3690 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 3691 3692 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 3693 3694 if (dst_entries_init(&ipv4_dst_ops) < 0) 3695 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 3696 3697 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 3698 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 3699 3700 ipv4_dst_ops.gc_thresh = ~0; 3701 ip_rt_max_size = INT_MAX; 3702 3703 devinet_init(); 3704 ip_fib_init(); 3705 3706 if (ip_rt_proc_init()) 3707 pr_err("Unable to create route proc files\n"); 3708 #ifdef CONFIG_XFRM 3709 xfrm_init(); 3710 xfrm4_init(); 3711 #endif 3712 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, 3713 RTNL_FLAG_DOIT_UNLOCKED); 3714 3715 #ifdef CONFIG_SYSCTL 3716 register_pernet_subsys(&sysctl_route_ops); 3717 #endif 3718 register_pernet_subsys(&rt_genid_ops); 3719 register_pernet_subsys(&ipv4_inetpeer_ops); 3720 return 0; 3721 } 3722 3723 #ifdef CONFIG_SYSCTL 3724 /* 3725 * We really need to sanitize the damn ipv4 init order, then all 3726 * this nonsense will go away. 3727 */ 3728 void __init ip_static_sysctl_init(void) 3729 { 3730 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table); 3731 } 3732 #endif 3733