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/bitops.h> 65 #include <linux/kernel.h> 66 #include <linux/mm.h> 67 #include <linux/memblock.h> 68 #include <linux/socket.h> 69 #include <linux/errno.h> 70 #include <linux/in.h> 71 #include <linux/inet.h> 72 #include <linux/netdevice.h> 73 #include <linux/proc_fs.h> 74 #include <linux/init.h> 75 #include <linux/skbuff.h> 76 #include <linux/inetdevice.h> 77 #include <linux/igmp.h> 78 #include <linux/pkt_sched.h> 79 #include <linux/mroute.h> 80 #include <linux/netfilter_ipv4.h> 81 #include <linux/random.h> 82 #include <linux/rcupdate.h> 83 #include <linux/slab.h> 84 #include <linux/jhash.h> 85 #include <net/dst.h> 86 #include <net/dst_metadata.h> 87 #include <net/inet_dscp.h> 88 #include <net/net_namespace.h> 89 #include <net/ip.h> 90 #include <net/route.h> 91 #include <net/inetpeer.h> 92 #include <net/sock.h> 93 #include <net/ip_fib.h> 94 #include <net/nexthop.h> 95 #include <net/tcp.h> 96 #include <net/icmp.h> 97 #include <net/xfrm.h> 98 #include <net/lwtunnel.h> 99 #include <net/netevent.h> 100 #include <net/rtnetlink.h> 101 #ifdef CONFIG_SYSCTL 102 #include <linux/sysctl.h> 103 #endif 104 #include <net/secure_seq.h> 105 #include <net/ip_tunnels.h> 106 107 #include "fib_lookup.h" 108 109 #define RT_FL_TOS(oldflp4) \ 110 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)) 111 112 #define RT_GC_TIMEOUT (300*HZ) 113 114 #define DEFAULT_MIN_PMTU (512 + 20 + 20) 115 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ) 116 #define DEFAULT_MIN_ADVMSS 256 117 static int ip_rt_max_size; 118 static int ip_rt_redirect_number __read_mostly = 9; 119 static int ip_rt_redirect_load __read_mostly = HZ / 50; 120 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1)); 121 static int ip_rt_error_cost __read_mostly = HZ; 122 static int ip_rt_error_burst __read_mostly = 5 * HZ; 123 124 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT; 125 126 /* 127 * Interface to generic destination cache. 128 */ 129 130 INDIRECT_CALLABLE_SCOPE 131 struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie); 132 static unsigned int ipv4_default_advmss(const struct dst_entry *dst); 133 INDIRECT_CALLABLE_SCOPE 134 unsigned int ipv4_mtu(const struct dst_entry *dst); 135 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst); 136 static void ipv4_link_failure(struct sk_buff *skb); 137 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 138 struct sk_buff *skb, u32 mtu, 139 bool confirm_neigh); 140 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, 141 struct sk_buff *skb); 142 static void ipv4_dst_destroy(struct dst_entry *dst); 143 144 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old) 145 { 146 WARN_ON(1); 147 return NULL; 148 } 149 150 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 151 struct sk_buff *skb, 152 const void *daddr); 153 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr); 154 155 static struct dst_ops ipv4_dst_ops = { 156 .family = AF_INET, 157 .check = ipv4_dst_check, 158 .default_advmss = ipv4_default_advmss, 159 .mtu = ipv4_mtu, 160 .cow_metrics = ipv4_cow_metrics, 161 .destroy = ipv4_dst_destroy, 162 .negative_advice = ipv4_negative_advice, 163 .link_failure = ipv4_link_failure, 164 .update_pmtu = ip_rt_update_pmtu, 165 .redirect = ip_do_redirect, 166 .local_out = __ip_local_out, 167 .neigh_lookup = ipv4_neigh_lookup, 168 .confirm_neigh = ipv4_confirm_neigh, 169 }; 170 171 #define ECN_OR_COST(class) TC_PRIO_##class 172 173 const __u8 ip_tos2prio[16] = { 174 TC_PRIO_BESTEFFORT, 175 ECN_OR_COST(BESTEFFORT), 176 TC_PRIO_BESTEFFORT, 177 ECN_OR_COST(BESTEFFORT), 178 TC_PRIO_BULK, 179 ECN_OR_COST(BULK), 180 TC_PRIO_BULK, 181 ECN_OR_COST(BULK), 182 TC_PRIO_INTERACTIVE, 183 ECN_OR_COST(INTERACTIVE), 184 TC_PRIO_INTERACTIVE, 185 ECN_OR_COST(INTERACTIVE), 186 TC_PRIO_INTERACTIVE_BULK, 187 ECN_OR_COST(INTERACTIVE_BULK), 188 TC_PRIO_INTERACTIVE_BULK, 189 ECN_OR_COST(INTERACTIVE_BULK) 190 }; 191 EXPORT_SYMBOL(ip_tos2prio); 192 193 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat); 194 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field) 195 196 #ifdef CONFIG_PROC_FS 197 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos) 198 { 199 if (*pos) 200 return NULL; 201 return SEQ_START_TOKEN; 202 } 203 204 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos) 205 { 206 ++*pos; 207 return NULL; 208 } 209 210 static void rt_cache_seq_stop(struct seq_file *seq, void *v) 211 { 212 } 213 214 static int rt_cache_seq_show(struct seq_file *seq, void *v) 215 { 216 if (v == SEQ_START_TOKEN) 217 seq_printf(seq, "%-127s\n", 218 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t" 219 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t" 220 "HHUptod\tSpecDst"); 221 return 0; 222 } 223 224 static const struct seq_operations rt_cache_seq_ops = { 225 .start = rt_cache_seq_start, 226 .next = rt_cache_seq_next, 227 .stop = rt_cache_seq_stop, 228 .show = rt_cache_seq_show, 229 }; 230 231 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos) 232 { 233 int cpu; 234 235 if (*pos == 0) 236 return SEQ_START_TOKEN; 237 238 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) { 239 if (!cpu_possible(cpu)) 240 continue; 241 *pos = cpu+1; 242 return &per_cpu(rt_cache_stat, cpu); 243 } 244 return NULL; 245 } 246 247 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos) 248 { 249 int cpu; 250 251 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) { 252 if (!cpu_possible(cpu)) 253 continue; 254 *pos = cpu+1; 255 return &per_cpu(rt_cache_stat, cpu); 256 } 257 (*pos)++; 258 return NULL; 259 260 } 261 262 static void rt_cpu_seq_stop(struct seq_file *seq, void *v) 263 { 264 265 } 266 267 static int rt_cpu_seq_show(struct seq_file *seq, void *v) 268 { 269 struct rt_cache_stat *st = v; 270 271 if (v == SEQ_START_TOKEN) { 272 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"); 273 return 0; 274 } 275 276 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x " 277 "%08x %08x %08x %08x %08x %08x " 278 "%08x %08x %08x %08x\n", 279 dst_entries_get_slow(&ipv4_dst_ops), 280 0, /* st->in_hit */ 281 st->in_slow_tot, 282 st->in_slow_mc, 283 st->in_no_route, 284 st->in_brd, 285 st->in_martian_dst, 286 st->in_martian_src, 287 288 0, /* st->out_hit */ 289 st->out_slow_tot, 290 st->out_slow_mc, 291 292 0, /* st->gc_total */ 293 0, /* st->gc_ignored */ 294 0, /* st->gc_goal_miss */ 295 0, /* st->gc_dst_overflow */ 296 0, /* st->in_hlist_search */ 297 0 /* st->out_hlist_search */ 298 ); 299 return 0; 300 } 301 302 static const struct seq_operations rt_cpu_seq_ops = { 303 .start = rt_cpu_seq_start, 304 .next = rt_cpu_seq_next, 305 .stop = rt_cpu_seq_stop, 306 .show = rt_cpu_seq_show, 307 }; 308 309 #ifdef CONFIG_IP_ROUTE_CLASSID 310 static int rt_acct_proc_show(struct seq_file *m, void *v) 311 { 312 struct ip_rt_acct *dst, *src; 313 unsigned int i, j; 314 315 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL); 316 if (!dst) 317 return -ENOMEM; 318 319 for_each_possible_cpu(i) { 320 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i); 321 for (j = 0; j < 256; j++) { 322 dst[j].o_bytes += src[j].o_bytes; 323 dst[j].o_packets += src[j].o_packets; 324 dst[j].i_bytes += src[j].i_bytes; 325 dst[j].i_packets += src[j].i_packets; 326 } 327 } 328 329 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct)); 330 kfree(dst); 331 return 0; 332 } 333 #endif 334 335 static int __net_init ip_rt_do_proc_init(struct net *net) 336 { 337 struct proc_dir_entry *pde; 338 339 pde = proc_create_seq("rt_cache", 0444, net->proc_net, 340 &rt_cache_seq_ops); 341 if (!pde) 342 goto err1; 343 344 pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat, 345 &rt_cpu_seq_ops); 346 if (!pde) 347 goto err2; 348 349 #ifdef CONFIG_IP_ROUTE_CLASSID 350 pde = proc_create_single("rt_acct", 0, net->proc_net, 351 rt_acct_proc_show); 352 if (!pde) 353 goto err3; 354 #endif 355 return 0; 356 357 #ifdef CONFIG_IP_ROUTE_CLASSID 358 err3: 359 remove_proc_entry("rt_cache", net->proc_net_stat); 360 #endif 361 err2: 362 remove_proc_entry("rt_cache", net->proc_net); 363 err1: 364 return -ENOMEM; 365 } 366 367 static void __net_exit ip_rt_do_proc_exit(struct net *net) 368 { 369 remove_proc_entry("rt_cache", net->proc_net_stat); 370 remove_proc_entry("rt_cache", net->proc_net); 371 #ifdef CONFIG_IP_ROUTE_CLASSID 372 remove_proc_entry("rt_acct", net->proc_net); 373 #endif 374 } 375 376 static struct pernet_operations ip_rt_proc_ops __net_initdata = { 377 .init = ip_rt_do_proc_init, 378 .exit = ip_rt_do_proc_exit, 379 }; 380 381 static int __init ip_rt_proc_init(void) 382 { 383 return register_pernet_subsys(&ip_rt_proc_ops); 384 } 385 386 #else 387 static inline int ip_rt_proc_init(void) 388 { 389 return 0; 390 } 391 #endif /* CONFIG_PROC_FS */ 392 393 static inline bool rt_is_expired(const struct rtable *rth) 394 { 395 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev)); 396 } 397 398 void rt_cache_flush(struct net *net) 399 { 400 rt_genid_bump_ipv4(net); 401 } 402 403 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 404 struct sk_buff *skb, 405 const void *daddr) 406 { 407 const struct rtable *rt = container_of(dst, struct rtable, dst); 408 struct net_device *dev = dst->dev; 409 struct neighbour *n; 410 411 rcu_read_lock_bh(); 412 413 if (likely(rt->rt_gw_family == AF_INET)) { 414 n = ip_neigh_gw4(dev, rt->rt_gw4); 415 } else if (rt->rt_gw_family == AF_INET6) { 416 n = ip_neigh_gw6(dev, &rt->rt_gw6); 417 } else { 418 __be32 pkey; 419 420 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr); 421 n = ip_neigh_gw4(dev, pkey); 422 } 423 424 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt)) 425 n = NULL; 426 427 rcu_read_unlock_bh(); 428 429 return n; 430 } 431 432 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr) 433 { 434 const struct rtable *rt = container_of(dst, struct rtable, dst); 435 struct net_device *dev = dst->dev; 436 const __be32 *pkey = daddr; 437 438 if (rt->rt_gw_family == AF_INET) { 439 pkey = (const __be32 *)&rt->rt_gw4; 440 } else if (rt->rt_gw_family == AF_INET6) { 441 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6); 442 } else if (!daddr || 443 (rt->rt_flags & 444 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) { 445 return; 446 } 447 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey); 448 } 449 450 /* Hash tables of size 2048..262144 depending on RAM size. 451 * Each bucket uses 8 bytes. 452 */ 453 static u32 ip_idents_mask __read_mostly; 454 static atomic_t *ip_idents __read_mostly; 455 static u32 *ip_tstamps __read_mostly; 456 457 /* In order to protect privacy, we add a perturbation to identifiers 458 * if one generator is seldom used. This makes hard for an attacker 459 * to infer how many packets were sent between two points in time. 460 */ 461 static u32 ip_idents_reserve(u32 hash, int segs) 462 { 463 u32 bucket, old, now = (u32)jiffies; 464 atomic_t *p_id; 465 u32 *p_tstamp; 466 u32 delta = 0; 467 468 bucket = hash & ip_idents_mask; 469 p_tstamp = ip_tstamps + bucket; 470 p_id = ip_idents + bucket; 471 old = READ_ONCE(*p_tstamp); 472 473 if (old != now && cmpxchg(p_tstamp, old, now) == old) 474 delta = prandom_u32_max(now - old); 475 476 /* If UBSAN reports an error there, please make sure your compiler 477 * supports -fno-strict-overflow before reporting it that was a bug 478 * in UBSAN, and it has been fixed in GCC-8. 479 */ 480 return atomic_add_return(segs + delta, p_id) - segs; 481 } 482 483 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs) 484 { 485 u32 hash, id; 486 487 /* Note the following code is not safe, but this is okay. */ 488 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key))) 489 get_random_bytes(&net->ipv4.ip_id_key, 490 sizeof(net->ipv4.ip_id_key)); 491 492 hash = siphash_3u32((__force u32)iph->daddr, 493 (__force u32)iph->saddr, 494 iph->protocol, 495 &net->ipv4.ip_id_key); 496 id = ip_idents_reserve(hash, segs); 497 iph->id = htons(id); 498 } 499 EXPORT_SYMBOL(__ip_select_ident); 500 501 static void __build_flow_key(const struct net *net, struct flowi4 *fl4, 502 const struct sock *sk, 503 const struct iphdr *iph, 504 int oif, u8 tos, 505 u8 prot, u32 mark, int flow_flags) 506 { 507 if (sk) { 508 const struct inet_sock *inet = inet_sk(sk); 509 510 oif = sk->sk_bound_dev_if; 511 mark = sk->sk_mark; 512 tos = RT_CONN_FLAGS(sk); 513 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol; 514 } 515 flowi4_init_output(fl4, oif, mark, tos, 516 RT_SCOPE_UNIVERSE, prot, 517 flow_flags, 518 iph->daddr, iph->saddr, 0, 0, 519 sock_net_uid(net, sk)); 520 } 521 522 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb, 523 const struct sock *sk) 524 { 525 const struct net *net = dev_net(skb->dev); 526 const struct iphdr *iph = ip_hdr(skb); 527 int oif = skb->dev->ifindex; 528 u8 tos = RT_TOS(iph->tos); 529 u8 prot = iph->protocol; 530 u32 mark = skb->mark; 531 532 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0); 533 } 534 535 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk) 536 { 537 const struct inet_sock *inet = inet_sk(sk); 538 const struct ip_options_rcu *inet_opt; 539 __be32 daddr = inet->inet_daddr; 540 541 rcu_read_lock(); 542 inet_opt = rcu_dereference(inet->inet_opt); 543 if (inet_opt && inet_opt->opt.srr) 544 daddr = inet_opt->opt.faddr; 545 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark, 546 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 547 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol, 548 inet_sk_flowi_flags(sk), 549 daddr, inet->inet_saddr, 0, 0, sk->sk_uid); 550 rcu_read_unlock(); 551 } 552 553 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk, 554 const struct sk_buff *skb) 555 { 556 if (skb) 557 build_skb_flow_key(fl4, skb, sk); 558 else 559 build_sk_flow_key(fl4, sk); 560 } 561 562 static DEFINE_SPINLOCK(fnhe_lock); 563 564 static void fnhe_flush_routes(struct fib_nh_exception *fnhe) 565 { 566 struct rtable *rt; 567 568 rt = rcu_dereference(fnhe->fnhe_rth_input); 569 if (rt) { 570 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL); 571 dst_dev_put(&rt->dst); 572 dst_release(&rt->dst); 573 } 574 rt = rcu_dereference(fnhe->fnhe_rth_output); 575 if (rt) { 576 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL); 577 dst_dev_put(&rt->dst); 578 dst_release(&rt->dst); 579 } 580 } 581 582 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash) 583 { 584 struct fib_nh_exception __rcu **fnhe_p, **oldest_p; 585 struct fib_nh_exception *fnhe, *oldest = NULL; 586 587 for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) { 588 fnhe = rcu_dereference_protected(*fnhe_p, 589 lockdep_is_held(&fnhe_lock)); 590 if (!fnhe) 591 break; 592 if (!oldest || 593 time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) { 594 oldest = fnhe; 595 oldest_p = fnhe_p; 596 } 597 } 598 fnhe_flush_routes(oldest); 599 *oldest_p = oldest->fnhe_next; 600 kfree_rcu(oldest, rcu); 601 } 602 603 static u32 fnhe_hashfun(__be32 daddr) 604 { 605 static siphash_aligned_key_t fnhe_hash_key; 606 u64 hval; 607 608 net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key)); 609 hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key); 610 return hash_64(hval, FNHE_HASH_SHIFT); 611 } 612 613 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe) 614 { 615 rt->rt_pmtu = fnhe->fnhe_pmtu; 616 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked; 617 rt->dst.expires = fnhe->fnhe_expires; 618 619 if (fnhe->fnhe_gw) { 620 rt->rt_flags |= RTCF_REDIRECTED; 621 rt->rt_uses_gateway = 1; 622 rt->rt_gw_family = AF_INET; 623 rt->rt_gw4 = fnhe->fnhe_gw; 624 } 625 } 626 627 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr, 628 __be32 gw, u32 pmtu, bool lock, 629 unsigned long expires) 630 { 631 struct fnhe_hash_bucket *hash; 632 struct fib_nh_exception *fnhe; 633 struct rtable *rt; 634 u32 genid, hval; 635 unsigned int i; 636 int depth; 637 638 genid = fnhe_genid(dev_net(nhc->nhc_dev)); 639 hval = fnhe_hashfun(daddr); 640 641 spin_lock_bh(&fnhe_lock); 642 643 hash = rcu_dereference(nhc->nhc_exceptions); 644 if (!hash) { 645 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC); 646 if (!hash) 647 goto out_unlock; 648 rcu_assign_pointer(nhc->nhc_exceptions, hash); 649 } 650 651 hash += hval; 652 653 depth = 0; 654 for (fnhe = rcu_dereference(hash->chain); fnhe; 655 fnhe = rcu_dereference(fnhe->fnhe_next)) { 656 if (fnhe->fnhe_daddr == daddr) 657 break; 658 depth++; 659 } 660 661 if (fnhe) { 662 if (fnhe->fnhe_genid != genid) 663 fnhe->fnhe_genid = genid; 664 if (gw) 665 fnhe->fnhe_gw = gw; 666 if (pmtu) { 667 fnhe->fnhe_pmtu = pmtu; 668 fnhe->fnhe_mtu_locked = lock; 669 } 670 fnhe->fnhe_expires = max(1UL, expires); 671 /* Update all cached dsts too */ 672 rt = rcu_dereference(fnhe->fnhe_rth_input); 673 if (rt) 674 fill_route_from_fnhe(rt, fnhe); 675 rt = rcu_dereference(fnhe->fnhe_rth_output); 676 if (rt) 677 fill_route_from_fnhe(rt, fnhe); 678 } else { 679 /* Randomize max depth to avoid some side channels attacks. */ 680 int max_depth = FNHE_RECLAIM_DEPTH + 681 prandom_u32_max(FNHE_RECLAIM_DEPTH); 682 683 while (depth > max_depth) { 684 fnhe_remove_oldest(hash); 685 depth--; 686 } 687 688 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC); 689 if (!fnhe) 690 goto out_unlock; 691 692 fnhe->fnhe_next = hash->chain; 693 694 fnhe->fnhe_genid = genid; 695 fnhe->fnhe_daddr = daddr; 696 fnhe->fnhe_gw = gw; 697 fnhe->fnhe_pmtu = pmtu; 698 fnhe->fnhe_mtu_locked = lock; 699 fnhe->fnhe_expires = max(1UL, expires); 700 701 rcu_assign_pointer(hash->chain, fnhe); 702 703 /* Exception created; mark the cached routes for the nexthop 704 * stale, so anyone caching it rechecks if this exception 705 * applies to them. 706 */ 707 rt = rcu_dereference(nhc->nhc_rth_input); 708 if (rt) 709 rt->dst.obsolete = DST_OBSOLETE_KILL; 710 711 for_each_possible_cpu(i) { 712 struct rtable __rcu **prt; 713 714 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i); 715 rt = rcu_dereference(*prt); 716 if (rt) 717 rt->dst.obsolete = DST_OBSOLETE_KILL; 718 } 719 } 720 721 fnhe->fnhe_stamp = jiffies; 722 723 out_unlock: 724 spin_unlock_bh(&fnhe_lock); 725 } 726 727 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4, 728 bool kill_route) 729 { 730 __be32 new_gw = icmp_hdr(skb)->un.gateway; 731 __be32 old_gw = ip_hdr(skb)->saddr; 732 struct net_device *dev = skb->dev; 733 struct in_device *in_dev; 734 struct fib_result res; 735 struct neighbour *n; 736 struct net *net; 737 738 switch (icmp_hdr(skb)->code & 7) { 739 case ICMP_REDIR_NET: 740 case ICMP_REDIR_NETTOS: 741 case ICMP_REDIR_HOST: 742 case ICMP_REDIR_HOSTTOS: 743 break; 744 745 default: 746 return; 747 } 748 749 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw) 750 return; 751 752 in_dev = __in_dev_get_rcu(dev); 753 if (!in_dev) 754 return; 755 756 net = dev_net(dev); 757 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) || 758 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) || 759 ipv4_is_zeronet(new_gw)) 760 goto reject_redirect; 761 762 if (!IN_DEV_SHARED_MEDIA(in_dev)) { 763 if (!inet_addr_onlink(in_dev, new_gw, old_gw)) 764 goto reject_redirect; 765 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) 766 goto reject_redirect; 767 } else { 768 if (inet_addr_type(net, new_gw) != RTN_UNICAST) 769 goto reject_redirect; 770 } 771 772 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw); 773 if (!n) 774 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev); 775 if (!IS_ERR(n)) { 776 if (!(n->nud_state & NUD_VALID)) { 777 neigh_event_send(n, NULL); 778 } else { 779 if (fib_lookup(net, fl4, &res, 0) == 0) { 780 struct fib_nh_common *nhc; 781 782 fib_select_path(net, &res, fl4, skb); 783 nhc = FIB_RES_NHC(res); 784 update_or_create_fnhe(nhc, fl4->daddr, new_gw, 785 0, false, 786 jiffies + ip_rt_gc_timeout); 787 } 788 if (kill_route) 789 rt->dst.obsolete = DST_OBSOLETE_KILL; 790 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n); 791 } 792 neigh_release(n); 793 } 794 return; 795 796 reject_redirect: 797 #ifdef CONFIG_IP_ROUTE_VERBOSE 798 if (IN_DEV_LOG_MARTIANS(in_dev)) { 799 const struct iphdr *iph = (const struct iphdr *) skb->data; 800 __be32 daddr = iph->daddr; 801 __be32 saddr = iph->saddr; 802 803 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n" 804 " Advised path = %pI4 -> %pI4\n", 805 &old_gw, dev->name, &new_gw, 806 &saddr, &daddr); 807 } 808 #endif 809 ; 810 } 811 812 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb) 813 { 814 struct rtable *rt; 815 struct flowi4 fl4; 816 const struct iphdr *iph = (const struct iphdr *) skb->data; 817 struct net *net = dev_net(skb->dev); 818 int oif = skb->dev->ifindex; 819 u8 tos = RT_TOS(iph->tos); 820 u8 prot = iph->protocol; 821 u32 mark = skb->mark; 822 823 rt = (struct rtable *) dst; 824 825 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0); 826 __ip_do_redirect(rt, skb, &fl4, true); 827 } 828 829 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst) 830 { 831 struct rtable *rt = (struct rtable *)dst; 832 struct dst_entry *ret = dst; 833 834 if (rt) { 835 if (dst->obsolete > 0) { 836 ip_rt_put(rt); 837 ret = NULL; 838 } else if ((rt->rt_flags & RTCF_REDIRECTED) || 839 rt->dst.expires) { 840 ip_rt_put(rt); 841 ret = NULL; 842 } 843 } 844 return ret; 845 } 846 847 /* 848 * Algorithm: 849 * 1. The first ip_rt_redirect_number redirects are sent 850 * with exponential backoff, then we stop sending them at all, 851 * assuming that the host ignores our redirects. 852 * 2. If we did not see packets requiring redirects 853 * during ip_rt_redirect_silence, we assume that the host 854 * forgot redirected route and start to send redirects again. 855 * 856 * This algorithm is much cheaper and more intelligent than dumb load limiting 857 * in icmp.c. 858 * 859 * NOTE. Do not forget to inhibit load limiting for redirects (redundant) 860 * and "frag. need" (breaks PMTU discovery) in icmp.c. 861 */ 862 863 void ip_rt_send_redirect(struct sk_buff *skb) 864 { 865 struct rtable *rt = skb_rtable(skb); 866 struct in_device *in_dev; 867 struct inet_peer *peer; 868 struct net *net; 869 int log_martians; 870 int vif; 871 872 rcu_read_lock(); 873 in_dev = __in_dev_get_rcu(rt->dst.dev); 874 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) { 875 rcu_read_unlock(); 876 return; 877 } 878 log_martians = IN_DEV_LOG_MARTIANS(in_dev); 879 vif = l3mdev_master_ifindex_rcu(rt->dst.dev); 880 rcu_read_unlock(); 881 882 net = dev_net(rt->dst.dev); 883 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1); 884 if (!peer) { 885 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, 886 rt_nexthop(rt, ip_hdr(skb)->daddr)); 887 return; 888 } 889 890 /* No redirected packets during ip_rt_redirect_silence; 891 * reset the algorithm. 892 */ 893 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) { 894 peer->rate_tokens = 0; 895 peer->n_redirects = 0; 896 } 897 898 /* Too many ignored redirects; do not send anything 899 * set dst.rate_last to the last seen redirected packet. 900 */ 901 if (peer->n_redirects >= ip_rt_redirect_number) { 902 peer->rate_last = jiffies; 903 goto out_put_peer; 904 } 905 906 /* Check for load limit; set rate_last to the latest sent 907 * redirect. 908 */ 909 if (peer->n_redirects == 0 || 910 time_after(jiffies, 911 (peer->rate_last + 912 (ip_rt_redirect_load << peer->n_redirects)))) { 913 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr); 914 915 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw); 916 peer->rate_last = jiffies; 917 ++peer->n_redirects; 918 #ifdef CONFIG_IP_ROUTE_VERBOSE 919 if (log_martians && 920 peer->n_redirects == ip_rt_redirect_number) 921 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n", 922 &ip_hdr(skb)->saddr, inet_iif(skb), 923 &ip_hdr(skb)->daddr, &gw); 924 #endif 925 } 926 out_put_peer: 927 inet_putpeer(peer); 928 } 929 930 static int ip_error(struct sk_buff *skb) 931 { 932 struct rtable *rt = skb_rtable(skb); 933 struct net_device *dev = skb->dev; 934 struct in_device *in_dev; 935 struct inet_peer *peer; 936 unsigned long now; 937 struct net *net; 938 bool send; 939 int code; 940 941 if (netif_is_l3_master(skb->dev)) { 942 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif); 943 if (!dev) 944 goto out; 945 } 946 947 in_dev = __in_dev_get_rcu(dev); 948 949 /* IP on this device is disabled. */ 950 if (!in_dev) 951 goto out; 952 953 net = dev_net(rt->dst.dev); 954 if (!IN_DEV_FORWARD(in_dev)) { 955 switch (rt->dst.error) { 956 case EHOSTUNREACH: 957 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS); 958 break; 959 960 case ENETUNREACH: 961 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 962 break; 963 } 964 goto out; 965 } 966 967 switch (rt->dst.error) { 968 case EINVAL: 969 default: 970 goto out; 971 case EHOSTUNREACH: 972 code = ICMP_HOST_UNREACH; 973 break; 974 case ENETUNREACH: 975 code = ICMP_NET_UNREACH; 976 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 977 break; 978 case EACCES: 979 code = ICMP_PKT_FILTERED; 980 break; 981 } 982 983 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 984 l3mdev_master_ifindex(skb->dev), 1); 985 986 send = true; 987 if (peer) { 988 now = jiffies; 989 peer->rate_tokens += now - peer->rate_last; 990 if (peer->rate_tokens > ip_rt_error_burst) 991 peer->rate_tokens = ip_rt_error_burst; 992 peer->rate_last = now; 993 if (peer->rate_tokens >= ip_rt_error_cost) 994 peer->rate_tokens -= ip_rt_error_cost; 995 else 996 send = false; 997 inet_putpeer(peer); 998 } 999 if (send) 1000 icmp_send(skb, ICMP_DEST_UNREACH, code, 0); 1001 1002 out: kfree_skb(skb); 1003 return 0; 1004 } 1005 1006 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu) 1007 { 1008 struct dst_entry *dst = &rt->dst; 1009 struct net *net = dev_net(dst->dev); 1010 struct fib_result res; 1011 bool lock = false; 1012 u32 old_mtu; 1013 1014 if (ip_mtu_locked(dst)) 1015 return; 1016 1017 old_mtu = ipv4_mtu(dst); 1018 if (old_mtu < mtu) 1019 return; 1020 1021 if (mtu < net->ipv4.ip_rt_min_pmtu) { 1022 lock = true; 1023 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu); 1024 } 1025 1026 if (rt->rt_pmtu == mtu && !lock && 1027 time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2)) 1028 return; 1029 1030 rcu_read_lock(); 1031 if (fib_lookup(net, fl4, &res, 0) == 0) { 1032 struct fib_nh_common *nhc; 1033 1034 fib_select_path(net, &res, fl4, NULL); 1035 nhc = FIB_RES_NHC(res); 1036 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1037 jiffies + net->ipv4.ip_rt_mtu_expires); 1038 } 1039 rcu_read_unlock(); 1040 } 1041 1042 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 1043 struct sk_buff *skb, u32 mtu, 1044 bool confirm_neigh) 1045 { 1046 struct rtable *rt = (struct rtable *) dst; 1047 struct flowi4 fl4; 1048 1049 ip_rt_build_flow_key(&fl4, sk, skb); 1050 1051 /* Don't make lookup fail for bridged encapsulations */ 1052 if (skb && netif_is_any_bridge_port(skb->dev)) 1053 fl4.flowi4_oif = 0; 1054 1055 __ip_rt_update_pmtu(rt, &fl4, mtu); 1056 } 1057 1058 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, 1059 int oif, u8 protocol) 1060 { 1061 const struct iphdr *iph = (const struct iphdr *)skb->data; 1062 struct flowi4 fl4; 1063 struct rtable *rt; 1064 u32 mark = IP4_REPLY_MARK(net, skb->mark); 1065 1066 __build_flow_key(net, &fl4, NULL, iph, oif, 1067 RT_TOS(iph->tos), protocol, mark, 0); 1068 rt = __ip_route_output_key(net, &fl4); 1069 if (!IS_ERR(rt)) { 1070 __ip_rt_update_pmtu(rt, &fl4, mtu); 1071 ip_rt_put(rt); 1072 } 1073 } 1074 EXPORT_SYMBOL_GPL(ipv4_update_pmtu); 1075 1076 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1077 { 1078 const struct iphdr *iph = (const struct iphdr *)skb->data; 1079 struct flowi4 fl4; 1080 struct rtable *rt; 1081 1082 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0); 1083 1084 if (!fl4.flowi4_mark) 1085 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark); 1086 1087 rt = __ip_route_output_key(sock_net(sk), &fl4); 1088 if (!IS_ERR(rt)) { 1089 __ip_rt_update_pmtu(rt, &fl4, mtu); 1090 ip_rt_put(rt); 1091 } 1092 } 1093 1094 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1095 { 1096 const struct iphdr *iph = (const struct iphdr *)skb->data; 1097 struct flowi4 fl4; 1098 struct rtable *rt; 1099 struct dst_entry *odst = NULL; 1100 bool new = false; 1101 struct net *net = sock_net(sk); 1102 1103 bh_lock_sock(sk); 1104 1105 if (!ip_sk_accept_pmtu(sk)) 1106 goto out; 1107 1108 odst = sk_dst_get(sk); 1109 1110 if (sock_owned_by_user(sk) || !odst) { 1111 __ipv4_sk_update_pmtu(skb, sk, mtu); 1112 goto out; 1113 } 1114 1115 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1116 1117 rt = (struct rtable *)odst; 1118 if (odst->obsolete && !odst->ops->check(odst, 0)) { 1119 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1120 if (IS_ERR(rt)) 1121 goto out; 1122 1123 new = true; 1124 } 1125 1126 __ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu); 1127 1128 if (!dst_check(&rt->dst, 0)) { 1129 if (new) 1130 dst_release(&rt->dst); 1131 1132 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1133 if (IS_ERR(rt)) 1134 goto out; 1135 1136 new = true; 1137 } 1138 1139 if (new) 1140 sk_dst_set(sk, &rt->dst); 1141 1142 out: 1143 bh_unlock_sock(sk); 1144 dst_release(odst); 1145 } 1146 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu); 1147 1148 void ipv4_redirect(struct sk_buff *skb, struct net *net, 1149 int oif, u8 protocol) 1150 { 1151 const struct iphdr *iph = (const struct iphdr *)skb->data; 1152 struct flowi4 fl4; 1153 struct rtable *rt; 1154 1155 __build_flow_key(net, &fl4, NULL, iph, oif, 1156 RT_TOS(iph->tos), protocol, 0, 0); 1157 rt = __ip_route_output_key(net, &fl4); 1158 if (!IS_ERR(rt)) { 1159 __ip_do_redirect(rt, skb, &fl4, false); 1160 ip_rt_put(rt); 1161 } 1162 } 1163 EXPORT_SYMBOL_GPL(ipv4_redirect); 1164 1165 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk) 1166 { 1167 const struct iphdr *iph = (const struct iphdr *)skb->data; 1168 struct flowi4 fl4; 1169 struct rtable *rt; 1170 struct net *net = sock_net(sk); 1171 1172 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1173 rt = __ip_route_output_key(net, &fl4); 1174 if (!IS_ERR(rt)) { 1175 __ip_do_redirect(rt, skb, &fl4, false); 1176 ip_rt_put(rt); 1177 } 1178 } 1179 EXPORT_SYMBOL_GPL(ipv4_sk_redirect); 1180 1181 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst, 1182 u32 cookie) 1183 { 1184 struct rtable *rt = (struct rtable *) dst; 1185 1186 /* All IPV4 dsts are created with ->obsolete set to the value 1187 * DST_OBSOLETE_FORCE_CHK which forces validation calls down 1188 * into this function always. 1189 * 1190 * When a PMTU/redirect information update invalidates a route, 1191 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or 1192 * DST_OBSOLETE_DEAD. 1193 */ 1194 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt)) 1195 return NULL; 1196 return dst; 1197 } 1198 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check); 1199 1200 static void ipv4_send_dest_unreach(struct sk_buff *skb) 1201 { 1202 struct ip_options opt; 1203 int res; 1204 1205 /* Recompile ip options since IPCB may not be valid anymore. 1206 * Also check we have a reasonable ipv4 header. 1207 */ 1208 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) || 1209 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5) 1210 return; 1211 1212 memset(&opt, 0, sizeof(opt)); 1213 if (ip_hdr(skb)->ihl > 5) { 1214 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4)) 1215 return; 1216 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr); 1217 1218 rcu_read_lock(); 1219 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL); 1220 rcu_read_unlock(); 1221 1222 if (res) 1223 return; 1224 } 1225 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt); 1226 } 1227 1228 static void ipv4_link_failure(struct sk_buff *skb) 1229 { 1230 struct rtable *rt; 1231 1232 ipv4_send_dest_unreach(skb); 1233 1234 rt = skb_rtable(skb); 1235 if (rt) 1236 dst_set_expires(&rt->dst, 0); 1237 } 1238 1239 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb) 1240 { 1241 pr_debug("%s: %pI4 -> %pI4, %s\n", 1242 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1243 skb->dev ? skb->dev->name : "?"); 1244 kfree_skb(skb); 1245 WARN_ON(1); 1246 return 0; 1247 } 1248 1249 /* 1250 * We do not cache source address of outgoing interface, 1251 * because it is used only by IP RR, TS and SRR options, 1252 * so that it out of fast path. 1253 * 1254 * BTW remember: "addr" is allowed to be not aligned 1255 * in IP options! 1256 */ 1257 1258 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt) 1259 { 1260 __be32 src; 1261 1262 if (rt_is_output_route(rt)) 1263 src = ip_hdr(skb)->saddr; 1264 else { 1265 struct fib_result res; 1266 struct iphdr *iph = ip_hdr(skb); 1267 struct flowi4 fl4 = { 1268 .daddr = iph->daddr, 1269 .saddr = iph->saddr, 1270 .flowi4_tos = RT_TOS(iph->tos), 1271 .flowi4_oif = rt->dst.dev->ifindex, 1272 .flowi4_iif = skb->dev->ifindex, 1273 .flowi4_mark = skb->mark, 1274 }; 1275 1276 rcu_read_lock(); 1277 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0) 1278 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res); 1279 else 1280 src = inet_select_addr(rt->dst.dev, 1281 rt_nexthop(rt, iph->daddr), 1282 RT_SCOPE_UNIVERSE); 1283 rcu_read_unlock(); 1284 } 1285 memcpy(addr, &src, 4); 1286 } 1287 1288 #ifdef CONFIG_IP_ROUTE_CLASSID 1289 static void set_class_tag(struct rtable *rt, u32 tag) 1290 { 1291 if (!(rt->dst.tclassid & 0xFFFF)) 1292 rt->dst.tclassid |= tag & 0xFFFF; 1293 if (!(rt->dst.tclassid & 0xFFFF0000)) 1294 rt->dst.tclassid |= tag & 0xFFFF0000; 1295 } 1296 #endif 1297 1298 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1299 { 1300 struct net *net = dev_net(dst->dev); 1301 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr); 1302 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size, 1303 net->ipv4.ip_rt_min_advmss); 1304 1305 return min(advmss, IPV4_MAX_PMTU - header_size); 1306 } 1307 1308 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst) 1309 { 1310 return ip_dst_mtu_maybe_forward(dst, false); 1311 } 1312 EXPORT_INDIRECT_CALLABLE(ipv4_mtu); 1313 1314 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr) 1315 { 1316 struct fnhe_hash_bucket *hash; 1317 struct fib_nh_exception *fnhe, __rcu **fnhe_p; 1318 u32 hval = fnhe_hashfun(daddr); 1319 1320 spin_lock_bh(&fnhe_lock); 1321 1322 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1323 lockdep_is_held(&fnhe_lock)); 1324 hash += hval; 1325 1326 fnhe_p = &hash->chain; 1327 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock)); 1328 while (fnhe) { 1329 if (fnhe->fnhe_daddr == daddr) { 1330 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected( 1331 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock))); 1332 /* set fnhe_daddr to 0 to ensure it won't bind with 1333 * new dsts in rt_bind_exception(). 1334 */ 1335 fnhe->fnhe_daddr = 0; 1336 fnhe_flush_routes(fnhe); 1337 kfree_rcu(fnhe, rcu); 1338 break; 1339 } 1340 fnhe_p = &fnhe->fnhe_next; 1341 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1342 lockdep_is_held(&fnhe_lock)); 1343 } 1344 1345 spin_unlock_bh(&fnhe_lock); 1346 } 1347 1348 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc, 1349 __be32 daddr) 1350 { 1351 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions); 1352 struct fib_nh_exception *fnhe; 1353 u32 hval; 1354 1355 if (!hash) 1356 return NULL; 1357 1358 hval = fnhe_hashfun(daddr); 1359 1360 for (fnhe = rcu_dereference(hash[hval].chain); fnhe; 1361 fnhe = rcu_dereference(fnhe->fnhe_next)) { 1362 if (fnhe->fnhe_daddr == daddr) { 1363 if (fnhe->fnhe_expires && 1364 time_after(jiffies, fnhe->fnhe_expires)) { 1365 ip_del_fnhe(nhc, daddr); 1366 break; 1367 } 1368 return fnhe; 1369 } 1370 } 1371 return NULL; 1372 } 1373 1374 /* MTU selection: 1375 * 1. mtu on route is locked - use it 1376 * 2. mtu from nexthop exception 1377 * 3. mtu from egress device 1378 */ 1379 1380 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr) 1381 { 1382 struct fib_nh_common *nhc = res->nhc; 1383 struct net_device *dev = nhc->nhc_dev; 1384 struct fib_info *fi = res->fi; 1385 u32 mtu = 0; 1386 1387 if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu || 1388 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU)) 1389 mtu = fi->fib_mtu; 1390 1391 if (likely(!mtu)) { 1392 struct fib_nh_exception *fnhe; 1393 1394 fnhe = find_exception(nhc, daddr); 1395 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires)) 1396 mtu = fnhe->fnhe_pmtu; 1397 } 1398 1399 if (likely(!mtu)) 1400 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU); 1401 1402 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu); 1403 } 1404 1405 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, 1406 __be32 daddr, const bool do_cache) 1407 { 1408 bool ret = false; 1409 1410 spin_lock_bh(&fnhe_lock); 1411 1412 if (daddr == fnhe->fnhe_daddr) { 1413 struct rtable __rcu **porig; 1414 struct rtable *orig; 1415 int genid = fnhe_genid(dev_net(rt->dst.dev)); 1416 1417 if (rt_is_input_route(rt)) 1418 porig = &fnhe->fnhe_rth_input; 1419 else 1420 porig = &fnhe->fnhe_rth_output; 1421 orig = rcu_dereference(*porig); 1422 1423 if (fnhe->fnhe_genid != genid) { 1424 fnhe->fnhe_genid = genid; 1425 fnhe->fnhe_gw = 0; 1426 fnhe->fnhe_pmtu = 0; 1427 fnhe->fnhe_expires = 0; 1428 fnhe->fnhe_mtu_locked = false; 1429 fnhe_flush_routes(fnhe); 1430 orig = NULL; 1431 } 1432 fill_route_from_fnhe(rt, fnhe); 1433 if (!rt->rt_gw4) { 1434 rt->rt_gw4 = daddr; 1435 rt->rt_gw_family = AF_INET; 1436 } 1437 1438 if (do_cache) { 1439 dst_hold(&rt->dst); 1440 rcu_assign_pointer(*porig, rt); 1441 if (orig) { 1442 dst_dev_put(&orig->dst); 1443 dst_release(&orig->dst); 1444 } 1445 ret = true; 1446 } 1447 1448 fnhe->fnhe_stamp = jiffies; 1449 } 1450 spin_unlock_bh(&fnhe_lock); 1451 1452 return ret; 1453 } 1454 1455 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt) 1456 { 1457 struct rtable *orig, *prev, **p; 1458 bool ret = true; 1459 1460 if (rt_is_input_route(rt)) { 1461 p = (struct rtable **)&nhc->nhc_rth_input; 1462 } else { 1463 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 1464 } 1465 orig = *p; 1466 1467 /* hold dst before doing cmpxchg() to avoid race condition 1468 * on this dst 1469 */ 1470 dst_hold(&rt->dst); 1471 prev = cmpxchg(p, orig, rt); 1472 if (prev == orig) { 1473 if (orig) { 1474 rt_add_uncached_list(orig); 1475 dst_release(&orig->dst); 1476 } 1477 } else { 1478 dst_release(&rt->dst); 1479 ret = false; 1480 } 1481 1482 return ret; 1483 } 1484 1485 struct uncached_list { 1486 spinlock_t lock; 1487 struct list_head head; 1488 struct list_head quarantine; 1489 }; 1490 1491 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list); 1492 1493 void rt_add_uncached_list(struct rtable *rt) 1494 { 1495 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list); 1496 1497 rt->rt_uncached_list = ul; 1498 1499 spin_lock_bh(&ul->lock); 1500 list_add_tail(&rt->rt_uncached, &ul->head); 1501 spin_unlock_bh(&ul->lock); 1502 } 1503 1504 void rt_del_uncached_list(struct rtable *rt) 1505 { 1506 if (!list_empty(&rt->rt_uncached)) { 1507 struct uncached_list *ul = rt->rt_uncached_list; 1508 1509 spin_lock_bh(&ul->lock); 1510 list_del_init(&rt->rt_uncached); 1511 spin_unlock_bh(&ul->lock); 1512 } 1513 } 1514 1515 static void ipv4_dst_destroy(struct dst_entry *dst) 1516 { 1517 struct rtable *rt = (struct rtable *)dst; 1518 1519 ip_dst_metrics_put(dst); 1520 rt_del_uncached_list(rt); 1521 } 1522 1523 void rt_flush_dev(struct net_device *dev) 1524 { 1525 struct rtable *rt, *safe; 1526 int cpu; 1527 1528 for_each_possible_cpu(cpu) { 1529 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 1530 1531 if (list_empty(&ul->head)) 1532 continue; 1533 1534 spin_lock_bh(&ul->lock); 1535 list_for_each_entry_safe(rt, safe, &ul->head, rt_uncached) { 1536 if (rt->dst.dev != dev) 1537 continue; 1538 rt->dst.dev = blackhole_netdev; 1539 dev_replace_track(dev, blackhole_netdev, 1540 &rt->dst.dev_tracker, 1541 GFP_ATOMIC); 1542 list_move(&rt->rt_uncached, &ul->quarantine); 1543 } 1544 spin_unlock_bh(&ul->lock); 1545 } 1546 } 1547 1548 static bool rt_cache_valid(const struct rtable *rt) 1549 { 1550 return rt && 1551 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK && 1552 !rt_is_expired(rt); 1553 } 1554 1555 static void rt_set_nexthop(struct rtable *rt, __be32 daddr, 1556 const struct fib_result *res, 1557 struct fib_nh_exception *fnhe, 1558 struct fib_info *fi, u16 type, u32 itag, 1559 const bool do_cache) 1560 { 1561 bool cached = false; 1562 1563 if (fi) { 1564 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1565 1566 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) { 1567 rt->rt_uses_gateway = 1; 1568 rt->rt_gw_family = nhc->nhc_gw_family; 1569 /* only INET and INET6 are supported */ 1570 if (likely(nhc->nhc_gw_family == AF_INET)) 1571 rt->rt_gw4 = nhc->nhc_gw.ipv4; 1572 else 1573 rt->rt_gw6 = nhc->nhc_gw.ipv6; 1574 } 1575 1576 ip_dst_init_metrics(&rt->dst, fi->fib_metrics); 1577 1578 #ifdef CONFIG_IP_ROUTE_CLASSID 1579 if (nhc->nhc_family == AF_INET) { 1580 struct fib_nh *nh; 1581 1582 nh = container_of(nhc, struct fib_nh, nh_common); 1583 rt->dst.tclassid = nh->nh_tclassid; 1584 } 1585 #endif 1586 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 1587 if (unlikely(fnhe)) 1588 cached = rt_bind_exception(rt, fnhe, daddr, do_cache); 1589 else if (do_cache) 1590 cached = rt_cache_route(nhc, rt); 1591 if (unlikely(!cached)) { 1592 /* Routes we intend to cache in nexthop exception or 1593 * FIB nexthop have the DST_NOCACHE bit clear. 1594 * However, if we are unsuccessful at storing this 1595 * route into the cache we really need to set it. 1596 */ 1597 if (!rt->rt_gw4) { 1598 rt->rt_gw_family = AF_INET; 1599 rt->rt_gw4 = daddr; 1600 } 1601 rt_add_uncached_list(rt); 1602 } 1603 } else 1604 rt_add_uncached_list(rt); 1605 1606 #ifdef CONFIG_IP_ROUTE_CLASSID 1607 #ifdef CONFIG_IP_MULTIPLE_TABLES 1608 set_class_tag(rt, res->tclassid); 1609 #endif 1610 set_class_tag(rt, itag); 1611 #endif 1612 } 1613 1614 struct rtable *rt_dst_alloc(struct net_device *dev, 1615 unsigned int flags, u16 type, 1616 bool nopolicy, bool noxfrm) 1617 { 1618 struct rtable *rt; 1619 1620 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1621 (nopolicy ? DST_NOPOLICY : 0) | 1622 (noxfrm ? DST_NOXFRM : 0)); 1623 1624 if (rt) { 1625 rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1626 rt->rt_flags = flags; 1627 rt->rt_type = type; 1628 rt->rt_is_input = 0; 1629 rt->rt_iif = 0; 1630 rt->rt_pmtu = 0; 1631 rt->rt_mtu_locked = 0; 1632 rt->rt_uses_gateway = 0; 1633 rt->rt_gw_family = 0; 1634 rt->rt_gw4 = 0; 1635 INIT_LIST_HEAD(&rt->rt_uncached); 1636 1637 rt->dst.output = ip_output; 1638 if (flags & RTCF_LOCAL) 1639 rt->dst.input = ip_local_deliver; 1640 } 1641 1642 return rt; 1643 } 1644 EXPORT_SYMBOL(rt_dst_alloc); 1645 1646 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt) 1647 { 1648 struct rtable *new_rt; 1649 1650 new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1651 rt->dst.flags); 1652 1653 if (new_rt) { 1654 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1655 new_rt->rt_flags = rt->rt_flags; 1656 new_rt->rt_type = rt->rt_type; 1657 new_rt->rt_is_input = rt->rt_is_input; 1658 new_rt->rt_iif = rt->rt_iif; 1659 new_rt->rt_pmtu = rt->rt_pmtu; 1660 new_rt->rt_mtu_locked = rt->rt_mtu_locked; 1661 new_rt->rt_gw_family = rt->rt_gw_family; 1662 if (rt->rt_gw_family == AF_INET) 1663 new_rt->rt_gw4 = rt->rt_gw4; 1664 else if (rt->rt_gw_family == AF_INET6) 1665 new_rt->rt_gw6 = rt->rt_gw6; 1666 INIT_LIST_HEAD(&new_rt->rt_uncached); 1667 1668 new_rt->dst.input = rt->dst.input; 1669 new_rt->dst.output = rt->dst.output; 1670 new_rt->dst.error = rt->dst.error; 1671 new_rt->dst.lastuse = jiffies; 1672 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate); 1673 } 1674 return new_rt; 1675 } 1676 EXPORT_SYMBOL(rt_dst_clone); 1677 1678 /* called in rcu_read_lock() section */ 1679 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1680 u8 tos, struct net_device *dev, 1681 struct in_device *in_dev, u32 *itag) 1682 { 1683 int err; 1684 1685 /* Primary sanity checks. */ 1686 if (!in_dev) 1687 return -EINVAL; 1688 1689 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) || 1690 skb->protocol != htons(ETH_P_IP)) 1691 return -EINVAL; 1692 1693 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev)) 1694 return -EINVAL; 1695 1696 if (ipv4_is_zeronet(saddr)) { 1697 if (!ipv4_is_local_multicast(daddr) && 1698 ip_hdr(skb)->protocol != IPPROTO_IGMP) 1699 return -EINVAL; 1700 } else { 1701 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 1702 in_dev, itag); 1703 if (err < 0) 1704 return err; 1705 } 1706 return 0; 1707 } 1708 1709 /* called in rcu_read_lock() section */ 1710 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1711 u8 tos, struct net_device *dev, int our) 1712 { 1713 struct in_device *in_dev = __in_dev_get_rcu(dev); 1714 unsigned int flags = RTCF_MULTICAST; 1715 struct rtable *rth; 1716 u32 itag = 0; 1717 int err; 1718 1719 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag); 1720 if (err) 1721 return err; 1722 1723 if (our) 1724 flags |= RTCF_LOCAL; 1725 1726 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST, 1727 IN_DEV_ORCONF(in_dev, NOPOLICY), false); 1728 if (!rth) 1729 return -ENOBUFS; 1730 1731 #ifdef CONFIG_IP_ROUTE_CLASSID 1732 rth->dst.tclassid = itag; 1733 #endif 1734 rth->dst.output = ip_rt_bug; 1735 rth->rt_is_input= 1; 1736 1737 #ifdef CONFIG_IP_MROUTE 1738 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1739 rth->dst.input = ip_mr_input; 1740 #endif 1741 RT_CACHE_STAT_INC(in_slow_mc); 1742 1743 skb_dst_set(skb, &rth->dst); 1744 return 0; 1745 } 1746 1747 1748 static void ip_handle_martian_source(struct net_device *dev, 1749 struct in_device *in_dev, 1750 struct sk_buff *skb, 1751 __be32 daddr, 1752 __be32 saddr) 1753 { 1754 RT_CACHE_STAT_INC(in_martian_src); 1755 #ifdef CONFIG_IP_ROUTE_VERBOSE 1756 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1757 /* 1758 * RFC1812 recommendation, if source is martian, 1759 * the only hint is MAC header. 1760 */ 1761 pr_warn("martian source %pI4 from %pI4, on dev %s\n", 1762 &daddr, &saddr, dev->name); 1763 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1764 print_hex_dump(KERN_WARNING, "ll header: ", 1765 DUMP_PREFIX_OFFSET, 16, 1, 1766 skb_mac_header(skb), 1767 dev->hard_header_len, false); 1768 } 1769 } 1770 #endif 1771 } 1772 1773 /* called in rcu_read_lock() section */ 1774 static int __mkroute_input(struct sk_buff *skb, 1775 const struct fib_result *res, 1776 struct in_device *in_dev, 1777 __be32 daddr, __be32 saddr, u32 tos) 1778 { 1779 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1780 struct net_device *dev = nhc->nhc_dev; 1781 struct fib_nh_exception *fnhe; 1782 struct rtable *rth; 1783 int err; 1784 struct in_device *out_dev; 1785 bool do_cache; 1786 u32 itag = 0; 1787 1788 /* get a working reference to the output device */ 1789 out_dev = __in_dev_get_rcu(dev); 1790 if (!out_dev) { 1791 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n"); 1792 return -EINVAL; 1793 } 1794 1795 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res), 1796 in_dev->dev, in_dev, &itag); 1797 if (err < 0) { 1798 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1799 saddr); 1800 1801 goto cleanup; 1802 } 1803 1804 do_cache = res->fi && !itag; 1805 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) && 1806 skb->protocol == htons(ETH_P_IP)) { 1807 __be32 gw; 1808 1809 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0; 1810 if (IN_DEV_SHARED_MEDIA(out_dev) || 1811 inet_addr_onlink(out_dev, saddr, gw)) 1812 IPCB(skb)->flags |= IPSKB_DOREDIRECT; 1813 } 1814 1815 if (skb->protocol != htons(ETH_P_IP)) { 1816 /* Not IP (i.e. ARP). Do not create route, if it is 1817 * invalid for proxy arp. DNAT routes are always valid. 1818 * 1819 * Proxy arp feature have been extended to allow, ARP 1820 * replies back to the same interface, to support 1821 * Private VLAN switch technologies. See arp.c. 1822 */ 1823 if (out_dev == in_dev && 1824 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 1825 err = -EINVAL; 1826 goto cleanup; 1827 } 1828 } 1829 1830 fnhe = find_exception(nhc, daddr); 1831 if (do_cache) { 1832 if (fnhe) 1833 rth = rcu_dereference(fnhe->fnhe_rth_input); 1834 else 1835 rth = rcu_dereference(nhc->nhc_rth_input); 1836 if (rt_cache_valid(rth)) { 1837 skb_dst_set_noref(skb, &rth->dst); 1838 goto out; 1839 } 1840 } 1841 1842 rth = rt_dst_alloc(out_dev->dev, 0, res->type, 1843 IN_DEV_ORCONF(in_dev, NOPOLICY), 1844 IN_DEV_ORCONF(out_dev, NOXFRM)); 1845 if (!rth) { 1846 err = -ENOBUFS; 1847 goto cleanup; 1848 } 1849 1850 rth->rt_is_input = 1; 1851 RT_CACHE_STAT_INC(in_slow_tot); 1852 1853 rth->dst.input = ip_forward; 1854 1855 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag, 1856 do_cache); 1857 lwtunnel_set_redirect(&rth->dst); 1858 skb_dst_set(skb, &rth->dst); 1859 out: 1860 err = 0; 1861 cleanup: 1862 return err; 1863 } 1864 1865 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1866 /* To make ICMP packets follow the right flow, the multipath hash is 1867 * calculated from the inner IP addresses. 1868 */ 1869 static void ip_multipath_l3_keys(const struct sk_buff *skb, 1870 struct flow_keys *hash_keys) 1871 { 1872 const struct iphdr *outer_iph = ip_hdr(skb); 1873 const struct iphdr *key_iph = outer_iph; 1874 const struct iphdr *inner_iph; 1875 const struct icmphdr *icmph; 1876 struct iphdr _inner_iph; 1877 struct icmphdr _icmph; 1878 1879 if (likely(outer_iph->protocol != IPPROTO_ICMP)) 1880 goto out; 1881 1882 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0)) 1883 goto out; 1884 1885 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph), 1886 &_icmph); 1887 if (!icmph) 1888 goto out; 1889 1890 if (!icmp_is_err(icmph->type)) 1891 goto out; 1892 1893 inner_iph = skb_header_pointer(skb, 1894 outer_iph->ihl * 4 + sizeof(_icmph), 1895 sizeof(_inner_iph), &_inner_iph); 1896 if (!inner_iph) 1897 goto out; 1898 1899 key_iph = inner_iph; 1900 out: 1901 hash_keys->addrs.v4addrs.src = key_iph->saddr; 1902 hash_keys->addrs.v4addrs.dst = key_iph->daddr; 1903 } 1904 1905 static u32 fib_multipath_custom_hash_outer(const struct net *net, 1906 const struct sk_buff *skb, 1907 bool *p_has_inner) 1908 { 1909 u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields; 1910 struct flow_keys keys, hash_keys; 1911 1912 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 1913 return 0; 1914 1915 memset(&hash_keys, 0, sizeof(hash_keys)); 1916 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP); 1917 1918 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1919 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 1920 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 1921 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 1922 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 1923 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 1924 hash_keys.basic.ip_proto = keys.basic.ip_proto; 1925 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) 1926 hash_keys.ports.src = keys.ports.src; 1927 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 1928 hash_keys.ports.dst = keys.ports.dst; 1929 1930 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION); 1931 return flow_hash_from_keys(&hash_keys); 1932 } 1933 1934 static u32 fib_multipath_custom_hash_inner(const struct net *net, 1935 const struct sk_buff *skb, 1936 bool has_inner) 1937 { 1938 u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields; 1939 struct flow_keys keys, hash_keys; 1940 1941 /* We assume the packet carries an encapsulation, but if none was 1942 * encountered during dissection of the outer flow, then there is no 1943 * point in calling the flow dissector again. 1944 */ 1945 if (!has_inner) 1946 return 0; 1947 1948 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK)) 1949 return 0; 1950 1951 memset(&hash_keys, 0, sizeof(hash_keys)); 1952 skb_flow_dissect_flow_keys(skb, &keys, 0); 1953 1954 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION)) 1955 return 0; 1956 1957 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 1958 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1959 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 1960 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 1961 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 1962 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 1963 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 1964 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 1965 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 1966 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 1967 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 1968 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 1969 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL) 1970 hash_keys.tags.flow_label = keys.tags.flow_label; 1971 } 1972 1973 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO) 1974 hash_keys.basic.ip_proto = keys.basic.ip_proto; 1975 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT) 1976 hash_keys.ports.src = keys.ports.src; 1977 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT) 1978 hash_keys.ports.dst = keys.ports.dst; 1979 1980 return flow_hash_from_keys(&hash_keys); 1981 } 1982 1983 static u32 fib_multipath_custom_hash_skb(const struct net *net, 1984 const struct sk_buff *skb) 1985 { 1986 u32 mhash, mhash_inner; 1987 bool has_inner = true; 1988 1989 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner); 1990 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner); 1991 1992 return jhash_2words(mhash, mhash_inner, 0); 1993 } 1994 1995 static u32 fib_multipath_custom_hash_fl4(const struct net *net, 1996 const struct flowi4 *fl4) 1997 { 1998 u32 hash_fields = net->ipv4.sysctl_fib_multipath_hash_fields; 1999 struct flow_keys hash_keys; 2000 2001 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 2002 return 0; 2003 2004 memset(&hash_keys, 0, sizeof(hash_keys)); 2005 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2006 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 2007 hash_keys.addrs.v4addrs.src = fl4->saddr; 2008 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 2009 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2010 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 2011 hash_keys.basic.ip_proto = fl4->flowi4_proto; 2012 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) 2013 hash_keys.ports.src = fl4->fl4_sport; 2014 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 2015 hash_keys.ports.dst = fl4->fl4_dport; 2016 2017 return flow_hash_from_keys(&hash_keys); 2018 } 2019 2020 /* if skb is set it will be used and fl4 can be NULL */ 2021 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4, 2022 const struct sk_buff *skb, struct flow_keys *flkeys) 2023 { 2024 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0; 2025 struct flow_keys hash_keys; 2026 u32 mhash = 0; 2027 2028 switch (net->ipv4.sysctl_fib_multipath_hash_policy) { 2029 case 0: 2030 memset(&hash_keys, 0, sizeof(hash_keys)); 2031 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2032 if (skb) { 2033 ip_multipath_l3_keys(skb, &hash_keys); 2034 } else { 2035 hash_keys.addrs.v4addrs.src = fl4->saddr; 2036 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2037 } 2038 mhash = flow_hash_from_keys(&hash_keys); 2039 break; 2040 case 1: 2041 /* skb is currently provided only when forwarding */ 2042 if (skb) { 2043 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 2044 struct flow_keys keys; 2045 2046 /* short-circuit if we already have L4 hash present */ 2047 if (skb->l4_hash) 2048 return skb_get_hash_raw(skb) >> 1; 2049 2050 memset(&hash_keys, 0, sizeof(hash_keys)); 2051 2052 if (!flkeys) { 2053 skb_flow_dissect_flow_keys(skb, &keys, flag); 2054 flkeys = &keys; 2055 } 2056 2057 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2058 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src; 2059 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst; 2060 hash_keys.ports.src = flkeys->ports.src; 2061 hash_keys.ports.dst = flkeys->ports.dst; 2062 hash_keys.basic.ip_proto = flkeys->basic.ip_proto; 2063 } else { 2064 memset(&hash_keys, 0, sizeof(hash_keys)); 2065 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2066 hash_keys.addrs.v4addrs.src = fl4->saddr; 2067 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2068 hash_keys.ports.src = fl4->fl4_sport; 2069 hash_keys.ports.dst = fl4->fl4_dport; 2070 hash_keys.basic.ip_proto = fl4->flowi4_proto; 2071 } 2072 mhash = flow_hash_from_keys(&hash_keys); 2073 break; 2074 case 2: 2075 memset(&hash_keys, 0, sizeof(hash_keys)); 2076 /* skb is currently provided only when forwarding */ 2077 if (skb) { 2078 struct flow_keys keys; 2079 2080 skb_flow_dissect_flow_keys(skb, &keys, 0); 2081 /* Inner can be v4 or v6 */ 2082 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 2083 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2084 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 2085 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 2086 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 2087 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 2088 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 2089 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 2090 hash_keys.tags.flow_label = keys.tags.flow_label; 2091 hash_keys.basic.ip_proto = keys.basic.ip_proto; 2092 } else { 2093 /* Same as case 0 */ 2094 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2095 ip_multipath_l3_keys(skb, &hash_keys); 2096 } 2097 } else { 2098 /* Same as case 0 */ 2099 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2100 hash_keys.addrs.v4addrs.src = fl4->saddr; 2101 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2102 } 2103 mhash = flow_hash_from_keys(&hash_keys); 2104 break; 2105 case 3: 2106 if (skb) 2107 mhash = fib_multipath_custom_hash_skb(net, skb); 2108 else 2109 mhash = fib_multipath_custom_hash_fl4(net, fl4); 2110 break; 2111 } 2112 2113 if (multipath_hash) 2114 mhash = jhash_2words(mhash, multipath_hash, 0); 2115 2116 return mhash >> 1; 2117 } 2118 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 2119 2120 static int ip_mkroute_input(struct sk_buff *skb, 2121 struct fib_result *res, 2122 struct in_device *in_dev, 2123 __be32 daddr, __be32 saddr, u32 tos, 2124 struct flow_keys *hkeys) 2125 { 2126 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2127 if (res->fi && fib_info_num_path(res->fi) > 1) { 2128 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys); 2129 2130 fib_select_multipath(res, h); 2131 } 2132 #endif 2133 2134 /* create a routing cache entry */ 2135 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos); 2136 } 2137 2138 /* Implements all the saddr-related checks as ip_route_input_slow(), 2139 * assuming daddr is valid and the destination is not a local broadcast one. 2140 * Uses the provided hint instead of performing a route lookup. 2141 */ 2142 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2143 u8 tos, struct net_device *dev, 2144 const struct sk_buff *hint) 2145 { 2146 struct in_device *in_dev = __in_dev_get_rcu(dev); 2147 struct rtable *rt = skb_rtable(hint); 2148 struct net *net = dev_net(dev); 2149 int err = -EINVAL; 2150 u32 tag = 0; 2151 2152 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 2153 goto martian_source; 2154 2155 if (ipv4_is_zeronet(saddr)) 2156 goto martian_source; 2157 2158 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2159 goto martian_source; 2160 2161 if (rt->rt_type != RTN_LOCAL) 2162 goto skip_validate_source; 2163 2164 tos &= IPTOS_RT_MASK; 2165 err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag); 2166 if (err < 0) 2167 goto martian_source; 2168 2169 skip_validate_source: 2170 skb_dst_copy(skb, hint); 2171 return 0; 2172 2173 martian_source: 2174 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2175 return err; 2176 } 2177 2178 /* get device for dst_alloc with local routes */ 2179 static struct net_device *ip_rt_get_dev(struct net *net, 2180 const struct fib_result *res) 2181 { 2182 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL; 2183 struct net_device *dev = NULL; 2184 2185 if (nhc) 2186 dev = l3mdev_master_dev_rcu(nhc->nhc_dev); 2187 2188 return dev ? : net->loopback_dev; 2189 } 2190 2191 /* 2192 * NOTE. We drop all the packets that has local source 2193 * addresses, because every properly looped back packet 2194 * must have correct destination already attached by output routine. 2195 * Changes in the enforced policies must be applied also to 2196 * ip_route_use_hint(). 2197 * 2198 * Such approach solves two big problems: 2199 * 1. Not simplex devices are handled properly. 2200 * 2. IP spoofing attempts are filtered with 100% of guarantee. 2201 * called with rcu_read_lock() 2202 */ 2203 2204 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2205 u8 tos, struct net_device *dev, 2206 struct fib_result *res) 2207 { 2208 struct in_device *in_dev = __in_dev_get_rcu(dev); 2209 struct flow_keys *flkeys = NULL, _flkeys; 2210 struct net *net = dev_net(dev); 2211 struct ip_tunnel_info *tun_info; 2212 int err = -EINVAL; 2213 unsigned int flags = 0; 2214 u32 itag = 0; 2215 struct rtable *rth; 2216 struct flowi4 fl4; 2217 bool do_cache = true; 2218 2219 /* IP on this device is disabled. */ 2220 2221 if (!in_dev) 2222 goto out; 2223 2224 /* Check for the most weird martians, which can be not detected 2225 * by fib_lookup. 2226 */ 2227 2228 tun_info = skb_tunnel_info(skb); 2229 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX)) 2230 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id; 2231 else 2232 fl4.flowi4_tun_key.tun_id = 0; 2233 skb_dst_drop(skb); 2234 2235 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 2236 goto martian_source; 2237 2238 res->fi = NULL; 2239 res->table = NULL; 2240 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 2241 goto brd_input; 2242 2243 /* Accept zero addresses only to limited broadcast; 2244 * I even do not know to fix it or not. Waiting for complains :-) 2245 */ 2246 if (ipv4_is_zeronet(saddr)) 2247 goto martian_source; 2248 2249 if (ipv4_is_zeronet(daddr)) 2250 goto martian_destination; 2251 2252 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(), 2253 * and call it once if daddr or/and saddr are loopback addresses 2254 */ 2255 if (ipv4_is_loopback(daddr)) { 2256 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2257 goto martian_destination; 2258 } else if (ipv4_is_loopback(saddr)) { 2259 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2260 goto martian_source; 2261 } 2262 2263 /* 2264 * Now we are ready to route packet. 2265 */ 2266 fl4.flowi4_oif = 0; 2267 fl4.flowi4_iif = dev->ifindex; 2268 fl4.flowi4_mark = skb->mark; 2269 fl4.flowi4_tos = tos; 2270 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 2271 fl4.flowi4_flags = 0; 2272 fl4.daddr = daddr; 2273 fl4.saddr = saddr; 2274 fl4.flowi4_uid = sock_net_uid(net, NULL); 2275 fl4.flowi4_multipath_hash = 0; 2276 2277 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) { 2278 flkeys = &_flkeys; 2279 } else { 2280 fl4.flowi4_proto = 0; 2281 fl4.fl4_sport = 0; 2282 fl4.fl4_dport = 0; 2283 } 2284 2285 err = fib_lookup(net, &fl4, res, 0); 2286 if (err != 0) { 2287 if (!IN_DEV_FORWARD(in_dev)) 2288 err = -EHOSTUNREACH; 2289 goto no_route; 2290 } 2291 2292 if (res->type == RTN_BROADCAST) { 2293 if (IN_DEV_BFORWARD(in_dev)) 2294 goto make_route; 2295 /* not do cache if bc_forwarding is enabled */ 2296 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING)) 2297 do_cache = false; 2298 goto brd_input; 2299 } 2300 2301 if (res->type == RTN_LOCAL) { 2302 err = fib_validate_source(skb, saddr, daddr, tos, 2303 0, dev, in_dev, &itag); 2304 if (err < 0) 2305 goto martian_source; 2306 goto local_input; 2307 } 2308 2309 if (!IN_DEV_FORWARD(in_dev)) { 2310 err = -EHOSTUNREACH; 2311 goto no_route; 2312 } 2313 if (res->type != RTN_UNICAST) 2314 goto martian_destination; 2315 2316 make_route: 2317 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys); 2318 out: return err; 2319 2320 brd_input: 2321 if (skb->protocol != htons(ETH_P_IP)) 2322 goto e_inval; 2323 2324 if (!ipv4_is_zeronet(saddr)) { 2325 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 2326 in_dev, &itag); 2327 if (err < 0) 2328 goto martian_source; 2329 } 2330 flags |= RTCF_BROADCAST; 2331 res->type = RTN_BROADCAST; 2332 RT_CACHE_STAT_INC(in_brd); 2333 2334 local_input: 2335 do_cache &= res->fi && !itag; 2336 if (do_cache) { 2337 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2338 2339 rth = rcu_dereference(nhc->nhc_rth_input); 2340 if (rt_cache_valid(rth)) { 2341 skb_dst_set_noref(skb, &rth->dst); 2342 err = 0; 2343 goto out; 2344 } 2345 } 2346 2347 rth = rt_dst_alloc(ip_rt_get_dev(net, res), 2348 flags | RTCF_LOCAL, res->type, 2349 IN_DEV_ORCONF(in_dev, NOPOLICY), false); 2350 if (!rth) 2351 goto e_nobufs; 2352 2353 rth->dst.output= ip_rt_bug; 2354 #ifdef CONFIG_IP_ROUTE_CLASSID 2355 rth->dst.tclassid = itag; 2356 #endif 2357 rth->rt_is_input = 1; 2358 2359 RT_CACHE_STAT_INC(in_slow_tot); 2360 if (res->type == RTN_UNREACHABLE) { 2361 rth->dst.input= ip_error; 2362 rth->dst.error= -err; 2363 rth->rt_flags &= ~RTCF_LOCAL; 2364 } 2365 2366 if (do_cache) { 2367 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2368 2369 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 2370 if (lwtunnel_input_redirect(rth->dst.lwtstate)) { 2371 WARN_ON(rth->dst.input == lwtunnel_input); 2372 rth->dst.lwtstate->orig_input = rth->dst.input; 2373 rth->dst.input = lwtunnel_input; 2374 } 2375 2376 if (unlikely(!rt_cache_route(nhc, rth))) 2377 rt_add_uncached_list(rth); 2378 } 2379 skb_dst_set(skb, &rth->dst); 2380 err = 0; 2381 goto out; 2382 2383 no_route: 2384 RT_CACHE_STAT_INC(in_no_route); 2385 res->type = RTN_UNREACHABLE; 2386 res->fi = NULL; 2387 res->table = NULL; 2388 goto local_input; 2389 2390 /* 2391 * Do not cache martian addresses: they should be logged (RFC1812) 2392 */ 2393 martian_destination: 2394 RT_CACHE_STAT_INC(in_martian_dst); 2395 #ifdef CONFIG_IP_ROUTE_VERBOSE 2396 if (IN_DEV_LOG_MARTIANS(in_dev)) 2397 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n", 2398 &daddr, &saddr, dev->name); 2399 #endif 2400 2401 e_inval: 2402 err = -EINVAL; 2403 goto out; 2404 2405 e_nobufs: 2406 err = -ENOBUFS; 2407 goto out; 2408 2409 martian_source: 2410 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2411 goto out; 2412 } 2413 2414 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2415 u8 tos, struct net_device *dev) 2416 { 2417 struct fib_result res; 2418 int err; 2419 2420 tos &= IPTOS_RT_MASK; 2421 rcu_read_lock(); 2422 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res); 2423 rcu_read_unlock(); 2424 2425 return err; 2426 } 2427 EXPORT_SYMBOL(ip_route_input_noref); 2428 2429 /* called with rcu_read_lock held */ 2430 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2431 u8 tos, struct net_device *dev, struct fib_result *res) 2432 { 2433 /* Multicast recognition logic is moved from route cache to here. 2434 * The problem was that too many Ethernet cards have broken/missing 2435 * hardware multicast filters :-( As result the host on multicasting 2436 * network acquires a lot of useless route cache entries, sort of 2437 * SDR messages from all the world. Now we try to get rid of them. 2438 * Really, provided software IP multicast filter is organized 2439 * reasonably (at least, hashed), it does not result in a slowdown 2440 * comparing with route cache reject entries. 2441 * Note, that multicast routers are not affected, because 2442 * route cache entry is created eventually. 2443 */ 2444 if (ipv4_is_multicast(daddr)) { 2445 struct in_device *in_dev = __in_dev_get_rcu(dev); 2446 int our = 0; 2447 int err = -EINVAL; 2448 2449 if (!in_dev) 2450 return err; 2451 our = ip_check_mc_rcu(in_dev, daddr, saddr, 2452 ip_hdr(skb)->protocol); 2453 2454 /* check l3 master if no match yet */ 2455 if (!our && netif_is_l3_slave(dev)) { 2456 struct in_device *l3_in_dev; 2457 2458 l3_in_dev = __in_dev_get_rcu(skb->dev); 2459 if (l3_in_dev) 2460 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr, 2461 ip_hdr(skb)->protocol); 2462 } 2463 2464 if (our 2465 #ifdef CONFIG_IP_MROUTE 2466 || 2467 (!ipv4_is_local_multicast(daddr) && 2468 IN_DEV_MFORWARD(in_dev)) 2469 #endif 2470 ) { 2471 err = ip_route_input_mc(skb, daddr, saddr, 2472 tos, dev, our); 2473 } 2474 return err; 2475 } 2476 2477 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res); 2478 } 2479 2480 /* called with rcu_read_lock() */ 2481 static struct rtable *__mkroute_output(const struct fib_result *res, 2482 const struct flowi4 *fl4, int orig_oif, 2483 struct net_device *dev_out, 2484 unsigned int flags) 2485 { 2486 struct fib_info *fi = res->fi; 2487 struct fib_nh_exception *fnhe; 2488 struct in_device *in_dev; 2489 u16 type = res->type; 2490 struct rtable *rth; 2491 bool do_cache; 2492 2493 in_dev = __in_dev_get_rcu(dev_out); 2494 if (!in_dev) 2495 return ERR_PTR(-EINVAL); 2496 2497 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 2498 if (ipv4_is_loopback(fl4->saddr) && 2499 !(dev_out->flags & IFF_LOOPBACK) && 2500 !netif_is_l3_master(dev_out)) 2501 return ERR_PTR(-EINVAL); 2502 2503 if (ipv4_is_lbcast(fl4->daddr)) 2504 type = RTN_BROADCAST; 2505 else if (ipv4_is_multicast(fl4->daddr)) 2506 type = RTN_MULTICAST; 2507 else if (ipv4_is_zeronet(fl4->daddr)) 2508 return ERR_PTR(-EINVAL); 2509 2510 if (dev_out->flags & IFF_LOOPBACK) 2511 flags |= RTCF_LOCAL; 2512 2513 do_cache = true; 2514 if (type == RTN_BROADCAST) { 2515 flags |= RTCF_BROADCAST | RTCF_LOCAL; 2516 fi = NULL; 2517 } else if (type == RTN_MULTICAST) { 2518 flags |= RTCF_MULTICAST | RTCF_LOCAL; 2519 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr, 2520 fl4->flowi4_proto)) 2521 flags &= ~RTCF_LOCAL; 2522 else 2523 do_cache = false; 2524 /* If multicast route do not exist use 2525 * default one, but do not gateway in this case. 2526 * Yes, it is hack. 2527 */ 2528 if (fi && res->prefixlen < 4) 2529 fi = NULL; 2530 } else if ((type == RTN_LOCAL) && (orig_oif != 0) && 2531 (orig_oif != dev_out->ifindex)) { 2532 /* For local routes that require a particular output interface 2533 * we do not want to cache the result. Caching the result 2534 * causes incorrect behaviour when there are multiple source 2535 * addresses on the interface, the end result being that if the 2536 * intended recipient is waiting on that interface for the 2537 * packet he won't receive it because it will be delivered on 2538 * the loopback interface and the IP_PKTINFO ipi_ifindex will 2539 * be set to the loopback interface as well. 2540 */ 2541 do_cache = false; 2542 } 2543 2544 fnhe = NULL; 2545 do_cache &= fi != NULL; 2546 if (fi) { 2547 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2548 struct rtable __rcu **prth; 2549 2550 fnhe = find_exception(nhc, fl4->daddr); 2551 if (!do_cache) 2552 goto add; 2553 if (fnhe) { 2554 prth = &fnhe->fnhe_rth_output; 2555 } else { 2556 if (unlikely(fl4->flowi4_flags & 2557 FLOWI_FLAG_KNOWN_NH && 2558 !(nhc->nhc_gw_family && 2559 nhc->nhc_scope == RT_SCOPE_LINK))) { 2560 do_cache = false; 2561 goto add; 2562 } 2563 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 2564 } 2565 rth = rcu_dereference(*prth); 2566 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst)) 2567 return rth; 2568 } 2569 2570 add: 2571 rth = rt_dst_alloc(dev_out, flags, type, 2572 IN_DEV_ORCONF(in_dev, NOPOLICY), 2573 IN_DEV_ORCONF(in_dev, NOXFRM)); 2574 if (!rth) 2575 return ERR_PTR(-ENOBUFS); 2576 2577 rth->rt_iif = orig_oif; 2578 2579 RT_CACHE_STAT_INC(out_slow_tot); 2580 2581 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 2582 if (flags & RTCF_LOCAL && 2583 !(dev_out->flags & IFF_LOOPBACK)) { 2584 rth->dst.output = ip_mc_output; 2585 RT_CACHE_STAT_INC(out_slow_mc); 2586 } 2587 #ifdef CONFIG_IP_MROUTE 2588 if (type == RTN_MULTICAST) { 2589 if (IN_DEV_MFORWARD(in_dev) && 2590 !ipv4_is_local_multicast(fl4->daddr)) { 2591 rth->dst.input = ip_mr_input; 2592 rth->dst.output = ip_mc_output; 2593 } 2594 } 2595 #endif 2596 } 2597 2598 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache); 2599 lwtunnel_set_redirect(&rth->dst); 2600 2601 return rth; 2602 } 2603 2604 /* 2605 * Major route resolver routine. 2606 */ 2607 2608 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4, 2609 const struct sk_buff *skb) 2610 { 2611 __u8 tos = RT_FL_TOS(fl4); 2612 struct fib_result res = { 2613 .type = RTN_UNSPEC, 2614 .fi = NULL, 2615 .table = NULL, 2616 .tclassid = 0, 2617 }; 2618 struct rtable *rth; 2619 2620 fl4->flowi4_iif = LOOPBACK_IFINDEX; 2621 fl4->flowi4_tos = tos & IPTOS_RT_MASK; 2622 fl4->flowi4_scope = ((tos & RTO_ONLINK) ? 2623 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE); 2624 2625 rcu_read_lock(); 2626 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb); 2627 rcu_read_unlock(); 2628 2629 return rth; 2630 } 2631 EXPORT_SYMBOL_GPL(ip_route_output_key_hash); 2632 2633 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4, 2634 struct fib_result *res, 2635 const struct sk_buff *skb) 2636 { 2637 struct net_device *dev_out = NULL; 2638 int orig_oif = fl4->flowi4_oif; 2639 unsigned int flags = 0; 2640 struct rtable *rth; 2641 int err; 2642 2643 if (fl4->saddr) { 2644 if (ipv4_is_multicast(fl4->saddr) || 2645 ipv4_is_lbcast(fl4->saddr) || 2646 ipv4_is_zeronet(fl4->saddr)) { 2647 rth = ERR_PTR(-EINVAL); 2648 goto out; 2649 } 2650 2651 rth = ERR_PTR(-ENETUNREACH); 2652 2653 /* I removed check for oif == dev_out->oif here. 2654 * It was wrong for two reasons: 2655 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 2656 * is assigned to multiple interfaces. 2657 * 2. Moreover, we are allowed to send packets with saddr 2658 * of another iface. --ANK 2659 */ 2660 2661 if (fl4->flowi4_oif == 0 && 2662 (ipv4_is_multicast(fl4->daddr) || 2663 ipv4_is_lbcast(fl4->daddr))) { 2664 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2665 dev_out = __ip_dev_find(net, fl4->saddr, false); 2666 if (!dev_out) 2667 goto out; 2668 2669 /* Special hack: user can direct multicasts 2670 * and limited broadcast via necessary interface 2671 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 2672 * This hack is not just for fun, it allows 2673 * vic,vat and friends to work. 2674 * They bind socket to loopback, set ttl to zero 2675 * and expect that it will work. 2676 * From the viewpoint of routing cache they are broken, 2677 * because we are not allowed to build multicast path 2678 * with loopback source addr (look, routing cache 2679 * cannot know, that ttl is zero, so that packet 2680 * will not leave this host and route is valid). 2681 * Luckily, this hack is good workaround. 2682 */ 2683 2684 fl4->flowi4_oif = dev_out->ifindex; 2685 goto make_route; 2686 } 2687 2688 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 2689 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2690 if (!__ip_dev_find(net, fl4->saddr, false)) 2691 goto out; 2692 } 2693 } 2694 2695 2696 if (fl4->flowi4_oif) { 2697 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif); 2698 rth = ERR_PTR(-ENODEV); 2699 if (!dev_out) 2700 goto out; 2701 2702 /* RACE: Check return value of inet_select_addr instead. */ 2703 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 2704 rth = ERR_PTR(-ENETUNREACH); 2705 goto out; 2706 } 2707 if (ipv4_is_local_multicast(fl4->daddr) || 2708 ipv4_is_lbcast(fl4->daddr) || 2709 fl4->flowi4_proto == IPPROTO_IGMP) { 2710 if (!fl4->saddr) 2711 fl4->saddr = inet_select_addr(dev_out, 0, 2712 RT_SCOPE_LINK); 2713 goto make_route; 2714 } 2715 if (!fl4->saddr) { 2716 if (ipv4_is_multicast(fl4->daddr)) 2717 fl4->saddr = inet_select_addr(dev_out, 0, 2718 fl4->flowi4_scope); 2719 else if (!fl4->daddr) 2720 fl4->saddr = inet_select_addr(dev_out, 0, 2721 RT_SCOPE_HOST); 2722 } 2723 } 2724 2725 if (!fl4->daddr) { 2726 fl4->daddr = fl4->saddr; 2727 if (!fl4->daddr) 2728 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK); 2729 dev_out = net->loopback_dev; 2730 fl4->flowi4_oif = LOOPBACK_IFINDEX; 2731 res->type = RTN_LOCAL; 2732 flags |= RTCF_LOCAL; 2733 goto make_route; 2734 } 2735 2736 err = fib_lookup(net, fl4, res, 0); 2737 if (err) { 2738 res->fi = NULL; 2739 res->table = NULL; 2740 if (fl4->flowi4_oif && 2741 (ipv4_is_multicast(fl4->daddr) || 2742 !netif_index_is_l3_master(net, fl4->flowi4_oif))) { 2743 /* Apparently, routing tables are wrong. Assume, 2744 * that the destination is on link. 2745 * 2746 * WHY? DW. 2747 * Because we are allowed to send to iface 2748 * even if it has NO routes and NO assigned 2749 * addresses. When oif is specified, routing 2750 * tables are looked up with only one purpose: 2751 * to catch if destination is gatewayed, rather than 2752 * direct. Moreover, if MSG_DONTROUTE is set, 2753 * we send packet, ignoring both routing tables 2754 * and ifaddr state. --ANK 2755 * 2756 * 2757 * We could make it even if oif is unknown, 2758 * likely IPv6, but we do not. 2759 */ 2760 2761 if (fl4->saddr == 0) 2762 fl4->saddr = inet_select_addr(dev_out, 0, 2763 RT_SCOPE_LINK); 2764 res->type = RTN_UNICAST; 2765 goto make_route; 2766 } 2767 rth = ERR_PTR(err); 2768 goto out; 2769 } 2770 2771 if (res->type == RTN_LOCAL) { 2772 if (!fl4->saddr) { 2773 if (res->fi->fib_prefsrc) 2774 fl4->saddr = res->fi->fib_prefsrc; 2775 else 2776 fl4->saddr = fl4->daddr; 2777 } 2778 2779 /* L3 master device is the loopback for that domain */ 2780 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? : 2781 net->loopback_dev; 2782 2783 /* make sure orig_oif points to fib result device even 2784 * though packet rx/tx happens over loopback or l3mdev 2785 */ 2786 orig_oif = FIB_RES_OIF(*res); 2787 2788 fl4->flowi4_oif = dev_out->ifindex; 2789 flags |= RTCF_LOCAL; 2790 goto make_route; 2791 } 2792 2793 fib_select_path(net, res, fl4, skb); 2794 2795 dev_out = FIB_RES_DEV(*res); 2796 2797 make_route: 2798 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags); 2799 2800 out: 2801 return rth; 2802 } 2803 2804 static struct dst_ops ipv4_dst_blackhole_ops = { 2805 .family = AF_INET, 2806 .default_advmss = ipv4_default_advmss, 2807 .neigh_lookup = ipv4_neigh_lookup, 2808 .check = dst_blackhole_check, 2809 .cow_metrics = dst_blackhole_cow_metrics, 2810 .update_pmtu = dst_blackhole_update_pmtu, 2811 .redirect = dst_blackhole_redirect, 2812 .mtu = dst_blackhole_mtu, 2813 }; 2814 2815 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2816 { 2817 struct rtable *ort = (struct rtable *) dst_orig; 2818 struct rtable *rt; 2819 2820 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0); 2821 if (rt) { 2822 struct dst_entry *new = &rt->dst; 2823 2824 new->__use = 1; 2825 new->input = dst_discard; 2826 new->output = dst_discard_out; 2827 2828 new->dev = net->loopback_dev; 2829 dev_hold_track(new->dev, &new->dev_tracker, GFP_ATOMIC); 2830 2831 rt->rt_is_input = ort->rt_is_input; 2832 rt->rt_iif = ort->rt_iif; 2833 rt->rt_pmtu = ort->rt_pmtu; 2834 rt->rt_mtu_locked = ort->rt_mtu_locked; 2835 2836 rt->rt_genid = rt_genid_ipv4(net); 2837 rt->rt_flags = ort->rt_flags; 2838 rt->rt_type = ort->rt_type; 2839 rt->rt_uses_gateway = ort->rt_uses_gateway; 2840 rt->rt_gw_family = ort->rt_gw_family; 2841 if (rt->rt_gw_family == AF_INET) 2842 rt->rt_gw4 = ort->rt_gw4; 2843 else if (rt->rt_gw_family == AF_INET6) 2844 rt->rt_gw6 = ort->rt_gw6; 2845 2846 INIT_LIST_HEAD(&rt->rt_uncached); 2847 } 2848 2849 dst_release(dst_orig); 2850 2851 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2852 } 2853 2854 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2855 const struct sock *sk) 2856 { 2857 struct rtable *rt = __ip_route_output_key(net, flp4); 2858 2859 if (IS_ERR(rt)) 2860 return rt; 2861 2862 if (flp4->flowi4_proto) { 2863 flp4->flowi4_oif = rt->dst.dev->ifindex; 2864 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst, 2865 flowi4_to_flowi(flp4), 2866 sk, 0); 2867 } 2868 2869 return rt; 2870 } 2871 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2872 2873 struct rtable *ip_route_output_tunnel(struct sk_buff *skb, 2874 struct net_device *dev, 2875 struct net *net, __be32 *saddr, 2876 const struct ip_tunnel_info *info, 2877 u8 protocol, bool use_cache) 2878 { 2879 #ifdef CONFIG_DST_CACHE 2880 struct dst_cache *dst_cache; 2881 #endif 2882 struct rtable *rt = NULL; 2883 struct flowi4 fl4; 2884 __u8 tos; 2885 2886 #ifdef CONFIG_DST_CACHE 2887 dst_cache = (struct dst_cache *)&info->dst_cache; 2888 if (use_cache) { 2889 rt = dst_cache_get_ip4(dst_cache, saddr); 2890 if (rt) 2891 return rt; 2892 } 2893 #endif 2894 memset(&fl4, 0, sizeof(fl4)); 2895 fl4.flowi4_mark = skb->mark; 2896 fl4.flowi4_proto = protocol; 2897 fl4.daddr = info->key.u.ipv4.dst; 2898 fl4.saddr = info->key.u.ipv4.src; 2899 tos = info->key.tos; 2900 fl4.flowi4_tos = RT_TOS(tos); 2901 2902 rt = ip_route_output_key(net, &fl4); 2903 if (IS_ERR(rt)) { 2904 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr); 2905 return ERR_PTR(-ENETUNREACH); 2906 } 2907 if (rt->dst.dev == dev) { /* is this necessary? */ 2908 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr); 2909 ip_rt_put(rt); 2910 return ERR_PTR(-ELOOP); 2911 } 2912 #ifdef CONFIG_DST_CACHE 2913 if (use_cache) 2914 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr); 2915 #endif 2916 *saddr = fl4.saddr; 2917 return rt; 2918 } 2919 EXPORT_SYMBOL_GPL(ip_route_output_tunnel); 2920 2921 /* called with rcu_read_lock held */ 2922 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, 2923 struct rtable *rt, u32 table_id, struct flowi4 *fl4, 2924 struct sk_buff *skb, u32 portid, u32 seq, 2925 unsigned int flags) 2926 { 2927 struct rtmsg *r; 2928 struct nlmsghdr *nlh; 2929 unsigned long expires = 0; 2930 u32 error; 2931 u32 metrics[RTAX_MAX]; 2932 2933 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags); 2934 if (!nlh) 2935 return -EMSGSIZE; 2936 2937 r = nlmsg_data(nlh); 2938 r->rtm_family = AF_INET; 2939 r->rtm_dst_len = 32; 2940 r->rtm_src_len = 0; 2941 r->rtm_tos = fl4 ? fl4->flowi4_tos : 0; 2942 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT; 2943 if (nla_put_u32(skb, RTA_TABLE, table_id)) 2944 goto nla_put_failure; 2945 r->rtm_type = rt->rt_type; 2946 r->rtm_scope = RT_SCOPE_UNIVERSE; 2947 r->rtm_protocol = RTPROT_UNSPEC; 2948 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 2949 if (rt->rt_flags & RTCF_NOTIFY) 2950 r->rtm_flags |= RTM_F_NOTIFY; 2951 if (IPCB(skb)->flags & IPSKB_DOREDIRECT) 2952 r->rtm_flags |= RTCF_DOREDIRECT; 2953 2954 if (nla_put_in_addr(skb, RTA_DST, dst)) 2955 goto nla_put_failure; 2956 if (src) { 2957 r->rtm_src_len = 32; 2958 if (nla_put_in_addr(skb, RTA_SRC, src)) 2959 goto nla_put_failure; 2960 } 2961 if (rt->dst.dev && 2962 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex)) 2963 goto nla_put_failure; 2964 if (rt->dst.lwtstate && 2965 lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0) 2966 goto nla_put_failure; 2967 #ifdef CONFIG_IP_ROUTE_CLASSID 2968 if (rt->dst.tclassid && 2969 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid)) 2970 goto nla_put_failure; 2971 #endif 2972 if (fl4 && !rt_is_input_route(rt) && 2973 fl4->saddr != src) { 2974 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr)) 2975 goto nla_put_failure; 2976 } 2977 if (rt->rt_uses_gateway) { 2978 if (rt->rt_gw_family == AF_INET && 2979 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) { 2980 goto nla_put_failure; 2981 } else if (rt->rt_gw_family == AF_INET6) { 2982 int alen = sizeof(struct in6_addr); 2983 struct nlattr *nla; 2984 struct rtvia *via; 2985 2986 nla = nla_reserve(skb, RTA_VIA, alen + 2); 2987 if (!nla) 2988 goto nla_put_failure; 2989 2990 via = nla_data(nla); 2991 via->rtvia_family = AF_INET6; 2992 memcpy(via->rtvia_addr, &rt->rt_gw6, alen); 2993 } 2994 } 2995 2996 expires = rt->dst.expires; 2997 if (expires) { 2998 unsigned long now = jiffies; 2999 3000 if (time_before(now, expires)) 3001 expires -= now; 3002 else 3003 expires = 0; 3004 } 3005 3006 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics)); 3007 if (rt->rt_pmtu && expires) 3008 metrics[RTAX_MTU - 1] = rt->rt_pmtu; 3009 if (rt->rt_mtu_locked && expires) 3010 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU); 3011 if (rtnetlink_put_metrics(skb, metrics) < 0) 3012 goto nla_put_failure; 3013 3014 if (fl4) { 3015 if (fl4->flowi4_mark && 3016 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark)) 3017 goto nla_put_failure; 3018 3019 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) && 3020 nla_put_u32(skb, RTA_UID, 3021 from_kuid_munged(current_user_ns(), 3022 fl4->flowi4_uid))) 3023 goto nla_put_failure; 3024 3025 if (rt_is_input_route(rt)) { 3026 #ifdef CONFIG_IP_MROUTE 3027 if (ipv4_is_multicast(dst) && 3028 !ipv4_is_local_multicast(dst) && 3029 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) { 3030 int err = ipmr_get_route(net, skb, 3031 fl4->saddr, fl4->daddr, 3032 r, portid); 3033 3034 if (err <= 0) { 3035 if (err == 0) 3036 return 0; 3037 goto nla_put_failure; 3038 } 3039 } else 3040 #endif 3041 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif)) 3042 goto nla_put_failure; 3043 } 3044 } 3045 3046 error = rt->dst.error; 3047 3048 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0) 3049 goto nla_put_failure; 3050 3051 nlmsg_end(skb, nlh); 3052 return 0; 3053 3054 nla_put_failure: 3055 nlmsg_cancel(skb, nlh); 3056 return -EMSGSIZE; 3057 } 3058 3059 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb, 3060 struct netlink_callback *cb, u32 table_id, 3061 struct fnhe_hash_bucket *bucket, int genid, 3062 int *fa_index, int fa_start, unsigned int flags) 3063 { 3064 int i; 3065 3066 for (i = 0; i < FNHE_HASH_SIZE; i++) { 3067 struct fib_nh_exception *fnhe; 3068 3069 for (fnhe = rcu_dereference(bucket[i].chain); fnhe; 3070 fnhe = rcu_dereference(fnhe->fnhe_next)) { 3071 struct rtable *rt; 3072 int err; 3073 3074 if (*fa_index < fa_start) 3075 goto next; 3076 3077 if (fnhe->fnhe_genid != genid) 3078 goto next; 3079 3080 if (fnhe->fnhe_expires && 3081 time_after(jiffies, fnhe->fnhe_expires)) 3082 goto next; 3083 3084 rt = rcu_dereference(fnhe->fnhe_rth_input); 3085 if (!rt) 3086 rt = rcu_dereference(fnhe->fnhe_rth_output); 3087 if (!rt) 3088 goto next; 3089 3090 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt, 3091 table_id, NULL, skb, 3092 NETLINK_CB(cb->skb).portid, 3093 cb->nlh->nlmsg_seq, flags); 3094 if (err) 3095 return err; 3096 next: 3097 (*fa_index)++; 3098 } 3099 } 3100 3101 return 0; 3102 } 3103 3104 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb, 3105 u32 table_id, struct fib_info *fi, 3106 int *fa_index, int fa_start, unsigned int flags) 3107 { 3108 struct net *net = sock_net(cb->skb->sk); 3109 int nhsel, genid = fnhe_genid(net); 3110 3111 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) { 3112 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel); 3113 struct fnhe_hash_bucket *bucket; 3114 int err; 3115 3116 if (nhc->nhc_flags & RTNH_F_DEAD) 3117 continue; 3118 3119 rcu_read_lock(); 3120 bucket = rcu_dereference(nhc->nhc_exceptions); 3121 err = 0; 3122 if (bucket) 3123 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket, 3124 genid, fa_index, fa_start, 3125 flags); 3126 rcu_read_unlock(); 3127 if (err) 3128 return err; 3129 } 3130 3131 return 0; 3132 } 3133 3134 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst, 3135 u8 ip_proto, __be16 sport, 3136 __be16 dport) 3137 { 3138 struct sk_buff *skb; 3139 struct iphdr *iph; 3140 3141 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 3142 if (!skb) 3143 return NULL; 3144 3145 /* Reserve room for dummy headers, this skb can pass 3146 * through good chunk of routing engine. 3147 */ 3148 skb_reset_mac_header(skb); 3149 skb_reset_network_header(skb); 3150 skb->protocol = htons(ETH_P_IP); 3151 iph = skb_put(skb, sizeof(struct iphdr)); 3152 iph->protocol = ip_proto; 3153 iph->saddr = src; 3154 iph->daddr = dst; 3155 iph->version = 0x4; 3156 iph->frag_off = 0; 3157 iph->ihl = 0x5; 3158 skb_set_transport_header(skb, skb->len); 3159 3160 switch (iph->protocol) { 3161 case IPPROTO_UDP: { 3162 struct udphdr *udph; 3163 3164 udph = skb_put_zero(skb, sizeof(struct udphdr)); 3165 udph->source = sport; 3166 udph->dest = dport; 3167 udph->len = htons(sizeof(struct udphdr)); 3168 udph->check = 0; 3169 break; 3170 } 3171 case IPPROTO_TCP: { 3172 struct tcphdr *tcph; 3173 3174 tcph = skb_put_zero(skb, sizeof(struct tcphdr)); 3175 tcph->source = sport; 3176 tcph->dest = dport; 3177 tcph->doff = sizeof(struct tcphdr) / 4; 3178 tcph->rst = 1; 3179 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr), 3180 src, dst, 0); 3181 break; 3182 } 3183 case IPPROTO_ICMP: { 3184 struct icmphdr *icmph; 3185 3186 icmph = skb_put_zero(skb, sizeof(struct icmphdr)); 3187 icmph->type = ICMP_ECHO; 3188 icmph->code = 0; 3189 } 3190 } 3191 3192 return skb; 3193 } 3194 3195 static int inet_rtm_valid_getroute_req(struct sk_buff *skb, 3196 const struct nlmsghdr *nlh, 3197 struct nlattr **tb, 3198 struct netlink_ext_ack *extack) 3199 { 3200 struct rtmsg *rtm; 3201 int i, err; 3202 3203 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) { 3204 NL_SET_ERR_MSG(extack, 3205 "ipv4: Invalid header for route get request"); 3206 return -EINVAL; 3207 } 3208 3209 if (!netlink_strict_get_check(skb)) 3210 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX, 3211 rtm_ipv4_policy, extack); 3212 3213 rtm = nlmsg_data(nlh); 3214 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) || 3215 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) || 3216 rtm->rtm_table || rtm->rtm_protocol || 3217 rtm->rtm_scope || rtm->rtm_type) { 3218 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request"); 3219 return -EINVAL; 3220 } 3221 3222 if (rtm->rtm_flags & ~(RTM_F_NOTIFY | 3223 RTM_F_LOOKUP_TABLE | 3224 RTM_F_FIB_MATCH)) { 3225 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request"); 3226 return -EINVAL; 3227 } 3228 3229 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX, 3230 rtm_ipv4_policy, extack); 3231 if (err) 3232 return err; 3233 3234 if ((tb[RTA_SRC] && !rtm->rtm_src_len) || 3235 (tb[RTA_DST] && !rtm->rtm_dst_len)) { 3236 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4"); 3237 return -EINVAL; 3238 } 3239 3240 for (i = 0; i <= RTA_MAX; i++) { 3241 if (!tb[i]) 3242 continue; 3243 3244 switch (i) { 3245 case RTA_IIF: 3246 case RTA_OIF: 3247 case RTA_SRC: 3248 case RTA_DST: 3249 case RTA_IP_PROTO: 3250 case RTA_SPORT: 3251 case RTA_DPORT: 3252 case RTA_MARK: 3253 case RTA_UID: 3254 break; 3255 default: 3256 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request"); 3257 return -EINVAL; 3258 } 3259 } 3260 3261 return 0; 3262 } 3263 3264 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, 3265 struct netlink_ext_ack *extack) 3266 { 3267 struct net *net = sock_net(in_skb->sk); 3268 struct nlattr *tb[RTA_MAX+1]; 3269 u32 table_id = RT_TABLE_MAIN; 3270 __be16 sport = 0, dport = 0; 3271 struct fib_result res = {}; 3272 u8 ip_proto = IPPROTO_UDP; 3273 struct rtable *rt = NULL; 3274 struct sk_buff *skb; 3275 struct rtmsg *rtm; 3276 struct flowi4 fl4 = {}; 3277 __be32 dst = 0; 3278 __be32 src = 0; 3279 kuid_t uid; 3280 u32 iif; 3281 int err; 3282 int mark; 3283 3284 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack); 3285 if (err < 0) 3286 return err; 3287 3288 rtm = nlmsg_data(nlh); 3289 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0; 3290 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0; 3291 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0; 3292 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0; 3293 if (tb[RTA_UID]) 3294 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID])); 3295 else 3296 uid = (iif ? INVALID_UID : current_uid()); 3297 3298 if (tb[RTA_IP_PROTO]) { 3299 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO], 3300 &ip_proto, AF_INET, extack); 3301 if (err) 3302 return err; 3303 } 3304 3305 if (tb[RTA_SPORT]) 3306 sport = nla_get_be16(tb[RTA_SPORT]); 3307 3308 if (tb[RTA_DPORT]) 3309 dport = nla_get_be16(tb[RTA_DPORT]); 3310 3311 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport); 3312 if (!skb) 3313 return -ENOBUFS; 3314 3315 fl4.daddr = dst; 3316 fl4.saddr = src; 3317 fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK; 3318 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0; 3319 fl4.flowi4_mark = mark; 3320 fl4.flowi4_uid = uid; 3321 if (sport) 3322 fl4.fl4_sport = sport; 3323 if (dport) 3324 fl4.fl4_dport = dport; 3325 fl4.flowi4_proto = ip_proto; 3326 3327 rcu_read_lock(); 3328 3329 if (iif) { 3330 struct net_device *dev; 3331 3332 dev = dev_get_by_index_rcu(net, iif); 3333 if (!dev) { 3334 err = -ENODEV; 3335 goto errout_rcu; 3336 } 3337 3338 fl4.flowi4_iif = iif; /* for rt_fill_info */ 3339 skb->dev = dev; 3340 skb->mark = mark; 3341 err = ip_route_input_rcu(skb, dst, src, 3342 rtm->rtm_tos & IPTOS_RT_MASK, dev, 3343 &res); 3344 3345 rt = skb_rtable(skb); 3346 if (err == 0 && rt->dst.error) 3347 err = -rt->dst.error; 3348 } else { 3349 fl4.flowi4_iif = LOOPBACK_IFINDEX; 3350 skb->dev = net->loopback_dev; 3351 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb); 3352 err = 0; 3353 if (IS_ERR(rt)) 3354 err = PTR_ERR(rt); 3355 else 3356 skb_dst_set(skb, &rt->dst); 3357 } 3358 3359 if (err) 3360 goto errout_rcu; 3361 3362 if (rtm->rtm_flags & RTM_F_NOTIFY) 3363 rt->rt_flags |= RTCF_NOTIFY; 3364 3365 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE) 3366 table_id = res.table ? res.table->tb_id : 0; 3367 3368 /* reset skb for netlink reply msg */ 3369 skb_trim(skb, 0); 3370 skb_reset_network_header(skb); 3371 skb_reset_transport_header(skb); 3372 skb_reset_mac_header(skb); 3373 3374 if (rtm->rtm_flags & RTM_F_FIB_MATCH) { 3375 struct fib_rt_info fri; 3376 3377 if (!res.fi) { 3378 err = fib_props[res.type].error; 3379 if (!err) 3380 err = -EHOSTUNREACH; 3381 goto errout_rcu; 3382 } 3383 fri.fi = res.fi; 3384 fri.tb_id = table_id; 3385 fri.dst = res.prefix; 3386 fri.dst_len = res.prefixlen; 3387 fri.tos = fl4.flowi4_tos; 3388 fri.type = rt->rt_type; 3389 fri.offload = 0; 3390 fri.trap = 0; 3391 fri.offload_failed = 0; 3392 if (res.fa_head) { 3393 struct fib_alias *fa; 3394 3395 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) { 3396 u8 slen = 32 - fri.dst_len; 3397 3398 if (fa->fa_slen == slen && 3399 fa->tb_id == fri.tb_id && 3400 fa->fa_dscp == inet_dsfield_to_dscp(fri.tos) && 3401 fa->fa_info == res.fi && 3402 fa->fa_type == fri.type) { 3403 fri.offload = READ_ONCE(fa->offload); 3404 fri.trap = READ_ONCE(fa->trap); 3405 break; 3406 } 3407 } 3408 } 3409 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid, 3410 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0); 3411 } else { 3412 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb, 3413 NETLINK_CB(in_skb).portid, 3414 nlh->nlmsg_seq, 0); 3415 } 3416 if (err < 0) 3417 goto errout_rcu; 3418 3419 rcu_read_unlock(); 3420 3421 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 3422 3423 errout_free: 3424 return err; 3425 errout_rcu: 3426 rcu_read_unlock(); 3427 kfree_skb(skb); 3428 goto errout_free; 3429 } 3430 3431 void ip_rt_multicast_event(struct in_device *in_dev) 3432 { 3433 rt_cache_flush(dev_net(in_dev->dev)); 3434 } 3435 3436 #ifdef CONFIG_SYSCTL 3437 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 3438 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 3439 static int ip_rt_gc_elasticity __read_mostly = 8; 3440 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU; 3441 3442 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write, 3443 void *buffer, size_t *lenp, loff_t *ppos) 3444 { 3445 struct net *net = (struct net *)__ctl->extra1; 3446 3447 if (write) { 3448 rt_cache_flush(net); 3449 fnhe_genid_bump(net); 3450 return 0; 3451 } 3452 3453 return -EINVAL; 3454 } 3455 3456 static struct ctl_table ipv4_route_table[] = { 3457 { 3458 .procname = "gc_thresh", 3459 .data = &ipv4_dst_ops.gc_thresh, 3460 .maxlen = sizeof(int), 3461 .mode = 0644, 3462 .proc_handler = proc_dointvec, 3463 }, 3464 { 3465 .procname = "max_size", 3466 .data = &ip_rt_max_size, 3467 .maxlen = sizeof(int), 3468 .mode = 0644, 3469 .proc_handler = proc_dointvec, 3470 }, 3471 { 3472 /* Deprecated. Use gc_min_interval_ms */ 3473 3474 .procname = "gc_min_interval", 3475 .data = &ip_rt_gc_min_interval, 3476 .maxlen = sizeof(int), 3477 .mode = 0644, 3478 .proc_handler = proc_dointvec_jiffies, 3479 }, 3480 { 3481 .procname = "gc_min_interval_ms", 3482 .data = &ip_rt_gc_min_interval, 3483 .maxlen = sizeof(int), 3484 .mode = 0644, 3485 .proc_handler = proc_dointvec_ms_jiffies, 3486 }, 3487 { 3488 .procname = "gc_timeout", 3489 .data = &ip_rt_gc_timeout, 3490 .maxlen = sizeof(int), 3491 .mode = 0644, 3492 .proc_handler = proc_dointvec_jiffies, 3493 }, 3494 { 3495 .procname = "gc_interval", 3496 .data = &ip_rt_gc_interval, 3497 .maxlen = sizeof(int), 3498 .mode = 0644, 3499 .proc_handler = proc_dointvec_jiffies, 3500 }, 3501 { 3502 .procname = "redirect_load", 3503 .data = &ip_rt_redirect_load, 3504 .maxlen = sizeof(int), 3505 .mode = 0644, 3506 .proc_handler = proc_dointvec, 3507 }, 3508 { 3509 .procname = "redirect_number", 3510 .data = &ip_rt_redirect_number, 3511 .maxlen = sizeof(int), 3512 .mode = 0644, 3513 .proc_handler = proc_dointvec, 3514 }, 3515 { 3516 .procname = "redirect_silence", 3517 .data = &ip_rt_redirect_silence, 3518 .maxlen = sizeof(int), 3519 .mode = 0644, 3520 .proc_handler = proc_dointvec, 3521 }, 3522 { 3523 .procname = "error_cost", 3524 .data = &ip_rt_error_cost, 3525 .maxlen = sizeof(int), 3526 .mode = 0644, 3527 .proc_handler = proc_dointvec, 3528 }, 3529 { 3530 .procname = "error_burst", 3531 .data = &ip_rt_error_burst, 3532 .maxlen = sizeof(int), 3533 .mode = 0644, 3534 .proc_handler = proc_dointvec, 3535 }, 3536 { 3537 .procname = "gc_elasticity", 3538 .data = &ip_rt_gc_elasticity, 3539 .maxlen = sizeof(int), 3540 .mode = 0644, 3541 .proc_handler = proc_dointvec, 3542 }, 3543 { } 3544 }; 3545 3546 static const char ipv4_route_flush_procname[] = "flush"; 3547 3548 static struct ctl_table ipv4_route_netns_table[] = { 3549 { 3550 .procname = ipv4_route_flush_procname, 3551 .maxlen = sizeof(int), 3552 .mode = 0200, 3553 .proc_handler = ipv4_sysctl_rtcache_flush, 3554 }, 3555 { 3556 .procname = "min_pmtu", 3557 .data = &init_net.ipv4.ip_rt_min_pmtu, 3558 .maxlen = sizeof(int), 3559 .mode = 0644, 3560 .proc_handler = proc_dointvec_minmax, 3561 .extra1 = &ip_min_valid_pmtu, 3562 }, 3563 { 3564 .procname = "mtu_expires", 3565 .data = &init_net.ipv4.ip_rt_mtu_expires, 3566 .maxlen = sizeof(int), 3567 .mode = 0644, 3568 .proc_handler = proc_dointvec_jiffies, 3569 }, 3570 { 3571 .procname = "min_adv_mss", 3572 .data = &init_net.ipv4.ip_rt_min_advmss, 3573 .maxlen = sizeof(int), 3574 .mode = 0644, 3575 .proc_handler = proc_dointvec, 3576 }, 3577 { }, 3578 }; 3579 3580 static __net_init int sysctl_route_net_init(struct net *net) 3581 { 3582 struct ctl_table *tbl; 3583 3584 tbl = ipv4_route_netns_table; 3585 if (!net_eq(net, &init_net)) { 3586 int i; 3587 3588 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL); 3589 if (!tbl) 3590 goto err_dup; 3591 3592 /* Don't export non-whitelisted sysctls to unprivileged users */ 3593 if (net->user_ns != &init_user_ns) { 3594 if (tbl[0].procname != ipv4_route_flush_procname) 3595 tbl[0].procname = NULL; 3596 } 3597 3598 /* Update the variables to point into the current struct net 3599 * except for the first element flush 3600 */ 3601 for (i = 1; i < ARRAY_SIZE(ipv4_route_netns_table) - 1; i++) 3602 tbl[i].data += (void *)net - (void *)&init_net; 3603 } 3604 tbl[0].extra1 = net; 3605 3606 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl); 3607 if (!net->ipv4.route_hdr) 3608 goto err_reg; 3609 return 0; 3610 3611 err_reg: 3612 if (tbl != ipv4_route_netns_table) 3613 kfree(tbl); 3614 err_dup: 3615 return -ENOMEM; 3616 } 3617 3618 static __net_exit void sysctl_route_net_exit(struct net *net) 3619 { 3620 struct ctl_table *tbl; 3621 3622 tbl = net->ipv4.route_hdr->ctl_table_arg; 3623 unregister_net_sysctl_table(net->ipv4.route_hdr); 3624 BUG_ON(tbl == ipv4_route_netns_table); 3625 kfree(tbl); 3626 } 3627 3628 static __net_initdata struct pernet_operations sysctl_route_ops = { 3629 .init = sysctl_route_net_init, 3630 .exit = sysctl_route_net_exit, 3631 }; 3632 #endif 3633 3634 static __net_init int netns_ip_rt_init(struct net *net) 3635 { 3636 /* Set default value for namespaceified sysctls */ 3637 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU; 3638 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES; 3639 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS; 3640 return 0; 3641 } 3642 3643 static struct pernet_operations __net_initdata ip_rt_ops = { 3644 .init = netns_ip_rt_init, 3645 }; 3646 3647 static __net_init int rt_genid_init(struct net *net) 3648 { 3649 atomic_set(&net->ipv4.rt_genid, 0); 3650 atomic_set(&net->fnhe_genid, 0); 3651 atomic_set(&net->ipv4.dev_addr_genid, get_random_int()); 3652 return 0; 3653 } 3654 3655 static __net_initdata struct pernet_operations rt_genid_ops = { 3656 .init = rt_genid_init, 3657 }; 3658 3659 static int __net_init ipv4_inetpeer_init(struct net *net) 3660 { 3661 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL); 3662 3663 if (!bp) 3664 return -ENOMEM; 3665 inet_peer_base_init(bp); 3666 net->ipv4.peers = bp; 3667 return 0; 3668 } 3669 3670 static void __net_exit ipv4_inetpeer_exit(struct net *net) 3671 { 3672 struct inet_peer_base *bp = net->ipv4.peers; 3673 3674 net->ipv4.peers = NULL; 3675 inetpeer_invalidate_tree(bp); 3676 kfree(bp); 3677 } 3678 3679 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = { 3680 .init = ipv4_inetpeer_init, 3681 .exit = ipv4_inetpeer_exit, 3682 }; 3683 3684 #ifdef CONFIG_IP_ROUTE_CLASSID 3685 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 3686 #endif /* CONFIG_IP_ROUTE_CLASSID */ 3687 3688 int __init ip_rt_init(void) 3689 { 3690 void *idents_hash; 3691 int cpu; 3692 3693 /* For modern hosts, this will use 2 MB of memory */ 3694 idents_hash = alloc_large_system_hash("IP idents", 3695 sizeof(*ip_idents) + sizeof(*ip_tstamps), 3696 0, 3697 16, /* one bucket per 64 KB */ 3698 HASH_ZERO, 3699 NULL, 3700 &ip_idents_mask, 3701 2048, 3702 256*1024); 3703 3704 ip_idents = idents_hash; 3705 3706 prandom_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents)); 3707 3708 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents); 3709 3710 for_each_possible_cpu(cpu) { 3711 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 3712 3713 INIT_LIST_HEAD(&ul->head); 3714 INIT_LIST_HEAD(&ul->quarantine); 3715 spin_lock_init(&ul->lock); 3716 } 3717 #ifdef CONFIG_IP_ROUTE_CLASSID 3718 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 3719 if (!ip_rt_acct) 3720 panic("IP: failed to allocate ip_rt_acct\n"); 3721 #endif 3722 3723 ipv4_dst_ops.kmem_cachep = 3724 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0, 3725 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 3726 3727 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 3728 3729 if (dst_entries_init(&ipv4_dst_ops) < 0) 3730 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 3731 3732 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 3733 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 3734 3735 ipv4_dst_ops.gc_thresh = ~0; 3736 ip_rt_max_size = INT_MAX; 3737 3738 devinet_init(); 3739 ip_fib_init(); 3740 3741 if (ip_rt_proc_init()) 3742 pr_err("Unable to create route proc files\n"); 3743 #ifdef CONFIG_XFRM 3744 xfrm_init(); 3745 xfrm4_init(); 3746 #endif 3747 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, 3748 RTNL_FLAG_DOIT_UNLOCKED); 3749 3750 #ifdef CONFIG_SYSCTL 3751 register_pernet_subsys(&sysctl_route_ops); 3752 #endif 3753 register_pernet_subsys(&ip_rt_ops); 3754 register_pernet_subsys(&rt_genid_ops); 3755 register_pernet_subsys(&ipv4_inetpeer_ops); 3756 return 0; 3757 } 3758 3759 #ifdef CONFIG_SYSCTL 3760 /* 3761 * We really need to sanitize the damn ipv4 init order, then all 3762 * this nonsense will go away. 3763 */ 3764 void __init ip_static_sysctl_init(void) 3765 { 3766 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table); 3767 } 3768 #endif 3769