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 void ipv4_negative_advice(struct sock *sk, 136 struct dst_entry *dst); 137 static void ipv4_link_failure(struct sk_buff *skb); 138 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 139 struct sk_buff *skb, u32 mtu, 140 bool confirm_neigh); 141 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, 142 struct sk_buff *skb); 143 static void ipv4_dst_destroy(struct dst_entry *dst); 144 145 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old) 146 { 147 WARN_ON(1); 148 return NULL; 149 } 150 151 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 152 struct sk_buff *skb, 153 const void *daddr); 154 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr); 155 156 static struct dst_ops ipv4_dst_ops = { 157 .family = AF_INET, 158 .check = ipv4_dst_check, 159 .default_advmss = ipv4_default_advmss, 160 .mtu = ipv4_mtu, 161 .cow_metrics = ipv4_cow_metrics, 162 .destroy = ipv4_dst_destroy, 163 .negative_advice = ipv4_negative_advice, 164 .link_failure = ipv4_link_failure, 165 .update_pmtu = ip_rt_update_pmtu, 166 .redirect = ip_do_redirect, 167 .local_out = __ip_local_out, 168 .neigh_lookup = ipv4_neigh_lookup, 169 .confirm_neigh = ipv4_confirm_neigh, 170 }; 171 172 #define ECN_OR_COST(class) TC_PRIO_##class 173 174 const __u8 ip_tos2prio[16] = { 175 TC_PRIO_BESTEFFORT, 176 ECN_OR_COST(BESTEFFORT), 177 TC_PRIO_BESTEFFORT, 178 ECN_OR_COST(BESTEFFORT), 179 TC_PRIO_BULK, 180 ECN_OR_COST(BULK), 181 TC_PRIO_BULK, 182 ECN_OR_COST(BULK), 183 TC_PRIO_INTERACTIVE, 184 ECN_OR_COST(INTERACTIVE), 185 TC_PRIO_INTERACTIVE, 186 ECN_OR_COST(INTERACTIVE), 187 TC_PRIO_INTERACTIVE_BULK, 188 ECN_OR_COST(INTERACTIVE_BULK), 189 TC_PRIO_INTERACTIVE_BULK, 190 ECN_OR_COST(INTERACTIVE_BULK) 191 }; 192 EXPORT_SYMBOL(ip_tos2prio); 193 194 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat); 195 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field) 196 197 #ifdef CONFIG_PROC_FS 198 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos) 199 { 200 if (*pos) 201 return NULL; 202 return SEQ_START_TOKEN; 203 } 204 205 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos) 206 { 207 ++*pos; 208 return NULL; 209 } 210 211 static void rt_cache_seq_stop(struct seq_file *seq, void *v) 212 { 213 } 214 215 static int rt_cache_seq_show(struct seq_file *seq, void *v) 216 { 217 if (v == SEQ_START_TOKEN) 218 seq_printf(seq, "%-127s\n", 219 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t" 220 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t" 221 "HHUptod\tSpecDst"); 222 return 0; 223 } 224 225 static const struct seq_operations rt_cache_seq_ops = { 226 .start = rt_cache_seq_start, 227 .next = rt_cache_seq_next, 228 .stop = rt_cache_seq_stop, 229 .show = rt_cache_seq_show, 230 }; 231 232 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos) 233 { 234 int cpu; 235 236 if (*pos == 0) 237 return SEQ_START_TOKEN; 238 239 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) { 240 if (!cpu_possible(cpu)) 241 continue; 242 *pos = cpu+1; 243 return &per_cpu(rt_cache_stat, cpu); 244 } 245 return NULL; 246 } 247 248 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos) 249 { 250 int cpu; 251 252 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) { 253 if (!cpu_possible(cpu)) 254 continue; 255 *pos = cpu+1; 256 return &per_cpu(rt_cache_stat, cpu); 257 } 258 (*pos)++; 259 return NULL; 260 261 } 262 263 static void rt_cpu_seq_stop(struct seq_file *seq, void *v) 264 { 265 266 } 267 268 static int rt_cpu_seq_show(struct seq_file *seq, void *v) 269 { 270 struct rt_cache_stat *st = v; 271 272 if (v == SEQ_START_TOKEN) { 273 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"); 274 return 0; 275 } 276 277 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x " 278 "%08x %08x %08x %08x %08x %08x " 279 "%08x %08x %08x %08x\n", 280 dst_entries_get_slow(&ipv4_dst_ops), 281 0, /* st->in_hit */ 282 st->in_slow_tot, 283 st->in_slow_mc, 284 st->in_no_route, 285 st->in_brd, 286 st->in_martian_dst, 287 st->in_martian_src, 288 289 0, /* st->out_hit */ 290 st->out_slow_tot, 291 st->out_slow_mc, 292 293 0, /* st->gc_total */ 294 0, /* st->gc_ignored */ 295 0, /* st->gc_goal_miss */ 296 0, /* st->gc_dst_overflow */ 297 0, /* st->in_hlist_search */ 298 0 /* st->out_hlist_search */ 299 ); 300 return 0; 301 } 302 303 static const struct seq_operations rt_cpu_seq_ops = { 304 .start = rt_cpu_seq_start, 305 .next = rt_cpu_seq_next, 306 .stop = rt_cpu_seq_stop, 307 .show = rt_cpu_seq_show, 308 }; 309 310 #ifdef CONFIG_IP_ROUTE_CLASSID 311 static int rt_acct_proc_show(struct seq_file *m, void *v) 312 { 313 struct ip_rt_acct *dst, *src; 314 unsigned int i, j; 315 316 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL); 317 if (!dst) 318 return -ENOMEM; 319 320 for_each_possible_cpu(i) { 321 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i); 322 for (j = 0; j < 256; j++) { 323 dst[j].o_bytes += src[j].o_bytes; 324 dst[j].o_packets += src[j].o_packets; 325 dst[j].i_bytes += src[j].i_bytes; 326 dst[j].i_packets += src[j].i_packets; 327 } 328 } 329 330 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct)); 331 kfree(dst); 332 return 0; 333 } 334 #endif 335 336 static int __net_init ip_rt_do_proc_init(struct net *net) 337 { 338 struct proc_dir_entry *pde; 339 340 pde = proc_create_seq("rt_cache", 0444, net->proc_net, 341 &rt_cache_seq_ops); 342 if (!pde) 343 goto err1; 344 345 pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat, 346 &rt_cpu_seq_ops); 347 if (!pde) 348 goto err2; 349 350 #ifdef CONFIG_IP_ROUTE_CLASSID 351 pde = proc_create_single("rt_acct", 0, net->proc_net, 352 rt_acct_proc_show); 353 if (!pde) 354 goto err3; 355 #endif 356 return 0; 357 358 #ifdef CONFIG_IP_ROUTE_CLASSID 359 err3: 360 remove_proc_entry("rt_cache", net->proc_net_stat); 361 #endif 362 err2: 363 remove_proc_entry("rt_cache", net->proc_net); 364 err1: 365 return -ENOMEM; 366 } 367 368 static void __net_exit ip_rt_do_proc_exit(struct net *net) 369 { 370 remove_proc_entry("rt_cache", net->proc_net_stat); 371 remove_proc_entry("rt_cache", net->proc_net); 372 #ifdef CONFIG_IP_ROUTE_CLASSID 373 remove_proc_entry("rt_acct", net->proc_net); 374 #endif 375 } 376 377 static struct pernet_operations ip_rt_proc_ops __net_initdata = { 378 .init = ip_rt_do_proc_init, 379 .exit = ip_rt_do_proc_exit, 380 }; 381 382 static int __init ip_rt_proc_init(void) 383 { 384 return register_pernet_subsys(&ip_rt_proc_ops); 385 } 386 387 #else 388 static inline int ip_rt_proc_init(void) 389 { 390 return 0; 391 } 392 #endif /* CONFIG_PROC_FS */ 393 394 static inline bool rt_is_expired(const struct rtable *rth) 395 { 396 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev)); 397 } 398 399 void rt_cache_flush(struct net *net) 400 { 401 rt_genid_bump_ipv4(net); 402 } 403 404 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 405 struct sk_buff *skb, 406 const void *daddr) 407 { 408 const struct rtable *rt = container_of(dst, struct rtable, dst); 409 struct net_device *dev = dst->dev; 410 struct neighbour *n; 411 412 rcu_read_lock(); 413 414 if (likely(rt->rt_gw_family == AF_INET)) { 415 n = ip_neigh_gw4(dev, rt->rt_gw4); 416 } else if (rt->rt_gw_family == AF_INET6) { 417 n = ip_neigh_gw6(dev, &rt->rt_gw6); 418 } else { 419 __be32 pkey; 420 421 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr); 422 n = ip_neigh_gw4(dev, pkey); 423 } 424 425 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt)) 426 n = NULL; 427 428 rcu_read_unlock(); 429 430 return n; 431 } 432 433 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr) 434 { 435 const struct rtable *rt = container_of(dst, struct rtable, dst); 436 struct net_device *dev = dst->dev; 437 const __be32 *pkey = daddr; 438 439 if (rt->rt_gw_family == AF_INET) { 440 pkey = (const __be32 *)&rt->rt_gw4; 441 } else if (rt->rt_gw_family == AF_INET6) { 442 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6); 443 } else if (!daddr || 444 (rt->rt_flags & 445 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) { 446 return; 447 } 448 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey); 449 } 450 451 /* Hash tables of size 2048..262144 depending on RAM size. 452 * Each bucket uses 8 bytes. 453 */ 454 static u32 ip_idents_mask __read_mostly; 455 static atomic_t *ip_idents __read_mostly; 456 static u32 *ip_tstamps __read_mostly; 457 458 /* In order to protect privacy, we add a perturbation to identifiers 459 * if one generator is seldom used. This makes hard for an attacker 460 * to infer how many packets were sent between two points in time. 461 */ 462 static u32 ip_idents_reserve(u32 hash, int segs) 463 { 464 u32 bucket, old, now = (u32)jiffies; 465 atomic_t *p_id; 466 u32 *p_tstamp; 467 u32 delta = 0; 468 469 bucket = hash & ip_idents_mask; 470 p_tstamp = ip_tstamps + bucket; 471 p_id = ip_idents + bucket; 472 old = READ_ONCE(*p_tstamp); 473 474 if (old != now && cmpxchg(p_tstamp, old, now) == old) 475 delta = get_random_u32_below(now - old); 476 477 /* If UBSAN reports an error there, please make sure your compiler 478 * supports -fno-strict-overflow before reporting it that was a bug 479 * in UBSAN, and it has been fixed in GCC-8. 480 */ 481 return atomic_add_return(segs + delta, p_id) - segs; 482 } 483 484 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs) 485 { 486 u32 hash, id; 487 488 /* Note the following code is not safe, but this is okay. */ 489 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key))) 490 get_random_bytes(&net->ipv4.ip_id_key, 491 sizeof(net->ipv4.ip_id_key)); 492 493 hash = siphash_3u32((__force u32)iph->daddr, 494 (__force u32)iph->saddr, 495 iph->protocol, 496 &net->ipv4.ip_id_key); 497 id = ip_idents_reserve(hash, segs); 498 iph->id = htons(id); 499 } 500 EXPORT_SYMBOL(__ip_select_ident); 501 502 static void ip_rt_fix_tos(struct flowi4 *fl4) 503 { 504 __u8 tos = RT_FL_TOS(fl4); 505 506 fl4->flowi4_tos = tos & IPTOS_RT_MASK; 507 if (tos & RTO_ONLINK) 508 fl4->flowi4_scope = RT_SCOPE_LINK; 509 } 510 511 static void __build_flow_key(const struct net *net, struct flowi4 *fl4, 512 const struct sock *sk, const struct iphdr *iph, 513 int oif, __u8 tos, u8 prot, u32 mark, 514 int flow_flags) 515 { 516 __u8 scope = RT_SCOPE_UNIVERSE; 517 518 if (sk) { 519 oif = sk->sk_bound_dev_if; 520 mark = READ_ONCE(sk->sk_mark); 521 tos = ip_sock_rt_tos(sk); 522 scope = ip_sock_rt_scope(sk); 523 prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW : 524 sk->sk_protocol; 525 } 526 527 flowi4_init_output(fl4, oif, mark, tos & IPTOS_RT_MASK, scope, 528 prot, flow_flags, iph->daddr, iph->saddr, 0, 0, 529 sock_net_uid(net, sk)); 530 } 531 532 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb, 533 const struct sock *sk) 534 { 535 const struct net *net = dev_net(skb->dev); 536 const struct iphdr *iph = ip_hdr(skb); 537 int oif = skb->dev->ifindex; 538 u8 prot = iph->protocol; 539 u32 mark = skb->mark; 540 __u8 tos = iph->tos; 541 542 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0); 543 } 544 545 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk) 546 { 547 const struct inet_sock *inet = inet_sk(sk); 548 const struct ip_options_rcu *inet_opt; 549 __be32 daddr = inet->inet_daddr; 550 551 rcu_read_lock(); 552 inet_opt = rcu_dereference(inet->inet_opt); 553 if (inet_opt && inet_opt->opt.srr) 554 daddr = inet_opt->opt.faddr; 555 flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark), 556 ip_sock_rt_tos(sk) & IPTOS_RT_MASK, 557 ip_sock_rt_scope(sk), 558 inet_test_bit(HDRINCL, sk) ? 559 IPPROTO_RAW : sk->sk_protocol, 560 inet_sk_flowi_flags(sk), 561 daddr, inet->inet_saddr, 0, 0, sk->sk_uid); 562 rcu_read_unlock(); 563 } 564 565 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk, 566 const struct sk_buff *skb) 567 { 568 if (skb) 569 build_skb_flow_key(fl4, skb, sk); 570 else 571 build_sk_flow_key(fl4, sk); 572 } 573 574 static DEFINE_SPINLOCK(fnhe_lock); 575 576 static void fnhe_flush_routes(struct fib_nh_exception *fnhe) 577 { 578 struct rtable *rt; 579 580 rt = rcu_dereference(fnhe->fnhe_rth_input); 581 if (rt) { 582 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL); 583 dst_dev_put(&rt->dst); 584 dst_release(&rt->dst); 585 } 586 rt = rcu_dereference(fnhe->fnhe_rth_output); 587 if (rt) { 588 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL); 589 dst_dev_put(&rt->dst); 590 dst_release(&rt->dst); 591 } 592 } 593 594 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash) 595 { 596 struct fib_nh_exception __rcu **fnhe_p, **oldest_p; 597 struct fib_nh_exception *fnhe, *oldest = NULL; 598 599 for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) { 600 fnhe = rcu_dereference_protected(*fnhe_p, 601 lockdep_is_held(&fnhe_lock)); 602 if (!fnhe) 603 break; 604 if (!oldest || 605 time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) { 606 oldest = fnhe; 607 oldest_p = fnhe_p; 608 } 609 } 610 fnhe_flush_routes(oldest); 611 *oldest_p = oldest->fnhe_next; 612 kfree_rcu(oldest, rcu); 613 } 614 615 static u32 fnhe_hashfun(__be32 daddr) 616 { 617 static siphash_aligned_key_t fnhe_hash_key; 618 u64 hval; 619 620 net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key)); 621 hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key); 622 return hash_64(hval, FNHE_HASH_SHIFT); 623 } 624 625 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe) 626 { 627 rt->rt_pmtu = fnhe->fnhe_pmtu; 628 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked; 629 rt->dst.expires = fnhe->fnhe_expires; 630 631 if (fnhe->fnhe_gw) { 632 rt->rt_flags |= RTCF_REDIRECTED; 633 rt->rt_uses_gateway = 1; 634 rt->rt_gw_family = AF_INET; 635 rt->rt_gw4 = fnhe->fnhe_gw; 636 } 637 } 638 639 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr, 640 __be32 gw, u32 pmtu, bool lock, 641 unsigned long expires) 642 { 643 struct fnhe_hash_bucket *hash; 644 struct fib_nh_exception *fnhe; 645 struct rtable *rt; 646 u32 genid, hval; 647 unsigned int i; 648 int depth; 649 650 genid = fnhe_genid(dev_net(nhc->nhc_dev)); 651 hval = fnhe_hashfun(daddr); 652 653 spin_lock_bh(&fnhe_lock); 654 655 hash = rcu_dereference(nhc->nhc_exceptions); 656 if (!hash) { 657 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC); 658 if (!hash) 659 goto out_unlock; 660 rcu_assign_pointer(nhc->nhc_exceptions, hash); 661 } 662 663 hash += hval; 664 665 depth = 0; 666 for (fnhe = rcu_dereference(hash->chain); fnhe; 667 fnhe = rcu_dereference(fnhe->fnhe_next)) { 668 if (fnhe->fnhe_daddr == daddr) 669 break; 670 depth++; 671 } 672 673 if (fnhe) { 674 if (fnhe->fnhe_genid != genid) 675 fnhe->fnhe_genid = genid; 676 if (gw) 677 fnhe->fnhe_gw = gw; 678 if (pmtu) { 679 fnhe->fnhe_pmtu = pmtu; 680 fnhe->fnhe_mtu_locked = lock; 681 } 682 fnhe->fnhe_expires = max(1UL, expires); 683 /* Update all cached dsts too */ 684 rt = rcu_dereference(fnhe->fnhe_rth_input); 685 if (rt) 686 fill_route_from_fnhe(rt, fnhe); 687 rt = rcu_dereference(fnhe->fnhe_rth_output); 688 if (rt) 689 fill_route_from_fnhe(rt, fnhe); 690 } else { 691 /* Randomize max depth to avoid some side channels attacks. */ 692 int max_depth = FNHE_RECLAIM_DEPTH + 693 get_random_u32_below(FNHE_RECLAIM_DEPTH); 694 695 while (depth > max_depth) { 696 fnhe_remove_oldest(hash); 697 depth--; 698 } 699 700 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC); 701 if (!fnhe) 702 goto out_unlock; 703 704 fnhe->fnhe_next = hash->chain; 705 706 fnhe->fnhe_genid = genid; 707 fnhe->fnhe_daddr = daddr; 708 fnhe->fnhe_gw = gw; 709 fnhe->fnhe_pmtu = pmtu; 710 fnhe->fnhe_mtu_locked = lock; 711 fnhe->fnhe_expires = max(1UL, expires); 712 713 rcu_assign_pointer(hash->chain, fnhe); 714 715 /* Exception created; mark the cached routes for the nexthop 716 * stale, so anyone caching it rechecks if this exception 717 * applies to them. 718 */ 719 rt = rcu_dereference(nhc->nhc_rth_input); 720 if (rt) 721 rt->dst.obsolete = DST_OBSOLETE_KILL; 722 723 for_each_possible_cpu(i) { 724 struct rtable __rcu **prt; 725 726 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i); 727 rt = rcu_dereference(*prt); 728 if (rt) 729 rt->dst.obsolete = DST_OBSOLETE_KILL; 730 } 731 } 732 733 fnhe->fnhe_stamp = jiffies; 734 735 out_unlock: 736 spin_unlock_bh(&fnhe_lock); 737 } 738 739 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4, 740 bool kill_route) 741 { 742 __be32 new_gw = icmp_hdr(skb)->un.gateway; 743 __be32 old_gw = ip_hdr(skb)->saddr; 744 struct net_device *dev = skb->dev; 745 struct in_device *in_dev; 746 struct fib_result res; 747 struct neighbour *n; 748 struct net *net; 749 750 switch (icmp_hdr(skb)->code & 7) { 751 case ICMP_REDIR_NET: 752 case ICMP_REDIR_NETTOS: 753 case ICMP_REDIR_HOST: 754 case ICMP_REDIR_HOSTTOS: 755 break; 756 757 default: 758 return; 759 } 760 761 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw) 762 return; 763 764 in_dev = __in_dev_get_rcu(dev); 765 if (!in_dev) 766 return; 767 768 net = dev_net(dev); 769 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) || 770 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) || 771 ipv4_is_zeronet(new_gw)) 772 goto reject_redirect; 773 774 if (!IN_DEV_SHARED_MEDIA(in_dev)) { 775 if (!inet_addr_onlink(in_dev, new_gw, old_gw)) 776 goto reject_redirect; 777 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) 778 goto reject_redirect; 779 } else { 780 if (inet_addr_type(net, new_gw) != RTN_UNICAST) 781 goto reject_redirect; 782 } 783 784 n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw); 785 if (!n) 786 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev); 787 if (!IS_ERR(n)) { 788 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) { 789 neigh_event_send(n, NULL); 790 } else { 791 if (fib_lookup(net, fl4, &res, 0) == 0) { 792 struct fib_nh_common *nhc; 793 794 fib_select_path(net, &res, fl4, skb); 795 nhc = FIB_RES_NHC(res); 796 update_or_create_fnhe(nhc, fl4->daddr, new_gw, 797 0, false, 798 jiffies + ip_rt_gc_timeout); 799 } 800 if (kill_route) 801 rt->dst.obsolete = DST_OBSOLETE_KILL; 802 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n); 803 } 804 neigh_release(n); 805 } 806 return; 807 808 reject_redirect: 809 #ifdef CONFIG_IP_ROUTE_VERBOSE 810 if (IN_DEV_LOG_MARTIANS(in_dev)) { 811 const struct iphdr *iph = (const struct iphdr *) skb->data; 812 __be32 daddr = iph->daddr; 813 __be32 saddr = iph->saddr; 814 815 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n" 816 " Advised path = %pI4 -> %pI4\n", 817 &old_gw, dev->name, &new_gw, 818 &saddr, &daddr); 819 } 820 #endif 821 ; 822 } 823 824 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb) 825 { 826 struct rtable *rt; 827 struct flowi4 fl4; 828 const struct iphdr *iph = (const struct iphdr *) skb->data; 829 struct net *net = dev_net(skb->dev); 830 int oif = skb->dev->ifindex; 831 u8 prot = iph->protocol; 832 u32 mark = skb->mark; 833 __u8 tos = iph->tos; 834 835 rt = (struct rtable *) dst; 836 837 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0); 838 __ip_do_redirect(rt, skb, &fl4, true); 839 } 840 841 static void ipv4_negative_advice(struct sock *sk, 842 struct dst_entry *dst) 843 { 844 struct rtable *rt = (struct rtable *)dst; 845 846 if ((dst->obsolete > 0) || 847 (rt->rt_flags & RTCF_REDIRECTED) || 848 rt->dst.expires) 849 sk_dst_reset(sk); 850 } 851 852 /* 853 * Algorithm: 854 * 1. The first ip_rt_redirect_number redirects are sent 855 * with exponential backoff, then we stop sending them at all, 856 * assuming that the host ignores our redirects. 857 * 2. If we did not see packets requiring redirects 858 * during ip_rt_redirect_silence, we assume that the host 859 * forgot redirected route and start to send redirects again. 860 * 861 * This algorithm is much cheaper and more intelligent than dumb load limiting 862 * in icmp.c. 863 * 864 * NOTE. Do not forget to inhibit load limiting for redirects (redundant) 865 * and "frag. need" (breaks PMTU discovery) in icmp.c. 866 */ 867 868 void ip_rt_send_redirect(struct sk_buff *skb) 869 { 870 struct rtable *rt = skb_rtable(skb); 871 struct in_device *in_dev; 872 struct inet_peer *peer; 873 struct net *net; 874 int log_martians; 875 int vif; 876 877 rcu_read_lock(); 878 in_dev = __in_dev_get_rcu(rt->dst.dev); 879 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) { 880 rcu_read_unlock(); 881 return; 882 } 883 log_martians = IN_DEV_LOG_MARTIANS(in_dev); 884 vif = l3mdev_master_ifindex_rcu(rt->dst.dev); 885 886 net = dev_net(rt->dst.dev); 887 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif); 888 if (!peer) { 889 rcu_read_unlock(); 890 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, 891 rt_nexthop(rt, ip_hdr(skb)->daddr)); 892 return; 893 } 894 895 /* No redirected packets during ip_rt_redirect_silence; 896 * reset the algorithm. 897 */ 898 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) { 899 peer->rate_tokens = 0; 900 peer->n_redirects = 0; 901 } 902 903 /* Too many ignored redirects; do not send anything 904 * set dst.rate_last to the last seen redirected packet. 905 */ 906 if (peer->n_redirects >= ip_rt_redirect_number) { 907 peer->rate_last = jiffies; 908 goto out_unlock; 909 } 910 911 /* Check for load limit; set rate_last to the latest sent 912 * redirect. 913 */ 914 if (peer->n_redirects == 0 || 915 time_after(jiffies, 916 (peer->rate_last + 917 (ip_rt_redirect_load << peer->n_redirects)))) { 918 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr); 919 920 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw); 921 peer->rate_last = jiffies; 922 ++peer->n_redirects; 923 if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians && 924 peer->n_redirects == ip_rt_redirect_number) 925 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n", 926 &ip_hdr(skb)->saddr, inet_iif(skb), 927 &ip_hdr(skb)->daddr, &gw); 928 } 929 out_unlock: 930 rcu_read_unlock(); 931 } 932 933 static int ip_error(struct sk_buff *skb) 934 { 935 struct rtable *rt = skb_rtable(skb); 936 struct net_device *dev = skb->dev; 937 struct in_device *in_dev; 938 struct inet_peer *peer; 939 unsigned long now; 940 struct net *net; 941 SKB_DR(reason); 942 bool send; 943 int code; 944 945 if (netif_is_l3_master(skb->dev)) { 946 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif); 947 if (!dev) 948 goto out; 949 } 950 951 in_dev = __in_dev_get_rcu(dev); 952 953 /* IP on this device is disabled. */ 954 if (!in_dev) 955 goto out; 956 957 net = dev_net(rt->dst.dev); 958 if (!IN_DEV_FORWARD(in_dev)) { 959 switch (rt->dst.error) { 960 case EHOSTUNREACH: 961 SKB_DR_SET(reason, IP_INADDRERRORS); 962 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS); 963 break; 964 965 case ENETUNREACH: 966 SKB_DR_SET(reason, IP_INNOROUTES); 967 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 968 break; 969 } 970 goto out; 971 } 972 973 switch (rt->dst.error) { 974 case EINVAL: 975 default: 976 goto out; 977 case EHOSTUNREACH: 978 code = ICMP_HOST_UNREACH; 979 break; 980 case ENETUNREACH: 981 code = ICMP_NET_UNREACH; 982 SKB_DR_SET(reason, IP_INNOROUTES); 983 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 984 break; 985 case EACCES: 986 code = ICMP_PKT_FILTERED; 987 break; 988 } 989 990 rcu_read_lock(); 991 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 992 l3mdev_master_ifindex_rcu(skb->dev)); 993 send = true; 994 if (peer) { 995 now = jiffies; 996 peer->rate_tokens += now - peer->rate_last; 997 if (peer->rate_tokens > ip_rt_error_burst) 998 peer->rate_tokens = ip_rt_error_burst; 999 peer->rate_last = now; 1000 if (peer->rate_tokens >= ip_rt_error_cost) 1001 peer->rate_tokens -= ip_rt_error_cost; 1002 else 1003 send = false; 1004 } 1005 rcu_read_unlock(); 1006 1007 if (send) 1008 icmp_send(skb, ICMP_DEST_UNREACH, code, 0); 1009 1010 out: kfree_skb_reason(skb, reason); 1011 return 0; 1012 } 1013 1014 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu) 1015 { 1016 struct dst_entry *dst = &rt->dst; 1017 struct net *net = dev_net(dst->dev); 1018 struct fib_result res; 1019 bool lock = false; 1020 u32 old_mtu; 1021 1022 if (ip_mtu_locked(dst)) 1023 return; 1024 1025 old_mtu = ipv4_mtu(dst); 1026 if (old_mtu < mtu) 1027 return; 1028 1029 if (mtu < net->ipv4.ip_rt_min_pmtu) { 1030 lock = true; 1031 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu); 1032 } 1033 1034 if (rt->rt_pmtu == mtu && !lock && 1035 time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2)) 1036 return; 1037 1038 rcu_read_lock(); 1039 if (fib_lookup(net, fl4, &res, 0) == 0) { 1040 struct fib_nh_common *nhc; 1041 1042 fib_select_path(net, &res, fl4, NULL); 1043 nhc = FIB_RES_NHC(res); 1044 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1045 jiffies + net->ipv4.ip_rt_mtu_expires); 1046 } 1047 rcu_read_unlock(); 1048 } 1049 1050 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 1051 struct sk_buff *skb, u32 mtu, 1052 bool confirm_neigh) 1053 { 1054 struct rtable *rt = (struct rtable *) dst; 1055 struct flowi4 fl4; 1056 1057 ip_rt_build_flow_key(&fl4, sk, skb); 1058 1059 /* Don't make lookup fail for bridged encapsulations */ 1060 if (skb && netif_is_any_bridge_port(skb->dev)) 1061 fl4.flowi4_oif = 0; 1062 1063 __ip_rt_update_pmtu(rt, &fl4, mtu); 1064 } 1065 1066 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, 1067 int oif, u8 protocol) 1068 { 1069 const struct iphdr *iph = (const struct iphdr *)skb->data; 1070 struct flowi4 fl4; 1071 struct rtable *rt; 1072 u32 mark = IP4_REPLY_MARK(net, skb->mark); 1073 1074 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark, 1075 0); 1076 rt = __ip_route_output_key(net, &fl4); 1077 if (!IS_ERR(rt)) { 1078 __ip_rt_update_pmtu(rt, &fl4, mtu); 1079 ip_rt_put(rt); 1080 } 1081 } 1082 EXPORT_SYMBOL_GPL(ipv4_update_pmtu); 1083 1084 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1085 { 1086 const struct iphdr *iph = (const struct iphdr *)skb->data; 1087 struct flowi4 fl4; 1088 struct rtable *rt; 1089 1090 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0); 1091 1092 if (!fl4.flowi4_mark) 1093 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark); 1094 1095 rt = __ip_route_output_key(sock_net(sk), &fl4); 1096 if (!IS_ERR(rt)) { 1097 __ip_rt_update_pmtu(rt, &fl4, mtu); 1098 ip_rt_put(rt); 1099 } 1100 } 1101 1102 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1103 { 1104 const struct iphdr *iph = (const struct iphdr *)skb->data; 1105 struct flowi4 fl4; 1106 struct rtable *rt; 1107 struct dst_entry *odst = NULL; 1108 bool new = false; 1109 struct net *net = sock_net(sk); 1110 1111 bh_lock_sock(sk); 1112 1113 if (!ip_sk_accept_pmtu(sk)) 1114 goto out; 1115 1116 odst = sk_dst_get(sk); 1117 1118 if (sock_owned_by_user(sk) || !odst) { 1119 __ipv4_sk_update_pmtu(skb, sk, mtu); 1120 goto out; 1121 } 1122 1123 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1124 1125 rt = (struct rtable *)odst; 1126 if (odst->obsolete && !odst->ops->check(odst, 0)) { 1127 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1128 if (IS_ERR(rt)) 1129 goto out; 1130 1131 new = true; 1132 } 1133 1134 __ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu); 1135 1136 if (!dst_check(&rt->dst, 0)) { 1137 if (new) 1138 dst_release(&rt->dst); 1139 1140 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1141 if (IS_ERR(rt)) 1142 goto out; 1143 1144 new = true; 1145 } 1146 1147 if (new) 1148 sk_dst_set(sk, &rt->dst); 1149 1150 out: 1151 bh_unlock_sock(sk); 1152 dst_release(odst); 1153 } 1154 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu); 1155 1156 void ipv4_redirect(struct sk_buff *skb, struct net *net, 1157 int oif, u8 protocol) 1158 { 1159 const struct iphdr *iph = (const struct iphdr *)skb->data; 1160 struct flowi4 fl4; 1161 struct rtable *rt; 1162 1163 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0); 1164 rt = __ip_route_output_key(net, &fl4); 1165 if (!IS_ERR(rt)) { 1166 __ip_do_redirect(rt, skb, &fl4, false); 1167 ip_rt_put(rt); 1168 } 1169 } 1170 EXPORT_SYMBOL_GPL(ipv4_redirect); 1171 1172 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk) 1173 { 1174 const struct iphdr *iph = (const struct iphdr *)skb->data; 1175 struct flowi4 fl4; 1176 struct rtable *rt; 1177 struct net *net = sock_net(sk); 1178 1179 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1180 rt = __ip_route_output_key(net, &fl4); 1181 if (!IS_ERR(rt)) { 1182 __ip_do_redirect(rt, skb, &fl4, false); 1183 ip_rt_put(rt); 1184 } 1185 } 1186 EXPORT_SYMBOL_GPL(ipv4_sk_redirect); 1187 1188 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst, 1189 u32 cookie) 1190 { 1191 struct rtable *rt = (struct rtable *) dst; 1192 1193 /* All IPV4 dsts are created with ->obsolete set to the value 1194 * DST_OBSOLETE_FORCE_CHK which forces validation calls down 1195 * into this function always. 1196 * 1197 * When a PMTU/redirect information update invalidates a route, 1198 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or 1199 * DST_OBSOLETE_DEAD. 1200 */ 1201 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt)) 1202 return NULL; 1203 return dst; 1204 } 1205 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check); 1206 1207 static void ipv4_send_dest_unreach(struct sk_buff *skb) 1208 { 1209 struct net_device *dev; 1210 struct ip_options opt; 1211 int res; 1212 1213 /* Recompile ip options since IPCB may not be valid anymore. 1214 * Also check we have a reasonable ipv4 header. 1215 */ 1216 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) || 1217 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5) 1218 return; 1219 1220 memset(&opt, 0, sizeof(opt)); 1221 if (ip_hdr(skb)->ihl > 5) { 1222 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4)) 1223 return; 1224 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr); 1225 1226 rcu_read_lock(); 1227 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev; 1228 res = __ip_options_compile(dev_net(dev), &opt, skb, NULL); 1229 rcu_read_unlock(); 1230 1231 if (res) 1232 return; 1233 } 1234 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt); 1235 } 1236 1237 static void ipv4_link_failure(struct sk_buff *skb) 1238 { 1239 struct rtable *rt; 1240 1241 ipv4_send_dest_unreach(skb); 1242 1243 rt = skb_rtable(skb); 1244 if (rt) 1245 dst_set_expires(&rt->dst, 0); 1246 } 1247 1248 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb) 1249 { 1250 pr_debug("%s: %pI4 -> %pI4, %s\n", 1251 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1252 skb->dev ? skb->dev->name : "?"); 1253 kfree_skb(skb); 1254 WARN_ON(1); 1255 return 0; 1256 } 1257 1258 /* 1259 * We do not cache source address of outgoing interface, 1260 * because it is used only by IP RR, TS and SRR options, 1261 * so that it out of fast path. 1262 * 1263 * BTW remember: "addr" is allowed to be not aligned 1264 * in IP options! 1265 */ 1266 1267 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt) 1268 { 1269 __be32 src; 1270 1271 if (rt_is_output_route(rt)) 1272 src = ip_hdr(skb)->saddr; 1273 else { 1274 struct fib_result res; 1275 struct iphdr *iph = ip_hdr(skb); 1276 struct flowi4 fl4 = { 1277 .daddr = iph->daddr, 1278 .saddr = iph->saddr, 1279 .flowi4_tos = iph->tos & IPTOS_RT_MASK, 1280 .flowi4_oif = rt->dst.dev->ifindex, 1281 .flowi4_iif = skb->dev->ifindex, 1282 .flowi4_mark = skb->mark, 1283 }; 1284 1285 rcu_read_lock(); 1286 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0) 1287 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res); 1288 else 1289 src = inet_select_addr(rt->dst.dev, 1290 rt_nexthop(rt, iph->daddr), 1291 RT_SCOPE_UNIVERSE); 1292 rcu_read_unlock(); 1293 } 1294 memcpy(addr, &src, 4); 1295 } 1296 1297 #ifdef CONFIG_IP_ROUTE_CLASSID 1298 static void set_class_tag(struct rtable *rt, u32 tag) 1299 { 1300 if (!(rt->dst.tclassid & 0xFFFF)) 1301 rt->dst.tclassid |= tag & 0xFFFF; 1302 if (!(rt->dst.tclassid & 0xFFFF0000)) 1303 rt->dst.tclassid |= tag & 0xFFFF0000; 1304 } 1305 #endif 1306 1307 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1308 { 1309 struct net *net = dev_net(dst->dev); 1310 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr); 1311 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size, 1312 net->ipv4.ip_rt_min_advmss); 1313 1314 return min(advmss, IPV4_MAX_PMTU - header_size); 1315 } 1316 1317 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst) 1318 { 1319 return ip_dst_mtu_maybe_forward(dst, false); 1320 } 1321 EXPORT_INDIRECT_CALLABLE(ipv4_mtu); 1322 1323 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr) 1324 { 1325 struct fnhe_hash_bucket *hash; 1326 struct fib_nh_exception *fnhe, __rcu **fnhe_p; 1327 u32 hval = fnhe_hashfun(daddr); 1328 1329 spin_lock_bh(&fnhe_lock); 1330 1331 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1332 lockdep_is_held(&fnhe_lock)); 1333 hash += hval; 1334 1335 fnhe_p = &hash->chain; 1336 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock)); 1337 while (fnhe) { 1338 if (fnhe->fnhe_daddr == daddr) { 1339 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected( 1340 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock))); 1341 /* set fnhe_daddr to 0 to ensure it won't bind with 1342 * new dsts in rt_bind_exception(). 1343 */ 1344 fnhe->fnhe_daddr = 0; 1345 fnhe_flush_routes(fnhe); 1346 kfree_rcu(fnhe, rcu); 1347 break; 1348 } 1349 fnhe_p = &fnhe->fnhe_next; 1350 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1351 lockdep_is_held(&fnhe_lock)); 1352 } 1353 1354 spin_unlock_bh(&fnhe_lock); 1355 } 1356 1357 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc, 1358 __be32 daddr) 1359 { 1360 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions); 1361 struct fib_nh_exception *fnhe; 1362 u32 hval; 1363 1364 if (!hash) 1365 return NULL; 1366 1367 hval = fnhe_hashfun(daddr); 1368 1369 for (fnhe = rcu_dereference(hash[hval].chain); fnhe; 1370 fnhe = rcu_dereference(fnhe->fnhe_next)) { 1371 if (fnhe->fnhe_daddr == daddr) { 1372 if (fnhe->fnhe_expires && 1373 time_after(jiffies, fnhe->fnhe_expires)) { 1374 ip_del_fnhe(nhc, daddr); 1375 break; 1376 } 1377 return fnhe; 1378 } 1379 } 1380 return NULL; 1381 } 1382 1383 /* MTU selection: 1384 * 1. mtu on route is locked - use it 1385 * 2. mtu from nexthop exception 1386 * 3. mtu from egress device 1387 */ 1388 1389 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr) 1390 { 1391 struct fib_nh_common *nhc = res->nhc; 1392 struct net_device *dev = nhc->nhc_dev; 1393 struct fib_info *fi = res->fi; 1394 u32 mtu = 0; 1395 1396 if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) || 1397 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU)) 1398 mtu = fi->fib_mtu; 1399 1400 if (likely(!mtu)) { 1401 struct fib_nh_exception *fnhe; 1402 1403 fnhe = find_exception(nhc, daddr); 1404 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires)) 1405 mtu = fnhe->fnhe_pmtu; 1406 } 1407 1408 if (likely(!mtu)) 1409 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU); 1410 1411 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu); 1412 } 1413 1414 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, 1415 __be32 daddr, const bool do_cache) 1416 { 1417 bool ret = false; 1418 1419 spin_lock_bh(&fnhe_lock); 1420 1421 if (daddr == fnhe->fnhe_daddr) { 1422 struct rtable __rcu **porig; 1423 struct rtable *orig; 1424 int genid = fnhe_genid(dev_net(rt->dst.dev)); 1425 1426 if (rt_is_input_route(rt)) 1427 porig = &fnhe->fnhe_rth_input; 1428 else 1429 porig = &fnhe->fnhe_rth_output; 1430 orig = rcu_dereference(*porig); 1431 1432 if (fnhe->fnhe_genid != genid) { 1433 fnhe->fnhe_genid = genid; 1434 fnhe->fnhe_gw = 0; 1435 fnhe->fnhe_pmtu = 0; 1436 fnhe->fnhe_expires = 0; 1437 fnhe->fnhe_mtu_locked = false; 1438 fnhe_flush_routes(fnhe); 1439 orig = NULL; 1440 } 1441 fill_route_from_fnhe(rt, fnhe); 1442 if (!rt->rt_gw4) { 1443 rt->rt_gw4 = daddr; 1444 rt->rt_gw_family = AF_INET; 1445 } 1446 1447 if (do_cache) { 1448 dst_hold(&rt->dst); 1449 rcu_assign_pointer(*porig, rt); 1450 if (orig) { 1451 dst_dev_put(&orig->dst); 1452 dst_release(&orig->dst); 1453 } 1454 ret = true; 1455 } 1456 1457 fnhe->fnhe_stamp = jiffies; 1458 } 1459 spin_unlock_bh(&fnhe_lock); 1460 1461 return ret; 1462 } 1463 1464 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt) 1465 { 1466 struct rtable *orig, *prev, **p; 1467 bool ret = true; 1468 1469 if (rt_is_input_route(rt)) { 1470 p = (struct rtable **)&nhc->nhc_rth_input; 1471 } else { 1472 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 1473 } 1474 orig = *p; 1475 1476 /* hold dst before doing cmpxchg() to avoid race condition 1477 * on this dst 1478 */ 1479 dst_hold(&rt->dst); 1480 prev = cmpxchg(p, orig, rt); 1481 if (prev == orig) { 1482 if (orig) { 1483 rt_add_uncached_list(orig); 1484 dst_release(&orig->dst); 1485 } 1486 } else { 1487 dst_release(&rt->dst); 1488 ret = false; 1489 } 1490 1491 return ret; 1492 } 1493 1494 struct uncached_list { 1495 spinlock_t lock; 1496 struct list_head head; 1497 struct list_head quarantine; 1498 }; 1499 1500 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list); 1501 1502 void rt_add_uncached_list(struct rtable *rt) 1503 { 1504 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list); 1505 1506 rt->dst.rt_uncached_list = ul; 1507 1508 spin_lock_bh(&ul->lock); 1509 list_add_tail(&rt->dst.rt_uncached, &ul->head); 1510 spin_unlock_bh(&ul->lock); 1511 } 1512 1513 void rt_del_uncached_list(struct rtable *rt) 1514 { 1515 if (!list_empty(&rt->dst.rt_uncached)) { 1516 struct uncached_list *ul = rt->dst.rt_uncached_list; 1517 1518 spin_lock_bh(&ul->lock); 1519 list_del_init(&rt->dst.rt_uncached); 1520 spin_unlock_bh(&ul->lock); 1521 } 1522 } 1523 1524 static void ipv4_dst_destroy(struct dst_entry *dst) 1525 { 1526 struct rtable *rt = (struct rtable *)dst; 1527 1528 ip_dst_metrics_put(dst); 1529 rt_del_uncached_list(rt); 1530 } 1531 1532 void rt_flush_dev(struct net_device *dev) 1533 { 1534 struct rtable *rt, *safe; 1535 int cpu; 1536 1537 for_each_possible_cpu(cpu) { 1538 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 1539 1540 if (list_empty(&ul->head)) 1541 continue; 1542 1543 spin_lock_bh(&ul->lock); 1544 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) { 1545 if (rt->dst.dev != dev) 1546 continue; 1547 rt->dst.dev = blackhole_netdev; 1548 netdev_ref_replace(dev, blackhole_netdev, 1549 &rt->dst.dev_tracker, GFP_ATOMIC); 1550 list_move(&rt->dst.rt_uncached, &ul->quarantine); 1551 } 1552 spin_unlock_bh(&ul->lock); 1553 } 1554 } 1555 1556 static bool rt_cache_valid(const struct rtable *rt) 1557 { 1558 return rt && 1559 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK && 1560 !rt_is_expired(rt); 1561 } 1562 1563 static void rt_set_nexthop(struct rtable *rt, __be32 daddr, 1564 const struct fib_result *res, 1565 struct fib_nh_exception *fnhe, 1566 struct fib_info *fi, u16 type, u32 itag, 1567 const bool do_cache) 1568 { 1569 bool cached = false; 1570 1571 if (fi) { 1572 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1573 1574 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) { 1575 rt->rt_uses_gateway = 1; 1576 rt->rt_gw_family = nhc->nhc_gw_family; 1577 /* only INET and INET6 are supported */ 1578 if (likely(nhc->nhc_gw_family == AF_INET)) 1579 rt->rt_gw4 = nhc->nhc_gw.ipv4; 1580 else 1581 rt->rt_gw6 = nhc->nhc_gw.ipv6; 1582 } 1583 1584 ip_dst_init_metrics(&rt->dst, fi->fib_metrics); 1585 1586 #ifdef CONFIG_IP_ROUTE_CLASSID 1587 if (nhc->nhc_family == AF_INET) { 1588 struct fib_nh *nh; 1589 1590 nh = container_of(nhc, struct fib_nh, nh_common); 1591 rt->dst.tclassid = nh->nh_tclassid; 1592 } 1593 #endif 1594 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 1595 if (unlikely(fnhe)) 1596 cached = rt_bind_exception(rt, fnhe, daddr, do_cache); 1597 else if (do_cache) 1598 cached = rt_cache_route(nhc, rt); 1599 if (unlikely(!cached)) { 1600 /* Routes we intend to cache in nexthop exception or 1601 * FIB nexthop have the DST_NOCACHE bit clear. 1602 * However, if we are unsuccessful at storing this 1603 * route into the cache we really need to set it. 1604 */ 1605 if (!rt->rt_gw4) { 1606 rt->rt_gw_family = AF_INET; 1607 rt->rt_gw4 = daddr; 1608 } 1609 rt_add_uncached_list(rt); 1610 } 1611 } else 1612 rt_add_uncached_list(rt); 1613 1614 #ifdef CONFIG_IP_ROUTE_CLASSID 1615 #ifdef CONFIG_IP_MULTIPLE_TABLES 1616 set_class_tag(rt, res->tclassid); 1617 #endif 1618 set_class_tag(rt, itag); 1619 #endif 1620 } 1621 1622 struct rtable *rt_dst_alloc(struct net_device *dev, 1623 unsigned int flags, u16 type, 1624 bool noxfrm) 1625 { 1626 struct rtable *rt; 1627 1628 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1629 (noxfrm ? DST_NOXFRM : 0)); 1630 1631 if (rt) { 1632 rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1633 rt->rt_flags = flags; 1634 rt->rt_type = type; 1635 rt->rt_is_input = 0; 1636 rt->rt_iif = 0; 1637 rt->rt_pmtu = 0; 1638 rt->rt_mtu_locked = 0; 1639 rt->rt_uses_gateway = 0; 1640 rt->rt_gw_family = 0; 1641 rt->rt_gw4 = 0; 1642 1643 rt->dst.output = ip_output; 1644 if (flags & RTCF_LOCAL) 1645 rt->dst.input = ip_local_deliver; 1646 } 1647 1648 return rt; 1649 } 1650 EXPORT_SYMBOL(rt_dst_alloc); 1651 1652 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt) 1653 { 1654 struct rtable *new_rt; 1655 1656 new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1657 rt->dst.flags); 1658 1659 if (new_rt) { 1660 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1661 new_rt->rt_flags = rt->rt_flags; 1662 new_rt->rt_type = rt->rt_type; 1663 new_rt->rt_is_input = rt->rt_is_input; 1664 new_rt->rt_iif = rt->rt_iif; 1665 new_rt->rt_pmtu = rt->rt_pmtu; 1666 new_rt->rt_mtu_locked = rt->rt_mtu_locked; 1667 new_rt->rt_gw_family = rt->rt_gw_family; 1668 if (rt->rt_gw_family == AF_INET) 1669 new_rt->rt_gw4 = rt->rt_gw4; 1670 else if (rt->rt_gw_family == AF_INET6) 1671 new_rt->rt_gw6 = rt->rt_gw6; 1672 1673 new_rt->dst.input = rt->dst.input; 1674 new_rt->dst.output = rt->dst.output; 1675 new_rt->dst.error = rt->dst.error; 1676 new_rt->dst.lastuse = jiffies; 1677 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate); 1678 } 1679 return new_rt; 1680 } 1681 EXPORT_SYMBOL(rt_dst_clone); 1682 1683 /* called in rcu_read_lock() section */ 1684 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1685 u8 tos, struct net_device *dev, 1686 struct in_device *in_dev, u32 *itag) 1687 { 1688 int err; 1689 1690 /* Primary sanity checks. */ 1691 if (!in_dev) 1692 return -EINVAL; 1693 1694 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) || 1695 skb->protocol != htons(ETH_P_IP)) 1696 return -EINVAL; 1697 1698 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev)) 1699 return -EINVAL; 1700 1701 if (ipv4_is_zeronet(saddr)) { 1702 if (!ipv4_is_local_multicast(daddr) && 1703 ip_hdr(skb)->protocol != IPPROTO_IGMP) 1704 return -EINVAL; 1705 } else { 1706 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 1707 in_dev, itag); 1708 if (err < 0) 1709 return err; 1710 } 1711 return 0; 1712 } 1713 1714 /* called in rcu_read_lock() section */ 1715 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1716 u8 tos, struct net_device *dev, int our) 1717 { 1718 struct in_device *in_dev = __in_dev_get_rcu(dev); 1719 unsigned int flags = RTCF_MULTICAST; 1720 struct rtable *rth; 1721 u32 itag = 0; 1722 int err; 1723 1724 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag); 1725 if (err) 1726 return err; 1727 1728 if (our) 1729 flags |= RTCF_LOCAL; 1730 1731 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 1732 IPCB(skb)->flags |= IPSKB_NOPOLICY; 1733 1734 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST, 1735 false); 1736 if (!rth) 1737 return -ENOBUFS; 1738 1739 #ifdef CONFIG_IP_ROUTE_CLASSID 1740 rth->dst.tclassid = itag; 1741 #endif 1742 rth->dst.output = ip_rt_bug; 1743 rth->rt_is_input= 1; 1744 1745 #ifdef CONFIG_IP_MROUTE 1746 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1747 rth->dst.input = ip_mr_input; 1748 #endif 1749 RT_CACHE_STAT_INC(in_slow_mc); 1750 1751 skb_dst_drop(skb); 1752 skb_dst_set(skb, &rth->dst); 1753 return 0; 1754 } 1755 1756 1757 static void ip_handle_martian_source(struct net_device *dev, 1758 struct in_device *in_dev, 1759 struct sk_buff *skb, 1760 __be32 daddr, 1761 __be32 saddr) 1762 { 1763 RT_CACHE_STAT_INC(in_martian_src); 1764 #ifdef CONFIG_IP_ROUTE_VERBOSE 1765 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1766 /* 1767 * RFC1812 recommendation, if source is martian, 1768 * the only hint is MAC header. 1769 */ 1770 pr_warn("martian source %pI4 from %pI4, on dev %s\n", 1771 &daddr, &saddr, dev->name); 1772 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1773 print_hex_dump(KERN_WARNING, "ll header: ", 1774 DUMP_PREFIX_OFFSET, 16, 1, 1775 skb_mac_header(skb), 1776 dev->hard_header_len, false); 1777 } 1778 } 1779 #endif 1780 } 1781 1782 /* called in rcu_read_lock() section */ 1783 static int __mkroute_input(struct sk_buff *skb, 1784 const struct fib_result *res, 1785 struct in_device *in_dev, 1786 __be32 daddr, __be32 saddr, u32 tos) 1787 { 1788 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1789 struct net_device *dev = nhc->nhc_dev; 1790 struct fib_nh_exception *fnhe; 1791 struct rtable *rth; 1792 int err; 1793 struct in_device *out_dev; 1794 bool do_cache; 1795 u32 itag = 0; 1796 1797 /* get a working reference to the output device */ 1798 out_dev = __in_dev_get_rcu(dev); 1799 if (!out_dev) { 1800 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n"); 1801 return -EINVAL; 1802 } 1803 1804 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res), 1805 in_dev->dev, in_dev, &itag); 1806 if (err < 0) { 1807 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1808 saddr); 1809 1810 goto cleanup; 1811 } 1812 1813 do_cache = res->fi && !itag; 1814 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) && 1815 skb->protocol == htons(ETH_P_IP)) { 1816 __be32 gw; 1817 1818 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0; 1819 if (IN_DEV_SHARED_MEDIA(out_dev) || 1820 inet_addr_onlink(out_dev, saddr, gw)) 1821 IPCB(skb)->flags |= IPSKB_DOREDIRECT; 1822 } 1823 1824 if (skb->protocol != htons(ETH_P_IP)) { 1825 /* Not IP (i.e. ARP). Do not create route, if it is 1826 * invalid for proxy arp. DNAT routes are always valid. 1827 * 1828 * Proxy arp feature have been extended to allow, ARP 1829 * replies back to the same interface, to support 1830 * Private VLAN switch technologies. See arp.c. 1831 */ 1832 if (out_dev == in_dev && 1833 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 1834 err = -EINVAL; 1835 goto cleanup; 1836 } 1837 } 1838 1839 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 1840 IPCB(skb)->flags |= IPSKB_NOPOLICY; 1841 1842 fnhe = find_exception(nhc, daddr); 1843 if (do_cache) { 1844 if (fnhe) 1845 rth = rcu_dereference(fnhe->fnhe_rth_input); 1846 else 1847 rth = rcu_dereference(nhc->nhc_rth_input); 1848 if (rt_cache_valid(rth)) { 1849 skb_dst_set_noref(skb, &rth->dst); 1850 goto out; 1851 } 1852 } 1853 1854 rth = rt_dst_alloc(out_dev->dev, 0, res->type, 1855 IN_DEV_ORCONF(out_dev, NOXFRM)); 1856 if (!rth) { 1857 err = -ENOBUFS; 1858 goto cleanup; 1859 } 1860 1861 rth->rt_is_input = 1; 1862 RT_CACHE_STAT_INC(in_slow_tot); 1863 1864 rth->dst.input = ip_forward; 1865 1866 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag, 1867 do_cache); 1868 lwtunnel_set_redirect(&rth->dst); 1869 skb_dst_set(skb, &rth->dst); 1870 out: 1871 err = 0; 1872 cleanup: 1873 return err; 1874 } 1875 1876 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1877 /* To make ICMP packets follow the right flow, the multipath hash is 1878 * calculated from the inner IP addresses. 1879 */ 1880 static void ip_multipath_l3_keys(const struct sk_buff *skb, 1881 struct flow_keys *hash_keys) 1882 { 1883 const struct iphdr *outer_iph = ip_hdr(skb); 1884 const struct iphdr *key_iph = outer_iph; 1885 const struct iphdr *inner_iph; 1886 const struct icmphdr *icmph; 1887 struct iphdr _inner_iph; 1888 struct icmphdr _icmph; 1889 1890 if (likely(outer_iph->protocol != IPPROTO_ICMP)) 1891 goto out; 1892 1893 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0)) 1894 goto out; 1895 1896 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph), 1897 &_icmph); 1898 if (!icmph) 1899 goto out; 1900 1901 if (!icmp_is_err(icmph->type)) 1902 goto out; 1903 1904 inner_iph = skb_header_pointer(skb, 1905 outer_iph->ihl * 4 + sizeof(_icmph), 1906 sizeof(_inner_iph), &_inner_iph); 1907 if (!inner_iph) 1908 goto out; 1909 1910 key_iph = inner_iph; 1911 out: 1912 hash_keys->addrs.v4addrs.src = key_iph->saddr; 1913 hash_keys->addrs.v4addrs.dst = key_iph->daddr; 1914 } 1915 1916 static u32 fib_multipath_custom_hash_outer(const struct net *net, 1917 const struct sk_buff *skb, 1918 bool *p_has_inner) 1919 { 1920 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 1921 struct flow_keys keys, hash_keys; 1922 1923 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 1924 return 0; 1925 1926 memset(&hash_keys, 0, sizeof(hash_keys)); 1927 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP); 1928 1929 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1930 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 1931 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 1932 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 1933 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 1934 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 1935 hash_keys.basic.ip_proto = keys.basic.ip_proto; 1936 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) 1937 hash_keys.ports.src = keys.ports.src; 1938 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 1939 hash_keys.ports.dst = keys.ports.dst; 1940 1941 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION); 1942 return flow_hash_from_keys(&hash_keys); 1943 } 1944 1945 static u32 fib_multipath_custom_hash_inner(const struct net *net, 1946 const struct sk_buff *skb, 1947 bool has_inner) 1948 { 1949 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 1950 struct flow_keys keys, hash_keys; 1951 1952 /* We assume the packet carries an encapsulation, but if none was 1953 * encountered during dissection of the outer flow, then there is no 1954 * point in calling the flow dissector again. 1955 */ 1956 if (!has_inner) 1957 return 0; 1958 1959 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK)) 1960 return 0; 1961 1962 memset(&hash_keys, 0, sizeof(hash_keys)); 1963 skb_flow_dissect_flow_keys(skb, &keys, 0); 1964 1965 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION)) 1966 return 0; 1967 1968 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 1969 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1970 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 1971 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 1972 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 1973 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 1974 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 1975 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 1976 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 1977 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 1978 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 1979 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 1980 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL) 1981 hash_keys.tags.flow_label = keys.tags.flow_label; 1982 } 1983 1984 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO) 1985 hash_keys.basic.ip_proto = keys.basic.ip_proto; 1986 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT) 1987 hash_keys.ports.src = keys.ports.src; 1988 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT) 1989 hash_keys.ports.dst = keys.ports.dst; 1990 1991 return flow_hash_from_keys(&hash_keys); 1992 } 1993 1994 static u32 fib_multipath_custom_hash_skb(const struct net *net, 1995 const struct sk_buff *skb) 1996 { 1997 u32 mhash, mhash_inner; 1998 bool has_inner = true; 1999 2000 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner); 2001 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner); 2002 2003 return jhash_2words(mhash, mhash_inner, 0); 2004 } 2005 2006 static u32 fib_multipath_custom_hash_fl4(const struct net *net, 2007 const struct flowi4 *fl4) 2008 { 2009 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 2010 struct flow_keys hash_keys; 2011 2012 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 2013 return 0; 2014 2015 memset(&hash_keys, 0, sizeof(hash_keys)); 2016 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2017 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 2018 hash_keys.addrs.v4addrs.src = fl4->saddr; 2019 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 2020 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2021 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 2022 hash_keys.basic.ip_proto = fl4->flowi4_proto; 2023 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) 2024 hash_keys.ports.src = fl4->fl4_sport; 2025 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 2026 hash_keys.ports.dst = fl4->fl4_dport; 2027 2028 return flow_hash_from_keys(&hash_keys); 2029 } 2030 2031 /* if skb is set it will be used and fl4 can be NULL */ 2032 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4, 2033 const struct sk_buff *skb, struct flow_keys *flkeys) 2034 { 2035 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0; 2036 struct flow_keys hash_keys; 2037 u32 mhash = 0; 2038 2039 switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) { 2040 case 0: 2041 memset(&hash_keys, 0, sizeof(hash_keys)); 2042 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2043 if (skb) { 2044 ip_multipath_l3_keys(skb, &hash_keys); 2045 } else { 2046 hash_keys.addrs.v4addrs.src = fl4->saddr; 2047 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2048 } 2049 mhash = flow_hash_from_keys(&hash_keys); 2050 break; 2051 case 1: 2052 /* skb is currently provided only when forwarding */ 2053 if (skb) { 2054 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 2055 struct flow_keys keys; 2056 2057 /* short-circuit if we already have L4 hash present */ 2058 if (skb->l4_hash) 2059 return skb_get_hash_raw(skb) >> 1; 2060 2061 memset(&hash_keys, 0, sizeof(hash_keys)); 2062 2063 if (!flkeys) { 2064 skb_flow_dissect_flow_keys(skb, &keys, flag); 2065 flkeys = &keys; 2066 } 2067 2068 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2069 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src; 2070 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst; 2071 hash_keys.ports.src = flkeys->ports.src; 2072 hash_keys.ports.dst = flkeys->ports.dst; 2073 hash_keys.basic.ip_proto = flkeys->basic.ip_proto; 2074 } else { 2075 memset(&hash_keys, 0, sizeof(hash_keys)); 2076 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2077 hash_keys.addrs.v4addrs.src = fl4->saddr; 2078 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2079 hash_keys.ports.src = fl4->fl4_sport; 2080 hash_keys.ports.dst = fl4->fl4_dport; 2081 hash_keys.basic.ip_proto = fl4->flowi4_proto; 2082 } 2083 mhash = flow_hash_from_keys(&hash_keys); 2084 break; 2085 case 2: 2086 memset(&hash_keys, 0, sizeof(hash_keys)); 2087 /* skb is currently provided only when forwarding */ 2088 if (skb) { 2089 struct flow_keys keys; 2090 2091 skb_flow_dissect_flow_keys(skb, &keys, 0); 2092 /* Inner can be v4 or v6 */ 2093 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 2094 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2095 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 2096 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 2097 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 2098 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 2099 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 2100 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 2101 hash_keys.tags.flow_label = keys.tags.flow_label; 2102 hash_keys.basic.ip_proto = keys.basic.ip_proto; 2103 } else { 2104 /* Same as case 0 */ 2105 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2106 ip_multipath_l3_keys(skb, &hash_keys); 2107 } 2108 } else { 2109 /* Same as case 0 */ 2110 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2111 hash_keys.addrs.v4addrs.src = fl4->saddr; 2112 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2113 } 2114 mhash = flow_hash_from_keys(&hash_keys); 2115 break; 2116 case 3: 2117 if (skb) 2118 mhash = fib_multipath_custom_hash_skb(net, skb); 2119 else 2120 mhash = fib_multipath_custom_hash_fl4(net, fl4); 2121 break; 2122 } 2123 2124 if (multipath_hash) 2125 mhash = jhash_2words(mhash, multipath_hash, 0); 2126 2127 return mhash >> 1; 2128 } 2129 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 2130 2131 static int ip_mkroute_input(struct sk_buff *skb, 2132 struct fib_result *res, 2133 struct in_device *in_dev, 2134 __be32 daddr, __be32 saddr, u32 tos, 2135 struct flow_keys *hkeys) 2136 { 2137 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2138 if (res->fi && fib_info_num_path(res->fi) > 1) { 2139 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys); 2140 2141 fib_select_multipath(res, h); 2142 IPCB(skb)->flags |= IPSKB_MULTIPATH; 2143 } 2144 #endif 2145 2146 /* create a routing cache entry */ 2147 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos); 2148 } 2149 2150 /* Implements all the saddr-related checks as ip_route_input_slow(), 2151 * assuming daddr is valid and the destination is not a local broadcast one. 2152 * Uses the provided hint instead of performing a route lookup. 2153 */ 2154 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2155 u8 tos, struct net_device *dev, 2156 const struct sk_buff *hint) 2157 { 2158 struct in_device *in_dev = __in_dev_get_rcu(dev); 2159 struct rtable *rt = skb_rtable(hint); 2160 struct net *net = dev_net(dev); 2161 int err = -EINVAL; 2162 u32 tag = 0; 2163 2164 if (!in_dev) 2165 return -EINVAL; 2166 2167 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 2168 goto martian_source; 2169 2170 if (ipv4_is_zeronet(saddr)) 2171 goto martian_source; 2172 2173 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2174 goto martian_source; 2175 2176 if (rt->rt_type != RTN_LOCAL) 2177 goto skip_validate_source; 2178 2179 tos &= IPTOS_RT_MASK; 2180 err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag); 2181 if (err < 0) 2182 goto martian_source; 2183 2184 skip_validate_source: 2185 skb_dst_copy(skb, hint); 2186 return 0; 2187 2188 martian_source: 2189 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2190 return err; 2191 } 2192 2193 /* get device for dst_alloc with local routes */ 2194 static struct net_device *ip_rt_get_dev(struct net *net, 2195 const struct fib_result *res) 2196 { 2197 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL; 2198 struct net_device *dev = NULL; 2199 2200 if (nhc) 2201 dev = l3mdev_master_dev_rcu(nhc->nhc_dev); 2202 2203 return dev ? : net->loopback_dev; 2204 } 2205 2206 /* 2207 * NOTE. We drop all the packets that has local source 2208 * addresses, because every properly looped back packet 2209 * must have correct destination already attached by output routine. 2210 * Changes in the enforced policies must be applied also to 2211 * ip_route_use_hint(). 2212 * 2213 * Such approach solves two big problems: 2214 * 1. Not simplex devices are handled properly. 2215 * 2. IP spoofing attempts are filtered with 100% of guarantee. 2216 * called with rcu_read_lock() 2217 */ 2218 2219 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2220 u8 tos, struct net_device *dev, 2221 struct fib_result *res) 2222 { 2223 struct in_device *in_dev = __in_dev_get_rcu(dev); 2224 struct flow_keys *flkeys = NULL, _flkeys; 2225 struct net *net = dev_net(dev); 2226 struct ip_tunnel_info *tun_info; 2227 int err = -EINVAL; 2228 unsigned int flags = 0; 2229 u32 itag = 0; 2230 struct rtable *rth; 2231 struct flowi4 fl4; 2232 bool do_cache = true; 2233 2234 /* IP on this device is disabled. */ 2235 2236 if (!in_dev) 2237 goto out; 2238 2239 /* Check for the most weird martians, which can be not detected 2240 * by fib_lookup. 2241 */ 2242 2243 tun_info = skb_tunnel_info(skb); 2244 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX)) 2245 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id; 2246 else 2247 fl4.flowi4_tun_key.tun_id = 0; 2248 skb_dst_drop(skb); 2249 2250 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 2251 goto martian_source; 2252 2253 res->fi = NULL; 2254 res->table = NULL; 2255 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 2256 goto brd_input; 2257 2258 /* Accept zero addresses only to limited broadcast; 2259 * I even do not know to fix it or not. Waiting for complains :-) 2260 */ 2261 if (ipv4_is_zeronet(saddr)) 2262 goto martian_source; 2263 2264 if (ipv4_is_zeronet(daddr)) 2265 goto martian_destination; 2266 2267 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(), 2268 * and call it once if daddr or/and saddr are loopback addresses 2269 */ 2270 if (ipv4_is_loopback(daddr)) { 2271 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2272 goto martian_destination; 2273 } else if (ipv4_is_loopback(saddr)) { 2274 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2275 goto martian_source; 2276 } 2277 2278 /* 2279 * Now we are ready to route packet. 2280 */ 2281 fl4.flowi4_l3mdev = 0; 2282 fl4.flowi4_oif = 0; 2283 fl4.flowi4_iif = dev->ifindex; 2284 fl4.flowi4_mark = skb->mark; 2285 fl4.flowi4_tos = tos; 2286 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 2287 fl4.flowi4_flags = 0; 2288 fl4.daddr = daddr; 2289 fl4.saddr = saddr; 2290 fl4.flowi4_uid = sock_net_uid(net, NULL); 2291 fl4.flowi4_multipath_hash = 0; 2292 2293 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) { 2294 flkeys = &_flkeys; 2295 } else { 2296 fl4.flowi4_proto = 0; 2297 fl4.fl4_sport = 0; 2298 fl4.fl4_dport = 0; 2299 } 2300 2301 err = fib_lookup(net, &fl4, res, 0); 2302 if (err != 0) { 2303 if (!IN_DEV_FORWARD(in_dev)) 2304 err = -EHOSTUNREACH; 2305 goto no_route; 2306 } 2307 2308 if (res->type == RTN_BROADCAST) { 2309 if (IN_DEV_BFORWARD(in_dev)) 2310 goto make_route; 2311 /* not do cache if bc_forwarding is enabled */ 2312 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING)) 2313 do_cache = false; 2314 goto brd_input; 2315 } 2316 2317 if (res->type == RTN_LOCAL) { 2318 err = fib_validate_source(skb, saddr, daddr, tos, 2319 0, dev, in_dev, &itag); 2320 if (err < 0) 2321 goto martian_source; 2322 goto local_input; 2323 } 2324 2325 if (!IN_DEV_FORWARD(in_dev)) { 2326 err = -EHOSTUNREACH; 2327 goto no_route; 2328 } 2329 if (res->type != RTN_UNICAST) 2330 goto martian_destination; 2331 2332 make_route: 2333 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys); 2334 out: return err; 2335 2336 brd_input: 2337 if (skb->protocol != htons(ETH_P_IP)) 2338 goto e_inval; 2339 2340 if (!ipv4_is_zeronet(saddr)) { 2341 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 2342 in_dev, &itag); 2343 if (err < 0) 2344 goto martian_source; 2345 } 2346 flags |= RTCF_BROADCAST; 2347 res->type = RTN_BROADCAST; 2348 RT_CACHE_STAT_INC(in_brd); 2349 2350 local_input: 2351 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 2352 IPCB(skb)->flags |= IPSKB_NOPOLICY; 2353 2354 do_cache &= res->fi && !itag; 2355 if (do_cache) { 2356 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2357 2358 rth = rcu_dereference(nhc->nhc_rth_input); 2359 if (rt_cache_valid(rth)) { 2360 skb_dst_set_noref(skb, &rth->dst); 2361 err = 0; 2362 goto out; 2363 } 2364 } 2365 2366 rth = rt_dst_alloc(ip_rt_get_dev(net, res), 2367 flags | RTCF_LOCAL, res->type, false); 2368 if (!rth) 2369 goto e_nobufs; 2370 2371 rth->dst.output= ip_rt_bug; 2372 #ifdef CONFIG_IP_ROUTE_CLASSID 2373 rth->dst.tclassid = itag; 2374 #endif 2375 rth->rt_is_input = 1; 2376 2377 RT_CACHE_STAT_INC(in_slow_tot); 2378 if (res->type == RTN_UNREACHABLE) { 2379 rth->dst.input= ip_error; 2380 rth->dst.error= -err; 2381 rth->rt_flags &= ~RTCF_LOCAL; 2382 } 2383 2384 if (do_cache) { 2385 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2386 2387 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 2388 if (lwtunnel_input_redirect(rth->dst.lwtstate)) { 2389 WARN_ON(rth->dst.input == lwtunnel_input); 2390 rth->dst.lwtstate->orig_input = rth->dst.input; 2391 rth->dst.input = lwtunnel_input; 2392 } 2393 2394 if (unlikely(!rt_cache_route(nhc, rth))) 2395 rt_add_uncached_list(rth); 2396 } 2397 skb_dst_set(skb, &rth->dst); 2398 err = 0; 2399 goto out; 2400 2401 no_route: 2402 RT_CACHE_STAT_INC(in_no_route); 2403 res->type = RTN_UNREACHABLE; 2404 res->fi = NULL; 2405 res->table = NULL; 2406 goto local_input; 2407 2408 /* 2409 * Do not cache martian addresses: they should be logged (RFC1812) 2410 */ 2411 martian_destination: 2412 RT_CACHE_STAT_INC(in_martian_dst); 2413 #ifdef CONFIG_IP_ROUTE_VERBOSE 2414 if (IN_DEV_LOG_MARTIANS(in_dev)) 2415 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n", 2416 &daddr, &saddr, dev->name); 2417 #endif 2418 2419 e_inval: 2420 err = -EINVAL; 2421 goto out; 2422 2423 e_nobufs: 2424 err = -ENOBUFS; 2425 goto out; 2426 2427 martian_source: 2428 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2429 goto out; 2430 } 2431 2432 /* called with rcu_read_lock held */ 2433 static int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2434 u8 tos, struct net_device *dev, struct fib_result *res) 2435 { 2436 /* Multicast recognition logic is moved from route cache to here. 2437 * The problem was that too many Ethernet cards have broken/missing 2438 * hardware multicast filters :-( As result the host on multicasting 2439 * network acquires a lot of useless route cache entries, sort of 2440 * SDR messages from all the world. Now we try to get rid of them. 2441 * Really, provided software IP multicast filter is organized 2442 * reasonably (at least, hashed), it does not result in a slowdown 2443 * comparing with route cache reject entries. 2444 * Note, that multicast routers are not affected, because 2445 * route cache entry is created eventually. 2446 */ 2447 if (ipv4_is_multicast(daddr)) { 2448 struct in_device *in_dev = __in_dev_get_rcu(dev); 2449 int our = 0; 2450 int err = -EINVAL; 2451 2452 if (!in_dev) 2453 return err; 2454 our = ip_check_mc_rcu(in_dev, daddr, saddr, 2455 ip_hdr(skb)->protocol); 2456 2457 /* check l3 master if no match yet */ 2458 if (!our && netif_is_l3_slave(dev)) { 2459 struct in_device *l3_in_dev; 2460 2461 l3_in_dev = __in_dev_get_rcu(skb->dev); 2462 if (l3_in_dev) 2463 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr, 2464 ip_hdr(skb)->protocol); 2465 } 2466 2467 if (our 2468 #ifdef CONFIG_IP_MROUTE 2469 || 2470 (!ipv4_is_local_multicast(daddr) && 2471 IN_DEV_MFORWARD(in_dev)) 2472 #endif 2473 ) { 2474 err = ip_route_input_mc(skb, daddr, saddr, 2475 tos, dev, our); 2476 } 2477 return err; 2478 } 2479 2480 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res); 2481 } 2482 2483 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2484 u8 tos, struct net_device *dev) 2485 { 2486 struct fib_result res; 2487 int err; 2488 2489 tos &= IPTOS_RT_MASK; 2490 rcu_read_lock(); 2491 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res); 2492 rcu_read_unlock(); 2493 2494 return err; 2495 } 2496 EXPORT_SYMBOL(ip_route_input_noref); 2497 2498 /* called with rcu_read_lock() */ 2499 static struct rtable *__mkroute_output(const struct fib_result *res, 2500 const struct flowi4 *fl4, int orig_oif, 2501 struct net_device *dev_out, 2502 unsigned int flags) 2503 { 2504 struct fib_info *fi = res->fi; 2505 struct fib_nh_exception *fnhe; 2506 struct in_device *in_dev; 2507 u16 type = res->type; 2508 struct rtable *rth; 2509 bool do_cache; 2510 2511 in_dev = __in_dev_get_rcu(dev_out); 2512 if (!in_dev) 2513 return ERR_PTR(-EINVAL); 2514 2515 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 2516 if (ipv4_is_loopback(fl4->saddr) && 2517 !(dev_out->flags & IFF_LOOPBACK) && 2518 !netif_is_l3_master(dev_out)) 2519 return ERR_PTR(-EINVAL); 2520 2521 if (ipv4_is_lbcast(fl4->daddr)) 2522 type = RTN_BROADCAST; 2523 else if (ipv4_is_multicast(fl4->daddr)) 2524 type = RTN_MULTICAST; 2525 else if (ipv4_is_zeronet(fl4->daddr)) 2526 return ERR_PTR(-EINVAL); 2527 2528 if (dev_out->flags & IFF_LOOPBACK) 2529 flags |= RTCF_LOCAL; 2530 2531 do_cache = true; 2532 if (type == RTN_BROADCAST) { 2533 flags |= RTCF_BROADCAST | RTCF_LOCAL; 2534 fi = NULL; 2535 } else if (type == RTN_MULTICAST) { 2536 flags |= RTCF_MULTICAST | RTCF_LOCAL; 2537 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr, 2538 fl4->flowi4_proto)) 2539 flags &= ~RTCF_LOCAL; 2540 else 2541 do_cache = false; 2542 /* If multicast route do not exist use 2543 * default one, but do not gateway in this case. 2544 * Yes, it is hack. 2545 */ 2546 if (fi && res->prefixlen < 4) 2547 fi = NULL; 2548 } else if ((type == RTN_LOCAL) && (orig_oif != 0) && 2549 (orig_oif != dev_out->ifindex)) { 2550 /* For local routes that require a particular output interface 2551 * we do not want to cache the result. Caching the result 2552 * causes incorrect behaviour when there are multiple source 2553 * addresses on the interface, the end result being that if the 2554 * intended recipient is waiting on that interface for the 2555 * packet he won't receive it because it will be delivered on 2556 * the loopback interface and the IP_PKTINFO ipi_ifindex will 2557 * be set to the loopback interface as well. 2558 */ 2559 do_cache = false; 2560 } 2561 2562 fnhe = NULL; 2563 do_cache &= fi != NULL; 2564 if (fi) { 2565 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2566 struct rtable __rcu **prth; 2567 2568 fnhe = find_exception(nhc, fl4->daddr); 2569 if (!do_cache) 2570 goto add; 2571 if (fnhe) { 2572 prth = &fnhe->fnhe_rth_output; 2573 } else { 2574 if (unlikely(fl4->flowi4_flags & 2575 FLOWI_FLAG_KNOWN_NH && 2576 !(nhc->nhc_gw_family && 2577 nhc->nhc_scope == RT_SCOPE_LINK))) { 2578 do_cache = false; 2579 goto add; 2580 } 2581 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 2582 } 2583 rth = rcu_dereference(*prth); 2584 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst)) 2585 return rth; 2586 } 2587 2588 add: 2589 rth = rt_dst_alloc(dev_out, flags, type, 2590 IN_DEV_ORCONF(in_dev, NOXFRM)); 2591 if (!rth) 2592 return ERR_PTR(-ENOBUFS); 2593 2594 rth->rt_iif = orig_oif; 2595 2596 RT_CACHE_STAT_INC(out_slow_tot); 2597 2598 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 2599 if (flags & RTCF_LOCAL && 2600 !(dev_out->flags & IFF_LOOPBACK)) { 2601 rth->dst.output = ip_mc_output; 2602 RT_CACHE_STAT_INC(out_slow_mc); 2603 } 2604 #ifdef CONFIG_IP_MROUTE 2605 if (type == RTN_MULTICAST) { 2606 if (IN_DEV_MFORWARD(in_dev) && 2607 !ipv4_is_local_multicast(fl4->daddr)) { 2608 rth->dst.input = ip_mr_input; 2609 rth->dst.output = ip_mc_output; 2610 } 2611 } 2612 #endif 2613 } 2614 2615 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache); 2616 lwtunnel_set_redirect(&rth->dst); 2617 2618 return rth; 2619 } 2620 2621 /* 2622 * Major route resolver routine. 2623 */ 2624 2625 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4, 2626 const struct sk_buff *skb) 2627 { 2628 struct fib_result res = { 2629 .type = RTN_UNSPEC, 2630 .fi = NULL, 2631 .table = NULL, 2632 .tclassid = 0, 2633 }; 2634 struct rtable *rth; 2635 2636 fl4->flowi4_iif = LOOPBACK_IFINDEX; 2637 ip_rt_fix_tos(fl4); 2638 2639 rcu_read_lock(); 2640 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb); 2641 rcu_read_unlock(); 2642 2643 return rth; 2644 } 2645 EXPORT_SYMBOL_GPL(ip_route_output_key_hash); 2646 2647 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4, 2648 struct fib_result *res, 2649 const struct sk_buff *skb) 2650 { 2651 struct net_device *dev_out = NULL; 2652 int orig_oif = fl4->flowi4_oif; 2653 unsigned int flags = 0; 2654 struct rtable *rth; 2655 int err; 2656 2657 if (fl4->saddr) { 2658 if (ipv4_is_multicast(fl4->saddr) || 2659 ipv4_is_lbcast(fl4->saddr) || 2660 ipv4_is_zeronet(fl4->saddr)) { 2661 rth = ERR_PTR(-EINVAL); 2662 goto out; 2663 } 2664 2665 rth = ERR_PTR(-ENETUNREACH); 2666 2667 /* I removed check for oif == dev_out->oif here. 2668 * It was wrong for two reasons: 2669 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 2670 * is assigned to multiple interfaces. 2671 * 2. Moreover, we are allowed to send packets with saddr 2672 * of another iface. --ANK 2673 */ 2674 2675 if (fl4->flowi4_oif == 0 && 2676 (ipv4_is_multicast(fl4->daddr) || 2677 ipv4_is_lbcast(fl4->daddr))) { 2678 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2679 dev_out = __ip_dev_find(net, fl4->saddr, false); 2680 if (!dev_out) 2681 goto out; 2682 2683 /* Special hack: user can direct multicasts 2684 * and limited broadcast via necessary interface 2685 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 2686 * This hack is not just for fun, it allows 2687 * vic,vat and friends to work. 2688 * They bind socket to loopback, set ttl to zero 2689 * and expect that it will work. 2690 * From the viewpoint of routing cache they are broken, 2691 * because we are not allowed to build multicast path 2692 * with loopback source addr (look, routing cache 2693 * cannot know, that ttl is zero, so that packet 2694 * will not leave this host and route is valid). 2695 * Luckily, this hack is good workaround. 2696 */ 2697 2698 fl4->flowi4_oif = dev_out->ifindex; 2699 goto make_route; 2700 } 2701 2702 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 2703 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2704 if (!__ip_dev_find(net, fl4->saddr, false)) 2705 goto out; 2706 } 2707 } 2708 2709 2710 if (fl4->flowi4_oif) { 2711 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif); 2712 rth = ERR_PTR(-ENODEV); 2713 if (!dev_out) 2714 goto out; 2715 2716 /* RACE: Check return value of inet_select_addr instead. */ 2717 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 2718 rth = ERR_PTR(-ENETUNREACH); 2719 goto out; 2720 } 2721 if (ipv4_is_local_multicast(fl4->daddr) || 2722 ipv4_is_lbcast(fl4->daddr) || 2723 fl4->flowi4_proto == IPPROTO_IGMP) { 2724 if (!fl4->saddr) 2725 fl4->saddr = inet_select_addr(dev_out, 0, 2726 RT_SCOPE_LINK); 2727 goto make_route; 2728 } 2729 if (!fl4->saddr) { 2730 if (ipv4_is_multicast(fl4->daddr)) 2731 fl4->saddr = inet_select_addr(dev_out, 0, 2732 fl4->flowi4_scope); 2733 else if (!fl4->daddr) 2734 fl4->saddr = inet_select_addr(dev_out, 0, 2735 RT_SCOPE_HOST); 2736 } 2737 } 2738 2739 if (!fl4->daddr) { 2740 fl4->daddr = fl4->saddr; 2741 if (!fl4->daddr) 2742 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK); 2743 dev_out = net->loopback_dev; 2744 fl4->flowi4_oif = LOOPBACK_IFINDEX; 2745 res->type = RTN_LOCAL; 2746 flags |= RTCF_LOCAL; 2747 goto make_route; 2748 } 2749 2750 err = fib_lookup(net, fl4, res, 0); 2751 if (err) { 2752 res->fi = NULL; 2753 res->table = NULL; 2754 if (fl4->flowi4_oif && 2755 (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) { 2756 /* Apparently, routing tables are wrong. Assume, 2757 * that the destination is on link. 2758 * 2759 * WHY? DW. 2760 * Because we are allowed to send to iface 2761 * even if it has NO routes and NO assigned 2762 * addresses. When oif is specified, routing 2763 * tables are looked up with only one purpose: 2764 * to catch if destination is gatewayed, rather than 2765 * direct. Moreover, if MSG_DONTROUTE is set, 2766 * we send packet, ignoring both routing tables 2767 * and ifaddr state. --ANK 2768 * 2769 * 2770 * We could make it even if oif is unknown, 2771 * likely IPv6, but we do not. 2772 */ 2773 2774 if (fl4->saddr == 0) 2775 fl4->saddr = inet_select_addr(dev_out, 0, 2776 RT_SCOPE_LINK); 2777 res->type = RTN_UNICAST; 2778 goto make_route; 2779 } 2780 rth = ERR_PTR(err); 2781 goto out; 2782 } 2783 2784 if (res->type == RTN_LOCAL) { 2785 if (!fl4->saddr) { 2786 if (res->fi->fib_prefsrc) 2787 fl4->saddr = res->fi->fib_prefsrc; 2788 else 2789 fl4->saddr = fl4->daddr; 2790 } 2791 2792 /* L3 master device is the loopback for that domain */ 2793 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? : 2794 net->loopback_dev; 2795 2796 /* make sure orig_oif points to fib result device even 2797 * though packet rx/tx happens over loopback or l3mdev 2798 */ 2799 orig_oif = FIB_RES_OIF(*res); 2800 2801 fl4->flowi4_oif = dev_out->ifindex; 2802 flags |= RTCF_LOCAL; 2803 goto make_route; 2804 } 2805 2806 fib_select_path(net, res, fl4, skb); 2807 2808 dev_out = FIB_RES_DEV(*res); 2809 2810 make_route: 2811 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags); 2812 2813 out: 2814 return rth; 2815 } 2816 2817 static struct dst_ops ipv4_dst_blackhole_ops = { 2818 .family = AF_INET, 2819 .default_advmss = ipv4_default_advmss, 2820 .neigh_lookup = ipv4_neigh_lookup, 2821 .check = dst_blackhole_check, 2822 .cow_metrics = dst_blackhole_cow_metrics, 2823 .update_pmtu = dst_blackhole_update_pmtu, 2824 .redirect = dst_blackhole_redirect, 2825 .mtu = dst_blackhole_mtu, 2826 }; 2827 2828 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2829 { 2830 struct rtable *ort = (struct rtable *) dst_orig; 2831 struct rtable *rt; 2832 2833 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0); 2834 if (rt) { 2835 struct dst_entry *new = &rt->dst; 2836 2837 new->__use = 1; 2838 new->input = dst_discard; 2839 new->output = dst_discard_out; 2840 2841 new->dev = net->loopback_dev; 2842 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC); 2843 2844 rt->rt_is_input = ort->rt_is_input; 2845 rt->rt_iif = ort->rt_iif; 2846 rt->rt_pmtu = ort->rt_pmtu; 2847 rt->rt_mtu_locked = ort->rt_mtu_locked; 2848 2849 rt->rt_genid = rt_genid_ipv4(net); 2850 rt->rt_flags = ort->rt_flags; 2851 rt->rt_type = ort->rt_type; 2852 rt->rt_uses_gateway = ort->rt_uses_gateway; 2853 rt->rt_gw_family = ort->rt_gw_family; 2854 if (rt->rt_gw_family == AF_INET) 2855 rt->rt_gw4 = ort->rt_gw4; 2856 else if (rt->rt_gw_family == AF_INET6) 2857 rt->rt_gw6 = ort->rt_gw6; 2858 } 2859 2860 dst_release(dst_orig); 2861 2862 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2863 } 2864 2865 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2866 const struct sock *sk) 2867 { 2868 struct rtable *rt = __ip_route_output_key(net, flp4); 2869 2870 if (IS_ERR(rt)) 2871 return rt; 2872 2873 if (flp4->flowi4_proto) { 2874 flp4->flowi4_oif = rt->dst.dev->ifindex; 2875 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst, 2876 flowi4_to_flowi(flp4), 2877 sk, 0); 2878 } 2879 2880 return rt; 2881 } 2882 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2883 2884 struct rtable *ip_route_output_tunnel(struct sk_buff *skb, 2885 struct net_device *dev, 2886 struct net *net, __be32 *saddr, 2887 const struct ip_tunnel_info *info, 2888 u8 protocol, bool use_cache) 2889 { 2890 #ifdef CONFIG_DST_CACHE 2891 struct dst_cache *dst_cache; 2892 #endif 2893 struct rtable *rt = NULL; 2894 struct flowi4 fl4; 2895 __u8 tos; 2896 2897 #ifdef CONFIG_DST_CACHE 2898 dst_cache = (struct dst_cache *)&info->dst_cache; 2899 if (use_cache) { 2900 rt = dst_cache_get_ip4(dst_cache, saddr); 2901 if (rt) 2902 return rt; 2903 } 2904 #endif 2905 memset(&fl4, 0, sizeof(fl4)); 2906 fl4.flowi4_mark = skb->mark; 2907 fl4.flowi4_proto = protocol; 2908 fl4.daddr = info->key.u.ipv4.dst; 2909 fl4.saddr = info->key.u.ipv4.src; 2910 tos = info->key.tos; 2911 fl4.flowi4_tos = RT_TOS(tos); 2912 2913 rt = ip_route_output_key(net, &fl4); 2914 if (IS_ERR(rt)) { 2915 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr); 2916 return ERR_PTR(-ENETUNREACH); 2917 } 2918 if (rt->dst.dev == dev) { /* is this necessary? */ 2919 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr); 2920 ip_rt_put(rt); 2921 return ERR_PTR(-ELOOP); 2922 } 2923 #ifdef CONFIG_DST_CACHE 2924 if (use_cache) 2925 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr); 2926 #endif 2927 *saddr = fl4.saddr; 2928 return rt; 2929 } 2930 EXPORT_SYMBOL_GPL(ip_route_output_tunnel); 2931 2932 /* called with rcu_read_lock held */ 2933 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, 2934 struct rtable *rt, u32 table_id, dscp_t dscp, 2935 struct flowi4 *fl4, struct sk_buff *skb, u32 portid, 2936 u32 seq, unsigned int flags) 2937 { 2938 struct rtmsg *r; 2939 struct nlmsghdr *nlh; 2940 unsigned long expires = 0; 2941 u32 error; 2942 u32 metrics[RTAX_MAX]; 2943 2944 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags); 2945 if (!nlh) 2946 return -EMSGSIZE; 2947 2948 r = nlmsg_data(nlh); 2949 r->rtm_family = AF_INET; 2950 r->rtm_dst_len = 32; 2951 r->rtm_src_len = 0; 2952 r->rtm_tos = inet_dscp_to_dsfield(dscp); 2953 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT; 2954 if (nla_put_u32(skb, RTA_TABLE, table_id)) 2955 goto nla_put_failure; 2956 r->rtm_type = rt->rt_type; 2957 r->rtm_scope = RT_SCOPE_UNIVERSE; 2958 r->rtm_protocol = RTPROT_UNSPEC; 2959 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 2960 if (rt->rt_flags & RTCF_NOTIFY) 2961 r->rtm_flags |= RTM_F_NOTIFY; 2962 if (IPCB(skb)->flags & IPSKB_DOREDIRECT) 2963 r->rtm_flags |= RTCF_DOREDIRECT; 2964 2965 if (nla_put_in_addr(skb, RTA_DST, dst)) 2966 goto nla_put_failure; 2967 if (src) { 2968 r->rtm_src_len = 32; 2969 if (nla_put_in_addr(skb, RTA_SRC, src)) 2970 goto nla_put_failure; 2971 } 2972 if (rt->dst.dev && 2973 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex)) 2974 goto nla_put_failure; 2975 if (rt->dst.lwtstate && 2976 lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0) 2977 goto nla_put_failure; 2978 #ifdef CONFIG_IP_ROUTE_CLASSID 2979 if (rt->dst.tclassid && 2980 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid)) 2981 goto nla_put_failure; 2982 #endif 2983 if (fl4 && !rt_is_input_route(rt) && 2984 fl4->saddr != src) { 2985 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr)) 2986 goto nla_put_failure; 2987 } 2988 if (rt->rt_uses_gateway) { 2989 if (rt->rt_gw_family == AF_INET && 2990 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) { 2991 goto nla_put_failure; 2992 } else if (rt->rt_gw_family == AF_INET6) { 2993 int alen = sizeof(struct in6_addr); 2994 struct nlattr *nla; 2995 struct rtvia *via; 2996 2997 nla = nla_reserve(skb, RTA_VIA, alen + 2); 2998 if (!nla) 2999 goto nla_put_failure; 3000 3001 via = nla_data(nla); 3002 via->rtvia_family = AF_INET6; 3003 memcpy(via->rtvia_addr, &rt->rt_gw6, alen); 3004 } 3005 } 3006 3007 expires = rt->dst.expires; 3008 if (expires) { 3009 unsigned long now = jiffies; 3010 3011 if (time_before(now, expires)) 3012 expires -= now; 3013 else 3014 expires = 0; 3015 } 3016 3017 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics)); 3018 if (rt->rt_pmtu && expires) 3019 metrics[RTAX_MTU - 1] = rt->rt_pmtu; 3020 if (rt->rt_mtu_locked && expires) 3021 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU); 3022 if (rtnetlink_put_metrics(skb, metrics) < 0) 3023 goto nla_put_failure; 3024 3025 if (fl4) { 3026 if (fl4->flowi4_mark && 3027 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark)) 3028 goto nla_put_failure; 3029 3030 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) && 3031 nla_put_u32(skb, RTA_UID, 3032 from_kuid_munged(current_user_ns(), 3033 fl4->flowi4_uid))) 3034 goto nla_put_failure; 3035 3036 if (rt_is_input_route(rt)) { 3037 #ifdef CONFIG_IP_MROUTE 3038 if (ipv4_is_multicast(dst) && 3039 !ipv4_is_local_multicast(dst) && 3040 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) { 3041 int err = ipmr_get_route(net, skb, 3042 fl4->saddr, fl4->daddr, 3043 r, portid); 3044 3045 if (err <= 0) { 3046 if (err == 0) 3047 return 0; 3048 goto nla_put_failure; 3049 } 3050 } else 3051 #endif 3052 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif)) 3053 goto nla_put_failure; 3054 } 3055 } 3056 3057 error = rt->dst.error; 3058 3059 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0) 3060 goto nla_put_failure; 3061 3062 nlmsg_end(skb, nlh); 3063 return 0; 3064 3065 nla_put_failure: 3066 nlmsg_cancel(skb, nlh); 3067 return -EMSGSIZE; 3068 } 3069 3070 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb, 3071 struct netlink_callback *cb, u32 table_id, 3072 struct fnhe_hash_bucket *bucket, int genid, 3073 int *fa_index, int fa_start, unsigned int flags) 3074 { 3075 int i; 3076 3077 for (i = 0; i < FNHE_HASH_SIZE; i++) { 3078 struct fib_nh_exception *fnhe; 3079 3080 for (fnhe = rcu_dereference(bucket[i].chain); fnhe; 3081 fnhe = rcu_dereference(fnhe->fnhe_next)) { 3082 struct rtable *rt; 3083 int err; 3084 3085 if (*fa_index < fa_start) 3086 goto next; 3087 3088 if (fnhe->fnhe_genid != genid) 3089 goto next; 3090 3091 if (fnhe->fnhe_expires && 3092 time_after(jiffies, fnhe->fnhe_expires)) 3093 goto next; 3094 3095 rt = rcu_dereference(fnhe->fnhe_rth_input); 3096 if (!rt) 3097 rt = rcu_dereference(fnhe->fnhe_rth_output); 3098 if (!rt) 3099 goto next; 3100 3101 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt, 3102 table_id, 0, NULL, skb, 3103 NETLINK_CB(cb->skb).portid, 3104 cb->nlh->nlmsg_seq, flags); 3105 if (err) 3106 return err; 3107 next: 3108 (*fa_index)++; 3109 } 3110 } 3111 3112 return 0; 3113 } 3114 3115 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb, 3116 u32 table_id, struct fib_info *fi, 3117 int *fa_index, int fa_start, unsigned int flags) 3118 { 3119 struct net *net = sock_net(cb->skb->sk); 3120 int nhsel, genid = fnhe_genid(net); 3121 3122 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) { 3123 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel); 3124 struct fnhe_hash_bucket *bucket; 3125 int err; 3126 3127 if (nhc->nhc_flags & RTNH_F_DEAD) 3128 continue; 3129 3130 rcu_read_lock(); 3131 bucket = rcu_dereference(nhc->nhc_exceptions); 3132 err = 0; 3133 if (bucket) 3134 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket, 3135 genid, fa_index, fa_start, 3136 flags); 3137 rcu_read_unlock(); 3138 if (err) 3139 return err; 3140 } 3141 3142 return 0; 3143 } 3144 3145 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst, 3146 u8 ip_proto, __be16 sport, 3147 __be16 dport) 3148 { 3149 struct sk_buff *skb; 3150 struct iphdr *iph; 3151 3152 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 3153 if (!skb) 3154 return NULL; 3155 3156 /* Reserve room for dummy headers, this skb can pass 3157 * through good chunk of routing engine. 3158 */ 3159 skb_reset_mac_header(skb); 3160 skb_reset_network_header(skb); 3161 skb->protocol = htons(ETH_P_IP); 3162 iph = skb_put(skb, sizeof(struct iphdr)); 3163 iph->protocol = ip_proto; 3164 iph->saddr = src; 3165 iph->daddr = dst; 3166 iph->version = 0x4; 3167 iph->frag_off = 0; 3168 iph->ihl = 0x5; 3169 skb_set_transport_header(skb, skb->len); 3170 3171 switch (iph->protocol) { 3172 case IPPROTO_UDP: { 3173 struct udphdr *udph; 3174 3175 udph = skb_put_zero(skb, sizeof(struct udphdr)); 3176 udph->source = sport; 3177 udph->dest = dport; 3178 udph->len = htons(sizeof(struct udphdr)); 3179 udph->check = 0; 3180 break; 3181 } 3182 case IPPROTO_TCP: { 3183 struct tcphdr *tcph; 3184 3185 tcph = skb_put_zero(skb, sizeof(struct tcphdr)); 3186 tcph->source = sport; 3187 tcph->dest = dport; 3188 tcph->doff = sizeof(struct tcphdr) / 4; 3189 tcph->rst = 1; 3190 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr), 3191 src, dst, 0); 3192 break; 3193 } 3194 case IPPROTO_ICMP: { 3195 struct icmphdr *icmph; 3196 3197 icmph = skb_put_zero(skb, sizeof(struct icmphdr)); 3198 icmph->type = ICMP_ECHO; 3199 icmph->code = 0; 3200 } 3201 } 3202 3203 return skb; 3204 } 3205 3206 static int inet_rtm_valid_getroute_req(struct sk_buff *skb, 3207 const struct nlmsghdr *nlh, 3208 struct nlattr **tb, 3209 struct netlink_ext_ack *extack) 3210 { 3211 struct rtmsg *rtm; 3212 int i, err; 3213 3214 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) { 3215 NL_SET_ERR_MSG(extack, 3216 "ipv4: Invalid header for route get request"); 3217 return -EINVAL; 3218 } 3219 3220 if (!netlink_strict_get_check(skb)) 3221 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX, 3222 rtm_ipv4_policy, extack); 3223 3224 rtm = nlmsg_data(nlh); 3225 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) || 3226 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) || 3227 rtm->rtm_table || rtm->rtm_protocol || 3228 rtm->rtm_scope || rtm->rtm_type) { 3229 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request"); 3230 return -EINVAL; 3231 } 3232 3233 if (rtm->rtm_flags & ~(RTM_F_NOTIFY | 3234 RTM_F_LOOKUP_TABLE | 3235 RTM_F_FIB_MATCH)) { 3236 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request"); 3237 return -EINVAL; 3238 } 3239 3240 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX, 3241 rtm_ipv4_policy, extack); 3242 if (err) 3243 return err; 3244 3245 if ((tb[RTA_SRC] && !rtm->rtm_src_len) || 3246 (tb[RTA_DST] && !rtm->rtm_dst_len)) { 3247 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4"); 3248 return -EINVAL; 3249 } 3250 3251 for (i = 0; i <= RTA_MAX; i++) { 3252 if (!tb[i]) 3253 continue; 3254 3255 switch (i) { 3256 case RTA_IIF: 3257 case RTA_OIF: 3258 case RTA_SRC: 3259 case RTA_DST: 3260 case RTA_IP_PROTO: 3261 case RTA_SPORT: 3262 case RTA_DPORT: 3263 case RTA_MARK: 3264 case RTA_UID: 3265 break; 3266 default: 3267 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request"); 3268 return -EINVAL; 3269 } 3270 } 3271 3272 return 0; 3273 } 3274 3275 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, 3276 struct netlink_ext_ack *extack) 3277 { 3278 struct net *net = sock_net(in_skb->sk); 3279 struct nlattr *tb[RTA_MAX+1]; 3280 u32 table_id = RT_TABLE_MAIN; 3281 __be16 sport = 0, dport = 0; 3282 struct fib_result res = {}; 3283 u8 ip_proto = IPPROTO_UDP; 3284 struct rtable *rt = NULL; 3285 struct sk_buff *skb; 3286 struct rtmsg *rtm; 3287 struct flowi4 fl4 = {}; 3288 __be32 dst = 0; 3289 __be32 src = 0; 3290 kuid_t uid; 3291 u32 iif; 3292 int err; 3293 int mark; 3294 3295 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack); 3296 if (err < 0) 3297 return err; 3298 3299 rtm = nlmsg_data(nlh); 3300 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0; 3301 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0; 3302 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0; 3303 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0; 3304 if (tb[RTA_UID]) 3305 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID])); 3306 else 3307 uid = (iif ? INVALID_UID : current_uid()); 3308 3309 if (tb[RTA_IP_PROTO]) { 3310 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO], 3311 &ip_proto, AF_INET, extack); 3312 if (err) 3313 return err; 3314 } 3315 3316 if (tb[RTA_SPORT]) 3317 sport = nla_get_be16(tb[RTA_SPORT]); 3318 3319 if (tb[RTA_DPORT]) 3320 dport = nla_get_be16(tb[RTA_DPORT]); 3321 3322 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport); 3323 if (!skb) 3324 return -ENOBUFS; 3325 3326 fl4.daddr = dst; 3327 fl4.saddr = src; 3328 fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK; 3329 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0; 3330 fl4.flowi4_mark = mark; 3331 fl4.flowi4_uid = uid; 3332 if (sport) 3333 fl4.fl4_sport = sport; 3334 if (dport) 3335 fl4.fl4_dport = dport; 3336 fl4.flowi4_proto = ip_proto; 3337 3338 rcu_read_lock(); 3339 3340 if (iif) { 3341 struct net_device *dev; 3342 3343 dev = dev_get_by_index_rcu(net, iif); 3344 if (!dev) { 3345 err = -ENODEV; 3346 goto errout_rcu; 3347 } 3348 3349 fl4.flowi4_iif = iif; /* for rt_fill_info */ 3350 skb->dev = dev; 3351 skb->mark = mark; 3352 err = ip_route_input_rcu(skb, dst, src, 3353 rtm->rtm_tos & IPTOS_RT_MASK, dev, 3354 &res); 3355 3356 rt = skb_rtable(skb); 3357 if (err == 0 && rt->dst.error) 3358 err = -rt->dst.error; 3359 } else { 3360 fl4.flowi4_iif = LOOPBACK_IFINDEX; 3361 skb->dev = net->loopback_dev; 3362 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb); 3363 err = 0; 3364 if (IS_ERR(rt)) 3365 err = PTR_ERR(rt); 3366 else 3367 skb_dst_set(skb, &rt->dst); 3368 } 3369 3370 if (err) 3371 goto errout_rcu; 3372 3373 if (rtm->rtm_flags & RTM_F_NOTIFY) 3374 rt->rt_flags |= RTCF_NOTIFY; 3375 3376 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE) 3377 table_id = res.table ? res.table->tb_id : 0; 3378 3379 /* reset skb for netlink reply msg */ 3380 skb_trim(skb, 0); 3381 skb_reset_network_header(skb); 3382 skb_reset_transport_header(skb); 3383 skb_reset_mac_header(skb); 3384 3385 if (rtm->rtm_flags & RTM_F_FIB_MATCH) { 3386 struct fib_rt_info fri; 3387 3388 if (!res.fi) { 3389 err = fib_props[res.type].error; 3390 if (!err) 3391 err = -EHOSTUNREACH; 3392 goto errout_rcu; 3393 } 3394 fri.fi = res.fi; 3395 fri.tb_id = table_id; 3396 fri.dst = res.prefix; 3397 fri.dst_len = res.prefixlen; 3398 fri.dscp = res.dscp; 3399 fri.type = rt->rt_type; 3400 fri.offload = 0; 3401 fri.trap = 0; 3402 fri.offload_failed = 0; 3403 if (res.fa_head) { 3404 struct fib_alias *fa; 3405 3406 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) { 3407 u8 slen = 32 - fri.dst_len; 3408 3409 if (fa->fa_slen == slen && 3410 fa->tb_id == fri.tb_id && 3411 fa->fa_dscp == fri.dscp && 3412 fa->fa_info == res.fi && 3413 fa->fa_type == fri.type) { 3414 fri.offload = READ_ONCE(fa->offload); 3415 fri.trap = READ_ONCE(fa->trap); 3416 fri.offload_failed = 3417 READ_ONCE(fa->offload_failed); 3418 break; 3419 } 3420 } 3421 } 3422 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid, 3423 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0); 3424 } else { 3425 err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4, 3426 skb, NETLINK_CB(in_skb).portid, 3427 nlh->nlmsg_seq, 0); 3428 } 3429 if (err < 0) 3430 goto errout_rcu; 3431 3432 rcu_read_unlock(); 3433 3434 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 3435 3436 errout_free: 3437 return err; 3438 errout_rcu: 3439 rcu_read_unlock(); 3440 kfree_skb(skb); 3441 goto errout_free; 3442 } 3443 3444 void ip_rt_multicast_event(struct in_device *in_dev) 3445 { 3446 rt_cache_flush(dev_net(in_dev->dev)); 3447 } 3448 3449 #ifdef CONFIG_SYSCTL 3450 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 3451 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 3452 static int ip_rt_gc_elasticity __read_mostly = 8; 3453 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU; 3454 3455 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write, 3456 void *buffer, size_t *lenp, loff_t *ppos) 3457 { 3458 struct net *net = (struct net *)__ctl->extra1; 3459 3460 if (write) { 3461 rt_cache_flush(net); 3462 fnhe_genid_bump(net); 3463 return 0; 3464 } 3465 3466 return -EINVAL; 3467 } 3468 3469 static struct ctl_table ipv4_route_table[] = { 3470 { 3471 .procname = "gc_thresh", 3472 .data = &ipv4_dst_ops.gc_thresh, 3473 .maxlen = sizeof(int), 3474 .mode = 0644, 3475 .proc_handler = proc_dointvec, 3476 }, 3477 { 3478 .procname = "max_size", 3479 .data = &ip_rt_max_size, 3480 .maxlen = sizeof(int), 3481 .mode = 0644, 3482 .proc_handler = proc_dointvec, 3483 }, 3484 { 3485 /* Deprecated. Use gc_min_interval_ms */ 3486 3487 .procname = "gc_min_interval", 3488 .data = &ip_rt_gc_min_interval, 3489 .maxlen = sizeof(int), 3490 .mode = 0644, 3491 .proc_handler = proc_dointvec_jiffies, 3492 }, 3493 { 3494 .procname = "gc_min_interval_ms", 3495 .data = &ip_rt_gc_min_interval, 3496 .maxlen = sizeof(int), 3497 .mode = 0644, 3498 .proc_handler = proc_dointvec_ms_jiffies, 3499 }, 3500 { 3501 .procname = "gc_timeout", 3502 .data = &ip_rt_gc_timeout, 3503 .maxlen = sizeof(int), 3504 .mode = 0644, 3505 .proc_handler = proc_dointvec_jiffies, 3506 }, 3507 { 3508 .procname = "gc_interval", 3509 .data = &ip_rt_gc_interval, 3510 .maxlen = sizeof(int), 3511 .mode = 0644, 3512 .proc_handler = proc_dointvec_jiffies, 3513 }, 3514 { 3515 .procname = "redirect_load", 3516 .data = &ip_rt_redirect_load, 3517 .maxlen = sizeof(int), 3518 .mode = 0644, 3519 .proc_handler = proc_dointvec, 3520 }, 3521 { 3522 .procname = "redirect_number", 3523 .data = &ip_rt_redirect_number, 3524 .maxlen = sizeof(int), 3525 .mode = 0644, 3526 .proc_handler = proc_dointvec, 3527 }, 3528 { 3529 .procname = "redirect_silence", 3530 .data = &ip_rt_redirect_silence, 3531 .maxlen = sizeof(int), 3532 .mode = 0644, 3533 .proc_handler = proc_dointvec, 3534 }, 3535 { 3536 .procname = "error_cost", 3537 .data = &ip_rt_error_cost, 3538 .maxlen = sizeof(int), 3539 .mode = 0644, 3540 .proc_handler = proc_dointvec, 3541 }, 3542 { 3543 .procname = "error_burst", 3544 .data = &ip_rt_error_burst, 3545 .maxlen = sizeof(int), 3546 .mode = 0644, 3547 .proc_handler = proc_dointvec, 3548 }, 3549 { 3550 .procname = "gc_elasticity", 3551 .data = &ip_rt_gc_elasticity, 3552 .maxlen = sizeof(int), 3553 .mode = 0644, 3554 .proc_handler = proc_dointvec, 3555 }, 3556 { } 3557 }; 3558 3559 static const char ipv4_route_flush_procname[] = "flush"; 3560 3561 static struct ctl_table ipv4_route_netns_table[] = { 3562 { 3563 .procname = ipv4_route_flush_procname, 3564 .maxlen = sizeof(int), 3565 .mode = 0200, 3566 .proc_handler = ipv4_sysctl_rtcache_flush, 3567 }, 3568 { 3569 .procname = "min_pmtu", 3570 .data = &init_net.ipv4.ip_rt_min_pmtu, 3571 .maxlen = sizeof(int), 3572 .mode = 0644, 3573 .proc_handler = proc_dointvec_minmax, 3574 .extra1 = &ip_min_valid_pmtu, 3575 }, 3576 { 3577 .procname = "mtu_expires", 3578 .data = &init_net.ipv4.ip_rt_mtu_expires, 3579 .maxlen = sizeof(int), 3580 .mode = 0644, 3581 .proc_handler = proc_dointvec_jiffies, 3582 }, 3583 { 3584 .procname = "min_adv_mss", 3585 .data = &init_net.ipv4.ip_rt_min_advmss, 3586 .maxlen = sizeof(int), 3587 .mode = 0644, 3588 .proc_handler = proc_dointvec, 3589 }, 3590 { }, 3591 }; 3592 3593 static __net_init int sysctl_route_net_init(struct net *net) 3594 { 3595 struct ctl_table *tbl; 3596 size_t table_size = ARRAY_SIZE(ipv4_route_netns_table); 3597 3598 tbl = ipv4_route_netns_table; 3599 if (!net_eq(net, &init_net)) { 3600 int i; 3601 3602 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL); 3603 if (!tbl) 3604 goto err_dup; 3605 3606 /* Don't export non-whitelisted sysctls to unprivileged users */ 3607 if (net->user_ns != &init_user_ns) { 3608 if (tbl[0].procname != ipv4_route_flush_procname) { 3609 tbl[0].procname = NULL; 3610 table_size = 0; 3611 } 3612 } 3613 3614 /* Update the variables to point into the current struct net 3615 * except for the first element flush 3616 */ 3617 for (i = 1; i < ARRAY_SIZE(ipv4_route_netns_table) - 1; i++) 3618 tbl[i].data += (void *)net - (void *)&init_net; 3619 } 3620 tbl[0].extra1 = net; 3621 3622 net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route", 3623 tbl, table_size); 3624 if (!net->ipv4.route_hdr) 3625 goto err_reg; 3626 return 0; 3627 3628 err_reg: 3629 if (tbl != ipv4_route_netns_table) 3630 kfree(tbl); 3631 err_dup: 3632 return -ENOMEM; 3633 } 3634 3635 static __net_exit void sysctl_route_net_exit(struct net *net) 3636 { 3637 struct ctl_table *tbl; 3638 3639 tbl = net->ipv4.route_hdr->ctl_table_arg; 3640 unregister_net_sysctl_table(net->ipv4.route_hdr); 3641 BUG_ON(tbl == ipv4_route_netns_table); 3642 kfree(tbl); 3643 } 3644 3645 static __net_initdata struct pernet_operations sysctl_route_ops = { 3646 .init = sysctl_route_net_init, 3647 .exit = sysctl_route_net_exit, 3648 }; 3649 #endif 3650 3651 static __net_init int netns_ip_rt_init(struct net *net) 3652 { 3653 /* Set default value for namespaceified sysctls */ 3654 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU; 3655 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES; 3656 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS; 3657 return 0; 3658 } 3659 3660 static struct pernet_operations __net_initdata ip_rt_ops = { 3661 .init = netns_ip_rt_init, 3662 }; 3663 3664 static __net_init int rt_genid_init(struct net *net) 3665 { 3666 atomic_set(&net->ipv4.rt_genid, 0); 3667 atomic_set(&net->fnhe_genid, 0); 3668 atomic_set(&net->ipv4.dev_addr_genid, get_random_u32()); 3669 return 0; 3670 } 3671 3672 static __net_initdata struct pernet_operations rt_genid_ops = { 3673 .init = rt_genid_init, 3674 }; 3675 3676 static int __net_init ipv4_inetpeer_init(struct net *net) 3677 { 3678 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL); 3679 3680 if (!bp) 3681 return -ENOMEM; 3682 inet_peer_base_init(bp); 3683 net->ipv4.peers = bp; 3684 return 0; 3685 } 3686 3687 static void __net_exit ipv4_inetpeer_exit(struct net *net) 3688 { 3689 struct inet_peer_base *bp = net->ipv4.peers; 3690 3691 net->ipv4.peers = NULL; 3692 inetpeer_invalidate_tree(bp); 3693 kfree(bp); 3694 } 3695 3696 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = { 3697 .init = ipv4_inetpeer_init, 3698 .exit = ipv4_inetpeer_exit, 3699 }; 3700 3701 #ifdef CONFIG_IP_ROUTE_CLASSID 3702 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 3703 #endif /* CONFIG_IP_ROUTE_CLASSID */ 3704 3705 int __init ip_rt_init(void) 3706 { 3707 void *idents_hash; 3708 int cpu; 3709 3710 /* For modern hosts, this will use 2 MB of memory */ 3711 idents_hash = alloc_large_system_hash("IP idents", 3712 sizeof(*ip_idents) + sizeof(*ip_tstamps), 3713 0, 3714 16, /* one bucket per 64 KB */ 3715 HASH_ZERO, 3716 NULL, 3717 &ip_idents_mask, 3718 2048, 3719 256*1024); 3720 3721 ip_idents = idents_hash; 3722 3723 get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents)); 3724 3725 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents); 3726 3727 for_each_possible_cpu(cpu) { 3728 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 3729 3730 INIT_LIST_HEAD(&ul->head); 3731 INIT_LIST_HEAD(&ul->quarantine); 3732 spin_lock_init(&ul->lock); 3733 } 3734 #ifdef CONFIG_IP_ROUTE_CLASSID 3735 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 3736 if (!ip_rt_acct) 3737 panic("IP: failed to allocate ip_rt_acct\n"); 3738 #endif 3739 3740 ipv4_dst_ops.kmem_cachep = 3741 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0, 3742 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 3743 3744 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 3745 3746 if (dst_entries_init(&ipv4_dst_ops) < 0) 3747 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 3748 3749 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 3750 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 3751 3752 ipv4_dst_ops.gc_thresh = ~0; 3753 ip_rt_max_size = INT_MAX; 3754 3755 devinet_init(); 3756 ip_fib_init(); 3757 3758 if (ip_rt_proc_init()) 3759 pr_err("Unable to create route proc files\n"); 3760 #ifdef CONFIG_XFRM 3761 xfrm_init(); 3762 xfrm4_init(); 3763 #endif 3764 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, 3765 RTNL_FLAG_DOIT_UNLOCKED); 3766 3767 #ifdef CONFIG_SYSCTL 3768 register_pernet_subsys(&sysctl_route_ops); 3769 #endif 3770 register_pernet_subsys(&ip_rt_ops); 3771 register_pernet_subsys(&rt_genid_ops); 3772 register_pernet_subsys(&ipv4_inetpeer_ops); 3773 return 0; 3774 } 3775 3776 #ifdef CONFIG_SYSCTL 3777 /* 3778 * We really need to sanitize the damn ipv4 init order, then all 3779 * this nonsense will go away. 3780 */ 3781 void __init ip_static_sysctl_init(void) 3782 { 3783 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table); 3784 } 3785 #endif 3786