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/flow.h> 88 #include <net/inet_dscp.h> 89 #include <net/net_namespace.h> 90 #include <net/ip.h> 91 #include <net/route.h> 92 #include <net/inetpeer.h> 93 #include <net/sock.h> 94 #include <net/ip_fib.h> 95 #include <net/nexthop.h> 96 #include <net/tcp.h> 97 #include <net/icmp.h> 98 #include <net/xfrm.h> 99 #include <net/lwtunnel.h> 100 #include <net/netevent.h> 101 #include <net/rtnetlink.h> 102 #ifdef CONFIG_SYSCTL 103 #include <linux/sysctl.h> 104 #endif 105 #include <net/secure_seq.h> 106 #include <net/ip_tunnels.h> 107 108 #include "fib_lookup.h" 109 110 #define RT_GC_TIMEOUT (300*HZ) 111 112 #define DEFAULT_MIN_PMTU (512 + 20 + 20) 113 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ) 114 #define DEFAULT_MIN_ADVMSS 256 115 static int ip_rt_max_size; 116 static int ip_rt_redirect_number __read_mostly = 9; 117 static int ip_rt_redirect_load __read_mostly = HZ / 50; 118 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1)); 119 static int ip_rt_error_cost __read_mostly = HZ; 120 static int ip_rt_error_burst __read_mostly = 5 * HZ; 121 122 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT; 123 124 /* 125 * Interface to generic destination cache. 126 */ 127 128 INDIRECT_CALLABLE_SCOPE 129 struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie); 130 static unsigned int ipv4_default_advmss(const struct dst_entry *dst); 131 INDIRECT_CALLABLE_SCOPE 132 unsigned int ipv4_mtu(const struct dst_entry *dst); 133 static void ipv4_negative_advice(struct sock *sk, 134 struct dst_entry *dst); 135 static void ipv4_link_failure(struct sk_buff *skb); 136 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 137 struct sk_buff *skb, u32 mtu, 138 bool confirm_neigh); 139 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, 140 struct sk_buff *skb); 141 static void ipv4_dst_destroy(struct dst_entry *dst); 142 143 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old) 144 { 145 WARN_ON(1); 146 return NULL; 147 } 148 149 static struct neighbour *ipv4_dst_neigh_lookup(const struct dst_entry *dst, 150 struct sk_buff *skb, 151 const void *daddr); 152 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr); 153 154 static struct dst_ops ipv4_dst_ops = { 155 .family = AF_INET, 156 .check = ipv4_dst_check, 157 .default_advmss = ipv4_default_advmss, 158 .mtu = ipv4_mtu, 159 .cow_metrics = ipv4_cow_metrics, 160 .destroy = ipv4_dst_destroy, 161 .negative_advice = ipv4_negative_advice, 162 .link_failure = ipv4_link_failure, 163 .update_pmtu = ip_rt_update_pmtu, 164 .redirect = ip_do_redirect, 165 .local_out = __ip_local_out, 166 .neigh_lookup = ipv4_dst_neigh_lookup, 167 .confirm_neigh = ipv4_confirm_neigh, 168 }; 169 170 #define ECN_OR_COST(class) TC_PRIO_##class 171 172 const __u8 ip_tos2prio[16] = { 173 TC_PRIO_BESTEFFORT, 174 ECN_OR_COST(BESTEFFORT), 175 TC_PRIO_BESTEFFORT, 176 ECN_OR_COST(BESTEFFORT), 177 TC_PRIO_BULK, 178 ECN_OR_COST(BULK), 179 TC_PRIO_BULK, 180 ECN_OR_COST(BULK), 181 TC_PRIO_INTERACTIVE, 182 ECN_OR_COST(INTERACTIVE), 183 TC_PRIO_INTERACTIVE, 184 ECN_OR_COST(INTERACTIVE), 185 TC_PRIO_INTERACTIVE_BULK, 186 ECN_OR_COST(INTERACTIVE_BULK), 187 TC_PRIO_INTERACTIVE_BULK, 188 ECN_OR_COST(INTERACTIVE_BULK) 189 }; 190 EXPORT_SYMBOL(ip_tos2prio); 191 192 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat); 193 #ifndef CONFIG_PREEMPT_RT 194 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field) 195 #else 196 #define RT_CACHE_STAT_INC(field) this_cpu_inc(rt_cache_stat.field) 197 #endif 198 199 #ifdef CONFIG_PROC_FS 200 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos) 201 { 202 if (*pos) 203 return NULL; 204 return SEQ_START_TOKEN; 205 } 206 207 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos) 208 { 209 ++*pos; 210 return NULL; 211 } 212 213 static void rt_cache_seq_stop(struct seq_file *seq, void *v) 214 { 215 } 216 217 static int rt_cache_seq_show(struct seq_file *seq, void *v) 218 { 219 if (v == SEQ_START_TOKEN) 220 seq_printf(seq, "%-127s\n", 221 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t" 222 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t" 223 "HHUptod\tSpecDst"); 224 return 0; 225 } 226 227 static const struct seq_operations rt_cache_seq_ops = { 228 .start = rt_cache_seq_start, 229 .next = rt_cache_seq_next, 230 .stop = rt_cache_seq_stop, 231 .show = rt_cache_seq_show, 232 }; 233 234 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos) 235 { 236 int cpu; 237 238 if (*pos == 0) 239 return SEQ_START_TOKEN; 240 241 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) { 242 if (!cpu_possible(cpu)) 243 continue; 244 *pos = cpu+1; 245 return &per_cpu(rt_cache_stat, cpu); 246 } 247 return NULL; 248 } 249 250 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos) 251 { 252 int cpu; 253 254 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) { 255 if (!cpu_possible(cpu)) 256 continue; 257 *pos = cpu+1; 258 return &per_cpu(rt_cache_stat, cpu); 259 } 260 (*pos)++; 261 return NULL; 262 263 } 264 265 static void rt_cpu_seq_stop(struct seq_file *seq, void *v) 266 { 267 268 } 269 270 static int rt_cpu_seq_show(struct seq_file *seq, void *v) 271 { 272 struct rt_cache_stat *st = v; 273 274 if (v == SEQ_START_TOKEN) { 275 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"); 276 return 0; 277 } 278 279 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x " 280 "%08x %08x %08x %08x %08x %08x " 281 "%08x %08x %08x %08x\n", 282 dst_entries_get_slow(&ipv4_dst_ops), 283 0, /* st->in_hit */ 284 st->in_slow_tot, 285 st->in_slow_mc, 286 st->in_no_route, 287 st->in_brd, 288 st->in_martian_dst, 289 st->in_martian_src, 290 291 0, /* st->out_hit */ 292 st->out_slow_tot, 293 st->out_slow_mc, 294 295 0, /* st->gc_total */ 296 0, /* st->gc_ignored */ 297 0, /* st->gc_goal_miss */ 298 0, /* st->gc_dst_overflow */ 299 0, /* st->in_hlist_search */ 300 0 /* st->out_hlist_search */ 301 ); 302 return 0; 303 } 304 305 static const struct seq_operations rt_cpu_seq_ops = { 306 .start = rt_cpu_seq_start, 307 .next = rt_cpu_seq_next, 308 .stop = rt_cpu_seq_stop, 309 .show = rt_cpu_seq_show, 310 }; 311 312 #ifdef CONFIG_IP_ROUTE_CLASSID 313 static int rt_acct_proc_show(struct seq_file *m, void *v) 314 { 315 struct ip_rt_acct *dst, *src; 316 unsigned int i, j; 317 318 dst = kzalloc_objs(struct ip_rt_acct, 256); 319 if (!dst) 320 return -ENOMEM; 321 322 for_each_possible_cpu(i) { 323 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i); 324 for (j = 0; j < 256; j++) { 325 dst[j].o_bytes += src[j].o_bytes; 326 dst[j].o_packets += src[j].o_packets; 327 dst[j].i_bytes += src[j].i_bytes; 328 dst[j].i_packets += src[j].i_packets; 329 } 330 } 331 332 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct)); 333 kfree(dst); 334 return 0; 335 } 336 #endif 337 338 static int __net_init ip_rt_do_proc_init(struct net *net) 339 { 340 struct proc_dir_entry *pde; 341 342 pde = proc_create_seq("rt_cache", 0444, net->proc_net, 343 &rt_cache_seq_ops); 344 if (!pde) 345 goto err1; 346 347 pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat, 348 &rt_cpu_seq_ops); 349 if (!pde) 350 goto err2; 351 352 #ifdef CONFIG_IP_ROUTE_CLASSID 353 pde = proc_create_single("rt_acct", 0, net->proc_net, 354 rt_acct_proc_show); 355 if (!pde) 356 goto err3; 357 #endif 358 return 0; 359 360 #ifdef CONFIG_IP_ROUTE_CLASSID 361 err3: 362 remove_proc_entry("rt_cache", net->proc_net_stat); 363 #endif 364 err2: 365 remove_proc_entry("rt_cache", net->proc_net); 366 err1: 367 return -ENOMEM; 368 } 369 370 static void __net_exit ip_rt_do_proc_exit(struct net *net) 371 { 372 remove_proc_entry("rt_cache", net->proc_net_stat); 373 remove_proc_entry("rt_cache", net->proc_net); 374 #ifdef CONFIG_IP_ROUTE_CLASSID 375 remove_proc_entry("rt_acct", net->proc_net); 376 #endif 377 } 378 379 static struct pernet_operations ip_rt_proc_ops __net_initdata = { 380 .init = ip_rt_do_proc_init, 381 .exit = ip_rt_do_proc_exit, 382 }; 383 384 static int __init ip_rt_proc_init(void) 385 { 386 return register_pernet_subsys(&ip_rt_proc_ops); 387 } 388 389 #else 390 static inline int ip_rt_proc_init(void) 391 { 392 return 0; 393 } 394 #endif /* CONFIG_PROC_FS */ 395 396 static inline bool rt_is_expired(const struct rtable *rth) 397 { 398 bool res; 399 400 rcu_read_lock(); 401 res = rth->rt_genid != rt_genid_ipv4(dev_net_rcu(rth->dst.dev)); 402 rcu_read_unlock(); 403 404 return res; 405 } 406 407 void rt_cache_flush(struct net *net) 408 { 409 rt_genid_bump_ipv4(net); 410 } 411 412 static struct neighbour *ipv4_dst_neigh_lookup(const struct dst_entry *dst, 413 struct sk_buff *skb, 414 const void *daddr) 415 { 416 const struct rtable *rt = container_of(dst, struct rtable, dst); 417 struct net_device *dev; 418 struct neighbour *n; 419 420 rcu_read_lock(); 421 dev = dst_dev_rcu(dst); 422 if (likely(rt->rt_gw_family == AF_INET)) { 423 n = ip_neigh_gw4(dev, rt->rt_gw4); 424 } else if (rt->rt_gw_family == AF_INET6) { 425 n = ip_neigh_gw6(dev, &rt->rt_gw6); 426 } else { 427 __be32 pkey; 428 429 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr); 430 n = ip_neigh_gw4(dev, pkey); 431 } 432 433 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt)) 434 n = NULL; 435 436 rcu_read_unlock(); 437 438 return n; 439 } 440 441 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr) 442 { 443 const struct rtable *rt = container_of(dst, struct rtable, dst); 444 struct net_device *dev = dst_dev(dst); 445 const __be32 *pkey = daddr; 446 447 if (rt->rt_gw_family == AF_INET) { 448 pkey = (const __be32 *)&rt->rt_gw4; 449 } else if (IS_ENABLED(CONFIG_IPV6) && rt->rt_gw_family == AF_INET6) { 450 return __ipv6_confirm_neigh(dev, &rt->rt_gw6); 451 } else if (!daddr || 452 (rt->rt_flags & 453 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) { 454 return; 455 } 456 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey); 457 } 458 459 /* Hash tables of size 2048..262144 depending on RAM size. 460 * Each bucket uses 8 bytes. 461 */ 462 static u32 ip_idents_mask __read_mostly; 463 static atomic_t *ip_idents __read_mostly; 464 static u32 *ip_tstamps __read_mostly; 465 466 /* In order to protect privacy, we add a perturbation to identifiers 467 * if one generator is seldom used. This makes hard for an attacker 468 * to infer how many packets were sent between two points in time. 469 */ 470 static u32 ip_idents_reserve(u32 hash, int segs) 471 { 472 u32 bucket, old, now = (u32)jiffies; 473 atomic_t *p_id; 474 u32 *p_tstamp; 475 u32 delta = 0; 476 477 bucket = hash & ip_idents_mask; 478 p_tstamp = ip_tstamps + bucket; 479 p_id = ip_idents + bucket; 480 old = READ_ONCE(*p_tstamp); 481 482 if (old != now && cmpxchg(p_tstamp, old, now) == old) 483 delta = get_random_u32_below(now - old); 484 485 /* If UBSAN reports an error there, please make sure your compiler 486 * supports -fno-strict-overflow before reporting it that was a bug 487 * in UBSAN, and it has been fixed in GCC-8. 488 */ 489 return atomic_add_return(segs + delta, p_id) - segs; 490 } 491 492 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs) 493 { 494 u32 hash, id; 495 496 /* Note the following code is not safe, but this is okay. */ 497 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key))) 498 get_random_bytes(&net->ipv4.ip_id_key, 499 sizeof(net->ipv4.ip_id_key)); 500 501 hash = siphash_3u32((__force u32)iph->daddr, 502 (__force u32)iph->saddr, 503 iph->protocol, 504 &net->ipv4.ip_id_key); 505 id = ip_idents_reserve(hash, segs); 506 iph->id = htons(id); 507 } 508 EXPORT_SYMBOL(__ip_select_ident); 509 510 static void __build_flow_key(const struct net *net, struct flowi4 *fl4, 511 const struct sock *sk, const struct iphdr *iph, 512 int oif, __u8 tos, u8 prot, u32 mark, 513 int flow_flags) 514 { 515 __u8 scope = RT_SCOPE_UNIVERSE; 516 517 if (sk) { 518 oif = sk->sk_bound_dev_if; 519 mark = READ_ONCE(sk->sk_mark); 520 tos = ip_sock_rt_tos(sk); 521 scope = ip_sock_rt_scope(sk); 522 prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW : 523 sk->sk_protocol; 524 } 525 526 flowi4_init_output(fl4, oif, mark, tos & INET_DSCP_MASK, scope, 527 prot, flow_flags, iph->daddr, iph->saddr, 0, 0, 528 sock_net_uid(net, sk)); 529 } 530 531 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb, 532 const struct sock *sk) 533 { 534 const struct net *net = dev_net(skb->dev); 535 const struct iphdr *iph = ip_hdr(skb); 536 int oif = skb->dev->ifindex; 537 u8 prot = iph->protocol; 538 u32 mark = skb->mark; 539 __u8 tos = iph->tos; 540 541 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0); 542 } 543 544 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk) 545 { 546 const struct inet_sock *inet = inet_sk(sk); 547 const struct ip_options_rcu *inet_opt; 548 __be32 daddr = inet->inet_daddr; 549 550 rcu_read_lock(); 551 inet_opt = rcu_dereference(inet->inet_opt); 552 if (inet_opt && inet_opt->opt.srr) 553 daddr = inet_opt->opt.faddr; 554 flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark), 555 ip_sock_rt_tos(sk), 556 ip_sock_rt_scope(sk), 557 inet_test_bit(HDRINCL, sk) ? 558 IPPROTO_RAW : sk->sk_protocol, 559 inet_sk_flowi_flags(sk), 560 daddr, inet->inet_saddr, 0, 0, 561 sk_uid(sk)); 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 } 610 611 /* Clear oldest->fnhe_daddr to prevent this fnhe from being 612 * rebound with new dsts in rt_bind_exception(). 613 */ 614 oldest->fnhe_daddr = 0; 615 fnhe_flush_routes(oldest); 616 *oldest_p = oldest->fnhe_next; 617 kfree_rcu(oldest, rcu); 618 } 619 620 static u32 fnhe_hashfun(__be32 daddr) 621 { 622 static siphash_aligned_key_t fnhe_hash_key; 623 u64 hval; 624 625 net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key)); 626 hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key); 627 return hash_64(hval, FNHE_HASH_SHIFT); 628 } 629 630 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe) 631 { 632 rt->rt_pmtu = fnhe->fnhe_pmtu; 633 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked; 634 rt->dst.expires = fnhe->fnhe_expires; 635 636 if (fnhe->fnhe_gw) { 637 rt->rt_flags |= RTCF_REDIRECTED; 638 rt->rt_uses_gateway = 1; 639 rt->rt_gw_family = AF_INET; 640 rt->rt_gw4 = fnhe->fnhe_gw; 641 } 642 } 643 644 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr, 645 __be32 gw, u32 pmtu, bool lock, 646 unsigned long expires) 647 { 648 struct fnhe_hash_bucket *hash; 649 struct fib_nh_exception *fnhe; 650 struct rtable *rt; 651 u32 genid, hval; 652 unsigned int i; 653 int depth; 654 655 genid = fnhe_genid(dev_net(nhc->nhc_dev)); 656 hval = fnhe_hashfun(daddr); 657 658 spin_lock_bh(&fnhe_lock); 659 660 hash = rcu_dereference(nhc->nhc_exceptions); 661 if (!hash) { 662 hash = kzalloc_objs(*hash, FNHE_HASH_SIZE, GFP_ATOMIC); 663 if (!hash) 664 goto out_unlock; 665 rcu_assign_pointer(nhc->nhc_exceptions, hash); 666 } 667 668 hash += hval; 669 670 depth = 0; 671 for (fnhe = rcu_dereference(hash->chain); fnhe; 672 fnhe = rcu_dereference(fnhe->fnhe_next)) { 673 if (fnhe->fnhe_daddr == daddr) 674 break; 675 depth++; 676 } 677 678 if (fnhe) { 679 if (fnhe->fnhe_genid != genid) 680 fnhe->fnhe_genid = genid; 681 if (gw) 682 fnhe->fnhe_gw = gw; 683 if (pmtu) { 684 fnhe->fnhe_pmtu = pmtu; 685 fnhe->fnhe_mtu_locked = lock; 686 } 687 fnhe->fnhe_expires = max(1UL, expires); 688 /* Update all cached dsts too */ 689 rt = rcu_dereference(fnhe->fnhe_rth_input); 690 if (rt) 691 fill_route_from_fnhe(rt, fnhe); 692 rt = rcu_dereference(fnhe->fnhe_rth_output); 693 if (rt) 694 fill_route_from_fnhe(rt, fnhe); 695 } else { 696 /* Randomize max depth to avoid some side channels attacks. */ 697 int max_depth = FNHE_RECLAIM_DEPTH + 698 get_random_u32_below(FNHE_RECLAIM_DEPTH); 699 700 while (depth > max_depth) { 701 fnhe_remove_oldest(hash); 702 depth--; 703 } 704 705 fnhe = kzalloc_obj(*fnhe, GFP_ATOMIC); 706 if (!fnhe) 707 goto out_unlock; 708 709 fnhe->fnhe_next = hash->chain; 710 711 fnhe->fnhe_genid = genid; 712 fnhe->fnhe_daddr = daddr; 713 fnhe->fnhe_gw = gw; 714 fnhe->fnhe_pmtu = pmtu; 715 fnhe->fnhe_mtu_locked = lock; 716 fnhe->fnhe_expires = max(1UL, expires); 717 718 rcu_assign_pointer(hash->chain, fnhe); 719 720 /* Exception created; mark the cached routes for the nexthop 721 * stale, so anyone caching it rechecks if this exception 722 * applies to them. 723 */ 724 rt = rcu_dereference(nhc->nhc_rth_input); 725 if (rt) 726 WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL); 727 728 for_each_possible_cpu(i) { 729 struct rtable __rcu **prt; 730 731 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i); 732 rt = rcu_dereference(*prt); 733 if (rt) 734 WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL); 735 } 736 } 737 738 fnhe->fnhe_stamp = jiffies; 739 740 out_unlock: 741 spin_unlock_bh(&fnhe_lock); 742 } 743 744 /* Update the PMTU of an exception when: 745 * - the new MTU of the first hop becomes smaller than the PMTU 746 * - the old MTU was the same as the PMTU, and it limited discovery of 747 * larger MTUs on the path. With that limit raised, we can now 748 * discover larger MTUs 749 * A special case is locked exceptions, for which the PMTU is smaller 750 * than the minimal accepted PMTU: 751 * - if the new MTU is greater than the PMTU, don't make any change 752 * - otherwise, unlock and set PMTU 753 * 754 * fnhe_lock keeps fnhe_pmtu and fnhe_mtu_locked consistent against 755 * update_or_create_fnhe(), which sets both under the same lock. 756 */ 757 void fnhe_update_pmtu(struct fib_nh_exception *fnhe, u32 new, u32 orig) 758 { 759 spin_lock_bh(&fnhe_lock); 760 761 if (fnhe->fnhe_mtu_locked) { 762 if (new <= fnhe->fnhe_pmtu) { 763 fnhe->fnhe_pmtu = new; 764 fnhe->fnhe_mtu_locked = false; 765 } 766 } else if (new < fnhe->fnhe_pmtu || orig == fnhe->fnhe_pmtu) { 767 fnhe->fnhe_pmtu = new; 768 } 769 770 spin_unlock_bh(&fnhe_lock); 771 } 772 773 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4, 774 bool kill_route) 775 { 776 __be32 new_gw = icmp_hdr(skb)->un.gateway; 777 __be32 old_gw = ip_hdr(skb)->saddr; 778 struct net_device *dev = skb->dev; 779 struct in_device *in_dev; 780 struct fib_result res; 781 struct neighbour *n; 782 struct net *net; 783 784 switch (icmp_hdr(skb)->code & 7) { 785 case ICMP_REDIR_NET: 786 case ICMP_REDIR_NETTOS: 787 case ICMP_REDIR_HOST: 788 case ICMP_REDIR_HOSTTOS: 789 break; 790 791 default: 792 return; 793 } 794 795 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw) 796 return; 797 798 in_dev = __in_dev_get_rcu(dev); 799 if (!in_dev) 800 return; 801 802 net = dev_net(dev); 803 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) || 804 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) || 805 ipv4_is_zeronet(new_gw)) 806 goto reject_redirect; 807 808 if (!IN_DEV_SHARED_MEDIA(in_dev)) { 809 if (!inet_addr_onlink(in_dev, new_gw, old_gw)) 810 goto reject_redirect; 811 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) 812 goto reject_redirect; 813 } else { 814 if (inet_addr_type(net, new_gw) != RTN_UNICAST) 815 goto reject_redirect; 816 } 817 818 n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw); 819 if (!n) 820 n = ipv4_neigh_create(rt->dst.dev, &new_gw); 821 if (!IS_ERR(n)) { 822 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) { 823 neigh_event_send(n, NULL); 824 } else { 825 if (fib_lookup(net, fl4, &res, 0) == 0) { 826 struct fib_nh_common *nhc; 827 828 fib_select_path(net, &res, fl4, skb); 829 nhc = FIB_RES_NHC(res); 830 update_or_create_fnhe(nhc, fl4->daddr, new_gw, 831 0, false, 832 jiffies + ip_rt_gc_timeout); 833 } 834 if (kill_route) 835 WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL); 836 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n); 837 } 838 neigh_release(n); 839 } 840 return; 841 842 reject_redirect: 843 #ifdef CONFIG_IP_ROUTE_VERBOSE 844 if (IN_DEV_LOG_MARTIANS(in_dev)) { 845 const struct iphdr *iph = (const struct iphdr *) skb->data; 846 __be32 daddr = iph->daddr; 847 __be32 saddr = iph->saddr; 848 849 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n" 850 " Advised path = %pI4 -> %pI4\n", 851 &old_gw, dev->name, &new_gw, 852 &saddr, &daddr); 853 } 854 #endif 855 ; 856 } 857 858 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb) 859 { 860 struct rtable *rt; 861 struct flowi4 fl4; 862 const struct iphdr *iph = (const struct iphdr *) skb->data; 863 struct net *net = dev_net(skb->dev); 864 int oif = skb->dev->ifindex; 865 u8 prot = iph->protocol; 866 u32 mark = skb->mark; 867 __u8 tos = iph->tos; 868 869 rt = dst_rtable(dst); 870 871 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0); 872 __ip_do_redirect(rt, skb, &fl4, true); 873 } 874 875 static void ipv4_negative_advice(struct sock *sk, 876 struct dst_entry *dst) 877 { 878 struct rtable *rt = dst_rtable(dst); 879 880 if ((READ_ONCE(dst->obsolete) > 0) || 881 (rt->rt_flags & RTCF_REDIRECTED) || 882 READ_ONCE(rt->dst.expires)) 883 sk_dst_reset(sk); 884 } 885 886 /* 887 * Algorithm: 888 * 1. The first ip_rt_redirect_number redirects are sent 889 * with exponential backoff, then we stop sending them at all, 890 * assuming that the host ignores our redirects. 891 * 2. If we did not see packets requiring redirects 892 * during ip_rt_redirect_silence, we assume that the host 893 * forgot redirected route and start to send redirects again. 894 * 895 * This algorithm is much cheaper and more intelligent than dumb load limiting 896 * in icmp.c. 897 * 898 * NOTE. Do not forget to inhibit load limiting for redirects (redundant) 899 * and "frag. need" (breaks PMTU discovery) in icmp.c. 900 */ 901 902 void ip_rt_send_redirect(struct sk_buff *skb) 903 { 904 struct rtable *rt = skb_rtable(skb); 905 struct in_device *in_dev; 906 struct net_device *dev; 907 struct inet_peer *peer; 908 int log_martians; 909 struct net *net; 910 int vif; 911 912 rcu_read_lock(); 913 dev = dst_dev_rcu(&rt->dst); 914 in_dev = __in_dev_get_rcu(dev); 915 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) { 916 rcu_read_unlock(); 917 return; 918 } 919 log_martians = IN_DEV_LOG_MARTIANS(in_dev); 920 vif = l3mdev_master_ifindex_rcu(dev); 921 922 net = dev_net_rcu(dev); 923 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif); 924 if (!peer) { 925 rcu_read_unlock(); 926 return; 927 } 928 929 /* No redirected packets during ip_rt_redirect_silence; 930 * reset the algorithm. 931 */ 932 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) { 933 peer->rate_tokens = 0; 934 peer->n_redirects = 0; 935 } 936 937 /* Too many ignored redirects; do not send anything 938 * set dst.rate_last to the last seen redirected packet. 939 */ 940 if (peer->n_redirects >= ip_rt_redirect_number) { 941 peer->rate_last = jiffies; 942 goto out_unlock; 943 } 944 945 /* Check for load limit; set rate_last to the latest sent 946 * redirect. 947 */ 948 if (peer->n_redirects == 0 || 949 time_after(jiffies, 950 (peer->rate_last + 951 (ip_rt_redirect_load << peer->n_redirects)))) { 952 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr); 953 954 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw); 955 peer->rate_last = jiffies; 956 ++peer->n_redirects; 957 if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians && 958 peer->n_redirects == ip_rt_redirect_number) 959 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n", 960 &ip_hdr(skb)->saddr, inet_iif(skb), 961 &ip_hdr(skb)->daddr, &gw); 962 } 963 out_unlock: 964 rcu_read_unlock(); 965 } 966 967 static int ip_error(struct sk_buff *skb) 968 { 969 struct rtable *rt = skb_rtable(skb); 970 struct net_device *dev = skb->dev; 971 struct in_device *in_dev; 972 struct inet_peer *peer; 973 unsigned long now; 974 struct net *net; 975 SKB_DR(reason); 976 bool send; 977 int code; 978 979 if (netif_is_l3_master(skb->dev)) { 980 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif); 981 if (!dev) 982 goto out; 983 } 984 985 in_dev = __in_dev_get_rcu(dev); 986 987 /* IP on this device is disabled. */ 988 if (!in_dev) 989 goto out; 990 991 net = dev_net(rt->dst.dev); 992 if (!IN_DEV_FORWARD(in_dev)) { 993 switch (rt->dst.error) { 994 case EHOSTUNREACH: 995 SKB_DR_SET(reason, IP_INADDRERRORS); 996 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS); 997 break; 998 999 case ENETUNREACH: 1000 SKB_DR_SET(reason, IP_INNOROUTES); 1001 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 1002 break; 1003 } 1004 goto out; 1005 } 1006 1007 switch (rt->dst.error) { 1008 case EINVAL: 1009 default: 1010 goto out; 1011 case EHOSTUNREACH: 1012 code = ICMP_HOST_UNREACH; 1013 break; 1014 case ENETUNREACH: 1015 code = ICMP_NET_UNREACH; 1016 SKB_DR_SET(reason, IP_INNOROUTES); 1017 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 1018 break; 1019 case EACCES: 1020 code = ICMP_PKT_FILTERED; 1021 break; 1022 } 1023 1024 rcu_read_lock(); 1025 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1026 l3mdev_master_ifindex_rcu(skb->dev)); 1027 send = true; 1028 if (peer) { 1029 now = jiffies; 1030 peer->rate_tokens += now - peer->rate_last; 1031 if (peer->rate_tokens > ip_rt_error_burst) 1032 peer->rate_tokens = ip_rt_error_burst; 1033 peer->rate_last = now; 1034 if (peer->rate_tokens >= ip_rt_error_cost) 1035 peer->rate_tokens -= ip_rt_error_cost; 1036 else 1037 send = false; 1038 } 1039 rcu_read_unlock(); 1040 1041 if (send) 1042 icmp_send(skb, ICMP_DEST_UNREACH, code, 0); 1043 1044 out: kfree_skb_reason(skb, reason); 1045 return 0; 1046 } 1047 1048 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu) 1049 { 1050 struct dst_entry *dst = &rt->dst; 1051 struct fib_result res; 1052 bool lock = false; 1053 struct net *net; 1054 u32 old_mtu; 1055 1056 if (ip_mtu_locked(dst)) 1057 return; 1058 1059 old_mtu = ipv4_mtu(dst); 1060 if (old_mtu < mtu) 1061 return; 1062 1063 rcu_read_lock(); 1064 net = dst_dev_net_rcu(dst); 1065 if (mtu < net->ipv4.ip_rt_min_pmtu) { 1066 lock = true; 1067 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu); 1068 } 1069 1070 if (rt->rt_pmtu == mtu && !lock && 1071 time_before(jiffies, READ_ONCE(dst->expires) - 1072 net->ipv4.ip_rt_mtu_expires / 2)) 1073 goto out; 1074 1075 if (fib_lookup(net, fl4, &res, 0) == 0) { 1076 struct fib_nh_common *nhc; 1077 1078 fib_select_path(net, &res, fl4, NULL); 1079 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1080 if (fib_info_num_path(res.fi) > 1) { 1081 int nhsel; 1082 1083 for (nhsel = 0; nhsel < fib_info_num_path(res.fi); nhsel++) { 1084 nhc = fib_info_nhc(res.fi, nhsel); 1085 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1086 jiffies + net->ipv4.ip_rt_mtu_expires); 1087 } 1088 goto out; 1089 } 1090 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 1091 nhc = FIB_RES_NHC(res); 1092 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1093 jiffies + net->ipv4.ip_rt_mtu_expires); 1094 } 1095 out: 1096 rcu_read_unlock(); 1097 } 1098 1099 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 1100 struct sk_buff *skb, u32 mtu, 1101 bool confirm_neigh) 1102 { 1103 struct rtable *rt = dst_rtable(dst); 1104 struct flowi4 fl4; 1105 1106 ip_rt_build_flow_key(&fl4, sk, skb); 1107 1108 /* Don't make lookup fail for bridged encapsulations */ 1109 if (skb && netif_is_any_bridge_port(skb->dev)) 1110 fl4.flowi4_oif = 0; 1111 1112 __ip_rt_update_pmtu(rt, &fl4, mtu); 1113 } 1114 1115 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, 1116 int oif, u8 protocol) 1117 { 1118 const struct iphdr *iph = (const struct iphdr *)skb->data; 1119 struct flowi4 fl4; 1120 struct rtable *rt; 1121 u32 mark = IP4_REPLY_MARK(net, skb->mark); 1122 1123 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark, 1124 0); 1125 rt = __ip_route_output_key(net, &fl4); 1126 if (!IS_ERR(rt)) { 1127 __ip_rt_update_pmtu(rt, &fl4, mtu); 1128 ip_rt_put(rt); 1129 } 1130 } 1131 EXPORT_SYMBOL_GPL(ipv4_update_pmtu); 1132 1133 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1134 { 1135 const struct iphdr *iph = (const struct iphdr *)skb->data; 1136 struct flowi4 fl4; 1137 struct rtable *rt; 1138 1139 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0); 1140 1141 if (!fl4.flowi4_mark) 1142 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark); 1143 1144 rt = __ip_route_output_key(sock_net(sk), &fl4); 1145 if (!IS_ERR(rt)) { 1146 __ip_rt_update_pmtu(rt, &fl4, mtu); 1147 ip_rt_put(rt); 1148 } 1149 } 1150 1151 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1152 { 1153 const struct iphdr *iph = (const struct iphdr *)skb->data; 1154 struct flowi4 fl4; 1155 struct rtable *rt; 1156 struct dst_entry *odst = NULL; 1157 bool new = false; 1158 struct net *net = sock_net(sk); 1159 1160 bh_lock_sock(sk); 1161 1162 if (!ip_sk_accept_pmtu(sk)) 1163 goto out; 1164 1165 odst = sk_dst_get(sk); 1166 1167 if (sock_owned_by_user(sk) || !odst) { 1168 __ipv4_sk_update_pmtu(skb, sk, mtu); 1169 goto out; 1170 } 1171 1172 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1173 1174 rt = dst_rtable(odst); 1175 if (READ_ONCE(odst->obsolete) && !odst->ops->check(odst, 0)) { 1176 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1177 if (IS_ERR(rt)) 1178 goto out; 1179 1180 new = true; 1181 } 1182 1183 __ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu); 1184 1185 if (!dst_check(&rt->dst, 0)) { 1186 if (new) 1187 dst_release(&rt->dst); 1188 1189 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1190 if (IS_ERR(rt)) 1191 goto out; 1192 1193 new = true; 1194 } 1195 1196 if (new) 1197 sk_dst_set(sk, &rt->dst); 1198 1199 out: 1200 bh_unlock_sock(sk); 1201 dst_release(odst); 1202 } 1203 1204 void ipv4_redirect(struct sk_buff *skb, struct net *net, 1205 int oif, u8 protocol) 1206 { 1207 const struct iphdr *iph = (const struct iphdr *)skb->data; 1208 struct flowi4 fl4; 1209 struct rtable *rt; 1210 1211 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0); 1212 rt = __ip_route_output_key(net, &fl4); 1213 if (!IS_ERR(rt)) { 1214 __ip_do_redirect(rt, skb, &fl4, false); 1215 ip_rt_put(rt); 1216 } 1217 } 1218 EXPORT_SYMBOL_GPL(ipv4_redirect); 1219 1220 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk) 1221 { 1222 const struct iphdr *iph = (const struct iphdr *)skb->data; 1223 struct flowi4 fl4; 1224 struct rtable *rt; 1225 struct net *net = sock_net(sk); 1226 1227 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1228 rt = __ip_route_output_key(net, &fl4); 1229 if (!IS_ERR(rt)) { 1230 __ip_do_redirect(rt, skb, &fl4, false); 1231 ip_rt_put(rt); 1232 } 1233 } 1234 1235 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst, 1236 u32 cookie) 1237 { 1238 struct rtable *rt = dst_rtable(dst); 1239 1240 /* All IPV4 dsts are created with ->obsolete set to the value 1241 * DST_OBSOLETE_FORCE_CHK which forces validation calls down 1242 * into this function always. 1243 * 1244 * When a PMTU/redirect information update invalidates a route, 1245 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or 1246 * DST_OBSOLETE_DEAD. 1247 */ 1248 if (READ_ONCE(dst->obsolete) != DST_OBSOLETE_FORCE_CHK || 1249 rt_is_expired(rt)) 1250 return NULL; 1251 return dst; 1252 } 1253 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check); 1254 1255 static void ipv4_send_dest_unreach(struct sk_buff *skb) 1256 { 1257 struct inet_skb_parm parm; 1258 struct net_device *dev; 1259 int res; 1260 1261 /* Recompile ip options since IPCB may not be valid anymore. 1262 * Also check we have a reasonable ipv4 header. 1263 */ 1264 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) || 1265 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5) 1266 return; 1267 1268 memset(&parm, 0, sizeof(parm)); 1269 if (ip_hdr(skb)->ihl > 5) { 1270 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4)) 1271 return; 1272 parm.opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr); 1273 1274 rcu_read_lock(); 1275 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev; 1276 res = __ip_options_compile(dev_net(dev), &parm.opt, skb, NULL); 1277 rcu_read_unlock(); 1278 1279 if (res) 1280 return; 1281 } 1282 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &parm); 1283 } 1284 1285 static void ipv4_link_failure(struct sk_buff *skb) 1286 { 1287 struct rtable *rt; 1288 1289 ipv4_send_dest_unreach(skb); 1290 1291 rt = skb_rtable(skb); 1292 if (rt) 1293 dst_set_expires(&rt->dst, 0); 1294 } 1295 1296 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb) 1297 { 1298 pr_debug("%s: %pI4 -> %pI4, %s\n", 1299 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1300 skb->dev ? skb->dev->name : "?"); 1301 kfree_skb(skb); 1302 WARN_ON_ONCE(1); 1303 return 0; 1304 } 1305 1306 /* 1307 * We do not cache source address of outgoing interface, 1308 * because it is used only by IP RR, TS and SRR options, 1309 * so that it out of fast path. 1310 * 1311 * BTW remember: "addr" is allowed to be not aligned 1312 * in IP options! 1313 */ 1314 1315 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt) 1316 { 1317 __be32 src; 1318 1319 rcu_read_lock(); 1320 if (rt_is_output_route(rt)) { 1321 src = ip_hdr(skb)->saddr; 1322 } else { 1323 struct net_device *dev = dst_dev_rcu(&rt->dst); 1324 struct net *net = dev_net_rcu(dev); 1325 struct iphdr *iph = ip_hdr(skb); 1326 struct fib_result res; 1327 struct flowi4 fl4 = { 1328 .daddr = iph->daddr, 1329 .saddr = iph->saddr, 1330 .flowi4_dscp = ip4h_dscp(iph), 1331 .flowi4_oif = dev->ifindex, 1332 .flowi4_iif = skb->dev->ifindex, 1333 .flowi4_mark = skb->mark, 1334 }; 1335 1336 if (fib_lookup(net, &fl4, &res, 0) == 0) 1337 src = fib_result_prefsrc(net, &res); 1338 else 1339 src = inet_select_addr(dev, 1340 rt_nexthop(rt, iph->daddr), 1341 RT_SCOPE_UNIVERSE); 1342 } 1343 rcu_read_unlock(); 1344 1345 memcpy(addr, &src, 4); 1346 } 1347 1348 #ifdef CONFIG_IP_ROUTE_CLASSID 1349 static void set_class_tag(struct rtable *rt, u32 tag) 1350 { 1351 if (!(rt->dst.tclassid & 0xFFFF)) 1352 rt->dst.tclassid |= tag & 0xFFFF; 1353 if (!(rt->dst.tclassid & 0xFFFF0000)) 1354 rt->dst.tclassid |= tag & 0xFFFF0000; 1355 } 1356 #endif 1357 1358 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1359 { 1360 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr); 1361 unsigned int advmss; 1362 struct net *net; 1363 1364 rcu_read_lock(); 1365 net = dst_dev_net_rcu(dst); 1366 advmss = max_t(unsigned int, ip_dst_mtu_configured(dst) - header_size, 1367 net->ipv4.ip_rt_min_advmss); 1368 rcu_read_unlock(); 1369 1370 return min(advmss, IPV4_MAX_PMTU - header_size); 1371 } 1372 1373 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst) 1374 { 1375 return ip_dst_mtu_maybe_forward(dst, false); 1376 } 1377 EXPORT_INDIRECT_CALLABLE(ipv4_mtu); 1378 1379 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr) 1380 { 1381 struct fnhe_hash_bucket *hash; 1382 struct fib_nh_exception *fnhe, __rcu **fnhe_p; 1383 u32 hval = fnhe_hashfun(daddr); 1384 1385 spin_lock_bh(&fnhe_lock); 1386 1387 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1388 lockdep_is_held(&fnhe_lock)); 1389 hash += hval; 1390 1391 fnhe_p = &hash->chain; 1392 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock)); 1393 while (fnhe) { 1394 if (fnhe->fnhe_daddr == daddr) { 1395 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected( 1396 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock))); 1397 /* set fnhe_daddr to 0 to ensure it won't bind with 1398 * new dsts in rt_bind_exception(). 1399 */ 1400 fnhe->fnhe_daddr = 0; 1401 fnhe_flush_routes(fnhe); 1402 kfree_rcu(fnhe, rcu); 1403 break; 1404 } 1405 fnhe_p = &fnhe->fnhe_next; 1406 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1407 lockdep_is_held(&fnhe_lock)); 1408 } 1409 1410 spin_unlock_bh(&fnhe_lock); 1411 } 1412 1413 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc, 1414 __be32 daddr) 1415 { 1416 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions); 1417 struct fib_nh_exception *fnhe; 1418 u32 hval; 1419 1420 if (!hash) 1421 return NULL; 1422 1423 hval = fnhe_hashfun(daddr); 1424 1425 for (fnhe = rcu_dereference(hash[hval].chain); fnhe; 1426 fnhe = rcu_dereference(fnhe->fnhe_next)) { 1427 if (fnhe->fnhe_daddr == daddr) { 1428 if (fnhe->fnhe_expires && 1429 time_after(jiffies, fnhe->fnhe_expires)) { 1430 ip_del_fnhe(nhc, daddr); 1431 break; 1432 } 1433 return fnhe; 1434 } 1435 } 1436 return NULL; 1437 } 1438 1439 /* MTU selection: 1440 * 1. mtu on route is locked - use it 1441 * 2. mtu from nexthop exception 1442 * 3. mtu from egress device 1443 */ 1444 1445 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr) 1446 { 1447 struct fib_nh_common *nhc = res->nhc; 1448 struct net_device *dev = nhc->nhc_dev; 1449 struct fib_info *fi = res->fi; 1450 u32 mtu = 0; 1451 1452 if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) || 1453 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU)) 1454 mtu = fi->fib_mtu; 1455 1456 if (likely(!mtu)) { 1457 struct fib_nh_exception *fnhe; 1458 1459 fnhe = find_exception(nhc, daddr); 1460 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires)) 1461 mtu = fnhe->fnhe_pmtu; 1462 } 1463 1464 if (likely(!mtu)) 1465 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU); 1466 1467 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu); 1468 } 1469 1470 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, 1471 __be32 daddr, const bool do_cache) 1472 { 1473 bool ret = false; 1474 1475 spin_lock_bh(&fnhe_lock); 1476 1477 if (daddr == fnhe->fnhe_daddr) { 1478 struct rtable __rcu **porig; 1479 struct rtable *orig; 1480 int genid = fnhe_genid(dev_net(rt->dst.dev)); 1481 1482 if (rt_is_input_route(rt)) 1483 porig = &fnhe->fnhe_rth_input; 1484 else 1485 porig = &fnhe->fnhe_rth_output; 1486 orig = rcu_dereference(*porig); 1487 1488 if (fnhe->fnhe_genid != genid) { 1489 fnhe->fnhe_genid = genid; 1490 fnhe->fnhe_gw = 0; 1491 fnhe->fnhe_pmtu = 0; 1492 fnhe->fnhe_expires = 0; 1493 fnhe->fnhe_mtu_locked = false; 1494 fnhe_flush_routes(fnhe); 1495 orig = NULL; 1496 } 1497 fill_route_from_fnhe(rt, fnhe); 1498 if (!rt->rt_gw4) { 1499 rt->rt_gw4 = daddr; 1500 rt->rt_gw_family = AF_INET; 1501 } 1502 1503 if (do_cache) { 1504 dst_hold(&rt->dst); 1505 rcu_assign_pointer(*porig, rt); 1506 if (orig) { 1507 dst_dev_put(&orig->dst); 1508 dst_release(&orig->dst); 1509 } 1510 ret = true; 1511 } 1512 1513 fnhe->fnhe_stamp = jiffies; 1514 } 1515 spin_unlock_bh(&fnhe_lock); 1516 1517 return ret; 1518 } 1519 1520 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt) 1521 { 1522 struct rtable *orig, *prev, **p; 1523 bool ret = true; 1524 1525 if (rt_is_input_route(rt)) { 1526 p = (struct rtable **)&nhc->nhc_rth_input; 1527 } else { 1528 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 1529 } 1530 orig = *p; 1531 1532 /* hold dst before doing cmpxchg() to avoid race condition 1533 * on this dst 1534 */ 1535 dst_hold(&rt->dst); 1536 prev = cmpxchg(p, orig, rt); 1537 if (prev == orig) { 1538 if (orig) { 1539 rt_add_uncached_list(orig); 1540 dst_release(&orig->dst); 1541 } 1542 } else { 1543 dst_release(&rt->dst); 1544 ret = false; 1545 } 1546 1547 return ret; 1548 } 1549 1550 struct uncached_list { 1551 spinlock_t lock; 1552 struct list_head head; 1553 }; 1554 1555 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list); 1556 1557 void rt_add_uncached_list(struct rtable *rt) 1558 { 1559 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list); 1560 1561 rt->dst.rt_uncached_list = ul; 1562 1563 spin_lock_bh(&ul->lock); 1564 list_add_tail(&rt->dst.rt_uncached, &ul->head); 1565 spin_unlock_bh(&ul->lock); 1566 } 1567 1568 void rt_del_uncached_list(struct rtable *rt) 1569 { 1570 struct uncached_list *ul = rt->dst.rt_uncached_list; 1571 1572 if (ul) { 1573 spin_lock_bh(&ul->lock); 1574 list_del_init(&rt->dst.rt_uncached); 1575 spin_unlock_bh(&ul->lock); 1576 } 1577 } 1578 1579 static void ipv4_dst_destroy(struct dst_entry *dst) 1580 { 1581 ip_dst_metrics_put(dst); 1582 rt_del_uncached_list(dst_rtable(dst)); 1583 } 1584 1585 void rt_flush_dev(struct net_device *dev) 1586 { 1587 struct rtable *rt, *safe; 1588 int cpu; 1589 1590 if (dev && dev->dismantle) 1591 return; 1592 1593 for_each_possible_cpu(cpu) { 1594 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 1595 1596 if (list_empty(&ul->head)) 1597 continue; 1598 1599 spin_lock_bh(&ul->lock); 1600 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) { 1601 struct net_device *rt_dev = rt->dst.dev; 1602 1603 if (dev ? rt_dev != dev : 1604 READ_ONCE(rt_dev->reg_state) != NETREG_UNREGISTERED) 1605 continue; 1606 1607 rcu_assign_pointer(rt->dst.dev_rcu, blackhole_netdev); 1608 netdev_ref_replace(rt_dev, blackhole_netdev, 1609 &rt->dst.dev_tracker, GFP_ATOMIC); 1610 list_del_init(&rt->dst.rt_uncached); 1611 } 1612 spin_unlock_bh(&ul->lock); 1613 } 1614 } 1615 1616 static bool rt_cache_valid(const struct rtable *rt) 1617 { 1618 return rt && 1619 READ_ONCE(rt->dst.obsolete) == DST_OBSOLETE_FORCE_CHK && 1620 !rt_is_expired(rt); 1621 } 1622 1623 static void rt_set_nexthop(struct rtable *rt, __be32 daddr, 1624 const struct fib_result *res, 1625 struct fib_nh_exception *fnhe, 1626 struct fib_info *fi, u16 type, u32 itag, 1627 const bool do_cache) 1628 { 1629 bool cached = false; 1630 1631 if (fi) { 1632 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1633 1634 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) { 1635 rt->rt_uses_gateway = 1; 1636 rt->rt_gw_family = nhc->nhc_gw_family; 1637 /* only INET and INET6 are supported */ 1638 if (likely(nhc->nhc_gw_family == AF_INET)) 1639 rt->rt_gw4 = nhc->nhc_gw.ipv4; 1640 else 1641 rt->rt_gw6 = nhc->nhc_gw.ipv6; 1642 } 1643 1644 ip_dst_init_metrics(&rt->dst, fi->fib_metrics); 1645 1646 #ifdef CONFIG_IP_ROUTE_CLASSID 1647 if (nhc->nhc_family == AF_INET) { 1648 struct fib_nh *nh; 1649 1650 nh = container_of(nhc, struct fib_nh, nh_common); 1651 rt->dst.tclassid = nh->nh_tclassid; 1652 } 1653 #endif 1654 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 1655 if (unlikely(fnhe)) 1656 cached = rt_bind_exception(rt, fnhe, daddr, do_cache); 1657 else if (do_cache) 1658 cached = rt_cache_route(nhc, rt); 1659 if (unlikely(!cached)) { 1660 /* Routes we intend to cache in nexthop exception or 1661 * FIB nexthop have the DST_NOCACHE bit clear. 1662 * However, if we are unsuccessful at storing this 1663 * route into the cache we really need to set it. 1664 */ 1665 if (!rt->rt_gw4) { 1666 rt->rt_gw_family = AF_INET; 1667 rt->rt_gw4 = daddr; 1668 } 1669 rt_add_uncached_list(rt); 1670 } 1671 } else 1672 rt_add_uncached_list(rt); 1673 1674 #ifdef CONFIG_IP_ROUTE_CLASSID 1675 #ifdef CONFIG_IP_MULTIPLE_TABLES 1676 set_class_tag(rt, res->tclassid); 1677 #endif 1678 set_class_tag(rt, itag); 1679 #endif 1680 } 1681 1682 struct rtable *rt_dst_alloc(struct net_device *dev, 1683 unsigned int flags, u16 type, 1684 bool noxfrm) 1685 { 1686 struct rtable *rt; 1687 1688 rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK, 1689 (noxfrm ? DST_NOXFRM : 0)); 1690 1691 if (rt) { 1692 rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1693 rt->rt_flags = flags; 1694 rt->rt_type = type; 1695 rt->rt_is_input = 0; 1696 rt->rt_iif = 0; 1697 rt->rt_pmtu = 0; 1698 rt->rt_mtu_locked = 0; 1699 rt->rt_uses_gateway = 0; 1700 rt->rt_gw_family = 0; 1701 rt->rt_gw4 = 0; 1702 1703 rt->dst.output = ip_output; 1704 if (flags & RTCF_LOCAL) 1705 rt->dst.input = ip_local_deliver; 1706 } 1707 1708 return rt; 1709 } 1710 EXPORT_SYMBOL(rt_dst_alloc); 1711 1712 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt) 1713 { 1714 struct rtable *new_rt; 1715 1716 new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK, 1717 rt->dst.flags); 1718 1719 if (new_rt) { 1720 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1721 new_rt->rt_flags = rt->rt_flags; 1722 new_rt->rt_type = rt->rt_type; 1723 new_rt->rt_is_input = rt->rt_is_input; 1724 new_rt->rt_iif = rt->rt_iif; 1725 new_rt->rt_pmtu = rt->rt_pmtu; 1726 new_rt->rt_mtu_locked = rt->rt_mtu_locked; 1727 new_rt->rt_gw_family = rt->rt_gw_family; 1728 if (rt->rt_gw_family == AF_INET) 1729 new_rt->rt_gw4 = rt->rt_gw4; 1730 else if (rt->rt_gw_family == AF_INET6) 1731 new_rt->rt_gw6 = rt->rt_gw6; 1732 1733 new_rt->dst.input = READ_ONCE(rt->dst.input); 1734 new_rt->dst.output = READ_ONCE(rt->dst.output); 1735 new_rt->dst.error = rt->dst.error; 1736 new_rt->dst.lastuse = jiffies; 1737 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate); 1738 } 1739 return new_rt; 1740 } 1741 1742 /* called in rcu_read_lock() section */ 1743 enum skb_drop_reason 1744 ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1745 dscp_t dscp, struct net_device *dev, 1746 struct in_device *in_dev, u32 *itag) 1747 { 1748 enum skb_drop_reason reason; 1749 1750 /* Primary sanity checks. */ 1751 if (!in_dev) 1752 return SKB_DROP_REASON_NOT_SPECIFIED; 1753 1754 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 1755 return SKB_DROP_REASON_IP_INVALID_SOURCE; 1756 1757 if (skb->protocol != htons(ETH_P_IP)) 1758 return SKB_DROP_REASON_INVALID_PROTO; 1759 1760 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev)) 1761 return SKB_DROP_REASON_IP_LOCALNET; 1762 1763 if (ipv4_is_zeronet(saddr)) { 1764 if (!ipv4_is_local_multicast(daddr) && 1765 ip_hdr(skb)->protocol != IPPROTO_IGMP) 1766 return SKB_DROP_REASON_IP_INVALID_SOURCE; 1767 } else { 1768 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0, 1769 dev, in_dev, itag); 1770 if (reason) 1771 return reason; 1772 } 1773 return SKB_NOT_DROPPED_YET; 1774 } 1775 1776 /* called in rcu_read_lock() section */ 1777 static enum skb_drop_reason 1778 ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1779 dscp_t dscp, struct net_device *dev, int our) 1780 { 1781 struct in_device *in_dev = __in_dev_get_rcu(dev); 1782 unsigned int flags = RTCF_MULTICAST; 1783 enum skb_drop_reason reason; 1784 struct rtable *rth; 1785 u32 itag = 0; 1786 1787 reason = ip_mc_validate_source(skb, daddr, saddr, dscp, dev, in_dev, 1788 &itag); 1789 if (reason) 1790 return reason; 1791 1792 if (our) 1793 flags |= RTCF_LOCAL; 1794 1795 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 1796 IPCB(skb)->flags |= IPSKB_NOPOLICY; 1797 1798 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST, 1799 false); 1800 if (!rth) 1801 return SKB_DROP_REASON_NOMEM; 1802 1803 #ifdef CONFIG_IP_ROUTE_CLASSID 1804 rth->dst.tclassid = itag; 1805 #endif 1806 rth->dst.output = ip_rt_bug; 1807 rth->rt_is_input= 1; 1808 1809 #ifdef CONFIG_IP_MROUTE 1810 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1811 rth->dst.input = ip_mr_input; 1812 #endif 1813 RT_CACHE_STAT_INC(in_slow_mc); 1814 1815 skb_dst_drop(skb); 1816 skb_dst_set(skb, &rth->dst); 1817 return SKB_NOT_DROPPED_YET; 1818 } 1819 1820 1821 static void ip_handle_martian_source(struct net_device *dev, 1822 struct in_device *in_dev, 1823 struct sk_buff *skb, 1824 __be32 daddr, 1825 __be32 saddr) 1826 { 1827 RT_CACHE_STAT_INC(in_martian_src); 1828 #ifdef CONFIG_IP_ROUTE_VERBOSE 1829 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1830 /* 1831 * RFC1812 recommendation, if source is martian, 1832 * the only hint is MAC header. 1833 */ 1834 pr_warn("martian source (src=%pI4, dst=%pI4, dev=%s)\n", 1835 &saddr, &daddr, dev->name); 1836 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1837 print_hex_dump(KERN_WARNING, "ll header: ", 1838 DUMP_PREFIX_OFFSET, 16, 1, 1839 skb_mac_header(skb), 1840 dev->hard_header_len, false); 1841 } 1842 } 1843 #endif 1844 } 1845 1846 /* called in rcu_read_lock() section */ 1847 static enum skb_drop_reason 1848 __mkroute_input(struct sk_buff *skb, const struct fib_result *res, 1849 struct in_device *in_dev, __be32 daddr, 1850 __be32 saddr, dscp_t dscp) 1851 { 1852 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 1853 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1854 struct net_device *dev = nhc->nhc_dev; 1855 struct fib_nh_exception *fnhe; 1856 struct rtable *rth; 1857 int err; 1858 struct in_device *out_dev; 1859 bool do_cache; 1860 u32 itag = 0; 1861 1862 /* get a working reference to the output device */ 1863 out_dev = __in_dev_get_rcu(dev); 1864 if (!out_dev) { 1865 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n"); 1866 return reason; 1867 } 1868 1869 err = fib_validate_source(skb, saddr, daddr, dscp, FIB_RES_OIF(*res), 1870 in_dev->dev, in_dev, &itag); 1871 if (err < 0) { 1872 reason = -err; 1873 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1874 saddr); 1875 1876 goto cleanup; 1877 } 1878 1879 do_cache = res->fi && !itag; 1880 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) && 1881 skb->protocol == htons(ETH_P_IP)) { 1882 __be32 gw; 1883 1884 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0; 1885 if (IN_DEV_SHARED_MEDIA(out_dev) || 1886 inet_addr_onlink(out_dev, saddr, gw)) 1887 IPCB(skb)->flags |= IPSKB_DOREDIRECT; 1888 } 1889 1890 if (skb->protocol != htons(ETH_P_IP)) { 1891 /* Not IP (i.e. ARP). Do not create route, if it is 1892 * invalid for proxy arp. DNAT routes are always valid. 1893 * 1894 * Proxy arp feature have been extended to allow, ARP 1895 * replies back to the same interface, to support 1896 * Private VLAN switch technologies. See arp.c. 1897 */ 1898 if (out_dev == in_dev && 1899 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 1900 reason = SKB_DROP_REASON_ARP_PVLAN_DISABLE; 1901 goto cleanup; 1902 } 1903 } 1904 1905 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 1906 IPCB(skb)->flags |= IPSKB_NOPOLICY; 1907 1908 fnhe = find_exception(nhc, daddr); 1909 if (do_cache) { 1910 if (fnhe) 1911 rth = rcu_dereference(fnhe->fnhe_rth_input); 1912 else 1913 rth = rcu_dereference(nhc->nhc_rth_input); 1914 if (rt_cache_valid(rth)) { 1915 skb_dst_set_noref(skb, &rth->dst); 1916 goto out; 1917 } 1918 } 1919 1920 rth = rt_dst_alloc(out_dev->dev, 0, res->type, 1921 IN_DEV_ORCONF(out_dev, NOXFRM)); 1922 if (!rth) { 1923 reason = SKB_DROP_REASON_NOMEM; 1924 goto cleanup; 1925 } 1926 1927 rth->rt_is_input = 1; 1928 RT_CACHE_STAT_INC(in_slow_tot); 1929 1930 rth->dst.input = ip_forward; 1931 1932 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag, 1933 do_cache); 1934 lwtunnel_set_redirect(&rth->dst); 1935 skb_dst_set(skb, &rth->dst); 1936 out: 1937 reason = SKB_NOT_DROPPED_YET; 1938 cleanup: 1939 return reason; 1940 } 1941 1942 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1943 /* To make ICMP packets follow the right flow, the multipath hash is 1944 * calculated from the inner IP addresses. 1945 */ 1946 static void ip_multipath_l3_keys(const struct sk_buff *skb, 1947 struct flow_keys *hash_keys) 1948 { 1949 const struct iphdr *outer_iph = ip_hdr(skb); 1950 const struct iphdr *key_iph = outer_iph; 1951 const struct iphdr *inner_iph; 1952 const struct icmphdr *icmph; 1953 struct iphdr _inner_iph; 1954 struct icmphdr _icmph; 1955 1956 if (likely(outer_iph->protocol != IPPROTO_ICMP)) 1957 goto out; 1958 1959 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0)) 1960 goto out; 1961 1962 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph), 1963 &_icmph); 1964 if (!icmph) 1965 goto out; 1966 1967 if (!icmp_is_err(icmph->type)) 1968 goto out; 1969 1970 inner_iph = skb_header_pointer(skb, 1971 outer_iph->ihl * 4 + sizeof(_icmph), 1972 sizeof(_inner_iph), &_inner_iph); 1973 if (!inner_iph) 1974 goto out; 1975 1976 key_iph = inner_iph; 1977 out: 1978 hash_keys->addrs.v4addrs.src = key_iph->saddr; 1979 hash_keys->addrs.v4addrs.dst = key_iph->daddr; 1980 } 1981 1982 static u32 fib_multipath_custom_hash_outer(const struct net *net, 1983 const struct sk_buff *skb, 1984 bool *p_has_inner) 1985 { 1986 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 1987 struct flow_keys keys, hash_keys; 1988 1989 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 1990 return 0; 1991 1992 memset(&hash_keys, 0, sizeof(hash_keys)); 1993 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP); 1994 1995 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1996 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 1997 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 1998 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 1999 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 2000 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 2001 hash_keys.basic.ip_proto = keys.basic.ip_proto; 2002 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) 2003 hash_keys.ports.src = keys.ports.src; 2004 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 2005 hash_keys.ports.dst = keys.ports.dst; 2006 2007 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION); 2008 return fib_multipath_hash_from_keys(net, &hash_keys); 2009 } 2010 2011 static u32 fib_multipath_custom_hash_inner(const struct net *net, 2012 const struct sk_buff *skb, 2013 bool has_inner) 2014 { 2015 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 2016 struct flow_keys keys, hash_keys; 2017 2018 /* We assume the packet carries an encapsulation, but if none was 2019 * encountered during dissection of the outer flow, then there is no 2020 * point in calling the flow dissector again. 2021 */ 2022 if (!has_inner) 2023 return 0; 2024 2025 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK)) 2026 return 0; 2027 2028 memset(&hash_keys, 0, sizeof(hash_keys)); 2029 skb_flow_dissect_flow_keys(skb, &keys, 0); 2030 2031 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION)) 2032 return 0; 2033 2034 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 2035 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2036 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 2037 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 2038 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 2039 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 2040 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 2041 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 2042 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 2043 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 2044 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 2045 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 2046 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL) 2047 hash_keys.tags.flow_label = keys.tags.flow_label; 2048 } 2049 2050 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO) 2051 hash_keys.basic.ip_proto = keys.basic.ip_proto; 2052 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT) 2053 hash_keys.ports.src = keys.ports.src; 2054 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT) 2055 hash_keys.ports.dst = keys.ports.dst; 2056 2057 return fib_multipath_hash_from_keys(net, &hash_keys); 2058 } 2059 2060 static u32 fib_multipath_custom_hash_skb(const struct net *net, 2061 const struct sk_buff *skb) 2062 { 2063 u32 mhash, mhash_inner; 2064 bool has_inner = true; 2065 2066 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner); 2067 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner); 2068 2069 return jhash_2words(mhash, mhash_inner, 0); 2070 } 2071 2072 static u32 fib_multipath_custom_hash_fl4(const struct net *net, 2073 const struct flowi4 *fl4) 2074 { 2075 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 2076 struct flow_keys hash_keys; 2077 2078 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 2079 return 0; 2080 2081 memset(&hash_keys, 0, sizeof(hash_keys)); 2082 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2083 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 2084 hash_keys.addrs.v4addrs.src = fl4->saddr; 2085 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 2086 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2087 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 2088 hash_keys.basic.ip_proto = fl4->flowi4_proto; 2089 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) { 2090 if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT) 2091 hash_keys.ports.src = (__force __be16)get_random_u16(); 2092 else 2093 hash_keys.ports.src = fl4->fl4_sport; 2094 } 2095 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 2096 hash_keys.ports.dst = fl4->fl4_dport; 2097 2098 return fib_multipath_hash_from_keys(net, &hash_keys); 2099 } 2100 2101 /* if skb is set it will be used and fl4 can be NULL */ 2102 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4, 2103 const struct sk_buff *skb, struct flow_keys *flkeys) 2104 { 2105 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0; 2106 struct flow_keys hash_keys; 2107 u32 mhash = 0; 2108 2109 switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) { 2110 case 0: 2111 memset(&hash_keys, 0, sizeof(hash_keys)); 2112 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2113 if (skb) { 2114 ip_multipath_l3_keys(skb, &hash_keys); 2115 } else { 2116 hash_keys.addrs.v4addrs.src = fl4->saddr; 2117 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2118 } 2119 mhash = fib_multipath_hash_from_keys(net, &hash_keys); 2120 break; 2121 case 1: 2122 /* skb is currently provided only when forwarding */ 2123 if (skb) { 2124 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 2125 struct flow_keys keys; 2126 2127 /* short-circuit if we already have L4 hash present */ 2128 if (skb->l4_hash) 2129 return skb_get_hash_raw(skb) >> 1; 2130 2131 memset(&hash_keys, 0, sizeof(hash_keys)); 2132 2133 if (!flkeys) { 2134 skb_flow_dissect_flow_keys(skb, &keys, flag); 2135 flkeys = &keys; 2136 } 2137 2138 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2139 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src; 2140 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst; 2141 hash_keys.ports.src = flkeys->ports.src; 2142 hash_keys.ports.dst = flkeys->ports.dst; 2143 hash_keys.basic.ip_proto = flkeys->basic.ip_proto; 2144 } else { 2145 memset(&hash_keys, 0, sizeof(hash_keys)); 2146 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2147 hash_keys.addrs.v4addrs.src = fl4->saddr; 2148 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2149 if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT) 2150 hash_keys.ports.src = (__force __be16)get_random_u16(); 2151 else 2152 hash_keys.ports.src = fl4->fl4_sport; 2153 hash_keys.ports.dst = fl4->fl4_dport; 2154 hash_keys.basic.ip_proto = fl4->flowi4_proto; 2155 } 2156 mhash = fib_multipath_hash_from_keys(net, &hash_keys); 2157 break; 2158 case 2: 2159 memset(&hash_keys, 0, sizeof(hash_keys)); 2160 /* skb is currently provided only when forwarding */ 2161 if (skb) { 2162 struct flow_keys keys; 2163 2164 skb_flow_dissect_flow_keys(skb, &keys, 0); 2165 /* Inner can be v4 or v6 */ 2166 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 2167 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2168 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 2169 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 2170 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 2171 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 2172 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 2173 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 2174 hash_keys.tags.flow_label = keys.tags.flow_label; 2175 hash_keys.basic.ip_proto = keys.basic.ip_proto; 2176 } else { 2177 /* Same as case 0 */ 2178 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2179 ip_multipath_l3_keys(skb, &hash_keys); 2180 } 2181 } else { 2182 /* Same as case 0 */ 2183 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2184 hash_keys.addrs.v4addrs.src = fl4->saddr; 2185 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2186 } 2187 mhash = fib_multipath_hash_from_keys(net, &hash_keys); 2188 break; 2189 case 3: 2190 if (skb) 2191 mhash = fib_multipath_custom_hash_skb(net, skb); 2192 else 2193 mhash = fib_multipath_custom_hash_fl4(net, fl4); 2194 break; 2195 } 2196 2197 if (multipath_hash) 2198 mhash = jhash_2words(mhash, multipath_hash, 0); 2199 2200 return mhash >> 1; 2201 } 2202 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 2203 2204 static enum skb_drop_reason 2205 ip_mkroute_input(struct sk_buff *skb, struct fib_result *res, 2206 struct in_device *in_dev, __be32 daddr, 2207 __be32 saddr, dscp_t dscp, struct flow_keys *hkeys) 2208 { 2209 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2210 if (res->fi && fib_info_num_path(res->fi) > 1) { 2211 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys); 2212 2213 fib_select_multipath(res, h, NULL); 2214 IPCB(skb)->flags |= IPSKB_MULTIPATH; 2215 } 2216 #endif 2217 2218 /* create a routing cache entry */ 2219 return __mkroute_input(skb, res, in_dev, daddr, saddr, dscp); 2220 } 2221 2222 /* Implements all the saddr-related checks as ip_route_input_slow(), 2223 * assuming daddr is valid and the destination is not a local broadcast one. 2224 * Uses the provided hint instead of performing a route lookup. 2225 */ 2226 enum skb_drop_reason 2227 ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2228 dscp_t dscp, struct net_device *dev, 2229 const struct sk_buff *hint) 2230 { 2231 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 2232 struct in_device *in_dev = __in_dev_get_rcu(dev); 2233 struct rtable *rt = skb_rtable(hint); 2234 struct net *net = dev_net(dev); 2235 u32 tag = 0; 2236 2237 if (!in_dev) 2238 return reason; 2239 2240 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) { 2241 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2242 goto martian_source; 2243 } 2244 2245 if (ipv4_is_zeronet(saddr)) { 2246 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2247 goto martian_source; 2248 } 2249 2250 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) { 2251 reason = SKB_DROP_REASON_IP_LOCALNET; 2252 goto martian_source; 2253 } 2254 2255 if (!(rt->rt_flags & RTCF_LOCAL)) 2256 goto skip_validate_source; 2257 2258 reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 0, dev, 2259 in_dev, &tag); 2260 if (reason) 2261 goto martian_source; 2262 2263 skip_validate_source: 2264 skb_dst_copy(skb, hint); 2265 return SKB_NOT_DROPPED_YET; 2266 2267 martian_source: 2268 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2269 return reason; 2270 } 2271 2272 /* get device for dst_alloc with local routes */ 2273 static struct net_device *ip_rt_get_dev(struct net *net, 2274 const struct fib_result *res) 2275 { 2276 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL; 2277 struct net_device *dev = NULL; 2278 2279 if (nhc) 2280 dev = l3mdev_master_dev_rcu(nhc->nhc_dev); 2281 2282 return dev ? : net->loopback_dev; 2283 } 2284 2285 /* 2286 * NOTE. We drop all the packets that has local source 2287 * addresses, because every properly looped back packet 2288 * must have correct destination already attached by output routine. 2289 * Changes in the enforced policies must be applied also to 2290 * ip_route_use_hint(). 2291 * 2292 * Such approach solves two big problems: 2293 * 1. Not simplex devices are handled properly. 2294 * 2. IP spoofing attempts are filtered with 100% of guarantee. 2295 * called with rcu_read_lock() 2296 */ 2297 2298 static enum skb_drop_reason 2299 ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2300 dscp_t dscp, struct net_device *dev, 2301 struct fib_result *res) 2302 { 2303 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 2304 struct in_device *in_dev = __in_dev_get_rcu(dev); 2305 struct flow_keys *flkeys = NULL, _flkeys; 2306 struct net *net = dev_net(dev); 2307 struct ip_tunnel_info *tun_info; 2308 int err = -EINVAL; 2309 unsigned int flags = 0; 2310 u32 itag = 0; 2311 struct rtable *rth; 2312 struct flowi4 fl4; 2313 bool do_cache = true; 2314 2315 /* IP on this device is disabled. */ 2316 2317 if (!in_dev) 2318 goto out; 2319 2320 /* Check for the most weird martians, which can be not detected 2321 * by fib_lookup. 2322 */ 2323 2324 tun_info = skb_tunnel_info(skb); 2325 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX)) 2326 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id; 2327 else 2328 fl4.flowi4_tun_key.tun_id = 0; 2329 skb_dst_drop(skb); 2330 2331 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) { 2332 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2333 goto martian_source; 2334 } 2335 2336 res->fi = NULL; 2337 res->table = NULL; 2338 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 2339 goto brd_input; 2340 2341 /* Accept zero addresses only to limited broadcast; 2342 * I even do not know to fix it or not. Waiting for complains :-) 2343 */ 2344 if (ipv4_is_zeronet(saddr)) { 2345 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2346 goto martian_source; 2347 } 2348 2349 if (ipv4_is_zeronet(daddr)) { 2350 reason = SKB_DROP_REASON_IP_INVALID_DEST; 2351 goto martian_destination; 2352 } 2353 2354 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(), 2355 * and call it once if daddr or/and saddr are loopback addresses 2356 */ 2357 if (ipv4_is_loopback(daddr)) { 2358 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) { 2359 reason = SKB_DROP_REASON_IP_LOCALNET; 2360 goto martian_destination; 2361 } 2362 } else if (ipv4_is_loopback(saddr)) { 2363 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) { 2364 reason = SKB_DROP_REASON_IP_LOCALNET; 2365 goto martian_source; 2366 } 2367 } 2368 2369 /* 2370 * Now we are ready to route packet. 2371 */ 2372 fl4.flowi4_l3mdev = 0; 2373 fl4.flowi4_oif = 0; 2374 fl4.flowi4_iif = dev->ifindex; 2375 fl4.flowi4_mark = skb->mark; 2376 fl4.flowi4_dscp = dscp; 2377 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 2378 fl4.flowi4_flags = 0; 2379 fl4.daddr = daddr; 2380 fl4.saddr = saddr; 2381 fl4.flowi4_uid = sock_net_uid(net, NULL); 2382 fl4.flowi4_multipath_hash = 0; 2383 2384 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) { 2385 flkeys = &_flkeys; 2386 } else { 2387 fl4.flowi4_proto = 0; 2388 fl4.fl4_sport = 0; 2389 fl4.fl4_dport = 0; 2390 } 2391 2392 err = fib_lookup(net, &fl4, res, 0); 2393 if (err != 0) { 2394 if (!IN_DEV_FORWARD(in_dev)) 2395 err = -EHOSTUNREACH; 2396 goto no_route; 2397 } 2398 2399 if (res->type == RTN_BROADCAST) { 2400 if (IN_DEV_BFORWARD(in_dev)) 2401 goto make_route; 2402 /* not do cache if bc_forwarding is enabled */ 2403 if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING)) 2404 do_cache = false; 2405 goto brd_input; 2406 } 2407 2408 err = -EINVAL; 2409 if (res->type == RTN_LOCAL) { 2410 reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 2411 0, dev, in_dev, &itag); 2412 if (reason) 2413 goto martian_source; 2414 goto local_input; 2415 } 2416 2417 if (!IN_DEV_FORWARD(in_dev)) { 2418 err = -EHOSTUNREACH; 2419 goto no_route; 2420 } 2421 if (res->type != RTN_UNICAST) { 2422 reason = SKB_DROP_REASON_IP_INVALID_DEST; 2423 goto martian_destination; 2424 } 2425 2426 make_route: 2427 reason = ip_mkroute_input(skb, res, in_dev, daddr, saddr, dscp, 2428 flkeys); 2429 2430 out: 2431 return reason; 2432 2433 brd_input: 2434 if (skb->protocol != htons(ETH_P_IP)) { 2435 reason = SKB_DROP_REASON_INVALID_PROTO; 2436 goto out; 2437 } 2438 2439 if (!ipv4_is_zeronet(saddr)) { 2440 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0, 2441 dev, in_dev, &itag); 2442 if (reason) 2443 goto martian_source; 2444 } 2445 flags |= RTCF_BROADCAST; 2446 res->type = RTN_BROADCAST; 2447 RT_CACHE_STAT_INC(in_brd); 2448 2449 local_input: 2450 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 2451 IPCB(skb)->flags |= IPSKB_NOPOLICY; 2452 2453 do_cache &= res->fi && !itag; 2454 if (do_cache) { 2455 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2456 2457 rth = rcu_dereference(nhc->nhc_rth_input); 2458 if (rt_cache_valid(rth)) { 2459 skb_dst_set_noref(skb, &rth->dst); 2460 reason = SKB_NOT_DROPPED_YET; 2461 goto out; 2462 } 2463 } 2464 2465 rth = rt_dst_alloc(ip_rt_get_dev(net, res), 2466 flags | RTCF_LOCAL, res->type, false); 2467 if (!rth) 2468 goto e_nobufs; 2469 2470 rth->dst.output= ip_rt_bug; 2471 #ifdef CONFIG_IP_ROUTE_CLASSID 2472 rth->dst.tclassid = itag; 2473 #endif 2474 rth->rt_is_input = 1; 2475 2476 RT_CACHE_STAT_INC(in_slow_tot); 2477 if (res->type == RTN_UNREACHABLE) { 2478 rth->dst.input= ip_error; 2479 rth->dst.error= -err; 2480 rth->rt_flags &= ~RTCF_LOCAL; 2481 } 2482 2483 if (do_cache) { 2484 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2485 2486 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 2487 if (lwtunnel_input_redirect(rth->dst.lwtstate)) { 2488 WARN_ON(rth->dst.input == lwtunnel_input); 2489 rth->dst.lwtstate->orig_input = rth->dst.input; 2490 rth->dst.input = lwtunnel_input; 2491 } 2492 2493 if (unlikely(!rt_cache_route(nhc, rth))) 2494 rt_add_uncached_list(rth); 2495 } 2496 skb_dst_set(skb, &rth->dst); 2497 reason = SKB_NOT_DROPPED_YET; 2498 goto out; 2499 2500 no_route: 2501 RT_CACHE_STAT_INC(in_no_route); 2502 res->type = RTN_UNREACHABLE; 2503 res->fi = NULL; 2504 res->table = NULL; 2505 goto local_input; 2506 2507 /* 2508 * Do not cache martian addresses: they should be logged (RFC1812) 2509 */ 2510 martian_destination: 2511 RT_CACHE_STAT_INC(in_martian_dst); 2512 #ifdef CONFIG_IP_ROUTE_VERBOSE 2513 if (IN_DEV_LOG_MARTIANS(in_dev)) 2514 net_warn_ratelimited("martian destination (src=%pI4, dst=%pI4, dev=%s)\n", 2515 &saddr, &daddr, dev->name); 2516 #endif 2517 goto out; 2518 2519 e_nobufs: 2520 reason = SKB_DROP_REASON_NOMEM; 2521 goto out; 2522 2523 martian_source: 2524 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2525 goto out; 2526 } 2527 2528 /* called with rcu_read_lock held */ 2529 static enum skb_drop_reason 2530 ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2531 dscp_t dscp, struct net_device *dev, 2532 struct fib_result *res) 2533 { 2534 /* Multicast recognition logic is moved from route cache to here. 2535 * The problem was that too many Ethernet cards have broken/missing 2536 * hardware multicast filters :-( As result the host on multicasting 2537 * network acquires a lot of useless route cache entries, sort of 2538 * SDR messages from all the world. Now we try to get rid of them. 2539 * Really, provided software IP multicast filter is organized 2540 * reasonably (at least, hashed), it does not result in a slowdown 2541 * comparing with route cache reject entries. 2542 * Note, that multicast routers are not affected, because 2543 * route cache entry is created eventually. 2544 */ 2545 if (ipv4_is_multicast(daddr)) { 2546 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 2547 struct in_device *in_dev = __in_dev_get_rcu(dev); 2548 int our = 0; 2549 2550 if (!in_dev) 2551 return reason; 2552 2553 our = ip_check_mc_rcu(in_dev, daddr, saddr, 2554 ip_hdr(skb)->protocol); 2555 2556 /* check l3 master if no match yet */ 2557 if (!our && netif_is_l3_slave(dev)) { 2558 struct in_device *l3_in_dev; 2559 2560 l3_in_dev = __in_dev_get_rcu(skb->dev); 2561 if (l3_in_dev) 2562 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr, 2563 ip_hdr(skb)->protocol); 2564 } 2565 2566 if (our 2567 #ifdef CONFIG_IP_MROUTE 2568 || 2569 (!ipv4_is_local_multicast(daddr) && 2570 IN_DEV_MFORWARD(in_dev)) 2571 #endif 2572 ) { 2573 reason = ip_route_input_mc(skb, daddr, saddr, dscp, 2574 dev, our); 2575 } 2576 return reason; 2577 } 2578 2579 return ip_route_input_slow(skb, daddr, saddr, dscp, dev, res); 2580 } 2581 2582 enum skb_drop_reason ip_route_input_noref(struct sk_buff *skb, __be32 daddr, 2583 __be32 saddr, dscp_t dscp, 2584 struct net_device *dev) 2585 { 2586 enum skb_drop_reason reason; 2587 struct fib_result res; 2588 2589 rcu_read_lock(); 2590 reason = ip_route_input_rcu(skb, daddr, saddr, dscp, dev, &res); 2591 rcu_read_unlock(); 2592 2593 return reason; 2594 } 2595 EXPORT_SYMBOL(ip_route_input_noref); 2596 2597 /* called with rcu_read_lock() */ 2598 static struct rtable *__mkroute_output(const struct fib_result *res, 2599 const struct flowi4 *fl4, int orig_oif, 2600 struct net_device *dev_out, 2601 unsigned int flags) 2602 { 2603 struct fib_info *fi = res->fi; 2604 struct fib_nh_exception *fnhe; 2605 struct in_device *in_dev; 2606 u16 type = res->type; 2607 struct rtable *rth; 2608 bool do_cache; 2609 2610 in_dev = __in_dev_get_rcu(dev_out); 2611 if (!in_dev) 2612 return ERR_PTR(-EINVAL); 2613 2614 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 2615 if (ipv4_is_loopback(fl4->saddr) && 2616 !(dev_out->flags & IFF_LOOPBACK) && 2617 !netif_is_l3_master(dev_out)) 2618 return ERR_PTR(-EINVAL); 2619 2620 if (ipv4_is_lbcast(fl4->daddr)) { 2621 type = RTN_BROADCAST; 2622 2623 /* reset fi to prevent gateway resolution */ 2624 fi = NULL; 2625 } else if (ipv4_is_multicast(fl4->daddr)) { 2626 type = RTN_MULTICAST; 2627 } else if (ipv4_is_zeronet(fl4->daddr)) { 2628 return ERR_PTR(-EINVAL); 2629 } 2630 2631 if (dev_out->flags & IFF_LOOPBACK) 2632 flags |= RTCF_LOCAL; 2633 2634 do_cache = true; 2635 if (type == RTN_BROADCAST) { 2636 flags |= RTCF_BROADCAST | RTCF_LOCAL; 2637 } else if (type == RTN_MULTICAST) { 2638 flags |= RTCF_MULTICAST | RTCF_LOCAL; 2639 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr, 2640 fl4->flowi4_proto)) 2641 flags &= ~RTCF_LOCAL; 2642 else 2643 do_cache = false; 2644 /* If multicast route do not exist use 2645 * default one, but do not gateway in this case. 2646 * Yes, it is hack. 2647 */ 2648 if (fi && res->prefixlen < 4) 2649 fi = NULL; 2650 } else if ((type == RTN_LOCAL) && (orig_oif != 0) && 2651 (orig_oif != dev_out->ifindex)) { 2652 /* For local routes that require a particular output interface 2653 * we do not want to cache the result. Caching the result 2654 * causes incorrect behaviour when there are multiple source 2655 * addresses on the interface, the end result being that if the 2656 * intended recipient is waiting on that interface for the 2657 * packet he won't receive it because it will be delivered on 2658 * the loopback interface and the IP_PKTINFO ipi_ifindex will 2659 * be set to the loopback interface as well. 2660 */ 2661 do_cache = false; 2662 } 2663 2664 fnhe = NULL; 2665 do_cache &= fi != NULL; 2666 if (fi) { 2667 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2668 struct rtable __rcu **prth; 2669 2670 fnhe = find_exception(nhc, fl4->daddr); 2671 if (!do_cache) 2672 goto add; 2673 if (fnhe) { 2674 prth = &fnhe->fnhe_rth_output; 2675 } else { 2676 if (unlikely(fl4->flowi4_flags & 2677 FLOWI_FLAG_KNOWN_NH && 2678 !(nhc->nhc_gw_family && 2679 nhc->nhc_scope == RT_SCOPE_LINK))) { 2680 do_cache = false; 2681 goto add; 2682 } 2683 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 2684 } 2685 rth = rcu_dereference(*prth); 2686 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst)) 2687 return rth; 2688 } 2689 2690 add: 2691 rth = rt_dst_alloc(dev_out, flags, type, 2692 IN_DEV_ORCONF(in_dev, NOXFRM)); 2693 if (!rth) 2694 return ERR_PTR(-ENOBUFS); 2695 2696 rth->rt_iif = orig_oif; 2697 2698 RT_CACHE_STAT_INC(out_slow_tot); 2699 2700 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 2701 if (flags & RTCF_LOCAL && 2702 !(dev_out->flags & IFF_LOOPBACK)) { 2703 rth->dst.output = ip_mc_output; 2704 RT_CACHE_STAT_INC(out_slow_mc); 2705 } 2706 #ifdef CONFIG_IP_MROUTE 2707 if (type == RTN_MULTICAST) { 2708 if (IN_DEV_MFORWARD(in_dev) && 2709 !ipv4_is_local_multicast(fl4->daddr)) { 2710 rth->dst.input = ip_mr_input; 2711 rth->dst.output = ip_mr_output; 2712 } 2713 } 2714 #endif 2715 } 2716 2717 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache); 2718 lwtunnel_set_redirect(&rth->dst); 2719 2720 return rth; 2721 } 2722 2723 /* 2724 * Major route resolver routine. 2725 */ 2726 2727 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4, 2728 const struct sk_buff *skb) 2729 { 2730 struct fib_result res = { 2731 .type = RTN_UNSPEC, 2732 .fi = NULL, 2733 .table = NULL, 2734 .tclassid = 0, 2735 }; 2736 struct rtable *rth; 2737 2738 fl4->flowi4_iif = LOOPBACK_IFINDEX; 2739 2740 rcu_read_lock(); 2741 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb); 2742 rcu_read_unlock(); 2743 2744 return rth; 2745 } 2746 EXPORT_SYMBOL_GPL(ip_route_output_key_hash); 2747 2748 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4, 2749 struct fib_result *res, 2750 const struct sk_buff *skb) 2751 { 2752 struct net_device *dev_out = NULL; 2753 int orig_oif = fl4->flowi4_oif; 2754 unsigned int flags = 0; 2755 struct rtable *rth; 2756 int err; 2757 2758 if (fl4->saddr) { 2759 if (ipv4_is_multicast(fl4->saddr) || 2760 ipv4_is_lbcast(fl4->saddr)) { 2761 rth = ERR_PTR(-EINVAL); 2762 goto out; 2763 } 2764 2765 rth = ERR_PTR(-ENETUNREACH); 2766 2767 /* I removed check for oif == dev_out->oif here. 2768 * It was wrong for two reasons: 2769 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 2770 * is assigned to multiple interfaces. 2771 * 2. Moreover, we are allowed to send packets with saddr 2772 * of another iface. --ANK 2773 */ 2774 2775 if (fl4->flowi4_oif == 0 && 2776 (ipv4_is_multicast(fl4->daddr) || 2777 ipv4_is_lbcast(fl4->daddr))) { 2778 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2779 dev_out = __ip_dev_find(net, fl4->saddr, false); 2780 if (!dev_out) 2781 goto out; 2782 2783 /* Special hack: user can direct multicasts 2784 * and limited broadcast via necessary interface 2785 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 2786 * This hack is not just for fun, it allows 2787 * vic,vat and friends to work. 2788 * They bind socket to loopback, set ttl to zero 2789 * and expect that it will work. 2790 * From the viewpoint of routing cache they are broken, 2791 * because we are not allowed to build multicast path 2792 * with loopback source addr (look, routing cache 2793 * cannot know, that ttl is zero, so that packet 2794 * will not leave this host and route is valid). 2795 * Luckily, this hack is good workaround. 2796 */ 2797 2798 fl4->flowi4_oif = dev_out->ifindex; 2799 goto make_route; 2800 } 2801 2802 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 2803 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2804 if (!__ip_dev_find(net, fl4->saddr, false)) 2805 goto out; 2806 } 2807 } 2808 2809 2810 if (fl4->flowi4_oif) { 2811 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif); 2812 rth = ERR_PTR(-ENODEV); 2813 if (!dev_out) 2814 goto out; 2815 2816 /* RACE: Check return value of inet_select_addr instead. */ 2817 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 2818 rth = ERR_PTR(-ENETUNREACH); 2819 goto out; 2820 } 2821 if (ipv4_is_local_multicast(fl4->daddr) || 2822 ipv4_is_lbcast(fl4->daddr) || 2823 fl4->flowi4_proto == IPPROTO_IGMP) { 2824 if (!fl4->saddr) 2825 fl4->saddr = inet_select_addr(dev_out, 0, 2826 RT_SCOPE_LINK); 2827 goto make_route; 2828 } 2829 if (!fl4->saddr) { 2830 if (ipv4_is_multicast(fl4->daddr)) 2831 fl4->saddr = inet_select_addr(dev_out, 0, 2832 fl4->flowi4_scope); 2833 else if (!fl4->daddr) 2834 fl4->saddr = inet_select_addr(dev_out, 0, 2835 RT_SCOPE_HOST); 2836 } 2837 } 2838 2839 if (!fl4->daddr) { 2840 fl4->daddr = fl4->saddr; 2841 if (!fl4->daddr) 2842 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK); 2843 dev_out = net->loopback_dev; 2844 fl4->flowi4_oif = LOOPBACK_IFINDEX; 2845 res->type = RTN_LOCAL; 2846 flags |= RTCF_LOCAL; 2847 goto make_route; 2848 } 2849 2850 err = fib_lookup(net, fl4, res, 0); 2851 if (err) { 2852 res->fi = NULL; 2853 res->table = NULL; 2854 if (fl4->flowi4_oif && 2855 (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) { 2856 /* Apparently, routing tables are wrong. Assume, 2857 * that the destination is on link. 2858 * 2859 * WHY? DW. 2860 * Because we are allowed to send to iface 2861 * even if it has NO routes and NO assigned 2862 * addresses. When oif is specified, routing 2863 * tables are looked up with only one purpose: 2864 * to catch if destination is gatewayed, rather than 2865 * direct. Moreover, if MSG_DONTROUTE is set, 2866 * we send packet, ignoring both routing tables 2867 * and ifaddr state. --ANK 2868 * 2869 * 2870 * We could make it even if oif is unknown, 2871 * likely IPv6, but we do not. 2872 */ 2873 2874 if (fl4->saddr == 0) 2875 fl4->saddr = inet_select_addr(dev_out, 0, 2876 RT_SCOPE_LINK); 2877 res->type = RTN_UNICAST; 2878 goto make_route; 2879 } 2880 rth = ERR_PTR(err); 2881 goto out; 2882 } 2883 2884 if (res->type == RTN_LOCAL) { 2885 if (!fl4->saddr) { 2886 if (res->fi->fib_prefsrc) 2887 fl4->saddr = res->fi->fib_prefsrc; 2888 else 2889 fl4->saddr = fl4->daddr; 2890 } 2891 2892 /* L3 master device is the loopback for that domain */ 2893 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? : 2894 net->loopback_dev; 2895 2896 /* make sure orig_oif points to fib result device even 2897 * though packet rx/tx happens over loopback or l3mdev 2898 */ 2899 orig_oif = FIB_RES_OIF(*res); 2900 2901 fl4->flowi4_oif = dev_out->ifindex; 2902 flags |= RTCF_LOCAL; 2903 goto make_route; 2904 } 2905 2906 fib_select_path(net, res, fl4, skb); 2907 2908 dev_out = FIB_RES_DEV(*res); 2909 2910 make_route: 2911 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags); 2912 2913 out: 2914 return rth; 2915 } 2916 2917 static struct dst_ops ipv4_dst_blackhole_ops = { 2918 .family = AF_INET, 2919 .default_advmss = ipv4_default_advmss, 2920 .neigh_lookup = ipv4_dst_neigh_lookup, 2921 .check = dst_blackhole_check, 2922 .cow_metrics = dst_blackhole_cow_metrics, 2923 .update_pmtu = dst_blackhole_update_pmtu, 2924 .redirect = dst_blackhole_redirect, 2925 .mtu = dst_blackhole_mtu, 2926 }; 2927 2928 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2929 { 2930 struct rtable *ort = dst_rtable(dst_orig); 2931 struct rtable *rt; 2932 2933 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0); 2934 if (rt) { 2935 struct dst_entry *new = &rt->dst; 2936 2937 new->__use = 1; 2938 new->input = dst_discard; 2939 new->output = dst_discard_out; 2940 2941 new->dev = net->loopback_dev; 2942 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC); 2943 2944 rt->rt_is_input = ort->rt_is_input; 2945 rt->rt_iif = ort->rt_iif; 2946 rt->rt_pmtu = ort->rt_pmtu; 2947 rt->rt_mtu_locked = ort->rt_mtu_locked; 2948 2949 rt->rt_genid = rt_genid_ipv4(net); 2950 rt->rt_flags = ort->rt_flags; 2951 rt->rt_type = ort->rt_type; 2952 rt->rt_uses_gateway = ort->rt_uses_gateway; 2953 rt->rt_gw_family = ort->rt_gw_family; 2954 if (rt->rt_gw_family == AF_INET) 2955 rt->rt_gw4 = ort->rt_gw4; 2956 else if (rt->rt_gw_family == AF_INET6) 2957 rt->rt_gw6 = ort->rt_gw6; 2958 } 2959 2960 dst_release(dst_orig); 2961 2962 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2963 } 2964 2965 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2966 const struct sock *sk) 2967 { 2968 struct rtable *rt = __ip_route_output_key(net, flp4); 2969 2970 if (IS_ERR(rt)) 2971 return rt; 2972 2973 if (flp4->flowi4_proto) { 2974 flp4->flowi4_oif = rt->dst.dev->ifindex; 2975 rt = dst_rtable(xfrm_lookup_route(net, &rt->dst, 2976 flowi4_to_flowi(flp4), 2977 sk, 0)); 2978 } 2979 2980 return rt; 2981 } 2982 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2983 2984 /* called with rcu_read_lock held */ 2985 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, 2986 struct rtable *rt, u32 table_id, dscp_t dscp, 2987 struct flowi4 *fl4, struct sk_buff *skb, u32 portid, 2988 u32 seq, unsigned int flags) 2989 { 2990 struct rtmsg *r; 2991 struct nlmsghdr *nlh; 2992 unsigned long expires = 0; 2993 u32 error; 2994 u32 metrics[RTAX_MAX]; 2995 2996 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags); 2997 if (!nlh) 2998 return -EMSGSIZE; 2999 3000 r = nlmsg_data(nlh); 3001 r->rtm_family = AF_INET; 3002 r->rtm_dst_len = 32; 3003 r->rtm_src_len = 0; 3004 r->rtm_tos = inet_dscp_to_dsfield(dscp); 3005 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT; 3006 if (nla_put_u32(skb, RTA_TABLE, table_id)) 3007 goto nla_put_failure; 3008 r->rtm_type = rt->rt_type; 3009 r->rtm_scope = RT_SCOPE_UNIVERSE; 3010 r->rtm_protocol = RTPROT_UNSPEC; 3011 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 3012 if (rt->rt_flags & RTCF_NOTIFY) 3013 r->rtm_flags |= RTM_F_NOTIFY; 3014 if (IPCB(skb)->flags & IPSKB_DOREDIRECT) 3015 r->rtm_flags |= RTCF_DOREDIRECT; 3016 3017 if (nla_put_in_addr(skb, RTA_DST, dst)) 3018 goto nla_put_failure; 3019 if (src) { 3020 r->rtm_src_len = 32; 3021 if (nla_put_in_addr(skb, RTA_SRC, src)) 3022 goto nla_put_failure; 3023 } 3024 if (rt->dst.dev && 3025 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex)) 3026 goto nla_put_failure; 3027 if (lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0) 3028 goto nla_put_failure; 3029 #ifdef CONFIG_IP_ROUTE_CLASSID 3030 if (rt->dst.tclassid && 3031 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid)) 3032 goto nla_put_failure; 3033 #endif 3034 if (fl4 && !rt_is_input_route(rt) && 3035 fl4->saddr != src) { 3036 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr)) 3037 goto nla_put_failure; 3038 } 3039 if (rt->rt_uses_gateway) { 3040 if (rt->rt_gw_family == AF_INET && 3041 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) { 3042 goto nla_put_failure; 3043 } else if (rt->rt_gw_family == AF_INET6) { 3044 int alen = sizeof(struct in6_addr); 3045 struct nlattr *nla; 3046 struct rtvia *via; 3047 3048 nla = nla_reserve(skb, RTA_VIA, alen + 2); 3049 if (!nla) 3050 goto nla_put_failure; 3051 3052 via = nla_data(nla); 3053 via->rtvia_family = AF_INET6; 3054 memcpy(via->rtvia_addr, &rt->rt_gw6, alen); 3055 } 3056 } 3057 3058 expires = READ_ONCE(rt->dst.expires); 3059 if (expires) { 3060 unsigned long now = jiffies; 3061 3062 if (time_before(now, expires)) 3063 expires -= now; 3064 else 3065 expires = 0; 3066 } 3067 3068 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics)); 3069 if (rt->rt_pmtu && expires) 3070 metrics[RTAX_MTU - 1] = rt->rt_pmtu; 3071 if (rt->rt_mtu_locked && expires) 3072 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU); 3073 if (rtnetlink_put_metrics(skb, metrics) < 0) 3074 goto nla_put_failure; 3075 3076 if (fl4) { 3077 if (fl4->flowi4_mark && 3078 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark)) 3079 goto nla_put_failure; 3080 3081 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) && 3082 nla_put_u32(skb, RTA_UID, 3083 from_kuid_munged(current_user_ns(), 3084 fl4->flowi4_uid))) 3085 goto nla_put_failure; 3086 3087 if (rt_is_input_route(rt)) { 3088 #ifdef CONFIG_IP_MROUTE 3089 if (ipv4_is_multicast(dst) && 3090 !ipv4_is_local_multicast(dst) && 3091 IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) { 3092 int err = ipmr_get_route(net, skb, 3093 fl4->saddr, fl4->daddr, 3094 r, portid); 3095 3096 if (err <= 0) { 3097 if (err == 0) 3098 return 0; 3099 goto nla_put_failure; 3100 } 3101 } else 3102 #endif 3103 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif)) 3104 goto nla_put_failure; 3105 } 3106 } 3107 3108 error = rt->dst.error; 3109 3110 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0) 3111 goto nla_put_failure; 3112 3113 nlmsg_end(skb, nlh); 3114 return 0; 3115 3116 nla_put_failure: 3117 nlmsg_cancel(skb, nlh); 3118 return -EMSGSIZE; 3119 } 3120 3121 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb, 3122 struct netlink_callback *cb, u32 table_id, 3123 struct fnhe_hash_bucket *bucket, int genid, 3124 int *fa_index, int fa_start, unsigned int flags) 3125 { 3126 int i; 3127 3128 for (i = 0; i < FNHE_HASH_SIZE; i++) { 3129 struct fib_nh_exception *fnhe; 3130 3131 for (fnhe = rcu_dereference(bucket[i].chain); fnhe; 3132 fnhe = rcu_dereference(fnhe->fnhe_next)) { 3133 struct rtable *rt; 3134 int err; 3135 3136 if (*fa_index < fa_start) 3137 goto next; 3138 3139 if (fnhe->fnhe_genid != genid) 3140 goto next; 3141 3142 if (fnhe->fnhe_expires && 3143 time_after(jiffies, fnhe->fnhe_expires)) 3144 goto next; 3145 3146 rt = rcu_dereference(fnhe->fnhe_rth_input); 3147 if (!rt) 3148 rt = rcu_dereference(fnhe->fnhe_rth_output); 3149 if (!rt) 3150 goto next; 3151 3152 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt, 3153 table_id, 0, NULL, skb, 3154 NETLINK_CB(cb->skb).portid, 3155 cb->nlh->nlmsg_seq, flags); 3156 if (err) 3157 return err; 3158 next: 3159 (*fa_index)++; 3160 } 3161 } 3162 3163 return 0; 3164 } 3165 3166 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb, 3167 u32 table_id, struct fib_info *fi, 3168 int *fa_index, int fa_start, unsigned int flags) 3169 { 3170 struct net *net = sock_net(cb->skb->sk); 3171 int nhsel, genid = fnhe_genid(net); 3172 3173 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) { 3174 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel); 3175 struct fnhe_hash_bucket *bucket; 3176 int err; 3177 3178 if (nhc->nhc_flags & RTNH_F_DEAD) 3179 continue; 3180 3181 rcu_read_lock(); 3182 bucket = rcu_dereference(nhc->nhc_exceptions); 3183 err = 0; 3184 if (bucket) 3185 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket, 3186 genid, fa_index, fa_start, 3187 flags); 3188 rcu_read_unlock(); 3189 if (err) 3190 return err; 3191 } 3192 3193 return 0; 3194 } 3195 3196 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst, 3197 u8 ip_proto, __be16 sport, 3198 __be16 dport) 3199 { 3200 struct sk_buff *skb; 3201 struct iphdr *iph; 3202 3203 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 3204 if (!skb) 3205 return NULL; 3206 3207 /* Reserve room for dummy headers, this skb can pass 3208 * through good chunk of routing engine. 3209 */ 3210 skb_reset_mac_header(skb); 3211 skb_reset_network_header(skb); 3212 skb->protocol = htons(ETH_P_IP); 3213 iph = skb_put(skb, sizeof(struct iphdr)); 3214 iph->protocol = ip_proto; 3215 iph->saddr = src; 3216 iph->daddr = dst; 3217 iph->version = 0x4; 3218 iph->frag_off = 0; 3219 iph->ihl = 0x5; 3220 skb_set_transport_header(skb, skb->len); 3221 3222 switch (iph->protocol) { 3223 case IPPROTO_UDP: { 3224 struct udphdr *udph; 3225 3226 udph = skb_put_zero(skb, sizeof(struct udphdr)); 3227 udph->source = sport; 3228 udph->dest = dport; 3229 udp_set_len_short(udph, sizeof(struct udphdr)); 3230 udph->check = 0; 3231 break; 3232 } 3233 case IPPROTO_TCP: { 3234 struct tcphdr *tcph; 3235 3236 tcph = skb_put_zero(skb, sizeof(struct tcphdr)); 3237 tcph->source = sport; 3238 tcph->dest = dport; 3239 tcph->doff = sizeof(struct tcphdr) / 4; 3240 tcph->rst = 1; 3241 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr), 3242 src, dst, 0); 3243 break; 3244 } 3245 case IPPROTO_ICMP: { 3246 struct icmphdr *icmph; 3247 3248 icmph = skb_put_zero(skb, sizeof(struct icmphdr)); 3249 icmph->type = ICMP_ECHO; 3250 icmph->code = 0; 3251 } 3252 } 3253 3254 return skb; 3255 } 3256 3257 static int inet_rtm_valid_getroute_req(struct sk_buff *skb, 3258 const struct nlmsghdr *nlh, 3259 struct nlattr **tb, 3260 struct netlink_ext_ack *extack) 3261 { 3262 struct rtmsg *rtm; 3263 int i, err; 3264 3265 rtm = nlmsg_payload(nlh, sizeof(*rtm)); 3266 if (!rtm) { 3267 NL_SET_ERR_MSG(extack, 3268 "ipv4: Invalid header for route get request"); 3269 return -EINVAL; 3270 } 3271 3272 if (!netlink_strict_get_check(skb)) 3273 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX, 3274 rtm_ipv4_policy, extack); 3275 3276 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) || 3277 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) || 3278 rtm->rtm_table || rtm->rtm_protocol || 3279 rtm->rtm_scope || rtm->rtm_type) { 3280 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request"); 3281 return -EINVAL; 3282 } 3283 3284 if (rtm->rtm_flags & ~(RTM_F_NOTIFY | 3285 RTM_F_LOOKUP_TABLE | 3286 RTM_F_FIB_MATCH)) { 3287 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request"); 3288 return -EINVAL; 3289 } 3290 3291 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX, 3292 rtm_ipv4_policy, extack); 3293 if (err) 3294 return err; 3295 3296 if ((tb[RTA_SRC] && !rtm->rtm_src_len) || 3297 (tb[RTA_DST] && !rtm->rtm_dst_len)) { 3298 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4"); 3299 return -EINVAL; 3300 } 3301 3302 for (i = 0; i <= RTA_MAX; i++) { 3303 if (!tb[i]) 3304 continue; 3305 3306 switch (i) { 3307 case RTA_IIF: 3308 case RTA_OIF: 3309 case RTA_SRC: 3310 case RTA_DST: 3311 case RTA_IP_PROTO: 3312 case RTA_SPORT: 3313 case RTA_DPORT: 3314 case RTA_MARK: 3315 case RTA_UID: 3316 break; 3317 default: 3318 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request"); 3319 return -EINVAL; 3320 } 3321 } 3322 3323 return 0; 3324 } 3325 3326 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, 3327 struct netlink_ext_ack *extack) 3328 { 3329 struct net *net = sock_net(in_skb->sk); 3330 struct nlattr *tb[RTA_MAX+1]; 3331 u32 table_id = RT_TABLE_MAIN; 3332 __be16 sport = 0, dport = 0; 3333 struct fib_result res = {}; 3334 u8 ip_proto = IPPROTO_UDP; 3335 struct rtable *rt = NULL; 3336 struct sk_buff *skb; 3337 struct rtmsg *rtm; 3338 struct flowi4 fl4 = {}; 3339 __be32 dst = 0; 3340 __be32 src = 0; 3341 dscp_t dscp; 3342 kuid_t uid; 3343 u32 iif; 3344 int err; 3345 int mark; 3346 3347 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack); 3348 if (err < 0) 3349 return err; 3350 3351 rtm = nlmsg_data(nlh); 3352 src = nla_get_in_addr_default(tb[RTA_SRC], 0); 3353 dst = nla_get_in_addr_default(tb[RTA_DST], 0); 3354 iif = nla_get_u32_default(tb[RTA_IIF], 0); 3355 mark = nla_get_u32_default(tb[RTA_MARK], 0); 3356 dscp = inet_dsfield_to_dscp(rtm->rtm_tos); 3357 if (tb[RTA_UID]) 3358 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID])); 3359 else 3360 uid = (iif ? INVALID_UID : current_uid()); 3361 3362 if (tb[RTA_IP_PROTO]) { 3363 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO], 3364 &ip_proto, AF_INET, extack); 3365 if (err) 3366 return err; 3367 } 3368 3369 if (tb[RTA_SPORT]) 3370 sport = nla_get_be16(tb[RTA_SPORT]); 3371 3372 if (tb[RTA_DPORT]) 3373 dport = nla_get_be16(tb[RTA_DPORT]); 3374 3375 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport); 3376 if (!skb) 3377 return -ENOBUFS; 3378 3379 fl4.daddr = dst; 3380 fl4.saddr = src; 3381 fl4.flowi4_dscp = dscp; 3382 fl4.flowi4_oif = nla_get_u32_default(tb[RTA_OIF], 0); 3383 fl4.flowi4_mark = mark; 3384 fl4.flowi4_uid = uid; 3385 if (sport) 3386 fl4.fl4_sport = sport; 3387 if (dport) 3388 fl4.fl4_dport = dport; 3389 fl4.flowi4_proto = ip_proto; 3390 3391 rcu_read_lock(); 3392 3393 if (iif) { 3394 struct net_device *dev; 3395 3396 dev = dev_get_by_index_rcu(net, iif); 3397 if (!dev) { 3398 err = -ENODEV; 3399 goto errout_rcu; 3400 } 3401 3402 fl4.flowi4_iif = iif; /* for rt_fill_info */ 3403 skb->dev = dev; 3404 skb->mark = mark; 3405 err = ip_route_input_rcu(skb, dst, src, dscp, dev, 3406 &res) ? -EINVAL : 0; 3407 3408 rt = skb_rtable(skb); 3409 if (err == 0 && rt->dst.error) 3410 err = -rt->dst.error; 3411 } else { 3412 fl4.flowi4_iif = LOOPBACK_IFINDEX; 3413 skb->dev = net->loopback_dev; 3414 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb); 3415 err = 0; 3416 if (IS_ERR(rt)) 3417 err = PTR_ERR(rt); 3418 else 3419 skb_dst_set(skb, &rt->dst); 3420 } 3421 3422 if (err) 3423 goto errout_rcu; 3424 3425 if (rtm->rtm_flags & RTM_F_NOTIFY) 3426 rt->rt_flags |= RTCF_NOTIFY; 3427 3428 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE) 3429 table_id = res.table ? res.table->tb_id : 0; 3430 3431 /* reset skb for netlink reply msg */ 3432 skb_trim(skb, 0); 3433 skb_reset_network_header(skb); 3434 skb_reset_transport_header(skb); 3435 skb_reset_mac_header(skb); 3436 3437 if (rtm->rtm_flags & RTM_F_FIB_MATCH) { 3438 struct fib_rt_info fri; 3439 3440 if (!res.fi) { 3441 err = fib_props[res.type].error; 3442 if (!err) 3443 err = -EHOSTUNREACH; 3444 goto errout_rcu; 3445 } 3446 fri.fi = res.fi; 3447 fri.tb_id = table_id; 3448 fri.dst = res.prefix; 3449 fri.dst_len = res.prefixlen; 3450 fri.dscp = res.dscp; 3451 fri.type = rt->rt_type; 3452 fri.offload = 0; 3453 fri.trap = 0; 3454 fri.offload_failed = 0; 3455 if (res.fa_head) { 3456 struct fib_alias *fa; 3457 3458 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) { 3459 u8 slen = 32 - fri.dst_len; 3460 3461 if (fa->fa_slen == slen && 3462 fa->tb_id == fri.tb_id && 3463 fa->fa_dscp == fri.dscp && 3464 fa->fa_info == res.fi && 3465 fa->fa_type == fri.type) { 3466 fri.offload = READ_ONCE(fa->offload); 3467 fri.trap = READ_ONCE(fa->trap); 3468 fri.offload_failed = 3469 READ_ONCE(fa->offload_failed); 3470 break; 3471 } 3472 } 3473 } 3474 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid, 3475 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0); 3476 } else { 3477 err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4, 3478 skb, NETLINK_CB(in_skb).portid, 3479 nlh->nlmsg_seq, 0); 3480 } 3481 if (err < 0) 3482 goto errout_rcu; 3483 3484 rcu_read_unlock(); 3485 3486 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 3487 3488 errout_free: 3489 return err; 3490 errout_rcu: 3491 rcu_read_unlock(); 3492 kfree_skb(skb); 3493 goto errout_free; 3494 } 3495 3496 void ip_rt_multicast_event(struct in_device *in_dev) 3497 { 3498 rt_cache_flush(dev_net(in_dev->dev)); 3499 } 3500 3501 #ifdef CONFIG_SYSCTL 3502 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 3503 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 3504 static int ip_rt_gc_elasticity __read_mostly = 8; 3505 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU; 3506 3507 static int ipv4_sysctl_rtcache_flush(const struct ctl_table *__ctl, int write, 3508 void *buffer, size_t *lenp, loff_t *ppos) 3509 { 3510 struct net *net = (struct net *)__ctl->extra1; 3511 3512 if (write) { 3513 rt_cache_flush(net); 3514 fnhe_genid_bump(net); 3515 return 0; 3516 } 3517 3518 return -EINVAL; 3519 } 3520 3521 static struct ctl_table ipv4_route_table[] = { 3522 { 3523 .procname = "gc_thresh", 3524 .data = &ipv4_dst_ops.gc_thresh, 3525 .maxlen = sizeof(int), 3526 .mode = 0644, 3527 .proc_handler = proc_dointvec, 3528 }, 3529 { 3530 .procname = "max_size", 3531 .data = &ip_rt_max_size, 3532 .maxlen = sizeof(int), 3533 .mode = 0644, 3534 .proc_handler = proc_dointvec, 3535 }, 3536 { 3537 /* Deprecated. Use gc_min_interval_ms */ 3538 3539 .procname = "gc_min_interval", 3540 .data = &ip_rt_gc_min_interval, 3541 .maxlen = sizeof(int), 3542 .mode = 0644, 3543 .proc_handler = proc_dointvec_jiffies, 3544 }, 3545 { 3546 .procname = "gc_min_interval_ms", 3547 .data = &ip_rt_gc_min_interval, 3548 .maxlen = sizeof(int), 3549 .mode = 0644, 3550 .proc_handler = proc_dointvec_ms_jiffies, 3551 }, 3552 { 3553 .procname = "gc_timeout", 3554 .data = &ip_rt_gc_timeout, 3555 .maxlen = sizeof(int), 3556 .mode = 0644, 3557 .proc_handler = proc_dointvec_jiffies, 3558 }, 3559 { 3560 .procname = "gc_interval", 3561 .data = &ip_rt_gc_interval, 3562 .maxlen = sizeof(int), 3563 .mode = 0644, 3564 .proc_handler = proc_dointvec_jiffies, 3565 }, 3566 { 3567 .procname = "redirect_load", 3568 .data = &ip_rt_redirect_load, 3569 .maxlen = sizeof(int), 3570 .mode = 0644, 3571 .proc_handler = proc_dointvec, 3572 }, 3573 { 3574 .procname = "redirect_number", 3575 .data = &ip_rt_redirect_number, 3576 .maxlen = sizeof(int), 3577 .mode = 0644, 3578 .proc_handler = proc_dointvec, 3579 }, 3580 { 3581 .procname = "redirect_silence", 3582 .data = &ip_rt_redirect_silence, 3583 .maxlen = sizeof(int), 3584 .mode = 0644, 3585 .proc_handler = proc_dointvec, 3586 }, 3587 { 3588 .procname = "error_cost", 3589 .data = &ip_rt_error_cost, 3590 .maxlen = sizeof(int), 3591 .mode = 0644, 3592 .proc_handler = proc_dointvec, 3593 }, 3594 { 3595 .procname = "error_burst", 3596 .data = &ip_rt_error_burst, 3597 .maxlen = sizeof(int), 3598 .mode = 0644, 3599 .proc_handler = proc_dointvec, 3600 }, 3601 { 3602 .procname = "gc_elasticity", 3603 .data = &ip_rt_gc_elasticity, 3604 .maxlen = sizeof(int), 3605 .mode = 0644, 3606 .proc_handler = proc_dointvec, 3607 }, 3608 }; 3609 3610 static const char ipv4_route_flush_procname[] = "flush"; 3611 3612 static struct ctl_table ipv4_route_netns_table[] = { 3613 { 3614 .procname = ipv4_route_flush_procname, 3615 .maxlen = sizeof(int), 3616 .mode = 0200, 3617 .proc_handler = ipv4_sysctl_rtcache_flush, 3618 }, 3619 { 3620 .procname = "min_pmtu", 3621 .data = &init_net.ipv4.ip_rt_min_pmtu, 3622 .maxlen = sizeof(int), 3623 .mode = 0644, 3624 .proc_handler = proc_dointvec_minmax, 3625 .extra1 = &ip_min_valid_pmtu, 3626 }, 3627 { 3628 .procname = "mtu_expires", 3629 .data = &init_net.ipv4.ip_rt_mtu_expires, 3630 .maxlen = sizeof(int), 3631 .mode = 0644, 3632 .proc_handler = proc_dointvec_jiffies, 3633 }, 3634 { 3635 .procname = "min_adv_mss", 3636 .data = &init_net.ipv4.ip_rt_min_advmss, 3637 .maxlen = sizeof(int), 3638 .mode = 0644, 3639 .proc_handler = proc_dointvec, 3640 }, 3641 }; 3642 3643 static __net_init int sysctl_route_net_init(struct net *net) 3644 { 3645 struct ctl_table *tbl; 3646 size_t table_size = ARRAY_SIZE(ipv4_route_netns_table); 3647 3648 tbl = ipv4_route_netns_table; 3649 if (!net_eq(net, &init_net)) { 3650 int i; 3651 3652 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL); 3653 if (!tbl) 3654 goto err_dup; 3655 3656 /* Don't export non-whitelisted sysctls to unprivileged users */ 3657 if (net->user_ns != &init_user_ns) { 3658 if (tbl[0].procname != ipv4_route_flush_procname) 3659 table_size = 0; 3660 } 3661 3662 /* Update the variables to point into the current struct net 3663 * except for the first element flush 3664 */ 3665 for (i = 1; i < table_size; i++) 3666 tbl[i].data += (void *)net - (void *)&init_net; 3667 } 3668 tbl[0].extra1 = net; 3669 3670 net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route", 3671 tbl, table_size); 3672 if (!net->ipv4.route_hdr) 3673 goto err_reg; 3674 return 0; 3675 3676 err_reg: 3677 if (tbl != ipv4_route_netns_table) 3678 kfree(tbl); 3679 err_dup: 3680 return -ENOMEM; 3681 } 3682 3683 static __net_exit void sysctl_route_net_exit(struct net *net) 3684 { 3685 const struct ctl_table *tbl; 3686 3687 tbl = net->ipv4.route_hdr->ctl_table_arg; 3688 unregister_net_sysctl_table(net->ipv4.route_hdr); 3689 BUG_ON(tbl == ipv4_route_netns_table); 3690 kfree(tbl); 3691 } 3692 3693 static __net_initdata struct pernet_operations sysctl_route_ops = { 3694 .init = sysctl_route_net_init, 3695 .exit = sysctl_route_net_exit, 3696 }; 3697 #endif 3698 3699 static __net_init int netns_ip_rt_init(struct net *net) 3700 { 3701 /* Set default value for namespaceified sysctls */ 3702 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU; 3703 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES; 3704 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS; 3705 return 0; 3706 } 3707 3708 static struct pernet_operations __net_initdata ip_rt_ops = { 3709 .init = netns_ip_rt_init, 3710 }; 3711 3712 static __net_init int rt_genid_init(struct net *net) 3713 { 3714 atomic_set(&net->ipv4.rt_genid, 0); 3715 atomic_set(&net->fnhe_genid, 0); 3716 atomic_set(&net->ipv4.dev_addr_genid, get_random_u32()); 3717 return 0; 3718 } 3719 3720 static __net_initdata struct pernet_operations rt_genid_ops = { 3721 .init = rt_genid_init, 3722 }; 3723 3724 static int __net_init ipv4_inetpeer_init(struct net *net) 3725 { 3726 struct inet_peer_base *bp = kmalloc_obj(*bp); 3727 3728 if (!bp) 3729 return -ENOMEM; 3730 inet_peer_base_init(bp); 3731 net->ipv4.peers = bp; 3732 return 0; 3733 } 3734 3735 static void __net_exit ipv4_inetpeer_exit(struct net *net) 3736 { 3737 struct inet_peer_base *bp = net->ipv4.peers; 3738 3739 net->ipv4.peers = NULL; 3740 inetpeer_invalidate_tree(bp); 3741 kfree(bp); 3742 } 3743 3744 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = { 3745 .init = ipv4_inetpeer_init, 3746 .exit = ipv4_inetpeer_exit, 3747 }; 3748 3749 #ifdef CONFIG_IP_ROUTE_CLASSID 3750 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 3751 #endif /* CONFIG_IP_ROUTE_CLASSID */ 3752 3753 static const struct rtnl_msg_handler ip_rt_rtnl_msg_handlers[] __initconst = { 3754 {.protocol = PF_INET, .msgtype = RTM_GETROUTE, 3755 .doit = inet_rtm_getroute, .flags = RTNL_FLAG_DOIT_UNLOCKED}, 3756 }; 3757 3758 int __init ip_rt_init(void) 3759 { 3760 void *idents_hash; 3761 int cpu; 3762 3763 /* For modern hosts, this will use 2 MB of memory */ 3764 idents_hash = alloc_large_system_hash("IP idents", 3765 sizeof(*ip_idents) + sizeof(*ip_tstamps), 3766 0, 3767 16, /* one bucket per 64 KB */ 3768 HASH_ZERO, 3769 NULL, 3770 &ip_idents_mask, 3771 2048, 3772 256*1024); 3773 3774 ip_idents = idents_hash; 3775 3776 get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents)); 3777 3778 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents); 3779 3780 for_each_possible_cpu(cpu) { 3781 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 3782 3783 INIT_LIST_HEAD(&ul->head); 3784 spin_lock_init(&ul->lock); 3785 } 3786 #ifdef CONFIG_IP_ROUTE_CLASSID 3787 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 3788 if (!ip_rt_acct) 3789 panic("IP: failed to allocate ip_rt_acct\n"); 3790 #endif 3791 3792 ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable, 3793 SLAB_HWCACHE_ALIGN | SLAB_PANIC); 3794 3795 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 3796 3797 if (dst_entries_init(&ipv4_dst_ops) < 0) 3798 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 3799 3800 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 3801 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 3802 3803 ipv4_dst_ops.gc_thresh = ~0; 3804 ip_rt_max_size = INT_MAX; 3805 3806 devinet_init(); 3807 ip_fib_init(); 3808 3809 if (ip_rt_proc_init()) 3810 pr_err("Unable to create route proc files\n"); 3811 #ifdef CONFIG_XFRM 3812 xfrm_init(); 3813 xfrm4_init(); 3814 #endif 3815 rtnl_register_many(ip_rt_rtnl_msg_handlers); 3816 3817 #ifdef CONFIG_SYSCTL 3818 register_pernet_subsys(&sysctl_route_ops); 3819 #endif 3820 register_pernet_subsys(&ip_rt_ops); 3821 register_pernet_subsys(&rt_genid_ops); 3822 register_pernet_subsys(&ipv4_inetpeer_ops); 3823 return 0; 3824 } 3825 3826 #ifdef CONFIG_SYSCTL 3827 /* 3828 * We really need to sanitize the damn ipv4 init order, then all 3829 * this nonsense will go away. 3830 */ 3831 void __init ip_static_sysctl_init(void) 3832 { 3833 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table); 3834 } 3835 #endif 3836