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 if (!nhc) 1086 break; 1087 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1088 jiffies + net->ipv4.ip_rt_mtu_expires); 1089 } 1090 goto out; 1091 } 1092 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 1093 nhc = FIB_RES_NHC(res); 1094 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1095 jiffies + net->ipv4.ip_rt_mtu_expires); 1096 } 1097 out: 1098 rcu_read_unlock(); 1099 } 1100 1101 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 1102 struct sk_buff *skb, u32 mtu, 1103 bool confirm_neigh) 1104 { 1105 struct rtable *rt = dst_rtable(dst); 1106 struct flowi4 fl4; 1107 1108 ip_rt_build_flow_key(&fl4, sk, skb); 1109 1110 /* Don't make lookup fail for bridged encapsulations */ 1111 if (skb && netif_is_any_bridge_port(skb->dev)) 1112 fl4.flowi4_oif = 0; 1113 1114 __ip_rt_update_pmtu(rt, &fl4, mtu); 1115 } 1116 1117 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, 1118 int oif, u8 protocol) 1119 { 1120 const struct iphdr *iph = (const struct iphdr *)skb->data; 1121 struct flowi4 fl4; 1122 struct rtable *rt; 1123 u32 mark = IP4_REPLY_MARK(net, skb->mark); 1124 1125 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark, 1126 0); 1127 rt = __ip_route_output_key(net, &fl4); 1128 if (!IS_ERR(rt)) { 1129 __ip_rt_update_pmtu(rt, &fl4, mtu); 1130 ip_rt_put(rt); 1131 } 1132 } 1133 EXPORT_SYMBOL_GPL(ipv4_update_pmtu); 1134 1135 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1136 { 1137 const struct iphdr *iph = (const struct iphdr *)skb->data; 1138 struct flowi4 fl4; 1139 struct rtable *rt; 1140 1141 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0); 1142 1143 if (!fl4.flowi4_mark) 1144 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark); 1145 1146 rt = __ip_route_output_key(sock_net(sk), &fl4); 1147 if (!IS_ERR(rt)) { 1148 __ip_rt_update_pmtu(rt, &fl4, mtu); 1149 ip_rt_put(rt); 1150 } 1151 } 1152 1153 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1154 { 1155 const struct iphdr *iph = (const struct iphdr *)skb->data; 1156 struct flowi4 fl4; 1157 struct rtable *rt; 1158 struct dst_entry *odst = NULL; 1159 bool new = false; 1160 struct net *net = sock_net(sk); 1161 1162 bh_lock_sock(sk); 1163 1164 if (!ip_sk_accept_pmtu(sk)) 1165 goto out; 1166 1167 odst = sk_dst_get(sk); 1168 1169 if (sock_owned_by_user(sk) || !odst) { 1170 __ipv4_sk_update_pmtu(skb, sk, mtu); 1171 goto out; 1172 } 1173 1174 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1175 1176 rt = dst_rtable(odst); 1177 if (READ_ONCE(odst->obsolete) && !odst->ops->check(odst, 0)) { 1178 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1179 if (IS_ERR(rt)) 1180 goto out; 1181 1182 new = true; 1183 } 1184 1185 __ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu); 1186 1187 if (!dst_check(&rt->dst, 0)) { 1188 if (new) 1189 dst_release(&rt->dst); 1190 1191 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1192 if (IS_ERR(rt)) 1193 goto out; 1194 1195 new = true; 1196 } 1197 1198 if (new) 1199 sk_dst_set(sk, &rt->dst); 1200 1201 out: 1202 bh_unlock_sock(sk); 1203 dst_release(odst); 1204 } 1205 1206 void ipv4_redirect(struct sk_buff *skb, struct net *net, 1207 int oif, u8 protocol) 1208 { 1209 const struct iphdr *iph = (const struct iphdr *)skb->data; 1210 struct flowi4 fl4; 1211 struct rtable *rt; 1212 1213 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0); 1214 rt = __ip_route_output_key(net, &fl4); 1215 if (!IS_ERR(rt)) { 1216 __ip_do_redirect(rt, skb, &fl4, false); 1217 ip_rt_put(rt); 1218 } 1219 } 1220 EXPORT_SYMBOL_GPL(ipv4_redirect); 1221 1222 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk) 1223 { 1224 const struct iphdr *iph = (const struct iphdr *)skb->data; 1225 struct flowi4 fl4; 1226 struct rtable *rt; 1227 struct net *net = sock_net(sk); 1228 1229 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1230 rt = __ip_route_output_key(net, &fl4); 1231 if (!IS_ERR(rt)) { 1232 __ip_do_redirect(rt, skb, &fl4, false); 1233 ip_rt_put(rt); 1234 } 1235 } 1236 1237 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst, 1238 u32 cookie) 1239 { 1240 struct rtable *rt = dst_rtable(dst); 1241 1242 /* All IPV4 dsts are created with ->obsolete set to the value 1243 * DST_OBSOLETE_FORCE_CHK which forces validation calls down 1244 * into this function always. 1245 * 1246 * When a PMTU/redirect information update invalidates a route, 1247 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or 1248 * DST_OBSOLETE_DEAD. 1249 */ 1250 if (READ_ONCE(dst->obsolete) != DST_OBSOLETE_FORCE_CHK || 1251 rt_is_expired(rt)) 1252 return NULL; 1253 return dst; 1254 } 1255 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check); 1256 1257 static void ipv4_send_dest_unreach(struct sk_buff *skb) 1258 { 1259 struct inet_skb_parm parm; 1260 struct net_device *dev; 1261 int res; 1262 1263 /* Recompile ip options since IPCB may not be valid anymore. 1264 * Also check we have a reasonable ipv4 header. 1265 */ 1266 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) || 1267 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5) 1268 return; 1269 1270 memset(&parm, 0, sizeof(parm)); 1271 if (ip_hdr(skb)->ihl > 5) { 1272 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4)) 1273 return; 1274 parm.opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr); 1275 1276 rcu_read_lock(); 1277 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev; 1278 res = __ip_options_compile(dev_net(dev), &parm.opt, skb, NULL); 1279 rcu_read_unlock(); 1280 1281 if (res) 1282 return; 1283 } 1284 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &parm); 1285 } 1286 1287 static void ipv4_link_failure(struct sk_buff *skb) 1288 { 1289 struct rtable *rt; 1290 1291 ipv4_send_dest_unreach(skb); 1292 1293 rt = skb_rtable(skb); 1294 if (rt) 1295 dst_set_expires(&rt->dst, 0); 1296 } 1297 1298 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb) 1299 { 1300 pr_debug("%s: %pI4 -> %pI4, %s\n", 1301 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1302 skb->dev ? skb->dev->name : "?"); 1303 kfree_skb(skb); 1304 WARN_ON_ONCE(1); 1305 return 0; 1306 } 1307 1308 /* 1309 * We do not cache source address of outgoing interface, 1310 * because it is used only by IP RR, TS and SRR options, 1311 * so that it out of fast path. 1312 * 1313 * BTW remember: "addr" is allowed to be not aligned 1314 * in IP options! 1315 */ 1316 1317 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt) 1318 { 1319 __be32 src; 1320 1321 rcu_read_lock(); 1322 if (rt_is_output_route(rt)) { 1323 src = ip_hdr(skb)->saddr; 1324 } else { 1325 struct net_device *dev = dst_dev_rcu(&rt->dst); 1326 struct net *net = dev_net_rcu(dev); 1327 struct iphdr *iph = ip_hdr(skb); 1328 struct fib_result res; 1329 struct flowi4 fl4 = { 1330 .daddr = iph->daddr, 1331 .saddr = iph->saddr, 1332 .flowi4_dscp = ip4h_dscp(iph), 1333 .flowi4_oif = dev->ifindex, 1334 .flowi4_iif = skb->dev->ifindex, 1335 .flowi4_mark = skb->mark, 1336 }; 1337 1338 if (fib_lookup(net, &fl4, &res, 0) == 0) 1339 src = fib_result_prefsrc(net, &res); 1340 else 1341 src = inet_select_addr(dev, 1342 rt_nexthop(rt, iph->daddr), 1343 RT_SCOPE_UNIVERSE); 1344 } 1345 rcu_read_unlock(); 1346 1347 memcpy(addr, &src, 4); 1348 } 1349 1350 #ifdef CONFIG_IP_ROUTE_CLASSID 1351 static void set_class_tag(struct rtable *rt, u32 tag) 1352 { 1353 if (!(rt->dst.tclassid & 0xFFFF)) 1354 rt->dst.tclassid |= tag & 0xFFFF; 1355 if (!(rt->dst.tclassid & 0xFFFF0000)) 1356 rt->dst.tclassid |= tag & 0xFFFF0000; 1357 } 1358 #endif 1359 1360 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1361 { 1362 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr); 1363 unsigned int advmss; 1364 struct net *net; 1365 1366 rcu_read_lock(); 1367 net = dst_dev_net_rcu(dst); 1368 advmss = max_t(unsigned int, ip_dst_mtu_configured(dst) - header_size, 1369 net->ipv4.ip_rt_min_advmss); 1370 rcu_read_unlock(); 1371 1372 return min(advmss, IPV4_MAX_PMTU - header_size); 1373 } 1374 1375 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst) 1376 { 1377 return ip_dst_mtu_maybe_forward(dst, false); 1378 } 1379 EXPORT_INDIRECT_CALLABLE(ipv4_mtu); 1380 1381 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr) 1382 { 1383 struct fnhe_hash_bucket *hash; 1384 struct fib_nh_exception *fnhe, __rcu **fnhe_p; 1385 u32 hval = fnhe_hashfun(daddr); 1386 1387 spin_lock_bh(&fnhe_lock); 1388 1389 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1390 lockdep_is_held(&fnhe_lock)); 1391 hash += hval; 1392 1393 fnhe_p = &hash->chain; 1394 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock)); 1395 while (fnhe) { 1396 if (fnhe->fnhe_daddr == daddr) { 1397 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected( 1398 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock))); 1399 /* set fnhe_daddr to 0 to ensure it won't bind with 1400 * new dsts in rt_bind_exception(). 1401 */ 1402 fnhe->fnhe_daddr = 0; 1403 fnhe_flush_routes(fnhe); 1404 kfree_rcu(fnhe, rcu); 1405 break; 1406 } 1407 fnhe_p = &fnhe->fnhe_next; 1408 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1409 lockdep_is_held(&fnhe_lock)); 1410 } 1411 1412 spin_unlock_bh(&fnhe_lock); 1413 } 1414 1415 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc, 1416 __be32 daddr) 1417 { 1418 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions); 1419 struct fib_nh_exception *fnhe; 1420 u32 hval; 1421 1422 if (!hash) 1423 return NULL; 1424 1425 hval = fnhe_hashfun(daddr); 1426 1427 for (fnhe = rcu_dereference(hash[hval].chain); fnhe; 1428 fnhe = rcu_dereference(fnhe->fnhe_next)) { 1429 if (fnhe->fnhe_daddr == daddr) { 1430 if (fnhe->fnhe_expires && 1431 time_after(jiffies, fnhe->fnhe_expires)) { 1432 ip_del_fnhe(nhc, daddr); 1433 break; 1434 } 1435 return fnhe; 1436 } 1437 } 1438 return NULL; 1439 } 1440 1441 /* MTU selection: 1442 * 1. mtu on route is locked - use it 1443 * 2. mtu from nexthop exception 1444 * 3. mtu from egress device 1445 */ 1446 1447 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr) 1448 { 1449 struct fib_nh_common *nhc = res->nhc; 1450 struct net_device *dev = nhc->nhc_dev; 1451 struct fib_info *fi = res->fi; 1452 u32 mtu = 0; 1453 1454 if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) || 1455 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU)) 1456 mtu = fi->fib_mtu; 1457 1458 if (likely(!mtu)) { 1459 struct fib_nh_exception *fnhe; 1460 1461 fnhe = find_exception(nhc, daddr); 1462 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires)) 1463 mtu = fnhe->fnhe_pmtu; 1464 } 1465 1466 if (likely(!mtu)) 1467 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU); 1468 1469 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu); 1470 } 1471 1472 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, 1473 __be32 daddr, const bool do_cache) 1474 { 1475 bool ret = false; 1476 1477 spin_lock_bh(&fnhe_lock); 1478 1479 if (daddr == fnhe->fnhe_daddr) { 1480 struct rtable __rcu **porig; 1481 struct rtable *orig; 1482 int genid = fnhe_genid(dev_net(rt->dst.dev)); 1483 1484 if (rt_is_input_route(rt)) 1485 porig = &fnhe->fnhe_rth_input; 1486 else 1487 porig = &fnhe->fnhe_rth_output; 1488 orig = rcu_dereference(*porig); 1489 1490 if (fnhe->fnhe_genid != genid) { 1491 fnhe->fnhe_genid = genid; 1492 fnhe->fnhe_gw = 0; 1493 fnhe->fnhe_pmtu = 0; 1494 fnhe->fnhe_expires = 0; 1495 fnhe->fnhe_mtu_locked = false; 1496 fnhe_flush_routes(fnhe); 1497 orig = NULL; 1498 } 1499 fill_route_from_fnhe(rt, fnhe); 1500 if (!rt->rt_gw4) { 1501 rt->rt_gw4 = daddr; 1502 rt->rt_gw_family = AF_INET; 1503 } 1504 1505 if (do_cache) { 1506 dst_hold(&rt->dst); 1507 rcu_assign_pointer(*porig, rt); 1508 if (orig) { 1509 dst_dev_put(&orig->dst); 1510 dst_release(&orig->dst); 1511 } 1512 ret = true; 1513 } 1514 1515 fnhe->fnhe_stamp = jiffies; 1516 } 1517 spin_unlock_bh(&fnhe_lock); 1518 1519 return ret; 1520 } 1521 1522 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt) 1523 { 1524 struct rtable *orig, *prev, **p; 1525 bool ret = true; 1526 1527 if (rt_is_input_route(rt)) { 1528 p = (struct rtable **)&nhc->nhc_rth_input; 1529 } else { 1530 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 1531 } 1532 orig = *p; 1533 1534 /* hold dst before doing cmpxchg() to avoid race condition 1535 * on this dst 1536 */ 1537 dst_hold(&rt->dst); 1538 prev = cmpxchg(p, orig, rt); 1539 if (prev == orig) { 1540 if (orig) { 1541 rt_add_uncached_list(orig); 1542 dst_release(&orig->dst); 1543 } 1544 } else { 1545 dst_release(&rt->dst); 1546 ret = false; 1547 } 1548 1549 return ret; 1550 } 1551 1552 struct uncached_list { 1553 spinlock_t lock; 1554 struct list_head head; 1555 }; 1556 1557 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list); 1558 1559 void rt_add_uncached_list(struct rtable *rt) 1560 { 1561 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list); 1562 1563 rt->dst.rt_uncached_list = ul; 1564 1565 spin_lock_bh(&ul->lock); 1566 list_add_tail(&rt->dst.rt_uncached, &ul->head); 1567 spin_unlock_bh(&ul->lock); 1568 } 1569 1570 void rt_del_uncached_list(struct rtable *rt) 1571 { 1572 struct uncached_list *ul = rt->dst.rt_uncached_list; 1573 1574 if (ul) { 1575 spin_lock_bh(&ul->lock); 1576 list_del_init(&rt->dst.rt_uncached); 1577 spin_unlock_bh(&ul->lock); 1578 } 1579 } 1580 1581 static void ipv4_dst_destroy(struct dst_entry *dst) 1582 { 1583 ip_dst_metrics_put(dst); 1584 rt_del_uncached_list(dst_rtable(dst)); 1585 } 1586 1587 void rt_flush_dev(struct net_device *dev) 1588 { 1589 struct rtable *rt, *safe; 1590 int cpu; 1591 1592 if (dev && dev->dismantle) 1593 return; 1594 1595 for_each_possible_cpu(cpu) { 1596 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 1597 1598 if (list_empty(&ul->head)) 1599 continue; 1600 1601 spin_lock_bh(&ul->lock); 1602 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) { 1603 struct net_device *rt_dev = rt->dst.dev; 1604 1605 if (dev ? rt_dev != dev : 1606 READ_ONCE(rt_dev->reg_state) != NETREG_UNREGISTERED) 1607 continue; 1608 1609 rcu_assign_pointer(rt->dst.dev_rcu, blackhole_netdev); 1610 netdev_ref_replace(rt_dev, blackhole_netdev, 1611 &rt->dst.dev_tracker, GFP_ATOMIC); 1612 list_del_init(&rt->dst.rt_uncached); 1613 } 1614 spin_unlock_bh(&ul->lock); 1615 } 1616 } 1617 1618 static bool rt_cache_valid(const struct rtable *rt) 1619 { 1620 return rt && 1621 READ_ONCE(rt->dst.obsolete) == DST_OBSOLETE_FORCE_CHK && 1622 !rt_is_expired(rt); 1623 } 1624 1625 static void rt_set_nexthop(struct rtable *rt, __be32 daddr, 1626 const struct fib_result *res, 1627 struct fib_nh_exception *fnhe, 1628 struct fib_info *fi, u16 type, u32 itag, 1629 const bool do_cache) 1630 { 1631 bool cached = false; 1632 1633 if (fi) { 1634 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1635 1636 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) { 1637 rt->rt_uses_gateway = 1; 1638 rt->rt_gw_family = nhc->nhc_gw_family; 1639 /* only INET and INET6 are supported */ 1640 if (likely(nhc->nhc_gw_family == AF_INET)) 1641 rt->rt_gw4 = nhc->nhc_gw.ipv4; 1642 else 1643 rt->rt_gw6 = nhc->nhc_gw.ipv6; 1644 } 1645 1646 ip_dst_init_metrics(&rt->dst, fi->fib_metrics); 1647 1648 #ifdef CONFIG_IP_ROUTE_CLASSID 1649 if (nhc->nhc_family == AF_INET) { 1650 struct fib_nh *nh; 1651 1652 nh = container_of(nhc, struct fib_nh, nh_common); 1653 rt->dst.tclassid = nh->nh_tclassid; 1654 } 1655 #endif 1656 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 1657 if (unlikely(fnhe)) 1658 cached = rt_bind_exception(rt, fnhe, daddr, do_cache); 1659 else if (do_cache) 1660 cached = rt_cache_route(nhc, rt); 1661 if (unlikely(!cached)) { 1662 /* Routes we intend to cache in nexthop exception or 1663 * FIB nexthop have the DST_NOCACHE bit clear. 1664 * However, if we are unsuccessful at storing this 1665 * route into the cache we really need to set it. 1666 */ 1667 if (!rt->rt_gw4) { 1668 rt->rt_gw_family = AF_INET; 1669 rt->rt_gw4 = daddr; 1670 } 1671 rt_add_uncached_list(rt); 1672 } 1673 } else 1674 rt_add_uncached_list(rt); 1675 1676 #ifdef CONFIG_IP_ROUTE_CLASSID 1677 #ifdef CONFIG_IP_MULTIPLE_TABLES 1678 set_class_tag(rt, res->tclassid); 1679 #endif 1680 set_class_tag(rt, itag); 1681 #endif 1682 } 1683 1684 struct rtable *rt_dst_alloc(struct net_device *dev, 1685 unsigned int flags, u16 type, 1686 bool noxfrm) 1687 { 1688 struct rtable *rt; 1689 1690 rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK, 1691 (noxfrm ? DST_NOXFRM : 0)); 1692 1693 if (rt) { 1694 rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1695 rt->rt_flags = flags; 1696 rt->rt_type = type; 1697 rt->rt_is_input = 0; 1698 rt->rt_iif = 0; 1699 rt->rt_pmtu = 0; 1700 rt->rt_mtu_locked = 0; 1701 rt->rt_uses_gateway = 0; 1702 rt->rt_gw_family = 0; 1703 rt->rt_gw4 = 0; 1704 1705 rt->dst.output = ip_output; 1706 if (flags & RTCF_LOCAL) 1707 rt->dst.input = ip_local_deliver; 1708 } 1709 1710 return rt; 1711 } 1712 EXPORT_SYMBOL(rt_dst_alloc); 1713 1714 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt) 1715 { 1716 struct rtable *new_rt; 1717 1718 new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK, 1719 rt->dst.flags); 1720 1721 if (new_rt) { 1722 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1723 new_rt->rt_flags = rt->rt_flags; 1724 new_rt->rt_type = rt->rt_type; 1725 new_rt->rt_is_input = rt->rt_is_input; 1726 new_rt->rt_iif = rt->rt_iif; 1727 new_rt->rt_pmtu = rt->rt_pmtu; 1728 new_rt->rt_mtu_locked = rt->rt_mtu_locked; 1729 new_rt->rt_gw_family = rt->rt_gw_family; 1730 if (rt->rt_gw_family == AF_INET) 1731 new_rt->rt_gw4 = rt->rt_gw4; 1732 else if (rt->rt_gw_family == AF_INET6) 1733 new_rt->rt_gw6 = rt->rt_gw6; 1734 1735 new_rt->dst.input = READ_ONCE(rt->dst.input); 1736 new_rt->dst.output = READ_ONCE(rt->dst.output); 1737 new_rt->dst.error = rt->dst.error; 1738 new_rt->dst.lastuse = jiffies; 1739 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate); 1740 } 1741 return new_rt; 1742 } 1743 1744 /* called in rcu_read_lock() section */ 1745 enum skb_drop_reason 1746 ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1747 dscp_t dscp, struct net_device *dev, 1748 struct in_device *in_dev, u32 *itag) 1749 { 1750 enum skb_drop_reason reason; 1751 1752 /* Primary sanity checks. */ 1753 if (!in_dev) 1754 return SKB_DROP_REASON_NOT_SPECIFIED; 1755 1756 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 1757 return SKB_DROP_REASON_IP_INVALID_SOURCE; 1758 1759 if (skb->protocol != htons(ETH_P_IP)) 1760 return SKB_DROP_REASON_INVALID_PROTO; 1761 1762 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev)) 1763 return SKB_DROP_REASON_IP_LOCALNET; 1764 1765 if (ipv4_is_zeronet(saddr)) { 1766 if (!ipv4_is_local_multicast(daddr) && 1767 ip_hdr(skb)->protocol != IPPROTO_IGMP) 1768 return SKB_DROP_REASON_IP_INVALID_SOURCE; 1769 } else { 1770 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0, 1771 dev, in_dev, itag); 1772 if (reason) 1773 return reason; 1774 } 1775 return SKB_NOT_DROPPED_YET; 1776 } 1777 1778 /* called in rcu_read_lock() section */ 1779 static enum skb_drop_reason 1780 ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1781 dscp_t dscp, struct net_device *dev, int our) 1782 { 1783 struct in_device *in_dev = __in_dev_get_rcu(dev); 1784 unsigned int flags = RTCF_MULTICAST; 1785 enum skb_drop_reason reason; 1786 struct rtable *rth; 1787 u32 itag = 0; 1788 1789 reason = ip_mc_validate_source(skb, daddr, saddr, dscp, dev, in_dev, 1790 &itag); 1791 if (reason) 1792 return reason; 1793 1794 if (our) 1795 flags |= RTCF_LOCAL; 1796 1797 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 1798 IPCB(skb)->flags |= IPSKB_NOPOLICY; 1799 1800 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST, 1801 false); 1802 if (!rth) 1803 return SKB_DROP_REASON_NOMEM; 1804 1805 #ifdef CONFIG_IP_ROUTE_CLASSID 1806 rth->dst.tclassid = itag; 1807 #endif 1808 rth->dst.output = ip_rt_bug; 1809 rth->rt_is_input= 1; 1810 1811 #ifdef CONFIG_IP_MROUTE 1812 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1813 rth->dst.input = ip_mr_input; 1814 #endif 1815 RT_CACHE_STAT_INC(in_slow_mc); 1816 1817 skb_dst_drop(skb); 1818 skb_dst_set(skb, &rth->dst); 1819 return SKB_NOT_DROPPED_YET; 1820 } 1821 1822 1823 static void ip_handle_martian_source(struct net_device *dev, 1824 struct in_device *in_dev, 1825 struct sk_buff *skb, 1826 __be32 daddr, 1827 __be32 saddr) 1828 { 1829 RT_CACHE_STAT_INC(in_martian_src); 1830 #ifdef CONFIG_IP_ROUTE_VERBOSE 1831 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1832 /* 1833 * RFC1812 recommendation, if source is martian, 1834 * the only hint is MAC header. 1835 */ 1836 pr_warn("martian source (src=%pI4, dst=%pI4, dev=%s)\n", 1837 &saddr, &daddr, dev->name); 1838 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1839 print_hex_dump(KERN_WARNING, "ll header: ", 1840 DUMP_PREFIX_OFFSET, 16, 1, 1841 skb_mac_header(skb), 1842 dev->hard_header_len, false); 1843 } 1844 } 1845 #endif 1846 } 1847 1848 /* called in rcu_read_lock() section */ 1849 static enum skb_drop_reason 1850 __mkroute_input(struct sk_buff *skb, const struct fib_result *res, 1851 struct in_device *in_dev, __be32 daddr, 1852 __be32 saddr, dscp_t dscp) 1853 { 1854 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 1855 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1856 struct net_device *dev = nhc->nhc_dev; 1857 struct fib_nh_exception *fnhe; 1858 struct rtable *rth; 1859 int err; 1860 struct in_device *out_dev; 1861 bool do_cache; 1862 u32 itag = 0; 1863 1864 /* get a working reference to the output device */ 1865 out_dev = __in_dev_get_rcu(dev); 1866 if (!out_dev) { 1867 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n"); 1868 return reason; 1869 } 1870 1871 err = fib_validate_source(skb, saddr, daddr, dscp, FIB_RES_OIF(*res), 1872 in_dev->dev, in_dev, &itag); 1873 if (err < 0) { 1874 reason = -err; 1875 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1876 saddr); 1877 1878 goto cleanup; 1879 } 1880 1881 do_cache = res->fi && !itag; 1882 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) && 1883 skb->protocol == htons(ETH_P_IP)) { 1884 __be32 gw; 1885 1886 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0; 1887 if (IN_DEV_SHARED_MEDIA(out_dev) || 1888 inet_addr_onlink(out_dev, saddr, gw)) 1889 IPCB(skb)->flags |= IPSKB_DOREDIRECT; 1890 } 1891 1892 if (skb->protocol != htons(ETH_P_IP)) { 1893 /* Not IP (i.e. ARP). Do not create route, if it is 1894 * invalid for proxy arp. DNAT routes are always valid. 1895 * 1896 * Proxy arp feature have been extended to allow, ARP 1897 * replies back to the same interface, to support 1898 * Private VLAN switch technologies. See arp.c. 1899 */ 1900 if (out_dev == in_dev && 1901 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 1902 reason = SKB_DROP_REASON_ARP_PVLAN_DISABLE; 1903 goto cleanup; 1904 } 1905 } 1906 1907 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 1908 IPCB(skb)->flags |= IPSKB_NOPOLICY; 1909 1910 fnhe = find_exception(nhc, daddr); 1911 if (do_cache) { 1912 if (fnhe) 1913 rth = rcu_dereference(fnhe->fnhe_rth_input); 1914 else 1915 rth = rcu_dereference(nhc->nhc_rth_input); 1916 if (rt_cache_valid(rth)) { 1917 skb_dst_set_noref(skb, &rth->dst); 1918 goto out; 1919 } 1920 } 1921 1922 rth = rt_dst_alloc(out_dev->dev, 0, res->type, 1923 IN_DEV_ORCONF(out_dev, NOXFRM)); 1924 if (!rth) { 1925 reason = SKB_DROP_REASON_NOMEM; 1926 goto cleanup; 1927 } 1928 1929 rth->rt_is_input = 1; 1930 RT_CACHE_STAT_INC(in_slow_tot); 1931 1932 rth->dst.input = ip_forward; 1933 1934 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag, 1935 do_cache); 1936 lwtunnel_set_redirect(&rth->dst); 1937 skb_dst_set(skb, &rth->dst); 1938 out: 1939 reason = SKB_NOT_DROPPED_YET; 1940 cleanup: 1941 return reason; 1942 } 1943 1944 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1945 /* To make ICMP packets follow the right flow, the multipath hash is 1946 * calculated from the inner IP addresses. 1947 */ 1948 static void ip_multipath_l3_keys(const struct sk_buff *skb, 1949 struct flow_keys *hash_keys) 1950 { 1951 const struct iphdr *outer_iph = ip_hdr(skb); 1952 const struct iphdr *key_iph = outer_iph; 1953 const struct iphdr *inner_iph; 1954 const struct icmphdr *icmph; 1955 struct iphdr _inner_iph; 1956 struct icmphdr _icmph; 1957 1958 if (likely(outer_iph->protocol != IPPROTO_ICMP)) 1959 goto out; 1960 1961 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0)) 1962 goto out; 1963 1964 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph), 1965 &_icmph); 1966 if (!icmph) 1967 goto out; 1968 1969 if (!icmp_is_err(icmph->type)) 1970 goto out; 1971 1972 inner_iph = skb_header_pointer(skb, 1973 outer_iph->ihl * 4 + sizeof(_icmph), 1974 sizeof(_inner_iph), &_inner_iph); 1975 if (!inner_iph) 1976 goto out; 1977 1978 key_iph = inner_iph; 1979 out: 1980 hash_keys->addrs.v4addrs.src = key_iph->saddr; 1981 hash_keys->addrs.v4addrs.dst = key_iph->daddr; 1982 } 1983 1984 static u32 fib_multipath_custom_hash_outer(const struct net *net, 1985 const struct sk_buff *skb, 1986 bool *p_has_inner) 1987 { 1988 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 1989 struct flow_keys keys, hash_keys; 1990 1991 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 1992 return 0; 1993 1994 memset(&hash_keys, 0, sizeof(hash_keys)); 1995 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP); 1996 1997 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1998 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 1999 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 2000 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 2001 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 2002 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 2003 hash_keys.basic.ip_proto = keys.basic.ip_proto; 2004 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) 2005 hash_keys.ports.src = keys.ports.src; 2006 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 2007 hash_keys.ports.dst = keys.ports.dst; 2008 2009 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION); 2010 return fib_multipath_hash_from_keys(net, &hash_keys); 2011 } 2012 2013 static u32 fib_multipath_custom_hash_inner(const struct net *net, 2014 const struct sk_buff *skb, 2015 bool has_inner) 2016 { 2017 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 2018 struct flow_keys keys, hash_keys; 2019 2020 /* We assume the packet carries an encapsulation, but if none was 2021 * encountered during dissection of the outer flow, then there is no 2022 * point in calling the flow dissector again. 2023 */ 2024 if (!has_inner) 2025 return 0; 2026 2027 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK)) 2028 return 0; 2029 2030 memset(&hash_keys, 0, sizeof(hash_keys)); 2031 skb_flow_dissect_flow_keys(skb, &keys, 0); 2032 2033 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION)) 2034 return 0; 2035 2036 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 2037 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2038 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 2039 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 2040 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 2041 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 2042 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 2043 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 2044 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 2045 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 2046 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 2047 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 2048 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL) 2049 hash_keys.tags.flow_label = keys.tags.flow_label; 2050 } 2051 2052 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO) 2053 hash_keys.basic.ip_proto = keys.basic.ip_proto; 2054 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT) 2055 hash_keys.ports.src = keys.ports.src; 2056 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT) 2057 hash_keys.ports.dst = keys.ports.dst; 2058 2059 return fib_multipath_hash_from_keys(net, &hash_keys); 2060 } 2061 2062 static u32 fib_multipath_custom_hash_skb(const struct net *net, 2063 const struct sk_buff *skb) 2064 { 2065 u32 mhash, mhash_inner; 2066 bool has_inner = true; 2067 2068 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner); 2069 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner); 2070 2071 return jhash_2words(mhash, mhash_inner, 0); 2072 } 2073 2074 static u32 fib_multipath_custom_hash_fl4(const struct net *net, 2075 const struct flowi4 *fl4) 2076 { 2077 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 2078 struct flow_keys hash_keys; 2079 2080 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 2081 return 0; 2082 2083 memset(&hash_keys, 0, sizeof(hash_keys)); 2084 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2085 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 2086 hash_keys.addrs.v4addrs.src = fl4->saddr; 2087 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 2088 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2089 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 2090 hash_keys.basic.ip_proto = fl4->flowi4_proto; 2091 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) { 2092 if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT) 2093 hash_keys.ports.src = (__force __be16)get_random_u16(); 2094 else 2095 hash_keys.ports.src = fl4->fl4_sport; 2096 } 2097 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 2098 hash_keys.ports.dst = fl4->fl4_dport; 2099 2100 return fib_multipath_hash_from_keys(net, &hash_keys); 2101 } 2102 2103 /* if skb is set it will be used and fl4 can be NULL */ 2104 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4, 2105 const struct sk_buff *skb, struct flow_keys *flkeys) 2106 { 2107 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0; 2108 struct flow_keys hash_keys; 2109 u32 mhash = 0; 2110 2111 switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) { 2112 case 0: 2113 memset(&hash_keys, 0, sizeof(hash_keys)); 2114 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2115 if (skb) { 2116 ip_multipath_l3_keys(skb, &hash_keys); 2117 } else { 2118 hash_keys.addrs.v4addrs.src = fl4->saddr; 2119 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2120 } 2121 mhash = fib_multipath_hash_from_keys(net, &hash_keys); 2122 break; 2123 case 1: 2124 /* skb is currently provided only when forwarding */ 2125 if (skb) { 2126 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 2127 struct flow_keys keys; 2128 2129 /* short-circuit if we already have L4 hash present */ 2130 if (skb->l4_hash) 2131 return skb_get_hash_raw(skb) >> 1; 2132 2133 memset(&hash_keys, 0, sizeof(hash_keys)); 2134 2135 if (!flkeys) { 2136 skb_flow_dissect_flow_keys(skb, &keys, flag); 2137 flkeys = &keys; 2138 } 2139 2140 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2141 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src; 2142 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst; 2143 hash_keys.ports.src = flkeys->ports.src; 2144 hash_keys.ports.dst = flkeys->ports.dst; 2145 hash_keys.basic.ip_proto = flkeys->basic.ip_proto; 2146 } else { 2147 memset(&hash_keys, 0, sizeof(hash_keys)); 2148 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2149 hash_keys.addrs.v4addrs.src = fl4->saddr; 2150 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2151 if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT) 2152 hash_keys.ports.src = (__force __be16)get_random_u16(); 2153 else 2154 hash_keys.ports.src = fl4->fl4_sport; 2155 hash_keys.ports.dst = fl4->fl4_dport; 2156 hash_keys.basic.ip_proto = fl4->flowi4_proto; 2157 } 2158 mhash = fib_multipath_hash_from_keys(net, &hash_keys); 2159 break; 2160 case 2: 2161 memset(&hash_keys, 0, sizeof(hash_keys)); 2162 /* skb is currently provided only when forwarding */ 2163 if (skb) { 2164 struct flow_keys keys; 2165 2166 skb_flow_dissect_flow_keys(skb, &keys, 0); 2167 /* Inner can be v4 or v6 */ 2168 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 2169 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2170 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 2171 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 2172 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 2173 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 2174 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 2175 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 2176 hash_keys.tags.flow_label = keys.tags.flow_label; 2177 hash_keys.basic.ip_proto = keys.basic.ip_proto; 2178 } else { 2179 /* Same as case 0 */ 2180 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2181 ip_multipath_l3_keys(skb, &hash_keys); 2182 } 2183 } else { 2184 /* Same as case 0 */ 2185 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2186 hash_keys.addrs.v4addrs.src = fl4->saddr; 2187 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2188 } 2189 mhash = fib_multipath_hash_from_keys(net, &hash_keys); 2190 break; 2191 case 3: 2192 if (skb) 2193 mhash = fib_multipath_custom_hash_skb(net, skb); 2194 else 2195 mhash = fib_multipath_custom_hash_fl4(net, fl4); 2196 break; 2197 } 2198 2199 if (multipath_hash) 2200 mhash = jhash_2words(mhash, multipath_hash, 0); 2201 2202 return mhash >> 1; 2203 } 2204 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 2205 2206 static enum skb_drop_reason 2207 ip_mkroute_input(struct sk_buff *skb, struct fib_result *res, 2208 struct in_device *in_dev, __be32 daddr, 2209 __be32 saddr, dscp_t dscp, struct flow_keys *hkeys) 2210 { 2211 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2212 if (res->fi && fib_info_num_path(res->fi) > 1) { 2213 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys); 2214 2215 fib_select_multipath(res, h, NULL); 2216 IPCB(skb)->flags |= IPSKB_MULTIPATH; 2217 } 2218 #endif 2219 2220 /* create a routing cache entry */ 2221 return __mkroute_input(skb, res, in_dev, daddr, saddr, dscp); 2222 } 2223 2224 /* Implements all the saddr-related checks as ip_route_input_slow(), 2225 * assuming daddr is valid and the destination is not a local broadcast one. 2226 * Uses the provided hint instead of performing a route lookup. 2227 */ 2228 enum skb_drop_reason 2229 ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2230 dscp_t dscp, struct net_device *dev, 2231 const struct sk_buff *hint) 2232 { 2233 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 2234 struct in_device *in_dev = __in_dev_get_rcu(dev); 2235 struct rtable *rt = skb_rtable(hint); 2236 struct net *net = dev_net(dev); 2237 u32 tag = 0; 2238 2239 if (!in_dev) 2240 return reason; 2241 2242 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) { 2243 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2244 goto martian_source; 2245 } 2246 2247 if (ipv4_is_zeronet(saddr)) { 2248 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2249 goto martian_source; 2250 } 2251 2252 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) { 2253 reason = SKB_DROP_REASON_IP_LOCALNET; 2254 goto martian_source; 2255 } 2256 2257 if (!(rt->rt_flags & RTCF_LOCAL)) 2258 goto skip_validate_source; 2259 2260 reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 0, dev, 2261 in_dev, &tag); 2262 if (reason) 2263 goto martian_source; 2264 2265 skip_validate_source: 2266 skb_dst_copy(skb, hint); 2267 return SKB_NOT_DROPPED_YET; 2268 2269 martian_source: 2270 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2271 return reason; 2272 } 2273 2274 /* get device for dst_alloc with local routes */ 2275 static struct net_device *ip_rt_get_dev(struct net *net, 2276 const struct fib_result *res) 2277 { 2278 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL; 2279 struct net_device *dev = NULL; 2280 2281 if (nhc) 2282 dev = l3mdev_master_dev_rcu(nhc->nhc_dev); 2283 2284 return dev ? : net->loopback_dev; 2285 } 2286 2287 /* 2288 * NOTE. We drop all the packets that has local source 2289 * addresses, because every properly looped back packet 2290 * must have correct destination already attached by output routine. 2291 * Changes in the enforced policies must be applied also to 2292 * ip_route_use_hint(). 2293 * 2294 * Such approach solves two big problems: 2295 * 1. Not simplex devices are handled properly. 2296 * 2. IP spoofing attempts are filtered with 100% of guarantee. 2297 * called with rcu_read_lock() 2298 */ 2299 2300 static enum skb_drop_reason 2301 ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2302 dscp_t dscp, struct net_device *dev, 2303 struct fib_result *res) 2304 { 2305 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 2306 struct in_device *in_dev = __in_dev_get_rcu(dev); 2307 struct flow_keys *flkeys = NULL, _flkeys; 2308 struct net *net = dev_net(dev); 2309 struct ip_tunnel_info *tun_info; 2310 int err = -EINVAL; 2311 unsigned int flags = 0; 2312 u32 itag = 0; 2313 struct rtable *rth; 2314 struct flowi4 fl4; 2315 bool do_cache = true; 2316 2317 /* IP on this device is disabled. */ 2318 2319 if (!in_dev) 2320 goto out; 2321 2322 /* Check for the most weird martians, which can be not detected 2323 * by fib_lookup. 2324 */ 2325 2326 tun_info = skb_tunnel_info(skb); 2327 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX)) 2328 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id; 2329 else 2330 fl4.flowi4_tun_key.tun_id = 0; 2331 skb_dst_drop(skb); 2332 2333 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) { 2334 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2335 goto martian_source; 2336 } 2337 2338 res->fi = NULL; 2339 res->table = NULL; 2340 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 2341 goto brd_input; 2342 2343 /* Accept zero addresses only to limited broadcast; 2344 * I even do not know to fix it or not. Waiting for complains :-) 2345 */ 2346 if (ipv4_is_zeronet(saddr)) { 2347 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2348 goto martian_source; 2349 } 2350 2351 if (ipv4_is_zeronet(daddr)) { 2352 reason = SKB_DROP_REASON_IP_INVALID_DEST; 2353 goto martian_destination; 2354 } 2355 2356 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(), 2357 * and call it once if daddr or/and saddr are loopback addresses 2358 */ 2359 if (ipv4_is_loopback(daddr)) { 2360 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) { 2361 reason = SKB_DROP_REASON_IP_LOCALNET; 2362 goto martian_destination; 2363 } 2364 } else if (ipv4_is_loopback(saddr)) { 2365 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) { 2366 reason = SKB_DROP_REASON_IP_LOCALNET; 2367 goto martian_source; 2368 } 2369 } 2370 2371 /* 2372 * Now we are ready to route packet. 2373 */ 2374 fl4.flowi4_l3mdev = 0; 2375 fl4.flowi4_oif = 0; 2376 fl4.flowi4_iif = dev->ifindex; 2377 fl4.flowi4_mark = skb->mark; 2378 fl4.flowi4_dscp = dscp; 2379 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 2380 fl4.flowi4_flags = 0; 2381 fl4.daddr = daddr; 2382 fl4.saddr = saddr; 2383 fl4.flowi4_uid = sock_net_uid(net, NULL); 2384 fl4.flowi4_multipath_hash = 0; 2385 2386 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) { 2387 flkeys = &_flkeys; 2388 } else { 2389 fl4.flowi4_proto = 0; 2390 fl4.fl4_sport = 0; 2391 fl4.fl4_dport = 0; 2392 } 2393 2394 err = fib_lookup(net, &fl4, res, 0); 2395 if (err != 0) { 2396 if (!IN_DEV_FORWARD(in_dev)) 2397 err = -EHOSTUNREACH; 2398 goto no_route; 2399 } 2400 2401 if (res->type == RTN_BROADCAST) { 2402 if (IN_DEV_BFORWARD(in_dev)) 2403 goto make_route; 2404 /* not do cache if bc_forwarding is enabled */ 2405 if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING)) 2406 do_cache = false; 2407 goto brd_input; 2408 } 2409 2410 err = -EINVAL; 2411 if (res->type == RTN_LOCAL) { 2412 reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 2413 0, dev, in_dev, &itag); 2414 if (reason) 2415 goto martian_source; 2416 goto local_input; 2417 } 2418 2419 if (!IN_DEV_FORWARD(in_dev)) { 2420 err = -EHOSTUNREACH; 2421 goto no_route; 2422 } 2423 if (res->type != RTN_UNICAST) { 2424 reason = SKB_DROP_REASON_IP_INVALID_DEST; 2425 goto martian_destination; 2426 } 2427 2428 make_route: 2429 reason = ip_mkroute_input(skb, res, in_dev, daddr, saddr, dscp, 2430 flkeys); 2431 2432 out: 2433 return reason; 2434 2435 brd_input: 2436 if (skb->protocol != htons(ETH_P_IP)) { 2437 reason = SKB_DROP_REASON_INVALID_PROTO; 2438 goto out; 2439 } 2440 2441 if (!ipv4_is_zeronet(saddr)) { 2442 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0, 2443 dev, in_dev, &itag); 2444 if (reason) 2445 goto martian_source; 2446 } 2447 flags |= RTCF_BROADCAST; 2448 res->type = RTN_BROADCAST; 2449 RT_CACHE_STAT_INC(in_brd); 2450 2451 local_input: 2452 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 2453 IPCB(skb)->flags |= IPSKB_NOPOLICY; 2454 2455 do_cache &= res->fi && !itag; 2456 if (do_cache) { 2457 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2458 2459 rth = rcu_dereference(nhc->nhc_rth_input); 2460 if (rt_cache_valid(rth)) { 2461 skb_dst_set_noref(skb, &rth->dst); 2462 reason = SKB_NOT_DROPPED_YET; 2463 goto out; 2464 } 2465 } 2466 2467 rth = rt_dst_alloc(ip_rt_get_dev(net, res), 2468 flags | RTCF_LOCAL, res->type, false); 2469 if (!rth) 2470 goto e_nobufs; 2471 2472 rth->dst.output= ip_rt_bug; 2473 #ifdef CONFIG_IP_ROUTE_CLASSID 2474 rth->dst.tclassid = itag; 2475 #endif 2476 rth->rt_is_input = 1; 2477 2478 RT_CACHE_STAT_INC(in_slow_tot); 2479 if (res->type == RTN_UNREACHABLE) { 2480 rth->dst.input= ip_error; 2481 rth->dst.error= -err; 2482 rth->rt_flags &= ~RTCF_LOCAL; 2483 } 2484 2485 if (do_cache) { 2486 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2487 2488 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 2489 if (lwtunnel_input_redirect(rth->dst.lwtstate)) { 2490 WARN_ON(rth->dst.input == lwtunnel_input); 2491 rth->dst.lwtstate->orig_input = rth->dst.input; 2492 rth->dst.input = lwtunnel_input; 2493 } 2494 2495 if (unlikely(!rt_cache_route(nhc, rth))) 2496 rt_add_uncached_list(rth); 2497 } 2498 skb_dst_set(skb, &rth->dst); 2499 reason = SKB_NOT_DROPPED_YET; 2500 goto out; 2501 2502 no_route: 2503 RT_CACHE_STAT_INC(in_no_route); 2504 res->type = RTN_UNREACHABLE; 2505 res->fi = NULL; 2506 res->table = NULL; 2507 goto local_input; 2508 2509 /* 2510 * Do not cache martian addresses: they should be logged (RFC1812) 2511 */ 2512 martian_destination: 2513 RT_CACHE_STAT_INC(in_martian_dst); 2514 #ifdef CONFIG_IP_ROUTE_VERBOSE 2515 if (IN_DEV_LOG_MARTIANS(in_dev)) 2516 net_warn_ratelimited("martian destination (src=%pI4, dst=%pI4, dev=%s)\n", 2517 &saddr, &daddr, dev->name); 2518 #endif 2519 goto out; 2520 2521 e_nobufs: 2522 reason = SKB_DROP_REASON_NOMEM; 2523 goto out; 2524 2525 martian_source: 2526 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2527 goto out; 2528 } 2529 2530 /* called with rcu_read_lock held */ 2531 static enum skb_drop_reason 2532 ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2533 dscp_t dscp, struct net_device *dev, 2534 struct fib_result *res) 2535 { 2536 /* Multicast recognition logic is moved from route cache to here. 2537 * The problem was that too many Ethernet cards have broken/missing 2538 * hardware multicast filters :-( As result the host on multicasting 2539 * network acquires a lot of useless route cache entries, sort of 2540 * SDR messages from all the world. Now we try to get rid of them. 2541 * Really, provided software IP multicast filter is organized 2542 * reasonably (at least, hashed), it does not result in a slowdown 2543 * comparing with route cache reject entries. 2544 * Note, that multicast routers are not affected, because 2545 * route cache entry is created eventually. 2546 */ 2547 if (ipv4_is_multicast(daddr)) { 2548 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 2549 struct in_device *in_dev = __in_dev_get_rcu(dev); 2550 int our = 0; 2551 2552 if (!in_dev) 2553 return reason; 2554 2555 our = ip_check_mc_rcu(in_dev, daddr, saddr, 2556 ip_hdr(skb)->protocol); 2557 2558 /* check l3 master if no match yet */ 2559 if (!our && netif_is_l3_slave(dev)) { 2560 struct in_device *l3_in_dev; 2561 2562 l3_in_dev = __in_dev_get_rcu(skb->dev); 2563 if (l3_in_dev) 2564 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr, 2565 ip_hdr(skb)->protocol); 2566 } 2567 2568 if (our 2569 #ifdef CONFIG_IP_MROUTE 2570 || 2571 (!ipv4_is_local_multicast(daddr) && 2572 IN_DEV_MFORWARD(in_dev)) 2573 #endif 2574 ) { 2575 reason = ip_route_input_mc(skb, daddr, saddr, dscp, 2576 dev, our); 2577 } 2578 return reason; 2579 } 2580 2581 return ip_route_input_slow(skb, daddr, saddr, dscp, dev, res); 2582 } 2583 2584 enum skb_drop_reason ip_route_input_noref(struct sk_buff *skb, __be32 daddr, 2585 __be32 saddr, dscp_t dscp, 2586 struct net_device *dev) 2587 { 2588 enum skb_drop_reason reason; 2589 struct fib_result res; 2590 2591 rcu_read_lock(); 2592 reason = ip_route_input_rcu(skb, daddr, saddr, dscp, dev, &res); 2593 rcu_read_unlock(); 2594 2595 return reason; 2596 } 2597 EXPORT_SYMBOL(ip_route_input_noref); 2598 2599 /* called with rcu_read_lock() */ 2600 static struct rtable *__mkroute_output(const struct fib_result *res, 2601 const struct flowi4 *fl4, int orig_oif, 2602 struct net_device *dev_out, 2603 unsigned int flags) 2604 { 2605 struct fib_info *fi = res->fi; 2606 struct fib_nh_exception *fnhe; 2607 struct in_device *in_dev; 2608 u16 type = res->type; 2609 struct rtable *rth; 2610 bool do_cache; 2611 2612 in_dev = __in_dev_get_rcu(dev_out); 2613 if (!in_dev) 2614 return ERR_PTR(-EINVAL); 2615 2616 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 2617 if (ipv4_is_loopback(fl4->saddr) && 2618 !(dev_out->flags & IFF_LOOPBACK) && 2619 !netif_is_l3_master(dev_out)) 2620 return ERR_PTR(-EINVAL); 2621 2622 if (ipv4_is_lbcast(fl4->daddr)) { 2623 type = RTN_BROADCAST; 2624 2625 /* reset fi to prevent gateway resolution */ 2626 fi = NULL; 2627 } else if (ipv4_is_multicast(fl4->daddr)) { 2628 type = RTN_MULTICAST; 2629 } else if (ipv4_is_zeronet(fl4->daddr)) { 2630 return ERR_PTR(-EINVAL); 2631 } 2632 2633 if (dev_out->flags & IFF_LOOPBACK) 2634 flags |= RTCF_LOCAL; 2635 2636 do_cache = true; 2637 if (type == RTN_BROADCAST) { 2638 flags |= RTCF_BROADCAST | RTCF_LOCAL; 2639 } else if (type == RTN_MULTICAST) { 2640 flags |= RTCF_MULTICAST | RTCF_LOCAL; 2641 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr, 2642 fl4->flowi4_proto)) 2643 flags &= ~RTCF_LOCAL; 2644 else 2645 do_cache = false; 2646 /* If multicast route do not exist use 2647 * default one, but do not gateway in this case. 2648 * Yes, it is hack. 2649 */ 2650 if (fi && res->prefixlen < 4) 2651 fi = NULL; 2652 } else if ((type == RTN_LOCAL) && (orig_oif != 0) && 2653 (orig_oif != dev_out->ifindex)) { 2654 /* For local routes that require a particular output interface 2655 * we do not want to cache the result. Caching the result 2656 * causes incorrect behaviour when there are multiple source 2657 * addresses on the interface, the end result being that if the 2658 * intended recipient is waiting on that interface for the 2659 * packet he won't receive it because it will be delivered on 2660 * the loopback interface and the IP_PKTINFO ipi_ifindex will 2661 * be set to the loopback interface as well. 2662 */ 2663 do_cache = false; 2664 } 2665 2666 fnhe = NULL; 2667 do_cache &= fi != NULL; 2668 if (fi) { 2669 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2670 struct rtable __rcu **prth; 2671 2672 fnhe = find_exception(nhc, fl4->daddr); 2673 if (!do_cache) 2674 goto add; 2675 if (fnhe) { 2676 prth = &fnhe->fnhe_rth_output; 2677 } else { 2678 if (unlikely(fl4->flowi4_flags & 2679 FLOWI_FLAG_KNOWN_NH && 2680 !(nhc->nhc_gw_family && 2681 nhc->nhc_scope == RT_SCOPE_LINK))) { 2682 do_cache = false; 2683 goto add; 2684 } 2685 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 2686 } 2687 rth = rcu_dereference(*prth); 2688 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst)) 2689 return rth; 2690 } 2691 2692 add: 2693 rth = rt_dst_alloc(dev_out, flags, type, 2694 IN_DEV_ORCONF(in_dev, NOXFRM)); 2695 if (!rth) 2696 return ERR_PTR(-ENOBUFS); 2697 2698 rth->rt_iif = orig_oif; 2699 2700 RT_CACHE_STAT_INC(out_slow_tot); 2701 2702 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 2703 if (flags & RTCF_LOCAL && 2704 !(dev_out->flags & IFF_LOOPBACK)) { 2705 rth->dst.output = ip_mc_output; 2706 RT_CACHE_STAT_INC(out_slow_mc); 2707 } 2708 #ifdef CONFIG_IP_MROUTE 2709 if (type == RTN_MULTICAST) { 2710 if (IN_DEV_MFORWARD(in_dev) && 2711 !ipv4_is_local_multicast(fl4->daddr)) { 2712 rth->dst.input = ip_mr_input; 2713 rth->dst.output = ip_mr_output; 2714 } 2715 } 2716 #endif 2717 } 2718 2719 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache); 2720 lwtunnel_set_redirect(&rth->dst); 2721 2722 return rth; 2723 } 2724 2725 /* 2726 * Major route resolver routine. 2727 */ 2728 2729 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4, 2730 const struct sk_buff *skb) 2731 { 2732 struct fib_result res = { 2733 .type = RTN_UNSPEC, 2734 .fi = NULL, 2735 .table = NULL, 2736 .tclassid = 0, 2737 }; 2738 struct rtable *rth; 2739 2740 fl4->flowi4_iif = LOOPBACK_IFINDEX; 2741 2742 rcu_read_lock(); 2743 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb); 2744 rcu_read_unlock(); 2745 2746 return rth; 2747 } 2748 EXPORT_SYMBOL_GPL(ip_route_output_key_hash); 2749 2750 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4, 2751 struct fib_result *res, 2752 const struct sk_buff *skb) 2753 { 2754 struct net_device *dev_out = NULL; 2755 int orig_oif = fl4->flowi4_oif; 2756 unsigned int flags = 0; 2757 struct rtable *rth; 2758 int err; 2759 2760 if (fl4->saddr) { 2761 if (ipv4_is_multicast(fl4->saddr) || 2762 ipv4_is_lbcast(fl4->saddr)) { 2763 rth = ERR_PTR(-EINVAL); 2764 goto out; 2765 } 2766 2767 rth = ERR_PTR(-ENETUNREACH); 2768 2769 /* I removed check for oif == dev_out->oif here. 2770 * It was wrong for two reasons: 2771 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 2772 * is assigned to multiple interfaces. 2773 * 2. Moreover, we are allowed to send packets with saddr 2774 * of another iface. --ANK 2775 */ 2776 2777 if (fl4->flowi4_oif == 0 && 2778 (ipv4_is_multicast(fl4->daddr) || 2779 ipv4_is_lbcast(fl4->daddr))) { 2780 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2781 dev_out = __ip_dev_find(net, fl4->saddr, false); 2782 if (!dev_out) 2783 goto out; 2784 2785 /* Special hack: user can direct multicasts 2786 * and limited broadcast via necessary interface 2787 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 2788 * This hack is not just for fun, it allows 2789 * vic,vat and friends to work. 2790 * They bind socket to loopback, set ttl to zero 2791 * and expect that it will work. 2792 * From the viewpoint of routing cache they are broken, 2793 * because we are not allowed to build multicast path 2794 * with loopback source addr (look, routing cache 2795 * cannot know, that ttl is zero, so that packet 2796 * will not leave this host and route is valid). 2797 * Luckily, this hack is good workaround. 2798 */ 2799 2800 fl4->flowi4_oif = dev_out->ifindex; 2801 goto make_route; 2802 } 2803 2804 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 2805 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2806 if (!__ip_dev_find(net, fl4->saddr, false)) 2807 goto out; 2808 } 2809 } 2810 2811 2812 if (fl4->flowi4_oif) { 2813 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif); 2814 rth = ERR_PTR(-ENODEV); 2815 if (!dev_out) 2816 goto out; 2817 2818 /* RACE: Check return value of inet_select_addr instead. */ 2819 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 2820 rth = ERR_PTR(-ENETUNREACH); 2821 goto out; 2822 } 2823 if (ipv4_is_local_multicast(fl4->daddr) || 2824 ipv4_is_lbcast(fl4->daddr) || 2825 fl4->flowi4_proto == IPPROTO_IGMP) { 2826 if (!fl4->saddr) 2827 fl4->saddr = inet_select_addr(dev_out, 0, 2828 RT_SCOPE_LINK); 2829 goto make_route; 2830 } 2831 if (!fl4->saddr) { 2832 if (ipv4_is_multicast(fl4->daddr)) 2833 fl4->saddr = inet_select_addr(dev_out, 0, 2834 fl4->flowi4_scope); 2835 else if (!fl4->daddr) 2836 fl4->saddr = inet_select_addr(dev_out, 0, 2837 RT_SCOPE_HOST); 2838 } 2839 } 2840 2841 if (!fl4->daddr) { 2842 fl4->daddr = fl4->saddr; 2843 if (!fl4->daddr) 2844 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK); 2845 dev_out = net->loopback_dev; 2846 fl4->flowi4_oif = LOOPBACK_IFINDEX; 2847 res->type = RTN_LOCAL; 2848 flags |= RTCF_LOCAL; 2849 goto make_route; 2850 } 2851 2852 err = fib_lookup(net, fl4, res, 0); 2853 if (err) { 2854 res->fi = NULL; 2855 res->table = NULL; 2856 if (fl4->flowi4_oif && 2857 (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) { 2858 /* Apparently, routing tables are wrong. Assume, 2859 * that the destination is on link. 2860 * 2861 * WHY? DW. 2862 * Because we are allowed to send to iface 2863 * even if it has NO routes and NO assigned 2864 * addresses. When oif is specified, routing 2865 * tables are looked up with only one purpose: 2866 * to catch if destination is gatewayed, rather than 2867 * direct. Moreover, if MSG_DONTROUTE is set, 2868 * we send packet, ignoring both routing tables 2869 * and ifaddr state. --ANK 2870 * 2871 * 2872 * We could make it even if oif is unknown, 2873 * likely IPv6, but we do not. 2874 */ 2875 2876 if (fl4->saddr == 0) 2877 fl4->saddr = inet_select_addr(dev_out, 0, 2878 RT_SCOPE_LINK); 2879 res->type = RTN_UNICAST; 2880 goto make_route; 2881 } 2882 rth = ERR_PTR(err); 2883 goto out; 2884 } 2885 2886 if (res->type == RTN_LOCAL) { 2887 if (!fl4->saddr) { 2888 if (res->fi->fib_prefsrc) 2889 fl4->saddr = res->fi->fib_prefsrc; 2890 else 2891 fl4->saddr = fl4->daddr; 2892 } 2893 2894 /* L3 master device is the loopback for that domain */ 2895 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? : 2896 net->loopback_dev; 2897 2898 /* make sure orig_oif points to fib result device even 2899 * though packet rx/tx happens over loopback or l3mdev 2900 */ 2901 orig_oif = FIB_RES_OIF(*res); 2902 2903 fl4->flowi4_oif = dev_out->ifindex; 2904 flags |= RTCF_LOCAL; 2905 goto make_route; 2906 } 2907 2908 fib_select_path(net, res, fl4, skb); 2909 2910 dev_out = FIB_RES_DEV(*res); 2911 2912 make_route: 2913 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags); 2914 2915 out: 2916 return rth; 2917 } 2918 2919 static struct dst_ops ipv4_dst_blackhole_ops = { 2920 .family = AF_INET, 2921 .default_advmss = ipv4_default_advmss, 2922 .neigh_lookup = ipv4_dst_neigh_lookup, 2923 .check = dst_blackhole_check, 2924 .cow_metrics = dst_blackhole_cow_metrics, 2925 .update_pmtu = dst_blackhole_update_pmtu, 2926 .redirect = dst_blackhole_redirect, 2927 .mtu = dst_blackhole_mtu, 2928 }; 2929 2930 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2931 { 2932 struct rtable *ort = dst_rtable(dst_orig); 2933 struct rtable *rt; 2934 2935 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0); 2936 if (rt) { 2937 struct dst_entry *new = &rt->dst; 2938 2939 new->__use = 1; 2940 new->input = dst_discard; 2941 new->output = dst_discard_out; 2942 2943 new->dev = net->loopback_dev; 2944 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC); 2945 2946 rt->rt_is_input = ort->rt_is_input; 2947 rt->rt_iif = ort->rt_iif; 2948 rt->rt_pmtu = ort->rt_pmtu; 2949 rt->rt_mtu_locked = ort->rt_mtu_locked; 2950 2951 rt->rt_genid = rt_genid_ipv4(net); 2952 rt->rt_flags = ort->rt_flags; 2953 rt->rt_type = ort->rt_type; 2954 rt->rt_uses_gateway = ort->rt_uses_gateway; 2955 rt->rt_gw_family = ort->rt_gw_family; 2956 if (rt->rt_gw_family == AF_INET) 2957 rt->rt_gw4 = ort->rt_gw4; 2958 else if (rt->rt_gw_family == AF_INET6) 2959 rt->rt_gw6 = ort->rt_gw6; 2960 } 2961 2962 dst_release(dst_orig); 2963 2964 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2965 } 2966 2967 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2968 const struct sock *sk) 2969 { 2970 struct rtable *rt = __ip_route_output_key(net, flp4); 2971 2972 if (IS_ERR(rt)) 2973 return rt; 2974 2975 if (flp4->flowi4_proto) { 2976 flp4->flowi4_oif = rt->dst.dev->ifindex; 2977 rt = dst_rtable(xfrm_lookup_route(net, &rt->dst, 2978 flowi4_to_flowi(flp4), 2979 sk, 0)); 2980 } 2981 2982 return rt; 2983 } 2984 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2985 2986 /* called with rcu_read_lock held */ 2987 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, 2988 struct rtable *rt, u32 table_id, dscp_t dscp, 2989 struct flowi4 *fl4, struct sk_buff *skb, u32 portid, 2990 u32 seq, unsigned int flags) 2991 { 2992 struct rtmsg *r; 2993 struct nlmsghdr *nlh; 2994 unsigned long expires = 0; 2995 u32 error; 2996 u32 metrics[RTAX_MAX]; 2997 2998 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags); 2999 if (!nlh) 3000 return -EMSGSIZE; 3001 3002 r = nlmsg_data(nlh); 3003 r->rtm_family = AF_INET; 3004 r->rtm_dst_len = 32; 3005 r->rtm_src_len = 0; 3006 r->rtm_tos = inet_dscp_to_dsfield(dscp); 3007 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT; 3008 if (nla_put_u32(skb, RTA_TABLE, table_id)) 3009 goto nla_put_failure; 3010 r->rtm_type = rt->rt_type; 3011 r->rtm_scope = RT_SCOPE_UNIVERSE; 3012 r->rtm_protocol = RTPROT_UNSPEC; 3013 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 3014 if (rt->rt_flags & RTCF_NOTIFY) 3015 r->rtm_flags |= RTM_F_NOTIFY; 3016 if (IPCB(skb)->flags & IPSKB_DOREDIRECT) 3017 r->rtm_flags |= RTCF_DOREDIRECT; 3018 3019 if (nla_put_in_addr(skb, RTA_DST, dst)) 3020 goto nla_put_failure; 3021 if (src) { 3022 r->rtm_src_len = 32; 3023 if (nla_put_in_addr(skb, RTA_SRC, src)) 3024 goto nla_put_failure; 3025 } 3026 if (rt->dst.dev && 3027 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex)) 3028 goto nla_put_failure; 3029 if (lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0) 3030 goto nla_put_failure; 3031 #ifdef CONFIG_IP_ROUTE_CLASSID 3032 if (rt->dst.tclassid && 3033 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid)) 3034 goto nla_put_failure; 3035 #endif 3036 if (fl4 && !rt_is_input_route(rt) && 3037 fl4->saddr != src) { 3038 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr)) 3039 goto nla_put_failure; 3040 } 3041 if (rt->rt_uses_gateway) { 3042 if (rt->rt_gw_family == AF_INET && 3043 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) { 3044 goto nla_put_failure; 3045 } else if (rt->rt_gw_family == AF_INET6) { 3046 int alen = sizeof(struct in6_addr); 3047 struct nlattr *nla; 3048 struct rtvia *via; 3049 3050 nla = nla_reserve(skb, RTA_VIA, alen + 2); 3051 if (!nla) 3052 goto nla_put_failure; 3053 3054 via = nla_data(nla); 3055 via->rtvia_family = AF_INET6; 3056 memcpy(via->rtvia_addr, &rt->rt_gw6, alen); 3057 } 3058 } 3059 3060 expires = READ_ONCE(rt->dst.expires); 3061 if (expires) { 3062 unsigned long now = jiffies; 3063 3064 if (time_before(now, expires)) 3065 expires -= now; 3066 else 3067 expires = 0; 3068 } 3069 3070 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics)); 3071 if (rt->rt_pmtu && expires) 3072 metrics[RTAX_MTU - 1] = rt->rt_pmtu; 3073 if (rt->rt_mtu_locked && expires) 3074 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU); 3075 if (rtnetlink_put_metrics(skb, metrics) < 0) 3076 goto nla_put_failure; 3077 3078 if (fl4) { 3079 if (fl4->flowi4_mark && 3080 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark)) 3081 goto nla_put_failure; 3082 3083 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) && 3084 nla_put_u32(skb, RTA_UID, 3085 from_kuid_munged(current_user_ns(), 3086 fl4->flowi4_uid))) 3087 goto nla_put_failure; 3088 3089 if (rt_is_input_route(rt)) { 3090 #ifdef CONFIG_IP_MROUTE 3091 if (ipv4_is_multicast(dst) && 3092 !ipv4_is_local_multicast(dst) && 3093 IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) { 3094 int err = ipmr_get_route(net, skb, 3095 fl4->saddr, fl4->daddr, 3096 r, portid); 3097 3098 if (err <= 0) { 3099 if (err == 0) 3100 return 0; 3101 goto nla_put_failure; 3102 } 3103 } else 3104 #endif 3105 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif)) 3106 goto nla_put_failure; 3107 } 3108 } 3109 3110 error = rt->dst.error; 3111 3112 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0) 3113 goto nla_put_failure; 3114 3115 nlmsg_end(skb, nlh); 3116 return 0; 3117 3118 nla_put_failure: 3119 nlmsg_cancel(skb, nlh); 3120 return -EMSGSIZE; 3121 } 3122 3123 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb, 3124 struct netlink_callback *cb, u32 table_id, 3125 struct fnhe_hash_bucket *bucket, int genid, 3126 int *fa_index, int fa_start, unsigned int flags) 3127 { 3128 int i; 3129 3130 for (i = 0; i < FNHE_HASH_SIZE; i++) { 3131 struct fib_nh_exception *fnhe; 3132 3133 for (fnhe = rcu_dereference(bucket[i].chain); fnhe; 3134 fnhe = rcu_dereference(fnhe->fnhe_next)) { 3135 struct rtable *rt; 3136 int err; 3137 3138 if (*fa_index < fa_start) 3139 goto next; 3140 3141 if (fnhe->fnhe_genid != genid) 3142 goto next; 3143 3144 if (fnhe->fnhe_expires && 3145 time_after(jiffies, fnhe->fnhe_expires)) 3146 goto next; 3147 3148 rt = rcu_dereference(fnhe->fnhe_rth_input); 3149 if (!rt) 3150 rt = rcu_dereference(fnhe->fnhe_rth_output); 3151 if (!rt) 3152 goto next; 3153 3154 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt, 3155 table_id, 0, NULL, skb, 3156 NETLINK_CB(cb->skb).portid, 3157 cb->nlh->nlmsg_seq, flags); 3158 if (err) 3159 return err; 3160 next: 3161 (*fa_index)++; 3162 } 3163 } 3164 3165 return 0; 3166 } 3167 3168 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb, 3169 u32 table_id, struct fib_info *fi, 3170 int *fa_index, int fa_start, unsigned int flags) 3171 { 3172 struct net *net = sock_net(cb->skb->sk); 3173 int nhsel, genid = fnhe_genid(net); 3174 3175 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) { 3176 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel); 3177 struct fnhe_hash_bucket *bucket; 3178 int err; 3179 3180 if (!nhc) 3181 break; 3182 3183 if (nhc->nhc_flags & RTNH_F_DEAD) 3184 continue; 3185 3186 rcu_read_lock(); 3187 bucket = rcu_dereference(nhc->nhc_exceptions); 3188 err = 0; 3189 if (bucket) 3190 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket, 3191 genid, fa_index, fa_start, 3192 flags); 3193 rcu_read_unlock(); 3194 if (err) 3195 return err; 3196 } 3197 3198 return 0; 3199 } 3200 3201 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst, 3202 u8 ip_proto, __be16 sport, 3203 __be16 dport) 3204 { 3205 struct sk_buff *skb; 3206 struct iphdr *iph; 3207 3208 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 3209 if (!skb) 3210 return NULL; 3211 3212 /* Reserve room for dummy headers, this skb can pass 3213 * through good chunk of routing engine. 3214 */ 3215 skb_reset_mac_header(skb); 3216 skb_reset_network_header(skb); 3217 skb->protocol = htons(ETH_P_IP); 3218 iph = skb_put(skb, sizeof(struct iphdr)); 3219 iph->protocol = ip_proto; 3220 iph->saddr = src; 3221 iph->daddr = dst; 3222 iph->version = 0x4; 3223 iph->frag_off = 0; 3224 iph->ihl = 0x5; 3225 skb_set_transport_header(skb, skb->len); 3226 3227 switch (iph->protocol) { 3228 case IPPROTO_UDP: { 3229 struct udphdr *udph; 3230 3231 udph = skb_put_zero(skb, sizeof(struct udphdr)); 3232 udph->source = sport; 3233 udph->dest = dport; 3234 udp_set_len_short(udph, sizeof(struct udphdr)); 3235 udph->check = 0; 3236 break; 3237 } 3238 case IPPROTO_TCP: { 3239 struct tcphdr *tcph; 3240 3241 tcph = skb_put_zero(skb, sizeof(struct tcphdr)); 3242 tcph->source = sport; 3243 tcph->dest = dport; 3244 tcph->doff = sizeof(struct tcphdr) / 4; 3245 tcph->rst = 1; 3246 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr), 3247 src, dst, 0); 3248 break; 3249 } 3250 case IPPROTO_ICMP: { 3251 struct icmphdr *icmph; 3252 3253 icmph = skb_put_zero(skb, sizeof(struct icmphdr)); 3254 icmph->type = ICMP_ECHO; 3255 icmph->code = 0; 3256 } 3257 } 3258 3259 return skb; 3260 } 3261 3262 static int inet_rtm_valid_getroute_req(struct sk_buff *skb, 3263 const struct nlmsghdr *nlh, 3264 struct nlattr **tb, 3265 struct netlink_ext_ack *extack) 3266 { 3267 struct rtmsg *rtm; 3268 int i, err; 3269 3270 rtm = nlmsg_payload(nlh, sizeof(*rtm)); 3271 if (!rtm) { 3272 NL_SET_ERR_MSG(extack, 3273 "ipv4: Invalid header for route get request"); 3274 return -EINVAL; 3275 } 3276 3277 if (!netlink_strict_get_check(skb)) 3278 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX, 3279 rtm_ipv4_policy, extack); 3280 3281 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) || 3282 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) || 3283 rtm->rtm_table || rtm->rtm_protocol || 3284 rtm->rtm_scope || rtm->rtm_type) { 3285 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request"); 3286 return -EINVAL; 3287 } 3288 3289 if (rtm->rtm_flags & ~(RTM_F_NOTIFY | 3290 RTM_F_LOOKUP_TABLE | 3291 RTM_F_FIB_MATCH)) { 3292 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request"); 3293 return -EINVAL; 3294 } 3295 3296 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX, 3297 rtm_ipv4_policy, extack); 3298 if (err) 3299 return err; 3300 3301 if ((tb[RTA_SRC] && !rtm->rtm_src_len) || 3302 (tb[RTA_DST] && !rtm->rtm_dst_len)) { 3303 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4"); 3304 return -EINVAL; 3305 } 3306 3307 for (i = 0; i <= RTA_MAX; i++) { 3308 if (!tb[i]) 3309 continue; 3310 3311 switch (i) { 3312 case RTA_IIF: 3313 case RTA_OIF: 3314 case RTA_SRC: 3315 case RTA_DST: 3316 case RTA_IP_PROTO: 3317 case RTA_SPORT: 3318 case RTA_DPORT: 3319 case RTA_MARK: 3320 case RTA_UID: 3321 break; 3322 default: 3323 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request"); 3324 return -EINVAL; 3325 } 3326 } 3327 3328 return 0; 3329 } 3330 3331 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, 3332 struct netlink_ext_ack *extack) 3333 { 3334 struct net *net = sock_net(in_skb->sk); 3335 struct nlattr *tb[RTA_MAX+1]; 3336 u32 table_id = RT_TABLE_MAIN; 3337 __be16 sport = 0, dport = 0; 3338 struct fib_result res = {}; 3339 u8 ip_proto = IPPROTO_UDP; 3340 struct rtable *rt = NULL; 3341 struct sk_buff *skb; 3342 struct rtmsg *rtm; 3343 struct flowi4 fl4 = {}; 3344 __be32 dst = 0; 3345 __be32 src = 0; 3346 dscp_t dscp; 3347 kuid_t uid; 3348 u32 iif; 3349 int err; 3350 int mark; 3351 3352 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack); 3353 if (err < 0) 3354 return err; 3355 3356 rtm = nlmsg_data(nlh); 3357 src = nla_get_in_addr_default(tb[RTA_SRC], 0); 3358 dst = nla_get_in_addr_default(tb[RTA_DST], 0); 3359 iif = nla_get_u32_default(tb[RTA_IIF], 0); 3360 mark = nla_get_u32_default(tb[RTA_MARK], 0); 3361 dscp = inet_dsfield_to_dscp(rtm->rtm_tos); 3362 if (tb[RTA_UID]) 3363 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID])); 3364 else 3365 uid = (iif ? INVALID_UID : current_uid()); 3366 3367 if (tb[RTA_IP_PROTO]) { 3368 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO], 3369 &ip_proto, AF_INET, extack); 3370 if (err) 3371 return err; 3372 } 3373 3374 if (tb[RTA_SPORT]) 3375 sport = nla_get_be16(tb[RTA_SPORT]); 3376 3377 if (tb[RTA_DPORT]) 3378 dport = nla_get_be16(tb[RTA_DPORT]); 3379 3380 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport); 3381 if (!skb) 3382 return -ENOBUFS; 3383 3384 fl4.daddr = dst; 3385 fl4.saddr = src; 3386 fl4.flowi4_dscp = dscp; 3387 fl4.flowi4_oif = nla_get_u32_default(tb[RTA_OIF], 0); 3388 fl4.flowi4_mark = mark; 3389 fl4.flowi4_uid = uid; 3390 if (sport) 3391 fl4.fl4_sport = sport; 3392 if (dport) 3393 fl4.fl4_dport = dport; 3394 fl4.flowi4_proto = ip_proto; 3395 3396 rcu_read_lock(); 3397 3398 if (iif) { 3399 struct net_device *dev; 3400 3401 dev = dev_get_by_index_rcu(net, iif); 3402 if (!dev) { 3403 err = -ENODEV; 3404 goto errout_rcu; 3405 } 3406 3407 fl4.flowi4_iif = iif; /* for rt_fill_info */ 3408 skb->dev = dev; 3409 skb->mark = mark; 3410 err = ip_route_input_rcu(skb, dst, src, dscp, dev, 3411 &res) ? -EINVAL : 0; 3412 3413 rt = skb_rtable(skb); 3414 if (err == 0 && rt->dst.error) 3415 err = -rt->dst.error; 3416 } else { 3417 fl4.flowi4_iif = LOOPBACK_IFINDEX; 3418 skb->dev = net->loopback_dev; 3419 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb); 3420 err = 0; 3421 if (IS_ERR(rt)) 3422 err = PTR_ERR(rt); 3423 else 3424 skb_dst_set(skb, &rt->dst); 3425 } 3426 3427 if (err) 3428 goto errout_rcu; 3429 3430 if (rtm->rtm_flags & RTM_F_NOTIFY) 3431 rt->rt_flags |= RTCF_NOTIFY; 3432 3433 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE) 3434 table_id = res.table ? res.table->tb_id : 0; 3435 3436 /* reset skb for netlink reply msg */ 3437 skb_trim(skb, 0); 3438 skb_reset_network_header(skb); 3439 skb_reset_transport_header(skb); 3440 skb_reset_mac_header(skb); 3441 3442 if (rtm->rtm_flags & RTM_F_FIB_MATCH) { 3443 struct fib_rt_info fri; 3444 3445 if (!res.fi) { 3446 err = fib_props[res.type].error; 3447 if (!err) 3448 err = -EHOSTUNREACH; 3449 goto errout_rcu; 3450 } 3451 fri.fi = res.fi; 3452 fri.tb_id = table_id; 3453 fri.dst = res.prefix; 3454 fri.dst_len = res.prefixlen; 3455 fri.dscp = res.dscp; 3456 fri.type = rt->rt_type; 3457 fri.offload = 0; 3458 fri.trap = 0; 3459 fri.offload_failed = 0; 3460 if (res.fa_head) { 3461 struct fib_alias *fa; 3462 3463 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) { 3464 u8 slen = 32 - fri.dst_len; 3465 3466 if (fa->fa_slen == slen && 3467 fa->tb_id == fri.tb_id && 3468 fa->fa_dscp == fri.dscp && 3469 fa->fa_info == res.fi && 3470 fa->fa_type == fri.type) { 3471 fri.offload = READ_ONCE(fa->offload); 3472 fri.trap = READ_ONCE(fa->trap); 3473 fri.offload_failed = 3474 READ_ONCE(fa->offload_failed); 3475 break; 3476 } 3477 } 3478 } 3479 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid, 3480 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0); 3481 } else { 3482 err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4, 3483 skb, NETLINK_CB(in_skb).portid, 3484 nlh->nlmsg_seq, 0); 3485 } 3486 if (err < 0) 3487 goto errout_rcu; 3488 3489 rcu_read_unlock(); 3490 3491 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 3492 3493 errout_free: 3494 return err; 3495 errout_rcu: 3496 rcu_read_unlock(); 3497 kfree_skb(skb); 3498 goto errout_free; 3499 } 3500 3501 void ip_rt_multicast_event(struct in_device *in_dev) 3502 { 3503 rt_cache_flush(dev_net(in_dev->dev)); 3504 } 3505 3506 #ifdef CONFIG_SYSCTL 3507 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 3508 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 3509 static int ip_rt_gc_elasticity __read_mostly = 8; 3510 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU; 3511 3512 static int ipv4_sysctl_rtcache_flush(const struct ctl_table *__ctl, int write, 3513 void *buffer, size_t *lenp, loff_t *ppos) 3514 { 3515 struct net *net = (struct net *)__ctl->extra1; 3516 3517 if (write) { 3518 rt_cache_flush(net); 3519 fnhe_genid_bump(net); 3520 return 0; 3521 } 3522 3523 return -EINVAL; 3524 } 3525 3526 static struct ctl_table ipv4_route_table[] = { 3527 { 3528 .procname = "gc_thresh", 3529 .data = &ipv4_dst_ops.gc_thresh, 3530 .maxlen = sizeof(int), 3531 .mode = 0644, 3532 .proc_handler = proc_dointvec, 3533 }, 3534 { 3535 .procname = "max_size", 3536 .data = &ip_rt_max_size, 3537 .maxlen = sizeof(int), 3538 .mode = 0644, 3539 .proc_handler = proc_dointvec, 3540 }, 3541 { 3542 /* Deprecated. Use gc_min_interval_ms */ 3543 3544 .procname = "gc_min_interval", 3545 .data = &ip_rt_gc_min_interval, 3546 .maxlen = sizeof(int), 3547 .mode = 0644, 3548 .proc_handler = proc_dointvec_jiffies, 3549 }, 3550 { 3551 .procname = "gc_min_interval_ms", 3552 .data = &ip_rt_gc_min_interval, 3553 .maxlen = sizeof(int), 3554 .mode = 0644, 3555 .proc_handler = proc_dointvec_ms_jiffies, 3556 }, 3557 { 3558 .procname = "gc_timeout", 3559 .data = &ip_rt_gc_timeout, 3560 .maxlen = sizeof(int), 3561 .mode = 0644, 3562 .proc_handler = proc_dointvec_jiffies, 3563 }, 3564 { 3565 .procname = "gc_interval", 3566 .data = &ip_rt_gc_interval, 3567 .maxlen = sizeof(int), 3568 .mode = 0644, 3569 .proc_handler = proc_dointvec_jiffies, 3570 }, 3571 { 3572 .procname = "redirect_load", 3573 .data = &ip_rt_redirect_load, 3574 .maxlen = sizeof(int), 3575 .mode = 0644, 3576 .proc_handler = proc_dointvec, 3577 }, 3578 { 3579 .procname = "redirect_number", 3580 .data = &ip_rt_redirect_number, 3581 .maxlen = sizeof(int), 3582 .mode = 0644, 3583 .proc_handler = proc_dointvec, 3584 }, 3585 { 3586 .procname = "redirect_silence", 3587 .data = &ip_rt_redirect_silence, 3588 .maxlen = sizeof(int), 3589 .mode = 0644, 3590 .proc_handler = proc_dointvec, 3591 }, 3592 { 3593 .procname = "error_cost", 3594 .data = &ip_rt_error_cost, 3595 .maxlen = sizeof(int), 3596 .mode = 0644, 3597 .proc_handler = proc_dointvec, 3598 }, 3599 { 3600 .procname = "error_burst", 3601 .data = &ip_rt_error_burst, 3602 .maxlen = sizeof(int), 3603 .mode = 0644, 3604 .proc_handler = proc_dointvec, 3605 }, 3606 { 3607 .procname = "gc_elasticity", 3608 .data = &ip_rt_gc_elasticity, 3609 .maxlen = sizeof(int), 3610 .mode = 0644, 3611 .proc_handler = proc_dointvec, 3612 }, 3613 }; 3614 3615 static const char ipv4_route_flush_procname[] = "flush"; 3616 3617 static struct ctl_table ipv4_route_netns_table[] = { 3618 { 3619 .procname = ipv4_route_flush_procname, 3620 .maxlen = sizeof(int), 3621 .mode = 0200, 3622 .proc_handler = ipv4_sysctl_rtcache_flush, 3623 }, 3624 { 3625 .procname = "min_pmtu", 3626 .data = &init_net.ipv4.ip_rt_min_pmtu, 3627 .maxlen = sizeof(int), 3628 .mode = 0644, 3629 .proc_handler = proc_dointvec_minmax, 3630 .extra1 = &ip_min_valid_pmtu, 3631 }, 3632 { 3633 .procname = "mtu_expires", 3634 .data = &init_net.ipv4.ip_rt_mtu_expires, 3635 .maxlen = sizeof(int), 3636 .mode = 0644, 3637 .proc_handler = proc_dointvec_jiffies, 3638 }, 3639 { 3640 .procname = "min_adv_mss", 3641 .data = &init_net.ipv4.ip_rt_min_advmss, 3642 .maxlen = sizeof(int), 3643 .mode = 0644, 3644 .proc_handler = proc_dointvec, 3645 }, 3646 }; 3647 3648 static __net_init int sysctl_route_net_init(struct net *net) 3649 { 3650 struct ctl_table *tbl; 3651 size_t table_size = ARRAY_SIZE(ipv4_route_netns_table); 3652 3653 tbl = ipv4_route_netns_table; 3654 if (!net_eq(net, &init_net)) { 3655 int i; 3656 3657 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL); 3658 if (!tbl) 3659 goto err_dup; 3660 3661 /* Don't export non-whitelisted sysctls to unprivileged users */ 3662 if (net->user_ns != &init_user_ns) { 3663 if (tbl[0].procname != ipv4_route_flush_procname) 3664 table_size = 0; 3665 } 3666 3667 /* Update the variables to point into the current struct net 3668 * except for the first element flush 3669 */ 3670 for (i = 1; i < table_size; i++) 3671 tbl[i].data += (void *)net - (void *)&init_net; 3672 } 3673 tbl[0].extra1 = net; 3674 3675 net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route", 3676 tbl, table_size); 3677 if (!net->ipv4.route_hdr) 3678 goto err_reg; 3679 return 0; 3680 3681 err_reg: 3682 if (tbl != ipv4_route_netns_table) 3683 kfree(tbl); 3684 err_dup: 3685 return -ENOMEM; 3686 } 3687 3688 static __net_exit void sysctl_route_net_exit(struct net *net) 3689 { 3690 const struct ctl_table *tbl; 3691 3692 tbl = net->ipv4.route_hdr->ctl_table_arg; 3693 unregister_net_sysctl_table(net->ipv4.route_hdr); 3694 BUG_ON(tbl == ipv4_route_netns_table); 3695 kfree(tbl); 3696 } 3697 3698 static __net_initdata struct pernet_operations sysctl_route_ops = { 3699 .init = sysctl_route_net_init, 3700 .exit = sysctl_route_net_exit, 3701 }; 3702 #endif 3703 3704 static __net_init int netns_ip_rt_init(struct net *net) 3705 { 3706 /* Set default value for namespaceified sysctls */ 3707 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU; 3708 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES; 3709 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS; 3710 return 0; 3711 } 3712 3713 static struct pernet_operations __net_initdata ip_rt_ops = { 3714 .init = netns_ip_rt_init, 3715 }; 3716 3717 static __net_init int rt_genid_init(struct net *net) 3718 { 3719 atomic_set(&net->ipv4.rt_genid, 0); 3720 atomic_set(&net->fnhe_genid, 0); 3721 atomic_set(&net->ipv4.dev_addr_genid, get_random_u32()); 3722 return 0; 3723 } 3724 3725 static __net_initdata struct pernet_operations rt_genid_ops = { 3726 .init = rt_genid_init, 3727 }; 3728 3729 static int __net_init ipv4_inetpeer_init(struct net *net) 3730 { 3731 struct inet_peer_base *bp = kmalloc_obj(*bp); 3732 3733 if (!bp) 3734 return -ENOMEM; 3735 inet_peer_base_init(bp); 3736 net->ipv4.peers = bp; 3737 return 0; 3738 } 3739 3740 static void __net_exit ipv4_inetpeer_exit(struct net *net) 3741 { 3742 struct inet_peer_base *bp = net->ipv4.peers; 3743 3744 net->ipv4.peers = NULL; 3745 inetpeer_invalidate_tree(bp); 3746 kfree(bp); 3747 } 3748 3749 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = { 3750 .init = ipv4_inetpeer_init, 3751 .exit = ipv4_inetpeer_exit, 3752 }; 3753 3754 #ifdef CONFIG_IP_ROUTE_CLASSID 3755 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 3756 #endif /* CONFIG_IP_ROUTE_CLASSID */ 3757 3758 static const struct rtnl_msg_handler ip_rt_rtnl_msg_handlers[] __initconst = { 3759 {.protocol = PF_INET, .msgtype = RTM_GETROUTE, 3760 .doit = inet_rtm_getroute, .flags = RTNL_FLAG_DOIT_UNLOCKED}, 3761 }; 3762 3763 int __init ip_rt_init(void) 3764 { 3765 void *idents_hash; 3766 int cpu; 3767 3768 /* For modern hosts, this will use 2 MB of memory */ 3769 idents_hash = alloc_large_system_hash("IP idents", 3770 sizeof(*ip_idents) + sizeof(*ip_tstamps), 3771 0, 3772 16, /* one bucket per 64 KB */ 3773 HASH_ZERO, 3774 NULL, 3775 &ip_idents_mask, 3776 2048, 3777 256*1024); 3778 3779 ip_idents = idents_hash; 3780 3781 get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents)); 3782 3783 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents); 3784 3785 for_each_possible_cpu(cpu) { 3786 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 3787 3788 INIT_LIST_HEAD(&ul->head); 3789 spin_lock_init(&ul->lock); 3790 } 3791 #ifdef CONFIG_IP_ROUTE_CLASSID 3792 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 3793 if (!ip_rt_acct) 3794 panic("IP: failed to allocate ip_rt_acct\n"); 3795 #endif 3796 3797 ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable, 3798 SLAB_HWCACHE_ALIGN | SLAB_PANIC); 3799 3800 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 3801 3802 if (dst_entries_init(&ipv4_dst_ops) < 0) 3803 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 3804 3805 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 3806 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 3807 3808 ipv4_dst_ops.gc_thresh = ~0; 3809 ip_rt_max_size = INT_MAX; 3810 3811 devinet_init(); 3812 ip_fib_init(); 3813 3814 if (ip_rt_proc_init()) 3815 pr_err("Unable to create route proc files\n"); 3816 #ifdef CONFIG_XFRM 3817 xfrm_init(); 3818 xfrm4_init(); 3819 #endif 3820 rtnl_register_many(ip_rt_rtnl_msg_handlers); 3821 3822 #ifdef CONFIG_SYSCTL 3823 register_pernet_subsys(&sysctl_route_ops); 3824 #endif 3825 register_pernet_subsys(&ip_rt_ops); 3826 register_pernet_subsys(&rt_genid_ops); 3827 register_pernet_subsys(&ipv4_inetpeer_ops); 3828 return 0; 3829 } 3830 3831 #ifdef CONFIG_SYSCTL 3832 /* 3833 * We really need to sanitize the damn ipv4 init order, then all 3834 * this nonsense will go away. 3835 */ 3836 void __init ip_static_sysctl_init(void) 3837 { 3838 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table); 3839 } 3840 #endif 3841