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 /* Other CPUs may still be using @rt, so don't dst_dev_put() it. 577 * rt_flush_dev() handles it if its device is unregistered. 578 * Marking it dead makes ipv4_dst_check() fail for users that cached it. 579 */ 580 static void fnhe_retire_route(struct rtable *rt) 581 { 582 WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_DEAD); 583 rt_add_uncached_list(rt); 584 dst_release(&rt->dst); 585 } 586 587 static void fnhe_flush_routes(struct fib_nh_exception *fnhe) 588 { 589 struct rtable *rt; 590 591 rt = rcu_dereference(fnhe->fnhe_rth_input); 592 if (rt) { 593 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL); 594 fnhe_retire_route(rt); 595 } 596 rt = rcu_dereference(fnhe->fnhe_rth_output); 597 if (rt) { 598 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL); 599 fnhe_retire_route(rt); 600 } 601 } 602 603 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash) 604 { 605 struct fib_nh_exception __rcu **fnhe_p, **oldest_p; 606 struct fib_nh_exception *fnhe, *oldest = NULL; 607 608 for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) { 609 fnhe = rcu_dereference_protected(*fnhe_p, 610 lockdep_is_held(&fnhe_lock)); 611 if (!fnhe) 612 break; 613 if (!oldest || 614 time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) { 615 oldest = fnhe; 616 oldest_p = fnhe_p; 617 } 618 } 619 620 /* Clear oldest->fnhe_daddr to prevent this fnhe from being 621 * rebound with new dsts in rt_bind_exception(). 622 */ 623 oldest->fnhe_daddr = 0; 624 fnhe_flush_routes(oldest); 625 *oldest_p = oldest->fnhe_next; 626 kfree_rcu(oldest, rcu); 627 } 628 629 static u32 fnhe_hashfun(__be32 daddr) 630 { 631 static siphash_aligned_key_t fnhe_hash_key; 632 u64 hval; 633 634 net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key)); 635 hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key); 636 return hash_64(hval, FNHE_HASH_SHIFT); 637 } 638 639 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe) 640 { 641 rt->rt_pmtu = fnhe->fnhe_pmtu; 642 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked; 643 rt->dst.expires = fnhe->fnhe_expires; 644 645 if (fnhe->fnhe_gw) { 646 rt->rt_flags |= RTCF_REDIRECTED; 647 rt->rt_uses_gateway = 1; 648 rt->rt_gw_family = AF_INET; 649 rt->rt_gw4 = fnhe->fnhe_gw; 650 } 651 } 652 653 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr, 654 __be32 gw, u32 pmtu, bool lock, 655 unsigned long expires) 656 { 657 struct fnhe_hash_bucket *hash; 658 struct fib_nh_exception *fnhe; 659 struct rtable *rt; 660 u32 genid, hval; 661 unsigned int i; 662 int depth; 663 664 genid = fnhe_genid(dev_net(nhc->nhc_dev)); 665 hval = fnhe_hashfun(daddr); 666 667 spin_lock_bh(&fnhe_lock); 668 669 hash = rcu_dereference(nhc->nhc_exceptions); 670 if (!hash) { 671 hash = kzalloc_objs(*hash, FNHE_HASH_SIZE, GFP_ATOMIC); 672 if (!hash) 673 goto out_unlock; 674 rcu_assign_pointer(nhc->nhc_exceptions, hash); 675 } 676 677 hash += hval; 678 679 depth = 0; 680 for (fnhe = rcu_dereference(hash->chain); fnhe; 681 fnhe = rcu_dereference(fnhe->fnhe_next)) { 682 if (fnhe->fnhe_daddr == daddr) 683 break; 684 depth++; 685 } 686 687 if (fnhe) { 688 if (fnhe->fnhe_genid != genid) 689 fnhe->fnhe_genid = genid; 690 if (gw) 691 fnhe->fnhe_gw = gw; 692 if (pmtu) { 693 fnhe->fnhe_pmtu = pmtu; 694 fnhe->fnhe_mtu_locked = lock; 695 } 696 fnhe->fnhe_expires = max(1UL, expires); 697 /* Update all cached dsts too */ 698 rt = rcu_dereference(fnhe->fnhe_rth_input); 699 if (rt) 700 fill_route_from_fnhe(rt, fnhe); 701 rt = rcu_dereference(fnhe->fnhe_rth_output); 702 if (rt) 703 fill_route_from_fnhe(rt, fnhe); 704 } else { 705 /* Randomize max depth to avoid some side channels attacks. */ 706 int max_depth = FNHE_RECLAIM_DEPTH + 707 get_random_u32_below(FNHE_RECLAIM_DEPTH); 708 709 while (depth > max_depth) { 710 fnhe_remove_oldest(hash); 711 depth--; 712 } 713 714 fnhe = kzalloc_obj(*fnhe, GFP_ATOMIC); 715 if (!fnhe) 716 goto out_unlock; 717 718 fnhe->fnhe_next = hash->chain; 719 720 fnhe->fnhe_genid = genid; 721 fnhe->fnhe_daddr = daddr; 722 fnhe->fnhe_gw = gw; 723 fnhe->fnhe_pmtu = pmtu; 724 fnhe->fnhe_mtu_locked = lock; 725 fnhe->fnhe_expires = max(1UL, expires); 726 727 rcu_assign_pointer(hash->chain, fnhe); 728 729 /* Exception created; mark the cached routes for the nexthop 730 * stale, so anyone caching it rechecks if this exception 731 * applies to them. 732 */ 733 rt = rcu_dereference(nhc->nhc_rth_input); 734 if (rt) 735 WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL); 736 737 for_each_possible_cpu(i) { 738 struct rtable __rcu **prt; 739 740 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i); 741 rt = rcu_dereference(*prt); 742 if (rt) 743 WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL); 744 } 745 } 746 747 fnhe->fnhe_stamp = jiffies; 748 749 out_unlock: 750 spin_unlock_bh(&fnhe_lock); 751 } 752 753 /* Update the PMTU of an exception when: 754 * - the new MTU of the first hop becomes smaller than the PMTU 755 * - the old MTU was the same as the PMTU, and it limited discovery of 756 * larger MTUs on the path. With that limit raised, we can now 757 * discover larger MTUs 758 * A special case is locked exceptions, for which the PMTU is smaller 759 * than the minimal accepted PMTU: 760 * - if the new MTU is greater than the PMTU, don't make any change 761 * - otherwise, unlock and set PMTU 762 * 763 * fnhe_lock keeps fnhe_pmtu and fnhe_mtu_locked consistent against 764 * update_or_create_fnhe(), which sets both under the same lock. 765 */ 766 void fnhe_update_pmtu(struct fib_nh_exception *fnhe, u32 new, u32 orig) 767 { 768 spin_lock_bh(&fnhe_lock); 769 770 if (fnhe->fnhe_mtu_locked) { 771 if (new <= fnhe->fnhe_pmtu) { 772 fnhe->fnhe_pmtu = new; 773 fnhe->fnhe_mtu_locked = false; 774 } 775 } else if (new < fnhe->fnhe_pmtu || orig == fnhe->fnhe_pmtu) { 776 fnhe->fnhe_pmtu = new; 777 } 778 779 spin_unlock_bh(&fnhe_lock); 780 } 781 782 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4, 783 bool kill_route) 784 { 785 __be32 new_gw = icmp_hdr(skb)->un.gateway; 786 __be32 old_gw = ip_hdr(skb)->saddr; 787 struct net_device *dev = skb->dev; 788 struct in_device *in_dev; 789 struct fib_result res; 790 struct neighbour *n; 791 struct net *net; 792 793 switch (icmp_hdr(skb)->code & 7) { 794 case ICMP_REDIR_NET: 795 case ICMP_REDIR_NETTOS: 796 case ICMP_REDIR_HOST: 797 case ICMP_REDIR_HOSTTOS: 798 break; 799 800 default: 801 return; 802 } 803 804 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw) 805 return; 806 807 in_dev = __in_dev_get_rcu(dev); 808 if (!in_dev) 809 return; 810 811 net = dev_net(dev); 812 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) || 813 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) || 814 ipv4_is_zeronet(new_gw)) 815 goto reject_redirect; 816 817 if (!IN_DEV_SHARED_MEDIA(in_dev)) { 818 if (!inet_addr_onlink(in_dev, new_gw, old_gw)) 819 goto reject_redirect; 820 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) 821 goto reject_redirect; 822 } else { 823 if (inet_addr_type(net, new_gw) != RTN_UNICAST) 824 goto reject_redirect; 825 } 826 827 n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw); 828 if (!n) 829 n = ipv4_neigh_create(rt->dst.dev, &new_gw); 830 if (!IS_ERR(n)) { 831 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) { 832 neigh_event_send(n, NULL); 833 } else { 834 if (fib_lookup(net, fl4, &res, 0) == 0) { 835 struct fib_nh_common *nhc; 836 837 fib_select_path(net, &res, fl4, skb); 838 nhc = FIB_RES_NHC(res); 839 update_or_create_fnhe(nhc, fl4->daddr, new_gw, 840 0, false, 841 jiffies + ip_rt_gc_timeout); 842 } 843 if (kill_route) 844 WRITE_ONCE(rt->dst.obsolete, DST_OBSOLETE_KILL); 845 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n); 846 } 847 neigh_release(n); 848 } 849 return; 850 851 reject_redirect: 852 #ifdef CONFIG_IP_ROUTE_VERBOSE 853 if (IN_DEV_LOG_MARTIANS(in_dev)) { 854 const struct iphdr *iph = (const struct iphdr *) skb->data; 855 __be32 daddr = iph->daddr; 856 __be32 saddr = iph->saddr; 857 858 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n" 859 " Advised path = %pI4 -> %pI4\n", 860 &old_gw, dev->name, &new_gw, 861 &saddr, &daddr); 862 } 863 #endif 864 ; 865 } 866 867 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb) 868 { 869 struct rtable *rt; 870 struct flowi4 fl4; 871 const struct iphdr *iph = (const struct iphdr *) skb->data; 872 struct net *net = dev_net(skb->dev); 873 int oif = skb->dev->ifindex; 874 u8 prot = iph->protocol; 875 u32 mark = skb->mark; 876 __u8 tos = iph->tos; 877 878 rt = dst_rtable(dst); 879 880 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0); 881 __ip_do_redirect(rt, skb, &fl4, true); 882 } 883 884 static void ipv4_negative_advice(struct sock *sk, 885 struct dst_entry *dst) 886 { 887 struct rtable *rt = dst_rtable(dst); 888 889 if ((READ_ONCE(dst->obsolete) > 0) || 890 (rt->rt_flags & RTCF_REDIRECTED) || 891 READ_ONCE(rt->dst.expires)) 892 sk_dst_reset(sk); 893 } 894 895 /* 896 * Algorithm: 897 * 1. The first ip_rt_redirect_number redirects are sent 898 * with exponential backoff, then we stop sending them at all, 899 * assuming that the host ignores our redirects. 900 * 2. If we did not see packets requiring redirects 901 * during ip_rt_redirect_silence, we assume that the host 902 * forgot redirected route and start to send redirects again. 903 * 904 * This algorithm is much cheaper and more intelligent than dumb load limiting 905 * in icmp.c. 906 * 907 * NOTE. Do not forget to inhibit load limiting for redirects (redundant) 908 * and "frag. need" (breaks PMTU discovery) in icmp.c. 909 */ 910 911 void ip_rt_send_redirect(struct sk_buff *skb) 912 { 913 struct rtable *rt = skb_rtable(skb); 914 struct in_device *in_dev; 915 struct net_device *dev; 916 struct inet_peer *peer; 917 int log_martians; 918 struct net *net; 919 int vif; 920 921 rcu_read_lock(); 922 dev = dst_dev_rcu(&rt->dst); 923 in_dev = __in_dev_get_rcu(dev); 924 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) { 925 rcu_read_unlock(); 926 return; 927 } 928 log_martians = IN_DEV_LOG_MARTIANS(in_dev); 929 vif = l3mdev_master_ifindex_rcu(dev); 930 931 net = dev_net_rcu(dev); 932 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif); 933 if (!peer) { 934 rcu_read_unlock(); 935 return; 936 } 937 938 /* No redirected packets during ip_rt_redirect_silence; 939 * reset the algorithm. 940 */ 941 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) { 942 peer->rate_tokens = 0; 943 peer->n_redirects = 0; 944 } 945 946 /* Too many ignored redirects; do not send anything 947 * set dst.rate_last to the last seen redirected packet. 948 */ 949 if (peer->n_redirects >= ip_rt_redirect_number) { 950 peer->rate_last = jiffies; 951 goto out_unlock; 952 } 953 954 /* Check for load limit; set rate_last to the latest sent 955 * redirect. 956 */ 957 if (peer->n_redirects == 0 || 958 time_after(jiffies, 959 (peer->rate_last + 960 (ip_rt_redirect_load << peer->n_redirects)))) { 961 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr); 962 963 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw); 964 peer->rate_last = jiffies; 965 ++peer->n_redirects; 966 if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians && 967 peer->n_redirects == ip_rt_redirect_number) 968 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n", 969 &ip_hdr(skb)->saddr, inet_iif(skb), 970 &ip_hdr(skb)->daddr, &gw); 971 } 972 out_unlock: 973 rcu_read_unlock(); 974 } 975 976 static int ip_error(struct sk_buff *skb) 977 { 978 struct rtable *rt = skb_rtable(skb); 979 struct net_device *dev = skb->dev; 980 struct in_device *in_dev; 981 struct inet_peer *peer; 982 unsigned long now; 983 struct net *net; 984 SKB_DR(reason); 985 bool send; 986 int code; 987 988 if (netif_is_l3_master(skb->dev)) { 989 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif); 990 if (!dev) 991 goto out; 992 } 993 994 in_dev = __in_dev_get_rcu(dev); 995 996 /* IP on this device is disabled. */ 997 if (!in_dev) 998 goto out; 999 1000 net = dev_net(rt->dst.dev); 1001 if (!IN_DEV_FORWARD(in_dev)) { 1002 switch (rt->dst.error) { 1003 case EHOSTUNREACH: 1004 SKB_DR_SET(reason, IP_INADDRERRORS); 1005 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS); 1006 break; 1007 1008 case ENETUNREACH: 1009 SKB_DR_SET(reason, IP_INNOROUTES); 1010 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 1011 break; 1012 } 1013 goto out; 1014 } 1015 1016 switch (rt->dst.error) { 1017 case EINVAL: 1018 default: 1019 goto out; 1020 case EHOSTUNREACH: 1021 code = ICMP_HOST_UNREACH; 1022 break; 1023 case ENETUNREACH: 1024 code = ICMP_NET_UNREACH; 1025 SKB_DR_SET(reason, IP_INNOROUTES); 1026 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 1027 break; 1028 case EACCES: 1029 code = ICMP_PKT_FILTERED; 1030 break; 1031 } 1032 1033 rcu_read_lock(); 1034 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 1035 l3mdev_master_ifindex_rcu(skb->dev)); 1036 send = true; 1037 if (peer) { 1038 now = jiffies; 1039 peer->rate_tokens += now - peer->rate_last; 1040 if (peer->rate_tokens > ip_rt_error_burst) 1041 peer->rate_tokens = ip_rt_error_burst; 1042 peer->rate_last = now; 1043 if (peer->rate_tokens >= ip_rt_error_cost) 1044 peer->rate_tokens -= ip_rt_error_cost; 1045 else 1046 send = false; 1047 } 1048 rcu_read_unlock(); 1049 1050 if (send) 1051 icmp_send(skb, ICMP_DEST_UNREACH, code, 0); 1052 1053 out: kfree_skb_reason(skb, reason); 1054 return 0; 1055 } 1056 1057 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu) 1058 { 1059 struct dst_entry *dst = &rt->dst; 1060 struct fib_result res; 1061 bool lock = false; 1062 struct net *net; 1063 u32 old_mtu; 1064 1065 if (ip_mtu_locked(dst)) 1066 return; 1067 1068 old_mtu = ipv4_mtu(dst); 1069 if (old_mtu < mtu) 1070 return; 1071 1072 rcu_read_lock(); 1073 net = dst_dev_net_rcu(dst); 1074 if (mtu < net->ipv4.ip_rt_min_pmtu) { 1075 lock = true; 1076 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu); 1077 } 1078 1079 if (rt->rt_pmtu == mtu && !lock && 1080 time_before(jiffies, READ_ONCE(dst->expires) - 1081 net->ipv4.ip_rt_mtu_expires / 2)) 1082 goto out; 1083 1084 if (fib_lookup(net, fl4, &res, 0) == 0) { 1085 struct fib_nh_common *nhc; 1086 1087 fib_select_path(net, &res, fl4, NULL); 1088 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1089 if (fib_info_num_path(res.fi) > 1) { 1090 int nhsel; 1091 1092 for (nhsel = 0; nhsel < fib_info_num_path(res.fi); nhsel++) { 1093 nhc = fib_info_nhc(res.fi, nhsel); 1094 if (!nhc) 1095 break; 1096 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1097 jiffies + net->ipv4.ip_rt_mtu_expires); 1098 } 1099 goto out; 1100 } 1101 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 1102 nhc = FIB_RES_NHC(res); 1103 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1104 jiffies + net->ipv4.ip_rt_mtu_expires); 1105 } 1106 out: 1107 rcu_read_unlock(); 1108 } 1109 1110 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 1111 struct sk_buff *skb, u32 mtu, 1112 bool confirm_neigh) 1113 { 1114 struct rtable *rt = dst_rtable(dst); 1115 struct flowi4 fl4; 1116 1117 ip_rt_build_flow_key(&fl4, sk, skb); 1118 1119 /* Don't make lookup fail for bridged encapsulations */ 1120 if (skb && netif_is_any_bridge_port(skb->dev)) 1121 fl4.flowi4_oif = 0; 1122 1123 __ip_rt_update_pmtu(rt, &fl4, mtu); 1124 } 1125 1126 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, 1127 int oif, u8 protocol) 1128 { 1129 const struct iphdr *iph = (const struct iphdr *)skb->data; 1130 struct flowi4 fl4; 1131 struct rtable *rt; 1132 u32 mark = IP4_REPLY_MARK(net, skb->mark); 1133 1134 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark, 1135 0); 1136 rt = __ip_route_output_key(net, &fl4); 1137 if (!IS_ERR(rt)) { 1138 __ip_rt_update_pmtu(rt, &fl4, mtu); 1139 ip_rt_put(rt); 1140 } 1141 } 1142 EXPORT_SYMBOL_GPL(ipv4_update_pmtu); 1143 1144 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1145 { 1146 const struct iphdr *iph = (const struct iphdr *)skb->data; 1147 struct flowi4 fl4; 1148 struct rtable *rt; 1149 1150 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0); 1151 1152 if (!fl4.flowi4_mark) 1153 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark); 1154 1155 rt = __ip_route_output_key(sock_net(sk), &fl4); 1156 if (!IS_ERR(rt)) { 1157 __ip_rt_update_pmtu(rt, &fl4, mtu); 1158 ip_rt_put(rt); 1159 } 1160 } 1161 1162 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1163 { 1164 const struct iphdr *iph = (const struct iphdr *)skb->data; 1165 struct flowi4 fl4; 1166 struct rtable *rt; 1167 struct dst_entry *odst = NULL; 1168 bool new = false; 1169 struct net *net = sock_net(sk); 1170 1171 bh_lock_sock(sk); 1172 1173 if (!ip_sk_accept_pmtu(sk)) 1174 goto out; 1175 1176 odst = sk_dst_get(sk); 1177 1178 if (sock_owned_by_user(sk) || !odst) { 1179 __ipv4_sk_update_pmtu(skb, sk, mtu); 1180 goto out; 1181 } 1182 1183 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1184 1185 rt = dst_rtable(odst); 1186 if (READ_ONCE(odst->obsolete) && !odst->ops->check(odst, 0)) { 1187 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1188 if (IS_ERR(rt)) 1189 goto out; 1190 1191 new = true; 1192 } 1193 1194 __ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu); 1195 1196 if (!dst_check(&rt->dst, 0)) { 1197 if (new) 1198 dst_release(&rt->dst); 1199 1200 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1201 if (IS_ERR(rt)) 1202 goto out; 1203 1204 new = true; 1205 } 1206 1207 if (new) 1208 sk_dst_set(sk, &rt->dst); 1209 1210 out: 1211 bh_unlock_sock(sk); 1212 dst_release(odst); 1213 } 1214 1215 void ipv4_redirect(struct sk_buff *skb, struct net *net, 1216 int oif, u8 protocol) 1217 { 1218 const struct iphdr *iph = (const struct iphdr *)skb->data; 1219 struct flowi4 fl4; 1220 struct rtable *rt; 1221 1222 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0); 1223 rt = __ip_route_output_key(net, &fl4); 1224 if (!IS_ERR(rt)) { 1225 __ip_do_redirect(rt, skb, &fl4, false); 1226 ip_rt_put(rt); 1227 } 1228 } 1229 EXPORT_SYMBOL_GPL(ipv4_redirect); 1230 1231 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk) 1232 { 1233 const struct iphdr *iph = (const struct iphdr *)skb->data; 1234 struct flowi4 fl4; 1235 struct rtable *rt; 1236 struct net *net = sock_net(sk); 1237 1238 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1239 rt = __ip_route_output_key(net, &fl4); 1240 if (!IS_ERR(rt)) { 1241 __ip_do_redirect(rt, skb, &fl4, false); 1242 ip_rt_put(rt); 1243 } 1244 } 1245 1246 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst, 1247 u32 cookie) 1248 { 1249 struct rtable *rt = dst_rtable(dst); 1250 1251 /* All IPV4 dsts are created with ->obsolete set to the value 1252 * DST_OBSOLETE_FORCE_CHK which forces validation calls down 1253 * into this function always. 1254 * 1255 * When a PMTU/redirect information update invalidates a route, 1256 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or 1257 * DST_OBSOLETE_DEAD. 1258 */ 1259 if (READ_ONCE(dst->obsolete) != DST_OBSOLETE_FORCE_CHK || 1260 rt_is_expired(rt)) 1261 return NULL; 1262 return dst; 1263 } 1264 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check); 1265 1266 static void ipv4_send_dest_unreach(struct sk_buff *skb) 1267 { 1268 struct inet_skb_parm parm; 1269 struct net_device *dev; 1270 int res; 1271 1272 /* Recompile ip options since IPCB may not be valid anymore. 1273 * Also check we have a reasonable ipv4 header. 1274 */ 1275 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) || 1276 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5) 1277 return; 1278 1279 memset(&parm, 0, sizeof(parm)); 1280 if (ip_hdr(skb)->ihl > 5) { 1281 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4)) 1282 return; 1283 parm.opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr); 1284 1285 rcu_read_lock(); 1286 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev; 1287 res = __ip_options_compile(dev_net(dev), &parm.opt, skb, NULL); 1288 rcu_read_unlock(); 1289 1290 if (res) 1291 return; 1292 } 1293 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &parm); 1294 } 1295 1296 static void ipv4_link_failure(struct sk_buff *skb) 1297 { 1298 struct rtable *rt; 1299 1300 ipv4_send_dest_unreach(skb); 1301 1302 rt = skb_rtable(skb); 1303 if (rt) 1304 dst_set_expires(&rt->dst, 0); 1305 } 1306 1307 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb) 1308 { 1309 pr_debug("%s: %pI4 -> %pI4, %s\n", 1310 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1311 skb->dev ? skb->dev->name : "?"); 1312 kfree_skb(skb); 1313 WARN_ON_ONCE(1); 1314 return 0; 1315 } 1316 1317 /* 1318 * We do not cache source address of outgoing interface, 1319 * because it is used only by IP RR, TS and SRR options, 1320 * so that it out of fast path. 1321 * 1322 * BTW remember: "addr" is allowed to be not aligned 1323 * in IP options! 1324 */ 1325 1326 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt) 1327 { 1328 __be32 src; 1329 1330 rcu_read_lock(); 1331 if (rt_is_output_route(rt)) { 1332 src = ip_hdr(skb)->saddr; 1333 } else { 1334 struct net_device *dev = dst_dev_rcu(&rt->dst); 1335 struct net *net = dev_net_rcu(dev); 1336 struct iphdr *iph = ip_hdr(skb); 1337 struct fib_result res; 1338 struct flowi4 fl4 = { 1339 .daddr = iph->daddr, 1340 .saddr = iph->saddr, 1341 .flowi4_dscp = ip4h_dscp(iph), 1342 .flowi4_oif = dev->ifindex, 1343 .flowi4_iif = skb->dev->ifindex, 1344 .flowi4_mark = skb->mark, 1345 }; 1346 1347 if (fib_lookup(net, &fl4, &res, 0) == 0) 1348 src = fib_result_prefsrc(net, &res); 1349 else 1350 src = inet_select_addr(dev, 1351 rt_nexthop(rt, iph->daddr), 1352 RT_SCOPE_UNIVERSE); 1353 } 1354 rcu_read_unlock(); 1355 1356 memcpy(addr, &src, 4); 1357 } 1358 1359 #ifdef CONFIG_IP_ROUTE_CLASSID 1360 static void set_class_tag(struct rtable *rt, u32 tag) 1361 { 1362 if (!(rt->dst.tclassid & 0xFFFF)) 1363 rt->dst.tclassid |= tag & 0xFFFF; 1364 if (!(rt->dst.tclassid & 0xFFFF0000)) 1365 rt->dst.tclassid |= tag & 0xFFFF0000; 1366 } 1367 #endif 1368 1369 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1370 { 1371 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr); 1372 unsigned int advmss; 1373 struct net *net; 1374 1375 rcu_read_lock(); 1376 net = dst_dev_net_rcu(dst); 1377 advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size, 1378 net->ipv4.ip_rt_min_advmss); 1379 rcu_read_unlock(); 1380 1381 return min(advmss, IPV4_MAX_PMTU - header_size); 1382 } 1383 1384 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst) 1385 { 1386 return ip_dst_mtu_maybe_forward(dst, false); 1387 } 1388 EXPORT_INDIRECT_CALLABLE(ipv4_mtu); 1389 1390 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr) 1391 { 1392 struct fnhe_hash_bucket *hash; 1393 struct fib_nh_exception *fnhe, __rcu **fnhe_p; 1394 u32 hval = fnhe_hashfun(daddr); 1395 1396 spin_lock_bh(&fnhe_lock); 1397 1398 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1399 lockdep_is_held(&fnhe_lock)); 1400 hash += hval; 1401 1402 fnhe_p = &hash->chain; 1403 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock)); 1404 while (fnhe) { 1405 if (fnhe->fnhe_daddr == daddr) { 1406 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected( 1407 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock))); 1408 /* set fnhe_daddr to 0 to ensure it won't bind with 1409 * new dsts in rt_bind_exception(). 1410 */ 1411 fnhe->fnhe_daddr = 0; 1412 fnhe_flush_routes(fnhe); 1413 kfree_rcu(fnhe, rcu); 1414 break; 1415 } 1416 fnhe_p = &fnhe->fnhe_next; 1417 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1418 lockdep_is_held(&fnhe_lock)); 1419 } 1420 1421 spin_unlock_bh(&fnhe_lock); 1422 } 1423 1424 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc, 1425 __be32 daddr) 1426 { 1427 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions); 1428 struct fib_nh_exception *fnhe; 1429 u32 hval; 1430 1431 if (!hash) 1432 return NULL; 1433 1434 hval = fnhe_hashfun(daddr); 1435 1436 for (fnhe = rcu_dereference(hash[hval].chain); fnhe; 1437 fnhe = rcu_dereference(fnhe->fnhe_next)) { 1438 if (fnhe->fnhe_daddr == daddr) { 1439 if (fnhe->fnhe_expires && 1440 time_after(jiffies, fnhe->fnhe_expires)) { 1441 ip_del_fnhe(nhc, daddr); 1442 break; 1443 } 1444 return fnhe; 1445 } 1446 } 1447 return NULL; 1448 } 1449 1450 /* MTU selection: 1451 * 1. mtu on route is locked - use it 1452 * 2. mtu from nexthop exception 1453 * 3. mtu from egress device 1454 */ 1455 1456 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr) 1457 { 1458 struct fib_nh_common *nhc = res->nhc; 1459 struct net_device *dev = nhc->nhc_dev; 1460 struct fib_info *fi = res->fi; 1461 u32 mtu = 0; 1462 1463 if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) || 1464 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU)) 1465 mtu = fi->fib_mtu; 1466 1467 if (likely(!mtu)) { 1468 struct fib_nh_exception *fnhe; 1469 1470 fnhe = find_exception(nhc, daddr); 1471 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires)) 1472 mtu = fnhe->fnhe_pmtu; 1473 } 1474 1475 if (likely(!mtu)) 1476 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU); 1477 1478 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu); 1479 } 1480 1481 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, 1482 __be32 daddr, const bool do_cache) 1483 { 1484 bool ret = false; 1485 1486 spin_lock_bh(&fnhe_lock); 1487 1488 if (daddr == fnhe->fnhe_daddr) { 1489 struct rtable __rcu **porig; 1490 struct rtable *orig; 1491 int genid = fnhe_genid(dev_net(rt->dst.dev)); 1492 1493 if (rt_is_input_route(rt)) 1494 porig = &fnhe->fnhe_rth_input; 1495 else 1496 porig = &fnhe->fnhe_rth_output; 1497 orig = rcu_dereference(*porig); 1498 1499 if (fnhe->fnhe_genid != genid) { 1500 fnhe->fnhe_genid = genid; 1501 fnhe->fnhe_gw = 0; 1502 fnhe->fnhe_pmtu = 0; 1503 fnhe->fnhe_expires = 0; 1504 fnhe->fnhe_mtu_locked = false; 1505 fnhe_flush_routes(fnhe); 1506 orig = NULL; 1507 } 1508 fill_route_from_fnhe(rt, fnhe); 1509 if (!rt->rt_gw4) { 1510 rt->rt_gw4 = daddr; 1511 rt->rt_gw_family = AF_INET; 1512 } 1513 1514 if (do_cache) { 1515 dst_hold(&rt->dst); 1516 rcu_assign_pointer(*porig, rt); 1517 if (orig) 1518 fnhe_retire_route(orig); 1519 ret = true; 1520 } 1521 1522 fnhe->fnhe_stamp = jiffies; 1523 } 1524 spin_unlock_bh(&fnhe_lock); 1525 1526 return ret; 1527 } 1528 1529 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt) 1530 { 1531 struct rtable *orig, *prev, **p; 1532 bool ret = true; 1533 1534 if (rt_is_input_route(rt)) { 1535 p = (struct rtable **)&nhc->nhc_rth_input; 1536 } else { 1537 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 1538 } 1539 orig = *p; 1540 1541 /* hold dst before doing cmpxchg() to avoid race condition 1542 * on this dst 1543 */ 1544 dst_hold(&rt->dst); 1545 prev = cmpxchg(p, orig, rt); 1546 if (prev == orig) { 1547 if (orig) { 1548 rt_add_uncached_list(orig); 1549 dst_release(&orig->dst); 1550 } 1551 } else { 1552 dst_release(&rt->dst); 1553 ret = false; 1554 } 1555 1556 return ret; 1557 } 1558 1559 struct uncached_list { 1560 spinlock_t lock; 1561 struct list_head head; 1562 }; 1563 1564 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list); 1565 1566 void rt_add_uncached_list(struct rtable *rt) 1567 { 1568 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list); 1569 1570 rt->dst.rt_uncached_list = ul; 1571 1572 spin_lock_bh(&ul->lock); 1573 list_add_tail(&rt->dst.rt_uncached, &ul->head); 1574 spin_unlock_bh(&ul->lock); 1575 } 1576 1577 void rt_del_uncached_list(struct rtable *rt) 1578 { 1579 struct uncached_list *ul = rt->dst.rt_uncached_list; 1580 1581 if (ul) { 1582 spin_lock_bh(&ul->lock); 1583 list_del_init(&rt->dst.rt_uncached); 1584 spin_unlock_bh(&ul->lock); 1585 } 1586 } 1587 1588 static void ipv4_dst_destroy(struct dst_entry *dst) 1589 { 1590 ip_dst_metrics_put(dst); 1591 rt_del_uncached_list(dst_rtable(dst)); 1592 } 1593 1594 void rt_flush_dev(struct net_device *dev) 1595 { 1596 struct rtable *rt, *safe; 1597 int cpu; 1598 1599 if (dev && dev->dismantle) 1600 return; 1601 1602 for_each_possible_cpu(cpu) { 1603 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 1604 1605 if (list_empty(&ul->head)) 1606 continue; 1607 1608 spin_lock_bh(&ul->lock); 1609 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) { 1610 struct net_device *rt_dev = rt->dst.dev; 1611 1612 if (dev ? rt_dev != dev : 1613 READ_ONCE(rt_dev->reg_state) != NETREG_UNREGISTERED) 1614 continue; 1615 1616 rcu_assign_pointer(rt->dst.dev_rcu, blackhole_netdev); 1617 netdev_ref_replace(rt_dev, blackhole_netdev, 1618 &rt->dst.dev_tracker, GFP_ATOMIC); 1619 list_del_init(&rt->dst.rt_uncached); 1620 } 1621 spin_unlock_bh(&ul->lock); 1622 } 1623 } 1624 1625 static bool rt_cache_valid(const struct rtable *rt) 1626 { 1627 return rt && 1628 READ_ONCE(rt->dst.obsolete) == DST_OBSOLETE_FORCE_CHK && 1629 !rt_is_expired(rt); 1630 } 1631 1632 static void rt_set_nexthop(struct rtable *rt, __be32 daddr, 1633 const struct fib_result *res, 1634 struct fib_nh_exception *fnhe, 1635 struct fib_info *fi, u16 type, u32 itag, 1636 const bool do_cache) 1637 { 1638 bool cached = false; 1639 1640 if (fi) { 1641 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1642 1643 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) { 1644 rt->rt_uses_gateway = 1; 1645 rt->rt_gw_family = nhc->nhc_gw_family; 1646 /* only INET and INET6 are supported */ 1647 if (likely(nhc->nhc_gw_family == AF_INET)) 1648 rt->rt_gw4 = nhc->nhc_gw.ipv4; 1649 else 1650 rt->rt_gw6 = nhc->nhc_gw.ipv6; 1651 } 1652 1653 ip_dst_init_metrics(&rt->dst, fi->fib_metrics); 1654 1655 #ifdef CONFIG_IP_ROUTE_CLASSID 1656 if (nhc->nhc_family == AF_INET) { 1657 struct fib_nh *nh; 1658 1659 nh = container_of(nhc, struct fib_nh, nh_common); 1660 rt->dst.tclassid = nh->nh_tclassid; 1661 } 1662 #endif 1663 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 1664 if (unlikely(fnhe)) 1665 cached = rt_bind_exception(rt, fnhe, daddr, do_cache); 1666 else if (do_cache) 1667 cached = rt_cache_route(nhc, rt); 1668 if (unlikely(!cached)) { 1669 /* Routes we intend to cache in nexthop exception or 1670 * FIB nexthop have the DST_NOCACHE bit clear. 1671 * However, if we are unsuccessful at storing this 1672 * route into the cache we really need to set it. 1673 */ 1674 if (!rt->rt_gw4) { 1675 rt->rt_gw_family = AF_INET; 1676 rt->rt_gw4 = daddr; 1677 } 1678 rt_add_uncached_list(rt); 1679 } 1680 } else 1681 rt_add_uncached_list(rt); 1682 1683 #ifdef CONFIG_IP_ROUTE_CLASSID 1684 #ifdef CONFIG_IP_MULTIPLE_TABLES 1685 set_class_tag(rt, res->tclassid); 1686 #endif 1687 set_class_tag(rt, itag); 1688 #endif 1689 } 1690 1691 struct rtable *rt_dst_alloc(struct net_device *dev, 1692 unsigned int flags, u16 type, 1693 bool noxfrm) 1694 { 1695 struct rtable *rt; 1696 1697 rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK, 1698 (noxfrm ? DST_NOXFRM : 0)); 1699 1700 if (rt) { 1701 rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1702 rt->rt_flags = flags; 1703 rt->rt_type = type; 1704 rt->rt_is_input = 0; 1705 rt->rt_iif = 0; 1706 rt->rt_pmtu = 0; 1707 rt->rt_mtu_locked = 0; 1708 rt->rt_uses_gateway = 0; 1709 rt->rt_gw_family = 0; 1710 rt->rt_gw4 = 0; 1711 1712 rt->dst.output = ip_output; 1713 if (flags & RTCF_LOCAL) 1714 rt->dst.input = ip_local_deliver; 1715 } 1716 1717 return rt; 1718 } 1719 EXPORT_SYMBOL(rt_dst_alloc); 1720 1721 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt) 1722 { 1723 struct rtable *new_rt; 1724 1725 new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK, 1726 rt->dst.flags); 1727 1728 if (new_rt) { 1729 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1730 new_rt->rt_flags = rt->rt_flags; 1731 new_rt->rt_type = rt->rt_type; 1732 new_rt->rt_is_input = rt->rt_is_input; 1733 new_rt->rt_iif = rt->rt_iif; 1734 new_rt->rt_pmtu = rt->rt_pmtu; 1735 new_rt->rt_mtu_locked = rt->rt_mtu_locked; 1736 new_rt->rt_gw_family = rt->rt_gw_family; 1737 if (rt->rt_gw_family == AF_INET) 1738 new_rt->rt_gw4 = rt->rt_gw4; 1739 else if (rt->rt_gw_family == AF_INET6) 1740 new_rt->rt_gw6 = rt->rt_gw6; 1741 1742 new_rt->dst.input = READ_ONCE(rt->dst.input); 1743 new_rt->dst.output = READ_ONCE(rt->dst.output); 1744 new_rt->dst.error = rt->dst.error; 1745 new_rt->dst.lastuse = jiffies; 1746 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate); 1747 } 1748 return new_rt; 1749 } 1750 1751 /* called in rcu_read_lock() section */ 1752 enum skb_drop_reason 1753 ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1754 dscp_t dscp, struct net_device *dev, 1755 struct in_device *in_dev, u32 *itag) 1756 { 1757 enum skb_drop_reason reason; 1758 1759 /* Primary sanity checks. */ 1760 if (!in_dev) 1761 return SKB_DROP_REASON_NOT_SPECIFIED; 1762 1763 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 1764 return SKB_DROP_REASON_IP_INVALID_SOURCE; 1765 1766 if (skb->protocol != htons(ETH_P_IP)) 1767 return SKB_DROP_REASON_INVALID_PROTO; 1768 1769 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev)) 1770 return SKB_DROP_REASON_IP_LOCALNET; 1771 1772 if (ipv4_is_zeronet(saddr)) { 1773 if (!ipv4_is_local_multicast(daddr) && 1774 ip_hdr(skb)->protocol != IPPROTO_IGMP) 1775 return SKB_DROP_REASON_IP_INVALID_SOURCE; 1776 } else { 1777 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0, 1778 dev, in_dev, itag); 1779 if (reason) 1780 return reason; 1781 } 1782 return SKB_NOT_DROPPED_YET; 1783 } 1784 1785 /* called in rcu_read_lock() section */ 1786 static enum skb_drop_reason 1787 ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1788 dscp_t dscp, struct net_device *dev, int our) 1789 { 1790 struct in_device *in_dev = __in_dev_get_rcu(dev); 1791 unsigned int flags = RTCF_MULTICAST; 1792 enum skb_drop_reason reason; 1793 struct rtable *rth; 1794 u32 itag = 0; 1795 1796 reason = ip_mc_validate_source(skb, daddr, saddr, dscp, dev, in_dev, 1797 &itag); 1798 if (reason) 1799 return reason; 1800 1801 if (our) 1802 flags |= RTCF_LOCAL; 1803 1804 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 1805 IPCB(skb)->flags |= IPSKB_NOPOLICY; 1806 1807 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST, 1808 false); 1809 if (!rth) 1810 return SKB_DROP_REASON_NOMEM; 1811 1812 #ifdef CONFIG_IP_ROUTE_CLASSID 1813 rth->dst.tclassid = itag; 1814 #endif 1815 rth->dst.output = ip_rt_bug; 1816 rth->rt_is_input= 1; 1817 1818 #ifdef CONFIG_IP_MROUTE 1819 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1820 rth->dst.input = ip_mr_input; 1821 #endif 1822 RT_CACHE_STAT_INC(in_slow_mc); 1823 1824 skb_dst_drop(skb); 1825 skb_dst_set(skb, &rth->dst); 1826 return SKB_NOT_DROPPED_YET; 1827 } 1828 1829 1830 static void ip_handle_martian_source(struct net_device *dev, 1831 struct in_device *in_dev, 1832 struct sk_buff *skb, 1833 __be32 daddr, 1834 __be32 saddr) 1835 { 1836 RT_CACHE_STAT_INC(in_martian_src); 1837 #ifdef CONFIG_IP_ROUTE_VERBOSE 1838 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1839 /* 1840 * RFC1812 recommendation, if source is martian, 1841 * the only hint is MAC header. 1842 */ 1843 pr_warn("martian source (src=%pI4, dst=%pI4, dev=%s)\n", 1844 &saddr, &daddr, dev->name); 1845 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1846 print_hex_dump(KERN_WARNING, "ll header: ", 1847 DUMP_PREFIX_OFFSET, 16, 1, 1848 skb_mac_header(skb), 1849 dev->hard_header_len, false); 1850 } 1851 } 1852 #endif 1853 } 1854 1855 /* called in rcu_read_lock() section */ 1856 static enum skb_drop_reason 1857 __mkroute_input(struct sk_buff *skb, const struct fib_result *res, 1858 struct in_device *in_dev, __be32 daddr, 1859 __be32 saddr, dscp_t dscp) 1860 { 1861 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 1862 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1863 struct net_device *dev = nhc->nhc_dev; 1864 struct fib_nh_exception *fnhe; 1865 struct rtable *rth; 1866 int err; 1867 struct in_device *out_dev; 1868 bool do_cache; 1869 u32 itag = 0; 1870 1871 /* get a working reference to the output device */ 1872 out_dev = __in_dev_get_rcu(dev); 1873 if (!out_dev) { 1874 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n"); 1875 return reason; 1876 } 1877 1878 err = fib_validate_source(skb, saddr, daddr, dscp, FIB_RES_OIF(*res), 1879 in_dev->dev, in_dev, &itag); 1880 if (err < 0) { 1881 reason = -err; 1882 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1883 saddr); 1884 1885 goto cleanup; 1886 } 1887 1888 do_cache = res->fi && !itag; 1889 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) && 1890 skb->protocol == htons(ETH_P_IP)) { 1891 __be32 gw; 1892 1893 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0; 1894 if (IN_DEV_SHARED_MEDIA(out_dev) || 1895 inet_addr_onlink(out_dev, saddr, gw)) 1896 IPCB(skb)->flags |= IPSKB_DOREDIRECT; 1897 } 1898 1899 if (skb->protocol != htons(ETH_P_IP)) { 1900 /* Not IP (i.e. ARP). Do not create route, if it is 1901 * invalid for proxy arp. DNAT routes are always valid. 1902 * 1903 * Proxy arp feature have been extended to allow, ARP 1904 * replies back to the same interface, to support 1905 * Private VLAN switch technologies. See arp.c. 1906 */ 1907 if (out_dev == in_dev && 1908 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 1909 reason = SKB_DROP_REASON_ARP_PVLAN_DISABLE; 1910 goto cleanup; 1911 } 1912 } 1913 1914 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 1915 IPCB(skb)->flags |= IPSKB_NOPOLICY; 1916 1917 fnhe = find_exception(nhc, daddr); 1918 if (do_cache) { 1919 if (fnhe) 1920 rth = rcu_dereference(fnhe->fnhe_rth_input); 1921 else 1922 rth = rcu_dereference(nhc->nhc_rth_input); 1923 if (rt_cache_valid(rth)) { 1924 skb_dst_set_noref(skb, &rth->dst); 1925 goto out; 1926 } 1927 } 1928 1929 rth = rt_dst_alloc(out_dev->dev, 0, res->type, 1930 IN_DEV_ORCONF(out_dev, NOXFRM)); 1931 if (!rth) { 1932 reason = SKB_DROP_REASON_NOMEM; 1933 goto cleanup; 1934 } 1935 1936 rth->rt_is_input = 1; 1937 RT_CACHE_STAT_INC(in_slow_tot); 1938 1939 rth->dst.input = ip_forward; 1940 1941 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag, 1942 do_cache); 1943 lwtunnel_set_redirect(&rth->dst); 1944 skb_dst_set(skb, &rth->dst); 1945 out: 1946 reason = SKB_NOT_DROPPED_YET; 1947 cleanup: 1948 return reason; 1949 } 1950 1951 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1952 /* To make ICMP packets follow the right flow, the multipath hash is 1953 * calculated from the inner IP addresses. 1954 */ 1955 static void ip_multipath_l3_keys(const struct sk_buff *skb, 1956 struct flow_keys *hash_keys) 1957 { 1958 const struct iphdr *outer_iph = ip_hdr(skb); 1959 const struct iphdr *key_iph = outer_iph; 1960 const struct iphdr *inner_iph; 1961 const struct icmphdr *icmph; 1962 struct iphdr _inner_iph; 1963 struct icmphdr _icmph; 1964 1965 if (likely(outer_iph->protocol != IPPROTO_ICMP)) 1966 goto out; 1967 1968 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0)) 1969 goto out; 1970 1971 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph), 1972 &_icmph); 1973 if (!icmph) 1974 goto out; 1975 1976 if (!icmp_is_err(icmph->type)) 1977 goto out; 1978 1979 inner_iph = skb_header_pointer(skb, 1980 outer_iph->ihl * 4 + sizeof(_icmph), 1981 sizeof(_inner_iph), &_inner_iph); 1982 if (!inner_iph) 1983 goto out; 1984 1985 key_iph = inner_iph; 1986 out: 1987 hash_keys->addrs.v4addrs.src = key_iph->saddr; 1988 hash_keys->addrs.v4addrs.dst = key_iph->daddr; 1989 } 1990 1991 static u32 fib_multipath_custom_hash_outer(const struct net *net, 1992 const struct sk_buff *skb, 1993 bool *p_has_inner) 1994 { 1995 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 1996 struct flow_keys keys, hash_keys; 1997 1998 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 1999 return 0; 2000 2001 memset(&hash_keys, 0, sizeof(hash_keys)); 2002 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP); 2003 2004 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2005 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 2006 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 2007 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 2008 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 2009 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 2010 hash_keys.basic.ip_proto = keys.basic.ip_proto; 2011 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) 2012 hash_keys.ports.src = keys.ports.src; 2013 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 2014 hash_keys.ports.dst = keys.ports.dst; 2015 2016 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION); 2017 return fib_multipath_hash_from_keys(net, &hash_keys); 2018 } 2019 2020 static u32 fib_multipath_custom_hash_inner(const struct net *net, 2021 const struct sk_buff *skb, 2022 bool has_inner) 2023 { 2024 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 2025 struct flow_keys keys, hash_keys; 2026 2027 /* We assume the packet carries an encapsulation, but if none was 2028 * encountered during dissection of the outer flow, then there is no 2029 * point in calling the flow dissector again. 2030 */ 2031 if (!has_inner) 2032 return 0; 2033 2034 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK)) 2035 return 0; 2036 2037 memset(&hash_keys, 0, sizeof(hash_keys)); 2038 skb_flow_dissect_flow_keys(skb, &keys, 0); 2039 2040 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION)) 2041 return 0; 2042 2043 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 2044 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2045 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 2046 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 2047 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 2048 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 2049 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 2050 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 2051 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 2052 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 2053 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 2054 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 2055 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL) 2056 hash_keys.tags.flow_label = keys.tags.flow_label; 2057 } 2058 2059 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO) 2060 hash_keys.basic.ip_proto = keys.basic.ip_proto; 2061 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT) 2062 hash_keys.ports.src = keys.ports.src; 2063 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT) 2064 hash_keys.ports.dst = keys.ports.dst; 2065 2066 return fib_multipath_hash_from_keys(net, &hash_keys); 2067 } 2068 2069 static u32 fib_multipath_custom_hash_skb(const struct net *net, 2070 const struct sk_buff *skb) 2071 { 2072 u32 mhash, mhash_inner; 2073 bool has_inner = true; 2074 2075 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner); 2076 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner); 2077 2078 return jhash_2words(mhash, mhash_inner, 0); 2079 } 2080 2081 static u32 fib_multipath_custom_hash_fl4(const struct net *net, 2082 const struct flowi4 *fl4) 2083 { 2084 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 2085 struct flow_keys hash_keys; 2086 2087 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 2088 return 0; 2089 2090 memset(&hash_keys, 0, sizeof(hash_keys)); 2091 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2092 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 2093 hash_keys.addrs.v4addrs.src = fl4->saddr; 2094 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 2095 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2096 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 2097 hash_keys.basic.ip_proto = fl4->flowi4_proto; 2098 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) { 2099 if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT) 2100 hash_keys.ports.src = (__force __be16)get_random_u16(); 2101 else 2102 hash_keys.ports.src = fl4->fl4_sport; 2103 } 2104 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 2105 hash_keys.ports.dst = fl4->fl4_dport; 2106 2107 return fib_multipath_hash_from_keys(net, &hash_keys); 2108 } 2109 2110 /* if skb is set it will be used and fl4 can be NULL */ 2111 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4, 2112 const struct sk_buff *skb, struct flow_keys *flkeys) 2113 { 2114 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0; 2115 struct flow_keys hash_keys; 2116 u32 mhash = 0; 2117 2118 switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) { 2119 case 0: 2120 memset(&hash_keys, 0, sizeof(hash_keys)); 2121 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2122 if (skb) { 2123 ip_multipath_l3_keys(skb, &hash_keys); 2124 } else { 2125 hash_keys.addrs.v4addrs.src = fl4->saddr; 2126 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2127 } 2128 mhash = fib_multipath_hash_from_keys(net, &hash_keys); 2129 break; 2130 case 1: 2131 /* skb is currently provided only when forwarding */ 2132 if (skb) { 2133 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 2134 struct flow_keys keys; 2135 2136 /* short-circuit if we already have L4 hash present */ 2137 if (skb->l4_hash) 2138 return skb_get_hash_raw(skb) >> 1; 2139 2140 memset(&hash_keys, 0, sizeof(hash_keys)); 2141 2142 if (!flkeys) { 2143 skb_flow_dissect_flow_keys(skb, &keys, flag); 2144 flkeys = &keys; 2145 } 2146 2147 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2148 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src; 2149 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst; 2150 hash_keys.ports.src = flkeys->ports.src; 2151 hash_keys.ports.dst = flkeys->ports.dst; 2152 hash_keys.basic.ip_proto = flkeys->basic.ip_proto; 2153 } else { 2154 memset(&hash_keys, 0, sizeof(hash_keys)); 2155 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2156 hash_keys.addrs.v4addrs.src = fl4->saddr; 2157 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2158 if (fl4->flowi4_flags & FLOWI_FLAG_ANY_SPORT) 2159 hash_keys.ports.src = (__force __be16)get_random_u16(); 2160 else 2161 hash_keys.ports.src = fl4->fl4_sport; 2162 hash_keys.ports.dst = fl4->fl4_dport; 2163 hash_keys.basic.ip_proto = fl4->flowi4_proto; 2164 } 2165 mhash = fib_multipath_hash_from_keys(net, &hash_keys); 2166 break; 2167 case 2: 2168 memset(&hash_keys, 0, sizeof(hash_keys)); 2169 /* skb is currently provided only when forwarding */ 2170 if (skb) { 2171 struct flow_keys keys; 2172 2173 skb_flow_dissect_flow_keys(skb, &keys, 0); 2174 /* Inner can be v4 or v6 */ 2175 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 2176 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2177 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 2178 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 2179 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 2180 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 2181 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 2182 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 2183 hash_keys.tags.flow_label = keys.tags.flow_label; 2184 hash_keys.basic.ip_proto = keys.basic.ip_proto; 2185 } else { 2186 /* Same as case 0 */ 2187 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2188 ip_multipath_l3_keys(skb, &hash_keys); 2189 } 2190 } else { 2191 /* Same as case 0 */ 2192 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2193 hash_keys.addrs.v4addrs.src = fl4->saddr; 2194 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2195 } 2196 mhash = fib_multipath_hash_from_keys(net, &hash_keys); 2197 break; 2198 case 3: 2199 if (skb) 2200 mhash = fib_multipath_custom_hash_skb(net, skb); 2201 else 2202 mhash = fib_multipath_custom_hash_fl4(net, fl4); 2203 break; 2204 } 2205 2206 if (multipath_hash) 2207 mhash = jhash_2words(mhash, multipath_hash, 0); 2208 2209 return mhash >> 1; 2210 } 2211 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 2212 2213 static enum skb_drop_reason 2214 ip_mkroute_input(struct sk_buff *skb, struct fib_result *res, 2215 struct in_device *in_dev, __be32 daddr, 2216 __be32 saddr, dscp_t dscp, struct flow_keys *hkeys) 2217 { 2218 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2219 if (res->fi && fib_info_num_path(res->fi) > 1) { 2220 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys); 2221 2222 fib_select_multipath(res, h, NULL); 2223 IPCB(skb)->flags |= IPSKB_MULTIPATH; 2224 } 2225 #endif 2226 2227 /* create a routing cache entry */ 2228 return __mkroute_input(skb, res, in_dev, daddr, saddr, dscp); 2229 } 2230 2231 /* Implements all the saddr-related checks as ip_route_input_slow(), 2232 * assuming daddr is valid and the destination is not a local broadcast one. 2233 * Uses the provided hint instead of performing a route lookup. 2234 */ 2235 enum skb_drop_reason 2236 ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2237 dscp_t dscp, struct net_device *dev, 2238 const struct sk_buff *hint) 2239 { 2240 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 2241 struct in_device *in_dev = __in_dev_get_rcu(dev); 2242 struct rtable *rt = skb_rtable(hint); 2243 struct net *net = dev_net(dev); 2244 u32 tag = 0; 2245 2246 if (!in_dev) 2247 return reason; 2248 2249 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) { 2250 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2251 goto martian_source; 2252 } 2253 2254 if (ipv4_is_zeronet(saddr)) { 2255 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2256 goto martian_source; 2257 } 2258 2259 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) { 2260 reason = SKB_DROP_REASON_IP_LOCALNET; 2261 goto martian_source; 2262 } 2263 2264 if (!(rt->rt_flags & RTCF_LOCAL)) 2265 goto skip_validate_source; 2266 2267 reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 0, dev, 2268 in_dev, &tag); 2269 if (reason) 2270 goto martian_source; 2271 2272 skip_validate_source: 2273 skb_dst_copy(skb, hint); 2274 return SKB_NOT_DROPPED_YET; 2275 2276 martian_source: 2277 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2278 return reason; 2279 } 2280 2281 /* get device for dst_alloc with local routes */ 2282 static struct net_device *ip_rt_get_dev(struct net *net, 2283 const struct fib_result *res) 2284 { 2285 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL; 2286 struct net_device *dev = NULL; 2287 2288 if (nhc) 2289 dev = l3mdev_master_dev_rcu(nhc->nhc_dev); 2290 2291 return dev ? : net->loopback_dev; 2292 } 2293 2294 /* 2295 * NOTE. We drop all the packets that has local source 2296 * addresses, because every properly looped back packet 2297 * must have correct destination already attached by output routine. 2298 * Changes in the enforced policies must be applied also to 2299 * ip_route_use_hint(). 2300 * 2301 * Such approach solves two big problems: 2302 * 1. Not simplex devices are handled properly. 2303 * 2. IP spoofing attempts are filtered with 100% of guarantee. 2304 * called with rcu_read_lock() 2305 */ 2306 2307 static enum skb_drop_reason 2308 ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2309 dscp_t dscp, struct net_device *dev, 2310 struct fib_result *res) 2311 { 2312 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 2313 struct in_device *in_dev = __in_dev_get_rcu(dev); 2314 struct flow_keys *flkeys = NULL, _flkeys; 2315 struct net *net = dev_net(dev); 2316 struct ip_tunnel_info *tun_info; 2317 int err = -EINVAL; 2318 unsigned int flags = 0; 2319 u32 itag = 0; 2320 struct rtable *rth; 2321 struct flowi4 fl4; 2322 bool do_cache = true; 2323 2324 /* IP on this device is disabled. */ 2325 2326 if (!in_dev) 2327 goto out; 2328 2329 /* Check for the most weird martians, which can be not detected 2330 * by fib_lookup. 2331 */ 2332 2333 tun_info = skb_tunnel_info(skb); 2334 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX)) 2335 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id; 2336 else 2337 fl4.flowi4_tun_key.tun_id = 0; 2338 skb_dst_drop(skb); 2339 2340 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) { 2341 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2342 goto martian_source; 2343 } 2344 2345 res->fi = NULL; 2346 res->table = NULL; 2347 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 2348 goto brd_input; 2349 2350 /* Accept zero addresses only to limited broadcast; 2351 * I even do not know to fix it or not. Waiting for complains :-) 2352 */ 2353 if (ipv4_is_zeronet(saddr)) { 2354 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2355 goto martian_source; 2356 } 2357 2358 if (ipv4_is_zeronet(daddr)) { 2359 reason = SKB_DROP_REASON_IP_INVALID_DEST; 2360 goto martian_destination; 2361 } 2362 2363 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(), 2364 * and call it once if daddr or/and saddr are loopback addresses 2365 */ 2366 if (ipv4_is_loopback(daddr)) { 2367 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) { 2368 reason = SKB_DROP_REASON_IP_LOCALNET; 2369 goto martian_destination; 2370 } 2371 } else if (ipv4_is_loopback(saddr)) { 2372 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) { 2373 reason = SKB_DROP_REASON_IP_LOCALNET; 2374 goto martian_source; 2375 } 2376 } 2377 2378 /* 2379 * Now we are ready to route packet. 2380 */ 2381 fl4.flowi4_l3mdev = 0; 2382 fl4.flowi4_oif = 0; 2383 fl4.flowi4_iif = dev->ifindex; 2384 fl4.flowi4_mark = skb->mark; 2385 fl4.flowi4_dscp = dscp; 2386 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 2387 fl4.flowi4_flags = 0; 2388 fl4.daddr = daddr; 2389 fl4.saddr = saddr; 2390 fl4.flowi4_uid = sock_net_uid(net, NULL); 2391 fl4.flowi4_multipath_hash = 0; 2392 2393 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) { 2394 flkeys = &_flkeys; 2395 } else { 2396 fl4.flowi4_proto = 0; 2397 fl4.fl4_sport = 0; 2398 fl4.fl4_dport = 0; 2399 } 2400 2401 err = fib_lookup(net, &fl4, res, 0); 2402 if (err != 0) { 2403 if (!IN_DEV_FORWARD(in_dev)) 2404 err = -EHOSTUNREACH; 2405 goto no_route; 2406 } 2407 2408 if (res->type == RTN_BROADCAST) { 2409 if (IN_DEV_BFORWARD(in_dev)) 2410 goto make_route; 2411 /* not do cache if bc_forwarding is enabled */ 2412 if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING)) 2413 do_cache = false; 2414 goto brd_input; 2415 } 2416 2417 err = -EINVAL; 2418 if (res->type == RTN_LOCAL) { 2419 reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 2420 0, dev, in_dev, &itag); 2421 if (reason) 2422 goto martian_source; 2423 goto local_input; 2424 } 2425 2426 if (!IN_DEV_FORWARD(in_dev)) { 2427 err = -EHOSTUNREACH; 2428 goto no_route; 2429 } 2430 if (res->type != RTN_UNICAST) { 2431 reason = SKB_DROP_REASON_IP_INVALID_DEST; 2432 goto martian_destination; 2433 } 2434 2435 make_route: 2436 reason = ip_mkroute_input(skb, res, in_dev, daddr, saddr, dscp, 2437 flkeys); 2438 2439 out: 2440 return reason; 2441 2442 brd_input: 2443 if (skb->protocol != htons(ETH_P_IP)) { 2444 reason = SKB_DROP_REASON_INVALID_PROTO; 2445 goto out; 2446 } 2447 2448 if (!ipv4_is_zeronet(saddr)) { 2449 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0, 2450 dev, in_dev, &itag); 2451 if (reason) 2452 goto martian_source; 2453 } 2454 flags |= RTCF_BROADCAST; 2455 res->type = RTN_BROADCAST; 2456 RT_CACHE_STAT_INC(in_brd); 2457 2458 local_input: 2459 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 2460 IPCB(skb)->flags |= IPSKB_NOPOLICY; 2461 2462 do_cache &= res->fi && !itag; 2463 if (do_cache) { 2464 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2465 2466 rth = rcu_dereference(nhc->nhc_rth_input); 2467 if (rt_cache_valid(rth)) { 2468 skb_dst_set_noref(skb, &rth->dst); 2469 reason = SKB_NOT_DROPPED_YET; 2470 goto out; 2471 } 2472 } 2473 2474 rth = rt_dst_alloc(ip_rt_get_dev(net, res), 2475 flags | RTCF_LOCAL, res->type, false); 2476 if (!rth) 2477 goto e_nobufs; 2478 2479 rth->dst.output= ip_rt_bug; 2480 #ifdef CONFIG_IP_ROUTE_CLASSID 2481 rth->dst.tclassid = itag; 2482 #endif 2483 rth->rt_is_input = 1; 2484 2485 RT_CACHE_STAT_INC(in_slow_tot); 2486 if (res->type == RTN_UNREACHABLE) { 2487 rth->dst.input= ip_error; 2488 rth->dst.error= -err; 2489 rth->rt_flags &= ~RTCF_LOCAL; 2490 } 2491 2492 if (do_cache) { 2493 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2494 2495 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 2496 if (lwtunnel_input_redirect(rth->dst.lwtstate)) { 2497 WARN_ON(rth->dst.input == lwtunnel_input); 2498 rth->dst.lwtstate->orig_input = rth->dst.input; 2499 rth->dst.input = lwtunnel_input; 2500 } 2501 2502 if (unlikely(!rt_cache_route(nhc, rth))) 2503 rt_add_uncached_list(rth); 2504 } else { 2505 rt_add_uncached_list(rth); 2506 } 2507 skb_dst_set(skb, &rth->dst); 2508 reason = SKB_NOT_DROPPED_YET; 2509 goto out; 2510 2511 no_route: 2512 RT_CACHE_STAT_INC(in_no_route); 2513 res->type = RTN_UNREACHABLE; 2514 res->fi = NULL; 2515 res->table = NULL; 2516 goto local_input; 2517 2518 /* 2519 * Do not cache martian addresses: they should be logged (RFC1812) 2520 */ 2521 martian_destination: 2522 RT_CACHE_STAT_INC(in_martian_dst); 2523 #ifdef CONFIG_IP_ROUTE_VERBOSE 2524 if (IN_DEV_LOG_MARTIANS(in_dev)) 2525 net_warn_ratelimited("martian destination (src=%pI4, dst=%pI4, dev=%s)\n", 2526 &saddr, &daddr, dev->name); 2527 #endif 2528 goto out; 2529 2530 e_nobufs: 2531 reason = SKB_DROP_REASON_NOMEM; 2532 goto out; 2533 2534 martian_source: 2535 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2536 goto out; 2537 } 2538 2539 /* called with rcu_read_lock held */ 2540 static enum skb_drop_reason 2541 ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2542 dscp_t dscp, struct net_device *dev, 2543 struct fib_result *res) 2544 { 2545 /* Multicast recognition logic is moved from route cache to here. 2546 * The problem was that too many Ethernet cards have broken/missing 2547 * hardware multicast filters :-( As result the host on multicasting 2548 * network acquires a lot of useless route cache entries, sort of 2549 * SDR messages from all the world. Now we try to get rid of them. 2550 * Really, provided software IP multicast filter is organized 2551 * reasonably (at least, hashed), it does not result in a slowdown 2552 * comparing with route cache reject entries. 2553 * Note, that multicast routers are not affected, because 2554 * route cache entry is created eventually. 2555 */ 2556 if (ipv4_is_multicast(daddr)) { 2557 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 2558 struct in_device *in_dev = __in_dev_get_rcu(dev); 2559 int our = 0; 2560 2561 if (!in_dev) 2562 return reason; 2563 2564 our = ip_check_mc_rcu(in_dev, daddr, saddr, 2565 ip_hdr(skb)->protocol); 2566 2567 /* check l3 master if no match yet */ 2568 if (!our && netif_is_l3_slave(dev)) { 2569 struct in_device *l3_in_dev; 2570 2571 l3_in_dev = __in_dev_get_rcu(skb->dev); 2572 if (l3_in_dev) 2573 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr, 2574 ip_hdr(skb)->protocol); 2575 } 2576 2577 if (our 2578 #ifdef CONFIG_IP_MROUTE 2579 || 2580 (!ipv4_is_local_multicast(daddr) && 2581 IN_DEV_MFORWARD(in_dev)) 2582 #endif 2583 ) { 2584 reason = ip_route_input_mc(skb, daddr, saddr, dscp, 2585 dev, our); 2586 } 2587 return reason; 2588 } 2589 2590 return ip_route_input_slow(skb, daddr, saddr, dscp, dev, res); 2591 } 2592 2593 enum skb_drop_reason ip_route_input_noref(struct sk_buff *skb, __be32 daddr, 2594 __be32 saddr, dscp_t dscp, 2595 struct net_device *dev) 2596 { 2597 enum skb_drop_reason reason; 2598 struct fib_result res; 2599 2600 rcu_read_lock(); 2601 reason = ip_route_input_rcu(skb, daddr, saddr, dscp, dev, &res); 2602 rcu_read_unlock(); 2603 2604 return reason; 2605 } 2606 EXPORT_SYMBOL(ip_route_input_noref); 2607 2608 /* called with rcu_read_lock() */ 2609 static struct rtable *__mkroute_output(const struct fib_result *res, 2610 const struct flowi4 *fl4, int orig_oif, 2611 struct net_device *dev_out, 2612 unsigned int flags) 2613 { 2614 struct fib_info *fi = res->fi; 2615 struct fib_nh_exception *fnhe; 2616 struct in_device *in_dev; 2617 u16 type = res->type; 2618 struct rtable *rth; 2619 bool do_cache; 2620 2621 in_dev = __in_dev_get_rcu(dev_out); 2622 if (!in_dev) 2623 return ERR_PTR(-EINVAL); 2624 2625 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 2626 if (ipv4_is_loopback(fl4->saddr) && 2627 !(dev_out->flags & IFF_LOOPBACK) && 2628 !netif_is_l3_master(dev_out)) 2629 return ERR_PTR(-EINVAL); 2630 2631 if (ipv4_is_lbcast(fl4->daddr)) { 2632 type = RTN_BROADCAST; 2633 2634 /* reset fi to prevent gateway resolution */ 2635 fi = NULL; 2636 } else if (ipv4_is_multicast(fl4->daddr)) { 2637 type = RTN_MULTICAST; 2638 } else if (ipv4_is_zeronet(fl4->daddr)) { 2639 return ERR_PTR(-EINVAL); 2640 } 2641 2642 if (dev_out->flags & IFF_LOOPBACK) 2643 flags |= RTCF_LOCAL; 2644 2645 do_cache = true; 2646 if (type == RTN_BROADCAST) { 2647 flags |= RTCF_BROADCAST | RTCF_LOCAL; 2648 } else if (type == RTN_MULTICAST) { 2649 flags |= RTCF_MULTICAST | RTCF_LOCAL; 2650 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr, 2651 fl4->flowi4_proto)) 2652 flags &= ~RTCF_LOCAL; 2653 else 2654 do_cache = false; 2655 /* If multicast route do not exist use 2656 * default one, but do not gateway in this case. 2657 * Yes, it is hack. 2658 */ 2659 if (fi && res->prefixlen < 4) 2660 fi = NULL; 2661 } else if ((type == RTN_LOCAL) && (orig_oif != 0) && 2662 (orig_oif != dev_out->ifindex)) { 2663 /* For local routes that require a particular output interface 2664 * we do not want to cache the result. Caching the result 2665 * causes incorrect behaviour when there are multiple source 2666 * addresses on the interface, the end result being that if the 2667 * intended recipient is waiting on that interface for the 2668 * packet he won't receive it because it will be delivered on 2669 * the loopback interface and the IP_PKTINFO ipi_ifindex will 2670 * be set to the loopback interface as well. 2671 */ 2672 do_cache = false; 2673 } 2674 2675 fnhe = NULL; 2676 do_cache &= fi != NULL; 2677 if (fi) { 2678 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2679 struct rtable __rcu **prth; 2680 2681 fnhe = find_exception(nhc, fl4->daddr); 2682 if (!do_cache) 2683 goto add; 2684 if (fnhe) { 2685 prth = &fnhe->fnhe_rth_output; 2686 } else { 2687 if (unlikely(fl4->flowi4_flags & 2688 FLOWI_FLAG_KNOWN_NH && 2689 !(nhc->nhc_gw_family && 2690 nhc->nhc_scope == RT_SCOPE_LINK))) { 2691 do_cache = false; 2692 goto add; 2693 } 2694 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 2695 } 2696 rth = rcu_dereference(*prth); 2697 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst)) 2698 return rth; 2699 } 2700 2701 add: 2702 rth = rt_dst_alloc(dev_out, flags, type, 2703 IN_DEV_ORCONF(in_dev, NOXFRM)); 2704 if (!rth) 2705 return ERR_PTR(-ENOBUFS); 2706 2707 rth->rt_iif = orig_oif; 2708 2709 RT_CACHE_STAT_INC(out_slow_tot); 2710 2711 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 2712 if (flags & RTCF_LOCAL && 2713 !(dev_out->flags & IFF_LOOPBACK)) { 2714 rth->dst.output = ip_mc_output; 2715 RT_CACHE_STAT_INC(out_slow_mc); 2716 } 2717 #ifdef CONFIG_IP_MROUTE 2718 if (type == RTN_MULTICAST) { 2719 if (IN_DEV_MFORWARD(in_dev) && 2720 !ipv4_is_local_multicast(fl4->daddr)) { 2721 rth->dst.input = ip_mr_input; 2722 rth->dst.output = ip_mr_output; 2723 } 2724 } 2725 #endif 2726 } 2727 2728 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache); 2729 lwtunnel_set_redirect(&rth->dst); 2730 2731 return rth; 2732 } 2733 2734 /* 2735 * Major route resolver routine. 2736 */ 2737 2738 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4, 2739 const struct sk_buff *skb) 2740 { 2741 struct fib_result res = { 2742 .type = RTN_UNSPEC, 2743 .fi = NULL, 2744 .table = NULL, 2745 .tclassid = 0, 2746 }; 2747 struct rtable *rth; 2748 2749 fl4->flowi4_iif = LOOPBACK_IFINDEX; 2750 2751 rcu_read_lock(); 2752 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb); 2753 rcu_read_unlock(); 2754 2755 return rth; 2756 } 2757 EXPORT_SYMBOL_GPL(ip_route_output_key_hash); 2758 2759 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4, 2760 struct fib_result *res, 2761 const struct sk_buff *skb) 2762 { 2763 struct net_device *dev_out = NULL; 2764 int orig_oif = fl4->flowi4_oif; 2765 unsigned int flags = 0; 2766 struct rtable *rth; 2767 int err; 2768 2769 if (fl4->saddr) { 2770 if (ipv4_is_multicast(fl4->saddr) || 2771 ipv4_is_lbcast(fl4->saddr)) { 2772 rth = ERR_PTR(-EINVAL); 2773 goto out; 2774 } 2775 2776 rth = ERR_PTR(-ENETUNREACH); 2777 2778 /* I removed check for oif == dev_out->oif here. 2779 * It was wrong for two reasons: 2780 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 2781 * is assigned to multiple interfaces. 2782 * 2. Moreover, we are allowed to send packets with saddr 2783 * of another iface. --ANK 2784 */ 2785 2786 if (fl4->flowi4_oif == 0 && 2787 (ipv4_is_multicast(fl4->daddr) || 2788 ipv4_is_lbcast(fl4->daddr))) { 2789 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2790 dev_out = __ip_dev_find(net, fl4->saddr, false); 2791 if (!dev_out) 2792 goto out; 2793 2794 /* Special hack: user can direct multicasts 2795 * and limited broadcast via necessary interface 2796 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 2797 * This hack is not just for fun, it allows 2798 * vic,vat and friends to work. 2799 * They bind socket to loopback, set ttl to zero 2800 * and expect that it will work. 2801 * From the viewpoint of routing cache they are broken, 2802 * because we are not allowed to build multicast path 2803 * with loopback source addr (look, routing cache 2804 * cannot know, that ttl is zero, so that packet 2805 * will not leave this host and route is valid). 2806 * Luckily, this hack is good workaround. 2807 */ 2808 2809 fl4->flowi4_oif = dev_out->ifindex; 2810 goto make_route; 2811 } 2812 2813 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 2814 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2815 if (!__ip_dev_find(net, fl4->saddr, false)) 2816 goto out; 2817 } 2818 } 2819 2820 2821 if (fl4->flowi4_oif) { 2822 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif); 2823 rth = ERR_PTR(-ENODEV); 2824 if (!dev_out) 2825 goto out; 2826 2827 /* RACE: Check return value of inet_select_addr instead. */ 2828 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 2829 rth = ERR_PTR(-ENETUNREACH); 2830 goto out; 2831 } 2832 if (ipv4_is_local_multicast(fl4->daddr) || 2833 ipv4_is_lbcast(fl4->daddr) || 2834 fl4->flowi4_proto == IPPROTO_IGMP) { 2835 if (!fl4->saddr) 2836 fl4->saddr = inet_select_addr(dev_out, 0, 2837 RT_SCOPE_LINK); 2838 goto make_route; 2839 } 2840 if (!fl4->saddr) { 2841 if (ipv4_is_multicast(fl4->daddr)) 2842 fl4->saddr = inet_select_addr(dev_out, 0, 2843 fl4->flowi4_scope); 2844 else if (!fl4->daddr) 2845 fl4->saddr = inet_select_addr(dev_out, 0, 2846 RT_SCOPE_HOST); 2847 } 2848 } 2849 2850 if (!fl4->daddr) { 2851 fl4->daddr = fl4->saddr; 2852 if (!fl4->daddr) 2853 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK); 2854 dev_out = net->loopback_dev; 2855 fl4->flowi4_oif = LOOPBACK_IFINDEX; 2856 res->type = RTN_LOCAL; 2857 flags |= RTCF_LOCAL; 2858 goto make_route; 2859 } 2860 2861 err = fib_lookup(net, fl4, res, 0); 2862 if (err) { 2863 res->fi = NULL; 2864 res->table = NULL; 2865 if (fl4->flowi4_oif && 2866 (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) { 2867 /* Apparently, routing tables are wrong. Assume, 2868 * that the destination is on link. 2869 * 2870 * WHY? DW. 2871 * Because we are allowed to send to iface 2872 * even if it has NO routes and NO assigned 2873 * addresses. When oif is specified, routing 2874 * tables are looked up with only one purpose: 2875 * to catch if destination is gatewayed, rather than 2876 * direct. Moreover, if MSG_DONTROUTE is set, 2877 * we send packet, ignoring both routing tables 2878 * and ifaddr state. --ANK 2879 * 2880 * 2881 * We could make it even if oif is unknown, 2882 * likely IPv6, but we do not. 2883 */ 2884 2885 if (fl4->saddr == 0) 2886 fl4->saddr = inet_select_addr(dev_out, 0, 2887 RT_SCOPE_LINK); 2888 res->type = RTN_UNICAST; 2889 goto make_route; 2890 } 2891 rth = ERR_PTR(err); 2892 goto out; 2893 } 2894 2895 if (res->type == RTN_LOCAL) { 2896 if (!fl4->saddr) { 2897 if (res->fi->fib_prefsrc) 2898 fl4->saddr = res->fi->fib_prefsrc; 2899 else 2900 fl4->saddr = fl4->daddr; 2901 } 2902 2903 /* L3 master device is the loopback for that domain */ 2904 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? : 2905 net->loopback_dev; 2906 2907 /* make sure orig_oif points to fib result device even 2908 * though packet rx/tx happens over loopback or l3mdev 2909 */ 2910 orig_oif = FIB_RES_OIF(*res); 2911 2912 fl4->flowi4_oif = dev_out->ifindex; 2913 flags |= RTCF_LOCAL; 2914 goto make_route; 2915 } 2916 2917 fib_select_path(net, res, fl4, skb); 2918 2919 dev_out = FIB_RES_DEV(*res); 2920 2921 make_route: 2922 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags); 2923 2924 out: 2925 return rth; 2926 } 2927 2928 static struct dst_ops ipv4_dst_blackhole_ops = { 2929 .family = AF_INET, 2930 .default_advmss = ipv4_default_advmss, 2931 .neigh_lookup = ipv4_dst_neigh_lookup, 2932 .check = dst_blackhole_check, 2933 .cow_metrics = dst_blackhole_cow_metrics, 2934 .update_pmtu = dst_blackhole_update_pmtu, 2935 .redirect = dst_blackhole_redirect, 2936 .mtu = dst_blackhole_mtu, 2937 }; 2938 2939 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2940 { 2941 struct rtable *ort = dst_rtable(dst_orig); 2942 struct rtable *rt; 2943 2944 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0); 2945 if (rt) { 2946 struct dst_entry *new = &rt->dst; 2947 2948 new->__use = 1; 2949 new->input = dst_discard; 2950 new->output = dst_discard_out; 2951 2952 new->dev = net->loopback_dev; 2953 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC); 2954 2955 rt->rt_is_input = ort->rt_is_input; 2956 rt->rt_iif = ort->rt_iif; 2957 rt->rt_pmtu = ort->rt_pmtu; 2958 rt->rt_mtu_locked = ort->rt_mtu_locked; 2959 2960 rt->rt_genid = rt_genid_ipv4(net); 2961 rt->rt_flags = ort->rt_flags; 2962 rt->rt_type = ort->rt_type; 2963 rt->rt_uses_gateway = ort->rt_uses_gateway; 2964 rt->rt_gw_family = ort->rt_gw_family; 2965 if (rt->rt_gw_family == AF_INET) 2966 rt->rt_gw4 = ort->rt_gw4; 2967 else if (rt->rt_gw_family == AF_INET6) 2968 rt->rt_gw6 = ort->rt_gw6; 2969 } 2970 2971 dst_release(dst_orig); 2972 2973 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2974 } 2975 2976 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2977 const struct sock *sk) 2978 { 2979 struct rtable *rt = __ip_route_output_key(net, flp4); 2980 2981 if (IS_ERR(rt)) 2982 return rt; 2983 2984 if (flp4->flowi4_proto) { 2985 flp4->flowi4_oif = rt->dst.dev->ifindex; 2986 rt = dst_rtable(xfrm_lookup_route(net, &rt->dst, 2987 flowi4_to_flowi(flp4), 2988 sk, 0)); 2989 } 2990 2991 return rt; 2992 } 2993 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2994 2995 /* called with rcu_read_lock held */ 2996 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, 2997 struct rtable *rt, u32 table_id, dscp_t dscp, 2998 struct flowi4 *fl4, struct sk_buff *skb, u32 portid, 2999 u32 seq, unsigned int flags) 3000 { 3001 struct rtmsg *r; 3002 struct nlmsghdr *nlh; 3003 unsigned long expires = 0; 3004 u32 error; 3005 u32 metrics[RTAX_MAX]; 3006 3007 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags); 3008 if (!nlh) 3009 return -EMSGSIZE; 3010 3011 r = nlmsg_data(nlh); 3012 r->rtm_family = AF_INET; 3013 r->rtm_dst_len = 32; 3014 r->rtm_src_len = 0; 3015 r->rtm_tos = inet_dscp_to_dsfield(dscp); 3016 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT; 3017 if (nla_put_u32(skb, RTA_TABLE, table_id)) 3018 goto nla_put_failure; 3019 r->rtm_type = rt->rt_type; 3020 r->rtm_scope = RT_SCOPE_UNIVERSE; 3021 r->rtm_protocol = RTPROT_UNSPEC; 3022 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 3023 if (rt->rt_flags & RTCF_NOTIFY) 3024 r->rtm_flags |= RTM_F_NOTIFY; 3025 if (IPCB(skb)->flags & IPSKB_DOREDIRECT) 3026 r->rtm_flags |= RTCF_DOREDIRECT; 3027 3028 if (nla_put_in_addr(skb, RTA_DST, dst)) 3029 goto nla_put_failure; 3030 if (src) { 3031 r->rtm_src_len = 32; 3032 if (nla_put_in_addr(skb, RTA_SRC, src)) 3033 goto nla_put_failure; 3034 } 3035 if (rt->dst.dev && 3036 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex)) 3037 goto nla_put_failure; 3038 if (lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0) 3039 goto nla_put_failure; 3040 #ifdef CONFIG_IP_ROUTE_CLASSID 3041 if (rt->dst.tclassid && 3042 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid)) 3043 goto nla_put_failure; 3044 #endif 3045 if (fl4 && !rt_is_input_route(rt) && 3046 fl4->saddr != src) { 3047 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr)) 3048 goto nla_put_failure; 3049 } 3050 if (rt->rt_uses_gateway) { 3051 if (rt->rt_gw_family == AF_INET && 3052 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) { 3053 goto nla_put_failure; 3054 } else if (rt->rt_gw_family == AF_INET6) { 3055 int alen = sizeof(struct in6_addr); 3056 struct nlattr *nla; 3057 struct rtvia *via; 3058 3059 nla = nla_reserve(skb, RTA_VIA, alen + 2); 3060 if (!nla) 3061 goto nla_put_failure; 3062 3063 via = nla_data(nla); 3064 via->rtvia_family = AF_INET6; 3065 memcpy(via->rtvia_addr, &rt->rt_gw6, alen); 3066 } 3067 } 3068 3069 expires = READ_ONCE(rt->dst.expires); 3070 if (expires) { 3071 unsigned long now = jiffies; 3072 3073 if (time_before(now, expires)) 3074 expires -= now; 3075 else 3076 expires = 0; 3077 } 3078 3079 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics)); 3080 if (rt->rt_pmtu && expires) 3081 metrics[RTAX_MTU - 1] = rt->rt_pmtu; 3082 if (rt->rt_mtu_locked && expires) 3083 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU); 3084 if (rtnetlink_put_metrics(skb, metrics) < 0) 3085 goto nla_put_failure; 3086 3087 if (fl4) { 3088 if (fl4->flowi4_mark && 3089 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark)) 3090 goto nla_put_failure; 3091 3092 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) && 3093 nla_put_u32(skb, RTA_UID, 3094 from_kuid_munged(current_user_ns(), 3095 fl4->flowi4_uid))) 3096 goto nla_put_failure; 3097 3098 if (rt_is_input_route(rt)) { 3099 #ifdef CONFIG_IP_MROUTE 3100 if (ipv4_is_multicast(dst) && 3101 !ipv4_is_local_multicast(dst) && 3102 IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) { 3103 int err = ipmr_get_route(net, skb, 3104 fl4->saddr, fl4->daddr, 3105 r, portid); 3106 3107 if (err <= 0) { 3108 if (err == 0) 3109 return 0; 3110 goto nla_put_failure; 3111 } 3112 } else 3113 #endif 3114 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif)) 3115 goto nla_put_failure; 3116 } 3117 } 3118 3119 error = rt->dst.error; 3120 3121 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0) 3122 goto nla_put_failure; 3123 3124 nlmsg_end(skb, nlh); 3125 return 0; 3126 3127 nla_put_failure: 3128 nlmsg_cancel(skb, nlh); 3129 return -EMSGSIZE; 3130 } 3131 3132 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb, 3133 struct netlink_callback *cb, u32 table_id, 3134 struct fnhe_hash_bucket *bucket, int genid, 3135 int *fa_index, int fa_start, unsigned int flags) 3136 { 3137 int i; 3138 3139 for (i = 0; i < FNHE_HASH_SIZE; i++) { 3140 struct fib_nh_exception *fnhe; 3141 3142 for (fnhe = rcu_dereference(bucket[i].chain); fnhe; 3143 fnhe = rcu_dereference(fnhe->fnhe_next)) { 3144 struct rtable *rt; 3145 int err; 3146 3147 if (*fa_index < fa_start) 3148 goto next; 3149 3150 if (fnhe->fnhe_genid != genid) 3151 goto next; 3152 3153 if (fnhe->fnhe_expires && 3154 time_after(jiffies, fnhe->fnhe_expires)) 3155 goto next; 3156 3157 rt = rcu_dereference(fnhe->fnhe_rth_input); 3158 if (!rt) 3159 rt = rcu_dereference(fnhe->fnhe_rth_output); 3160 if (!rt) 3161 goto next; 3162 3163 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt, 3164 table_id, 0, NULL, skb, 3165 NETLINK_CB(cb->skb).portid, 3166 cb->nlh->nlmsg_seq, flags); 3167 if (err) 3168 return err; 3169 next: 3170 (*fa_index)++; 3171 } 3172 } 3173 3174 return 0; 3175 } 3176 3177 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb, 3178 u32 table_id, struct fib_info *fi, 3179 int *fa_index, int fa_start, unsigned int flags) 3180 { 3181 struct net *net = sock_net(cb->skb->sk); 3182 int nhsel, genid = fnhe_genid(net); 3183 3184 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) { 3185 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel); 3186 struct fnhe_hash_bucket *bucket; 3187 int err; 3188 3189 if (!nhc) 3190 break; 3191 3192 if (nhc->nhc_flags & RTNH_F_DEAD) 3193 continue; 3194 3195 rcu_read_lock(); 3196 bucket = rcu_dereference(nhc->nhc_exceptions); 3197 err = 0; 3198 if (bucket) 3199 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket, 3200 genid, fa_index, fa_start, 3201 flags); 3202 rcu_read_unlock(); 3203 if (err) 3204 return err; 3205 } 3206 3207 return 0; 3208 } 3209 3210 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst, 3211 u8 ip_proto, __be16 sport, 3212 __be16 dport) 3213 { 3214 struct sk_buff *skb; 3215 struct iphdr *iph; 3216 3217 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 3218 if (!skb) 3219 return NULL; 3220 3221 /* Reserve room for dummy headers, this skb can pass 3222 * through good chunk of routing engine. 3223 */ 3224 skb_reset_mac_header(skb); 3225 skb_reset_network_header(skb); 3226 skb->protocol = htons(ETH_P_IP); 3227 iph = skb_put(skb, sizeof(struct iphdr)); 3228 iph->protocol = ip_proto; 3229 iph->saddr = src; 3230 iph->daddr = dst; 3231 iph->version = 0x4; 3232 iph->frag_off = 0; 3233 iph->ihl = 0x5; 3234 skb_set_transport_header(skb, skb->len); 3235 3236 switch (iph->protocol) { 3237 case IPPROTO_UDP: { 3238 struct udphdr *udph; 3239 3240 udph = skb_put_zero(skb, sizeof(struct udphdr)); 3241 udph->source = sport; 3242 udph->dest = dport; 3243 udp_set_len_short(udph, sizeof(struct udphdr)); 3244 udph->check = 0; 3245 break; 3246 } 3247 case IPPROTO_TCP: { 3248 struct tcphdr *tcph; 3249 3250 tcph = skb_put_zero(skb, sizeof(struct tcphdr)); 3251 tcph->source = sport; 3252 tcph->dest = dport; 3253 tcph->doff = sizeof(struct tcphdr) / 4; 3254 tcph->rst = 1; 3255 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr), 3256 src, dst, 0); 3257 break; 3258 } 3259 case IPPROTO_ICMP: { 3260 struct icmphdr *icmph; 3261 3262 icmph = skb_put_zero(skb, sizeof(struct icmphdr)); 3263 icmph->type = ICMP_ECHO; 3264 icmph->code = 0; 3265 } 3266 } 3267 3268 return skb; 3269 } 3270 3271 static int inet_rtm_valid_getroute_req(struct sk_buff *skb, 3272 const struct nlmsghdr *nlh, 3273 struct nlattr **tb, 3274 struct netlink_ext_ack *extack) 3275 { 3276 struct rtmsg *rtm; 3277 int i, err; 3278 3279 rtm = nlmsg_payload(nlh, sizeof(*rtm)); 3280 if (!rtm) { 3281 NL_SET_ERR_MSG(extack, 3282 "ipv4: Invalid header for route get request"); 3283 return -EINVAL; 3284 } 3285 3286 if (!netlink_strict_get_check(skb)) 3287 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX, 3288 rtm_ipv4_policy, extack); 3289 3290 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) || 3291 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) || 3292 rtm->rtm_table || rtm->rtm_protocol || 3293 rtm->rtm_scope || rtm->rtm_type) { 3294 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request"); 3295 return -EINVAL; 3296 } 3297 3298 if (rtm->rtm_flags & ~(RTM_F_NOTIFY | 3299 RTM_F_LOOKUP_TABLE | 3300 RTM_F_FIB_MATCH)) { 3301 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request"); 3302 return -EINVAL; 3303 } 3304 3305 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX, 3306 rtm_ipv4_policy, extack); 3307 if (err) 3308 return err; 3309 3310 if ((tb[RTA_SRC] && !rtm->rtm_src_len) || 3311 (tb[RTA_DST] && !rtm->rtm_dst_len)) { 3312 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4"); 3313 return -EINVAL; 3314 } 3315 3316 for (i = 0; i <= RTA_MAX; i++) { 3317 if (!tb[i]) 3318 continue; 3319 3320 switch (i) { 3321 case RTA_IIF: 3322 case RTA_OIF: 3323 case RTA_SRC: 3324 case RTA_DST: 3325 case RTA_IP_PROTO: 3326 case RTA_SPORT: 3327 case RTA_DPORT: 3328 case RTA_MARK: 3329 case RTA_UID: 3330 break; 3331 default: 3332 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request"); 3333 return -EINVAL; 3334 } 3335 } 3336 3337 return 0; 3338 } 3339 3340 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, 3341 struct netlink_ext_ack *extack) 3342 { 3343 struct net *net = sock_net(in_skb->sk); 3344 struct nlattr *tb[RTA_MAX+1]; 3345 u32 table_id = RT_TABLE_MAIN; 3346 __be16 sport = 0, dport = 0; 3347 struct fib_result res = {}; 3348 u8 ip_proto = IPPROTO_UDP; 3349 struct rtable *rt = NULL; 3350 struct sk_buff *skb; 3351 struct rtmsg *rtm; 3352 struct flowi4 fl4 = {}; 3353 __be32 dst = 0; 3354 __be32 src = 0; 3355 dscp_t dscp; 3356 kuid_t uid; 3357 u32 iif; 3358 int err; 3359 int mark; 3360 3361 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack); 3362 if (err < 0) 3363 return err; 3364 3365 rtm = nlmsg_data(nlh); 3366 src = nla_get_in_addr_default(tb[RTA_SRC], 0); 3367 dst = nla_get_in_addr_default(tb[RTA_DST], 0); 3368 iif = nla_get_u32_default(tb[RTA_IIF], 0); 3369 mark = nla_get_u32_default(tb[RTA_MARK], 0); 3370 dscp = inet_dsfield_to_dscp(rtm->rtm_tos); 3371 if (tb[RTA_UID]) 3372 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID])); 3373 else 3374 uid = (iif ? INVALID_UID : current_uid()); 3375 3376 if (tb[RTA_IP_PROTO]) { 3377 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO], 3378 &ip_proto, AF_INET, extack); 3379 if (err) 3380 return err; 3381 } 3382 3383 if (tb[RTA_SPORT]) 3384 sport = nla_get_be16(tb[RTA_SPORT]); 3385 3386 if (tb[RTA_DPORT]) 3387 dport = nla_get_be16(tb[RTA_DPORT]); 3388 3389 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport); 3390 if (!skb) 3391 return -ENOBUFS; 3392 3393 fl4.daddr = dst; 3394 fl4.saddr = src; 3395 fl4.flowi4_dscp = dscp; 3396 fl4.flowi4_oif = nla_get_u32_default(tb[RTA_OIF], 0); 3397 fl4.flowi4_mark = mark; 3398 fl4.flowi4_uid = uid; 3399 if (sport) 3400 fl4.fl4_sport = sport; 3401 if (dport) 3402 fl4.fl4_dport = dport; 3403 fl4.flowi4_proto = ip_proto; 3404 3405 rcu_read_lock(); 3406 3407 if (iif) { 3408 struct net_device *dev; 3409 3410 dev = dev_get_by_index_rcu(net, iif); 3411 if (!dev) { 3412 err = -ENODEV; 3413 goto errout_rcu; 3414 } 3415 3416 fl4.flowi4_iif = iif; /* for rt_fill_info */ 3417 skb->dev = dev; 3418 skb->mark = mark; 3419 err = ip_route_input_rcu(skb, dst, src, dscp, dev, 3420 &res) ? -EINVAL : 0; 3421 3422 rt = skb_rtable(skb); 3423 if (err == 0 && rt->dst.error) 3424 err = -rt->dst.error; 3425 } else { 3426 fl4.flowi4_iif = LOOPBACK_IFINDEX; 3427 skb->dev = net->loopback_dev; 3428 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb); 3429 err = 0; 3430 if (IS_ERR(rt)) 3431 err = PTR_ERR(rt); 3432 else 3433 skb_dst_set(skb, &rt->dst); 3434 } 3435 3436 if (err) 3437 goto errout_rcu; 3438 3439 if (rtm->rtm_flags & RTM_F_NOTIFY) 3440 rt->rt_flags |= RTCF_NOTIFY; 3441 3442 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE) 3443 table_id = res.table ? res.table->tb_id : 0; 3444 3445 /* reset skb for netlink reply msg */ 3446 skb_trim(skb, 0); 3447 skb_reset_network_header(skb); 3448 skb_reset_transport_header(skb); 3449 skb_reset_mac_header(skb); 3450 3451 if (rtm->rtm_flags & RTM_F_FIB_MATCH) { 3452 struct fib_rt_info fri; 3453 3454 if (!res.fi) { 3455 err = fib_props[res.type].error; 3456 if (!err) 3457 err = -EHOSTUNREACH; 3458 goto errout_rcu; 3459 } 3460 fri.fi = res.fi; 3461 fri.tb_id = table_id; 3462 fri.dst = res.prefix; 3463 fri.dst_len = res.prefixlen; 3464 fri.dscp = res.dscp; 3465 fri.type = rt->rt_type; 3466 fri.offload = 0; 3467 fri.trap = 0; 3468 fri.offload_failed = 0; 3469 if (res.fa_head) { 3470 struct fib_alias *fa; 3471 3472 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) { 3473 u8 slen = 32 - fri.dst_len; 3474 3475 if (fa->fa_slen == slen && 3476 fa->tb_id == fri.tb_id && 3477 fa->fa_dscp == fri.dscp && 3478 fa->fa_info == res.fi && 3479 fa->fa_type == fri.type) { 3480 fri.offload = READ_ONCE(fa->offload); 3481 fri.trap = READ_ONCE(fa->trap); 3482 fri.offload_failed = 3483 READ_ONCE(fa->offload_failed); 3484 break; 3485 } 3486 } 3487 } 3488 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid, 3489 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0); 3490 } else { 3491 err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4, 3492 skb, NETLINK_CB(in_skb).portid, 3493 nlh->nlmsg_seq, 0); 3494 } 3495 if (err < 0) 3496 goto errout_rcu; 3497 3498 rcu_read_unlock(); 3499 3500 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 3501 3502 errout_free: 3503 return err; 3504 errout_rcu: 3505 rcu_read_unlock(); 3506 kfree_skb(skb); 3507 goto errout_free; 3508 } 3509 3510 void ip_rt_multicast_event(struct in_device *in_dev) 3511 { 3512 rt_cache_flush(dev_net(in_dev->dev)); 3513 } 3514 3515 #ifdef CONFIG_SYSCTL 3516 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 3517 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 3518 static int ip_rt_gc_elasticity __read_mostly = 8; 3519 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU; 3520 3521 static int ipv4_sysctl_rtcache_flush(const struct ctl_table *__ctl, int write, 3522 void *buffer, size_t *lenp, loff_t *ppos) 3523 { 3524 struct net *net = (struct net *)__ctl->extra1; 3525 3526 if (write) { 3527 rt_cache_flush(net); 3528 fnhe_genid_bump(net); 3529 return 0; 3530 } 3531 3532 return -EINVAL; 3533 } 3534 3535 static struct ctl_table ipv4_route_table[] = { 3536 { 3537 .procname = "gc_thresh", 3538 .data = &ipv4_dst_ops.gc_thresh, 3539 .maxlen = sizeof(int), 3540 .mode = 0644, 3541 .proc_handler = proc_dointvec, 3542 }, 3543 { 3544 .procname = "max_size", 3545 .data = &ip_rt_max_size, 3546 .maxlen = sizeof(int), 3547 .mode = 0644, 3548 .proc_handler = proc_dointvec, 3549 }, 3550 { 3551 /* Deprecated. Use gc_min_interval_ms */ 3552 3553 .procname = "gc_min_interval", 3554 .data = &ip_rt_gc_min_interval, 3555 .maxlen = sizeof(int), 3556 .mode = 0644, 3557 .proc_handler = proc_dointvec_jiffies, 3558 }, 3559 { 3560 .procname = "gc_min_interval_ms", 3561 .data = &ip_rt_gc_min_interval, 3562 .maxlen = sizeof(int), 3563 .mode = 0644, 3564 .proc_handler = proc_dointvec_ms_jiffies, 3565 }, 3566 { 3567 .procname = "gc_timeout", 3568 .data = &ip_rt_gc_timeout, 3569 .maxlen = sizeof(int), 3570 .mode = 0644, 3571 .proc_handler = proc_dointvec_jiffies, 3572 }, 3573 { 3574 .procname = "gc_interval", 3575 .data = &ip_rt_gc_interval, 3576 .maxlen = sizeof(int), 3577 .mode = 0644, 3578 .proc_handler = proc_dointvec_jiffies, 3579 }, 3580 { 3581 .procname = "redirect_load", 3582 .data = &ip_rt_redirect_load, 3583 .maxlen = sizeof(int), 3584 .mode = 0644, 3585 .proc_handler = proc_dointvec, 3586 }, 3587 { 3588 .procname = "redirect_number", 3589 .data = &ip_rt_redirect_number, 3590 .maxlen = sizeof(int), 3591 .mode = 0644, 3592 .proc_handler = proc_dointvec, 3593 }, 3594 { 3595 .procname = "redirect_silence", 3596 .data = &ip_rt_redirect_silence, 3597 .maxlen = sizeof(int), 3598 .mode = 0644, 3599 .proc_handler = proc_dointvec, 3600 }, 3601 { 3602 .procname = "error_cost", 3603 .data = &ip_rt_error_cost, 3604 .maxlen = sizeof(int), 3605 .mode = 0644, 3606 .proc_handler = proc_dointvec, 3607 }, 3608 { 3609 .procname = "error_burst", 3610 .data = &ip_rt_error_burst, 3611 .maxlen = sizeof(int), 3612 .mode = 0644, 3613 .proc_handler = proc_dointvec, 3614 }, 3615 { 3616 .procname = "gc_elasticity", 3617 .data = &ip_rt_gc_elasticity, 3618 .maxlen = sizeof(int), 3619 .mode = 0644, 3620 .proc_handler = proc_dointvec, 3621 }, 3622 }; 3623 3624 static const char ipv4_route_flush_procname[] = "flush"; 3625 3626 static struct ctl_table ipv4_route_netns_table[] = { 3627 { 3628 .procname = ipv4_route_flush_procname, 3629 .maxlen = sizeof(int), 3630 .mode = 0200, 3631 .proc_handler = ipv4_sysctl_rtcache_flush, 3632 }, 3633 { 3634 .procname = "min_pmtu", 3635 .data = &init_net.ipv4.ip_rt_min_pmtu, 3636 .maxlen = sizeof(int), 3637 .mode = 0644, 3638 .proc_handler = proc_dointvec_minmax, 3639 .extra1 = &ip_min_valid_pmtu, 3640 }, 3641 { 3642 .procname = "mtu_expires", 3643 .data = &init_net.ipv4.ip_rt_mtu_expires, 3644 .maxlen = sizeof(int), 3645 .mode = 0644, 3646 .proc_handler = proc_dointvec_jiffies, 3647 }, 3648 { 3649 .procname = "min_adv_mss", 3650 .data = &init_net.ipv4.ip_rt_min_advmss, 3651 .maxlen = sizeof(int), 3652 .mode = 0644, 3653 .proc_handler = proc_dointvec, 3654 }, 3655 }; 3656 3657 static __net_init int sysctl_route_net_init(struct net *net) 3658 { 3659 struct ctl_table *tbl; 3660 size_t table_size = ARRAY_SIZE(ipv4_route_netns_table); 3661 3662 tbl = ipv4_route_netns_table; 3663 if (!net_eq(net, &init_net)) { 3664 int i; 3665 3666 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL); 3667 if (!tbl) 3668 goto err_dup; 3669 3670 /* Don't export non-whitelisted sysctls to unprivileged users */ 3671 if (net->user_ns != &init_user_ns) { 3672 if (tbl[0].procname != ipv4_route_flush_procname) 3673 table_size = 0; 3674 } 3675 3676 /* Update the variables to point into the current struct net 3677 * except for the first element flush 3678 */ 3679 for (i = 1; i < table_size; i++) 3680 tbl[i].data += (void *)net - (void *)&init_net; 3681 } 3682 tbl[0].extra1 = net; 3683 3684 net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route", 3685 tbl, table_size); 3686 if (!net->ipv4.route_hdr) 3687 goto err_reg; 3688 return 0; 3689 3690 err_reg: 3691 if (tbl != ipv4_route_netns_table) 3692 kfree(tbl); 3693 err_dup: 3694 return -ENOMEM; 3695 } 3696 3697 static __net_exit void sysctl_route_net_exit(struct net *net) 3698 { 3699 const struct ctl_table *tbl; 3700 3701 tbl = net->ipv4.route_hdr->ctl_table_arg; 3702 unregister_net_sysctl_table(net->ipv4.route_hdr); 3703 BUG_ON(tbl == ipv4_route_netns_table); 3704 kfree(tbl); 3705 } 3706 3707 static __net_initdata struct pernet_operations sysctl_route_ops = { 3708 .init = sysctl_route_net_init, 3709 .exit = sysctl_route_net_exit, 3710 }; 3711 #endif 3712 3713 static __net_init int netns_ip_rt_init(struct net *net) 3714 { 3715 /* Set default value for namespaceified sysctls */ 3716 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU; 3717 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES; 3718 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS; 3719 return 0; 3720 } 3721 3722 static struct pernet_operations __net_initdata ip_rt_ops = { 3723 .init = netns_ip_rt_init, 3724 }; 3725 3726 static __net_init int rt_genid_init(struct net *net) 3727 { 3728 atomic_set(&net->ipv4.rt_genid, 0); 3729 atomic_set(&net->fnhe_genid, 0); 3730 atomic_set(&net->ipv4.dev_addr_genid, get_random_u32()); 3731 return 0; 3732 } 3733 3734 static __net_initdata struct pernet_operations rt_genid_ops = { 3735 .init = rt_genid_init, 3736 }; 3737 3738 static int __net_init ipv4_inetpeer_init(struct net *net) 3739 { 3740 struct inet_peer_base *bp = kmalloc_obj(*bp); 3741 3742 if (!bp) 3743 return -ENOMEM; 3744 inet_peer_base_init(bp); 3745 net->ipv4.peers = bp; 3746 return 0; 3747 } 3748 3749 static void __net_exit ipv4_inetpeer_exit(struct net *net) 3750 { 3751 struct inet_peer_base *bp = net->ipv4.peers; 3752 3753 net->ipv4.peers = NULL; 3754 inetpeer_invalidate_tree(bp); 3755 kfree(bp); 3756 } 3757 3758 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = { 3759 .init = ipv4_inetpeer_init, 3760 .exit = ipv4_inetpeer_exit, 3761 }; 3762 3763 #ifdef CONFIG_IP_ROUTE_CLASSID 3764 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 3765 #endif /* CONFIG_IP_ROUTE_CLASSID */ 3766 3767 static const struct rtnl_msg_handler ip_rt_rtnl_msg_handlers[] __initconst = { 3768 {.protocol = PF_INET, .msgtype = RTM_GETROUTE, 3769 .doit = inet_rtm_getroute, .flags = RTNL_FLAG_DOIT_UNLOCKED}, 3770 }; 3771 3772 int __init ip_rt_init(void) 3773 { 3774 void *idents_hash; 3775 int cpu; 3776 3777 /* For modern hosts, this will use 2 MB of memory */ 3778 idents_hash = alloc_large_system_hash("IP idents", 3779 sizeof(*ip_idents) + sizeof(*ip_tstamps), 3780 0, 3781 16, /* one bucket per 64 KB */ 3782 HASH_ZERO, 3783 NULL, 3784 &ip_idents_mask, 3785 2048, 3786 256*1024); 3787 3788 ip_idents = idents_hash; 3789 3790 get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents)); 3791 3792 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents); 3793 3794 for_each_possible_cpu(cpu) { 3795 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 3796 3797 INIT_LIST_HEAD(&ul->head); 3798 spin_lock_init(&ul->lock); 3799 } 3800 #ifdef CONFIG_IP_ROUTE_CLASSID 3801 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 3802 if (!ip_rt_acct) 3803 panic("IP: failed to allocate ip_rt_acct\n"); 3804 #endif 3805 3806 ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable, 3807 SLAB_HWCACHE_ALIGN | SLAB_PANIC); 3808 3809 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 3810 3811 if (dst_entries_init(&ipv4_dst_ops) < 0) 3812 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 3813 3814 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 3815 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 3816 3817 ipv4_dst_ops.gc_thresh = ~0; 3818 ip_rt_max_size = INT_MAX; 3819 3820 devinet_init(); 3821 ip_fib_init(); 3822 3823 if (ip_rt_proc_init()) 3824 pr_err("Unable to create route proc files\n"); 3825 #ifdef CONFIG_XFRM 3826 xfrm_init(); 3827 xfrm4_init(); 3828 #endif 3829 rtnl_register_many(ip_rt_rtnl_msg_handlers); 3830 3831 #ifdef CONFIG_SYSCTL 3832 register_pernet_subsys(&sysctl_route_ops); 3833 #endif 3834 register_pernet_subsys(&ip_rt_ops); 3835 register_pernet_subsys(&rt_genid_ops); 3836 register_pernet_subsys(&ipv4_inetpeer_ops); 3837 return 0; 3838 } 3839 3840 #ifdef CONFIG_SYSCTL 3841 /* 3842 * We really need to sanitize the damn ipv4 init order, then all 3843 * this nonsense will go away. 3844 */ 3845 void __init ip_static_sysctl_init(void) 3846 { 3847 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table); 3848 } 3849 #endif 3850