1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * ROUTE - implementation of the IP router. 8 * 9 * Authors: Ross Biro 10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 11 * Alan Cox, <gw4pts@gw4pts.ampr.org> 12 * Linus Torvalds, <Linus.Torvalds@helsinki.fi> 13 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 14 * 15 * Fixes: 16 * Alan Cox : Verify area fixes. 17 * Alan Cox : cli() protects routing changes 18 * Rui Oliveira : ICMP routing table updates 19 * (rco@di.uminho.pt) Routing table insertion and update 20 * Linus Torvalds : Rewrote bits to be sensible 21 * Alan Cox : Added BSD route gw semantics 22 * Alan Cox : Super /proc >4K 23 * Alan Cox : MTU in route table 24 * Alan Cox : MSS actually. Also added the window 25 * clamper. 26 * Sam Lantinga : Fixed route matching in rt_del() 27 * Alan Cox : Routing cache support. 28 * Alan Cox : Removed compatibility cruft. 29 * Alan Cox : RTF_REJECT support. 30 * Alan Cox : TCP irtt support. 31 * Jonathan Naylor : Added Metric support. 32 * Miquel van Smoorenburg : BSD API fixes. 33 * Miquel van Smoorenburg : Metrics. 34 * Alan Cox : Use __u32 properly 35 * Alan Cox : Aligned routing errors more closely with BSD 36 * our system is still very different. 37 * Alan Cox : Faster /proc handling 38 * Alexey Kuznetsov : Massive rework to support tree based routing, 39 * routing caches and better behaviour. 40 * 41 * Olaf Erb : irtt wasn't being copied right. 42 * Bjorn Ekwall : Kerneld route support. 43 * Alan Cox : Multicast fixed (I hope) 44 * Pavel Krauz : Limited broadcast fixed 45 * Mike McLagan : Routing by source 46 * Alexey Kuznetsov : End of old history. Split to fib.c and 47 * route.c and rewritten from scratch. 48 * Andi Kleen : Load-limit warning messages. 49 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 50 * Vitaly E. Lavrov : Race condition in ip_route_input_slow. 51 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow. 52 * Vladimir V. Ivanov : IP rule info (flowid) is really useful. 53 * Marc Boucher : routing by fwmark 54 * Robert Olsson : Added rt_cache statistics 55 * Arnaldo C. Melo : Convert proc stuff to seq_file 56 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes. 57 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect 58 * Ilia Sotnikov : Removed TOS from hash calculations 59 */ 60 61 #define pr_fmt(fmt) "IPv4: " fmt 62 63 #include <linux/module.h> 64 #include <linux/bitops.h> 65 #include <linux/kernel.h> 66 #include <linux/mm.h> 67 #include <linux/memblock.h> 68 #include <linux/socket.h> 69 #include <linux/errno.h> 70 #include <linux/in.h> 71 #include <linux/inet.h> 72 #include <linux/netdevice.h> 73 #include <linux/proc_fs.h> 74 #include <linux/init.h> 75 #include <linux/skbuff.h> 76 #include <linux/inetdevice.h> 77 #include <linux/igmp.h> 78 #include <linux/pkt_sched.h> 79 #include <linux/mroute.h> 80 #include <linux/netfilter_ipv4.h> 81 #include <linux/random.h> 82 #include <linux/rcupdate.h> 83 #include <linux/slab.h> 84 #include <linux/jhash.h> 85 #include <net/dst.h> 86 #include <net/dst_metadata.h> 87 #include <net/inet_dscp.h> 88 #include <net/net_namespace.h> 89 #include <net/ip.h> 90 #include <net/route.h> 91 #include <net/inetpeer.h> 92 #include <net/sock.h> 93 #include <net/ip_fib.h> 94 #include <net/nexthop.h> 95 #include <net/tcp.h> 96 #include <net/icmp.h> 97 #include <net/xfrm.h> 98 #include <net/lwtunnel.h> 99 #include <net/netevent.h> 100 #include <net/rtnetlink.h> 101 #ifdef CONFIG_SYSCTL 102 #include <linux/sysctl.h> 103 #endif 104 #include <net/secure_seq.h> 105 #include <net/ip_tunnels.h> 106 107 #include "fib_lookup.h" 108 109 #define RT_FL_TOS(oldflp4) \ 110 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)) 111 112 #define RT_GC_TIMEOUT (300*HZ) 113 114 #define DEFAULT_MIN_PMTU (512 + 20 + 20) 115 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ) 116 #define DEFAULT_MIN_ADVMSS 256 117 static int ip_rt_max_size; 118 static int ip_rt_redirect_number __read_mostly = 9; 119 static int ip_rt_redirect_load __read_mostly = HZ / 50; 120 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1)); 121 static int ip_rt_error_cost __read_mostly = HZ; 122 static int ip_rt_error_burst __read_mostly = 5 * HZ; 123 124 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT; 125 126 /* 127 * Interface to generic destination cache. 128 */ 129 130 INDIRECT_CALLABLE_SCOPE 131 struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie); 132 static unsigned int ipv4_default_advmss(const struct dst_entry *dst); 133 INDIRECT_CALLABLE_SCOPE 134 unsigned int ipv4_mtu(const struct dst_entry *dst); 135 static void ipv4_negative_advice(struct sock *sk, 136 struct dst_entry *dst); 137 static void ipv4_link_failure(struct sk_buff *skb); 138 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 139 struct sk_buff *skb, u32 mtu, 140 bool confirm_neigh); 141 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, 142 struct sk_buff *skb); 143 static void ipv4_dst_destroy(struct dst_entry *dst); 144 145 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old) 146 { 147 WARN_ON(1); 148 return NULL; 149 } 150 151 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 152 struct sk_buff *skb, 153 const void *daddr); 154 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr); 155 156 static struct dst_ops ipv4_dst_ops = { 157 .family = AF_INET, 158 .check = ipv4_dst_check, 159 .default_advmss = ipv4_default_advmss, 160 .mtu = ipv4_mtu, 161 .cow_metrics = ipv4_cow_metrics, 162 .destroy = ipv4_dst_destroy, 163 .negative_advice = ipv4_negative_advice, 164 .link_failure = ipv4_link_failure, 165 .update_pmtu = ip_rt_update_pmtu, 166 .redirect = ip_do_redirect, 167 .local_out = __ip_local_out, 168 .neigh_lookup = ipv4_neigh_lookup, 169 .confirm_neigh = ipv4_confirm_neigh, 170 }; 171 172 #define ECN_OR_COST(class) TC_PRIO_##class 173 174 const __u8 ip_tos2prio[16] = { 175 TC_PRIO_BESTEFFORT, 176 ECN_OR_COST(BESTEFFORT), 177 TC_PRIO_BESTEFFORT, 178 ECN_OR_COST(BESTEFFORT), 179 TC_PRIO_BULK, 180 ECN_OR_COST(BULK), 181 TC_PRIO_BULK, 182 ECN_OR_COST(BULK), 183 TC_PRIO_INTERACTIVE, 184 ECN_OR_COST(INTERACTIVE), 185 TC_PRIO_INTERACTIVE, 186 ECN_OR_COST(INTERACTIVE), 187 TC_PRIO_INTERACTIVE_BULK, 188 ECN_OR_COST(INTERACTIVE_BULK), 189 TC_PRIO_INTERACTIVE_BULK, 190 ECN_OR_COST(INTERACTIVE_BULK) 191 }; 192 EXPORT_SYMBOL(ip_tos2prio); 193 194 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat); 195 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field) 196 197 #ifdef CONFIG_PROC_FS 198 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos) 199 { 200 if (*pos) 201 return NULL; 202 return SEQ_START_TOKEN; 203 } 204 205 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos) 206 { 207 ++*pos; 208 return NULL; 209 } 210 211 static void rt_cache_seq_stop(struct seq_file *seq, void *v) 212 { 213 } 214 215 static int rt_cache_seq_show(struct seq_file *seq, void *v) 216 { 217 if (v == SEQ_START_TOKEN) 218 seq_printf(seq, "%-127s\n", 219 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t" 220 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t" 221 "HHUptod\tSpecDst"); 222 return 0; 223 } 224 225 static const struct seq_operations rt_cache_seq_ops = { 226 .start = rt_cache_seq_start, 227 .next = rt_cache_seq_next, 228 .stop = rt_cache_seq_stop, 229 .show = rt_cache_seq_show, 230 }; 231 232 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos) 233 { 234 int cpu; 235 236 if (*pos == 0) 237 return SEQ_START_TOKEN; 238 239 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) { 240 if (!cpu_possible(cpu)) 241 continue; 242 *pos = cpu+1; 243 return &per_cpu(rt_cache_stat, cpu); 244 } 245 return NULL; 246 } 247 248 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos) 249 { 250 int cpu; 251 252 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) { 253 if (!cpu_possible(cpu)) 254 continue; 255 *pos = cpu+1; 256 return &per_cpu(rt_cache_stat, cpu); 257 } 258 (*pos)++; 259 return NULL; 260 261 } 262 263 static void rt_cpu_seq_stop(struct seq_file *seq, void *v) 264 { 265 266 } 267 268 static int rt_cpu_seq_show(struct seq_file *seq, void *v) 269 { 270 struct rt_cache_stat *st = v; 271 272 if (v == SEQ_START_TOKEN) { 273 seq_puts(seq, "entries in_hit in_slow_tot in_slow_mc in_no_route in_brd in_martian_dst in_martian_src out_hit out_slow_tot out_slow_mc gc_total gc_ignored gc_goal_miss gc_dst_overflow in_hlist_search out_hlist_search\n"); 274 return 0; 275 } 276 277 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x " 278 "%08x %08x %08x %08x %08x %08x " 279 "%08x %08x %08x %08x\n", 280 dst_entries_get_slow(&ipv4_dst_ops), 281 0, /* st->in_hit */ 282 st->in_slow_tot, 283 st->in_slow_mc, 284 st->in_no_route, 285 st->in_brd, 286 st->in_martian_dst, 287 st->in_martian_src, 288 289 0, /* st->out_hit */ 290 st->out_slow_tot, 291 st->out_slow_mc, 292 293 0, /* st->gc_total */ 294 0, /* st->gc_ignored */ 295 0, /* st->gc_goal_miss */ 296 0, /* st->gc_dst_overflow */ 297 0, /* st->in_hlist_search */ 298 0 /* st->out_hlist_search */ 299 ); 300 return 0; 301 } 302 303 static const struct seq_operations rt_cpu_seq_ops = { 304 .start = rt_cpu_seq_start, 305 .next = rt_cpu_seq_next, 306 .stop = rt_cpu_seq_stop, 307 .show = rt_cpu_seq_show, 308 }; 309 310 #ifdef CONFIG_IP_ROUTE_CLASSID 311 static int rt_acct_proc_show(struct seq_file *m, void *v) 312 { 313 struct ip_rt_acct *dst, *src; 314 unsigned int i, j; 315 316 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL); 317 if (!dst) 318 return -ENOMEM; 319 320 for_each_possible_cpu(i) { 321 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i); 322 for (j = 0; j < 256; j++) { 323 dst[j].o_bytes += src[j].o_bytes; 324 dst[j].o_packets += src[j].o_packets; 325 dst[j].i_bytes += src[j].i_bytes; 326 dst[j].i_packets += src[j].i_packets; 327 } 328 } 329 330 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct)); 331 kfree(dst); 332 return 0; 333 } 334 #endif 335 336 static int __net_init ip_rt_do_proc_init(struct net *net) 337 { 338 struct proc_dir_entry *pde; 339 340 pde = proc_create_seq("rt_cache", 0444, net->proc_net, 341 &rt_cache_seq_ops); 342 if (!pde) 343 goto err1; 344 345 pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat, 346 &rt_cpu_seq_ops); 347 if (!pde) 348 goto err2; 349 350 #ifdef CONFIG_IP_ROUTE_CLASSID 351 pde = proc_create_single("rt_acct", 0, net->proc_net, 352 rt_acct_proc_show); 353 if (!pde) 354 goto err3; 355 #endif 356 return 0; 357 358 #ifdef CONFIG_IP_ROUTE_CLASSID 359 err3: 360 remove_proc_entry("rt_cache", net->proc_net_stat); 361 #endif 362 err2: 363 remove_proc_entry("rt_cache", net->proc_net); 364 err1: 365 return -ENOMEM; 366 } 367 368 static void __net_exit ip_rt_do_proc_exit(struct net *net) 369 { 370 remove_proc_entry("rt_cache", net->proc_net_stat); 371 remove_proc_entry("rt_cache", net->proc_net); 372 #ifdef CONFIG_IP_ROUTE_CLASSID 373 remove_proc_entry("rt_acct", net->proc_net); 374 #endif 375 } 376 377 static struct pernet_operations ip_rt_proc_ops __net_initdata = { 378 .init = ip_rt_do_proc_init, 379 .exit = ip_rt_do_proc_exit, 380 }; 381 382 static int __init ip_rt_proc_init(void) 383 { 384 return register_pernet_subsys(&ip_rt_proc_ops); 385 } 386 387 #else 388 static inline int ip_rt_proc_init(void) 389 { 390 return 0; 391 } 392 #endif /* CONFIG_PROC_FS */ 393 394 static inline bool rt_is_expired(const struct rtable *rth) 395 { 396 bool res; 397 398 rcu_read_lock(); 399 res = rth->rt_genid != rt_genid_ipv4(dev_net_rcu(rth->dst.dev)); 400 rcu_read_unlock(); 401 402 return res; 403 } 404 405 void rt_cache_flush(struct net *net) 406 { 407 rt_genid_bump_ipv4(net); 408 } 409 410 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 411 struct sk_buff *skb, 412 const void *daddr) 413 { 414 const struct rtable *rt = container_of(dst, struct rtable, dst); 415 struct net_device *dev = dst->dev; 416 struct neighbour *n; 417 418 rcu_read_lock(); 419 420 if (likely(rt->rt_gw_family == AF_INET)) { 421 n = ip_neigh_gw4(dev, rt->rt_gw4); 422 } else if (rt->rt_gw_family == AF_INET6) { 423 n = ip_neigh_gw6(dev, &rt->rt_gw6); 424 } else { 425 __be32 pkey; 426 427 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr); 428 n = ip_neigh_gw4(dev, pkey); 429 } 430 431 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt)) 432 n = NULL; 433 434 rcu_read_unlock(); 435 436 return n; 437 } 438 439 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr) 440 { 441 const struct rtable *rt = container_of(dst, struct rtable, dst); 442 struct net_device *dev = dst->dev; 443 const __be32 *pkey = daddr; 444 445 if (rt->rt_gw_family == AF_INET) { 446 pkey = (const __be32 *)&rt->rt_gw4; 447 } else if (rt->rt_gw_family == AF_INET6) { 448 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6); 449 } else if (!daddr || 450 (rt->rt_flags & 451 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) { 452 return; 453 } 454 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey); 455 } 456 457 /* Hash tables of size 2048..262144 depending on RAM size. 458 * Each bucket uses 8 bytes. 459 */ 460 static u32 ip_idents_mask __read_mostly; 461 static atomic_t *ip_idents __read_mostly; 462 static u32 *ip_tstamps __read_mostly; 463 464 /* In order to protect privacy, we add a perturbation to identifiers 465 * if one generator is seldom used. This makes hard for an attacker 466 * to infer how many packets were sent between two points in time. 467 */ 468 static u32 ip_idents_reserve(u32 hash, int segs) 469 { 470 u32 bucket, old, now = (u32)jiffies; 471 atomic_t *p_id; 472 u32 *p_tstamp; 473 u32 delta = 0; 474 475 bucket = hash & ip_idents_mask; 476 p_tstamp = ip_tstamps + bucket; 477 p_id = ip_idents + bucket; 478 old = READ_ONCE(*p_tstamp); 479 480 if (old != now && cmpxchg(p_tstamp, old, now) == old) 481 delta = get_random_u32_below(now - old); 482 483 /* If UBSAN reports an error there, please make sure your compiler 484 * supports -fno-strict-overflow before reporting it that was a bug 485 * in UBSAN, and it has been fixed in GCC-8. 486 */ 487 return atomic_add_return(segs + delta, p_id) - segs; 488 } 489 490 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs) 491 { 492 u32 hash, id; 493 494 /* Note the following code is not safe, but this is okay. */ 495 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key))) 496 get_random_bytes(&net->ipv4.ip_id_key, 497 sizeof(net->ipv4.ip_id_key)); 498 499 hash = siphash_3u32((__force u32)iph->daddr, 500 (__force u32)iph->saddr, 501 iph->protocol, 502 &net->ipv4.ip_id_key); 503 id = ip_idents_reserve(hash, segs); 504 iph->id = htons(id); 505 } 506 EXPORT_SYMBOL(__ip_select_ident); 507 508 static void ip_rt_fix_tos(struct flowi4 *fl4) 509 { 510 __u8 tos = RT_FL_TOS(fl4); 511 512 fl4->flowi4_tos = tos & IPTOS_RT_MASK; 513 if (tos & RTO_ONLINK) 514 fl4->flowi4_scope = RT_SCOPE_LINK; 515 } 516 517 static void __build_flow_key(const struct net *net, struct flowi4 *fl4, 518 const struct sock *sk, const struct iphdr *iph, 519 int oif, __u8 tos, u8 prot, u32 mark, 520 int flow_flags) 521 { 522 __u8 scope = RT_SCOPE_UNIVERSE; 523 524 if (sk) { 525 oif = sk->sk_bound_dev_if; 526 mark = READ_ONCE(sk->sk_mark); 527 tos = ip_sock_rt_tos(sk); 528 scope = ip_sock_rt_scope(sk); 529 prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW : 530 sk->sk_protocol; 531 } 532 533 flowi4_init_output(fl4, oif, mark, tos & IPTOS_RT_MASK, scope, 534 prot, flow_flags, iph->daddr, iph->saddr, 0, 0, 535 sock_net_uid(net, sk)); 536 } 537 538 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb, 539 const struct sock *sk) 540 { 541 const struct net *net = dev_net(skb->dev); 542 const struct iphdr *iph = ip_hdr(skb); 543 int oif = skb->dev->ifindex; 544 u8 prot = iph->protocol; 545 u32 mark = skb->mark; 546 __u8 tos = iph->tos; 547 548 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0); 549 } 550 551 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk) 552 { 553 const struct inet_sock *inet = inet_sk(sk); 554 const struct ip_options_rcu *inet_opt; 555 __be32 daddr = inet->inet_daddr; 556 557 rcu_read_lock(); 558 inet_opt = rcu_dereference(inet->inet_opt); 559 if (inet_opt && inet_opt->opt.srr) 560 daddr = inet_opt->opt.faddr; 561 flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark), 562 ip_sock_rt_tos(sk) & IPTOS_RT_MASK, 563 ip_sock_rt_scope(sk), 564 inet_test_bit(HDRINCL, sk) ? 565 IPPROTO_RAW : sk->sk_protocol, 566 inet_sk_flowi_flags(sk), 567 daddr, inet->inet_saddr, 0, 0, sk->sk_uid); 568 rcu_read_unlock(); 569 } 570 571 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk, 572 const struct sk_buff *skb) 573 { 574 if (skb) 575 build_skb_flow_key(fl4, skb, sk); 576 else 577 build_sk_flow_key(fl4, sk); 578 } 579 580 static DEFINE_SPINLOCK(fnhe_lock); 581 582 static void fnhe_flush_routes(struct fib_nh_exception *fnhe) 583 { 584 struct rtable *rt; 585 586 rt = rcu_dereference(fnhe->fnhe_rth_input); 587 if (rt) { 588 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL); 589 dst_dev_put(&rt->dst); 590 dst_release(&rt->dst); 591 } 592 rt = rcu_dereference(fnhe->fnhe_rth_output); 593 if (rt) { 594 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL); 595 dst_dev_put(&rt->dst); 596 dst_release(&rt->dst); 597 } 598 } 599 600 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash) 601 { 602 struct fib_nh_exception __rcu **fnhe_p, **oldest_p; 603 struct fib_nh_exception *fnhe, *oldest = NULL; 604 605 for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) { 606 fnhe = rcu_dereference_protected(*fnhe_p, 607 lockdep_is_held(&fnhe_lock)); 608 if (!fnhe) 609 break; 610 if (!oldest || 611 time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) { 612 oldest = fnhe; 613 oldest_p = fnhe_p; 614 } 615 } 616 fnhe_flush_routes(oldest); 617 *oldest_p = oldest->fnhe_next; 618 kfree_rcu(oldest, rcu); 619 } 620 621 static u32 fnhe_hashfun(__be32 daddr) 622 { 623 static siphash_aligned_key_t fnhe_hash_key; 624 u64 hval; 625 626 net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key)); 627 hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key); 628 return hash_64(hval, FNHE_HASH_SHIFT); 629 } 630 631 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe) 632 { 633 rt->rt_pmtu = fnhe->fnhe_pmtu; 634 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked; 635 rt->dst.expires = fnhe->fnhe_expires; 636 637 if (fnhe->fnhe_gw) { 638 rt->rt_flags |= RTCF_REDIRECTED; 639 rt->rt_uses_gateway = 1; 640 rt->rt_gw_family = AF_INET; 641 rt->rt_gw4 = fnhe->fnhe_gw; 642 } 643 } 644 645 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr, 646 __be32 gw, u32 pmtu, bool lock, 647 unsigned long expires) 648 { 649 struct fnhe_hash_bucket *hash; 650 struct fib_nh_exception *fnhe; 651 struct rtable *rt; 652 u32 genid, hval; 653 unsigned int i; 654 int depth; 655 656 genid = fnhe_genid(dev_net(nhc->nhc_dev)); 657 hval = fnhe_hashfun(daddr); 658 659 spin_lock_bh(&fnhe_lock); 660 661 hash = rcu_dereference(nhc->nhc_exceptions); 662 if (!hash) { 663 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC); 664 if (!hash) 665 goto out_unlock; 666 rcu_assign_pointer(nhc->nhc_exceptions, hash); 667 } 668 669 hash += hval; 670 671 depth = 0; 672 for (fnhe = rcu_dereference(hash->chain); fnhe; 673 fnhe = rcu_dereference(fnhe->fnhe_next)) { 674 if (fnhe->fnhe_daddr == daddr) 675 break; 676 depth++; 677 } 678 679 if (fnhe) { 680 if (fnhe->fnhe_genid != genid) 681 fnhe->fnhe_genid = genid; 682 if (gw) 683 fnhe->fnhe_gw = gw; 684 if (pmtu) { 685 fnhe->fnhe_pmtu = pmtu; 686 fnhe->fnhe_mtu_locked = lock; 687 } 688 fnhe->fnhe_expires = max(1UL, expires); 689 /* Update all cached dsts too */ 690 rt = rcu_dereference(fnhe->fnhe_rth_input); 691 if (rt) 692 fill_route_from_fnhe(rt, fnhe); 693 rt = rcu_dereference(fnhe->fnhe_rth_output); 694 if (rt) 695 fill_route_from_fnhe(rt, fnhe); 696 } else { 697 /* Randomize max depth to avoid some side channels attacks. */ 698 int max_depth = FNHE_RECLAIM_DEPTH + 699 get_random_u32_below(FNHE_RECLAIM_DEPTH); 700 701 while (depth > max_depth) { 702 fnhe_remove_oldest(hash); 703 depth--; 704 } 705 706 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC); 707 if (!fnhe) 708 goto out_unlock; 709 710 fnhe->fnhe_next = hash->chain; 711 712 fnhe->fnhe_genid = genid; 713 fnhe->fnhe_daddr = daddr; 714 fnhe->fnhe_gw = gw; 715 fnhe->fnhe_pmtu = pmtu; 716 fnhe->fnhe_mtu_locked = lock; 717 fnhe->fnhe_expires = max(1UL, expires); 718 719 rcu_assign_pointer(hash->chain, fnhe); 720 721 /* Exception created; mark the cached routes for the nexthop 722 * stale, so anyone caching it rechecks if this exception 723 * applies to them. 724 */ 725 rt = rcu_dereference(nhc->nhc_rth_input); 726 if (rt) 727 rt->dst.obsolete = DST_OBSOLETE_KILL; 728 729 for_each_possible_cpu(i) { 730 struct rtable __rcu **prt; 731 732 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i); 733 rt = rcu_dereference(*prt); 734 if (rt) 735 rt->dst.obsolete = DST_OBSOLETE_KILL; 736 } 737 } 738 739 fnhe->fnhe_stamp = jiffies; 740 741 out_unlock: 742 spin_unlock_bh(&fnhe_lock); 743 } 744 745 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4, 746 bool kill_route) 747 { 748 __be32 new_gw = icmp_hdr(skb)->un.gateway; 749 __be32 old_gw = ip_hdr(skb)->saddr; 750 struct net_device *dev = skb->dev; 751 struct in_device *in_dev; 752 struct fib_result res; 753 struct neighbour *n; 754 struct net *net; 755 756 switch (icmp_hdr(skb)->code & 7) { 757 case ICMP_REDIR_NET: 758 case ICMP_REDIR_NETTOS: 759 case ICMP_REDIR_HOST: 760 case ICMP_REDIR_HOSTTOS: 761 break; 762 763 default: 764 return; 765 } 766 767 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw) 768 return; 769 770 in_dev = __in_dev_get_rcu(dev); 771 if (!in_dev) 772 return; 773 774 net = dev_net(dev); 775 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) || 776 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) || 777 ipv4_is_zeronet(new_gw)) 778 goto reject_redirect; 779 780 if (!IN_DEV_SHARED_MEDIA(in_dev)) { 781 if (!inet_addr_onlink(in_dev, new_gw, old_gw)) 782 goto reject_redirect; 783 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) 784 goto reject_redirect; 785 } else { 786 if (inet_addr_type(net, new_gw) != RTN_UNICAST) 787 goto reject_redirect; 788 } 789 790 n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw); 791 if (!n) 792 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev); 793 if (!IS_ERR(n)) { 794 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) { 795 neigh_event_send(n, NULL); 796 } else { 797 if (fib_lookup(net, fl4, &res, 0) == 0) { 798 struct fib_nh_common *nhc; 799 800 fib_select_path(net, &res, fl4, skb); 801 nhc = FIB_RES_NHC(res); 802 update_or_create_fnhe(nhc, fl4->daddr, new_gw, 803 0, false, 804 jiffies + ip_rt_gc_timeout); 805 } 806 if (kill_route) 807 rt->dst.obsolete = DST_OBSOLETE_KILL; 808 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n); 809 } 810 neigh_release(n); 811 } 812 return; 813 814 reject_redirect: 815 #ifdef CONFIG_IP_ROUTE_VERBOSE 816 if (IN_DEV_LOG_MARTIANS(in_dev)) { 817 const struct iphdr *iph = (const struct iphdr *) skb->data; 818 __be32 daddr = iph->daddr; 819 __be32 saddr = iph->saddr; 820 821 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n" 822 " Advised path = %pI4 -> %pI4\n", 823 &old_gw, dev->name, &new_gw, 824 &saddr, &daddr); 825 } 826 #endif 827 ; 828 } 829 830 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb) 831 { 832 struct rtable *rt; 833 struct flowi4 fl4; 834 const struct iphdr *iph = (const struct iphdr *) skb->data; 835 struct net *net = dev_net(skb->dev); 836 int oif = skb->dev->ifindex; 837 u8 prot = iph->protocol; 838 u32 mark = skb->mark; 839 __u8 tos = iph->tos; 840 841 rt = (struct rtable *) dst; 842 843 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0); 844 __ip_do_redirect(rt, skb, &fl4, true); 845 } 846 847 static void ipv4_negative_advice(struct sock *sk, 848 struct dst_entry *dst) 849 { 850 struct rtable *rt = (struct rtable *)dst; 851 852 if ((dst->obsolete > 0) || 853 (rt->rt_flags & RTCF_REDIRECTED) || 854 rt->dst.expires) 855 sk_dst_reset(sk); 856 } 857 858 /* 859 * Algorithm: 860 * 1. The first ip_rt_redirect_number redirects are sent 861 * with exponential backoff, then we stop sending them at all, 862 * assuming that the host ignores our redirects. 863 * 2. If we did not see packets requiring redirects 864 * during ip_rt_redirect_silence, we assume that the host 865 * forgot redirected route and start to send redirects again. 866 * 867 * This algorithm is much cheaper and more intelligent than dumb load limiting 868 * in icmp.c. 869 * 870 * NOTE. Do not forget to inhibit load limiting for redirects (redundant) 871 * and "frag. need" (breaks PMTU discovery) in icmp.c. 872 */ 873 874 void ip_rt_send_redirect(struct sk_buff *skb) 875 { 876 struct rtable *rt = skb_rtable(skb); 877 struct in_device *in_dev; 878 struct inet_peer *peer; 879 struct net *net; 880 int log_martians; 881 int vif; 882 883 rcu_read_lock(); 884 in_dev = __in_dev_get_rcu(rt->dst.dev); 885 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) { 886 rcu_read_unlock(); 887 return; 888 } 889 log_martians = IN_DEV_LOG_MARTIANS(in_dev); 890 vif = l3mdev_master_ifindex_rcu(rt->dst.dev); 891 892 net = dev_net(rt->dst.dev); 893 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif); 894 if (!peer) { 895 rcu_read_unlock(); 896 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, 897 rt_nexthop(rt, ip_hdr(skb)->daddr)); 898 return; 899 } 900 901 /* No redirected packets during ip_rt_redirect_silence; 902 * reset the algorithm. 903 */ 904 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) { 905 peer->rate_tokens = 0; 906 peer->n_redirects = 0; 907 } 908 909 /* Too many ignored redirects; do not send anything 910 * set dst.rate_last to the last seen redirected packet. 911 */ 912 if (peer->n_redirects >= ip_rt_redirect_number) { 913 peer->rate_last = jiffies; 914 goto out_unlock; 915 } 916 917 /* Check for load limit; set rate_last to the latest sent 918 * redirect. 919 */ 920 if (peer->n_redirects == 0 || 921 time_after(jiffies, 922 (peer->rate_last + 923 (ip_rt_redirect_load << peer->n_redirects)))) { 924 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr); 925 926 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw); 927 peer->rate_last = jiffies; 928 ++peer->n_redirects; 929 if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians && 930 peer->n_redirects == ip_rt_redirect_number) 931 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n", 932 &ip_hdr(skb)->saddr, inet_iif(skb), 933 &ip_hdr(skb)->daddr, &gw); 934 } 935 out_unlock: 936 rcu_read_unlock(); 937 } 938 939 static int ip_error(struct sk_buff *skb) 940 { 941 struct rtable *rt = skb_rtable(skb); 942 struct net_device *dev = skb->dev; 943 struct in_device *in_dev; 944 struct inet_peer *peer; 945 unsigned long now; 946 struct net *net; 947 SKB_DR(reason); 948 bool send; 949 int code; 950 951 if (netif_is_l3_master(skb->dev)) { 952 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif); 953 if (!dev) 954 goto out; 955 } 956 957 in_dev = __in_dev_get_rcu(dev); 958 959 /* IP on this device is disabled. */ 960 if (!in_dev) 961 goto out; 962 963 net = dev_net(rt->dst.dev); 964 if (!IN_DEV_FORWARD(in_dev)) { 965 switch (rt->dst.error) { 966 case EHOSTUNREACH: 967 SKB_DR_SET(reason, IP_INADDRERRORS); 968 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS); 969 break; 970 971 case ENETUNREACH: 972 SKB_DR_SET(reason, IP_INNOROUTES); 973 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 974 break; 975 } 976 goto out; 977 } 978 979 switch (rt->dst.error) { 980 case EINVAL: 981 default: 982 goto out; 983 case EHOSTUNREACH: 984 code = ICMP_HOST_UNREACH; 985 break; 986 case ENETUNREACH: 987 code = ICMP_NET_UNREACH; 988 SKB_DR_SET(reason, IP_INNOROUTES); 989 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 990 break; 991 case EACCES: 992 code = ICMP_PKT_FILTERED; 993 break; 994 } 995 996 rcu_read_lock(); 997 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 998 l3mdev_master_ifindex_rcu(skb->dev)); 999 send = true; 1000 if (peer) { 1001 now = jiffies; 1002 peer->rate_tokens += now - peer->rate_last; 1003 if (peer->rate_tokens > ip_rt_error_burst) 1004 peer->rate_tokens = ip_rt_error_burst; 1005 peer->rate_last = now; 1006 if (peer->rate_tokens >= ip_rt_error_cost) 1007 peer->rate_tokens -= ip_rt_error_cost; 1008 else 1009 send = false; 1010 } 1011 rcu_read_unlock(); 1012 1013 if (send) 1014 icmp_send(skb, ICMP_DEST_UNREACH, code, 0); 1015 1016 out: kfree_skb_reason(skb, reason); 1017 return 0; 1018 } 1019 1020 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu) 1021 { 1022 struct dst_entry *dst = &rt->dst; 1023 struct fib_result res; 1024 bool lock = false; 1025 struct net *net; 1026 u32 old_mtu; 1027 1028 if (ip_mtu_locked(dst)) 1029 return; 1030 1031 old_mtu = ipv4_mtu(dst); 1032 if (old_mtu < mtu) 1033 return; 1034 1035 rcu_read_lock(); 1036 net = dev_net_rcu(dst->dev); 1037 if (mtu < net->ipv4.ip_rt_min_pmtu) { 1038 lock = true; 1039 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu); 1040 } 1041 1042 if (rt->rt_pmtu == mtu && !lock && 1043 time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2)) 1044 goto out; 1045 1046 if (fib_lookup(net, fl4, &res, 0) == 0) { 1047 struct fib_nh_common *nhc; 1048 1049 fib_select_path(net, &res, fl4, NULL); 1050 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1051 if (fib_info_num_path(res.fi) > 1) { 1052 int nhsel; 1053 1054 for (nhsel = 0; nhsel < fib_info_num_path(res.fi); nhsel++) { 1055 nhc = fib_info_nhc(res.fi, nhsel); 1056 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1057 jiffies + net->ipv4.ip_rt_mtu_expires); 1058 } 1059 goto out; 1060 } 1061 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 1062 nhc = FIB_RES_NHC(res); 1063 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1064 jiffies + net->ipv4.ip_rt_mtu_expires); 1065 } 1066 out: 1067 rcu_read_unlock(); 1068 } 1069 1070 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 1071 struct sk_buff *skb, u32 mtu, 1072 bool confirm_neigh) 1073 { 1074 struct rtable *rt = (struct rtable *) dst; 1075 struct flowi4 fl4; 1076 1077 ip_rt_build_flow_key(&fl4, sk, skb); 1078 1079 /* Don't make lookup fail for bridged encapsulations */ 1080 if (skb && netif_is_any_bridge_port(skb->dev)) 1081 fl4.flowi4_oif = 0; 1082 1083 __ip_rt_update_pmtu(rt, &fl4, mtu); 1084 } 1085 1086 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, 1087 int oif, u8 protocol) 1088 { 1089 const struct iphdr *iph = (const struct iphdr *)skb->data; 1090 struct flowi4 fl4; 1091 struct rtable *rt; 1092 u32 mark = IP4_REPLY_MARK(net, skb->mark); 1093 1094 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark, 1095 0); 1096 rt = __ip_route_output_key(net, &fl4); 1097 if (!IS_ERR(rt)) { 1098 __ip_rt_update_pmtu(rt, &fl4, mtu); 1099 ip_rt_put(rt); 1100 } 1101 } 1102 EXPORT_SYMBOL_GPL(ipv4_update_pmtu); 1103 1104 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1105 { 1106 const struct iphdr *iph = (const struct iphdr *)skb->data; 1107 struct flowi4 fl4; 1108 struct rtable *rt; 1109 1110 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0); 1111 1112 if (!fl4.flowi4_mark) 1113 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark); 1114 1115 rt = __ip_route_output_key(sock_net(sk), &fl4); 1116 if (!IS_ERR(rt)) { 1117 __ip_rt_update_pmtu(rt, &fl4, mtu); 1118 ip_rt_put(rt); 1119 } 1120 } 1121 1122 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1123 { 1124 const struct iphdr *iph = (const struct iphdr *)skb->data; 1125 struct flowi4 fl4; 1126 struct rtable *rt; 1127 struct dst_entry *odst = NULL; 1128 bool new = false; 1129 struct net *net = sock_net(sk); 1130 1131 bh_lock_sock(sk); 1132 1133 if (!ip_sk_accept_pmtu(sk)) 1134 goto out; 1135 1136 odst = sk_dst_get(sk); 1137 1138 if (sock_owned_by_user(sk) || !odst) { 1139 __ipv4_sk_update_pmtu(skb, sk, mtu); 1140 goto out; 1141 } 1142 1143 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1144 1145 rt = (struct rtable *)odst; 1146 if (odst->obsolete && !odst->ops->check(odst, 0)) { 1147 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1148 if (IS_ERR(rt)) 1149 goto out; 1150 1151 new = true; 1152 } 1153 1154 __ip_rt_update_pmtu((struct rtable *)xfrm_dst_path(&rt->dst), &fl4, mtu); 1155 1156 if (!dst_check(&rt->dst, 0)) { 1157 if (new) 1158 dst_release(&rt->dst); 1159 1160 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1161 if (IS_ERR(rt)) 1162 goto out; 1163 1164 new = true; 1165 } 1166 1167 if (new) 1168 sk_dst_set(sk, &rt->dst); 1169 1170 out: 1171 bh_unlock_sock(sk); 1172 dst_release(odst); 1173 } 1174 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu); 1175 1176 void ipv4_redirect(struct sk_buff *skb, struct net *net, 1177 int oif, u8 protocol) 1178 { 1179 const struct iphdr *iph = (const struct iphdr *)skb->data; 1180 struct flowi4 fl4; 1181 struct rtable *rt; 1182 1183 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0); 1184 rt = __ip_route_output_key(net, &fl4); 1185 if (!IS_ERR(rt)) { 1186 __ip_do_redirect(rt, skb, &fl4, false); 1187 ip_rt_put(rt); 1188 } 1189 } 1190 EXPORT_SYMBOL_GPL(ipv4_redirect); 1191 1192 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk) 1193 { 1194 const struct iphdr *iph = (const struct iphdr *)skb->data; 1195 struct flowi4 fl4; 1196 struct rtable *rt; 1197 struct net *net = sock_net(sk); 1198 1199 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1200 rt = __ip_route_output_key(net, &fl4); 1201 if (!IS_ERR(rt)) { 1202 __ip_do_redirect(rt, skb, &fl4, false); 1203 ip_rt_put(rt); 1204 } 1205 } 1206 EXPORT_SYMBOL_GPL(ipv4_sk_redirect); 1207 1208 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst, 1209 u32 cookie) 1210 { 1211 struct rtable *rt = (struct rtable *) dst; 1212 1213 /* All IPV4 dsts are created with ->obsolete set to the value 1214 * DST_OBSOLETE_FORCE_CHK which forces validation calls down 1215 * into this function always. 1216 * 1217 * When a PMTU/redirect information update invalidates a route, 1218 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or 1219 * DST_OBSOLETE_DEAD. 1220 */ 1221 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt)) 1222 return NULL; 1223 return dst; 1224 } 1225 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check); 1226 1227 static void ipv4_send_dest_unreach(struct sk_buff *skb) 1228 { 1229 struct net_device *dev; 1230 struct ip_options opt; 1231 int res; 1232 1233 /* Recompile ip options since IPCB may not be valid anymore. 1234 * Also check we have a reasonable ipv4 header. 1235 */ 1236 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) || 1237 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5) 1238 return; 1239 1240 memset(&opt, 0, sizeof(opt)); 1241 if (ip_hdr(skb)->ihl > 5) { 1242 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4)) 1243 return; 1244 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr); 1245 1246 rcu_read_lock(); 1247 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev; 1248 res = __ip_options_compile(dev_net(dev), &opt, skb, NULL); 1249 rcu_read_unlock(); 1250 1251 if (res) 1252 return; 1253 } 1254 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt); 1255 } 1256 1257 static void ipv4_link_failure(struct sk_buff *skb) 1258 { 1259 struct rtable *rt; 1260 1261 ipv4_send_dest_unreach(skb); 1262 1263 rt = skb_rtable(skb); 1264 if (rt) 1265 dst_set_expires(&rt->dst, 0); 1266 } 1267 1268 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb) 1269 { 1270 pr_debug("%s: %pI4 -> %pI4, %s\n", 1271 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1272 skb->dev ? skb->dev->name : "?"); 1273 kfree_skb(skb); 1274 WARN_ON(1); 1275 return 0; 1276 } 1277 1278 /* 1279 * We do not cache source address of outgoing interface, 1280 * because it is used only by IP RR, TS and SRR options, 1281 * so that it out of fast path. 1282 * 1283 * BTW remember: "addr" is allowed to be not aligned 1284 * in IP options! 1285 */ 1286 1287 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt) 1288 { 1289 __be32 src; 1290 1291 if (rt_is_output_route(rt)) 1292 src = ip_hdr(skb)->saddr; 1293 else { 1294 struct fib_result res; 1295 struct iphdr *iph = ip_hdr(skb); 1296 struct flowi4 fl4 = { 1297 .daddr = iph->daddr, 1298 .saddr = iph->saddr, 1299 .flowi4_tos = iph->tos & IPTOS_RT_MASK, 1300 .flowi4_oif = rt->dst.dev->ifindex, 1301 .flowi4_iif = skb->dev->ifindex, 1302 .flowi4_mark = skb->mark, 1303 }; 1304 1305 rcu_read_lock(); 1306 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0) 1307 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res); 1308 else 1309 src = inet_select_addr(rt->dst.dev, 1310 rt_nexthop(rt, iph->daddr), 1311 RT_SCOPE_UNIVERSE); 1312 rcu_read_unlock(); 1313 } 1314 memcpy(addr, &src, 4); 1315 } 1316 1317 #ifdef CONFIG_IP_ROUTE_CLASSID 1318 static void set_class_tag(struct rtable *rt, u32 tag) 1319 { 1320 if (!(rt->dst.tclassid & 0xFFFF)) 1321 rt->dst.tclassid |= tag & 0xFFFF; 1322 if (!(rt->dst.tclassid & 0xFFFF0000)) 1323 rt->dst.tclassid |= tag & 0xFFFF0000; 1324 } 1325 #endif 1326 1327 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1328 { 1329 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr); 1330 unsigned int advmss; 1331 struct net *net; 1332 1333 rcu_read_lock(); 1334 net = dev_net_rcu(dst->dev); 1335 advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size, 1336 net->ipv4.ip_rt_min_advmss); 1337 rcu_read_unlock(); 1338 1339 return min(advmss, IPV4_MAX_PMTU - header_size); 1340 } 1341 1342 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst) 1343 { 1344 return ip_dst_mtu_maybe_forward(dst, false); 1345 } 1346 EXPORT_INDIRECT_CALLABLE(ipv4_mtu); 1347 1348 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr) 1349 { 1350 struct fnhe_hash_bucket *hash; 1351 struct fib_nh_exception *fnhe, __rcu **fnhe_p; 1352 u32 hval = fnhe_hashfun(daddr); 1353 1354 spin_lock_bh(&fnhe_lock); 1355 1356 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1357 lockdep_is_held(&fnhe_lock)); 1358 hash += hval; 1359 1360 fnhe_p = &hash->chain; 1361 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock)); 1362 while (fnhe) { 1363 if (fnhe->fnhe_daddr == daddr) { 1364 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected( 1365 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock))); 1366 /* set fnhe_daddr to 0 to ensure it won't bind with 1367 * new dsts in rt_bind_exception(). 1368 */ 1369 fnhe->fnhe_daddr = 0; 1370 fnhe_flush_routes(fnhe); 1371 kfree_rcu(fnhe, rcu); 1372 break; 1373 } 1374 fnhe_p = &fnhe->fnhe_next; 1375 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1376 lockdep_is_held(&fnhe_lock)); 1377 } 1378 1379 spin_unlock_bh(&fnhe_lock); 1380 } 1381 1382 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc, 1383 __be32 daddr) 1384 { 1385 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions); 1386 struct fib_nh_exception *fnhe; 1387 u32 hval; 1388 1389 if (!hash) 1390 return NULL; 1391 1392 hval = fnhe_hashfun(daddr); 1393 1394 for (fnhe = rcu_dereference(hash[hval].chain); fnhe; 1395 fnhe = rcu_dereference(fnhe->fnhe_next)) { 1396 if (fnhe->fnhe_daddr == daddr) { 1397 if (fnhe->fnhe_expires && 1398 time_after(jiffies, fnhe->fnhe_expires)) { 1399 ip_del_fnhe(nhc, daddr); 1400 break; 1401 } 1402 return fnhe; 1403 } 1404 } 1405 return NULL; 1406 } 1407 1408 /* MTU selection: 1409 * 1. mtu on route is locked - use it 1410 * 2. mtu from nexthop exception 1411 * 3. mtu from egress device 1412 */ 1413 1414 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr) 1415 { 1416 struct fib_nh_common *nhc = res->nhc; 1417 struct net_device *dev = nhc->nhc_dev; 1418 struct fib_info *fi = res->fi; 1419 u32 mtu = 0; 1420 1421 if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) || 1422 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU)) 1423 mtu = fi->fib_mtu; 1424 1425 if (likely(!mtu)) { 1426 struct fib_nh_exception *fnhe; 1427 1428 fnhe = find_exception(nhc, daddr); 1429 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires)) 1430 mtu = fnhe->fnhe_pmtu; 1431 } 1432 1433 if (likely(!mtu)) 1434 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU); 1435 1436 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu); 1437 } 1438 1439 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, 1440 __be32 daddr, const bool do_cache) 1441 { 1442 bool ret = false; 1443 1444 spin_lock_bh(&fnhe_lock); 1445 1446 if (daddr == fnhe->fnhe_daddr) { 1447 struct rtable __rcu **porig; 1448 struct rtable *orig; 1449 int genid = fnhe_genid(dev_net(rt->dst.dev)); 1450 1451 if (rt_is_input_route(rt)) 1452 porig = &fnhe->fnhe_rth_input; 1453 else 1454 porig = &fnhe->fnhe_rth_output; 1455 orig = rcu_dereference(*porig); 1456 1457 if (fnhe->fnhe_genid != genid) { 1458 fnhe->fnhe_genid = genid; 1459 fnhe->fnhe_gw = 0; 1460 fnhe->fnhe_pmtu = 0; 1461 fnhe->fnhe_expires = 0; 1462 fnhe->fnhe_mtu_locked = false; 1463 fnhe_flush_routes(fnhe); 1464 orig = NULL; 1465 } 1466 fill_route_from_fnhe(rt, fnhe); 1467 if (!rt->rt_gw4) { 1468 rt->rt_gw4 = daddr; 1469 rt->rt_gw_family = AF_INET; 1470 } 1471 1472 if (do_cache) { 1473 dst_hold(&rt->dst); 1474 rcu_assign_pointer(*porig, rt); 1475 if (orig) { 1476 dst_dev_put(&orig->dst); 1477 dst_release(&orig->dst); 1478 } 1479 ret = true; 1480 } 1481 1482 fnhe->fnhe_stamp = jiffies; 1483 } 1484 spin_unlock_bh(&fnhe_lock); 1485 1486 return ret; 1487 } 1488 1489 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt) 1490 { 1491 struct rtable *orig, *prev, **p; 1492 bool ret = true; 1493 1494 if (rt_is_input_route(rt)) { 1495 p = (struct rtable **)&nhc->nhc_rth_input; 1496 } else { 1497 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 1498 } 1499 orig = *p; 1500 1501 /* hold dst before doing cmpxchg() to avoid race condition 1502 * on this dst 1503 */ 1504 dst_hold(&rt->dst); 1505 prev = cmpxchg(p, orig, rt); 1506 if (prev == orig) { 1507 if (orig) { 1508 rt_add_uncached_list(orig); 1509 dst_release(&orig->dst); 1510 } 1511 } else { 1512 dst_release(&rt->dst); 1513 ret = false; 1514 } 1515 1516 return ret; 1517 } 1518 1519 struct uncached_list { 1520 spinlock_t lock; 1521 struct list_head head; 1522 struct list_head quarantine; 1523 }; 1524 1525 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list); 1526 1527 void rt_add_uncached_list(struct rtable *rt) 1528 { 1529 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list); 1530 1531 rt->dst.rt_uncached_list = ul; 1532 1533 spin_lock_bh(&ul->lock); 1534 list_add_tail(&rt->dst.rt_uncached, &ul->head); 1535 spin_unlock_bh(&ul->lock); 1536 } 1537 1538 void rt_del_uncached_list(struct rtable *rt) 1539 { 1540 if (!list_empty(&rt->dst.rt_uncached)) { 1541 struct uncached_list *ul = rt->dst.rt_uncached_list; 1542 1543 spin_lock_bh(&ul->lock); 1544 list_del_init(&rt->dst.rt_uncached); 1545 spin_unlock_bh(&ul->lock); 1546 } 1547 } 1548 1549 static void ipv4_dst_destroy(struct dst_entry *dst) 1550 { 1551 struct rtable *rt = (struct rtable *)dst; 1552 1553 ip_dst_metrics_put(dst); 1554 rt_del_uncached_list(rt); 1555 } 1556 1557 void rt_flush_dev(struct net_device *dev) 1558 { 1559 struct rtable *rt, *safe; 1560 int cpu; 1561 1562 for_each_possible_cpu(cpu) { 1563 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 1564 1565 if (list_empty(&ul->head)) 1566 continue; 1567 1568 spin_lock_bh(&ul->lock); 1569 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) { 1570 if (rt->dst.dev != dev) 1571 continue; 1572 rt->dst.dev = blackhole_netdev; 1573 netdev_ref_replace(dev, blackhole_netdev, 1574 &rt->dst.dev_tracker, GFP_ATOMIC); 1575 list_move(&rt->dst.rt_uncached, &ul->quarantine); 1576 } 1577 spin_unlock_bh(&ul->lock); 1578 } 1579 } 1580 1581 static bool rt_cache_valid(const struct rtable *rt) 1582 { 1583 return rt && 1584 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK && 1585 !rt_is_expired(rt); 1586 } 1587 1588 static void rt_set_nexthop(struct rtable *rt, __be32 daddr, 1589 const struct fib_result *res, 1590 struct fib_nh_exception *fnhe, 1591 struct fib_info *fi, u16 type, u32 itag, 1592 const bool do_cache) 1593 { 1594 bool cached = false; 1595 1596 if (fi) { 1597 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1598 1599 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) { 1600 rt->rt_uses_gateway = 1; 1601 rt->rt_gw_family = nhc->nhc_gw_family; 1602 /* only INET and INET6 are supported */ 1603 if (likely(nhc->nhc_gw_family == AF_INET)) 1604 rt->rt_gw4 = nhc->nhc_gw.ipv4; 1605 else 1606 rt->rt_gw6 = nhc->nhc_gw.ipv6; 1607 } 1608 1609 ip_dst_init_metrics(&rt->dst, fi->fib_metrics); 1610 1611 #ifdef CONFIG_IP_ROUTE_CLASSID 1612 if (nhc->nhc_family == AF_INET) { 1613 struct fib_nh *nh; 1614 1615 nh = container_of(nhc, struct fib_nh, nh_common); 1616 rt->dst.tclassid = nh->nh_tclassid; 1617 } 1618 #endif 1619 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 1620 if (unlikely(fnhe)) 1621 cached = rt_bind_exception(rt, fnhe, daddr, do_cache); 1622 else if (do_cache) 1623 cached = rt_cache_route(nhc, rt); 1624 if (unlikely(!cached)) { 1625 /* Routes we intend to cache in nexthop exception or 1626 * FIB nexthop have the DST_NOCACHE bit clear. 1627 * However, if we are unsuccessful at storing this 1628 * route into the cache we really need to set it. 1629 */ 1630 if (!rt->rt_gw4) { 1631 rt->rt_gw_family = AF_INET; 1632 rt->rt_gw4 = daddr; 1633 } 1634 rt_add_uncached_list(rt); 1635 } 1636 } else 1637 rt_add_uncached_list(rt); 1638 1639 #ifdef CONFIG_IP_ROUTE_CLASSID 1640 #ifdef CONFIG_IP_MULTIPLE_TABLES 1641 set_class_tag(rt, res->tclassid); 1642 #endif 1643 set_class_tag(rt, itag); 1644 #endif 1645 } 1646 1647 struct rtable *rt_dst_alloc(struct net_device *dev, 1648 unsigned int flags, u16 type, 1649 bool noxfrm) 1650 { 1651 struct rtable *rt; 1652 1653 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1654 (noxfrm ? DST_NOXFRM : 0)); 1655 1656 if (rt) { 1657 rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1658 rt->rt_flags = flags; 1659 rt->rt_type = type; 1660 rt->rt_is_input = 0; 1661 rt->rt_iif = 0; 1662 rt->rt_pmtu = 0; 1663 rt->rt_mtu_locked = 0; 1664 rt->rt_uses_gateway = 0; 1665 rt->rt_gw_family = 0; 1666 rt->rt_gw4 = 0; 1667 1668 rt->dst.output = ip_output; 1669 if (flags & RTCF_LOCAL) 1670 rt->dst.input = ip_local_deliver; 1671 } 1672 1673 return rt; 1674 } 1675 EXPORT_SYMBOL(rt_dst_alloc); 1676 1677 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt) 1678 { 1679 struct rtable *new_rt; 1680 1681 new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1682 rt->dst.flags); 1683 1684 if (new_rt) { 1685 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1686 new_rt->rt_flags = rt->rt_flags; 1687 new_rt->rt_type = rt->rt_type; 1688 new_rt->rt_is_input = rt->rt_is_input; 1689 new_rt->rt_iif = rt->rt_iif; 1690 new_rt->rt_pmtu = rt->rt_pmtu; 1691 new_rt->rt_mtu_locked = rt->rt_mtu_locked; 1692 new_rt->rt_gw_family = rt->rt_gw_family; 1693 if (rt->rt_gw_family == AF_INET) 1694 new_rt->rt_gw4 = rt->rt_gw4; 1695 else if (rt->rt_gw_family == AF_INET6) 1696 new_rt->rt_gw6 = rt->rt_gw6; 1697 1698 new_rt->dst.input = rt->dst.input; 1699 new_rt->dst.output = rt->dst.output; 1700 new_rt->dst.error = rt->dst.error; 1701 new_rt->dst.lastuse = jiffies; 1702 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate); 1703 } 1704 return new_rt; 1705 } 1706 EXPORT_SYMBOL(rt_dst_clone); 1707 1708 /* called in rcu_read_lock() section */ 1709 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1710 u8 tos, struct net_device *dev, 1711 struct in_device *in_dev, u32 *itag) 1712 { 1713 int err; 1714 1715 /* Primary sanity checks. */ 1716 if (!in_dev) 1717 return -EINVAL; 1718 1719 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) || 1720 skb->protocol != htons(ETH_P_IP)) 1721 return -EINVAL; 1722 1723 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev)) 1724 return -EINVAL; 1725 1726 if (ipv4_is_zeronet(saddr)) { 1727 if (!ipv4_is_local_multicast(daddr) && 1728 ip_hdr(skb)->protocol != IPPROTO_IGMP) 1729 return -EINVAL; 1730 } else { 1731 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 1732 in_dev, itag); 1733 if (err < 0) 1734 return err; 1735 } 1736 return 0; 1737 } 1738 1739 /* called in rcu_read_lock() section */ 1740 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1741 u8 tos, struct net_device *dev, int our) 1742 { 1743 struct in_device *in_dev = __in_dev_get_rcu(dev); 1744 unsigned int flags = RTCF_MULTICAST; 1745 struct rtable *rth; 1746 u32 itag = 0; 1747 int err; 1748 1749 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag); 1750 if (err) 1751 return err; 1752 1753 if (our) 1754 flags |= RTCF_LOCAL; 1755 1756 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 1757 IPCB(skb)->flags |= IPSKB_NOPOLICY; 1758 1759 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST, 1760 false); 1761 if (!rth) 1762 return -ENOBUFS; 1763 1764 #ifdef CONFIG_IP_ROUTE_CLASSID 1765 rth->dst.tclassid = itag; 1766 #endif 1767 rth->dst.output = ip_rt_bug; 1768 rth->rt_is_input= 1; 1769 1770 #ifdef CONFIG_IP_MROUTE 1771 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1772 rth->dst.input = ip_mr_input; 1773 #endif 1774 RT_CACHE_STAT_INC(in_slow_mc); 1775 1776 skb_dst_drop(skb); 1777 skb_dst_set(skb, &rth->dst); 1778 return 0; 1779 } 1780 1781 1782 static void ip_handle_martian_source(struct net_device *dev, 1783 struct in_device *in_dev, 1784 struct sk_buff *skb, 1785 __be32 daddr, 1786 __be32 saddr) 1787 { 1788 RT_CACHE_STAT_INC(in_martian_src); 1789 #ifdef CONFIG_IP_ROUTE_VERBOSE 1790 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1791 /* 1792 * RFC1812 recommendation, if source is martian, 1793 * the only hint is MAC header. 1794 */ 1795 pr_warn("martian source %pI4 from %pI4, on dev %s\n", 1796 &daddr, &saddr, dev->name); 1797 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1798 print_hex_dump(KERN_WARNING, "ll header: ", 1799 DUMP_PREFIX_OFFSET, 16, 1, 1800 skb_mac_header(skb), 1801 dev->hard_header_len, false); 1802 } 1803 } 1804 #endif 1805 } 1806 1807 /* called in rcu_read_lock() section */ 1808 static int __mkroute_input(struct sk_buff *skb, 1809 const struct fib_result *res, 1810 struct in_device *in_dev, 1811 __be32 daddr, __be32 saddr, u32 tos) 1812 { 1813 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1814 struct net_device *dev = nhc->nhc_dev; 1815 struct fib_nh_exception *fnhe; 1816 struct rtable *rth; 1817 int err; 1818 struct in_device *out_dev; 1819 bool do_cache; 1820 u32 itag = 0; 1821 1822 /* get a working reference to the output device */ 1823 out_dev = __in_dev_get_rcu(dev); 1824 if (!out_dev) { 1825 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n"); 1826 return -EINVAL; 1827 } 1828 1829 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res), 1830 in_dev->dev, in_dev, &itag); 1831 if (err < 0) { 1832 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1833 saddr); 1834 1835 goto cleanup; 1836 } 1837 1838 do_cache = res->fi && !itag; 1839 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) && 1840 skb->protocol == htons(ETH_P_IP)) { 1841 __be32 gw; 1842 1843 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0; 1844 if (IN_DEV_SHARED_MEDIA(out_dev) || 1845 inet_addr_onlink(out_dev, saddr, gw)) 1846 IPCB(skb)->flags |= IPSKB_DOREDIRECT; 1847 } 1848 1849 if (skb->protocol != htons(ETH_P_IP)) { 1850 /* Not IP (i.e. ARP). Do not create route, if it is 1851 * invalid for proxy arp. DNAT routes are always valid. 1852 * 1853 * Proxy arp feature have been extended to allow, ARP 1854 * replies back to the same interface, to support 1855 * Private VLAN switch technologies. See arp.c. 1856 */ 1857 if (out_dev == in_dev && 1858 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 1859 err = -EINVAL; 1860 goto cleanup; 1861 } 1862 } 1863 1864 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 1865 IPCB(skb)->flags |= IPSKB_NOPOLICY; 1866 1867 fnhe = find_exception(nhc, daddr); 1868 if (do_cache) { 1869 if (fnhe) 1870 rth = rcu_dereference(fnhe->fnhe_rth_input); 1871 else 1872 rth = rcu_dereference(nhc->nhc_rth_input); 1873 if (rt_cache_valid(rth)) { 1874 skb_dst_set_noref(skb, &rth->dst); 1875 goto out; 1876 } 1877 } 1878 1879 rth = rt_dst_alloc(out_dev->dev, 0, res->type, 1880 IN_DEV_ORCONF(out_dev, NOXFRM)); 1881 if (!rth) { 1882 err = -ENOBUFS; 1883 goto cleanup; 1884 } 1885 1886 rth->rt_is_input = 1; 1887 RT_CACHE_STAT_INC(in_slow_tot); 1888 1889 rth->dst.input = ip_forward; 1890 1891 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag, 1892 do_cache); 1893 lwtunnel_set_redirect(&rth->dst); 1894 skb_dst_set(skb, &rth->dst); 1895 out: 1896 err = 0; 1897 cleanup: 1898 return err; 1899 } 1900 1901 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1902 /* To make ICMP packets follow the right flow, the multipath hash is 1903 * calculated from the inner IP addresses. 1904 */ 1905 static void ip_multipath_l3_keys(const struct sk_buff *skb, 1906 struct flow_keys *hash_keys) 1907 { 1908 const struct iphdr *outer_iph = ip_hdr(skb); 1909 const struct iphdr *key_iph = outer_iph; 1910 const struct iphdr *inner_iph; 1911 const struct icmphdr *icmph; 1912 struct iphdr _inner_iph; 1913 struct icmphdr _icmph; 1914 1915 if (likely(outer_iph->protocol != IPPROTO_ICMP)) 1916 goto out; 1917 1918 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0)) 1919 goto out; 1920 1921 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph), 1922 &_icmph); 1923 if (!icmph) 1924 goto out; 1925 1926 if (!icmp_is_err(icmph->type)) 1927 goto out; 1928 1929 inner_iph = skb_header_pointer(skb, 1930 outer_iph->ihl * 4 + sizeof(_icmph), 1931 sizeof(_inner_iph), &_inner_iph); 1932 if (!inner_iph) 1933 goto out; 1934 1935 key_iph = inner_iph; 1936 out: 1937 hash_keys->addrs.v4addrs.src = key_iph->saddr; 1938 hash_keys->addrs.v4addrs.dst = key_iph->daddr; 1939 } 1940 1941 static u32 fib_multipath_custom_hash_outer(const struct net *net, 1942 const struct sk_buff *skb, 1943 bool *p_has_inner) 1944 { 1945 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 1946 struct flow_keys keys, hash_keys; 1947 1948 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 1949 return 0; 1950 1951 memset(&hash_keys, 0, sizeof(hash_keys)); 1952 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP); 1953 1954 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1955 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 1956 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 1957 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 1958 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 1959 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 1960 hash_keys.basic.ip_proto = keys.basic.ip_proto; 1961 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) 1962 hash_keys.ports.src = keys.ports.src; 1963 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 1964 hash_keys.ports.dst = keys.ports.dst; 1965 1966 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION); 1967 return flow_hash_from_keys(&hash_keys); 1968 } 1969 1970 static u32 fib_multipath_custom_hash_inner(const struct net *net, 1971 const struct sk_buff *skb, 1972 bool has_inner) 1973 { 1974 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 1975 struct flow_keys keys, hash_keys; 1976 1977 /* We assume the packet carries an encapsulation, but if none was 1978 * encountered during dissection of the outer flow, then there is no 1979 * point in calling the flow dissector again. 1980 */ 1981 if (!has_inner) 1982 return 0; 1983 1984 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK)) 1985 return 0; 1986 1987 memset(&hash_keys, 0, sizeof(hash_keys)); 1988 skb_flow_dissect_flow_keys(skb, &keys, 0); 1989 1990 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION)) 1991 return 0; 1992 1993 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 1994 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1995 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 1996 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 1997 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 1998 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 1999 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 2000 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 2001 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 2002 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 2003 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 2004 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 2005 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL) 2006 hash_keys.tags.flow_label = keys.tags.flow_label; 2007 } 2008 2009 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO) 2010 hash_keys.basic.ip_proto = keys.basic.ip_proto; 2011 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT) 2012 hash_keys.ports.src = keys.ports.src; 2013 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT) 2014 hash_keys.ports.dst = keys.ports.dst; 2015 2016 return flow_hash_from_keys(&hash_keys); 2017 } 2018 2019 static u32 fib_multipath_custom_hash_skb(const struct net *net, 2020 const struct sk_buff *skb) 2021 { 2022 u32 mhash, mhash_inner; 2023 bool has_inner = true; 2024 2025 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner); 2026 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner); 2027 2028 return jhash_2words(mhash, mhash_inner, 0); 2029 } 2030 2031 static u32 fib_multipath_custom_hash_fl4(const struct net *net, 2032 const struct flowi4 *fl4) 2033 { 2034 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 2035 struct flow_keys hash_keys; 2036 2037 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 2038 return 0; 2039 2040 memset(&hash_keys, 0, sizeof(hash_keys)); 2041 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2042 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 2043 hash_keys.addrs.v4addrs.src = fl4->saddr; 2044 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 2045 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2046 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 2047 hash_keys.basic.ip_proto = fl4->flowi4_proto; 2048 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) 2049 hash_keys.ports.src = fl4->fl4_sport; 2050 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 2051 hash_keys.ports.dst = fl4->fl4_dport; 2052 2053 return flow_hash_from_keys(&hash_keys); 2054 } 2055 2056 /* if skb is set it will be used and fl4 can be NULL */ 2057 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4, 2058 const struct sk_buff *skb, struct flow_keys *flkeys) 2059 { 2060 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0; 2061 struct flow_keys hash_keys; 2062 u32 mhash = 0; 2063 2064 switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) { 2065 case 0: 2066 memset(&hash_keys, 0, sizeof(hash_keys)); 2067 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2068 if (skb) { 2069 ip_multipath_l3_keys(skb, &hash_keys); 2070 } else { 2071 hash_keys.addrs.v4addrs.src = fl4->saddr; 2072 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2073 } 2074 mhash = flow_hash_from_keys(&hash_keys); 2075 break; 2076 case 1: 2077 /* skb is currently provided only when forwarding */ 2078 if (skb) { 2079 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 2080 struct flow_keys keys; 2081 2082 /* short-circuit if we already have L4 hash present */ 2083 if (skb->l4_hash) 2084 return skb_get_hash_raw(skb) >> 1; 2085 2086 memset(&hash_keys, 0, sizeof(hash_keys)); 2087 2088 if (!flkeys) { 2089 skb_flow_dissect_flow_keys(skb, &keys, flag); 2090 flkeys = &keys; 2091 } 2092 2093 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2094 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src; 2095 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst; 2096 hash_keys.ports.src = flkeys->ports.src; 2097 hash_keys.ports.dst = flkeys->ports.dst; 2098 hash_keys.basic.ip_proto = flkeys->basic.ip_proto; 2099 } else { 2100 memset(&hash_keys, 0, sizeof(hash_keys)); 2101 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2102 hash_keys.addrs.v4addrs.src = fl4->saddr; 2103 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2104 hash_keys.ports.src = fl4->fl4_sport; 2105 hash_keys.ports.dst = fl4->fl4_dport; 2106 hash_keys.basic.ip_proto = fl4->flowi4_proto; 2107 } 2108 mhash = flow_hash_from_keys(&hash_keys); 2109 break; 2110 case 2: 2111 memset(&hash_keys, 0, sizeof(hash_keys)); 2112 /* skb is currently provided only when forwarding */ 2113 if (skb) { 2114 struct flow_keys keys; 2115 2116 skb_flow_dissect_flow_keys(skb, &keys, 0); 2117 /* Inner can be v4 or v6 */ 2118 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 2119 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2120 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 2121 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 2122 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 2123 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 2124 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 2125 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 2126 hash_keys.tags.flow_label = keys.tags.flow_label; 2127 hash_keys.basic.ip_proto = keys.basic.ip_proto; 2128 } else { 2129 /* Same as case 0 */ 2130 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2131 ip_multipath_l3_keys(skb, &hash_keys); 2132 } 2133 } else { 2134 /* Same as case 0 */ 2135 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2136 hash_keys.addrs.v4addrs.src = fl4->saddr; 2137 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2138 } 2139 mhash = flow_hash_from_keys(&hash_keys); 2140 break; 2141 case 3: 2142 if (skb) 2143 mhash = fib_multipath_custom_hash_skb(net, skb); 2144 else 2145 mhash = fib_multipath_custom_hash_fl4(net, fl4); 2146 break; 2147 } 2148 2149 if (multipath_hash) 2150 mhash = jhash_2words(mhash, multipath_hash, 0); 2151 2152 return mhash >> 1; 2153 } 2154 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 2155 2156 static int ip_mkroute_input(struct sk_buff *skb, 2157 struct fib_result *res, 2158 struct in_device *in_dev, 2159 __be32 daddr, __be32 saddr, u32 tos, 2160 struct flow_keys *hkeys) 2161 { 2162 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2163 if (res->fi && fib_info_num_path(res->fi) > 1) { 2164 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys); 2165 2166 fib_select_multipath(res, h); 2167 IPCB(skb)->flags |= IPSKB_MULTIPATH; 2168 } 2169 #endif 2170 2171 /* create a routing cache entry */ 2172 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos); 2173 } 2174 2175 /* Implements all the saddr-related checks as ip_route_input_slow(), 2176 * assuming daddr is valid and the destination is not a local broadcast one. 2177 * Uses the provided hint instead of performing a route lookup. 2178 */ 2179 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2180 u8 tos, struct net_device *dev, 2181 const struct sk_buff *hint) 2182 { 2183 struct in_device *in_dev = __in_dev_get_rcu(dev); 2184 struct rtable *rt = skb_rtable(hint); 2185 struct net *net = dev_net(dev); 2186 int err = -EINVAL; 2187 u32 tag = 0; 2188 2189 if (!in_dev) 2190 return -EINVAL; 2191 2192 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 2193 goto martian_source; 2194 2195 if (ipv4_is_zeronet(saddr)) 2196 goto martian_source; 2197 2198 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2199 goto martian_source; 2200 2201 if (rt->rt_type != RTN_LOCAL) 2202 goto skip_validate_source; 2203 2204 tos &= IPTOS_RT_MASK; 2205 err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag); 2206 if (err < 0) 2207 goto martian_source; 2208 2209 skip_validate_source: 2210 skb_dst_copy(skb, hint); 2211 return 0; 2212 2213 martian_source: 2214 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2215 return err; 2216 } 2217 2218 /* get device for dst_alloc with local routes */ 2219 static struct net_device *ip_rt_get_dev(struct net *net, 2220 const struct fib_result *res) 2221 { 2222 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL; 2223 struct net_device *dev = NULL; 2224 2225 if (nhc) 2226 dev = l3mdev_master_dev_rcu(nhc->nhc_dev); 2227 2228 return dev ? : net->loopback_dev; 2229 } 2230 2231 /* 2232 * NOTE. We drop all the packets that has local source 2233 * addresses, because every properly looped back packet 2234 * must have correct destination already attached by output routine. 2235 * Changes in the enforced policies must be applied also to 2236 * ip_route_use_hint(). 2237 * 2238 * Such approach solves two big problems: 2239 * 1. Not simplex devices are handled properly. 2240 * 2. IP spoofing attempts are filtered with 100% of guarantee. 2241 * called with rcu_read_lock() 2242 */ 2243 2244 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2245 u8 tos, struct net_device *dev, 2246 struct fib_result *res) 2247 { 2248 struct in_device *in_dev = __in_dev_get_rcu(dev); 2249 struct flow_keys *flkeys = NULL, _flkeys; 2250 struct net *net = dev_net(dev); 2251 struct ip_tunnel_info *tun_info; 2252 int err = -EINVAL; 2253 unsigned int flags = 0; 2254 u32 itag = 0; 2255 struct rtable *rth; 2256 struct flowi4 fl4; 2257 bool do_cache = true; 2258 2259 /* IP on this device is disabled. */ 2260 2261 if (!in_dev) 2262 goto out; 2263 2264 /* Check for the most weird martians, which can be not detected 2265 * by fib_lookup. 2266 */ 2267 2268 tun_info = skb_tunnel_info(skb); 2269 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX)) 2270 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id; 2271 else 2272 fl4.flowi4_tun_key.tun_id = 0; 2273 skb_dst_drop(skb); 2274 2275 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 2276 goto martian_source; 2277 2278 res->fi = NULL; 2279 res->table = NULL; 2280 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 2281 goto brd_input; 2282 2283 /* Accept zero addresses only to limited broadcast; 2284 * I even do not know to fix it or not. Waiting for complains :-) 2285 */ 2286 if (ipv4_is_zeronet(saddr)) 2287 goto martian_source; 2288 2289 if (ipv4_is_zeronet(daddr)) 2290 goto martian_destination; 2291 2292 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(), 2293 * and call it once if daddr or/and saddr are loopback addresses 2294 */ 2295 if (ipv4_is_loopback(daddr)) { 2296 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2297 goto martian_destination; 2298 } else if (ipv4_is_loopback(saddr)) { 2299 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2300 goto martian_source; 2301 } 2302 2303 /* 2304 * Now we are ready to route packet. 2305 */ 2306 fl4.flowi4_l3mdev = 0; 2307 fl4.flowi4_oif = 0; 2308 fl4.flowi4_iif = dev->ifindex; 2309 fl4.flowi4_mark = skb->mark; 2310 fl4.flowi4_tos = tos; 2311 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 2312 fl4.flowi4_flags = 0; 2313 fl4.daddr = daddr; 2314 fl4.saddr = saddr; 2315 fl4.flowi4_uid = sock_net_uid(net, NULL); 2316 fl4.flowi4_multipath_hash = 0; 2317 2318 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) { 2319 flkeys = &_flkeys; 2320 } else { 2321 fl4.flowi4_proto = 0; 2322 fl4.fl4_sport = 0; 2323 fl4.fl4_dport = 0; 2324 } 2325 2326 err = fib_lookup(net, &fl4, res, 0); 2327 if (err != 0) { 2328 if (!IN_DEV_FORWARD(in_dev)) 2329 err = -EHOSTUNREACH; 2330 goto no_route; 2331 } 2332 2333 if (res->type == RTN_BROADCAST) { 2334 if (IN_DEV_BFORWARD(in_dev)) 2335 goto make_route; 2336 /* not do cache if bc_forwarding is enabled */ 2337 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING)) 2338 do_cache = false; 2339 goto brd_input; 2340 } 2341 2342 if (res->type == RTN_LOCAL) { 2343 err = fib_validate_source(skb, saddr, daddr, tos, 2344 0, dev, in_dev, &itag); 2345 if (err < 0) 2346 goto martian_source; 2347 goto local_input; 2348 } 2349 2350 if (!IN_DEV_FORWARD(in_dev)) { 2351 err = -EHOSTUNREACH; 2352 goto no_route; 2353 } 2354 if (res->type != RTN_UNICAST) 2355 goto martian_destination; 2356 2357 make_route: 2358 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys); 2359 out: return err; 2360 2361 brd_input: 2362 if (skb->protocol != htons(ETH_P_IP)) 2363 goto e_inval; 2364 2365 if (!ipv4_is_zeronet(saddr)) { 2366 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 2367 in_dev, &itag); 2368 if (err < 0) 2369 goto martian_source; 2370 } 2371 flags |= RTCF_BROADCAST; 2372 res->type = RTN_BROADCAST; 2373 RT_CACHE_STAT_INC(in_brd); 2374 2375 local_input: 2376 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 2377 IPCB(skb)->flags |= IPSKB_NOPOLICY; 2378 2379 do_cache &= res->fi && !itag; 2380 if (do_cache) { 2381 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2382 2383 rth = rcu_dereference(nhc->nhc_rth_input); 2384 if (rt_cache_valid(rth)) { 2385 skb_dst_set_noref(skb, &rth->dst); 2386 err = 0; 2387 goto out; 2388 } 2389 } 2390 2391 rth = rt_dst_alloc(ip_rt_get_dev(net, res), 2392 flags | RTCF_LOCAL, res->type, false); 2393 if (!rth) 2394 goto e_nobufs; 2395 2396 rth->dst.output= ip_rt_bug; 2397 #ifdef CONFIG_IP_ROUTE_CLASSID 2398 rth->dst.tclassid = itag; 2399 #endif 2400 rth->rt_is_input = 1; 2401 2402 RT_CACHE_STAT_INC(in_slow_tot); 2403 if (res->type == RTN_UNREACHABLE) { 2404 rth->dst.input= ip_error; 2405 rth->dst.error= -err; 2406 rth->rt_flags &= ~RTCF_LOCAL; 2407 } 2408 2409 if (do_cache) { 2410 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2411 2412 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 2413 if (lwtunnel_input_redirect(rth->dst.lwtstate)) { 2414 WARN_ON(rth->dst.input == lwtunnel_input); 2415 rth->dst.lwtstate->orig_input = rth->dst.input; 2416 rth->dst.input = lwtunnel_input; 2417 } 2418 2419 if (unlikely(!rt_cache_route(nhc, rth))) 2420 rt_add_uncached_list(rth); 2421 } 2422 skb_dst_set(skb, &rth->dst); 2423 err = 0; 2424 goto out; 2425 2426 no_route: 2427 RT_CACHE_STAT_INC(in_no_route); 2428 res->type = RTN_UNREACHABLE; 2429 res->fi = NULL; 2430 res->table = NULL; 2431 goto local_input; 2432 2433 /* 2434 * Do not cache martian addresses: they should be logged (RFC1812) 2435 */ 2436 martian_destination: 2437 RT_CACHE_STAT_INC(in_martian_dst); 2438 #ifdef CONFIG_IP_ROUTE_VERBOSE 2439 if (IN_DEV_LOG_MARTIANS(in_dev)) 2440 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n", 2441 &daddr, &saddr, dev->name); 2442 #endif 2443 2444 e_inval: 2445 err = -EINVAL; 2446 goto out; 2447 2448 e_nobufs: 2449 err = -ENOBUFS; 2450 goto out; 2451 2452 martian_source: 2453 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2454 goto out; 2455 } 2456 2457 /* called with rcu_read_lock held */ 2458 static int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2459 u8 tos, struct net_device *dev, struct fib_result *res) 2460 { 2461 /* Multicast recognition logic is moved from route cache to here. 2462 * The problem was that too many Ethernet cards have broken/missing 2463 * hardware multicast filters :-( As result the host on multicasting 2464 * network acquires a lot of useless route cache entries, sort of 2465 * SDR messages from all the world. Now we try to get rid of them. 2466 * Really, provided software IP multicast filter is organized 2467 * reasonably (at least, hashed), it does not result in a slowdown 2468 * comparing with route cache reject entries. 2469 * Note, that multicast routers are not affected, because 2470 * route cache entry is created eventually. 2471 */ 2472 if (ipv4_is_multicast(daddr)) { 2473 struct in_device *in_dev = __in_dev_get_rcu(dev); 2474 int our = 0; 2475 int err = -EINVAL; 2476 2477 if (!in_dev) 2478 return err; 2479 our = ip_check_mc_rcu(in_dev, daddr, saddr, 2480 ip_hdr(skb)->protocol); 2481 2482 /* check l3 master if no match yet */ 2483 if (!our && netif_is_l3_slave(dev)) { 2484 struct in_device *l3_in_dev; 2485 2486 l3_in_dev = __in_dev_get_rcu(skb->dev); 2487 if (l3_in_dev) 2488 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr, 2489 ip_hdr(skb)->protocol); 2490 } 2491 2492 if (our 2493 #ifdef CONFIG_IP_MROUTE 2494 || 2495 (!ipv4_is_local_multicast(daddr) && 2496 IN_DEV_MFORWARD(in_dev)) 2497 #endif 2498 ) { 2499 err = ip_route_input_mc(skb, daddr, saddr, 2500 tos, dev, our); 2501 } 2502 return err; 2503 } 2504 2505 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res); 2506 } 2507 2508 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2509 u8 tos, struct net_device *dev) 2510 { 2511 struct fib_result res; 2512 int err; 2513 2514 tos &= IPTOS_RT_MASK; 2515 rcu_read_lock(); 2516 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res); 2517 rcu_read_unlock(); 2518 2519 return err; 2520 } 2521 EXPORT_SYMBOL(ip_route_input_noref); 2522 2523 /* called with rcu_read_lock() */ 2524 static struct rtable *__mkroute_output(const struct fib_result *res, 2525 const struct flowi4 *fl4, int orig_oif, 2526 struct net_device *dev_out, 2527 unsigned int flags) 2528 { 2529 struct fib_info *fi = res->fi; 2530 struct fib_nh_exception *fnhe; 2531 struct in_device *in_dev; 2532 u16 type = res->type; 2533 struct rtable *rth; 2534 bool do_cache; 2535 2536 in_dev = __in_dev_get_rcu(dev_out); 2537 if (!in_dev) 2538 return ERR_PTR(-EINVAL); 2539 2540 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 2541 if (ipv4_is_loopback(fl4->saddr) && 2542 !(dev_out->flags & IFF_LOOPBACK) && 2543 !netif_is_l3_master(dev_out)) 2544 return ERR_PTR(-EINVAL); 2545 2546 if (ipv4_is_lbcast(fl4->daddr)) 2547 type = RTN_BROADCAST; 2548 else if (ipv4_is_multicast(fl4->daddr)) 2549 type = RTN_MULTICAST; 2550 else if (ipv4_is_zeronet(fl4->daddr)) 2551 return ERR_PTR(-EINVAL); 2552 2553 if (dev_out->flags & IFF_LOOPBACK) 2554 flags |= RTCF_LOCAL; 2555 2556 do_cache = true; 2557 if (type == RTN_BROADCAST) { 2558 flags |= RTCF_BROADCAST | RTCF_LOCAL; 2559 fi = NULL; 2560 } else if (type == RTN_MULTICAST) { 2561 flags |= RTCF_MULTICAST | RTCF_LOCAL; 2562 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr, 2563 fl4->flowi4_proto)) 2564 flags &= ~RTCF_LOCAL; 2565 else 2566 do_cache = false; 2567 /* If multicast route do not exist use 2568 * default one, but do not gateway in this case. 2569 * Yes, it is hack. 2570 */ 2571 if (fi && res->prefixlen < 4) 2572 fi = NULL; 2573 } else if ((type == RTN_LOCAL) && (orig_oif != 0) && 2574 (orig_oif != dev_out->ifindex)) { 2575 /* For local routes that require a particular output interface 2576 * we do not want to cache the result. Caching the result 2577 * causes incorrect behaviour when there are multiple source 2578 * addresses on the interface, the end result being that if the 2579 * intended recipient is waiting on that interface for the 2580 * packet he won't receive it because it will be delivered on 2581 * the loopback interface and the IP_PKTINFO ipi_ifindex will 2582 * be set to the loopback interface as well. 2583 */ 2584 do_cache = false; 2585 } 2586 2587 fnhe = NULL; 2588 do_cache &= fi != NULL; 2589 if (fi) { 2590 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2591 struct rtable __rcu **prth; 2592 2593 fnhe = find_exception(nhc, fl4->daddr); 2594 if (!do_cache) 2595 goto add; 2596 if (fnhe) { 2597 prth = &fnhe->fnhe_rth_output; 2598 } else { 2599 if (unlikely(fl4->flowi4_flags & 2600 FLOWI_FLAG_KNOWN_NH && 2601 !(nhc->nhc_gw_family && 2602 nhc->nhc_scope == RT_SCOPE_LINK))) { 2603 do_cache = false; 2604 goto add; 2605 } 2606 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 2607 } 2608 rth = rcu_dereference(*prth); 2609 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst)) 2610 return rth; 2611 } 2612 2613 add: 2614 rth = rt_dst_alloc(dev_out, flags, type, 2615 IN_DEV_ORCONF(in_dev, NOXFRM)); 2616 if (!rth) 2617 return ERR_PTR(-ENOBUFS); 2618 2619 rth->rt_iif = orig_oif; 2620 2621 RT_CACHE_STAT_INC(out_slow_tot); 2622 2623 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 2624 if (flags & RTCF_LOCAL && 2625 !(dev_out->flags & IFF_LOOPBACK)) { 2626 rth->dst.output = ip_mc_output; 2627 RT_CACHE_STAT_INC(out_slow_mc); 2628 } 2629 #ifdef CONFIG_IP_MROUTE 2630 if (type == RTN_MULTICAST) { 2631 if (IN_DEV_MFORWARD(in_dev) && 2632 !ipv4_is_local_multicast(fl4->daddr)) { 2633 rth->dst.input = ip_mr_input; 2634 rth->dst.output = ip_mc_output; 2635 } 2636 } 2637 #endif 2638 } 2639 2640 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache); 2641 lwtunnel_set_redirect(&rth->dst); 2642 2643 return rth; 2644 } 2645 2646 /* 2647 * Major route resolver routine. 2648 */ 2649 2650 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4, 2651 const struct sk_buff *skb) 2652 { 2653 struct fib_result res = { 2654 .type = RTN_UNSPEC, 2655 .fi = NULL, 2656 .table = NULL, 2657 .tclassid = 0, 2658 }; 2659 struct rtable *rth; 2660 2661 fl4->flowi4_iif = LOOPBACK_IFINDEX; 2662 ip_rt_fix_tos(fl4); 2663 2664 rcu_read_lock(); 2665 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb); 2666 rcu_read_unlock(); 2667 2668 return rth; 2669 } 2670 EXPORT_SYMBOL_GPL(ip_route_output_key_hash); 2671 2672 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4, 2673 struct fib_result *res, 2674 const struct sk_buff *skb) 2675 { 2676 struct net_device *dev_out = NULL; 2677 int orig_oif = fl4->flowi4_oif; 2678 unsigned int flags = 0; 2679 struct rtable *rth; 2680 int err; 2681 2682 if (fl4->saddr) { 2683 if (ipv4_is_multicast(fl4->saddr) || 2684 ipv4_is_lbcast(fl4->saddr) || 2685 ipv4_is_zeronet(fl4->saddr)) { 2686 rth = ERR_PTR(-EINVAL); 2687 goto out; 2688 } 2689 2690 rth = ERR_PTR(-ENETUNREACH); 2691 2692 /* I removed check for oif == dev_out->oif here. 2693 * It was wrong for two reasons: 2694 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 2695 * is assigned to multiple interfaces. 2696 * 2. Moreover, we are allowed to send packets with saddr 2697 * of another iface. --ANK 2698 */ 2699 2700 if (fl4->flowi4_oif == 0 && 2701 (ipv4_is_multicast(fl4->daddr) || 2702 ipv4_is_lbcast(fl4->daddr))) { 2703 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2704 dev_out = __ip_dev_find(net, fl4->saddr, false); 2705 if (!dev_out) 2706 goto out; 2707 2708 /* Special hack: user can direct multicasts 2709 * and limited broadcast via necessary interface 2710 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 2711 * This hack is not just for fun, it allows 2712 * vic,vat and friends to work. 2713 * They bind socket to loopback, set ttl to zero 2714 * and expect that it will work. 2715 * From the viewpoint of routing cache they are broken, 2716 * because we are not allowed to build multicast path 2717 * with loopback source addr (look, routing cache 2718 * cannot know, that ttl is zero, so that packet 2719 * will not leave this host and route is valid). 2720 * Luckily, this hack is good workaround. 2721 */ 2722 2723 fl4->flowi4_oif = dev_out->ifindex; 2724 goto make_route; 2725 } 2726 2727 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 2728 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2729 if (!__ip_dev_find(net, fl4->saddr, false)) 2730 goto out; 2731 } 2732 } 2733 2734 2735 if (fl4->flowi4_oif) { 2736 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif); 2737 rth = ERR_PTR(-ENODEV); 2738 if (!dev_out) 2739 goto out; 2740 2741 /* RACE: Check return value of inet_select_addr instead. */ 2742 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 2743 rth = ERR_PTR(-ENETUNREACH); 2744 goto out; 2745 } 2746 if (ipv4_is_local_multicast(fl4->daddr) || 2747 ipv4_is_lbcast(fl4->daddr) || 2748 fl4->flowi4_proto == IPPROTO_IGMP) { 2749 if (!fl4->saddr) 2750 fl4->saddr = inet_select_addr(dev_out, 0, 2751 RT_SCOPE_LINK); 2752 goto make_route; 2753 } 2754 if (!fl4->saddr) { 2755 if (ipv4_is_multicast(fl4->daddr)) 2756 fl4->saddr = inet_select_addr(dev_out, 0, 2757 fl4->flowi4_scope); 2758 else if (!fl4->daddr) 2759 fl4->saddr = inet_select_addr(dev_out, 0, 2760 RT_SCOPE_HOST); 2761 } 2762 } 2763 2764 if (!fl4->daddr) { 2765 fl4->daddr = fl4->saddr; 2766 if (!fl4->daddr) 2767 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK); 2768 dev_out = net->loopback_dev; 2769 fl4->flowi4_oif = LOOPBACK_IFINDEX; 2770 res->type = RTN_LOCAL; 2771 flags |= RTCF_LOCAL; 2772 goto make_route; 2773 } 2774 2775 err = fib_lookup(net, fl4, res, 0); 2776 if (err) { 2777 res->fi = NULL; 2778 res->table = NULL; 2779 if (fl4->flowi4_oif && 2780 (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) { 2781 /* Apparently, routing tables are wrong. Assume, 2782 * that the destination is on link. 2783 * 2784 * WHY? DW. 2785 * Because we are allowed to send to iface 2786 * even if it has NO routes and NO assigned 2787 * addresses. When oif is specified, routing 2788 * tables are looked up with only one purpose: 2789 * to catch if destination is gatewayed, rather than 2790 * direct. Moreover, if MSG_DONTROUTE is set, 2791 * we send packet, ignoring both routing tables 2792 * and ifaddr state. --ANK 2793 * 2794 * 2795 * We could make it even if oif is unknown, 2796 * likely IPv6, but we do not. 2797 */ 2798 2799 if (fl4->saddr == 0) 2800 fl4->saddr = inet_select_addr(dev_out, 0, 2801 RT_SCOPE_LINK); 2802 res->type = RTN_UNICAST; 2803 goto make_route; 2804 } 2805 rth = ERR_PTR(err); 2806 goto out; 2807 } 2808 2809 if (res->type == RTN_LOCAL) { 2810 if (!fl4->saddr) { 2811 if (res->fi->fib_prefsrc) 2812 fl4->saddr = res->fi->fib_prefsrc; 2813 else 2814 fl4->saddr = fl4->daddr; 2815 } 2816 2817 /* L3 master device is the loopback for that domain */ 2818 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? : 2819 net->loopback_dev; 2820 2821 /* make sure orig_oif points to fib result device even 2822 * though packet rx/tx happens over loopback or l3mdev 2823 */ 2824 orig_oif = FIB_RES_OIF(*res); 2825 2826 fl4->flowi4_oif = dev_out->ifindex; 2827 flags |= RTCF_LOCAL; 2828 goto make_route; 2829 } 2830 2831 fib_select_path(net, res, fl4, skb); 2832 2833 dev_out = FIB_RES_DEV(*res); 2834 2835 make_route: 2836 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags); 2837 2838 out: 2839 return rth; 2840 } 2841 2842 static struct dst_ops ipv4_dst_blackhole_ops = { 2843 .family = AF_INET, 2844 .default_advmss = ipv4_default_advmss, 2845 .neigh_lookup = ipv4_neigh_lookup, 2846 .check = dst_blackhole_check, 2847 .cow_metrics = dst_blackhole_cow_metrics, 2848 .update_pmtu = dst_blackhole_update_pmtu, 2849 .redirect = dst_blackhole_redirect, 2850 .mtu = dst_blackhole_mtu, 2851 }; 2852 2853 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2854 { 2855 struct rtable *ort = (struct rtable *) dst_orig; 2856 struct rtable *rt; 2857 2858 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0); 2859 if (rt) { 2860 struct dst_entry *new = &rt->dst; 2861 2862 new->__use = 1; 2863 new->input = dst_discard; 2864 new->output = dst_discard_out; 2865 2866 new->dev = net->loopback_dev; 2867 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC); 2868 2869 rt->rt_is_input = ort->rt_is_input; 2870 rt->rt_iif = ort->rt_iif; 2871 rt->rt_pmtu = ort->rt_pmtu; 2872 rt->rt_mtu_locked = ort->rt_mtu_locked; 2873 2874 rt->rt_genid = rt_genid_ipv4(net); 2875 rt->rt_flags = ort->rt_flags; 2876 rt->rt_type = ort->rt_type; 2877 rt->rt_uses_gateway = ort->rt_uses_gateway; 2878 rt->rt_gw_family = ort->rt_gw_family; 2879 if (rt->rt_gw_family == AF_INET) 2880 rt->rt_gw4 = ort->rt_gw4; 2881 else if (rt->rt_gw_family == AF_INET6) 2882 rt->rt_gw6 = ort->rt_gw6; 2883 } 2884 2885 dst_release(dst_orig); 2886 2887 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2888 } 2889 2890 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2891 const struct sock *sk) 2892 { 2893 struct rtable *rt = __ip_route_output_key(net, flp4); 2894 2895 if (IS_ERR(rt)) 2896 return rt; 2897 2898 if (flp4->flowi4_proto) { 2899 flp4->flowi4_oif = rt->dst.dev->ifindex; 2900 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst, 2901 flowi4_to_flowi(flp4), 2902 sk, 0); 2903 } 2904 2905 return rt; 2906 } 2907 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2908 2909 struct rtable *ip_route_output_tunnel(struct sk_buff *skb, 2910 struct net_device *dev, 2911 struct net *net, __be32 *saddr, 2912 const struct ip_tunnel_info *info, 2913 u8 protocol, bool use_cache) 2914 { 2915 #ifdef CONFIG_DST_CACHE 2916 struct dst_cache *dst_cache; 2917 #endif 2918 struct rtable *rt = NULL; 2919 struct flowi4 fl4; 2920 __u8 tos; 2921 2922 #ifdef CONFIG_DST_CACHE 2923 dst_cache = (struct dst_cache *)&info->dst_cache; 2924 if (use_cache) { 2925 rt = dst_cache_get_ip4(dst_cache, saddr); 2926 if (rt) 2927 return rt; 2928 } 2929 #endif 2930 memset(&fl4, 0, sizeof(fl4)); 2931 fl4.flowi4_mark = skb->mark; 2932 fl4.flowi4_proto = protocol; 2933 fl4.daddr = info->key.u.ipv4.dst; 2934 fl4.saddr = info->key.u.ipv4.src; 2935 tos = info->key.tos; 2936 fl4.flowi4_tos = RT_TOS(tos); 2937 2938 rt = ip_route_output_key(net, &fl4); 2939 if (IS_ERR(rt)) { 2940 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr); 2941 return ERR_PTR(-ENETUNREACH); 2942 } 2943 if (rt->dst.dev == dev) { /* is this necessary? */ 2944 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr); 2945 ip_rt_put(rt); 2946 return ERR_PTR(-ELOOP); 2947 } 2948 #ifdef CONFIG_DST_CACHE 2949 if (use_cache) 2950 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr); 2951 #endif 2952 *saddr = fl4.saddr; 2953 return rt; 2954 } 2955 EXPORT_SYMBOL_GPL(ip_route_output_tunnel); 2956 2957 /* called with rcu_read_lock held */ 2958 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, 2959 struct rtable *rt, u32 table_id, dscp_t dscp, 2960 struct flowi4 *fl4, struct sk_buff *skb, u32 portid, 2961 u32 seq, unsigned int flags) 2962 { 2963 struct rtmsg *r; 2964 struct nlmsghdr *nlh; 2965 unsigned long expires = 0; 2966 u32 error; 2967 u32 metrics[RTAX_MAX]; 2968 2969 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags); 2970 if (!nlh) 2971 return -EMSGSIZE; 2972 2973 r = nlmsg_data(nlh); 2974 r->rtm_family = AF_INET; 2975 r->rtm_dst_len = 32; 2976 r->rtm_src_len = 0; 2977 r->rtm_tos = inet_dscp_to_dsfield(dscp); 2978 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT; 2979 if (nla_put_u32(skb, RTA_TABLE, table_id)) 2980 goto nla_put_failure; 2981 r->rtm_type = rt->rt_type; 2982 r->rtm_scope = RT_SCOPE_UNIVERSE; 2983 r->rtm_protocol = RTPROT_UNSPEC; 2984 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 2985 if (rt->rt_flags & RTCF_NOTIFY) 2986 r->rtm_flags |= RTM_F_NOTIFY; 2987 if (IPCB(skb)->flags & IPSKB_DOREDIRECT) 2988 r->rtm_flags |= RTCF_DOREDIRECT; 2989 2990 if (nla_put_in_addr(skb, RTA_DST, dst)) 2991 goto nla_put_failure; 2992 if (src) { 2993 r->rtm_src_len = 32; 2994 if (nla_put_in_addr(skb, RTA_SRC, src)) 2995 goto nla_put_failure; 2996 } 2997 if (rt->dst.dev && 2998 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex)) 2999 goto nla_put_failure; 3000 if (rt->dst.lwtstate && 3001 lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0) 3002 goto nla_put_failure; 3003 #ifdef CONFIG_IP_ROUTE_CLASSID 3004 if (rt->dst.tclassid && 3005 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid)) 3006 goto nla_put_failure; 3007 #endif 3008 if (fl4 && !rt_is_input_route(rt) && 3009 fl4->saddr != src) { 3010 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr)) 3011 goto nla_put_failure; 3012 } 3013 if (rt->rt_uses_gateway) { 3014 if (rt->rt_gw_family == AF_INET && 3015 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) { 3016 goto nla_put_failure; 3017 } else if (rt->rt_gw_family == AF_INET6) { 3018 int alen = sizeof(struct in6_addr); 3019 struct nlattr *nla; 3020 struct rtvia *via; 3021 3022 nla = nla_reserve(skb, RTA_VIA, alen + 2); 3023 if (!nla) 3024 goto nla_put_failure; 3025 3026 via = nla_data(nla); 3027 via->rtvia_family = AF_INET6; 3028 memcpy(via->rtvia_addr, &rt->rt_gw6, alen); 3029 } 3030 } 3031 3032 expires = rt->dst.expires; 3033 if (expires) { 3034 unsigned long now = jiffies; 3035 3036 if (time_before(now, expires)) 3037 expires -= now; 3038 else 3039 expires = 0; 3040 } 3041 3042 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics)); 3043 if (rt->rt_pmtu && expires) 3044 metrics[RTAX_MTU - 1] = rt->rt_pmtu; 3045 if (rt->rt_mtu_locked && expires) 3046 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU); 3047 if (rtnetlink_put_metrics(skb, metrics) < 0) 3048 goto nla_put_failure; 3049 3050 if (fl4) { 3051 if (fl4->flowi4_mark && 3052 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark)) 3053 goto nla_put_failure; 3054 3055 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) && 3056 nla_put_u32(skb, RTA_UID, 3057 from_kuid_munged(current_user_ns(), 3058 fl4->flowi4_uid))) 3059 goto nla_put_failure; 3060 3061 if (rt_is_input_route(rt)) { 3062 #ifdef CONFIG_IP_MROUTE 3063 if (ipv4_is_multicast(dst) && 3064 !ipv4_is_local_multicast(dst) && 3065 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) { 3066 int err = ipmr_get_route(net, skb, 3067 fl4->saddr, fl4->daddr, 3068 r, portid); 3069 3070 if (err <= 0) { 3071 if (err == 0) 3072 return 0; 3073 goto nla_put_failure; 3074 } 3075 } else 3076 #endif 3077 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif)) 3078 goto nla_put_failure; 3079 } 3080 } 3081 3082 error = rt->dst.error; 3083 3084 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0) 3085 goto nla_put_failure; 3086 3087 nlmsg_end(skb, nlh); 3088 return 0; 3089 3090 nla_put_failure: 3091 nlmsg_cancel(skb, nlh); 3092 return -EMSGSIZE; 3093 } 3094 3095 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb, 3096 struct netlink_callback *cb, u32 table_id, 3097 struct fnhe_hash_bucket *bucket, int genid, 3098 int *fa_index, int fa_start, unsigned int flags) 3099 { 3100 int i; 3101 3102 for (i = 0; i < FNHE_HASH_SIZE; i++) { 3103 struct fib_nh_exception *fnhe; 3104 3105 for (fnhe = rcu_dereference(bucket[i].chain); fnhe; 3106 fnhe = rcu_dereference(fnhe->fnhe_next)) { 3107 struct rtable *rt; 3108 int err; 3109 3110 if (*fa_index < fa_start) 3111 goto next; 3112 3113 if (fnhe->fnhe_genid != genid) 3114 goto next; 3115 3116 if (fnhe->fnhe_expires && 3117 time_after(jiffies, fnhe->fnhe_expires)) 3118 goto next; 3119 3120 rt = rcu_dereference(fnhe->fnhe_rth_input); 3121 if (!rt) 3122 rt = rcu_dereference(fnhe->fnhe_rth_output); 3123 if (!rt) 3124 goto next; 3125 3126 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt, 3127 table_id, 0, NULL, skb, 3128 NETLINK_CB(cb->skb).portid, 3129 cb->nlh->nlmsg_seq, flags); 3130 if (err) 3131 return err; 3132 next: 3133 (*fa_index)++; 3134 } 3135 } 3136 3137 return 0; 3138 } 3139 3140 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb, 3141 u32 table_id, struct fib_info *fi, 3142 int *fa_index, int fa_start, unsigned int flags) 3143 { 3144 struct net *net = sock_net(cb->skb->sk); 3145 int nhsel, genid = fnhe_genid(net); 3146 3147 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) { 3148 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel); 3149 struct fnhe_hash_bucket *bucket; 3150 int err; 3151 3152 if (nhc->nhc_flags & RTNH_F_DEAD) 3153 continue; 3154 3155 rcu_read_lock(); 3156 bucket = rcu_dereference(nhc->nhc_exceptions); 3157 err = 0; 3158 if (bucket) 3159 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket, 3160 genid, fa_index, fa_start, 3161 flags); 3162 rcu_read_unlock(); 3163 if (err) 3164 return err; 3165 } 3166 3167 return 0; 3168 } 3169 3170 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst, 3171 u8 ip_proto, __be16 sport, 3172 __be16 dport) 3173 { 3174 struct sk_buff *skb; 3175 struct iphdr *iph; 3176 3177 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 3178 if (!skb) 3179 return NULL; 3180 3181 /* Reserve room for dummy headers, this skb can pass 3182 * through good chunk of routing engine. 3183 */ 3184 skb_reset_mac_header(skb); 3185 skb_reset_network_header(skb); 3186 skb->protocol = htons(ETH_P_IP); 3187 iph = skb_put(skb, sizeof(struct iphdr)); 3188 iph->protocol = ip_proto; 3189 iph->saddr = src; 3190 iph->daddr = dst; 3191 iph->version = 0x4; 3192 iph->frag_off = 0; 3193 iph->ihl = 0x5; 3194 skb_set_transport_header(skb, skb->len); 3195 3196 switch (iph->protocol) { 3197 case IPPROTO_UDP: { 3198 struct udphdr *udph; 3199 3200 udph = skb_put_zero(skb, sizeof(struct udphdr)); 3201 udph->source = sport; 3202 udph->dest = dport; 3203 udph->len = htons(sizeof(struct udphdr)); 3204 udph->check = 0; 3205 break; 3206 } 3207 case IPPROTO_TCP: { 3208 struct tcphdr *tcph; 3209 3210 tcph = skb_put_zero(skb, sizeof(struct tcphdr)); 3211 tcph->source = sport; 3212 tcph->dest = dport; 3213 tcph->doff = sizeof(struct tcphdr) / 4; 3214 tcph->rst = 1; 3215 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr), 3216 src, dst, 0); 3217 break; 3218 } 3219 case IPPROTO_ICMP: { 3220 struct icmphdr *icmph; 3221 3222 icmph = skb_put_zero(skb, sizeof(struct icmphdr)); 3223 icmph->type = ICMP_ECHO; 3224 icmph->code = 0; 3225 } 3226 } 3227 3228 return skb; 3229 } 3230 3231 static int inet_rtm_valid_getroute_req(struct sk_buff *skb, 3232 const struct nlmsghdr *nlh, 3233 struct nlattr **tb, 3234 struct netlink_ext_ack *extack) 3235 { 3236 struct rtmsg *rtm; 3237 int i, err; 3238 3239 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) { 3240 NL_SET_ERR_MSG(extack, 3241 "ipv4: Invalid header for route get request"); 3242 return -EINVAL; 3243 } 3244 3245 if (!netlink_strict_get_check(skb)) 3246 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX, 3247 rtm_ipv4_policy, extack); 3248 3249 rtm = nlmsg_data(nlh); 3250 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) || 3251 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) || 3252 rtm->rtm_table || rtm->rtm_protocol || 3253 rtm->rtm_scope || rtm->rtm_type) { 3254 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request"); 3255 return -EINVAL; 3256 } 3257 3258 if (rtm->rtm_flags & ~(RTM_F_NOTIFY | 3259 RTM_F_LOOKUP_TABLE | 3260 RTM_F_FIB_MATCH)) { 3261 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request"); 3262 return -EINVAL; 3263 } 3264 3265 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX, 3266 rtm_ipv4_policy, extack); 3267 if (err) 3268 return err; 3269 3270 if ((tb[RTA_SRC] && !rtm->rtm_src_len) || 3271 (tb[RTA_DST] && !rtm->rtm_dst_len)) { 3272 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4"); 3273 return -EINVAL; 3274 } 3275 3276 for (i = 0; i <= RTA_MAX; i++) { 3277 if (!tb[i]) 3278 continue; 3279 3280 switch (i) { 3281 case RTA_IIF: 3282 case RTA_OIF: 3283 case RTA_SRC: 3284 case RTA_DST: 3285 case RTA_IP_PROTO: 3286 case RTA_SPORT: 3287 case RTA_DPORT: 3288 case RTA_MARK: 3289 case RTA_UID: 3290 break; 3291 default: 3292 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request"); 3293 return -EINVAL; 3294 } 3295 } 3296 3297 return 0; 3298 } 3299 3300 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, 3301 struct netlink_ext_ack *extack) 3302 { 3303 struct net *net = sock_net(in_skb->sk); 3304 struct nlattr *tb[RTA_MAX+1]; 3305 u32 table_id = RT_TABLE_MAIN; 3306 __be16 sport = 0, dport = 0; 3307 struct fib_result res = {}; 3308 u8 ip_proto = IPPROTO_UDP; 3309 struct rtable *rt = NULL; 3310 struct sk_buff *skb; 3311 struct rtmsg *rtm; 3312 struct flowi4 fl4 = {}; 3313 __be32 dst = 0; 3314 __be32 src = 0; 3315 kuid_t uid; 3316 u32 iif; 3317 int err; 3318 int mark; 3319 3320 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack); 3321 if (err < 0) 3322 return err; 3323 3324 rtm = nlmsg_data(nlh); 3325 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0; 3326 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0; 3327 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0; 3328 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0; 3329 if (tb[RTA_UID]) 3330 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID])); 3331 else 3332 uid = (iif ? INVALID_UID : current_uid()); 3333 3334 if (tb[RTA_IP_PROTO]) { 3335 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO], 3336 &ip_proto, AF_INET, extack); 3337 if (err) 3338 return err; 3339 } 3340 3341 if (tb[RTA_SPORT]) 3342 sport = nla_get_be16(tb[RTA_SPORT]); 3343 3344 if (tb[RTA_DPORT]) 3345 dport = nla_get_be16(tb[RTA_DPORT]); 3346 3347 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport); 3348 if (!skb) 3349 return -ENOBUFS; 3350 3351 fl4.daddr = dst; 3352 fl4.saddr = src; 3353 fl4.flowi4_tos = rtm->rtm_tos & IPTOS_RT_MASK; 3354 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0; 3355 fl4.flowi4_mark = mark; 3356 fl4.flowi4_uid = uid; 3357 if (sport) 3358 fl4.fl4_sport = sport; 3359 if (dport) 3360 fl4.fl4_dport = dport; 3361 fl4.flowi4_proto = ip_proto; 3362 3363 rcu_read_lock(); 3364 3365 if (iif) { 3366 struct net_device *dev; 3367 3368 dev = dev_get_by_index_rcu(net, iif); 3369 if (!dev) { 3370 err = -ENODEV; 3371 goto errout_rcu; 3372 } 3373 3374 fl4.flowi4_iif = iif; /* for rt_fill_info */ 3375 skb->dev = dev; 3376 skb->mark = mark; 3377 err = ip_route_input_rcu(skb, dst, src, 3378 rtm->rtm_tos & IPTOS_RT_MASK, dev, 3379 &res); 3380 3381 rt = skb_rtable(skb); 3382 if (err == 0 && rt->dst.error) 3383 err = -rt->dst.error; 3384 } else { 3385 fl4.flowi4_iif = LOOPBACK_IFINDEX; 3386 skb->dev = net->loopback_dev; 3387 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb); 3388 err = 0; 3389 if (IS_ERR(rt)) 3390 err = PTR_ERR(rt); 3391 else 3392 skb_dst_set(skb, &rt->dst); 3393 } 3394 3395 if (err) 3396 goto errout_rcu; 3397 3398 if (rtm->rtm_flags & RTM_F_NOTIFY) 3399 rt->rt_flags |= RTCF_NOTIFY; 3400 3401 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE) 3402 table_id = res.table ? res.table->tb_id : 0; 3403 3404 /* reset skb for netlink reply msg */ 3405 skb_trim(skb, 0); 3406 skb_reset_network_header(skb); 3407 skb_reset_transport_header(skb); 3408 skb_reset_mac_header(skb); 3409 3410 if (rtm->rtm_flags & RTM_F_FIB_MATCH) { 3411 struct fib_rt_info fri; 3412 3413 if (!res.fi) { 3414 err = fib_props[res.type].error; 3415 if (!err) 3416 err = -EHOSTUNREACH; 3417 goto errout_rcu; 3418 } 3419 fri.fi = res.fi; 3420 fri.tb_id = table_id; 3421 fri.dst = res.prefix; 3422 fri.dst_len = res.prefixlen; 3423 fri.dscp = res.dscp; 3424 fri.type = rt->rt_type; 3425 fri.offload = 0; 3426 fri.trap = 0; 3427 fri.offload_failed = 0; 3428 if (res.fa_head) { 3429 struct fib_alias *fa; 3430 3431 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) { 3432 u8 slen = 32 - fri.dst_len; 3433 3434 if (fa->fa_slen == slen && 3435 fa->tb_id == fri.tb_id && 3436 fa->fa_dscp == fri.dscp && 3437 fa->fa_info == res.fi && 3438 fa->fa_type == fri.type) { 3439 fri.offload = READ_ONCE(fa->offload); 3440 fri.trap = READ_ONCE(fa->trap); 3441 fri.offload_failed = 3442 READ_ONCE(fa->offload_failed); 3443 break; 3444 } 3445 } 3446 } 3447 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid, 3448 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0); 3449 } else { 3450 err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4, 3451 skb, NETLINK_CB(in_skb).portid, 3452 nlh->nlmsg_seq, 0); 3453 } 3454 if (err < 0) 3455 goto errout_rcu; 3456 3457 rcu_read_unlock(); 3458 3459 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 3460 3461 errout_free: 3462 return err; 3463 errout_rcu: 3464 rcu_read_unlock(); 3465 kfree_skb(skb); 3466 goto errout_free; 3467 } 3468 3469 void ip_rt_multicast_event(struct in_device *in_dev) 3470 { 3471 rt_cache_flush(dev_net(in_dev->dev)); 3472 } 3473 3474 #ifdef CONFIG_SYSCTL 3475 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 3476 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 3477 static int ip_rt_gc_elasticity __read_mostly = 8; 3478 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU; 3479 3480 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write, 3481 void *buffer, size_t *lenp, loff_t *ppos) 3482 { 3483 struct net *net = (struct net *)__ctl->extra1; 3484 3485 if (write) { 3486 rt_cache_flush(net); 3487 fnhe_genid_bump(net); 3488 return 0; 3489 } 3490 3491 return -EINVAL; 3492 } 3493 3494 static struct ctl_table ipv4_route_table[] = { 3495 { 3496 .procname = "gc_thresh", 3497 .data = &ipv4_dst_ops.gc_thresh, 3498 .maxlen = sizeof(int), 3499 .mode = 0644, 3500 .proc_handler = proc_dointvec, 3501 }, 3502 { 3503 .procname = "max_size", 3504 .data = &ip_rt_max_size, 3505 .maxlen = sizeof(int), 3506 .mode = 0644, 3507 .proc_handler = proc_dointvec, 3508 }, 3509 { 3510 /* Deprecated. Use gc_min_interval_ms */ 3511 3512 .procname = "gc_min_interval", 3513 .data = &ip_rt_gc_min_interval, 3514 .maxlen = sizeof(int), 3515 .mode = 0644, 3516 .proc_handler = proc_dointvec_jiffies, 3517 }, 3518 { 3519 .procname = "gc_min_interval_ms", 3520 .data = &ip_rt_gc_min_interval, 3521 .maxlen = sizeof(int), 3522 .mode = 0644, 3523 .proc_handler = proc_dointvec_ms_jiffies, 3524 }, 3525 { 3526 .procname = "gc_timeout", 3527 .data = &ip_rt_gc_timeout, 3528 .maxlen = sizeof(int), 3529 .mode = 0644, 3530 .proc_handler = proc_dointvec_jiffies, 3531 }, 3532 { 3533 .procname = "gc_interval", 3534 .data = &ip_rt_gc_interval, 3535 .maxlen = sizeof(int), 3536 .mode = 0644, 3537 .proc_handler = proc_dointvec_jiffies, 3538 }, 3539 { 3540 .procname = "redirect_load", 3541 .data = &ip_rt_redirect_load, 3542 .maxlen = sizeof(int), 3543 .mode = 0644, 3544 .proc_handler = proc_dointvec, 3545 }, 3546 { 3547 .procname = "redirect_number", 3548 .data = &ip_rt_redirect_number, 3549 .maxlen = sizeof(int), 3550 .mode = 0644, 3551 .proc_handler = proc_dointvec, 3552 }, 3553 { 3554 .procname = "redirect_silence", 3555 .data = &ip_rt_redirect_silence, 3556 .maxlen = sizeof(int), 3557 .mode = 0644, 3558 .proc_handler = proc_dointvec, 3559 }, 3560 { 3561 .procname = "error_cost", 3562 .data = &ip_rt_error_cost, 3563 .maxlen = sizeof(int), 3564 .mode = 0644, 3565 .proc_handler = proc_dointvec, 3566 }, 3567 { 3568 .procname = "error_burst", 3569 .data = &ip_rt_error_burst, 3570 .maxlen = sizeof(int), 3571 .mode = 0644, 3572 .proc_handler = proc_dointvec, 3573 }, 3574 { 3575 .procname = "gc_elasticity", 3576 .data = &ip_rt_gc_elasticity, 3577 .maxlen = sizeof(int), 3578 .mode = 0644, 3579 .proc_handler = proc_dointvec, 3580 }, 3581 { } 3582 }; 3583 3584 static const char ipv4_route_flush_procname[] = "flush"; 3585 3586 static struct ctl_table ipv4_route_netns_table[] = { 3587 { 3588 .procname = ipv4_route_flush_procname, 3589 .maxlen = sizeof(int), 3590 .mode = 0200, 3591 .proc_handler = ipv4_sysctl_rtcache_flush, 3592 }, 3593 { 3594 .procname = "min_pmtu", 3595 .data = &init_net.ipv4.ip_rt_min_pmtu, 3596 .maxlen = sizeof(int), 3597 .mode = 0644, 3598 .proc_handler = proc_dointvec_minmax, 3599 .extra1 = &ip_min_valid_pmtu, 3600 }, 3601 { 3602 .procname = "mtu_expires", 3603 .data = &init_net.ipv4.ip_rt_mtu_expires, 3604 .maxlen = sizeof(int), 3605 .mode = 0644, 3606 .proc_handler = proc_dointvec_jiffies, 3607 }, 3608 { 3609 .procname = "min_adv_mss", 3610 .data = &init_net.ipv4.ip_rt_min_advmss, 3611 .maxlen = sizeof(int), 3612 .mode = 0644, 3613 .proc_handler = proc_dointvec, 3614 }, 3615 { }, 3616 }; 3617 3618 static __net_init int sysctl_route_net_init(struct net *net) 3619 { 3620 struct ctl_table *tbl; 3621 size_t table_size = ARRAY_SIZE(ipv4_route_netns_table); 3622 3623 tbl = ipv4_route_netns_table; 3624 if (!net_eq(net, &init_net)) { 3625 int i; 3626 3627 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL); 3628 if (!tbl) 3629 goto err_dup; 3630 3631 /* Don't export non-whitelisted sysctls to unprivileged users */ 3632 if (net->user_ns != &init_user_ns) { 3633 if (tbl[0].procname != ipv4_route_flush_procname) { 3634 tbl[0].procname = NULL; 3635 table_size = 0; 3636 } 3637 } 3638 3639 /* Update the variables to point into the current struct net 3640 * except for the first element flush 3641 */ 3642 for (i = 1; i < ARRAY_SIZE(ipv4_route_netns_table) - 1; i++) 3643 tbl[i].data += (void *)net - (void *)&init_net; 3644 } 3645 tbl[0].extra1 = net; 3646 3647 net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route", 3648 tbl, table_size); 3649 if (!net->ipv4.route_hdr) 3650 goto err_reg; 3651 return 0; 3652 3653 err_reg: 3654 if (tbl != ipv4_route_netns_table) 3655 kfree(tbl); 3656 err_dup: 3657 return -ENOMEM; 3658 } 3659 3660 static __net_exit void sysctl_route_net_exit(struct net *net) 3661 { 3662 struct ctl_table *tbl; 3663 3664 tbl = net->ipv4.route_hdr->ctl_table_arg; 3665 unregister_net_sysctl_table(net->ipv4.route_hdr); 3666 BUG_ON(tbl == ipv4_route_netns_table); 3667 kfree(tbl); 3668 } 3669 3670 static __net_initdata struct pernet_operations sysctl_route_ops = { 3671 .init = sysctl_route_net_init, 3672 .exit = sysctl_route_net_exit, 3673 }; 3674 #endif 3675 3676 static __net_init int netns_ip_rt_init(struct net *net) 3677 { 3678 /* Set default value for namespaceified sysctls */ 3679 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU; 3680 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES; 3681 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS; 3682 return 0; 3683 } 3684 3685 static struct pernet_operations __net_initdata ip_rt_ops = { 3686 .init = netns_ip_rt_init, 3687 }; 3688 3689 static __net_init int rt_genid_init(struct net *net) 3690 { 3691 atomic_set(&net->ipv4.rt_genid, 0); 3692 atomic_set(&net->fnhe_genid, 0); 3693 atomic_set(&net->ipv4.dev_addr_genid, get_random_u32()); 3694 return 0; 3695 } 3696 3697 static __net_initdata struct pernet_operations rt_genid_ops = { 3698 .init = rt_genid_init, 3699 }; 3700 3701 static int __net_init ipv4_inetpeer_init(struct net *net) 3702 { 3703 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL); 3704 3705 if (!bp) 3706 return -ENOMEM; 3707 inet_peer_base_init(bp); 3708 net->ipv4.peers = bp; 3709 return 0; 3710 } 3711 3712 static void __net_exit ipv4_inetpeer_exit(struct net *net) 3713 { 3714 struct inet_peer_base *bp = net->ipv4.peers; 3715 3716 net->ipv4.peers = NULL; 3717 inetpeer_invalidate_tree(bp); 3718 kfree(bp); 3719 } 3720 3721 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = { 3722 .init = ipv4_inetpeer_init, 3723 .exit = ipv4_inetpeer_exit, 3724 }; 3725 3726 #ifdef CONFIG_IP_ROUTE_CLASSID 3727 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 3728 #endif /* CONFIG_IP_ROUTE_CLASSID */ 3729 3730 int __init ip_rt_init(void) 3731 { 3732 void *idents_hash; 3733 int cpu; 3734 3735 /* For modern hosts, this will use 2 MB of memory */ 3736 idents_hash = alloc_large_system_hash("IP idents", 3737 sizeof(*ip_idents) + sizeof(*ip_tstamps), 3738 0, 3739 16, /* one bucket per 64 KB */ 3740 HASH_ZERO, 3741 NULL, 3742 &ip_idents_mask, 3743 2048, 3744 256*1024); 3745 3746 ip_idents = idents_hash; 3747 3748 get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents)); 3749 3750 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents); 3751 3752 for_each_possible_cpu(cpu) { 3753 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 3754 3755 INIT_LIST_HEAD(&ul->head); 3756 INIT_LIST_HEAD(&ul->quarantine); 3757 spin_lock_init(&ul->lock); 3758 } 3759 #ifdef CONFIG_IP_ROUTE_CLASSID 3760 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 3761 if (!ip_rt_acct) 3762 panic("IP: failed to allocate ip_rt_acct\n"); 3763 #endif 3764 3765 ipv4_dst_ops.kmem_cachep = 3766 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0, 3767 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 3768 3769 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 3770 3771 if (dst_entries_init(&ipv4_dst_ops) < 0) 3772 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 3773 3774 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 3775 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 3776 3777 ipv4_dst_ops.gc_thresh = ~0; 3778 ip_rt_max_size = INT_MAX; 3779 3780 devinet_init(); 3781 ip_fib_init(); 3782 3783 if (ip_rt_proc_init()) 3784 pr_err("Unable to create route proc files\n"); 3785 #ifdef CONFIG_XFRM 3786 xfrm_init(); 3787 xfrm4_init(); 3788 #endif 3789 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, 3790 RTNL_FLAG_DOIT_UNLOCKED); 3791 3792 #ifdef CONFIG_SYSCTL 3793 register_pernet_subsys(&sysctl_route_ops); 3794 #endif 3795 register_pernet_subsys(&ip_rt_ops); 3796 register_pernet_subsys(&rt_genid_ops); 3797 register_pernet_subsys(&ipv4_inetpeer_ops); 3798 return 0; 3799 } 3800 3801 #ifdef CONFIG_SYSCTL 3802 /* 3803 * We really need to sanitize the damn ipv4 init order, then all 3804 * this nonsense will go away. 3805 */ 3806 void __init ip_static_sysctl_init(void) 3807 { 3808 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table); 3809 } 3810 #endif 3811