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_GC_TIMEOUT (300*HZ) 110 111 #define DEFAULT_MIN_PMTU (512 + 20 + 20) 112 #define DEFAULT_MTU_EXPIRES (10 * 60 * HZ) 113 #define DEFAULT_MIN_ADVMSS 256 114 static int ip_rt_max_size; 115 static int ip_rt_redirect_number __read_mostly = 9; 116 static int ip_rt_redirect_load __read_mostly = HZ / 50; 117 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1)); 118 static int ip_rt_error_cost __read_mostly = HZ; 119 static int ip_rt_error_burst __read_mostly = 5 * HZ; 120 121 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT; 122 123 /* 124 * Interface to generic destination cache. 125 */ 126 127 INDIRECT_CALLABLE_SCOPE 128 struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie); 129 static unsigned int ipv4_default_advmss(const struct dst_entry *dst); 130 INDIRECT_CALLABLE_SCOPE 131 unsigned int ipv4_mtu(const struct dst_entry *dst); 132 static void ipv4_negative_advice(struct sock *sk, 133 struct dst_entry *dst); 134 static void ipv4_link_failure(struct sk_buff *skb); 135 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 136 struct sk_buff *skb, u32 mtu, 137 bool confirm_neigh); 138 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, 139 struct sk_buff *skb); 140 static void ipv4_dst_destroy(struct dst_entry *dst); 141 142 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old) 143 { 144 WARN_ON(1); 145 return NULL; 146 } 147 148 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 149 struct sk_buff *skb, 150 const void *daddr); 151 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr); 152 153 static struct dst_ops ipv4_dst_ops = { 154 .family = AF_INET, 155 .check = ipv4_dst_check, 156 .default_advmss = ipv4_default_advmss, 157 .mtu = ipv4_mtu, 158 .cow_metrics = ipv4_cow_metrics, 159 .destroy = ipv4_dst_destroy, 160 .negative_advice = ipv4_negative_advice, 161 .link_failure = ipv4_link_failure, 162 .update_pmtu = ip_rt_update_pmtu, 163 .redirect = ip_do_redirect, 164 .local_out = __ip_local_out, 165 .neigh_lookup = ipv4_neigh_lookup, 166 .confirm_neigh = ipv4_confirm_neigh, 167 }; 168 169 #define ECN_OR_COST(class) TC_PRIO_##class 170 171 const __u8 ip_tos2prio[16] = { 172 TC_PRIO_BESTEFFORT, 173 ECN_OR_COST(BESTEFFORT), 174 TC_PRIO_BESTEFFORT, 175 ECN_OR_COST(BESTEFFORT), 176 TC_PRIO_BULK, 177 ECN_OR_COST(BULK), 178 TC_PRIO_BULK, 179 ECN_OR_COST(BULK), 180 TC_PRIO_INTERACTIVE, 181 ECN_OR_COST(INTERACTIVE), 182 TC_PRIO_INTERACTIVE, 183 ECN_OR_COST(INTERACTIVE), 184 TC_PRIO_INTERACTIVE_BULK, 185 ECN_OR_COST(INTERACTIVE_BULK), 186 TC_PRIO_INTERACTIVE_BULK, 187 ECN_OR_COST(INTERACTIVE_BULK) 188 }; 189 EXPORT_SYMBOL(ip_tos2prio); 190 191 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat); 192 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field) 193 194 #ifdef CONFIG_PROC_FS 195 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos) 196 { 197 if (*pos) 198 return NULL; 199 return SEQ_START_TOKEN; 200 } 201 202 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos) 203 { 204 ++*pos; 205 return NULL; 206 } 207 208 static void rt_cache_seq_stop(struct seq_file *seq, void *v) 209 { 210 } 211 212 static int rt_cache_seq_show(struct seq_file *seq, void *v) 213 { 214 if (v == SEQ_START_TOKEN) 215 seq_printf(seq, "%-127s\n", 216 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t" 217 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t" 218 "HHUptod\tSpecDst"); 219 return 0; 220 } 221 222 static const struct seq_operations rt_cache_seq_ops = { 223 .start = rt_cache_seq_start, 224 .next = rt_cache_seq_next, 225 .stop = rt_cache_seq_stop, 226 .show = rt_cache_seq_show, 227 }; 228 229 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos) 230 { 231 int cpu; 232 233 if (*pos == 0) 234 return SEQ_START_TOKEN; 235 236 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) { 237 if (!cpu_possible(cpu)) 238 continue; 239 *pos = cpu+1; 240 return &per_cpu(rt_cache_stat, cpu); 241 } 242 return NULL; 243 } 244 245 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos) 246 { 247 int cpu; 248 249 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) { 250 if (!cpu_possible(cpu)) 251 continue; 252 *pos = cpu+1; 253 return &per_cpu(rt_cache_stat, cpu); 254 } 255 (*pos)++; 256 return NULL; 257 258 } 259 260 static void rt_cpu_seq_stop(struct seq_file *seq, void *v) 261 { 262 263 } 264 265 static int rt_cpu_seq_show(struct seq_file *seq, void *v) 266 { 267 struct rt_cache_stat *st = v; 268 269 if (v == SEQ_START_TOKEN) { 270 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"); 271 return 0; 272 } 273 274 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x " 275 "%08x %08x %08x %08x %08x %08x " 276 "%08x %08x %08x %08x\n", 277 dst_entries_get_slow(&ipv4_dst_ops), 278 0, /* st->in_hit */ 279 st->in_slow_tot, 280 st->in_slow_mc, 281 st->in_no_route, 282 st->in_brd, 283 st->in_martian_dst, 284 st->in_martian_src, 285 286 0, /* st->out_hit */ 287 st->out_slow_tot, 288 st->out_slow_mc, 289 290 0, /* st->gc_total */ 291 0, /* st->gc_ignored */ 292 0, /* st->gc_goal_miss */ 293 0, /* st->gc_dst_overflow */ 294 0, /* st->in_hlist_search */ 295 0 /* st->out_hlist_search */ 296 ); 297 return 0; 298 } 299 300 static const struct seq_operations rt_cpu_seq_ops = { 301 .start = rt_cpu_seq_start, 302 .next = rt_cpu_seq_next, 303 .stop = rt_cpu_seq_stop, 304 .show = rt_cpu_seq_show, 305 }; 306 307 #ifdef CONFIG_IP_ROUTE_CLASSID 308 static int rt_acct_proc_show(struct seq_file *m, void *v) 309 { 310 struct ip_rt_acct *dst, *src; 311 unsigned int i, j; 312 313 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL); 314 if (!dst) 315 return -ENOMEM; 316 317 for_each_possible_cpu(i) { 318 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i); 319 for (j = 0; j < 256; j++) { 320 dst[j].o_bytes += src[j].o_bytes; 321 dst[j].o_packets += src[j].o_packets; 322 dst[j].i_bytes += src[j].i_bytes; 323 dst[j].i_packets += src[j].i_packets; 324 } 325 } 326 327 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct)); 328 kfree(dst); 329 return 0; 330 } 331 #endif 332 333 static int __net_init ip_rt_do_proc_init(struct net *net) 334 { 335 struct proc_dir_entry *pde; 336 337 pde = proc_create_seq("rt_cache", 0444, net->proc_net, 338 &rt_cache_seq_ops); 339 if (!pde) 340 goto err1; 341 342 pde = proc_create_seq("rt_cache", 0444, net->proc_net_stat, 343 &rt_cpu_seq_ops); 344 if (!pde) 345 goto err2; 346 347 #ifdef CONFIG_IP_ROUTE_CLASSID 348 pde = proc_create_single("rt_acct", 0, net->proc_net, 349 rt_acct_proc_show); 350 if (!pde) 351 goto err3; 352 #endif 353 return 0; 354 355 #ifdef CONFIG_IP_ROUTE_CLASSID 356 err3: 357 remove_proc_entry("rt_cache", net->proc_net_stat); 358 #endif 359 err2: 360 remove_proc_entry("rt_cache", net->proc_net); 361 err1: 362 return -ENOMEM; 363 } 364 365 static void __net_exit ip_rt_do_proc_exit(struct net *net) 366 { 367 remove_proc_entry("rt_cache", net->proc_net_stat); 368 remove_proc_entry("rt_cache", net->proc_net); 369 #ifdef CONFIG_IP_ROUTE_CLASSID 370 remove_proc_entry("rt_acct", net->proc_net); 371 #endif 372 } 373 374 static struct pernet_operations ip_rt_proc_ops __net_initdata = { 375 .init = ip_rt_do_proc_init, 376 .exit = ip_rt_do_proc_exit, 377 }; 378 379 static int __init ip_rt_proc_init(void) 380 { 381 return register_pernet_subsys(&ip_rt_proc_ops); 382 } 383 384 #else 385 static inline int ip_rt_proc_init(void) 386 { 387 return 0; 388 } 389 #endif /* CONFIG_PROC_FS */ 390 391 static inline bool rt_is_expired(const struct rtable *rth) 392 { 393 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev)); 394 } 395 396 void rt_cache_flush(struct net *net) 397 { 398 rt_genid_bump_ipv4(net); 399 } 400 401 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 402 struct sk_buff *skb, 403 const void *daddr) 404 { 405 const struct rtable *rt = container_of(dst, struct rtable, dst); 406 struct net_device *dev = dst->dev; 407 struct neighbour *n; 408 409 rcu_read_lock(); 410 411 if (likely(rt->rt_gw_family == AF_INET)) { 412 n = ip_neigh_gw4(dev, rt->rt_gw4); 413 } else if (rt->rt_gw_family == AF_INET6) { 414 n = ip_neigh_gw6(dev, &rt->rt_gw6); 415 } else { 416 __be32 pkey; 417 418 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr); 419 n = ip_neigh_gw4(dev, pkey); 420 } 421 422 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt)) 423 n = NULL; 424 425 rcu_read_unlock(); 426 427 return n; 428 } 429 430 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr) 431 { 432 const struct rtable *rt = container_of(dst, struct rtable, dst); 433 struct net_device *dev = dst->dev; 434 const __be32 *pkey = daddr; 435 436 if (rt->rt_gw_family == AF_INET) { 437 pkey = (const __be32 *)&rt->rt_gw4; 438 } else if (rt->rt_gw_family == AF_INET6) { 439 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6); 440 } else if (!daddr || 441 (rt->rt_flags & 442 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) { 443 return; 444 } 445 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey); 446 } 447 448 /* Hash tables of size 2048..262144 depending on RAM size. 449 * Each bucket uses 8 bytes. 450 */ 451 static u32 ip_idents_mask __read_mostly; 452 static atomic_t *ip_idents __read_mostly; 453 static u32 *ip_tstamps __read_mostly; 454 455 /* In order to protect privacy, we add a perturbation to identifiers 456 * if one generator is seldom used. This makes hard for an attacker 457 * to infer how many packets were sent between two points in time. 458 */ 459 static u32 ip_idents_reserve(u32 hash, int segs) 460 { 461 u32 bucket, old, now = (u32)jiffies; 462 atomic_t *p_id; 463 u32 *p_tstamp; 464 u32 delta = 0; 465 466 bucket = hash & ip_idents_mask; 467 p_tstamp = ip_tstamps + bucket; 468 p_id = ip_idents + bucket; 469 old = READ_ONCE(*p_tstamp); 470 471 if (old != now && cmpxchg(p_tstamp, old, now) == old) 472 delta = get_random_u32_below(now - old); 473 474 /* If UBSAN reports an error there, please make sure your compiler 475 * supports -fno-strict-overflow before reporting it that was a bug 476 * in UBSAN, and it has been fixed in GCC-8. 477 */ 478 return atomic_add_return(segs + delta, p_id) - segs; 479 } 480 481 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs) 482 { 483 u32 hash, id; 484 485 /* Note the following code is not safe, but this is okay. */ 486 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key))) 487 get_random_bytes(&net->ipv4.ip_id_key, 488 sizeof(net->ipv4.ip_id_key)); 489 490 hash = siphash_3u32((__force u32)iph->daddr, 491 (__force u32)iph->saddr, 492 iph->protocol, 493 &net->ipv4.ip_id_key); 494 id = ip_idents_reserve(hash, segs); 495 iph->id = htons(id); 496 } 497 EXPORT_SYMBOL(__ip_select_ident); 498 499 static void __build_flow_key(const struct net *net, struct flowi4 *fl4, 500 const struct sock *sk, const struct iphdr *iph, 501 int oif, __u8 tos, u8 prot, u32 mark, 502 int flow_flags) 503 { 504 __u8 scope = RT_SCOPE_UNIVERSE; 505 506 if (sk) { 507 oif = sk->sk_bound_dev_if; 508 mark = READ_ONCE(sk->sk_mark); 509 tos = ip_sock_rt_tos(sk); 510 scope = ip_sock_rt_scope(sk); 511 prot = inet_test_bit(HDRINCL, sk) ? IPPROTO_RAW : 512 sk->sk_protocol; 513 } 514 515 flowi4_init_output(fl4, oif, mark, tos & INET_DSCP_MASK, scope, 516 prot, flow_flags, iph->daddr, iph->saddr, 0, 0, 517 sock_net_uid(net, sk)); 518 } 519 520 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb, 521 const struct sock *sk) 522 { 523 const struct net *net = dev_net(skb->dev); 524 const struct iphdr *iph = ip_hdr(skb); 525 int oif = skb->dev->ifindex; 526 u8 prot = iph->protocol; 527 u32 mark = skb->mark; 528 __u8 tos = iph->tos; 529 530 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0); 531 } 532 533 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk) 534 { 535 const struct inet_sock *inet = inet_sk(sk); 536 const struct ip_options_rcu *inet_opt; 537 __be32 daddr = inet->inet_daddr; 538 539 rcu_read_lock(); 540 inet_opt = rcu_dereference(inet->inet_opt); 541 if (inet_opt && inet_opt->opt.srr) 542 daddr = inet_opt->opt.faddr; 543 flowi4_init_output(fl4, sk->sk_bound_dev_if, READ_ONCE(sk->sk_mark), 544 ip_sock_rt_tos(sk), 545 ip_sock_rt_scope(sk), 546 inet_test_bit(HDRINCL, sk) ? 547 IPPROTO_RAW : sk->sk_protocol, 548 inet_sk_flowi_flags(sk), 549 daddr, inet->inet_saddr, 0, 0, sk->sk_uid); 550 rcu_read_unlock(); 551 } 552 553 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk, 554 const struct sk_buff *skb) 555 { 556 if (skb) 557 build_skb_flow_key(fl4, skb, sk); 558 else 559 build_sk_flow_key(fl4, sk); 560 } 561 562 static DEFINE_SPINLOCK(fnhe_lock); 563 564 static void fnhe_flush_routes(struct fib_nh_exception *fnhe) 565 { 566 struct rtable *rt; 567 568 rt = rcu_dereference(fnhe->fnhe_rth_input); 569 if (rt) { 570 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL); 571 dst_dev_put(&rt->dst); 572 dst_release(&rt->dst); 573 } 574 rt = rcu_dereference(fnhe->fnhe_rth_output); 575 if (rt) { 576 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL); 577 dst_dev_put(&rt->dst); 578 dst_release(&rt->dst); 579 } 580 } 581 582 static void fnhe_remove_oldest(struct fnhe_hash_bucket *hash) 583 { 584 struct fib_nh_exception __rcu **fnhe_p, **oldest_p; 585 struct fib_nh_exception *fnhe, *oldest = NULL; 586 587 for (fnhe_p = &hash->chain; ; fnhe_p = &fnhe->fnhe_next) { 588 fnhe = rcu_dereference_protected(*fnhe_p, 589 lockdep_is_held(&fnhe_lock)); 590 if (!fnhe) 591 break; 592 if (!oldest || 593 time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) { 594 oldest = fnhe; 595 oldest_p = fnhe_p; 596 } 597 } 598 fnhe_flush_routes(oldest); 599 *oldest_p = oldest->fnhe_next; 600 kfree_rcu(oldest, rcu); 601 } 602 603 static u32 fnhe_hashfun(__be32 daddr) 604 { 605 static siphash_aligned_key_t fnhe_hash_key; 606 u64 hval; 607 608 net_get_random_once(&fnhe_hash_key, sizeof(fnhe_hash_key)); 609 hval = siphash_1u32((__force u32)daddr, &fnhe_hash_key); 610 return hash_64(hval, FNHE_HASH_SHIFT); 611 } 612 613 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe) 614 { 615 rt->rt_pmtu = fnhe->fnhe_pmtu; 616 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked; 617 rt->dst.expires = fnhe->fnhe_expires; 618 619 if (fnhe->fnhe_gw) { 620 rt->rt_flags |= RTCF_REDIRECTED; 621 rt->rt_uses_gateway = 1; 622 rt->rt_gw_family = AF_INET; 623 rt->rt_gw4 = fnhe->fnhe_gw; 624 } 625 } 626 627 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr, 628 __be32 gw, u32 pmtu, bool lock, 629 unsigned long expires) 630 { 631 struct fnhe_hash_bucket *hash; 632 struct fib_nh_exception *fnhe; 633 struct rtable *rt; 634 u32 genid, hval; 635 unsigned int i; 636 int depth; 637 638 genid = fnhe_genid(dev_net(nhc->nhc_dev)); 639 hval = fnhe_hashfun(daddr); 640 641 spin_lock_bh(&fnhe_lock); 642 643 hash = rcu_dereference(nhc->nhc_exceptions); 644 if (!hash) { 645 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC); 646 if (!hash) 647 goto out_unlock; 648 rcu_assign_pointer(nhc->nhc_exceptions, hash); 649 } 650 651 hash += hval; 652 653 depth = 0; 654 for (fnhe = rcu_dereference(hash->chain); fnhe; 655 fnhe = rcu_dereference(fnhe->fnhe_next)) { 656 if (fnhe->fnhe_daddr == daddr) 657 break; 658 depth++; 659 } 660 661 if (fnhe) { 662 if (fnhe->fnhe_genid != genid) 663 fnhe->fnhe_genid = genid; 664 if (gw) 665 fnhe->fnhe_gw = gw; 666 if (pmtu) { 667 fnhe->fnhe_pmtu = pmtu; 668 fnhe->fnhe_mtu_locked = lock; 669 } 670 fnhe->fnhe_expires = max(1UL, expires); 671 /* Update all cached dsts too */ 672 rt = rcu_dereference(fnhe->fnhe_rth_input); 673 if (rt) 674 fill_route_from_fnhe(rt, fnhe); 675 rt = rcu_dereference(fnhe->fnhe_rth_output); 676 if (rt) 677 fill_route_from_fnhe(rt, fnhe); 678 } else { 679 /* Randomize max depth to avoid some side channels attacks. */ 680 int max_depth = FNHE_RECLAIM_DEPTH + 681 get_random_u32_below(FNHE_RECLAIM_DEPTH); 682 683 while (depth > max_depth) { 684 fnhe_remove_oldest(hash); 685 depth--; 686 } 687 688 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC); 689 if (!fnhe) 690 goto out_unlock; 691 692 fnhe->fnhe_next = hash->chain; 693 694 fnhe->fnhe_genid = genid; 695 fnhe->fnhe_daddr = daddr; 696 fnhe->fnhe_gw = gw; 697 fnhe->fnhe_pmtu = pmtu; 698 fnhe->fnhe_mtu_locked = lock; 699 fnhe->fnhe_expires = max(1UL, expires); 700 701 rcu_assign_pointer(hash->chain, fnhe); 702 703 /* Exception created; mark the cached routes for the nexthop 704 * stale, so anyone caching it rechecks if this exception 705 * applies to them. 706 */ 707 rt = rcu_dereference(nhc->nhc_rth_input); 708 if (rt) 709 rt->dst.obsolete = DST_OBSOLETE_KILL; 710 711 for_each_possible_cpu(i) { 712 struct rtable __rcu **prt; 713 714 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i); 715 rt = rcu_dereference(*prt); 716 if (rt) 717 rt->dst.obsolete = DST_OBSOLETE_KILL; 718 } 719 } 720 721 fnhe->fnhe_stamp = jiffies; 722 723 out_unlock: 724 spin_unlock_bh(&fnhe_lock); 725 } 726 727 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4, 728 bool kill_route) 729 { 730 __be32 new_gw = icmp_hdr(skb)->un.gateway; 731 __be32 old_gw = ip_hdr(skb)->saddr; 732 struct net_device *dev = skb->dev; 733 struct in_device *in_dev; 734 struct fib_result res; 735 struct neighbour *n; 736 struct net *net; 737 738 switch (icmp_hdr(skb)->code & 7) { 739 case ICMP_REDIR_NET: 740 case ICMP_REDIR_NETTOS: 741 case ICMP_REDIR_HOST: 742 case ICMP_REDIR_HOSTTOS: 743 break; 744 745 default: 746 return; 747 } 748 749 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw) 750 return; 751 752 in_dev = __in_dev_get_rcu(dev); 753 if (!in_dev) 754 return; 755 756 net = dev_net(dev); 757 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) || 758 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) || 759 ipv4_is_zeronet(new_gw)) 760 goto reject_redirect; 761 762 if (!IN_DEV_SHARED_MEDIA(in_dev)) { 763 if (!inet_addr_onlink(in_dev, new_gw, old_gw)) 764 goto reject_redirect; 765 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) 766 goto reject_redirect; 767 } else { 768 if (inet_addr_type(net, new_gw) != RTN_UNICAST) 769 goto reject_redirect; 770 } 771 772 n = __ipv4_neigh_lookup(rt->dst.dev, (__force u32)new_gw); 773 if (!n) 774 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev); 775 if (!IS_ERR(n)) { 776 if (!(READ_ONCE(n->nud_state) & NUD_VALID)) { 777 neigh_event_send(n, NULL); 778 } else { 779 if (fib_lookup(net, fl4, &res, 0) == 0) { 780 struct fib_nh_common *nhc; 781 782 fib_select_path(net, &res, fl4, skb); 783 nhc = FIB_RES_NHC(res); 784 update_or_create_fnhe(nhc, fl4->daddr, new_gw, 785 0, false, 786 jiffies + ip_rt_gc_timeout); 787 } 788 if (kill_route) 789 rt->dst.obsolete = DST_OBSOLETE_KILL; 790 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n); 791 } 792 neigh_release(n); 793 } 794 return; 795 796 reject_redirect: 797 #ifdef CONFIG_IP_ROUTE_VERBOSE 798 if (IN_DEV_LOG_MARTIANS(in_dev)) { 799 const struct iphdr *iph = (const struct iphdr *) skb->data; 800 __be32 daddr = iph->daddr; 801 __be32 saddr = iph->saddr; 802 803 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n" 804 " Advised path = %pI4 -> %pI4\n", 805 &old_gw, dev->name, &new_gw, 806 &saddr, &daddr); 807 } 808 #endif 809 ; 810 } 811 812 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb) 813 { 814 struct rtable *rt; 815 struct flowi4 fl4; 816 const struct iphdr *iph = (const struct iphdr *) skb->data; 817 struct net *net = dev_net(skb->dev); 818 int oif = skb->dev->ifindex; 819 u8 prot = iph->protocol; 820 u32 mark = skb->mark; 821 __u8 tos = iph->tos; 822 823 rt = dst_rtable(dst); 824 825 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0); 826 __ip_do_redirect(rt, skb, &fl4, true); 827 } 828 829 static void ipv4_negative_advice(struct sock *sk, 830 struct dst_entry *dst) 831 { 832 struct rtable *rt = dst_rtable(dst); 833 834 if ((dst->obsolete > 0) || 835 (rt->rt_flags & RTCF_REDIRECTED) || 836 rt->dst.expires) 837 sk_dst_reset(sk); 838 } 839 840 /* 841 * Algorithm: 842 * 1. The first ip_rt_redirect_number redirects are sent 843 * with exponential backoff, then we stop sending them at all, 844 * assuming that the host ignores our redirects. 845 * 2. If we did not see packets requiring redirects 846 * during ip_rt_redirect_silence, we assume that the host 847 * forgot redirected route and start to send redirects again. 848 * 849 * This algorithm is much cheaper and more intelligent than dumb load limiting 850 * in icmp.c. 851 * 852 * NOTE. Do not forget to inhibit load limiting for redirects (redundant) 853 * and "frag. need" (breaks PMTU discovery) in icmp.c. 854 */ 855 856 void ip_rt_send_redirect(struct sk_buff *skb) 857 { 858 struct rtable *rt = skb_rtable(skb); 859 struct in_device *in_dev; 860 struct inet_peer *peer; 861 struct net *net; 862 int log_martians; 863 int vif; 864 865 rcu_read_lock(); 866 in_dev = __in_dev_get_rcu(rt->dst.dev); 867 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) { 868 rcu_read_unlock(); 869 return; 870 } 871 log_martians = IN_DEV_LOG_MARTIANS(in_dev); 872 vif = l3mdev_master_ifindex_rcu(rt->dst.dev); 873 874 net = dev_net(rt->dst.dev); 875 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif); 876 if (!peer) { 877 rcu_read_unlock(); 878 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, 879 rt_nexthop(rt, ip_hdr(skb)->daddr)); 880 return; 881 } 882 883 /* No redirected packets during ip_rt_redirect_silence; 884 * reset the algorithm. 885 */ 886 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) { 887 peer->rate_tokens = 0; 888 peer->n_redirects = 0; 889 } 890 891 /* Too many ignored redirects; do not send anything 892 * set dst.rate_last to the last seen redirected packet. 893 */ 894 if (peer->n_redirects >= ip_rt_redirect_number) { 895 peer->rate_last = jiffies; 896 goto out_unlock; 897 } 898 899 /* Check for load limit; set rate_last to the latest sent 900 * redirect. 901 */ 902 if (peer->n_redirects == 0 || 903 time_after(jiffies, 904 (peer->rate_last + 905 (ip_rt_redirect_load << peer->n_redirects)))) { 906 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr); 907 908 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw); 909 peer->rate_last = jiffies; 910 ++peer->n_redirects; 911 if (IS_ENABLED(CONFIG_IP_ROUTE_VERBOSE) && log_martians && 912 peer->n_redirects == ip_rt_redirect_number) 913 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n", 914 &ip_hdr(skb)->saddr, inet_iif(skb), 915 &ip_hdr(skb)->daddr, &gw); 916 } 917 out_unlock: 918 rcu_read_unlock(); 919 } 920 921 static int ip_error(struct sk_buff *skb) 922 { 923 struct rtable *rt = skb_rtable(skb); 924 struct net_device *dev = skb->dev; 925 struct in_device *in_dev; 926 struct inet_peer *peer; 927 unsigned long now; 928 struct net *net; 929 SKB_DR(reason); 930 bool send; 931 int code; 932 933 if (netif_is_l3_master(skb->dev)) { 934 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif); 935 if (!dev) 936 goto out; 937 } 938 939 in_dev = __in_dev_get_rcu(dev); 940 941 /* IP on this device is disabled. */ 942 if (!in_dev) 943 goto out; 944 945 net = dev_net(rt->dst.dev); 946 if (!IN_DEV_FORWARD(in_dev)) { 947 switch (rt->dst.error) { 948 case EHOSTUNREACH: 949 SKB_DR_SET(reason, IP_INADDRERRORS); 950 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS); 951 break; 952 953 case ENETUNREACH: 954 SKB_DR_SET(reason, IP_INNOROUTES); 955 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 956 break; 957 } 958 goto out; 959 } 960 961 switch (rt->dst.error) { 962 case EINVAL: 963 default: 964 goto out; 965 case EHOSTUNREACH: 966 code = ICMP_HOST_UNREACH; 967 break; 968 case ENETUNREACH: 969 code = ICMP_NET_UNREACH; 970 SKB_DR_SET(reason, IP_INNOROUTES); 971 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 972 break; 973 case EACCES: 974 code = ICMP_PKT_FILTERED; 975 break; 976 } 977 978 rcu_read_lock(); 979 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 980 l3mdev_master_ifindex_rcu(skb->dev)); 981 send = true; 982 if (peer) { 983 now = jiffies; 984 peer->rate_tokens += now - peer->rate_last; 985 if (peer->rate_tokens > ip_rt_error_burst) 986 peer->rate_tokens = ip_rt_error_burst; 987 peer->rate_last = now; 988 if (peer->rate_tokens >= ip_rt_error_cost) 989 peer->rate_tokens -= ip_rt_error_cost; 990 else 991 send = false; 992 } 993 rcu_read_unlock(); 994 995 if (send) 996 icmp_send(skb, ICMP_DEST_UNREACH, code, 0); 997 998 out: kfree_skb_reason(skb, reason); 999 return 0; 1000 } 1001 1002 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu) 1003 { 1004 struct dst_entry *dst = &rt->dst; 1005 struct net *net = dev_net(dst->dev); 1006 struct fib_result res; 1007 bool lock = false; 1008 u32 old_mtu; 1009 1010 if (ip_mtu_locked(dst)) 1011 return; 1012 1013 old_mtu = ipv4_mtu(dst); 1014 if (old_mtu < mtu) 1015 return; 1016 1017 if (mtu < net->ipv4.ip_rt_min_pmtu) { 1018 lock = true; 1019 mtu = min(old_mtu, net->ipv4.ip_rt_min_pmtu); 1020 } 1021 1022 if (rt->rt_pmtu == mtu && !lock && 1023 time_before(jiffies, dst->expires - net->ipv4.ip_rt_mtu_expires / 2)) 1024 return; 1025 1026 rcu_read_lock(); 1027 if (fib_lookup(net, fl4, &res, 0) == 0) { 1028 struct fib_nh_common *nhc; 1029 1030 fib_select_path(net, &res, fl4, NULL); 1031 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1032 if (fib_info_num_path(res.fi) > 1) { 1033 int nhsel; 1034 1035 for (nhsel = 0; nhsel < fib_info_num_path(res.fi); nhsel++) { 1036 nhc = fib_info_nhc(res.fi, nhsel); 1037 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1038 jiffies + net->ipv4.ip_rt_mtu_expires); 1039 } 1040 rcu_read_unlock(); 1041 return; 1042 } 1043 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 1044 nhc = FIB_RES_NHC(res); 1045 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1046 jiffies + net->ipv4.ip_rt_mtu_expires); 1047 } 1048 rcu_read_unlock(); 1049 } 1050 1051 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 1052 struct sk_buff *skb, u32 mtu, 1053 bool confirm_neigh) 1054 { 1055 struct rtable *rt = dst_rtable(dst); 1056 struct flowi4 fl4; 1057 1058 ip_rt_build_flow_key(&fl4, sk, skb); 1059 1060 /* Don't make lookup fail for bridged encapsulations */ 1061 if (skb && netif_is_any_bridge_port(skb->dev)) 1062 fl4.flowi4_oif = 0; 1063 1064 __ip_rt_update_pmtu(rt, &fl4, mtu); 1065 } 1066 1067 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, 1068 int oif, u8 protocol) 1069 { 1070 const struct iphdr *iph = (const struct iphdr *)skb->data; 1071 struct flowi4 fl4; 1072 struct rtable *rt; 1073 u32 mark = IP4_REPLY_MARK(net, skb->mark); 1074 1075 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, mark, 1076 0); 1077 rt = __ip_route_output_key(net, &fl4); 1078 if (!IS_ERR(rt)) { 1079 __ip_rt_update_pmtu(rt, &fl4, mtu); 1080 ip_rt_put(rt); 1081 } 1082 } 1083 EXPORT_SYMBOL_GPL(ipv4_update_pmtu); 1084 1085 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1086 { 1087 const struct iphdr *iph = (const struct iphdr *)skb->data; 1088 struct flowi4 fl4; 1089 struct rtable *rt; 1090 1091 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0); 1092 1093 if (!fl4.flowi4_mark) 1094 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark); 1095 1096 rt = __ip_route_output_key(sock_net(sk), &fl4); 1097 if (!IS_ERR(rt)) { 1098 __ip_rt_update_pmtu(rt, &fl4, mtu); 1099 ip_rt_put(rt); 1100 } 1101 } 1102 1103 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1104 { 1105 const struct iphdr *iph = (const struct iphdr *)skb->data; 1106 struct flowi4 fl4; 1107 struct rtable *rt; 1108 struct dst_entry *odst = NULL; 1109 bool new = false; 1110 struct net *net = sock_net(sk); 1111 1112 bh_lock_sock(sk); 1113 1114 if (!ip_sk_accept_pmtu(sk)) 1115 goto out; 1116 1117 odst = sk_dst_get(sk); 1118 1119 if (sock_owned_by_user(sk) || !odst) { 1120 __ipv4_sk_update_pmtu(skb, sk, mtu); 1121 goto out; 1122 } 1123 1124 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1125 1126 rt = dst_rtable(odst); 1127 if (odst->obsolete && !odst->ops->check(odst, 0)) { 1128 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1129 if (IS_ERR(rt)) 1130 goto out; 1131 1132 new = true; 1133 } 1134 1135 __ip_rt_update_pmtu(dst_rtable(xfrm_dst_path(&rt->dst)), &fl4, mtu); 1136 1137 if (!dst_check(&rt->dst, 0)) { 1138 if (new) 1139 dst_release(&rt->dst); 1140 1141 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1142 if (IS_ERR(rt)) 1143 goto out; 1144 1145 new = true; 1146 } 1147 1148 if (new) 1149 sk_dst_set(sk, &rt->dst); 1150 1151 out: 1152 bh_unlock_sock(sk); 1153 dst_release(odst); 1154 } 1155 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu); 1156 1157 void ipv4_redirect(struct sk_buff *skb, struct net *net, 1158 int oif, u8 protocol) 1159 { 1160 const struct iphdr *iph = (const struct iphdr *)skb->data; 1161 struct flowi4 fl4; 1162 struct rtable *rt; 1163 1164 __build_flow_key(net, &fl4, NULL, iph, oif, iph->tos, protocol, 0, 0); 1165 rt = __ip_route_output_key(net, &fl4); 1166 if (!IS_ERR(rt)) { 1167 __ip_do_redirect(rt, skb, &fl4, false); 1168 ip_rt_put(rt); 1169 } 1170 } 1171 EXPORT_SYMBOL_GPL(ipv4_redirect); 1172 1173 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk) 1174 { 1175 const struct iphdr *iph = (const struct iphdr *)skb->data; 1176 struct flowi4 fl4; 1177 struct rtable *rt; 1178 struct net *net = sock_net(sk); 1179 1180 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1181 rt = __ip_route_output_key(net, &fl4); 1182 if (!IS_ERR(rt)) { 1183 __ip_do_redirect(rt, skb, &fl4, false); 1184 ip_rt_put(rt); 1185 } 1186 } 1187 EXPORT_SYMBOL_GPL(ipv4_sk_redirect); 1188 1189 INDIRECT_CALLABLE_SCOPE struct dst_entry *ipv4_dst_check(struct dst_entry *dst, 1190 u32 cookie) 1191 { 1192 struct rtable *rt = dst_rtable(dst); 1193 1194 /* All IPV4 dsts are created with ->obsolete set to the value 1195 * DST_OBSOLETE_FORCE_CHK which forces validation calls down 1196 * into this function always. 1197 * 1198 * When a PMTU/redirect information update invalidates a route, 1199 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or 1200 * DST_OBSOLETE_DEAD. 1201 */ 1202 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt)) 1203 return NULL; 1204 return dst; 1205 } 1206 EXPORT_INDIRECT_CALLABLE(ipv4_dst_check); 1207 1208 static void ipv4_send_dest_unreach(struct sk_buff *skb) 1209 { 1210 struct net_device *dev; 1211 struct ip_options opt; 1212 int res; 1213 1214 /* Recompile ip options since IPCB may not be valid anymore. 1215 * Also check we have a reasonable ipv4 header. 1216 */ 1217 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) || 1218 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5) 1219 return; 1220 1221 memset(&opt, 0, sizeof(opt)); 1222 if (ip_hdr(skb)->ihl > 5) { 1223 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4)) 1224 return; 1225 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr); 1226 1227 rcu_read_lock(); 1228 dev = skb->dev ? skb->dev : skb_rtable(skb)->dst.dev; 1229 res = __ip_options_compile(dev_net(dev), &opt, skb, NULL); 1230 rcu_read_unlock(); 1231 1232 if (res) 1233 return; 1234 } 1235 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt); 1236 } 1237 1238 static void ipv4_link_failure(struct sk_buff *skb) 1239 { 1240 struct rtable *rt; 1241 1242 ipv4_send_dest_unreach(skb); 1243 1244 rt = skb_rtable(skb); 1245 if (rt) 1246 dst_set_expires(&rt->dst, 0); 1247 } 1248 1249 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb) 1250 { 1251 pr_debug("%s: %pI4 -> %pI4, %s\n", 1252 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1253 skb->dev ? skb->dev->name : "?"); 1254 kfree_skb(skb); 1255 WARN_ON(1); 1256 return 0; 1257 } 1258 1259 /* 1260 * We do not cache source address of outgoing interface, 1261 * because it is used only by IP RR, TS and SRR options, 1262 * so that it out of fast path. 1263 * 1264 * BTW remember: "addr" is allowed to be not aligned 1265 * in IP options! 1266 */ 1267 1268 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt) 1269 { 1270 __be32 src; 1271 1272 if (rt_is_output_route(rt)) 1273 src = ip_hdr(skb)->saddr; 1274 else { 1275 struct fib_result res; 1276 struct iphdr *iph = ip_hdr(skb); 1277 struct flowi4 fl4 = { 1278 .daddr = iph->daddr, 1279 .saddr = iph->saddr, 1280 .flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(iph)), 1281 .flowi4_oif = rt->dst.dev->ifindex, 1282 .flowi4_iif = skb->dev->ifindex, 1283 .flowi4_mark = skb->mark, 1284 }; 1285 1286 rcu_read_lock(); 1287 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0) 1288 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res); 1289 else 1290 src = inet_select_addr(rt->dst.dev, 1291 rt_nexthop(rt, iph->daddr), 1292 RT_SCOPE_UNIVERSE); 1293 rcu_read_unlock(); 1294 } 1295 memcpy(addr, &src, 4); 1296 } 1297 1298 #ifdef CONFIG_IP_ROUTE_CLASSID 1299 static void set_class_tag(struct rtable *rt, u32 tag) 1300 { 1301 if (!(rt->dst.tclassid & 0xFFFF)) 1302 rt->dst.tclassid |= tag & 0xFFFF; 1303 if (!(rt->dst.tclassid & 0xFFFF0000)) 1304 rt->dst.tclassid |= tag & 0xFFFF0000; 1305 } 1306 #endif 1307 1308 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1309 { 1310 struct net *net = dev_net(dst->dev); 1311 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr); 1312 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size, 1313 net->ipv4.ip_rt_min_advmss); 1314 1315 return min(advmss, IPV4_MAX_PMTU - header_size); 1316 } 1317 1318 INDIRECT_CALLABLE_SCOPE unsigned int ipv4_mtu(const struct dst_entry *dst) 1319 { 1320 return ip_dst_mtu_maybe_forward(dst, false); 1321 } 1322 EXPORT_INDIRECT_CALLABLE(ipv4_mtu); 1323 1324 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr) 1325 { 1326 struct fnhe_hash_bucket *hash; 1327 struct fib_nh_exception *fnhe, __rcu **fnhe_p; 1328 u32 hval = fnhe_hashfun(daddr); 1329 1330 spin_lock_bh(&fnhe_lock); 1331 1332 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1333 lockdep_is_held(&fnhe_lock)); 1334 hash += hval; 1335 1336 fnhe_p = &hash->chain; 1337 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock)); 1338 while (fnhe) { 1339 if (fnhe->fnhe_daddr == daddr) { 1340 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected( 1341 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock))); 1342 /* set fnhe_daddr to 0 to ensure it won't bind with 1343 * new dsts in rt_bind_exception(). 1344 */ 1345 fnhe->fnhe_daddr = 0; 1346 fnhe_flush_routes(fnhe); 1347 kfree_rcu(fnhe, rcu); 1348 break; 1349 } 1350 fnhe_p = &fnhe->fnhe_next; 1351 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1352 lockdep_is_held(&fnhe_lock)); 1353 } 1354 1355 spin_unlock_bh(&fnhe_lock); 1356 } 1357 1358 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc, 1359 __be32 daddr) 1360 { 1361 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions); 1362 struct fib_nh_exception *fnhe; 1363 u32 hval; 1364 1365 if (!hash) 1366 return NULL; 1367 1368 hval = fnhe_hashfun(daddr); 1369 1370 for (fnhe = rcu_dereference(hash[hval].chain); fnhe; 1371 fnhe = rcu_dereference(fnhe->fnhe_next)) { 1372 if (fnhe->fnhe_daddr == daddr) { 1373 if (fnhe->fnhe_expires && 1374 time_after(jiffies, fnhe->fnhe_expires)) { 1375 ip_del_fnhe(nhc, daddr); 1376 break; 1377 } 1378 return fnhe; 1379 } 1380 } 1381 return NULL; 1382 } 1383 1384 /* MTU selection: 1385 * 1. mtu on route is locked - use it 1386 * 2. mtu from nexthop exception 1387 * 3. mtu from egress device 1388 */ 1389 1390 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr) 1391 { 1392 struct fib_nh_common *nhc = res->nhc; 1393 struct net_device *dev = nhc->nhc_dev; 1394 struct fib_info *fi = res->fi; 1395 u32 mtu = 0; 1396 1397 if (READ_ONCE(dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu) || 1398 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU)) 1399 mtu = fi->fib_mtu; 1400 1401 if (likely(!mtu)) { 1402 struct fib_nh_exception *fnhe; 1403 1404 fnhe = find_exception(nhc, daddr); 1405 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires)) 1406 mtu = fnhe->fnhe_pmtu; 1407 } 1408 1409 if (likely(!mtu)) 1410 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU); 1411 1412 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu); 1413 } 1414 1415 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, 1416 __be32 daddr, const bool do_cache) 1417 { 1418 bool ret = false; 1419 1420 spin_lock_bh(&fnhe_lock); 1421 1422 if (daddr == fnhe->fnhe_daddr) { 1423 struct rtable __rcu **porig; 1424 struct rtable *orig; 1425 int genid = fnhe_genid(dev_net(rt->dst.dev)); 1426 1427 if (rt_is_input_route(rt)) 1428 porig = &fnhe->fnhe_rth_input; 1429 else 1430 porig = &fnhe->fnhe_rth_output; 1431 orig = rcu_dereference(*porig); 1432 1433 if (fnhe->fnhe_genid != genid) { 1434 fnhe->fnhe_genid = genid; 1435 fnhe->fnhe_gw = 0; 1436 fnhe->fnhe_pmtu = 0; 1437 fnhe->fnhe_expires = 0; 1438 fnhe->fnhe_mtu_locked = false; 1439 fnhe_flush_routes(fnhe); 1440 orig = NULL; 1441 } 1442 fill_route_from_fnhe(rt, fnhe); 1443 if (!rt->rt_gw4) { 1444 rt->rt_gw4 = daddr; 1445 rt->rt_gw_family = AF_INET; 1446 } 1447 1448 if (do_cache) { 1449 dst_hold(&rt->dst); 1450 rcu_assign_pointer(*porig, rt); 1451 if (orig) { 1452 dst_dev_put(&orig->dst); 1453 dst_release(&orig->dst); 1454 } 1455 ret = true; 1456 } 1457 1458 fnhe->fnhe_stamp = jiffies; 1459 } 1460 spin_unlock_bh(&fnhe_lock); 1461 1462 return ret; 1463 } 1464 1465 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt) 1466 { 1467 struct rtable *orig, *prev, **p; 1468 bool ret = true; 1469 1470 if (rt_is_input_route(rt)) { 1471 p = (struct rtable **)&nhc->nhc_rth_input; 1472 } else { 1473 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 1474 } 1475 orig = *p; 1476 1477 /* hold dst before doing cmpxchg() to avoid race condition 1478 * on this dst 1479 */ 1480 dst_hold(&rt->dst); 1481 prev = cmpxchg(p, orig, rt); 1482 if (prev == orig) { 1483 if (orig) { 1484 rt_add_uncached_list(orig); 1485 dst_release(&orig->dst); 1486 } 1487 } else { 1488 dst_release(&rt->dst); 1489 ret = false; 1490 } 1491 1492 return ret; 1493 } 1494 1495 struct uncached_list { 1496 spinlock_t lock; 1497 struct list_head head; 1498 }; 1499 1500 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list); 1501 1502 void rt_add_uncached_list(struct rtable *rt) 1503 { 1504 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list); 1505 1506 rt->dst.rt_uncached_list = ul; 1507 1508 spin_lock_bh(&ul->lock); 1509 list_add_tail(&rt->dst.rt_uncached, &ul->head); 1510 spin_unlock_bh(&ul->lock); 1511 } 1512 1513 void rt_del_uncached_list(struct rtable *rt) 1514 { 1515 if (!list_empty(&rt->dst.rt_uncached)) { 1516 struct uncached_list *ul = rt->dst.rt_uncached_list; 1517 1518 spin_lock_bh(&ul->lock); 1519 list_del_init(&rt->dst.rt_uncached); 1520 spin_unlock_bh(&ul->lock); 1521 } 1522 } 1523 1524 static void ipv4_dst_destroy(struct dst_entry *dst) 1525 { 1526 ip_dst_metrics_put(dst); 1527 rt_del_uncached_list(dst_rtable(dst)); 1528 } 1529 1530 void rt_flush_dev(struct net_device *dev) 1531 { 1532 struct rtable *rt, *safe; 1533 int cpu; 1534 1535 for_each_possible_cpu(cpu) { 1536 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 1537 1538 if (list_empty(&ul->head)) 1539 continue; 1540 1541 spin_lock_bh(&ul->lock); 1542 list_for_each_entry_safe(rt, safe, &ul->head, dst.rt_uncached) { 1543 if (rt->dst.dev != dev) 1544 continue; 1545 rt->dst.dev = blackhole_netdev; 1546 netdev_ref_replace(dev, blackhole_netdev, 1547 &rt->dst.dev_tracker, GFP_ATOMIC); 1548 list_del_init(&rt->dst.rt_uncached); 1549 } 1550 spin_unlock_bh(&ul->lock); 1551 } 1552 } 1553 1554 static bool rt_cache_valid(const struct rtable *rt) 1555 { 1556 return rt && 1557 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK && 1558 !rt_is_expired(rt); 1559 } 1560 1561 static void rt_set_nexthop(struct rtable *rt, __be32 daddr, 1562 const struct fib_result *res, 1563 struct fib_nh_exception *fnhe, 1564 struct fib_info *fi, u16 type, u32 itag, 1565 const bool do_cache) 1566 { 1567 bool cached = false; 1568 1569 if (fi) { 1570 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1571 1572 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) { 1573 rt->rt_uses_gateway = 1; 1574 rt->rt_gw_family = nhc->nhc_gw_family; 1575 /* only INET and INET6 are supported */ 1576 if (likely(nhc->nhc_gw_family == AF_INET)) 1577 rt->rt_gw4 = nhc->nhc_gw.ipv4; 1578 else 1579 rt->rt_gw6 = nhc->nhc_gw.ipv6; 1580 } 1581 1582 ip_dst_init_metrics(&rt->dst, fi->fib_metrics); 1583 1584 #ifdef CONFIG_IP_ROUTE_CLASSID 1585 if (nhc->nhc_family == AF_INET) { 1586 struct fib_nh *nh; 1587 1588 nh = container_of(nhc, struct fib_nh, nh_common); 1589 rt->dst.tclassid = nh->nh_tclassid; 1590 } 1591 #endif 1592 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 1593 if (unlikely(fnhe)) 1594 cached = rt_bind_exception(rt, fnhe, daddr, do_cache); 1595 else if (do_cache) 1596 cached = rt_cache_route(nhc, rt); 1597 if (unlikely(!cached)) { 1598 /* Routes we intend to cache in nexthop exception or 1599 * FIB nexthop have the DST_NOCACHE bit clear. 1600 * However, if we are unsuccessful at storing this 1601 * route into the cache we really need to set it. 1602 */ 1603 if (!rt->rt_gw4) { 1604 rt->rt_gw_family = AF_INET; 1605 rt->rt_gw4 = daddr; 1606 } 1607 rt_add_uncached_list(rt); 1608 } 1609 } else 1610 rt_add_uncached_list(rt); 1611 1612 #ifdef CONFIG_IP_ROUTE_CLASSID 1613 #ifdef CONFIG_IP_MULTIPLE_TABLES 1614 set_class_tag(rt, res->tclassid); 1615 #endif 1616 set_class_tag(rt, itag); 1617 #endif 1618 } 1619 1620 struct rtable *rt_dst_alloc(struct net_device *dev, 1621 unsigned int flags, u16 type, 1622 bool noxfrm) 1623 { 1624 struct rtable *rt; 1625 1626 rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK, 1627 (noxfrm ? DST_NOXFRM : 0)); 1628 1629 if (rt) { 1630 rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1631 rt->rt_flags = flags; 1632 rt->rt_type = type; 1633 rt->rt_is_input = 0; 1634 rt->rt_iif = 0; 1635 rt->rt_pmtu = 0; 1636 rt->rt_mtu_locked = 0; 1637 rt->rt_uses_gateway = 0; 1638 rt->rt_gw_family = 0; 1639 rt->rt_gw4 = 0; 1640 1641 rt->dst.output = ip_output; 1642 if (flags & RTCF_LOCAL) 1643 rt->dst.input = ip_local_deliver; 1644 } 1645 1646 return rt; 1647 } 1648 EXPORT_SYMBOL(rt_dst_alloc); 1649 1650 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt) 1651 { 1652 struct rtable *new_rt; 1653 1654 new_rt = dst_alloc(&ipv4_dst_ops, dev, DST_OBSOLETE_FORCE_CHK, 1655 rt->dst.flags); 1656 1657 if (new_rt) { 1658 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1659 new_rt->rt_flags = rt->rt_flags; 1660 new_rt->rt_type = rt->rt_type; 1661 new_rt->rt_is_input = rt->rt_is_input; 1662 new_rt->rt_iif = rt->rt_iif; 1663 new_rt->rt_pmtu = rt->rt_pmtu; 1664 new_rt->rt_mtu_locked = rt->rt_mtu_locked; 1665 new_rt->rt_gw_family = rt->rt_gw_family; 1666 if (rt->rt_gw_family == AF_INET) 1667 new_rt->rt_gw4 = rt->rt_gw4; 1668 else if (rt->rt_gw_family == AF_INET6) 1669 new_rt->rt_gw6 = rt->rt_gw6; 1670 1671 new_rt->dst.input = rt->dst.input; 1672 new_rt->dst.output = rt->dst.output; 1673 new_rt->dst.error = rt->dst.error; 1674 new_rt->dst.lastuse = jiffies; 1675 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate); 1676 } 1677 return new_rt; 1678 } 1679 EXPORT_SYMBOL(rt_dst_clone); 1680 1681 /* called in rcu_read_lock() section */ 1682 enum skb_drop_reason 1683 ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1684 dscp_t dscp, struct net_device *dev, 1685 struct in_device *in_dev, u32 *itag) 1686 { 1687 enum skb_drop_reason reason; 1688 1689 /* Primary sanity checks. */ 1690 if (!in_dev) 1691 return SKB_DROP_REASON_NOT_SPECIFIED; 1692 1693 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 1694 return SKB_DROP_REASON_IP_INVALID_SOURCE; 1695 1696 if (skb->protocol != htons(ETH_P_IP)) 1697 return SKB_DROP_REASON_INVALID_PROTO; 1698 1699 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev)) 1700 return SKB_DROP_REASON_IP_LOCALNET; 1701 1702 if (ipv4_is_zeronet(saddr)) { 1703 if (!ipv4_is_local_multicast(daddr) && 1704 ip_hdr(skb)->protocol != IPPROTO_IGMP) 1705 return SKB_DROP_REASON_IP_INVALID_SOURCE; 1706 } else { 1707 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0, 1708 dev, in_dev, itag); 1709 if (reason) 1710 return reason; 1711 } 1712 return SKB_NOT_DROPPED_YET; 1713 } 1714 1715 /* called in rcu_read_lock() section */ 1716 static enum skb_drop_reason 1717 ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1718 dscp_t dscp, struct net_device *dev, int our) 1719 { 1720 struct in_device *in_dev = __in_dev_get_rcu(dev); 1721 unsigned int flags = RTCF_MULTICAST; 1722 enum skb_drop_reason reason; 1723 struct rtable *rth; 1724 u32 itag = 0; 1725 1726 reason = ip_mc_validate_source(skb, daddr, saddr, dscp, dev, in_dev, 1727 &itag); 1728 if (reason) 1729 return reason; 1730 1731 if (our) 1732 flags |= RTCF_LOCAL; 1733 1734 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 1735 IPCB(skb)->flags |= IPSKB_NOPOLICY; 1736 1737 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST, 1738 false); 1739 if (!rth) 1740 return SKB_DROP_REASON_NOMEM; 1741 1742 #ifdef CONFIG_IP_ROUTE_CLASSID 1743 rth->dst.tclassid = itag; 1744 #endif 1745 rth->dst.output = ip_rt_bug; 1746 rth->rt_is_input= 1; 1747 1748 #ifdef CONFIG_IP_MROUTE 1749 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1750 rth->dst.input = ip_mr_input; 1751 #endif 1752 RT_CACHE_STAT_INC(in_slow_mc); 1753 1754 skb_dst_drop(skb); 1755 skb_dst_set(skb, &rth->dst); 1756 return SKB_NOT_DROPPED_YET; 1757 } 1758 1759 1760 static void ip_handle_martian_source(struct net_device *dev, 1761 struct in_device *in_dev, 1762 struct sk_buff *skb, 1763 __be32 daddr, 1764 __be32 saddr) 1765 { 1766 RT_CACHE_STAT_INC(in_martian_src); 1767 #ifdef CONFIG_IP_ROUTE_VERBOSE 1768 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1769 /* 1770 * RFC1812 recommendation, if source is martian, 1771 * the only hint is MAC header. 1772 */ 1773 pr_warn("martian source %pI4 from %pI4, on dev %s\n", 1774 &daddr, &saddr, dev->name); 1775 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1776 print_hex_dump(KERN_WARNING, "ll header: ", 1777 DUMP_PREFIX_OFFSET, 16, 1, 1778 skb_mac_header(skb), 1779 dev->hard_header_len, false); 1780 } 1781 } 1782 #endif 1783 } 1784 1785 /* called in rcu_read_lock() section */ 1786 static enum skb_drop_reason 1787 __mkroute_input(struct sk_buff *skb, const struct fib_result *res, 1788 struct in_device *in_dev, __be32 daddr, 1789 __be32 saddr, dscp_t dscp) 1790 { 1791 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 1792 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1793 struct net_device *dev = nhc->nhc_dev; 1794 struct fib_nh_exception *fnhe; 1795 struct rtable *rth; 1796 int err; 1797 struct in_device *out_dev; 1798 bool do_cache; 1799 u32 itag = 0; 1800 1801 /* get a working reference to the output device */ 1802 out_dev = __in_dev_get_rcu(dev); 1803 if (!out_dev) { 1804 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n"); 1805 return reason; 1806 } 1807 1808 err = fib_validate_source(skb, saddr, daddr, dscp, FIB_RES_OIF(*res), 1809 in_dev->dev, in_dev, &itag); 1810 if (err < 0) { 1811 reason = -err; 1812 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1813 saddr); 1814 1815 goto cleanup; 1816 } 1817 1818 do_cache = res->fi && !itag; 1819 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) && 1820 skb->protocol == htons(ETH_P_IP)) { 1821 __be32 gw; 1822 1823 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0; 1824 if (IN_DEV_SHARED_MEDIA(out_dev) || 1825 inet_addr_onlink(out_dev, saddr, gw)) 1826 IPCB(skb)->flags |= IPSKB_DOREDIRECT; 1827 } 1828 1829 if (skb->protocol != htons(ETH_P_IP)) { 1830 /* Not IP (i.e. ARP). Do not create route, if it is 1831 * invalid for proxy arp. DNAT routes are always valid. 1832 * 1833 * Proxy arp feature have been extended to allow, ARP 1834 * replies back to the same interface, to support 1835 * Private VLAN switch technologies. See arp.c. 1836 */ 1837 if (out_dev == in_dev && 1838 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 1839 reason = SKB_DROP_REASON_ARP_PVLAN_DISABLE; 1840 goto cleanup; 1841 } 1842 } 1843 1844 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 1845 IPCB(skb)->flags |= IPSKB_NOPOLICY; 1846 1847 fnhe = find_exception(nhc, daddr); 1848 if (do_cache) { 1849 if (fnhe) 1850 rth = rcu_dereference(fnhe->fnhe_rth_input); 1851 else 1852 rth = rcu_dereference(nhc->nhc_rth_input); 1853 if (rt_cache_valid(rth)) { 1854 skb_dst_set_noref(skb, &rth->dst); 1855 goto out; 1856 } 1857 } 1858 1859 rth = rt_dst_alloc(out_dev->dev, 0, res->type, 1860 IN_DEV_ORCONF(out_dev, NOXFRM)); 1861 if (!rth) { 1862 reason = SKB_DROP_REASON_NOMEM; 1863 goto cleanup; 1864 } 1865 1866 rth->rt_is_input = 1; 1867 RT_CACHE_STAT_INC(in_slow_tot); 1868 1869 rth->dst.input = ip_forward; 1870 1871 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag, 1872 do_cache); 1873 lwtunnel_set_redirect(&rth->dst); 1874 skb_dst_set(skb, &rth->dst); 1875 out: 1876 reason = SKB_NOT_DROPPED_YET; 1877 cleanup: 1878 return reason; 1879 } 1880 1881 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1882 /* To make ICMP packets follow the right flow, the multipath hash is 1883 * calculated from the inner IP addresses. 1884 */ 1885 static void ip_multipath_l3_keys(const struct sk_buff *skb, 1886 struct flow_keys *hash_keys) 1887 { 1888 const struct iphdr *outer_iph = ip_hdr(skb); 1889 const struct iphdr *key_iph = outer_iph; 1890 const struct iphdr *inner_iph; 1891 const struct icmphdr *icmph; 1892 struct iphdr _inner_iph; 1893 struct icmphdr _icmph; 1894 1895 if (likely(outer_iph->protocol != IPPROTO_ICMP)) 1896 goto out; 1897 1898 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0)) 1899 goto out; 1900 1901 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph), 1902 &_icmph); 1903 if (!icmph) 1904 goto out; 1905 1906 if (!icmp_is_err(icmph->type)) 1907 goto out; 1908 1909 inner_iph = skb_header_pointer(skb, 1910 outer_iph->ihl * 4 + sizeof(_icmph), 1911 sizeof(_inner_iph), &_inner_iph); 1912 if (!inner_iph) 1913 goto out; 1914 1915 key_iph = inner_iph; 1916 out: 1917 hash_keys->addrs.v4addrs.src = key_iph->saddr; 1918 hash_keys->addrs.v4addrs.dst = key_iph->daddr; 1919 } 1920 1921 static u32 fib_multipath_custom_hash_outer(const struct net *net, 1922 const struct sk_buff *skb, 1923 bool *p_has_inner) 1924 { 1925 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 1926 struct flow_keys keys, hash_keys; 1927 1928 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 1929 return 0; 1930 1931 memset(&hash_keys, 0, sizeof(hash_keys)); 1932 skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP); 1933 1934 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1935 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 1936 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 1937 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 1938 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 1939 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 1940 hash_keys.basic.ip_proto = keys.basic.ip_proto; 1941 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) 1942 hash_keys.ports.src = keys.ports.src; 1943 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 1944 hash_keys.ports.dst = keys.ports.dst; 1945 1946 *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION); 1947 return fib_multipath_hash_from_keys(net, &hash_keys); 1948 } 1949 1950 static u32 fib_multipath_custom_hash_inner(const struct net *net, 1951 const struct sk_buff *skb, 1952 bool has_inner) 1953 { 1954 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 1955 struct flow_keys keys, hash_keys; 1956 1957 /* We assume the packet carries an encapsulation, but if none was 1958 * encountered during dissection of the outer flow, then there is no 1959 * point in calling the flow dissector again. 1960 */ 1961 if (!has_inner) 1962 return 0; 1963 1964 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK)) 1965 return 0; 1966 1967 memset(&hash_keys, 0, sizeof(hash_keys)); 1968 skb_flow_dissect_flow_keys(skb, &keys, 0); 1969 1970 if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION)) 1971 return 0; 1972 1973 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 1974 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1975 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 1976 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 1977 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 1978 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 1979 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 1980 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 1981 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP) 1982 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 1983 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP) 1984 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 1985 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL) 1986 hash_keys.tags.flow_label = keys.tags.flow_label; 1987 } 1988 1989 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO) 1990 hash_keys.basic.ip_proto = keys.basic.ip_proto; 1991 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT) 1992 hash_keys.ports.src = keys.ports.src; 1993 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT) 1994 hash_keys.ports.dst = keys.ports.dst; 1995 1996 return fib_multipath_hash_from_keys(net, &hash_keys); 1997 } 1998 1999 static u32 fib_multipath_custom_hash_skb(const struct net *net, 2000 const struct sk_buff *skb) 2001 { 2002 u32 mhash, mhash_inner; 2003 bool has_inner = true; 2004 2005 mhash = fib_multipath_custom_hash_outer(net, skb, &has_inner); 2006 mhash_inner = fib_multipath_custom_hash_inner(net, skb, has_inner); 2007 2008 return jhash_2words(mhash, mhash_inner, 0); 2009 } 2010 2011 static u32 fib_multipath_custom_hash_fl4(const struct net *net, 2012 const struct flowi4 *fl4) 2013 { 2014 u32 hash_fields = READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_fields); 2015 struct flow_keys hash_keys; 2016 2017 if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK)) 2018 return 0; 2019 2020 memset(&hash_keys, 0, sizeof(hash_keys)); 2021 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2022 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP) 2023 hash_keys.addrs.v4addrs.src = fl4->saddr; 2024 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP) 2025 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2026 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO) 2027 hash_keys.basic.ip_proto = fl4->flowi4_proto; 2028 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT) 2029 hash_keys.ports.src = fl4->fl4_sport; 2030 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT) 2031 hash_keys.ports.dst = fl4->fl4_dport; 2032 2033 return fib_multipath_hash_from_keys(net, &hash_keys); 2034 } 2035 2036 /* if skb is set it will be used and fl4 can be NULL */ 2037 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4, 2038 const struct sk_buff *skb, struct flow_keys *flkeys) 2039 { 2040 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0; 2041 struct flow_keys hash_keys; 2042 u32 mhash = 0; 2043 2044 switch (READ_ONCE(net->ipv4.sysctl_fib_multipath_hash_policy)) { 2045 case 0: 2046 memset(&hash_keys, 0, sizeof(hash_keys)); 2047 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2048 if (skb) { 2049 ip_multipath_l3_keys(skb, &hash_keys); 2050 } else { 2051 hash_keys.addrs.v4addrs.src = fl4->saddr; 2052 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2053 } 2054 mhash = fib_multipath_hash_from_keys(net, &hash_keys); 2055 break; 2056 case 1: 2057 /* skb is currently provided only when forwarding */ 2058 if (skb) { 2059 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 2060 struct flow_keys keys; 2061 2062 /* short-circuit if we already have L4 hash present */ 2063 if (skb->l4_hash) 2064 return skb_get_hash_raw(skb) >> 1; 2065 2066 memset(&hash_keys, 0, sizeof(hash_keys)); 2067 2068 if (!flkeys) { 2069 skb_flow_dissect_flow_keys(skb, &keys, flag); 2070 flkeys = &keys; 2071 } 2072 2073 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2074 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src; 2075 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst; 2076 hash_keys.ports.src = flkeys->ports.src; 2077 hash_keys.ports.dst = flkeys->ports.dst; 2078 hash_keys.basic.ip_proto = flkeys->basic.ip_proto; 2079 } else { 2080 memset(&hash_keys, 0, sizeof(hash_keys)); 2081 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2082 hash_keys.addrs.v4addrs.src = fl4->saddr; 2083 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2084 hash_keys.ports.src = fl4->fl4_sport; 2085 hash_keys.ports.dst = fl4->fl4_dport; 2086 hash_keys.basic.ip_proto = fl4->flowi4_proto; 2087 } 2088 mhash = fib_multipath_hash_from_keys(net, &hash_keys); 2089 break; 2090 case 2: 2091 memset(&hash_keys, 0, sizeof(hash_keys)); 2092 /* skb is currently provided only when forwarding */ 2093 if (skb) { 2094 struct flow_keys keys; 2095 2096 skb_flow_dissect_flow_keys(skb, &keys, 0); 2097 /* Inner can be v4 or v6 */ 2098 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 2099 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2100 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 2101 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 2102 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 2103 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 2104 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 2105 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 2106 hash_keys.tags.flow_label = keys.tags.flow_label; 2107 hash_keys.basic.ip_proto = keys.basic.ip_proto; 2108 } else { 2109 /* Same as case 0 */ 2110 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2111 ip_multipath_l3_keys(skb, &hash_keys); 2112 } 2113 } else { 2114 /* Same as case 0 */ 2115 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 2116 hash_keys.addrs.v4addrs.src = fl4->saddr; 2117 hash_keys.addrs.v4addrs.dst = fl4->daddr; 2118 } 2119 mhash = fib_multipath_hash_from_keys(net, &hash_keys); 2120 break; 2121 case 3: 2122 if (skb) 2123 mhash = fib_multipath_custom_hash_skb(net, skb); 2124 else 2125 mhash = fib_multipath_custom_hash_fl4(net, fl4); 2126 break; 2127 } 2128 2129 if (multipath_hash) 2130 mhash = jhash_2words(mhash, multipath_hash, 0); 2131 2132 return mhash >> 1; 2133 } 2134 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 2135 2136 static enum skb_drop_reason 2137 ip_mkroute_input(struct sk_buff *skb, struct fib_result *res, 2138 struct in_device *in_dev, __be32 daddr, 2139 __be32 saddr, dscp_t dscp, struct flow_keys *hkeys) 2140 { 2141 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2142 if (res->fi && fib_info_num_path(res->fi) > 1) { 2143 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys); 2144 2145 fib_select_multipath(res, h); 2146 IPCB(skb)->flags |= IPSKB_MULTIPATH; 2147 } 2148 #endif 2149 2150 /* create a routing cache entry */ 2151 return __mkroute_input(skb, res, in_dev, daddr, saddr, dscp); 2152 } 2153 2154 /* Implements all the saddr-related checks as ip_route_input_slow(), 2155 * assuming daddr is valid and the destination is not a local broadcast one. 2156 * Uses the provided hint instead of performing a route lookup. 2157 */ 2158 enum skb_drop_reason 2159 ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2160 dscp_t dscp, struct net_device *dev, 2161 const struct sk_buff *hint) 2162 { 2163 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 2164 struct in_device *in_dev = __in_dev_get_rcu(dev); 2165 struct rtable *rt = skb_rtable(hint); 2166 struct net *net = dev_net(dev); 2167 u32 tag = 0; 2168 2169 if (!in_dev) 2170 return reason; 2171 2172 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) { 2173 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2174 goto martian_source; 2175 } 2176 2177 if (ipv4_is_zeronet(saddr)) { 2178 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2179 goto martian_source; 2180 } 2181 2182 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) { 2183 reason = SKB_DROP_REASON_IP_LOCALNET; 2184 goto martian_source; 2185 } 2186 2187 if (rt->rt_type != RTN_LOCAL) 2188 goto skip_validate_source; 2189 2190 reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 0, dev, 2191 in_dev, &tag); 2192 if (reason) 2193 goto martian_source; 2194 2195 skip_validate_source: 2196 skb_dst_copy(skb, hint); 2197 return SKB_NOT_DROPPED_YET; 2198 2199 martian_source: 2200 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2201 return reason; 2202 } 2203 2204 /* get device for dst_alloc with local routes */ 2205 static struct net_device *ip_rt_get_dev(struct net *net, 2206 const struct fib_result *res) 2207 { 2208 struct fib_nh_common *nhc = res->fi ? res->nhc : NULL; 2209 struct net_device *dev = NULL; 2210 2211 if (nhc) 2212 dev = l3mdev_master_dev_rcu(nhc->nhc_dev); 2213 2214 return dev ? : net->loopback_dev; 2215 } 2216 2217 /* 2218 * NOTE. We drop all the packets that has local source 2219 * addresses, because every properly looped back packet 2220 * must have correct destination already attached by output routine. 2221 * Changes in the enforced policies must be applied also to 2222 * ip_route_use_hint(). 2223 * 2224 * Such approach solves two big problems: 2225 * 1. Not simplex devices are handled properly. 2226 * 2. IP spoofing attempts are filtered with 100% of guarantee. 2227 * called with rcu_read_lock() 2228 */ 2229 2230 static enum skb_drop_reason 2231 ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2232 dscp_t dscp, struct net_device *dev, 2233 struct fib_result *res) 2234 { 2235 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 2236 struct in_device *in_dev = __in_dev_get_rcu(dev); 2237 struct flow_keys *flkeys = NULL, _flkeys; 2238 struct net *net = dev_net(dev); 2239 struct ip_tunnel_info *tun_info; 2240 int err = -EINVAL; 2241 unsigned int flags = 0; 2242 u32 itag = 0; 2243 struct rtable *rth; 2244 struct flowi4 fl4; 2245 bool do_cache = true; 2246 2247 /* IP on this device is disabled. */ 2248 2249 if (!in_dev) 2250 goto out; 2251 2252 /* Check for the most weird martians, which can be not detected 2253 * by fib_lookup. 2254 */ 2255 2256 tun_info = skb_tunnel_info(skb); 2257 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX)) 2258 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id; 2259 else 2260 fl4.flowi4_tun_key.tun_id = 0; 2261 skb_dst_drop(skb); 2262 2263 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) { 2264 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2265 goto martian_source; 2266 } 2267 2268 res->fi = NULL; 2269 res->table = NULL; 2270 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 2271 goto brd_input; 2272 2273 /* Accept zero addresses only to limited broadcast; 2274 * I even do not know to fix it or not. Waiting for complains :-) 2275 */ 2276 if (ipv4_is_zeronet(saddr)) { 2277 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 2278 goto martian_source; 2279 } 2280 2281 if (ipv4_is_zeronet(daddr)) { 2282 reason = SKB_DROP_REASON_IP_INVALID_DEST; 2283 goto martian_destination; 2284 } 2285 2286 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(), 2287 * and call it once if daddr or/and saddr are loopback addresses 2288 */ 2289 if (ipv4_is_loopback(daddr)) { 2290 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) { 2291 reason = SKB_DROP_REASON_IP_LOCALNET; 2292 goto martian_destination; 2293 } 2294 } else if (ipv4_is_loopback(saddr)) { 2295 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) { 2296 reason = SKB_DROP_REASON_IP_LOCALNET; 2297 goto martian_source; 2298 } 2299 } 2300 2301 /* 2302 * Now we are ready to route packet. 2303 */ 2304 fl4.flowi4_l3mdev = 0; 2305 fl4.flowi4_oif = 0; 2306 fl4.flowi4_iif = dev->ifindex; 2307 fl4.flowi4_mark = skb->mark; 2308 fl4.flowi4_tos = inet_dscp_to_dsfield(dscp); 2309 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 2310 fl4.flowi4_flags = 0; 2311 fl4.daddr = daddr; 2312 fl4.saddr = saddr; 2313 fl4.flowi4_uid = sock_net_uid(net, NULL); 2314 fl4.flowi4_multipath_hash = 0; 2315 2316 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) { 2317 flkeys = &_flkeys; 2318 } else { 2319 fl4.flowi4_proto = 0; 2320 fl4.fl4_sport = 0; 2321 fl4.fl4_dport = 0; 2322 } 2323 2324 err = fib_lookup(net, &fl4, res, 0); 2325 if (err != 0) { 2326 if (!IN_DEV_FORWARD(in_dev)) 2327 err = -EHOSTUNREACH; 2328 goto no_route; 2329 } 2330 2331 if (res->type == RTN_BROADCAST) { 2332 if (IN_DEV_BFORWARD(in_dev)) 2333 goto make_route; 2334 /* not do cache if bc_forwarding is enabled */ 2335 if (IPV4_DEVCONF_ALL_RO(net, BC_FORWARDING)) 2336 do_cache = false; 2337 goto brd_input; 2338 } 2339 2340 err = -EINVAL; 2341 if (res->type == RTN_LOCAL) { 2342 reason = fib_validate_source_reason(skb, saddr, daddr, dscp, 2343 0, dev, in_dev, &itag); 2344 if (reason) 2345 goto martian_source; 2346 goto local_input; 2347 } 2348 2349 if (!IN_DEV_FORWARD(in_dev)) { 2350 err = -EHOSTUNREACH; 2351 goto no_route; 2352 } 2353 if (res->type != RTN_UNICAST) { 2354 reason = SKB_DROP_REASON_IP_INVALID_DEST; 2355 goto martian_destination; 2356 } 2357 2358 make_route: 2359 reason = ip_mkroute_input(skb, res, in_dev, daddr, saddr, dscp, 2360 flkeys); 2361 2362 out: 2363 return reason; 2364 2365 brd_input: 2366 if (skb->protocol != htons(ETH_P_IP)) { 2367 reason = SKB_DROP_REASON_INVALID_PROTO; 2368 goto out; 2369 } 2370 2371 if (!ipv4_is_zeronet(saddr)) { 2372 reason = fib_validate_source_reason(skb, saddr, 0, dscp, 0, 2373 dev, in_dev, &itag); 2374 if (reason) 2375 goto martian_source; 2376 } 2377 flags |= RTCF_BROADCAST; 2378 res->type = RTN_BROADCAST; 2379 RT_CACHE_STAT_INC(in_brd); 2380 2381 local_input: 2382 if (IN_DEV_ORCONF(in_dev, NOPOLICY)) 2383 IPCB(skb)->flags |= IPSKB_NOPOLICY; 2384 2385 do_cache &= res->fi && !itag; 2386 if (do_cache) { 2387 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2388 2389 rth = rcu_dereference(nhc->nhc_rth_input); 2390 if (rt_cache_valid(rth)) { 2391 skb_dst_set_noref(skb, &rth->dst); 2392 reason = SKB_NOT_DROPPED_YET; 2393 goto out; 2394 } 2395 } 2396 2397 rth = rt_dst_alloc(ip_rt_get_dev(net, res), 2398 flags | RTCF_LOCAL, res->type, false); 2399 if (!rth) 2400 goto e_nobufs; 2401 2402 rth->dst.output= ip_rt_bug; 2403 #ifdef CONFIG_IP_ROUTE_CLASSID 2404 rth->dst.tclassid = itag; 2405 #endif 2406 rth->rt_is_input = 1; 2407 2408 RT_CACHE_STAT_INC(in_slow_tot); 2409 if (res->type == RTN_UNREACHABLE) { 2410 rth->dst.input= ip_error; 2411 rth->dst.error= -err; 2412 rth->rt_flags &= ~RTCF_LOCAL; 2413 } 2414 2415 if (do_cache) { 2416 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2417 2418 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 2419 if (lwtunnel_input_redirect(rth->dst.lwtstate)) { 2420 WARN_ON(rth->dst.input == lwtunnel_input); 2421 rth->dst.lwtstate->orig_input = rth->dst.input; 2422 rth->dst.input = lwtunnel_input; 2423 } 2424 2425 if (unlikely(!rt_cache_route(nhc, rth))) 2426 rt_add_uncached_list(rth); 2427 } 2428 skb_dst_set(skb, &rth->dst); 2429 reason = SKB_NOT_DROPPED_YET; 2430 goto out; 2431 2432 no_route: 2433 RT_CACHE_STAT_INC(in_no_route); 2434 res->type = RTN_UNREACHABLE; 2435 res->fi = NULL; 2436 res->table = NULL; 2437 goto local_input; 2438 2439 /* 2440 * Do not cache martian addresses: they should be logged (RFC1812) 2441 */ 2442 martian_destination: 2443 RT_CACHE_STAT_INC(in_martian_dst); 2444 #ifdef CONFIG_IP_ROUTE_VERBOSE 2445 if (IN_DEV_LOG_MARTIANS(in_dev)) 2446 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n", 2447 &daddr, &saddr, dev->name); 2448 #endif 2449 goto out; 2450 2451 e_nobufs: 2452 reason = SKB_DROP_REASON_NOMEM; 2453 goto out; 2454 2455 martian_source: 2456 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2457 goto out; 2458 } 2459 2460 /* called with rcu_read_lock held */ 2461 static enum skb_drop_reason 2462 ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2463 dscp_t dscp, struct net_device *dev, 2464 struct fib_result *res) 2465 { 2466 /* Multicast recognition logic is moved from route cache to here. 2467 * The problem was that too many Ethernet cards have broken/missing 2468 * hardware multicast filters :-( As result the host on multicasting 2469 * network acquires a lot of useless route cache entries, sort of 2470 * SDR messages from all the world. Now we try to get rid of them. 2471 * Really, provided software IP multicast filter is organized 2472 * reasonably (at least, hashed), it does not result in a slowdown 2473 * comparing with route cache reject entries. 2474 * Note, that multicast routers are not affected, because 2475 * route cache entry is created eventually. 2476 */ 2477 if (ipv4_is_multicast(daddr)) { 2478 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 2479 struct in_device *in_dev = __in_dev_get_rcu(dev); 2480 int our = 0; 2481 2482 if (!in_dev) 2483 return reason; 2484 2485 our = ip_check_mc_rcu(in_dev, daddr, saddr, 2486 ip_hdr(skb)->protocol); 2487 2488 /* check l3 master if no match yet */ 2489 if (!our && netif_is_l3_slave(dev)) { 2490 struct in_device *l3_in_dev; 2491 2492 l3_in_dev = __in_dev_get_rcu(skb->dev); 2493 if (l3_in_dev) 2494 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr, 2495 ip_hdr(skb)->protocol); 2496 } 2497 2498 if (our 2499 #ifdef CONFIG_IP_MROUTE 2500 || 2501 (!ipv4_is_local_multicast(daddr) && 2502 IN_DEV_MFORWARD(in_dev)) 2503 #endif 2504 ) { 2505 reason = ip_route_input_mc(skb, daddr, saddr, dscp, 2506 dev, our); 2507 } 2508 return reason; 2509 } 2510 2511 return ip_route_input_slow(skb, daddr, saddr, dscp, dev, res); 2512 } 2513 2514 enum skb_drop_reason ip_route_input_noref(struct sk_buff *skb, __be32 daddr, 2515 __be32 saddr, dscp_t dscp, 2516 struct net_device *dev) 2517 { 2518 enum skb_drop_reason reason; 2519 struct fib_result res; 2520 2521 rcu_read_lock(); 2522 reason = ip_route_input_rcu(skb, daddr, saddr, dscp, dev, &res); 2523 rcu_read_unlock(); 2524 2525 return reason; 2526 } 2527 EXPORT_SYMBOL(ip_route_input_noref); 2528 2529 /* called with rcu_read_lock() */ 2530 static struct rtable *__mkroute_output(const struct fib_result *res, 2531 const struct flowi4 *fl4, int orig_oif, 2532 struct net_device *dev_out, 2533 unsigned int flags) 2534 { 2535 struct fib_info *fi = res->fi; 2536 struct fib_nh_exception *fnhe; 2537 struct in_device *in_dev; 2538 u16 type = res->type; 2539 struct rtable *rth; 2540 bool do_cache; 2541 2542 in_dev = __in_dev_get_rcu(dev_out); 2543 if (!in_dev) 2544 return ERR_PTR(-EINVAL); 2545 2546 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 2547 if (ipv4_is_loopback(fl4->saddr) && 2548 !(dev_out->flags & IFF_LOOPBACK) && 2549 !netif_is_l3_master(dev_out)) 2550 return ERR_PTR(-EINVAL); 2551 2552 if (ipv4_is_lbcast(fl4->daddr)) 2553 type = RTN_BROADCAST; 2554 else if (ipv4_is_multicast(fl4->daddr)) 2555 type = RTN_MULTICAST; 2556 else if (ipv4_is_zeronet(fl4->daddr)) 2557 return ERR_PTR(-EINVAL); 2558 2559 if (dev_out->flags & IFF_LOOPBACK) 2560 flags |= RTCF_LOCAL; 2561 2562 do_cache = true; 2563 if (type == RTN_BROADCAST) { 2564 flags |= RTCF_BROADCAST | RTCF_LOCAL; 2565 fi = NULL; 2566 } else if (type == RTN_MULTICAST) { 2567 flags |= RTCF_MULTICAST | RTCF_LOCAL; 2568 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr, 2569 fl4->flowi4_proto)) 2570 flags &= ~RTCF_LOCAL; 2571 else 2572 do_cache = false; 2573 /* If multicast route do not exist use 2574 * default one, but do not gateway in this case. 2575 * Yes, it is hack. 2576 */ 2577 if (fi && res->prefixlen < 4) 2578 fi = NULL; 2579 } else if ((type == RTN_LOCAL) && (orig_oif != 0) && 2580 (orig_oif != dev_out->ifindex)) { 2581 /* For local routes that require a particular output interface 2582 * we do not want to cache the result. Caching the result 2583 * causes incorrect behaviour when there are multiple source 2584 * addresses on the interface, the end result being that if the 2585 * intended recipient is waiting on that interface for the 2586 * packet he won't receive it because it will be delivered on 2587 * the loopback interface and the IP_PKTINFO ipi_ifindex will 2588 * be set to the loopback interface as well. 2589 */ 2590 do_cache = false; 2591 } 2592 2593 fnhe = NULL; 2594 do_cache &= fi != NULL; 2595 if (fi) { 2596 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2597 struct rtable __rcu **prth; 2598 2599 fnhe = find_exception(nhc, fl4->daddr); 2600 if (!do_cache) 2601 goto add; 2602 if (fnhe) { 2603 prth = &fnhe->fnhe_rth_output; 2604 } else { 2605 if (unlikely(fl4->flowi4_flags & 2606 FLOWI_FLAG_KNOWN_NH && 2607 !(nhc->nhc_gw_family && 2608 nhc->nhc_scope == RT_SCOPE_LINK))) { 2609 do_cache = false; 2610 goto add; 2611 } 2612 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 2613 } 2614 rth = rcu_dereference(*prth); 2615 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst)) 2616 return rth; 2617 } 2618 2619 add: 2620 rth = rt_dst_alloc(dev_out, flags, type, 2621 IN_DEV_ORCONF(in_dev, NOXFRM)); 2622 if (!rth) 2623 return ERR_PTR(-ENOBUFS); 2624 2625 rth->rt_iif = orig_oif; 2626 2627 RT_CACHE_STAT_INC(out_slow_tot); 2628 2629 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 2630 if (flags & RTCF_LOCAL && 2631 !(dev_out->flags & IFF_LOOPBACK)) { 2632 rth->dst.output = ip_mc_output; 2633 RT_CACHE_STAT_INC(out_slow_mc); 2634 } 2635 #ifdef CONFIG_IP_MROUTE 2636 if (type == RTN_MULTICAST) { 2637 if (IN_DEV_MFORWARD(in_dev) && 2638 !ipv4_is_local_multicast(fl4->daddr)) { 2639 rth->dst.input = ip_mr_input; 2640 rth->dst.output = ip_mc_output; 2641 } 2642 } 2643 #endif 2644 } 2645 2646 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache); 2647 lwtunnel_set_redirect(&rth->dst); 2648 2649 return rth; 2650 } 2651 2652 /* 2653 * Major route resolver routine. 2654 */ 2655 2656 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4, 2657 const struct sk_buff *skb) 2658 { 2659 struct fib_result res = { 2660 .type = RTN_UNSPEC, 2661 .fi = NULL, 2662 .table = NULL, 2663 .tclassid = 0, 2664 }; 2665 struct rtable *rth; 2666 2667 fl4->flowi4_iif = LOOPBACK_IFINDEX; 2668 fl4->flowi4_tos &= INET_DSCP_MASK; 2669 2670 rcu_read_lock(); 2671 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb); 2672 rcu_read_unlock(); 2673 2674 return rth; 2675 } 2676 EXPORT_SYMBOL_GPL(ip_route_output_key_hash); 2677 2678 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4, 2679 struct fib_result *res, 2680 const struct sk_buff *skb) 2681 { 2682 struct net_device *dev_out = NULL; 2683 int orig_oif = fl4->flowi4_oif; 2684 unsigned int flags = 0; 2685 struct rtable *rth; 2686 int err; 2687 2688 if (fl4->saddr) { 2689 if (ipv4_is_multicast(fl4->saddr) || 2690 ipv4_is_lbcast(fl4->saddr) || 2691 ipv4_is_zeronet(fl4->saddr)) { 2692 rth = ERR_PTR(-EINVAL); 2693 goto out; 2694 } 2695 2696 rth = ERR_PTR(-ENETUNREACH); 2697 2698 /* I removed check for oif == dev_out->oif here. 2699 * It was wrong for two reasons: 2700 * 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 2701 * is assigned to multiple interfaces. 2702 * 2. Moreover, we are allowed to send packets with saddr 2703 * of another iface. --ANK 2704 */ 2705 2706 if (fl4->flowi4_oif == 0 && 2707 (ipv4_is_multicast(fl4->daddr) || 2708 ipv4_is_lbcast(fl4->daddr))) { 2709 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2710 dev_out = __ip_dev_find(net, fl4->saddr, false); 2711 if (!dev_out) 2712 goto out; 2713 2714 /* Special hack: user can direct multicasts 2715 * and limited broadcast via necessary interface 2716 * without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 2717 * This hack is not just for fun, it allows 2718 * vic,vat and friends to work. 2719 * They bind socket to loopback, set ttl to zero 2720 * and expect that it will work. 2721 * From the viewpoint of routing cache they are broken, 2722 * because we are not allowed to build multicast path 2723 * with loopback source addr (look, routing cache 2724 * cannot know, that ttl is zero, so that packet 2725 * will not leave this host and route is valid). 2726 * Luckily, this hack is good workaround. 2727 */ 2728 2729 fl4->flowi4_oif = dev_out->ifindex; 2730 goto make_route; 2731 } 2732 2733 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 2734 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2735 if (!__ip_dev_find(net, fl4->saddr, false)) 2736 goto out; 2737 } 2738 } 2739 2740 2741 if (fl4->flowi4_oif) { 2742 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif); 2743 rth = ERR_PTR(-ENODEV); 2744 if (!dev_out) 2745 goto out; 2746 2747 /* RACE: Check return value of inet_select_addr instead. */ 2748 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 2749 rth = ERR_PTR(-ENETUNREACH); 2750 goto out; 2751 } 2752 if (ipv4_is_local_multicast(fl4->daddr) || 2753 ipv4_is_lbcast(fl4->daddr) || 2754 fl4->flowi4_proto == IPPROTO_IGMP) { 2755 if (!fl4->saddr) 2756 fl4->saddr = inet_select_addr(dev_out, 0, 2757 RT_SCOPE_LINK); 2758 goto make_route; 2759 } 2760 if (!fl4->saddr) { 2761 if (ipv4_is_multicast(fl4->daddr)) 2762 fl4->saddr = inet_select_addr(dev_out, 0, 2763 fl4->flowi4_scope); 2764 else if (!fl4->daddr) 2765 fl4->saddr = inet_select_addr(dev_out, 0, 2766 RT_SCOPE_HOST); 2767 } 2768 } 2769 2770 if (!fl4->daddr) { 2771 fl4->daddr = fl4->saddr; 2772 if (!fl4->daddr) 2773 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK); 2774 dev_out = net->loopback_dev; 2775 fl4->flowi4_oif = LOOPBACK_IFINDEX; 2776 res->type = RTN_LOCAL; 2777 flags |= RTCF_LOCAL; 2778 goto make_route; 2779 } 2780 2781 err = fib_lookup(net, fl4, res, 0); 2782 if (err) { 2783 res->fi = NULL; 2784 res->table = NULL; 2785 if (fl4->flowi4_oif && 2786 (ipv4_is_multicast(fl4->daddr) || !fl4->flowi4_l3mdev)) { 2787 /* Apparently, routing tables are wrong. Assume, 2788 * that the destination is on link. 2789 * 2790 * WHY? DW. 2791 * Because we are allowed to send to iface 2792 * even if it has NO routes and NO assigned 2793 * addresses. When oif is specified, routing 2794 * tables are looked up with only one purpose: 2795 * to catch if destination is gatewayed, rather than 2796 * direct. Moreover, if MSG_DONTROUTE is set, 2797 * we send packet, ignoring both routing tables 2798 * and ifaddr state. --ANK 2799 * 2800 * 2801 * We could make it even if oif is unknown, 2802 * likely IPv6, but we do not. 2803 */ 2804 2805 if (fl4->saddr == 0) 2806 fl4->saddr = inet_select_addr(dev_out, 0, 2807 RT_SCOPE_LINK); 2808 res->type = RTN_UNICAST; 2809 goto make_route; 2810 } 2811 rth = ERR_PTR(err); 2812 goto out; 2813 } 2814 2815 if (res->type == RTN_LOCAL) { 2816 if (!fl4->saddr) { 2817 if (res->fi->fib_prefsrc) 2818 fl4->saddr = res->fi->fib_prefsrc; 2819 else 2820 fl4->saddr = fl4->daddr; 2821 } 2822 2823 /* L3 master device is the loopback for that domain */ 2824 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? : 2825 net->loopback_dev; 2826 2827 /* make sure orig_oif points to fib result device even 2828 * though packet rx/tx happens over loopback or l3mdev 2829 */ 2830 orig_oif = FIB_RES_OIF(*res); 2831 2832 fl4->flowi4_oif = dev_out->ifindex; 2833 flags |= RTCF_LOCAL; 2834 goto make_route; 2835 } 2836 2837 fib_select_path(net, res, fl4, skb); 2838 2839 dev_out = FIB_RES_DEV(*res); 2840 2841 make_route: 2842 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags); 2843 2844 out: 2845 return rth; 2846 } 2847 2848 static struct dst_ops ipv4_dst_blackhole_ops = { 2849 .family = AF_INET, 2850 .default_advmss = ipv4_default_advmss, 2851 .neigh_lookup = ipv4_neigh_lookup, 2852 .check = dst_blackhole_check, 2853 .cow_metrics = dst_blackhole_cow_metrics, 2854 .update_pmtu = dst_blackhole_update_pmtu, 2855 .redirect = dst_blackhole_redirect, 2856 .mtu = dst_blackhole_mtu, 2857 }; 2858 2859 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2860 { 2861 struct rtable *ort = dst_rtable(dst_orig); 2862 struct rtable *rt; 2863 2864 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, DST_OBSOLETE_DEAD, 0); 2865 if (rt) { 2866 struct dst_entry *new = &rt->dst; 2867 2868 new->__use = 1; 2869 new->input = dst_discard; 2870 new->output = dst_discard_out; 2871 2872 new->dev = net->loopback_dev; 2873 netdev_hold(new->dev, &new->dev_tracker, GFP_ATOMIC); 2874 2875 rt->rt_is_input = ort->rt_is_input; 2876 rt->rt_iif = ort->rt_iif; 2877 rt->rt_pmtu = ort->rt_pmtu; 2878 rt->rt_mtu_locked = ort->rt_mtu_locked; 2879 2880 rt->rt_genid = rt_genid_ipv4(net); 2881 rt->rt_flags = ort->rt_flags; 2882 rt->rt_type = ort->rt_type; 2883 rt->rt_uses_gateway = ort->rt_uses_gateway; 2884 rt->rt_gw_family = ort->rt_gw_family; 2885 if (rt->rt_gw_family == AF_INET) 2886 rt->rt_gw4 = ort->rt_gw4; 2887 else if (rt->rt_gw_family == AF_INET6) 2888 rt->rt_gw6 = ort->rt_gw6; 2889 } 2890 2891 dst_release(dst_orig); 2892 2893 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2894 } 2895 2896 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2897 const struct sock *sk) 2898 { 2899 struct rtable *rt = __ip_route_output_key(net, flp4); 2900 2901 if (IS_ERR(rt)) 2902 return rt; 2903 2904 if (flp4->flowi4_proto) { 2905 flp4->flowi4_oif = rt->dst.dev->ifindex; 2906 rt = dst_rtable(xfrm_lookup_route(net, &rt->dst, 2907 flowi4_to_flowi(flp4), 2908 sk, 0)); 2909 } 2910 2911 return rt; 2912 } 2913 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2914 2915 /* called with rcu_read_lock held */ 2916 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, 2917 struct rtable *rt, u32 table_id, dscp_t dscp, 2918 struct flowi4 *fl4, struct sk_buff *skb, u32 portid, 2919 u32 seq, unsigned int flags) 2920 { 2921 struct rtmsg *r; 2922 struct nlmsghdr *nlh; 2923 unsigned long expires = 0; 2924 u32 error; 2925 u32 metrics[RTAX_MAX]; 2926 2927 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags); 2928 if (!nlh) 2929 return -EMSGSIZE; 2930 2931 r = nlmsg_data(nlh); 2932 r->rtm_family = AF_INET; 2933 r->rtm_dst_len = 32; 2934 r->rtm_src_len = 0; 2935 r->rtm_tos = inet_dscp_to_dsfield(dscp); 2936 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT; 2937 if (nla_put_u32(skb, RTA_TABLE, table_id)) 2938 goto nla_put_failure; 2939 r->rtm_type = rt->rt_type; 2940 r->rtm_scope = RT_SCOPE_UNIVERSE; 2941 r->rtm_protocol = RTPROT_UNSPEC; 2942 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 2943 if (rt->rt_flags & RTCF_NOTIFY) 2944 r->rtm_flags |= RTM_F_NOTIFY; 2945 if (IPCB(skb)->flags & IPSKB_DOREDIRECT) 2946 r->rtm_flags |= RTCF_DOREDIRECT; 2947 2948 if (nla_put_in_addr(skb, RTA_DST, dst)) 2949 goto nla_put_failure; 2950 if (src) { 2951 r->rtm_src_len = 32; 2952 if (nla_put_in_addr(skb, RTA_SRC, src)) 2953 goto nla_put_failure; 2954 } 2955 if (rt->dst.dev && 2956 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex)) 2957 goto nla_put_failure; 2958 if (rt->dst.lwtstate && 2959 lwtunnel_fill_encap(skb, rt->dst.lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0) 2960 goto nla_put_failure; 2961 #ifdef CONFIG_IP_ROUTE_CLASSID 2962 if (rt->dst.tclassid && 2963 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid)) 2964 goto nla_put_failure; 2965 #endif 2966 if (fl4 && !rt_is_input_route(rt) && 2967 fl4->saddr != src) { 2968 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr)) 2969 goto nla_put_failure; 2970 } 2971 if (rt->rt_uses_gateway) { 2972 if (rt->rt_gw_family == AF_INET && 2973 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) { 2974 goto nla_put_failure; 2975 } else if (rt->rt_gw_family == AF_INET6) { 2976 int alen = sizeof(struct in6_addr); 2977 struct nlattr *nla; 2978 struct rtvia *via; 2979 2980 nla = nla_reserve(skb, RTA_VIA, alen + 2); 2981 if (!nla) 2982 goto nla_put_failure; 2983 2984 via = nla_data(nla); 2985 via->rtvia_family = AF_INET6; 2986 memcpy(via->rtvia_addr, &rt->rt_gw6, alen); 2987 } 2988 } 2989 2990 expires = rt->dst.expires; 2991 if (expires) { 2992 unsigned long now = jiffies; 2993 2994 if (time_before(now, expires)) 2995 expires -= now; 2996 else 2997 expires = 0; 2998 } 2999 3000 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics)); 3001 if (rt->rt_pmtu && expires) 3002 metrics[RTAX_MTU - 1] = rt->rt_pmtu; 3003 if (rt->rt_mtu_locked && expires) 3004 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU); 3005 if (rtnetlink_put_metrics(skb, metrics) < 0) 3006 goto nla_put_failure; 3007 3008 if (fl4) { 3009 if (fl4->flowi4_mark && 3010 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark)) 3011 goto nla_put_failure; 3012 3013 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) && 3014 nla_put_u32(skb, RTA_UID, 3015 from_kuid_munged(current_user_ns(), 3016 fl4->flowi4_uid))) 3017 goto nla_put_failure; 3018 3019 if (rt_is_input_route(rt)) { 3020 #ifdef CONFIG_IP_MROUTE 3021 if (ipv4_is_multicast(dst) && 3022 !ipv4_is_local_multicast(dst) && 3023 IPV4_DEVCONF_ALL_RO(net, MC_FORWARDING)) { 3024 int err = ipmr_get_route(net, skb, 3025 fl4->saddr, fl4->daddr, 3026 r, portid); 3027 3028 if (err <= 0) { 3029 if (err == 0) 3030 return 0; 3031 goto nla_put_failure; 3032 } 3033 } else 3034 #endif 3035 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif)) 3036 goto nla_put_failure; 3037 } 3038 } 3039 3040 error = rt->dst.error; 3041 3042 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0) 3043 goto nla_put_failure; 3044 3045 nlmsg_end(skb, nlh); 3046 return 0; 3047 3048 nla_put_failure: 3049 nlmsg_cancel(skb, nlh); 3050 return -EMSGSIZE; 3051 } 3052 3053 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb, 3054 struct netlink_callback *cb, u32 table_id, 3055 struct fnhe_hash_bucket *bucket, int genid, 3056 int *fa_index, int fa_start, unsigned int flags) 3057 { 3058 int i; 3059 3060 for (i = 0; i < FNHE_HASH_SIZE; i++) { 3061 struct fib_nh_exception *fnhe; 3062 3063 for (fnhe = rcu_dereference(bucket[i].chain); fnhe; 3064 fnhe = rcu_dereference(fnhe->fnhe_next)) { 3065 struct rtable *rt; 3066 int err; 3067 3068 if (*fa_index < fa_start) 3069 goto next; 3070 3071 if (fnhe->fnhe_genid != genid) 3072 goto next; 3073 3074 if (fnhe->fnhe_expires && 3075 time_after(jiffies, fnhe->fnhe_expires)) 3076 goto next; 3077 3078 rt = rcu_dereference(fnhe->fnhe_rth_input); 3079 if (!rt) 3080 rt = rcu_dereference(fnhe->fnhe_rth_output); 3081 if (!rt) 3082 goto next; 3083 3084 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt, 3085 table_id, 0, NULL, skb, 3086 NETLINK_CB(cb->skb).portid, 3087 cb->nlh->nlmsg_seq, flags); 3088 if (err) 3089 return err; 3090 next: 3091 (*fa_index)++; 3092 } 3093 } 3094 3095 return 0; 3096 } 3097 3098 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb, 3099 u32 table_id, struct fib_info *fi, 3100 int *fa_index, int fa_start, unsigned int flags) 3101 { 3102 struct net *net = sock_net(cb->skb->sk); 3103 int nhsel, genid = fnhe_genid(net); 3104 3105 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) { 3106 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel); 3107 struct fnhe_hash_bucket *bucket; 3108 int err; 3109 3110 if (nhc->nhc_flags & RTNH_F_DEAD) 3111 continue; 3112 3113 rcu_read_lock(); 3114 bucket = rcu_dereference(nhc->nhc_exceptions); 3115 err = 0; 3116 if (bucket) 3117 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket, 3118 genid, fa_index, fa_start, 3119 flags); 3120 rcu_read_unlock(); 3121 if (err) 3122 return err; 3123 } 3124 3125 return 0; 3126 } 3127 3128 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst, 3129 u8 ip_proto, __be16 sport, 3130 __be16 dport) 3131 { 3132 struct sk_buff *skb; 3133 struct iphdr *iph; 3134 3135 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 3136 if (!skb) 3137 return NULL; 3138 3139 /* Reserve room for dummy headers, this skb can pass 3140 * through good chunk of routing engine. 3141 */ 3142 skb_reset_mac_header(skb); 3143 skb_reset_network_header(skb); 3144 skb->protocol = htons(ETH_P_IP); 3145 iph = skb_put(skb, sizeof(struct iphdr)); 3146 iph->protocol = ip_proto; 3147 iph->saddr = src; 3148 iph->daddr = dst; 3149 iph->version = 0x4; 3150 iph->frag_off = 0; 3151 iph->ihl = 0x5; 3152 skb_set_transport_header(skb, skb->len); 3153 3154 switch (iph->protocol) { 3155 case IPPROTO_UDP: { 3156 struct udphdr *udph; 3157 3158 udph = skb_put_zero(skb, sizeof(struct udphdr)); 3159 udph->source = sport; 3160 udph->dest = dport; 3161 udph->len = htons(sizeof(struct udphdr)); 3162 udph->check = 0; 3163 break; 3164 } 3165 case IPPROTO_TCP: { 3166 struct tcphdr *tcph; 3167 3168 tcph = skb_put_zero(skb, sizeof(struct tcphdr)); 3169 tcph->source = sport; 3170 tcph->dest = dport; 3171 tcph->doff = sizeof(struct tcphdr) / 4; 3172 tcph->rst = 1; 3173 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr), 3174 src, dst, 0); 3175 break; 3176 } 3177 case IPPROTO_ICMP: { 3178 struct icmphdr *icmph; 3179 3180 icmph = skb_put_zero(skb, sizeof(struct icmphdr)); 3181 icmph->type = ICMP_ECHO; 3182 icmph->code = 0; 3183 } 3184 } 3185 3186 return skb; 3187 } 3188 3189 static int inet_rtm_valid_getroute_req(struct sk_buff *skb, 3190 const struct nlmsghdr *nlh, 3191 struct nlattr **tb, 3192 struct netlink_ext_ack *extack) 3193 { 3194 struct rtmsg *rtm; 3195 int i, err; 3196 3197 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) { 3198 NL_SET_ERR_MSG(extack, 3199 "ipv4: Invalid header for route get request"); 3200 return -EINVAL; 3201 } 3202 3203 if (!netlink_strict_get_check(skb)) 3204 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX, 3205 rtm_ipv4_policy, extack); 3206 3207 rtm = nlmsg_data(nlh); 3208 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) || 3209 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) || 3210 rtm->rtm_table || rtm->rtm_protocol || 3211 rtm->rtm_scope || rtm->rtm_type) { 3212 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request"); 3213 return -EINVAL; 3214 } 3215 3216 if (rtm->rtm_flags & ~(RTM_F_NOTIFY | 3217 RTM_F_LOOKUP_TABLE | 3218 RTM_F_FIB_MATCH)) { 3219 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request"); 3220 return -EINVAL; 3221 } 3222 3223 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX, 3224 rtm_ipv4_policy, extack); 3225 if (err) 3226 return err; 3227 3228 if ((tb[RTA_SRC] && !rtm->rtm_src_len) || 3229 (tb[RTA_DST] && !rtm->rtm_dst_len)) { 3230 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4"); 3231 return -EINVAL; 3232 } 3233 3234 for (i = 0; i <= RTA_MAX; i++) { 3235 if (!tb[i]) 3236 continue; 3237 3238 switch (i) { 3239 case RTA_IIF: 3240 case RTA_OIF: 3241 case RTA_SRC: 3242 case RTA_DST: 3243 case RTA_IP_PROTO: 3244 case RTA_SPORT: 3245 case RTA_DPORT: 3246 case RTA_MARK: 3247 case RTA_UID: 3248 break; 3249 default: 3250 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request"); 3251 return -EINVAL; 3252 } 3253 } 3254 3255 return 0; 3256 } 3257 3258 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, 3259 struct netlink_ext_ack *extack) 3260 { 3261 struct net *net = sock_net(in_skb->sk); 3262 struct nlattr *tb[RTA_MAX+1]; 3263 u32 table_id = RT_TABLE_MAIN; 3264 __be16 sport = 0, dport = 0; 3265 struct fib_result res = {}; 3266 u8 ip_proto = IPPROTO_UDP; 3267 struct rtable *rt = NULL; 3268 struct sk_buff *skb; 3269 struct rtmsg *rtm; 3270 struct flowi4 fl4 = {}; 3271 __be32 dst = 0; 3272 __be32 src = 0; 3273 dscp_t dscp; 3274 kuid_t uid; 3275 u32 iif; 3276 int err; 3277 int mark; 3278 3279 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack); 3280 if (err < 0) 3281 return err; 3282 3283 rtm = nlmsg_data(nlh); 3284 src = nla_get_in_addr_default(tb[RTA_SRC], 0); 3285 dst = nla_get_in_addr_default(tb[RTA_DST], 0); 3286 iif = nla_get_u32_default(tb[RTA_IIF], 0); 3287 mark = nla_get_u32_default(tb[RTA_MARK], 0); 3288 dscp = inet_dsfield_to_dscp(rtm->rtm_tos); 3289 if (tb[RTA_UID]) 3290 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID])); 3291 else 3292 uid = (iif ? INVALID_UID : current_uid()); 3293 3294 if (tb[RTA_IP_PROTO]) { 3295 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO], 3296 &ip_proto, AF_INET, extack); 3297 if (err) 3298 return err; 3299 } 3300 3301 if (tb[RTA_SPORT]) 3302 sport = nla_get_be16(tb[RTA_SPORT]); 3303 3304 if (tb[RTA_DPORT]) 3305 dport = nla_get_be16(tb[RTA_DPORT]); 3306 3307 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport); 3308 if (!skb) 3309 return -ENOBUFS; 3310 3311 fl4.daddr = dst; 3312 fl4.saddr = src; 3313 fl4.flowi4_tos = inet_dscp_to_dsfield(dscp); 3314 fl4.flowi4_oif = nla_get_u32_default(tb[RTA_OIF], 0); 3315 fl4.flowi4_mark = mark; 3316 fl4.flowi4_uid = uid; 3317 if (sport) 3318 fl4.fl4_sport = sport; 3319 if (dport) 3320 fl4.fl4_dport = dport; 3321 fl4.flowi4_proto = ip_proto; 3322 3323 rcu_read_lock(); 3324 3325 if (iif) { 3326 struct net_device *dev; 3327 3328 dev = dev_get_by_index_rcu(net, iif); 3329 if (!dev) { 3330 err = -ENODEV; 3331 goto errout_rcu; 3332 } 3333 3334 fl4.flowi4_iif = iif; /* for rt_fill_info */ 3335 skb->dev = dev; 3336 skb->mark = mark; 3337 err = ip_route_input_rcu(skb, dst, src, dscp, dev, 3338 &res) ? -EINVAL : 0; 3339 3340 rt = skb_rtable(skb); 3341 if (err == 0 && rt->dst.error) 3342 err = -rt->dst.error; 3343 } else { 3344 fl4.flowi4_iif = LOOPBACK_IFINDEX; 3345 skb->dev = net->loopback_dev; 3346 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb); 3347 err = 0; 3348 if (IS_ERR(rt)) 3349 err = PTR_ERR(rt); 3350 else 3351 skb_dst_set(skb, &rt->dst); 3352 } 3353 3354 if (err) 3355 goto errout_rcu; 3356 3357 if (rtm->rtm_flags & RTM_F_NOTIFY) 3358 rt->rt_flags |= RTCF_NOTIFY; 3359 3360 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE) 3361 table_id = res.table ? res.table->tb_id : 0; 3362 3363 /* reset skb for netlink reply msg */ 3364 skb_trim(skb, 0); 3365 skb_reset_network_header(skb); 3366 skb_reset_transport_header(skb); 3367 skb_reset_mac_header(skb); 3368 3369 if (rtm->rtm_flags & RTM_F_FIB_MATCH) { 3370 struct fib_rt_info fri; 3371 3372 if (!res.fi) { 3373 err = fib_props[res.type].error; 3374 if (!err) 3375 err = -EHOSTUNREACH; 3376 goto errout_rcu; 3377 } 3378 fri.fi = res.fi; 3379 fri.tb_id = table_id; 3380 fri.dst = res.prefix; 3381 fri.dst_len = res.prefixlen; 3382 fri.dscp = res.dscp; 3383 fri.type = rt->rt_type; 3384 fri.offload = 0; 3385 fri.trap = 0; 3386 fri.offload_failed = 0; 3387 if (res.fa_head) { 3388 struct fib_alias *fa; 3389 3390 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) { 3391 u8 slen = 32 - fri.dst_len; 3392 3393 if (fa->fa_slen == slen && 3394 fa->tb_id == fri.tb_id && 3395 fa->fa_dscp == fri.dscp && 3396 fa->fa_info == res.fi && 3397 fa->fa_type == fri.type) { 3398 fri.offload = READ_ONCE(fa->offload); 3399 fri.trap = READ_ONCE(fa->trap); 3400 fri.offload_failed = 3401 READ_ONCE(fa->offload_failed); 3402 break; 3403 } 3404 } 3405 } 3406 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid, 3407 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0); 3408 } else { 3409 err = rt_fill_info(net, dst, src, rt, table_id, res.dscp, &fl4, 3410 skb, NETLINK_CB(in_skb).portid, 3411 nlh->nlmsg_seq, 0); 3412 } 3413 if (err < 0) 3414 goto errout_rcu; 3415 3416 rcu_read_unlock(); 3417 3418 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 3419 3420 errout_free: 3421 return err; 3422 errout_rcu: 3423 rcu_read_unlock(); 3424 kfree_skb(skb); 3425 goto errout_free; 3426 } 3427 3428 void ip_rt_multicast_event(struct in_device *in_dev) 3429 { 3430 rt_cache_flush(dev_net(in_dev->dev)); 3431 } 3432 3433 #ifdef CONFIG_SYSCTL 3434 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 3435 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 3436 static int ip_rt_gc_elasticity __read_mostly = 8; 3437 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU; 3438 3439 static int ipv4_sysctl_rtcache_flush(const struct ctl_table *__ctl, int write, 3440 void *buffer, size_t *lenp, loff_t *ppos) 3441 { 3442 struct net *net = (struct net *)__ctl->extra1; 3443 3444 if (write) { 3445 rt_cache_flush(net); 3446 fnhe_genid_bump(net); 3447 return 0; 3448 } 3449 3450 return -EINVAL; 3451 } 3452 3453 static struct ctl_table ipv4_route_table[] = { 3454 { 3455 .procname = "gc_thresh", 3456 .data = &ipv4_dst_ops.gc_thresh, 3457 .maxlen = sizeof(int), 3458 .mode = 0644, 3459 .proc_handler = proc_dointvec, 3460 }, 3461 { 3462 .procname = "max_size", 3463 .data = &ip_rt_max_size, 3464 .maxlen = sizeof(int), 3465 .mode = 0644, 3466 .proc_handler = proc_dointvec, 3467 }, 3468 { 3469 /* Deprecated. Use gc_min_interval_ms */ 3470 3471 .procname = "gc_min_interval", 3472 .data = &ip_rt_gc_min_interval, 3473 .maxlen = sizeof(int), 3474 .mode = 0644, 3475 .proc_handler = proc_dointvec_jiffies, 3476 }, 3477 { 3478 .procname = "gc_min_interval_ms", 3479 .data = &ip_rt_gc_min_interval, 3480 .maxlen = sizeof(int), 3481 .mode = 0644, 3482 .proc_handler = proc_dointvec_ms_jiffies, 3483 }, 3484 { 3485 .procname = "gc_timeout", 3486 .data = &ip_rt_gc_timeout, 3487 .maxlen = sizeof(int), 3488 .mode = 0644, 3489 .proc_handler = proc_dointvec_jiffies, 3490 }, 3491 { 3492 .procname = "gc_interval", 3493 .data = &ip_rt_gc_interval, 3494 .maxlen = sizeof(int), 3495 .mode = 0644, 3496 .proc_handler = proc_dointvec_jiffies, 3497 }, 3498 { 3499 .procname = "redirect_load", 3500 .data = &ip_rt_redirect_load, 3501 .maxlen = sizeof(int), 3502 .mode = 0644, 3503 .proc_handler = proc_dointvec, 3504 }, 3505 { 3506 .procname = "redirect_number", 3507 .data = &ip_rt_redirect_number, 3508 .maxlen = sizeof(int), 3509 .mode = 0644, 3510 .proc_handler = proc_dointvec, 3511 }, 3512 { 3513 .procname = "redirect_silence", 3514 .data = &ip_rt_redirect_silence, 3515 .maxlen = sizeof(int), 3516 .mode = 0644, 3517 .proc_handler = proc_dointvec, 3518 }, 3519 { 3520 .procname = "error_cost", 3521 .data = &ip_rt_error_cost, 3522 .maxlen = sizeof(int), 3523 .mode = 0644, 3524 .proc_handler = proc_dointvec, 3525 }, 3526 { 3527 .procname = "error_burst", 3528 .data = &ip_rt_error_burst, 3529 .maxlen = sizeof(int), 3530 .mode = 0644, 3531 .proc_handler = proc_dointvec, 3532 }, 3533 { 3534 .procname = "gc_elasticity", 3535 .data = &ip_rt_gc_elasticity, 3536 .maxlen = sizeof(int), 3537 .mode = 0644, 3538 .proc_handler = proc_dointvec, 3539 }, 3540 }; 3541 3542 static const char ipv4_route_flush_procname[] = "flush"; 3543 3544 static struct ctl_table ipv4_route_netns_table[] = { 3545 { 3546 .procname = ipv4_route_flush_procname, 3547 .maxlen = sizeof(int), 3548 .mode = 0200, 3549 .proc_handler = ipv4_sysctl_rtcache_flush, 3550 }, 3551 { 3552 .procname = "min_pmtu", 3553 .data = &init_net.ipv4.ip_rt_min_pmtu, 3554 .maxlen = sizeof(int), 3555 .mode = 0644, 3556 .proc_handler = proc_dointvec_minmax, 3557 .extra1 = &ip_min_valid_pmtu, 3558 }, 3559 { 3560 .procname = "mtu_expires", 3561 .data = &init_net.ipv4.ip_rt_mtu_expires, 3562 .maxlen = sizeof(int), 3563 .mode = 0644, 3564 .proc_handler = proc_dointvec_jiffies, 3565 }, 3566 { 3567 .procname = "min_adv_mss", 3568 .data = &init_net.ipv4.ip_rt_min_advmss, 3569 .maxlen = sizeof(int), 3570 .mode = 0644, 3571 .proc_handler = proc_dointvec, 3572 }, 3573 }; 3574 3575 static __net_init int sysctl_route_net_init(struct net *net) 3576 { 3577 struct ctl_table *tbl; 3578 size_t table_size = ARRAY_SIZE(ipv4_route_netns_table); 3579 3580 tbl = ipv4_route_netns_table; 3581 if (!net_eq(net, &init_net)) { 3582 int i; 3583 3584 tbl = kmemdup(tbl, sizeof(ipv4_route_netns_table), GFP_KERNEL); 3585 if (!tbl) 3586 goto err_dup; 3587 3588 /* Don't export non-whitelisted sysctls to unprivileged users */ 3589 if (net->user_ns != &init_user_ns) { 3590 if (tbl[0].procname != ipv4_route_flush_procname) 3591 table_size = 0; 3592 } 3593 3594 /* Update the variables to point into the current struct net 3595 * except for the first element flush 3596 */ 3597 for (i = 1; i < table_size; i++) 3598 tbl[i].data += (void *)net - (void *)&init_net; 3599 } 3600 tbl[0].extra1 = net; 3601 3602 net->ipv4.route_hdr = register_net_sysctl_sz(net, "net/ipv4/route", 3603 tbl, table_size); 3604 if (!net->ipv4.route_hdr) 3605 goto err_reg; 3606 return 0; 3607 3608 err_reg: 3609 if (tbl != ipv4_route_netns_table) 3610 kfree(tbl); 3611 err_dup: 3612 return -ENOMEM; 3613 } 3614 3615 static __net_exit void sysctl_route_net_exit(struct net *net) 3616 { 3617 const struct ctl_table *tbl; 3618 3619 tbl = net->ipv4.route_hdr->ctl_table_arg; 3620 unregister_net_sysctl_table(net->ipv4.route_hdr); 3621 BUG_ON(tbl == ipv4_route_netns_table); 3622 kfree(tbl); 3623 } 3624 3625 static __net_initdata struct pernet_operations sysctl_route_ops = { 3626 .init = sysctl_route_net_init, 3627 .exit = sysctl_route_net_exit, 3628 }; 3629 #endif 3630 3631 static __net_init int netns_ip_rt_init(struct net *net) 3632 { 3633 /* Set default value for namespaceified sysctls */ 3634 net->ipv4.ip_rt_min_pmtu = DEFAULT_MIN_PMTU; 3635 net->ipv4.ip_rt_mtu_expires = DEFAULT_MTU_EXPIRES; 3636 net->ipv4.ip_rt_min_advmss = DEFAULT_MIN_ADVMSS; 3637 return 0; 3638 } 3639 3640 static struct pernet_operations __net_initdata ip_rt_ops = { 3641 .init = netns_ip_rt_init, 3642 }; 3643 3644 static __net_init int rt_genid_init(struct net *net) 3645 { 3646 atomic_set(&net->ipv4.rt_genid, 0); 3647 atomic_set(&net->fnhe_genid, 0); 3648 atomic_set(&net->ipv4.dev_addr_genid, get_random_u32()); 3649 return 0; 3650 } 3651 3652 static __net_initdata struct pernet_operations rt_genid_ops = { 3653 .init = rt_genid_init, 3654 }; 3655 3656 static int __net_init ipv4_inetpeer_init(struct net *net) 3657 { 3658 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL); 3659 3660 if (!bp) 3661 return -ENOMEM; 3662 inet_peer_base_init(bp); 3663 net->ipv4.peers = bp; 3664 return 0; 3665 } 3666 3667 static void __net_exit ipv4_inetpeer_exit(struct net *net) 3668 { 3669 struct inet_peer_base *bp = net->ipv4.peers; 3670 3671 net->ipv4.peers = NULL; 3672 inetpeer_invalidate_tree(bp); 3673 kfree(bp); 3674 } 3675 3676 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = { 3677 .init = ipv4_inetpeer_init, 3678 .exit = ipv4_inetpeer_exit, 3679 }; 3680 3681 #ifdef CONFIG_IP_ROUTE_CLASSID 3682 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 3683 #endif /* CONFIG_IP_ROUTE_CLASSID */ 3684 3685 static const struct rtnl_msg_handler ip_rt_rtnl_msg_handlers[] __initconst = { 3686 {.protocol = PF_INET, .msgtype = RTM_GETROUTE, 3687 .doit = inet_rtm_getroute, .flags = RTNL_FLAG_DOIT_UNLOCKED}, 3688 }; 3689 3690 int __init ip_rt_init(void) 3691 { 3692 void *idents_hash; 3693 int cpu; 3694 3695 /* For modern hosts, this will use 2 MB of memory */ 3696 idents_hash = alloc_large_system_hash("IP idents", 3697 sizeof(*ip_idents) + sizeof(*ip_tstamps), 3698 0, 3699 16, /* one bucket per 64 KB */ 3700 HASH_ZERO, 3701 NULL, 3702 &ip_idents_mask, 3703 2048, 3704 256*1024); 3705 3706 ip_idents = idents_hash; 3707 3708 get_random_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents)); 3709 3710 ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents); 3711 3712 for_each_possible_cpu(cpu) { 3713 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 3714 3715 INIT_LIST_HEAD(&ul->head); 3716 spin_lock_init(&ul->lock); 3717 } 3718 #ifdef CONFIG_IP_ROUTE_CLASSID 3719 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 3720 if (!ip_rt_acct) 3721 panic("IP: failed to allocate ip_rt_acct\n"); 3722 #endif 3723 3724 ipv4_dst_ops.kmem_cachep = KMEM_CACHE(rtable, 3725 SLAB_HWCACHE_ALIGN | SLAB_PANIC); 3726 3727 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 3728 3729 if (dst_entries_init(&ipv4_dst_ops) < 0) 3730 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 3731 3732 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 3733 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 3734 3735 ipv4_dst_ops.gc_thresh = ~0; 3736 ip_rt_max_size = INT_MAX; 3737 3738 devinet_init(); 3739 ip_fib_init(); 3740 3741 if (ip_rt_proc_init()) 3742 pr_err("Unable to create route proc files\n"); 3743 #ifdef CONFIG_XFRM 3744 xfrm_init(); 3745 xfrm4_init(); 3746 #endif 3747 rtnl_register_many(ip_rt_rtnl_msg_handlers); 3748 3749 #ifdef CONFIG_SYSCTL 3750 register_pernet_subsys(&sysctl_route_ops); 3751 #endif 3752 register_pernet_subsys(&ip_rt_ops); 3753 register_pernet_subsys(&rt_genid_ops); 3754 register_pernet_subsys(&ipv4_inetpeer_ops); 3755 return 0; 3756 } 3757 3758 #ifdef CONFIG_SYSCTL 3759 /* 3760 * We really need to sanitize the damn ipv4 init order, then all 3761 * this nonsense will go away. 3762 */ 3763 void __init ip_static_sysctl_init(void) 3764 { 3765 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table); 3766 } 3767 #endif 3768