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/uaccess.h> 65 #include <linux/bitops.h> 66 #include <linux/types.h> 67 #include <linux/kernel.h> 68 #include <linux/mm.h> 69 #include <linux/string.h> 70 #include <linux/socket.h> 71 #include <linux/sockios.h> 72 #include <linux/errno.h> 73 #include <linux/in.h> 74 #include <linux/inet.h> 75 #include <linux/netdevice.h> 76 #include <linux/proc_fs.h> 77 #include <linux/init.h> 78 #include <linux/skbuff.h> 79 #include <linux/inetdevice.h> 80 #include <linux/igmp.h> 81 #include <linux/pkt_sched.h> 82 #include <linux/mroute.h> 83 #include <linux/netfilter_ipv4.h> 84 #include <linux/random.h> 85 #include <linux/rcupdate.h> 86 #include <linux/times.h> 87 #include <linux/slab.h> 88 #include <linux/jhash.h> 89 #include <net/dst.h> 90 #include <net/dst_metadata.h> 91 #include <net/net_namespace.h> 92 #include <net/protocol.h> 93 #include <net/ip.h> 94 #include <net/route.h> 95 #include <net/inetpeer.h> 96 #include <net/sock.h> 97 #include <net/ip_fib.h> 98 #include <net/nexthop.h> 99 #include <net/arp.h> 100 #include <net/tcp.h> 101 #include <net/icmp.h> 102 #include <net/xfrm.h> 103 #include <net/lwtunnel.h> 104 #include <net/netevent.h> 105 #include <net/rtnetlink.h> 106 #ifdef CONFIG_SYSCTL 107 #include <linux/sysctl.h> 108 #endif 109 #include <net/secure_seq.h> 110 #include <net/ip_tunnels.h> 111 #include <net/l3mdev.h> 112 113 #include "fib_lookup.h" 114 115 #define RT_FL_TOS(oldflp4) \ 116 ((oldflp4)->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK)) 117 118 #define RT_GC_TIMEOUT (300*HZ) 119 120 static int ip_rt_max_size; 121 static int ip_rt_redirect_number __read_mostly = 9; 122 static int ip_rt_redirect_load __read_mostly = HZ / 50; 123 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1)); 124 static int ip_rt_error_cost __read_mostly = HZ; 125 static int ip_rt_error_burst __read_mostly = 5 * HZ; 126 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ; 127 static u32 ip_rt_min_pmtu __read_mostly = 512 + 20 + 20; 128 static int ip_rt_min_advmss __read_mostly = 256; 129 130 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT; 131 132 /* 133 * Interface to generic destination cache. 134 */ 135 136 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie); 137 static unsigned int ipv4_default_advmss(const struct dst_entry *dst); 138 static unsigned int ipv4_mtu(const struct dst_entry *dst); 139 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst); 140 static void ipv4_link_failure(struct sk_buff *skb); 141 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 142 struct sk_buff *skb, u32 mtu); 143 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, 144 struct sk_buff *skb); 145 static void ipv4_dst_destroy(struct dst_entry *dst); 146 147 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old) 148 { 149 WARN_ON(1); 150 return NULL; 151 } 152 153 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 154 struct sk_buff *skb, 155 const void *daddr); 156 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr); 157 158 static struct dst_ops ipv4_dst_ops = { 159 .family = AF_INET, 160 .check = ipv4_dst_check, 161 .default_advmss = ipv4_default_advmss, 162 .mtu = ipv4_mtu, 163 .cow_metrics = ipv4_cow_metrics, 164 .destroy = ipv4_dst_destroy, 165 .negative_advice = ipv4_negative_advice, 166 .link_failure = ipv4_link_failure, 167 .update_pmtu = ip_rt_update_pmtu, 168 .redirect = ip_do_redirect, 169 .local_out = __ip_local_out, 170 .neigh_lookup = ipv4_neigh_lookup, 171 .confirm_neigh = ipv4_confirm_neigh, 172 }; 173 174 #define ECN_OR_COST(class) TC_PRIO_##class 175 176 const __u8 ip_tos2prio[16] = { 177 TC_PRIO_BESTEFFORT, 178 ECN_OR_COST(BESTEFFORT), 179 TC_PRIO_BESTEFFORT, 180 ECN_OR_COST(BESTEFFORT), 181 TC_PRIO_BULK, 182 ECN_OR_COST(BULK), 183 TC_PRIO_BULK, 184 ECN_OR_COST(BULK), 185 TC_PRIO_INTERACTIVE, 186 ECN_OR_COST(INTERACTIVE), 187 TC_PRIO_INTERACTIVE, 188 ECN_OR_COST(INTERACTIVE), 189 TC_PRIO_INTERACTIVE_BULK, 190 ECN_OR_COST(INTERACTIVE_BULK), 191 TC_PRIO_INTERACTIVE_BULK, 192 ECN_OR_COST(INTERACTIVE_BULK) 193 }; 194 EXPORT_SYMBOL(ip_tos2prio); 195 196 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat); 197 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field) 198 199 #ifdef CONFIG_PROC_FS 200 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos) 201 { 202 if (*pos) 203 return NULL; 204 return SEQ_START_TOKEN; 205 } 206 207 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos) 208 { 209 ++*pos; 210 return NULL; 211 } 212 213 static void rt_cache_seq_stop(struct seq_file *seq, void *v) 214 { 215 } 216 217 static int rt_cache_seq_show(struct seq_file *seq, void *v) 218 { 219 if (v == SEQ_START_TOKEN) 220 seq_printf(seq, "%-127s\n", 221 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t" 222 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t" 223 "HHUptod\tSpecDst"); 224 return 0; 225 } 226 227 static const struct seq_operations rt_cache_seq_ops = { 228 .start = rt_cache_seq_start, 229 .next = rt_cache_seq_next, 230 .stop = rt_cache_seq_stop, 231 .show = rt_cache_seq_show, 232 }; 233 234 static int rt_cache_seq_open(struct inode *inode, struct file *file) 235 { 236 return seq_open(file, &rt_cache_seq_ops); 237 } 238 239 static const struct file_operations rt_cache_seq_fops = { 240 .open = rt_cache_seq_open, 241 .read = seq_read, 242 .llseek = seq_lseek, 243 .release = seq_release, 244 }; 245 246 247 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos) 248 { 249 int cpu; 250 251 if (*pos == 0) 252 return SEQ_START_TOKEN; 253 254 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) { 255 if (!cpu_possible(cpu)) 256 continue; 257 *pos = cpu+1; 258 return &per_cpu(rt_cache_stat, cpu); 259 } 260 return NULL; 261 } 262 263 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos) 264 { 265 int cpu; 266 267 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) { 268 if (!cpu_possible(cpu)) 269 continue; 270 *pos = cpu+1; 271 return &per_cpu(rt_cache_stat, cpu); 272 } 273 return NULL; 274 275 } 276 277 static void rt_cpu_seq_stop(struct seq_file *seq, void *v) 278 { 279 280 } 281 282 static int rt_cpu_seq_show(struct seq_file *seq, void *v) 283 { 284 struct rt_cache_stat *st = v; 285 286 if (v == SEQ_START_TOKEN) { 287 seq_printf(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"); 288 return 0; 289 } 290 291 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x " 292 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n", 293 dst_entries_get_slow(&ipv4_dst_ops), 294 0, /* st->in_hit */ 295 st->in_slow_tot, 296 st->in_slow_mc, 297 st->in_no_route, 298 st->in_brd, 299 st->in_martian_dst, 300 st->in_martian_src, 301 302 0, /* st->out_hit */ 303 st->out_slow_tot, 304 st->out_slow_mc, 305 306 0, /* st->gc_total */ 307 0, /* st->gc_ignored */ 308 0, /* st->gc_goal_miss */ 309 0, /* st->gc_dst_overflow */ 310 0, /* st->in_hlist_search */ 311 0 /* st->out_hlist_search */ 312 ); 313 return 0; 314 } 315 316 static const struct seq_operations rt_cpu_seq_ops = { 317 .start = rt_cpu_seq_start, 318 .next = rt_cpu_seq_next, 319 .stop = rt_cpu_seq_stop, 320 .show = rt_cpu_seq_show, 321 }; 322 323 324 static int rt_cpu_seq_open(struct inode *inode, struct file *file) 325 { 326 return seq_open(file, &rt_cpu_seq_ops); 327 } 328 329 static const struct file_operations rt_cpu_seq_fops = { 330 .open = rt_cpu_seq_open, 331 .read = seq_read, 332 .llseek = seq_lseek, 333 .release = seq_release, 334 }; 335 336 #ifdef CONFIG_IP_ROUTE_CLASSID 337 static int rt_acct_proc_show(struct seq_file *m, void *v) 338 { 339 struct ip_rt_acct *dst, *src; 340 unsigned int i, j; 341 342 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL); 343 if (!dst) 344 return -ENOMEM; 345 346 for_each_possible_cpu(i) { 347 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i); 348 for (j = 0; j < 256; j++) { 349 dst[j].o_bytes += src[j].o_bytes; 350 dst[j].o_packets += src[j].o_packets; 351 dst[j].i_bytes += src[j].i_bytes; 352 dst[j].i_packets += src[j].i_packets; 353 } 354 } 355 356 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct)); 357 kfree(dst); 358 return 0; 359 } 360 #endif 361 362 static int __net_init ip_rt_do_proc_init(struct net *net) 363 { 364 struct proc_dir_entry *pde; 365 366 pde = proc_create("rt_cache", 0444, net->proc_net, 367 &rt_cache_seq_fops); 368 if (!pde) 369 goto err1; 370 371 pde = proc_create("rt_cache", 0444, 372 net->proc_net_stat, &rt_cpu_seq_fops); 373 if (!pde) 374 goto err2; 375 376 #ifdef CONFIG_IP_ROUTE_CLASSID 377 pde = proc_create_single("rt_acct", 0, net->proc_net, 378 rt_acct_proc_show); 379 if (!pde) 380 goto err3; 381 #endif 382 return 0; 383 384 #ifdef CONFIG_IP_ROUTE_CLASSID 385 err3: 386 remove_proc_entry("rt_cache", net->proc_net_stat); 387 #endif 388 err2: 389 remove_proc_entry("rt_cache", net->proc_net); 390 err1: 391 return -ENOMEM; 392 } 393 394 static void __net_exit ip_rt_do_proc_exit(struct net *net) 395 { 396 remove_proc_entry("rt_cache", net->proc_net_stat); 397 remove_proc_entry("rt_cache", net->proc_net); 398 #ifdef CONFIG_IP_ROUTE_CLASSID 399 remove_proc_entry("rt_acct", net->proc_net); 400 #endif 401 } 402 403 static struct pernet_operations ip_rt_proc_ops __net_initdata = { 404 .init = ip_rt_do_proc_init, 405 .exit = ip_rt_do_proc_exit, 406 }; 407 408 static int __init ip_rt_proc_init(void) 409 { 410 return register_pernet_subsys(&ip_rt_proc_ops); 411 } 412 413 #else 414 static inline int ip_rt_proc_init(void) 415 { 416 return 0; 417 } 418 #endif /* CONFIG_PROC_FS */ 419 420 static inline bool rt_is_expired(const struct rtable *rth) 421 { 422 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev)); 423 } 424 425 void rt_cache_flush(struct net *net) 426 { 427 rt_genid_bump_ipv4(net); 428 } 429 430 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 431 struct sk_buff *skb, 432 const void *daddr) 433 { 434 const struct rtable *rt = container_of(dst, struct rtable, dst); 435 struct net_device *dev = dst->dev; 436 struct neighbour *n; 437 438 rcu_read_lock_bh(); 439 440 if (likely(rt->rt_gw_family == AF_INET)) { 441 n = ip_neigh_gw4(dev, rt->rt_gw4); 442 } else if (rt->rt_gw_family == AF_INET6) { 443 n = ip_neigh_gw6(dev, &rt->rt_gw6); 444 } else { 445 __be32 pkey; 446 447 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr); 448 n = ip_neigh_gw4(dev, pkey); 449 } 450 451 if (n && !refcount_inc_not_zero(&n->refcnt)) 452 n = NULL; 453 454 rcu_read_unlock_bh(); 455 456 return n; 457 } 458 459 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr) 460 { 461 const struct rtable *rt = container_of(dst, struct rtable, dst); 462 struct net_device *dev = dst->dev; 463 const __be32 *pkey = daddr; 464 465 if (rt->rt_gw_family == AF_INET) { 466 pkey = (const __be32 *)&rt->rt_gw4; 467 } else if (rt->rt_gw_family == AF_INET6) { 468 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6); 469 } else if (!daddr || 470 (rt->rt_flags & 471 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) { 472 return; 473 } 474 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey); 475 } 476 477 #define IP_IDENTS_SZ 2048u 478 479 static atomic_t *ip_idents __read_mostly; 480 static u32 *ip_tstamps __read_mostly; 481 482 /* In order to protect privacy, we add a perturbation to identifiers 483 * if one generator is seldom used. This makes hard for an attacker 484 * to infer how many packets were sent between two points in time. 485 */ 486 u32 ip_idents_reserve(u32 hash, int segs) 487 { 488 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ; 489 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ; 490 u32 old = READ_ONCE(*p_tstamp); 491 u32 now = (u32)jiffies; 492 u32 new, delta = 0; 493 494 if (old != now && cmpxchg(p_tstamp, old, now) == old) 495 delta = prandom_u32_max(now - old); 496 497 /* Do not use atomic_add_return() as it makes UBSAN unhappy */ 498 do { 499 old = (u32)atomic_read(p_id); 500 new = old + delta + segs; 501 } while (atomic_cmpxchg(p_id, old, new) != old); 502 503 return new - segs; 504 } 505 EXPORT_SYMBOL(ip_idents_reserve); 506 507 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs) 508 { 509 u32 hash, id; 510 511 /* Note the following code is not safe, but this is okay. */ 512 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key))) 513 get_random_bytes(&net->ipv4.ip_id_key, 514 sizeof(net->ipv4.ip_id_key)); 515 516 hash = siphash_3u32((__force u32)iph->daddr, 517 (__force u32)iph->saddr, 518 iph->protocol, 519 &net->ipv4.ip_id_key); 520 id = ip_idents_reserve(hash, segs); 521 iph->id = htons(id); 522 } 523 EXPORT_SYMBOL(__ip_select_ident); 524 525 static void __build_flow_key(const struct net *net, struct flowi4 *fl4, 526 const struct sock *sk, 527 const struct iphdr *iph, 528 int oif, u8 tos, 529 u8 prot, u32 mark, int flow_flags) 530 { 531 if (sk) { 532 const struct inet_sock *inet = inet_sk(sk); 533 534 oif = sk->sk_bound_dev_if; 535 mark = sk->sk_mark; 536 tos = RT_CONN_FLAGS(sk); 537 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol; 538 } 539 flowi4_init_output(fl4, oif, mark, tos, 540 RT_SCOPE_UNIVERSE, prot, 541 flow_flags, 542 iph->daddr, iph->saddr, 0, 0, 543 sock_net_uid(net, sk)); 544 } 545 546 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb, 547 const struct sock *sk) 548 { 549 const struct net *net = dev_net(skb->dev); 550 const struct iphdr *iph = ip_hdr(skb); 551 int oif = skb->dev->ifindex; 552 u8 tos = RT_TOS(iph->tos); 553 u8 prot = iph->protocol; 554 u32 mark = skb->mark; 555 556 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0); 557 } 558 559 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk) 560 { 561 const struct inet_sock *inet = inet_sk(sk); 562 const struct ip_options_rcu *inet_opt; 563 __be32 daddr = inet->inet_daddr; 564 565 rcu_read_lock(); 566 inet_opt = rcu_dereference(inet->inet_opt); 567 if (inet_opt && inet_opt->opt.srr) 568 daddr = inet_opt->opt.faddr; 569 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark, 570 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 571 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol, 572 inet_sk_flowi_flags(sk), 573 daddr, inet->inet_saddr, 0, 0, sk->sk_uid); 574 rcu_read_unlock(); 575 } 576 577 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk, 578 const struct sk_buff *skb) 579 { 580 if (skb) 581 build_skb_flow_key(fl4, skb, sk); 582 else 583 build_sk_flow_key(fl4, sk); 584 } 585 586 static DEFINE_SPINLOCK(fnhe_lock); 587 588 static void fnhe_flush_routes(struct fib_nh_exception *fnhe) 589 { 590 struct rtable *rt; 591 592 rt = rcu_dereference(fnhe->fnhe_rth_input); 593 if (rt) { 594 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL); 595 dst_dev_put(&rt->dst); 596 dst_release(&rt->dst); 597 } 598 rt = rcu_dereference(fnhe->fnhe_rth_output); 599 if (rt) { 600 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL); 601 dst_dev_put(&rt->dst); 602 dst_release(&rt->dst); 603 } 604 } 605 606 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash) 607 { 608 struct fib_nh_exception *fnhe, *oldest; 609 610 oldest = rcu_dereference(hash->chain); 611 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe; 612 fnhe = rcu_dereference(fnhe->fnhe_next)) { 613 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) 614 oldest = fnhe; 615 } 616 fnhe_flush_routes(oldest); 617 return oldest; 618 } 619 620 static inline u32 fnhe_hashfun(__be32 daddr) 621 { 622 static u32 fnhe_hashrnd __read_mostly; 623 u32 hval; 624 625 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd)); 626 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd); 627 return hash_32(hval, FNHE_HASH_SHIFT); 628 } 629 630 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe) 631 { 632 rt->rt_pmtu = fnhe->fnhe_pmtu; 633 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked; 634 rt->dst.expires = fnhe->fnhe_expires; 635 636 if (fnhe->fnhe_gw) { 637 rt->rt_flags |= RTCF_REDIRECTED; 638 rt->rt_gw_family = AF_INET; 639 rt->rt_gw4 = fnhe->fnhe_gw; 640 } 641 } 642 643 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr, 644 __be32 gw, u32 pmtu, bool lock, 645 unsigned long expires) 646 { 647 struct fnhe_hash_bucket *hash; 648 struct fib_nh_exception *fnhe; 649 struct rtable *rt; 650 u32 genid, hval; 651 unsigned int i; 652 int depth; 653 654 genid = fnhe_genid(dev_net(nhc->nhc_dev)); 655 hval = fnhe_hashfun(daddr); 656 657 spin_lock_bh(&fnhe_lock); 658 659 hash = rcu_dereference(nhc->nhc_exceptions); 660 if (!hash) { 661 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC); 662 if (!hash) 663 goto out_unlock; 664 rcu_assign_pointer(nhc->nhc_exceptions, hash); 665 } 666 667 hash += hval; 668 669 depth = 0; 670 for (fnhe = rcu_dereference(hash->chain); fnhe; 671 fnhe = rcu_dereference(fnhe->fnhe_next)) { 672 if (fnhe->fnhe_daddr == daddr) 673 break; 674 depth++; 675 } 676 677 if (fnhe) { 678 if (fnhe->fnhe_genid != genid) 679 fnhe->fnhe_genid = genid; 680 if (gw) 681 fnhe->fnhe_gw = gw; 682 if (pmtu) { 683 fnhe->fnhe_pmtu = pmtu; 684 fnhe->fnhe_mtu_locked = lock; 685 } 686 fnhe->fnhe_expires = max(1UL, expires); 687 /* Update all cached dsts too */ 688 rt = rcu_dereference(fnhe->fnhe_rth_input); 689 if (rt) 690 fill_route_from_fnhe(rt, fnhe); 691 rt = rcu_dereference(fnhe->fnhe_rth_output); 692 if (rt) 693 fill_route_from_fnhe(rt, fnhe); 694 } else { 695 if (depth > FNHE_RECLAIM_DEPTH) 696 fnhe = fnhe_oldest(hash); 697 else { 698 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC); 699 if (!fnhe) 700 goto out_unlock; 701 702 fnhe->fnhe_next = hash->chain; 703 rcu_assign_pointer(hash->chain, fnhe); 704 } 705 fnhe->fnhe_genid = genid; 706 fnhe->fnhe_daddr = daddr; 707 fnhe->fnhe_gw = gw; 708 fnhe->fnhe_pmtu = pmtu; 709 fnhe->fnhe_mtu_locked = lock; 710 fnhe->fnhe_expires = max(1UL, expires); 711 712 /* Exception created; mark the cached routes for the nexthop 713 * stale, so anyone caching it rechecks if this exception 714 * applies to them. 715 */ 716 rt = rcu_dereference(nhc->nhc_rth_input); 717 if (rt) 718 rt->dst.obsolete = DST_OBSOLETE_KILL; 719 720 for_each_possible_cpu(i) { 721 struct rtable __rcu **prt; 722 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i); 723 rt = rcu_dereference(*prt); 724 if (rt) 725 rt->dst.obsolete = DST_OBSOLETE_KILL; 726 } 727 } 728 729 fnhe->fnhe_stamp = jiffies; 730 731 out_unlock: 732 spin_unlock_bh(&fnhe_lock); 733 } 734 735 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4, 736 bool kill_route) 737 { 738 __be32 new_gw = icmp_hdr(skb)->un.gateway; 739 __be32 old_gw = ip_hdr(skb)->saddr; 740 struct net_device *dev = skb->dev; 741 struct in_device *in_dev; 742 struct fib_result res; 743 struct neighbour *n; 744 struct net *net; 745 746 switch (icmp_hdr(skb)->code & 7) { 747 case ICMP_REDIR_NET: 748 case ICMP_REDIR_NETTOS: 749 case ICMP_REDIR_HOST: 750 case ICMP_REDIR_HOSTTOS: 751 break; 752 753 default: 754 return; 755 } 756 757 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw) 758 return; 759 760 in_dev = __in_dev_get_rcu(dev); 761 if (!in_dev) 762 return; 763 764 net = dev_net(dev); 765 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) || 766 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) || 767 ipv4_is_zeronet(new_gw)) 768 goto reject_redirect; 769 770 if (!IN_DEV_SHARED_MEDIA(in_dev)) { 771 if (!inet_addr_onlink(in_dev, new_gw, old_gw)) 772 goto reject_redirect; 773 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) 774 goto reject_redirect; 775 } else { 776 if (inet_addr_type(net, new_gw) != RTN_UNICAST) 777 goto reject_redirect; 778 } 779 780 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw); 781 if (!n) 782 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev); 783 if (!IS_ERR(n)) { 784 if (!(n->nud_state & NUD_VALID)) { 785 neigh_event_send(n, NULL); 786 } else { 787 if (fib_lookup(net, fl4, &res, 0) == 0) { 788 struct fib_nh_common *nhc = FIB_RES_NHC(res); 789 790 update_or_create_fnhe(nhc, fl4->daddr, new_gw, 791 0, false, 792 jiffies + ip_rt_gc_timeout); 793 } 794 if (kill_route) 795 rt->dst.obsolete = DST_OBSOLETE_KILL; 796 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n); 797 } 798 neigh_release(n); 799 } 800 return; 801 802 reject_redirect: 803 #ifdef CONFIG_IP_ROUTE_VERBOSE 804 if (IN_DEV_LOG_MARTIANS(in_dev)) { 805 const struct iphdr *iph = (const struct iphdr *) skb->data; 806 __be32 daddr = iph->daddr; 807 __be32 saddr = iph->saddr; 808 809 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n" 810 " Advised path = %pI4 -> %pI4\n", 811 &old_gw, dev->name, &new_gw, 812 &saddr, &daddr); 813 } 814 #endif 815 ; 816 } 817 818 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb) 819 { 820 struct rtable *rt; 821 struct flowi4 fl4; 822 const struct iphdr *iph = (const struct iphdr *) skb->data; 823 struct net *net = dev_net(skb->dev); 824 int oif = skb->dev->ifindex; 825 u8 tos = RT_TOS(iph->tos); 826 u8 prot = iph->protocol; 827 u32 mark = skb->mark; 828 829 rt = (struct rtable *) dst; 830 831 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0); 832 __ip_do_redirect(rt, skb, &fl4, true); 833 } 834 835 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst) 836 { 837 struct rtable *rt = (struct rtable *)dst; 838 struct dst_entry *ret = dst; 839 840 if (rt) { 841 if (dst->obsolete > 0) { 842 ip_rt_put(rt); 843 ret = NULL; 844 } else if ((rt->rt_flags & RTCF_REDIRECTED) || 845 rt->dst.expires) { 846 ip_rt_put(rt); 847 ret = NULL; 848 } 849 } 850 return ret; 851 } 852 853 /* 854 * Algorithm: 855 * 1. The first ip_rt_redirect_number redirects are sent 856 * with exponential backoff, then we stop sending them at all, 857 * assuming that the host ignores our redirects. 858 * 2. If we did not see packets requiring redirects 859 * during ip_rt_redirect_silence, we assume that the host 860 * forgot redirected route and start to send redirects again. 861 * 862 * This algorithm is much cheaper and more intelligent than dumb load limiting 863 * in icmp.c. 864 * 865 * NOTE. Do not forget to inhibit load limiting for redirects (redundant) 866 * and "frag. need" (breaks PMTU discovery) in icmp.c. 867 */ 868 869 void ip_rt_send_redirect(struct sk_buff *skb) 870 { 871 struct rtable *rt = skb_rtable(skb); 872 struct in_device *in_dev; 873 struct inet_peer *peer; 874 struct net *net; 875 int log_martians; 876 int vif; 877 878 rcu_read_lock(); 879 in_dev = __in_dev_get_rcu(rt->dst.dev); 880 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) { 881 rcu_read_unlock(); 882 return; 883 } 884 log_martians = IN_DEV_LOG_MARTIANS(in_dev); 885 vif = l3mdev_master_ifindex_rcu(rt->dst.dev); 886 rcu_read_unlock(); 887 888 net = dev_net(rt->dst.dev); 889 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1); 890 if (!peer) { 891 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, 892 rt_nexthop(rt, ip_hdr(skb)->daddr)); 893 return; 894 } 895 896 /* No redirected packets during ip_rt_redirect_silence; 897 * reset the algorithm. 898 */ 899 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) { 900 peer->rate_tokens = 0; 901 peer->n_redirects = 0; 902 } 903 904 /* Too many ignored redirects; do not send anything 905 * set dst.rate_last to the last seen redirected packet. 906 */ 907 if (peer->n_redirects >= ip_rt_redirect_number) { 908 peer->rate_last = jiffies; 909 goto out_put_peer; 910 } 911 912 /* Check for load limit; set rate_last to the latest sent 913 * redirect. 914 */ 915 if (peer->rate_tokens == 0 || 916 time_after(jiffies, 917 (peer->rate_last + 918 (ip_rt_redirect_load << peer->rate_tokens)))) { 919 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr); 920 921 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw); 922 peer->rate_last = jiffies; 923 ++peer->rate_tokens; 924 ++peer->n_redirects; 925 #ifdef CONFIG_IP_ROUTE_VERBOSE 926 if (log_martians && 927 peer->rate_tokens == ip_rt_redirect_number) 928 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n", 929 &ip_hdr(skb)->saddr, inet_iif(skb), 930 &ip_hdr(skb)->daddr, &gw); 931 #endif 932 } 933 out_put_peer: 934 inet_putpeer(peer); 935 } 936 937 static int ip_error(struct sk_buff *skb) 938 { 939 struct rtable *rt = skb_rtable(skb); 940 struct net_device *dev = skb->dev; 941 struct in_device *in_dev; 942 struct inet_peer *peer; 943 unsigned long now; 944 struct net *net; 945 bool send; 946 int code; 947 948 if (netif_is_l3_master(skb->dev)) { 949 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif); 950 if (!dev) 951 goto out; 952 } 953 954 in_dev = __in_dev_get_rcu(dev); 955 956 /* IP on this device is disabled. */ 957 if (!in_dev) 958 goto out; 959 960 net = dev_net(rt->dst.dev); 961 if (!IN_DEV_FORWARD(in_dev)) { 962 switch (rt->dst.error) { 963 case EHOSTUNREACH: 964 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS); 965 break; 966 967 case ENETUNREACH: 968 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 969 break; 970 } 971 goto out; 972 } 973 974 switch (rt->dst.error) { 975 case EINVAL: 976 default: 977 goto out; 978 case EHOSTUNREACH: 979 code = ICMP_HOST_UNREACH; 980 break; 981 case ENETUNREACH: 982 code = ICMP_NET_UNREACH; 983 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 984 break; 985 case EACCES: 986 code = ICMP_PKT_FILTERED; 987 break; 988 } 989 990 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 991 l3mdev_master_ifindex(skb->dev), 1); 992 993 send = true; 994 if (peer) { 995 now = jiffies; 996 peer->rate_tokens += now - peer->rate_last; 997 if (peer->rate_tokens > ip_rt_error_burst) 998 peer->rate_tokens = ip_rt_error_burst; 999 peer->rate_last = now; 1000 if (peer->rate_tokens >= ip_rt_error_cost) 1001 peer->rate_tokens -= ip_rt_error_cost; 1002 else 1003 send = false; 1004 inet_putpeer(peer); 1005 } 1006 if (send) 1007 icmp_send(skb, ICMP_DEST_UNREACH, code, 0); 1008 1009 out: kfree_skb(skb); 1010 return 0; 1011 } 1012 1013 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu) 1014 { 1015 struct dst_entry *dst = &rt->dst; 1016 u32 old_mtu = ipv4_mtu(dst); 1017 struct fib_result res; 1018 bool lock = false; 1019 1020 if (ip_mtu_locked(dst)) 1021 return; 1022 1023 if (old_mtu < mtu) 1024 return; 1025 1026 if (mtu < ip_rt_min_pmtu) { 1027 lock = true; 1028 mtu = min(old_mtu, ip_rt_min_pmtu); 1029 } 1030 1031 if (rt->rt_pmtu == mtu && !lock && 1032 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2)) 1033 return; 1034 1035 rcu_read_lock(); 1036 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) { 1037 struct fib_nh_common *nhc = FIB_RES_NHC(res); 1038 1039 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1040 jiffies + ip_rt_mtu_expires); 1041 } 1042 rcu_read_unlock(); 1043 } 1044 1045 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 1046 struct sk_buff *skb, u32 mtu) 1047 { 1048 struct rtable *rt = (struct rtable *) dst; 1049 struct flowi4 fl4; 1050 1051 ip_rt_build_flow_key(&fl4, sk, skb); 1052 __ip_rt_update_pmtu(rt, &fl4, mtu); 1053 } 1054 1055 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, 1056 int oif, u8 protocol) 1057 { 1058 const struct iphdr *iph = (const struct iphdr *) skb->data; 1059 struct flowi4 fl4; 1060 struct rtable *rt; 1061 u32 mark = IP4_REPLY_MARK(net, skb->mark); 1062 1063 __build_flow_key(net, &fl4, NULL, iph, oif, 1064 RT_TOS(iph->tos), protocol, mark, 0); 1065 rt = __ip_route_output_key(net, &fl4); 1066 if (!IS_ERR(rt)) { 1067 __ip_rt_update_pmtu(rt, &fl4, mtu); 1068 ip_rt_put(rt); 1069 } 1070 } 1071 EXPORT_SYMBOL_GPL(ipv4_update_pmtu); 1072 1073 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1074 { 1075 const struct iphdr *iph = (const struct iphdr *) skb->data; 1076 struct flowi4 fl4; 1077 struct rtable *rt; 1078 1079 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0); 1080 1081 if (!fl4.flowi4_mark) 1082 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark); 1083 1084 rt = __ip_route_output_key(sock_net(sk), &fl4); 1085 if (!IS_ERR(rt)) { 1086 __ip_rt_update_pmtu(rt, &fl4, mtu); 1087 ip_rt_put(rt); 1088 } 1089 } 1090 1091 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1092 { 1093 const struct iphdr *iph = (const struct iphdr *) skb->data; 1094 struct flowi4 fl4; 1095 struct rtable *rt; 1096 struct dst_entry *odst = NULL; 1097 bool new = false; 1098 struct net *net = sock_net(sk); 1099 1100 bh_lock_sock(sk); 1101 1102 if (!ip_sk_accept_pmtu(sk)) 1103 goto out; 1104 1105 odst = sk_dst_get(sk); 1106 1107 if (sock_owned_by_user(sk) || !odst) { 1108 __ipv4_sk_update_pmtu(skb, sk, mtu); 1109 goto out; 1110 } 1111 1112 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1113 1114 rt = (struct rtable *)odst; 1115 if (odst->obsolete && !odst->ops->check(odst, 0)) { 1116 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1117 if (IS_ERR(rt)) 1118 goto out; 1119 1120 new = true; 1121 } 1122 1123 __ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu); 1124 1125 if (!dst_check(&rt->dst, 0)) { 1126 if (new) 1127 dst_release(&rt->dst); 1128 1129 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1130 if (IS_ERR(rt)) 1131 goto out; 1132 1133 new = true; 1134 } 1135 1136 if (new) 1137 sk_dst_set(sk, &rt->dst); 1138 1139 out: 1140 bh_unlock_sock(sk); 1141 dst_release(odst); 1142 } 1143 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu); 1144 1145 void ipv4_redirect(struct sk_buff *skb, struct net *net, 1146 int oif, u8 protocol) 1147 { 1148 const struct iphdr *iph = (const struct iphdr *) skb->data; 1149 struct flowi4 fl4; 1150 struct rtable *rt; 1151 1152 __build_flow_key(net, &fl4, NULL, iph, oif, 1153 RT_TOS(iph->tos), protocol, 0, 0); 1154 rt = __ip_route_output_key(net, &fl4); 1155 if (!IS_ERR(rt)) { 1156 __ip_do_redirect(rt, skb, &fl4, false); 1157 ip_rt_put(rt); 1158 } 1159 } 1160 EXPORT_SYMBOL_GPL(ipv4_redirect); 1161 1162 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk) 1163 { 1164 const struct iphdr *iph = (const struct iphdr *) skb->data; 1165 struct flowi4 fl4; 1166 struct rtable *rt; 1167 struct net *net = sock_net(sk); 1168 1169 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1170 rt = __ip_route_output_key(net, &fl4); 1171 if (!IS_ERR(rt)) { 1172 __ip_do_redirect(rt, skb, &fl4, false); 1173 ip_rt_put(rt); 1174 } 1175 } 1176 EXPORT_SYMBOL_GPL(ipv4_sk_redirect); 1177 1178 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie) 1179 { 1180 struct rtable *rt = (struct rtable *) dst; 1181 1182 /* All IPV4 dsts are created with ->obsolete set to the value 1183 * DST_OBSOLETE_FORCE_CHK which forces validation calls down 1184 * into this function always. 1185 * 1186 * When a PMTU/redirect information update invalidates a route, 1187 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or 1188 * DST_OBSOLETE_DEAD. 1189 */ 1190 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt)) 1191 return NULL; 1192 return dst; 1193 } 1194 1195 static void ipv4_send_dest_unreach(struct sk_buff *skb) 1196 { 1197 struct ip_options opt; 1198 int res; 1199 1200 /* Recompile ip options since IPCB may not be valid anymore. 1201 * Also check we have a reasonable ipv4 header. 1202 */ 1203 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) || 1204 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5) 1205 return; 1206 1207 memset(&opt, 0, sizeof(opt)); 1208 if (ip_hdr(skb)->ihl > 5) { 1209 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4)) 1210 return; 1211 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr); 1212 1213 rcu_read_lock(); 1214 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL); 1215 rcu_read_unlock(); 1216 1217 if (res) 1218 return; 1219 } 1220 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt); 1221 } 1222 1223 static void ipv4_link_failure(struct sk_buff *skb) 1224 { 1225 struct rtable *rt; 1226 1227 ipv4_send_dest_unreach(skb); 1228 1229 rt = skb_rtable(skb); 1230 if (rt) 1231 dst_set_expires(&rt->dst, 0); 1232 } 1233 1234 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb) 1235 { 1236 pr_debug("%s: %pI4 -> %pI4, %s\n", 1237 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1238 skb->dev ? skb->dev->name : "?"); 1239 kfree_skb(skb); 1240 WARN_ON(1); 1241 return 0; 1242 } 1243 1244 /* 1245 We do not cache source address of outgoing interface, 1246 because it is used only by IP RR, TS and SRR options, 1247 so that it out of fast path. 1248 1249 BTW remember: "addr" is allowed to be not aligned 1250 in IP options! 1251 */ 1252 1253 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt) 1254 { 1255 __be32 src; 1256 1257 if (rt_is_output_route(rt)) 1258 src = ip_hdr(skb)->saddr; 1259 else { 1260 struct fib_result res; 1261 struct iphdr *iph = ip_hdr(skb); 1262 struct flowi4 fl4 = { 1263 .daddr = iph->daddr, 1264 .saddr = iph->saddr, 1265 .flowi4_tos = RT_TOS(iph->tos), 1266 .flowi4_oif = rt->dst.dev->ifindex, 1267 .flowi4_iif = skb->dev->ifindex, 1268 .flowi4_mark = skb->mark, 1269 }; 1270 1271 rcu_read_lock(); 1272 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0) 1273 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res); 1274 else 1275 src = inet_select_addr(rt->dst.dev, 1276 rt_nexthop(rt, iph->daddr), 1277 RT_SCOPE_UNIVERSE); 1278 rcu_read_unlock(); 1279 } 1280 memcpy(addr, &src, 4); 1281 } 1282 1283 #ifdef CONFIG_IP_ROUTE_CLASSID 1284 static void set_class_tag(struct rtable *rt, u32 tag) 1285 { 1286 if (!(rt->dst.tclassid & 0xFFFF)) 1287 rt->dst.tclassid |= tag & 0xFFFF; 1288 if (!(rt->dst.tclassid & 0xFFFF0000)) 1289 rt->dst.tclassid |= tag & 0xFFFF0000; 1290 } 1291 #endif 1292 1293 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1294 { 1295 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr); 1296 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size, 1297 ip_rt_min_advmss); 1298 1299 return min(advmss, IPV4_MAX_PMTU - header_size); 1300 } 1301 1302 static unsigned int ipv4_mtu(const struct dst_entry *dst) 1303 { 1304 const struct rtable *rt = (const struct rtable *) dst; 1305 unsigned int mtu = rt->rt_pmtu; 1306 1307 if (!mtu || time_after_eq(jiffies, rt->dst.expires)) 1308 mtu = dst_metric_raw(dst, RTAX_MTU); 1309 1310 if (mtu) 1311 return mtu; 1312 1313 mtu = READ_ONCE(dst->dev->mtu); 1314 1315 if (unlikely(ip_mtu_locked(dst))) { 1316 if (rt->rt_gw_family && mtu > 576) 1317 mtu = 576; 1318 } 1319 1320 mtu = min_t(unsigned int, mtu, IP_MAX_MTU); 1321 1322 return mtu - lwtunnel_headroom(dst->lwtstate, mtu); 1323 } 1324 1325 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr) 1326 { 1327 struct fnhe_hash_bucket *hash; 1328 struct fib_nh_exception *fnhe, __rcu **fnhe_p; 1329 u32 hval = fnhe_hashfun(daddr); 1330 1331 spin_lock_bh(&fnhe_lock); 1332 1333 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1334 lockdep_is_held(&fnhe_lock)); 1335 hash += hval; 1336 1337 fnhe_p = &hash->chain; 1338 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock)); 1339 while (fnhe) { 1340 if (fnhe->fnhe_daddr == daddr) { 1341 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected( 1342 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock))); 1343 /* set fnhe_daddr to 0 to ensure it won't bind with 1344 * new dsts in rt_bind_exception(). 1345 */ 1346 fnhe->fnhe_daddr = 0; 1347 fnhe_flush_routes(fnhe); 1348 kfree_rcu(fnhe, rcu); 1349 break; 1350 } 1351 fnhe_p = &fnhe->fnhe_next; 1352 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1353 lockdep_is_held(&fnhe_lock)); 1354 } 1355 1356 spin_unlock_bh(&fnhe_lock); 1357 } 1358 1359 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc, 1360 __be32 daddr) 1361 { 1362 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions); 1363 struct fib_nh_exception *fnhe; 1364 u32 hval; 1365 1366 if (!hash) 1367 return NULL; 1368 1369 hval = fnhe_hashfun(daddr); 1370 1371 for (fnhe = rcu_dereference(hash[hval].chain); fnhe; 1372 fnhe = rcu_dereference(fnhe->fnhe_next)) { 1373 if (fnhe->fnhe_daddr == daddr) { 1374 if (fnhe->fnhe_expires && 1375 time_after(jiffies, fnhe->fnhe_expires)) { 1376 ip_del_fnhe(nhc, daddr); 1377 break; 1378 } 1379 return fnhe; 1380 } 1381 } 1382 return NULL; 1383 } 1384 1385 /* MTU selection: 1386 * 1. mtu on route is locked - use it 1387 * 2. mtu from nexthop exception 1388 * 3. mtu from egress device 1389 */ 1390 1391 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr) 1392 { 1393 struct fib_nh_common *nhc = res->nhc; 1394 struct net_device *dev = nhc->nhc_dev; 1395 struct fib_info *fi = res->fi; 1396 u32 mtu = 0; 1397 1398 if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu || 1399 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU)) 1400 mtu = fi->fib_mtu; 1401 1402 if (likely(!mtu)) { 1403 struct fib_nh_exception *fnhe; 1404 1405 fnhe = find_exception(nhc, daddr); 1406 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires)) 1407 mtu = fnhe->fnhe_pmtu; 1408 } 1409 1410 if (likely(!mtu)) 1411 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU); 1412 1413 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu); 1414 } 1415 1416 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, 1417 __be32 daddr, const bool do_cache) 1418 { 1419 bool ret = false; 1420 1421 spin_lock_bh(&fnhe_lock); 1422 1423 if (daddr == fnhe->fnhe_daddr) { 1424 struct rtable __rcu **porig; 1425 struct rtable *orig; 1426 int genid = fnhe_genid(dev_net(rt->dst.dev)); 1427 1428 if (rt_is_input_route(rt)) 1429 porig = &fnhe->fnhe_rth_input; 1430 else 1431 porig = &fnhe->fnhe_rth_output; 1432 orig = rcu_dereference(*porig); 1433 1434 if (fnhe->fnhe_genid != genid) { 1435 fnhe->fnhe_genid = genid; 1436 fnhe->fnhe_gw = 0; 1437 fnhe->fnhe_pmtu = 0; 1438 fnhe->fnhe_expires = 0; 1439 fnhe->fnhe_mtu_locked = false; 1440 fnhe_flush_routes(fnhe); 1441 orig = NULL; 1442 } 1443 fill_route_from_fnhe(rt, fnhe); 1444 if (!rt->rt_gw4) { 1445 rt->rt_gw4 = daddr; 1446 rt->rt_gw_family = AF_INET; 1447 } 1448 1449 if (do_cache) { 1450 dst_hold(&rt->dst); 1451 rcu_assign_pointer(*porig, rt); 1452 if (orig) { 1453 dst_dev_put(&orig->dst); 1454 dst_release(&orig->dst); 1455 } 1456 ret = true; 1457 } 1458 1459 fnhe->fnhe_stamp = jiffies; 1460 } 1461 spin_unlock_bh(&fnhe_lock); 1462 1463 return ret; 1464 } 1465 1466 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt) 1467 { 1468 struct rtable *orig, *prev, **p; 1469 bool ret = true; 1470 1471 if (rt_is_input_route(rt)) { 1472 p = (struct rtable **)&nhc->nhc_rth_input; 1473 } else { 1474 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 1475 } 1476 orig = *p; 1477 1478 /* hold dst before doing cmpxchg() to avoid race condition 1479 * on this dst 1480 */ 1481 dst_hold(&rt->dst); 1482 prev = cmpxchg(p, orig, rt); 1483 if (prev == orig) { 1484 if (orig) { 1485 dst_dev_put(&orig->dst); 1486 dst_release(&orig->dst); 1487 } 1488 } else { 1489 dst_release(&rt->dst); 1490 ret = false; 1491 } 1492 1493 return ret; 1494 } 1495 1496 struct uncached_list { 1497 spinlock_t lock; 1498 struct list_head head; 1499 }; 1500 1501 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list); 1502 1503 void rt_add_uncached_list(struct rtable *rt) 1504 { 1505 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list); 1506 1507 rt->rt_uncached_list = ul; 1508 1509 spin_lock_bh(&ul->lock); 1510 list_add_tail(&rt->rt_uncached, &ul->head); 1511 spin_unlock_bh(&ul->lock); 1512 } 1513 1514 void rt_del_uncached_list(struct rtable *rt) 1515 { 1516 if (!list_empty(&rt->rt_uncached)) { 1517 struct uncached_list *ul = rt->rt_uncached_list; 1518 1519 spin_lock_bh(&ul->lock); 1520 list_del(&rt->rt_uncached); 1521 spin_unlock_bh(&ul->lock); 1522 } 1523 } 1524 1525 static void ipv4_dst_destroy(struct dst_entry *dst) 1526 { 1527 struct rtable *rt = (struct rtable *)dst; 1528 1529 ip_dst_metrics_put(dst); 1530 rt_del_uncached_list(rt); 1531 } 1532 1533 void rt_flush_dev(struct net_device *dev) 1534 { 1535 struct rtable *rt; 1536 int cpu; 1537 1538 for_each_possible_cpu(cpu) { 1539 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 1540 1541 spin_lock_bh(&ul->lock); 1542 list_for_each_entry(rt, &ul->head, rt_uncached) { 1543 if (rt->dst.dev != dev) 1544 continue; 1545 rt->dst.dev = blackhole_netdev; 1546 dev_hold(rt->dst.dev); 1547 dev_put(dev); 1548 } 1549 spin_unlock_bh(&ul->lock); 1550 } 1551 } 1552 1553 static bool rt_cache_valid(const struct rtable *rt) 1554 { 1555 return rt && 1556 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK && 1557 !rt_is_expired(rt); 1558 } 1559 1560 static void rt_set_nexthop(struct rtable *rt, __be32 daddr, 1561 const struct fib_result *res, 1562 struct fib_nh_exception *fnhe, 1563 struct fib_info *fi, u16 type, u32 itag, 1564 const bool do_cache) 1565 { 1566 bool cached = false; 1567 1568 if (fi) { 1569 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1570 1571 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) { 1572 rt->rt_gw_family = nhc->nhc_gw_family; 1573 /* only INET and INET6 are supported */ 1574 if (likely(nhc->nhc_gw_family == AF_INET)) 1575 rt->rt_gw4 = nhc->nhc_gw.ipv4; 1576 else 1577 rt->rt_gw6 = nhc->nhc_gw.ipv6; 1578 } 1579 1580 ip_dst_init_metrics(&rt->dst, fi->fib_metrics); 1581 1582 #ifdef CONFIG_IP_ROUTE_CLASSID 1583 if (nhc->nhc_family == AF_INET) { 1584 struct fib_nh *nh; 1585 1586 nh = container_of(nhc, struct fib_nh, nh_common); 1587 rt->dst.tclassid = nh->nh_tclassid; 1588 } 1589 #endif 1590 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 1591 if (unlikely(fnhe)) 1592 cached = rt_bind_exception(rt, fnhe, daddr, do_cache); 1593 else if (do_cache) 1594 cached = rt_cache_route(nhc, rt); 1595 if (unlikely(!cached)) { 1596 /* Routes we intend to cache in nexthop exception or 1597 * FIB nexthop have the DST_NOCACHE bit clear. 1598 * However, if we are unsuccessful at storing this 1599 * route into the cache we really need to set it. 1600 */ 1601 if (!rt->rt_gw4) { 1602 rt->rt_gw_family = AF_INET; 1603 rt->rt_gw4 = daddr; 1604 } 1605 rt_add_uncached_list(rt); 1606 } 1607 } else 1608 rt_add_uncached_list(rt); 1609 1610 #ifdef CONFIG_IP_ROUTE_CLASSID 1611 #ifdef CONFIG_IP_MULTIPLE_TABLES 1612 set_class_tag(rt, res->tclassid); 1613 #endif 1614 set_class_tag(rt, itag); 1615 #endif 1616 } 1617 1618 struct rtable *rt_dst_alloc(struct net_device *dev, 1619 unsigned int flags, u16 type, 1620 bool nopolicy, bool noxfrm, bool will_cache) 1621 { 1622 struct rtable *rt; 1623 1624 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1625 (will_cache ? 0 : DST_HOST) | 1626 (nopolicy ? DST_NOPOLICY : 0) | 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_gw_family = 0; 1638 rt->rt_gw4 = 0; 1639 INIT_LIST_HEAD(&rt->rt_uncached); 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, 1, 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 INIT_LIST_HEAD(&new_rt->rt_uncached); 1671 1672 new_rt->dst.flags |= DST_HOST; 1673 new_rt->dst.input = rt->dst.input; 1674 new_rt->dst.output = rt->dst.output; 1675 new_rt->dst.error = rt->dst.error; 1676 new_rt->dst.lastuse = jiffies; 1677 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate); 1678 } 1679 return new_rt; 1680 } 1681 EXPORT_SYMBOL(rt_dst_clone); 1682 1683 /* called in rcu_read_lock() section */ 1684 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1685 u8 tos, struct net_device *dev, 1686 struct in_device *in_dev, u32 *itag) 1687 { 1688 int err; 1689 1690 /* Primary sanity checks. */ 1691 if (!in_dev) 1692 return -EINVAL; 1693 1694 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) || 1695 skb->protocol != htons(ETH_P_IP)) 1696 return -EINVAL; 1697 1698 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev)) 1699 return -EINVAL; 1700 1701 if (ipv4_is_zeronet(saddr)) { 1702 if (!ipv4_is_local_multicast(daddr) && 1703 ip_hdr(skb)->protocol != IPPROTO_IGMP) 1704 return -EINVAL; 1705 } else { 1706 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 1707 in_dev, itag); 1708 if (err < 0) 1709 return err; 1710 } 1711 return 0; 1712 } 1713 1714 /* called in rcu_read_lock() section */ 1715 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1716 u8 tos, struct net_device *dev, int our) 1717 { 1718 struct in_device *in_dev = __in_dev_get_rcu(dev); 1719 unsigned int flags = RTCF_MULTICAST; 1720 struct rtable *rth; 1721 u32 itag = 0; 1722 int err; 1723 1724 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag); 1725 if (err) 1726 return err; 1727 1728 if (our) 1729 flags |= RTCF_LOCAL; 1730 1731 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST, 1732 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, false); 1733 if (!rth) 1734 return -ENOBUFS; 1735 1736 #ifdef CONFIG_IP_ROUTE_CLASSID 1737 rth->dst.tclassid = itag; 1738 #endif 1739 rth->dst.output = ip_rt_bug; 1740 rth->rt_is_input= 1; 1741 1742 #ifdef CONFIG_IP_MROUTE 1743 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1744 rth->dst.input = ip_mr_input; 1745 #endif 1746 RT_CACHE_STAT_INC(in_slow_mc); 1747 1748 skb_dst_set(skb, &rth->dst); 1749 return 0; 1750 } 1751 1752 1753 static void ip_handle_martian_source(struct net_device *dev, 1754 struct in_device *in_dev, 1755 struct sk_buff *skb, 1756 __be32 daddr, 1757 __be32 saddr) 1758 { 1759 RT_CACHE_STAT_INC(in_martian_src); 1760 #ifdef CONFIG_IP_ROUTE_VERBOSE 1761 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1762 /* 1763 * RFC1812 recommendation, if source is martian, 1764 * the only hint is MAC header. 1765 */ 1766 pr_warn("martian source %pI4 from %pI4, on dev %s\n", 1767 &daddr, &saddr, dev->name); 1768 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1769 print_hex_dump(KERN_WARNING, "ll header: ", 1770 DUMP_PREFIX_OFFSET, 16, 1, 1771 skb_mac_header(skb), 1772 dev->hard_header_len, false); 1773 } 1774 } 1775 #endif 1776 } 1777 1778 /* called in rcu_read_lock() section */ 1779 static int __mkroute_input(struct sk_buff *skb, 1780 const struct fib_result *res, 1781 struct in_device *in_dev, 1782 __be32 daddr, __be32 saddr, u32 tos) 1783 { 1784 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1785 struct net_device *dev = nhc->nhc_dev; 1786 struct fib_nh_exception *fnhe; 1787 struct rtable *rth; 1788 int err; 1789 struct in_device *out_dev; 1790 bool do_cache; 1791 u32 itag = 0; 1792 1793 /* get a working reference to the output device */ 1794 out_dev = __in_dev_get_rcu(dev); 1795 if (!out_dev) { 1796 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n"); 1797 return -EINVAL; 1798 } 1799 1800 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res), 1801 in_dev->dev, in_dev, &itag); 1802 if (err < 0) { 1803 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1804 saddr); 1805 1806 goto cleanup; 1807 } 1808 1809 do_cache = res->fi && !itag; 1810 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) && 1811 skb->protocol == htons(ETH_P_IP)) { 1812 __be32 gw; 1813 1814 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0; 1815 if (IN_DEV_SHARED_MEDIA(out_dev) || 1816 inet_addr_onlink(out_dev, saddr, gw)) 1817 IPCB(skb)->flags |= IPSKB_DOREDIRECT; 1818 } 1819 1820 if (skb->protocol != htons(ETH_P_IP)) { 1821 /* Not IP (i.e. ARP). Do not create route, if it is 1822 * invalid for proxy arp. DNAT routes are always valid. 1823 * 1824 * Proxy arp feature have been extended to allow, ARP 1825 * replies back to the same interface, to support 1826 * Private VLAN switch technologies. See arp.c. 1827 */ 1828 if (out_dev == in_dev && 1829 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 1830 err = -EINVAL; 1831 goto cleanup; 1832 } 1833 } 1834 1835 fnhe = find_exception(nhc, daddr); 1836 if (do_cache) { 1837 if (fnhe) 1838 rth = rcu_dereference(fnhe->fnhe_rth_input); 1839 else 1840 rth = rcu_dereference(nhc->nhc_rth_input); 1841 if (rt_cache_valid(rth)) { 1842 skb_dst_set_noref(skb, &rth->dst); 1843 goto out; 1844 } 1845 } 1846 1847 rth = rt_dst_alloc(out_dev->dev, 0, res->type, 1848 IN_DEV_CONF_GET(in_dev, NOPOLICY), 1849 IN_DEV_CONF_GET(out_dev, NOXFRM), do_cache); 1850 if (!rth) { 1851 err = -ENOBUFS; 1852 goto cleanup; 1853 } 1854 1855 rth->rt_is_input = 1; 1856 RT_CACHE_STAT_INC(in_slow_tot); 1857 1858 rth->dst.input = ip_forward; 1859 1860 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag, 1861 do_cache); 1862 lwtunnel_set_redirect(&rth->dst); 1863 skb_dst_set(skb, &rth->dst); 1864 out: 1865 err = 0; 1866 cleanup: 1867 return err; 1868 } 1869 1870 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1871 /* To make ICMP packets follow the right flow, the multipath hash is 1872 * calculated from the inner IP addresses. 1873 */ 1874 static void ip_multipath_l3_keys(const struct sk_buff *skb, 1875 struct flow_keys *hash_keys) 1876 { 1877 const struct iphdr *outer_iph = ip_hdr(skb); 1878 const struct iphdr *key_iph = outer_iph; 1879 const struct iphdr *inner_iph; 1880 const struct icmphdr *icmph; 1881 struct iphdr _inner_iph; 1882 struct icmphdr _icmph; 1883 1884 if (likely(outer_iph->protocol != IPPROTO_ICMP)) 1885 goto out; 1886 1887 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0)) 1888 goto out; 1889 1890 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph), 1891 &_icmph); 1892 if (!icmph) 1893 goto out; 1894 1895 if (icmph->type != ICMP_DEST_UNREACH && 1896 icmph->type != ICMP_REDIRECT && 1897 icmph->type != ICMP_TIME_EXCEEDED && 1898 icmph->type != ICMP_PARAMETERPROB) 1899 goto out; 1900 1901 inner_iph = skb_header_pointer(skb, 1902 outer_iph->ihl * 4 + sizeof(_icmph), 1903 sizeof(_inner_iph), &_inner_iph); 1904 if (!inner_iph) 1905 goto out; 1906 1907 key_iph = inner_iph; 1908 out: 1909 hash_keys->addrs.v4addrs.src = key_iph->saddr; 1910 hash_keys->addrs.v4addrs.dst = key_iph->daddr; 1911 } 1912 1913 /* if skb is set it will be used and fl4 can be NULL */ 1914 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4, 1915 const struct sk_buff *skb, struct flow_keys *flkeys) 1916 { 1917 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0; 1918 struct flow_keys hash_keys; 1919 u32 mhash; 1920 1921 switch (net->ipv4.sysctl_fib_multipath_hash_policy) { 1922 case 0: 1923 memset(&hash_keys, 0, sizeof(hash_keys)); 1924 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1925 if (skb) { 1926 ip_multipath_l3_keys(skb, &hash_keys); 1927 } else { 1928 hash_keys.addrs.v4addrs.src = fl4->saddr; 1929 hash_keys.addrs.v4addrs.dst = fl4->daddr; 1930 } 1931 break; 1932 case 1: 1933 /* skb is currently provided only when forwarding */ 1934 if (skb) { 1935 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 1936 struct flow_keys keys; 1937 1938 /* short-circuit if we already have L4 hash present */ 1939 if (skb->l4_hash) 1940 return skb_get_hash_raw(skb) >> 1; 1941 1942 memset(&hash_keys, 0, sizeof(hash_keys)); 1943 1944 if (!flkeys) { 1945 skb_flow_dissect_flow_keys(skb, &keys, flag); 1946 flkeys = &keys; 1947 } 1948 1949 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1950 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src; 1951 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst; 1952 hash_keys.ports.src = flkeys->ports.src; 1953 hash_keys.ports.dst = flkeys->ports.dst; 1954 hash_keys.basic.ip_proto = flkeys->basic.ip_proto; 1955 } else { 1956 memset(&hash_keys, 0, sizeof(hash_keys)); 1957 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1958 hash_keys.addrs.v4addrs.src = fl4->saddr; 1959 hash_keys.addrs.v4addrs.dst = fl4->daddr; 1960 hash_keys.ports.src = fl4->fl4_sport; 1961 hash_keys.ports.dst = fl4->fl4_dport; 1962 hash_keys.basic.ip_proto = fl4->flowi4_proto; 1963 } 1964 break; 1965 case 2: 1966 memset(&hash_keys, 0, sizeof(hash_keys)); 1967 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1968 /* skb is currently provided only when forwarding */ 1969 if (skb) { 1970 struct flow_keys keys; 1971 1972 skb_flow_dissect_flow_keys(skb, &keys, 0); 1973 1974 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 1975 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 1976 } else { 1977 /* Same as case 0 */ 1978 hash_keys.addrs.v4addrs.src = fl4->saddr; 1979 hash_keys.addrs.v4addrs.dst = fl4->daddr; 1980 } 1981 break; 1982 } 1983 mhash = flow_hash_from_keys(&hash_keys); 1984 1985 if (multipath_hash) 1986 mhash = jhash_2words(mhash, multipath_hash, 0); 1987 1988 return mhash >> 1; 1989 } 1990 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 1991 1992 static int ip_mkroute_input(struct sk_buff *skb, 1993 struct fib_result *res, 1994 struct in_device *in_dev, 1995 __be32 daddr, __be32 saddr, u32 tos, 1996 struct flow_keys *hkeys) 1997 { 1998 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1999 if (res->fi && fib_info_num_path(res->fi) > 1) { 2000 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys); 2001 2002 fib_select_multipath(res, h); 2003 } 2004 #endif 2005 2006 /* create a routing cache entry */ 2007 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos); 2008 } 2009 2010 /* 2011 * NOTE. We drop all the packets that has local source 2012 * addresses, because every properly looped back packet 2013 * must have correct destination already attached by output routine. 2014 * 2015 * Such approach solves two big problems: 2016 * 1. Not simplex devices are handled properly. 2017 * 2. IP spoofing attempts are filtered with 100% of guarantee. 2018 * called with rcu_read_lock() 2019 */ 2020 2021 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2022 u8 tos, struct net_device *dev, 2023 struct fib_result *res) 2024 { 2025 struct in_device *in_dev = __in_dev_get_rcu(dev); 2026 struct flow_keys *flkeys = NULL, _flkeys; 2027 struct net *net = dev_net(dev); 2028 struct ip_tunnel_info *tun_info; 2029 int err = -EINVAL; 2030 unsigned int flags = 0; 2031 u32 itag = 0; 2032 struct rtable *rth; 2033 struct flowi4 fl4; 2034 bool do_cache = true; 2035 2036 /* IP on this device is disabled. */ 2037 2038 if (!in_dev) 2039 goto out; 2040 2041 /* Check for the most weird martians, which can be not detected 2042 by fib_lookup. 2043 */ 2044 2045 tun_info = skb_tunnel_info(skb); 2046 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX)) 2047 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id; 2048 else 2049 fl4.flowi4_tun_key.tun_id = 0; 2050 skb_dst_drop(skb); 2051 2052 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 2053 goto martian_source; 2054 2055 res->fi = NULL; 2056 res->table = NULL; 2057 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 2058 goto brd_input; 2059 2060 /* Accept zero addresses only to limited broadcast; 2061 * I even do not know to fix it or not. Waiting for complains :-) 2062 */ 2063 if (ipv4_is_zeronet(saddr)) 2064 goto martian_source; 2065 2066 if (ipv4_is_zeronet(daddr)) 2067 goto martian_destination; 2068 2069 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(), 2070 * and call it once if daddr or/and saddr are loopback addresses 2071 */ 2072 if (ipv4_is_loopback(daddr)) { 2073 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2074 goto martian_destination; 2075 } else if (ipv4_is_loopback(saddr)) { 2076 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2077 goto martian_source; 2078 } 2079 2080 /* 2081 * Now we are ready to route packet. 2082 */ 2083 fl4.flowi4_oif = 0; 2084 fl4.flowi4_iif = dev->ifindex; 2085 fl4.flowi4_mark = skb->mark; 2086 fl4.flowi4_tos = tos; 2087 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 2088 fl4.flowi4_flags = 0; 2089 fl4.daddr = daddr; 2090 fl4.saddr = saddr; 2091 fl4.flowi4_uid = sock_net_uid(net, NULL); 2092 2093 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) { 2094 flkeys = &_flkeys; 2095 } else { 2096 fl4.flowi4_proto = 0; 2097 fl4.fl4_sport = 0; 2098 fl4.fl4_dport = 0; 2099 } 2100 2101 err = fib_lookup(net, &fl4, res, 0); 2102 if (err != 0) { 2103 if (!IN_DEV_FORWARD(in_dev)) 2104 err = -EHOSTUNREACH; 2105 goto no_route; 2106 } 2107 2108 if (res->type == RTN_BROADCAST) { 2109 if (IN_DEV_BFORWARD(in_dev)) 2110 goto make_route; 2111 /* not do cache if bc_forwarding is enabled */ 2112 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING)) 2113 do_cache = false; 2114 goto brd_input; 2115 } 2116 2117 if (res->type == RTN_LOCAL) { 2118 err = fib_validate_source(skb, saddr, daddr, tos, 2119 0, dev, in_dev, &itag); 2120 if (err < 0) 2121 goto martian_source; 2122 goto local_input; 2123 } 2124 2125 if (!IN_DEV_FORWARD(in_dev)) { 2126 err = -EHOSTUNREACH; 2127 goto no_route; 2128 } 2129 if (res->type != RTN_UNICAST) 2130 goto martian_destination; 2131 2132 make_route: 2133 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys); 2134 out: return err; 2135 2136 brd_input: 2137 if (skb->protocol != htons(ETH_P_IP)) 2138 goto e_inval; 2139 2140 if (!ipv4_is_zeronet(saddr)) { 2141 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 2142 in_dev, &itag); 2143 if (err < 0) 2144 goto martian_source; 2145 } 2146 flags |= RTCF_BROADCAST; 2147 res->type = RTN_BROADCAST; 2148 RT_CACHE_STAT_INC(in_brd); 2149 2150 local_input: 2151 do_cache &= res->fi && !itag; 2152 if (do_cache) { 2153 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2154 2155 rth = rcu_dereference(nhc->nhc_rth_input); 2156 if (rt_cache_valid(rth)) { 2157 skb_dst_set_noref(skb, &rth->dst); 2158 err = 0; 2159 goto out; 2160 } 2161 } 2162 2163 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev, 2164 flags | RTCF_LOCAL, res->type, 2165 IN_DEV_CONF_GET(in_dev, NOPOLICY), false, do_cache); 2166 if (!rth) 2167 goto e_nobufs; 2168 2169 rth->dst.output= ip_rt_bug; 2170 #ifdef CONFIG_IP_ROUTE_CLASSID 2171 rth->dst.tclassid = itag; 2172 #endif 2173 rth->rt_is_input = 1; 2174 2175 RT_CACHE_STAT_INC(in_slow_tot); 2176 if (res->type == RTN_UNREACHABLE) { 2177 rth->dst.input= ip_error; 2178 rth->dst.error= -err; 2179 rth->rt_flags &= ~RTCF_LOCAL; 2180 } 2181 2182 if (do_cache) { 2183 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2184 2185 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 2186 if (lwtunnel_input_redirect(rth->dst.lwtstate)) { 2187 WARN_ON(rth->dst.input == lwtunnel_input); 2188 rth->dst.lwtstate->orig_input = rth->dst.input; 2189 rth->dst.input = lwtunnel_input; 2190 } 2191 2192 if (unlikely(!rt_cache_route(nhc, rth))) 2193 rt_add_uncached_list(rth); 2194 } 2195 skb_dst_set(skb, &rth->dst); 2196 err = 0; 2197 goto out; 2198 2199 no_route: 2200 RT_CACHE_STAT_INC(in_no_route); 2201 res->type = RTN_UNREACHABLE; 2202 res->fi = NULL; 2203 res->table = NULL; 2204 goto local_input; 2205 2206 /* 2207 * Do not cache martian addresses: they should be logged (RFC1812) 2208 */ 2209 martian_destination: 2210 RT_CACHE_STAT_INC(in_martian_dst); 2211 #ifdef CONFIG_IP_ROUTE_VERBOSE 2212 if (IN_DEV_LOG_MARTIANS(in_dev)) 2213 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n", 2214 &daddr, &saddr, dev->name); 2215 #endif 2216 2217 e_inval: 2218 err = -EINVAL; 2219 goto out; 2220 2221 e_nobufs: 2222 err = -ENOBUFS; 2223 goto out; 2224 2225 martian_source: 2226 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2227 goto out; 2228 } 2229 2230 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2231 u8 tos, struct net_device *dev) 2232 { 2233 struct fib_result res; 2234 int err; 2235 2236 tos &= IPTOS_RT_MASK; 2237 rcu_read_lock(); 2238 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res); 2239 rcu_read_unlock(); 2240 2241 return err; 2242 } 2243 EXPORT_SYMBOL(ip_route_input_noref); 2244 2245 /* called with rcu_read_lock held */ 2246 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2247 u8 tos, struct net_device *dev, struct fib_result *res) 2248 { 2249 /* Multicast recognition logic is moved from route cache to here. 2250 The problem was that too many Ethernet cards have broken/missing 2251 hardware multicast filters :-( As result the host on multicasting 2252 network acquires a lot of useless route cache entries, sort of 2253 SDR messages from all the world. Now we try to get rid of them. 2254 Really, provided software IP multicast filter is organized 2255 reasonably (at least, hashed), it does not result in a slowdown 2256 comparing with route cache reject entries. 2257 Note, that multicast routers are not affected, because 2258 route cache entry is created eventually. 2259 */ 2260 if (ipv4_is_multicast(daddr)) { 2261 struct in_device *in_dev = __in_dev_get_rcu(dev); 2262 int our = 0; 2263 int err = -EINVAL; 2264 2265 if (!in_dev) 2266 return err; 2267 our = ip_check_mc_rcu(in_dev, daddr, saddr, 2268 ip_hdr(skb)->protocol); 2269 2270 /* check l3 master if no match yet */ 2271 if (!our && netif_is_l3_slave(dev)) { 2272 struct in_device *l3_in_dev; 2273 2274 l3_in_dev = __in_dev_get_rcu(skb->dev); 2275 if (l3_in_dev) 2276 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr, 2277 ip_hdr(skb)->protocol); 2278 } 2279 2280 if (our 2281 #ifdef CONFIG_IP_MROUTE 2282 || 2283 (!ipv4_is_local_multicast(daddr) && 2284 IN_DEV_MFORWARD(in_dev)) 2285 #endif 2286 ) { 2287 err = ip_route_input_mc(skb, daddr, saddr, 2288 tos, dev, our); 2289 } 2290 return err; 2291 } 2292 2293 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res); 2294 } 2295 2296 /* called with rcu_read_lock() */ 2297 static struct rtable *__mkroute_output(const struct fib_result *res, 2298 const struct flowi4 *fl4, int orig_oif, 2299 struct net_device *dev_out, 2300 unsigned int flags) 2301 { 2302 struct fib_info *fi = res->fi; 2303 struct fib_nh_exception *fnhe; 2304 struct in_device *in_dev; 2305 u16 type = res->type; 2306 struct rtable *rth; 2307 bool do_cache; 2308 2309 in_dev = __in_dev_get_rcu(dev_out); 2310 if (!in_dev) 2311 return ERR_PTR(-EINVAL); 2312 2313 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 2314 if (ipv4_is_loopback(fl4->saddr) && 2315 !(dev_out->flags & IFF_LOOPBACK) && 2316 !netif_is_l3_master(dev_out)) 2317 return ERR_PTR(-EINVAL); 2318 2319 if (ipv4_is_lbcast(fl4->daddr)) 2320 type = RTN_BROADCAST; 2321 else if (ipv4_is_multicast(fl4->daddr)) 2322 type = RTN_MULTICAST; 2323 else if (ipv4_is_zeronet(fl4->daddr)) 2324 return ERR_PTR(-EINVAL); 2325 2326 if (dev_out->flags & IFF_LOOPBACK) 2327 flags |= RTCF_LOCAL; 2328 2329 do_cache = true; 2330 if (type == RTN_BROADCAST) { 2331 flags |= RTCF_BROADCAST | RTCF_LOCAL; 2332 fi = NULL; 2333 } else if (type == RTN_MULTICAST) { 2334 flags |= RTCF_MULTICAST | RTCF_LOCAL; 2335 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr, 2336 fl4->flowi4_proto)) 2337 flags &= ~RTCF_LOCAL; 2338 else 2339 do_cache = false; 2340 /* If multicast route do not exist use 2341 * default one, but do not gateway in this case. 2342 * Yes, it is hack. 2343 */ 2344 if (fi && res->prefixlen < 4) 2345 fi = NULL; 2346 } else if ((type == RTN_LOCAL) && (orig_oif != 0) && 2347 (orig_oif != dev_out->ifindex)) { 2348 /* For local routes that require a particular output interface 2349 * we do not want to cache the result. Caching the result 2350 * causes incorrect behaviour when there are multiple source 2351 * addresses on the interface, the end result being that if the 2352 * intended recipient is waiting on that interface for the 2353 * packet he won't receive it because it will be delivered on 2354 * the loopback interface and the IP_PKTINFO ipi_ifindex will 2355 * be set to the loopback interface as well. 2356 */ 2357 do_cache = false; 2358 } 2359 2360 fnhe = NULL; 2361 do_cache &= fi != NULL; 2362 if (fi) { 2363 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2364 struct rtable __rcu **prth; 2365 2366 fnhe = find_exception(nhc, fl4->daddr); 2367 if (!do_cache) 2368 goto add; 2369 if (fnhe) { 2370 prth = &fnhe->fnhe_rth_output; 2371 } else { 2372 if (unlikely(fl4->flowi4_flags & 2373 FLOWI_FLAG_KNOWN_NH && 2374 !(nhc->nhc_gw_family && 2375 nhc->nhc_scope == RT_SCOPE_LINK))) { 2376 do_cache = false; 2377 goto add; 2378 } 2379 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 2380 } 2381 rth = rcu_dereference(*prth); 2382 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst)) 2383 return rth; 2384 } 2385 2386 add: 2387 rth = rt_dst_alloc(dev_out, flags, type, 2388 IN_DEV_CONF_GET(in_dev, NOPOLICY), 2389 IN_DEV_CONF_GET(in_dev, NOXFRM), 2390 do_cache); 2391 if (!rth) 2392 return ERR_PTR(-ENOBUFS); 2393 2394 rth->rt_iif = orig_oif; 2395 2396 RT_CACHE_STAT_INC(out_slow_tot); 2397 2398 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 2399 if (flags & RTCF_LOCAL && 2400 !(dev_out->flags & IFF_LOOPBACK)) { 2401 rth->dst.output = ip_mc_output; 2402 RT_CACHE_STAT_INC(out_slow_mc); 2403 } 2404 #ifdef CONFIG_IP_MROUTE 2405 if (type == RTN_MULTICAST) { 2406 if (IN_DEV_MFORWARD(in_dev) && 2407 !ipv4_is_local_multicast(fl4->daddr)) { 2408 rth->dst.input = ip_mr_input; 2409 rth->dst.output = ip_mc_output; 2410 } 2411 } 2412 #endif 2413 } 2414 2415 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache); 2416 lwtunnel_set_redirect(&rth->dst); 2417 2418 return rth; 2419 } 2420 2421 /* 2422 * Major route resolver routine. 2423 */ 2424 2425 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4, 2426 const struct sk_buff *skb) 2427 { 2428 __u8 tos = RT_FL_TOS(fl4); 2429 struct fib_result res = { 2430 .type = RTN_UNSPEC, 2431 .fi = NULL, 2432 .table = NULL, 2433 .tclassid = 0, 2434 }; 2435 struct rtable *rth; 2436 2437 fl4->flowi4_iif = LOOPBACK_IFINDEX; 2438 fl4->flowi4_tos = tos & IPTOS_RT_MASK; 2439 fl4->flowi4_scope = ((tos & RTO_ONLINK) ? 2440 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE); 2441 2442 rcu_read_lock(); 2443 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb); 2444 rcu_read_unlock(); 2445 2446 return rth; 2447 } 2448 EXPORT_SYMBOL_GPL(ip_route_output_key_hash); 2449 2450 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4, 2451 struct fib_result *res, 2452 const struct sk_buff *skb) 2453 { 2454 struct net_device *dev_out = NULL; 2455 int orig_oif = fl4->flowi4_oif; 2456 unsigned int flags = 0; 2457 struct rtable *rth; 2458 int err = -ENETUNREACH; 2459 2460 if (fl4->saddr) { 2461 rth = ERR_PTR(-EINVAL); 2462 if (ipv4_is_multicast(fl4->saddr) || 2463 ipv4_is_lbcast(fl4->saddr) || 2464 ipv4_is_zeronet(fl4->saddr)) 2465 goto out; 2466 2467 /* I removed check for oif == dev_out->oif here. 2468 It was wrong for two reasons: 2469 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 2470 is assigned to multiple interfaces. 2471 2. Moreover, we are allowed to send packets with saddr 2472 of another iface. --ANK 2473 */ 2474 2475 if (fl4->flowi4_oif == 0 && 2476 (ipv4_is_multicast(fl4->daddr) || 2477 ipv4_is_lbcast(fl4->daddr))) { 2478 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2479 dev_out = __ip_dev_find(net, fl4->saddr, false); 2480 if (!dev_out) 2481 goto out; 2482 2483 /* Special hack: user can direct multicasts 2484 and limited broadcast via necessary interface 2485 without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 2486 This hack is not just for fun, it allows 2487 vic,vat and friends to work. 2488 They bind socket to loopback, set ttl to zero 2489 and expect that it will work. 2490 From the viewpoint of routing cache they are broken, 2491 because we are not allowed to build multicast path 2492 with loopback source addr (look, routing cache 2493 cannot know, that ttl is zero, so that packet 2494 will not leave this host and route is valid). 2495 Luckily, this hack is good workaround. 2496 */ 2497 2498 fl4->flowi4_oif = dev_out->ifindex; 2499 goto make_route; 2500 } 2501 2502 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 2503 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2504 if (!__ip_dev_find(net, fl4->saddr, false)) 2505 goto out; 2506 } 2507 } 2508 2509 2510 if (fl4->flowi4_oif) { 2511 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif); 2512 rth = ERR_PTR(-ENODEV); 2513 if (!dev_out) 2514 goto out; 2515 2516 /* RACE: Check return value of inet_select_addr instead. */ 2517 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 2518 rth = ERR_PTR(-ENETUNREACH); 2519 goto out; 2520 } 2521 if (ipv4_is_local_multicast(fl4->daddr) || 2522 ipv4_is_lbcast(fl4->daddr) || 2523 fl4->flowi4_proto == IPPROTO_IGMP) { 2524 if (!fl4->saddr) 2525 fl4->saddr = inet_select_addr(dev_out, 0, 2526 RT_SCOPE_LINK); 2527 goto make_route; 2528 } 2529 if (!fl4->saddr) { 2530 if (ipv4_is_multicast(fl4->daddr)) 2531 fl4->saddr = inet_select_addr(dev_out, 0, 2532 fl4->flowi4_scope); 2533 else if (!fl4->daddr) 2534 fl4->saddr = inet_select_addr(dev_out, 0, 2535 RT_SCOPE_HOST); 2536 } 2537 } 2538 2539 if (!fl4->daddr) { 2540 fl4->daddr = fl4->saddr; 2541 if (!fl4->daddr) 2542 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK); 2543 dev_out = net->loopback_dev; 2544 fl4->flowi4_oif = LOOPBACK_IFINDEX; 2545 res->type = RTN_LOCAL; 2546 flags |= RTCF_LOCAL; 2547 goto make_route; 2548 } 2549 2550 err = fib_lookup(net, fl4, res, 0); 2551 if (err) { 2552 res->fi = NULL; 2553 res->table = NULL; 2554 if (fl4->flowi4_oif && 2555 (ipv4_is_multicast(fl4->daddr) || 2556 !netif_index_is_l3_master(net, fl4->flowi4_oif))) { 2557 /* Apparently, routing tables are wrong. Assume, 2558 that the destination is on link. 2559 2560 WHY? DW. 2561 Because we are allowed to send to iface 2562 even if it has NO routes and NO assigned 2563 addresses. When oif is specified, routing 2564 tables are looked up with only one purpose: 2565 to catch if destination is gatewayed, rather than 2566 direct. Moreover, if MSG_DONTROUTE is set, 2567 we send packet, ignoring both routing tables 2568 and ifaddr state. --ANK 2569 2570 2571 We could make it even if oif is unknown, 2572 likely IPv6, but we do not. 2573 */ 2574 2575 if (fl4->saddr == 0) 2576 fl4->saddr = inet_select_addr(dev_out, 0, 2577 RT_SCOPE_LINK); 2578 res->type = RTN_UNICAST; 2579 goto make_route; 2580 } 2581 rth = ERR_PTR(err); 2582 goto out; 2583 } 2584 2585 if (res->type == RTN_LOCAL) { 2586 if (!fl4->saddr) { 2587 if (res->fi->fib_prefsrc) 2588 fl4->saddr = res->fi->fib_prefsrc; 2589 else 2590 fl4->saddr = fl4->daddr; 2591 } 2592 2593 /* L3 master device is the loopback for that domain */ 2594 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? : 2595 net->loopback_dev; 2596 2597 /* make sure orig_oif points to fib result device even 2598 * though packet rx/tx happens over loopback or l3mdev 2599 */ 2600 orig_oif = FIB_RES_OIF(*res); 2601 2602 fl4->flowi4_oif = dev_out->ifindex; 2603 flags |= RTCF_LOCAL; 2604 goto make_route; 2605 } 2606 2607 fib_select_path(net, res, fl4, skb); 2608 2609 dev_out = FIB_RES_DEV(*res); 2610 fl4->flowi4_oif = dev_out->ifindex; 2611 2612 2613 make_route: 2614 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags); 2615 2616 out: 2617 return rth; 2618 } 2619 2620 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie) 2621 { 2622 return NULL; 2623 } 2624 2625 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst) 2626 { 2627 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU); 2628 2629 return mtu ? : dst->dev->mtu; 2630 } 2631 2632 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk, 2633 struct sk_buff *skb, u32 mtu) 2634 { 2635 } 2636 2637 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk, 2638 struct sk_buff *skb) 2639 { 2640 } 2641 2642 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst, 2643 unsigned long old) 2644 { 2645 return NULL; 2646 } 2647 2648 static struct dst_ops ipv4_dst_blackhole_ops = { 2649 .family = AF_INET, 2650 .check = ipv4_blackhole_dst_check, 2651 .mtu = ipv4_blackhole_mtu, 2652 .default_advmss = ipv4_default_advmss, 2653 .update_pmtu = ipv4_rt_blackhole_update_pmtu, 2654 .redirect = ipv4_rt_blackhole_redirect, 2655 .cow_metrics = ipv4_rt_blackhole_cow_metrics, 2656 .neigh_lookup = ipv4_neigh_lookup, 2657 }; 2658 2659 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2660 { 2661 struct rtable *ort = (struct rtable *) dst_orig; 2662 struct rtable *rt; 2663 2664 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0); 2665 if (rt) { 2666 struct dst_entry *new = &rt->dst; 2667 2668 new->__use = 1; 2669 new->input = dst_discard; 2670 new->output = dst_discard_out; 2671 2672 new->dev = net->loopback_dev; 2673 if (new->dev) 2674 dev_hold(new->dev); 2675 2676 rt->rt_is_input = ort->rt_is_input; 2677 rt->rt_iif = ort->rt_iif; 2678 rt->rt_pmtu = ort->rt_pmtu; 2679 rt->rt_mtu_locked = ort->rt_mtu_locked; 2680 2681 rt->rt_genid = rt_genid_ipv4(net); 2682 rt->rt_flags = ort->rt_flags; 2683 rt->rt_type = ort->rt_type; 2684 rt->rt_gw_family = ort->rt_gw_family; 2685 if (rt->rt_gw_family == AF_INET) 2686 rt->rt_gw4 = ort->rt_gw4; 2687 else if (rt->rt_gw_family == AF_INET6) 2688 rt->rt_gw6 = ort->rt_gw6; 2689 2690 INIT_LIST_HEAD(&rt->rt_uncached); 2691 } 2692 2693 dst_release(dst_orig); 2694 2695 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2696 } 2697 2698 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2699 const struct sock *sk) 2700 { 2701 struct rtable *rt = __ip_route_output_key(net, flp4); 2702 2703 if (IS_ERR(rt)) 2704 return rt; 2705 2706 if (flp4->flowi4_proto) 2707 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst, 2708 flowi4_to_flowi(flp4), 2709 sk, 0); 2710 2711 return rt; 2712 } 2713 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2714 2715 /* called with rcu_read_lock held */ 2716 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, 2717 struct rtable *rt, u32 table_id, struct flowi4 *fl4, 2718 struct sk_buff *skb, u32 portid, u32 seq) 2719 { 2720 struct rtmsg *r; 2721 struct nlmsghdr *nlh; 2722 unsigned long expires = 0; 2723 u32 error; 2724 u32 metrics[RTAX_MAX]; 2725 2726 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), 0); 2727 if (!nlh) 2728 return -EMSGSIZE; 2729 2730 r = nlmsg_data(nlh); 2731 r->rtm_family = AF_INET; 2732 r->rtm_dst_len = 32; 2733 r->rtm_src_len = 0; 2734 r->rtm_tos = fl4 ? fl4->flowi4_tos : 0; 2735 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT; 2736 if (nla_put_u32(skb, RTA_TABLE, table_id)) 2737 goto nla_put_failure; 2738 r->rtm_type = rt->rt_type; 2739 r->rtm_scope = RT_SCOPE_UNIVERSE; 2740 r->rtm_protocol = RTPROT_UNSPEC; 2741 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 2742 if (rt->rt_flags & RTCF_NOTIFY) 2743 r->rtm_flags |= RTM_F_NOTIFY; 2744 if (IPCB(skb)->flags & IPSKB_DOREDIRECT) 2745 r->rtm_flags |= RTCF_DOREDIRECT; 2746 2747 if (nla_put_in_addr(skb, RTA_DST, dst)) 2748 goto nla_put_failure; 2749 if (src) { 2750 r->rtm_src_len = 32; 2751 if (nla_put_in_addr(skb, RTA_SRC, src)) 2752 goto nla_put_failure; 2753 } 2754 if (rt->dst.dev && 2755 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex)) 2756 goto nla_put_failure; 2757 #ifdef CONFIG_IP_ROUTE_CLASSID 2758 if (rt->dst.tclassid && 2759 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid)) 2760 goto nla_put_failure; 2761 #endif 2762 if (fl4 && !rt_is_input_route(rt) && 2763 fl4->saddr != src) { 2764 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr)) 2765 goto nla_put_failure; 2766 } 2767 if (rt->rt_gw_family == AF_INET && 2768 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) { 2769 goto nla_put_failure; 2770 } else if (rt->rt_gw_family == AF_INET6) { 2771 int alen = sizeof(struct in6_addr); 2772 struct nlattr *nla; 2773 struct rtvia *via; 2774 2775 nla = nla_reserve(skb, RTA_VIA, alen + 2); 2776 if (!nla) 2777 goto nla_put_failure; 2778 2779 via = nla_data(nla); 2780 via->rtvia_family = AF_INET6; 2781 memcpy(via->rtvia_addr, &rt->rt_gw6, alen); 2782 } 2783 2784 expires = rt->dst.expires; 2785 if (expires) { 2786 unsigned long now = jiffies; 2787 2788 if (time_before(now, expires)) 2789 expires -= now; 2790 else 2791 expires = 0; 2792 } 2793 2794 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics)); 2795 if (rt->rt_pmtu && expires) 2796 metrics[RTAX_MTU - 1] = rt->rt_pmtu; 2797 if (rt->rt_mtu_locked && expires) 2798 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU); 2799 if (rtnetlink_put_metrics(skb, metrics) < 0) 2800 goto nla_put_failure; 2801 2802 if (fl4) { 2803 if (fl4->flowi4_mark && 2804 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark)) 2805 goto nla_put_failure; 2806 2807 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) && 2808 nla_put_u32(skb, RTA_UID, 2809 from_kuid_munged(current_user_ns(), 2810 fl4->flowi4_uid))) 2811 goto nla_put_failure; 2812 2813 if (rt_is_input_route(rt)) { 2814 #ifdef CONFIG_IP_MROUTE 2815 if (ipv4_is_multicast(dst) && 2816 !ipv4_is_local_multicast(dst) && 2817 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) { 2818 int err = ipmr_get_route(net, skb, 2819 fl4->saddr, fl4->daddr, 2820 r, portid); 2821 2822 if (err <= 0) { 2823 if (err == 0) 2824 return 0; 2825 goto nla_put_failure; 2826 } 2827 } else 2828 #endif 2829 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif)) 2830 goto nla_put_failure; 2831 } 2832 } 2833 2834 error = rt->dst.error; 2835 2836 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0) 2837 goto nla_put_failure; 2838 2839 nlmsg_end(skb, nlh); 2840 return 0; 2841 2842 nla_put_failure: 2843 nlmsg_cancel(skb, nlh); 2844 return -EMSGSIZE; 2845 } 2846 2847 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb, 2848 struct netlink_callback *cb, u32 table_id, 2849 struct fnhe_hash_bucket *bucket, int genid, 2850 int *fa_index, int fa_start) 2851 { 2852 int i; 2853 2854 for (i = 0; i < FNHE_HASH_SIZE; i++) { 2855 struct fib_nh_exception *fnhe; 2856 2857 for (fnhe = rcu_dereference(bucket[i].chain); fnhe; 2858 fnhe = rcu_dereference(fnhe->fnhe_next)) { 2859 struct rtable *rt; 2860 int err; 2861 2862 if (*fa_index < fa_start) 2863 goto next; 2864 2865 if (fnhe->fnhe_genid != genid) 2866 goto next; 2867 2868 if (fnhe->fnhe_expires && 2869 time_after(jiffies, fnhe->fnhe_expires)) 2870 goto next; 2871 2872 rt = rcu_dereference(fnhe->fnhe_rth_input); 2873 if (!rt) 2874 rt = rcu_dereference(fnhe->fnhe_rth_output); 2875 if (!rt) 2876 goto next; 2877 2878 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt, 2879 table_id, NULL, skb, 2880 NETLINK_CB(cb->skb).portid, 2881 cb->nlh->nlmsg_seq); 2882 if (err) 2883 return err; 2884 next: 2885 (*fa_index)++; 2886 } 2887 } 2888 2889 return 0; 2890 } 2891 2892 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb, 2893 u32 table_id, struct fib_info *fi, 2894 int *fa_index, int fa_start) 2895 { 2896 struct net *net = sock_net(cb->skb->sk); 2897 int nhsel, genid = fnhe_genid(net); 2898 2899 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) { 2900 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel); 2901 struct fnhe_hash_bucket *bucket; 2902 int err; 2903 2904 if (nhc->nhc_flags & RTNH_F_DEAD) 2905 continue; 2906 2907 rcu_read_lock(); 2908 bucket = rcu_dereference(nhc->nhc_exceptions); 2909 err = 0; 2910 if (bucket) 2911 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket, 2912 genid, fa_index, fa_start); 2913 rcu_read_unlock(); 2914 if (err) 2915 return err; 2916 } 2917 2918 return 0; 2919 } 2920 2921 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst, 2922 u8 ip_proto, __be16 sport, 2923 __be16 dport) 2924 { 2925 struct sk_buff *skb; 2926 struct iphdr *iph; 2927 2928 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 2929 if (!skb) 2930 return NULL; 2931 2932 /* Reserve room for dummy headers, this skb can pass 2933 * through good chunk of routing engine. 2934 */ 2935 skb_reset_mac_header(skb); 2936 skb_reset_network_header(skb); 2937 skb->protocol = htons(ETH_P_IP); 2938 iph = skb_put(skb, sizeof(struct iphdr)); 2939 iph->protocol = ip_proto; 2940 iph->saddr = src; 2941 iph->daddr = dst; 2942 iph->version = 0x4; 2943 iph->frag_off = 0; 2944 iph->ihl = 0x5; 2945 skb_set_transport_header(skb, skb->len); 2946 2947 switch (iph->protocol) { 2948 case IPPROTO_UDP: { 2949 struct udphdr *udph; 2950 2951 udph = skb_put_zero(skb, sizeof(struct udphdr)); 2952 udph->source = sport; 2953 udph->dest = dport; 2954 udph->len = sizeof(struct udphdr); 2955 udph->check = 0; 2956 break; 2957 } 2958 case IPPROTO_TCP: { 2959 struct tcphdr *tcph; 2960 2961 tcph = skb_put_zero(skb, sizeof(struct tcphdr)); 2962 tcph->source = sport; 2963 tcph->dest = dport; 2964 tcph->doff = sizeof(struct tcphdr) / 4; 2965 tcph->rst = 1; 2966 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr), 2967 src, dst, 0); 2968 break; 2969 } 2970 case IPPROTO_ICMP: { 2971 struct icmphdr *icmph; 2972 2973 icmph = skb_put_zero(skb, sizeof(struct icmphdr)); 2974 icmph->type = ICMP_ECHO; 2975 icmph->code = 0; 2976 } 2977 } 2978 2979 return skb; 2980 } 2981 2982 static int inet_rtm_valid_getroute_req(struct sk_buff *skb, 2983 const struct nlmsghdr *nlh, 2984 struct nlattr **tb, 2985 struct netlink_ext_ack *extack) 2986 { 2987 struct rtmsg *rtm; 2988 int i, err; 2989 2990 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) { 2991 NL_SET_ERR_MSG(extack, 2992 "ipv4: Invalid header for route get request"); 2993 return -EINVAL; 2994 } 2995 2996 if (!netlink_strict_get_check(skb)) 2997 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX, 2998 rtm_ipv4_policy, extack); 2999 3000 rtm = nlmsg_data(nlh); 3001 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) || 3002 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) || 3003 rtm->rtm_table || rtm->rtm_protocol || 3004 rtm->rtm_scope || rtm->rtm_type) { 3005 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request"); 3006 return -EINVAL; 3007 } 3008 3009 if (rtm->rtm_flags & ~(RTM_F_NOTIFY | 3010 RTM_F_LOOKUP_TABLE | 3011 RTM_F_FIB_MATCH)) { 3012 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request"); 3013 return -EINVAL; 3014 } 3015 3016 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX, 3017 rtm_ipv4_policy, extack); 3018 if (err) 3019 return err; 3020 3021 if ((tb[RTA_SRC] && !rtm->rtm_src_len) || 3022 (tb[RTA_DST] && !rtm->rtm_dst_len)) { 3023 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4"); 3024 return -EINVAL; 3025 } 3026 3027 for (i = 0; i <= RTA_MAX; i++) { 3028 if (!tb[i]) 3029 continue; 3030 3031 switch (i) { 3032 case RTA_IIF: 3033 case RTA_OIF: 3034 case RTA_SRC: 3035 case RTA_DST: 3036 case RTA_IP_PROTO: 3037 case RTA_SPORT: 3038 case RTA_DPORT: 3039 case RTA_MARK: 3040 case RTA_UID: 3041 break; 3042 default: 3043 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request"); 3044 return -EINVAL; 3045 } 3046 } 3047 3048 return 0; 3049 } 3050 3051 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, 3052 struct netlink_ext_ack *extack) 3053 { 3054 struct net *net = sock_net(in_skb->sk); 3055 struct nlattr *tb[RTA_MAX+1]; 3056 u32 table_id = RT_TABLE_MAIN; 3057 __be16 sport = 0, dport = 0; 3058 struct fib_result res = {}; 3059 u8 ip_proto = IPPROTO_UDP; 3060 struct rtable *rt = NULL; 3061 struct sk_buff *skb; 3062 struct rtmsg *rtm; 3063 struct flowi4 fl4 = {}; 3064 __be32 dst = 0; 3065 __be32 src = 0; 3066 kuid_t uid; 3067 u32 iif; 3068 int err; 3069 int mark; 3070 3071 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack); 3072 if (err < 0) 3073 return err; 3074 3075 rtm = nlmsg_data(nlh); 3076 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0; 3077 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0; 3078 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0; 3079 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0; 3080 if (tb[RTA_UID]) 3081 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID])); 3082 else 3083 uid = (iif ? INVALID_UID : current_uid()); 3084 3085 if (tb[RTA_IP_PROTO]) { 3086 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO], 3087 &ip_proto, AF_INET, extack); 3088 if (err) 3089 return err; 3090 } 3091 3092 if (tb[RTA_SPORT]) 3093 sport = nla_get_be16(tb[RTA_SPORT]); 3094 3095 if (tb[RTA_DPORT]) 3096 dport = nla_get_be16(tb[RTA_DPORT]); 3097 3098 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport); 3099 if (!skb) 3100 return -ENOBUFS; 3101 3102 fl4.daddr = dst; 3103 fl4.saddr = src; 3104 fl4.flowi4_tos = rtm->rtm_tos; 3105 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0; 3106 fl4.flowi4_mark = mark; 3107 fl4.flowi4_uid = uid; 3108 if (sport) 3109 fl4.fl4_sport = sport; 3110 if (dport) 3111 fl4.fl4_dport = dport; 3112 fl4.flowi4_proto = ip_proto; 3113 3114 rcu_read_lock(); 3115 3116 if (iif) { 3117 struct net_device *dev; 3118 3119 dev = dev_get_by_index_rcu(net, iif); 3120 if (!dev) { 3121 err = -ENODEV; 3122 goto errout_rcu; 3123 } 3124 3125 fl4.flowi4_iif = iif; /* for rt_fill_info */ 3126 skb->dev = dev; 3127 skb->mark = mark; 3128 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos, 3129 dev, &res); 3130 3131 rt = skb_rtable(skb); 3132 if (err == 0 && rt->dst.error) 3133 err = -rt->dst.error; 3134 } else { 3135 fl4.flowi4_iif = LOOPBACK_IFINDEX; 3136 skb->dev = net->loopback_dev; 3137 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb); 3138 err = 0; 3139 if (IS_ERR(rt)) 3140 err = PTR_ERR(rt); 3141 else 3142 skb_dst_set(skb, &rt->dst); 3143 } 3144 3145 if (err) 3146 goto errout_rcu; 3147 3148 if (rtm->rtm_flags & RTM_F_NOTIFY) 3149 rt->rt_flags |= RTCF_NOTIFY; 3150 3151 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE) 3152 table_id = res.table ? res.table->tb_id : 0; 3153 3154 /* reset skb for netlink reply msg */ 3155 skb_trim(skb, 0); 3156 skb_reset_network_header(skb); 3157 skb_reset_transport_header(skb); 3158 skb_reset_mac_header(skb); 3159 3160 if (rtm->rtm_flags & RTM_F_FIB_MATCH) { 3161 if (!res.fi) { 3162 err = fib_props[res.type].error; 3163 if (!err) 3164 err = -EHOSTUNREACH; 3165 goto errout_rcu; 3166 } 3167 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid, 3168 nlh->nlmsg_seq, RTM_NEWROUTE, table_id, 3169 rt->rt_type, res.prefix, res.prefixlen, 3170 fl4.flowi4_tos, res.fi, 0); 3171 } else { 3172 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb, 3173 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq); 3174 } 3175 if (err < 0) 3176 goto errout_rcu; 3177 3178 rcu_read_unlock(); 3179 3180 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 3181 3182 errout_free: 3183 return err; 3184 errout_rcu: 3185 rcu_read_unlock(); 3186 kfree_skb(skb); 3187 goto errout_free; 3188 } 3189 3190 void ip_rt_multicast_event(struct in_device *in_dev) 3191 { 3192 rt_cache_flush(dev_net(in_dev->dev)); 3193 } 3194 3195 #ifdef CONFIG_SYSCTL 3196 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 3197 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 3198 static int ip_rt_gc_elasticity __read_mostly = 8; 3199 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU; 3200 3201 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write, 3202 void __user *buffer, 3203 size_t *lenp, loff_t *ppos) 3204 { 3205 struct net *net = (struct net *)__ctl->extra1; 3206 3207 if (write) { 3208 rt_cache_flush(net); 3209 fnhe_genid_bump(net); 3210 return 0; 3211 } 3212 3213 return -EINVAL; 3214 } 3215 3216 static struct ctl_table ipv4_route_table[] = { 3217 { 3218 .procname = "gc_thresh", 3219 .data = &ipv4_dst_ops.gc_thresh, 3220 .maxlen = sizeof(int), 3221 .mode = 0644, 3222 .proc_handler = proc_dointvec, 3223 }, 3224 { 3225 .procname = "max_size", 3226 .data = &ip_rt_max_size, 3227 .maxlen = sizeof(int), 3228 .mode = 0644, 3229 .proc_handler = proc_dointvec, 3230 }, 3231 { 3232 /* Deprecated. Use gc_min_interval_ms */ 3233 3234 .procname = "gc_min_interval", 3235 .data = &ip_rt_gc_min_interval, 3236 .maxlen = sizeof(int), 3237 .mode = 0644, 3238 .proc_handler = proc_dointvec_jiffies, 3239 }, 3240 { 3241 .procname = "gc_min_interval_ms", 3242 .data = &ip_rt_gc_min_interval, 3243 .maxlen = sizeof(int), 3244 .mode = 0644, 3245 .proc_handler = proc_dointvec_ms_jiffies, 3246 }, 3247 { 3248 .procname = "gc_timeout", 3249 .data = &ip_rt_gc_timeout, 3250 .maxlen = sizeof(int), 3251 .mode = 0644, 3252 .proc_handler = proc_dointvec_jiffies, 3253 }, 3254 { 3255 .procname = "gc_interval", 3256 .data = &ip_rt_gc_interval, 3257 .maxlen = sizeof(int), 3258 .mode = 0644, 3259 .proc_handler = proc_dointvec_jiffies, 3260 }, 3261 { 3262 .procname = "redirect_load", 3263 .data = &ip_rt_redirect_load, 3264 .maxlen = sizeof(int), 3265 .mode = 0644, 3266 .proc_handler = proc_dointvec, 3267 }, 3268 { 3269 .procname = "redirect_number", 3270 .data = &ip_rt_redirect_number, 3271 .maxlen = sizeof(int), 3272 .mode = 0644, 3273 .proc_handler = proc_dointvec, 3274 }, 3275 { 3276 .procname = "redirect_silence", 3277 .data = &ip_rt_redirect_silence, 3278 .maxlen = sizeof(int), 3279 .mode = 0644, 3280 .proc_handler = proc_dointvec, 3281 }, 3282 { 3283 .procname = "error_cost", 3284 .data = &ip_rt_error_cost, 3285 .maxlen = sizeof(int), 3286 .mode = 0644, 3287 .proc_handler = proc_dointvec, 3288 }, 3289 { 3290 .procname = "error_burst", 3291 .data = &ip_rt_error_burst, 3292 .maxlen = sizeof(int), 3293 .mode = 0644, 3294 .proc_handler = proc_dointvec, 3295 }, 3296 { 3297 .procname = "gc_elasticity", 3298 .data = &ip_rt_gc_elasticity, 3299 .maxlen = sizeof(int), 3300 .mode = 0644, 3301 .proc_handler = proc_dointvec, 3302 }, 3303 { 3304 .procname = "mtu_expires", 3305 .data = &ip_rt_mtu_expires, 3306 .maxlen = sizeof(int), 3307 .mode = 0644, 3308 .proc_handler = proc_dointvec_jiffies, 3309 }, 3310 { 3311 .procname = "min_pmtu", 3312 .data = &ip_rt_min_pmtu, 3313 .maxlen = sizeof(int), 3314 .mode = 0644, 3315 .proc_handler = proc_dointvec_minmax, 3316 .extra1 = &ip_min_valid_pmtu, 3317 }, 3318 { 3319 .procname = "min_adv_mss", 3320 .data = &ip_rt_min_advmss, 3321 .maxlen = sizeof(int), 3322 .mode = 0644, 3323 .proc_handler = proc_dointvec, 3324 }, 3325 { } 3326 }; 3327 3328 static const char ipv4_route_flush_procname[] = "flush"; 3329 3330 static struct ctl_table ipv4_route_flush_table[] = { 3331 { 3332 .procname = ipv4_route_flush_procname, 3333 .maxlen = sizeof(int), 3334 .mode = 0200, 3335 .proc_handler = ipv4_sysctl_rtcache_flush, 3336 }, 3337 { }, 3338 }; 3339 3340 static __net_init int sysctl_route_net_init(struct net *net) 3341 { 3342 struct ctl_table *tbl; 3343 3344 tbl = ipv4_route_flush_table; 3345 if (!net_eq(net, &init_net)) { 3346 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL); 3347 if (!tbl) 3348 goto err_dup; 3349 3350 /* Don't export non-whitelisted sysctls to unprivileged users */ 3351 if (net->user_ns != &init_user_ns) { 3352 if (tbl[0].procname != ipv4_route_flush_procname) 3353 tbl[0].procname = NULL; 3354 } 3355 } 3356 tbl[0].extra1 = net; 3357 3358 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl); 3359 if (!net->ipv4.route_hdr) 3360 goto err_reg; 3361 return 0; 3362 3363 err_reg: 3364 if (tbl != ipv4_route_flush_table) 3365 kfree(tbl); 3366 err_dup: 3367 return -ENOMEM; 3368 } 3369 3370 static __net_exit void sysctl_route_net_exit(struct net *net) 3371 { 3372 struct ctl_table *tbl; 3373 3374 tbl = net->ipv4.route_hdr->ctl_table_arg; 3375 unregister_net_sysctl_table(net->ipv4.route_hdr); 3376 BUG_ON(tbl == ipv4_route_flush_table); 3377 kfree(tbl); 3378 } 3379 3380 static __net_initdata struct pernet_operations sysctl_route_ops = { 3381 .init = sysctl_route_net_init, 3382 .exit = sysctl_route_net_exit, 3383 }; 3384 #endif 3385 3386 static __net_init int rt_genid_init(struct net *net) 3387 { 3388 atomic_set(&net->ipv4.rt_genid, 0); 3389 atomic_set(&net->fnhe_genid, 0); 3390 atomic_set(&net->ipv4.dev_addr_genid, get_random_int()); 3391 return 0; 3392 } 3393 3394 static __net_initdata struct pernet_operations rt_genid_ops = { 3395 .init = rt_genid_init, 3396 }; 3397 3398 static int __net_init ipv4_inetpeer_init(struct net *net) 3399 { 3400 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL); 3401 3402 if (!bp) 3403 return -ENOMEM; 3404 inet_peer_base_init(bp); 3405 net->ipv4.peers = bp; 3406 return 0; 3407 } 3408 3409 static void __net_exit ipv4_inetpeer_exit(struct net *net) 3410 { 3411 struct inet_peer_base *bp = net->ipv4.peers; 3412 3413 net->ipv4.peers = NULL; 3414 inetpeer_invalidate_tree(bp); 3415 kfree(bp); 3416 } 3417 3418 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = { 3419 .init = ipv4_inetpeer_init, 3420 .exit = ipv4_inetpeer_exit, 3421 }; 3422 3423 #ifdef CONFIG_IP_ROUTE_CLASSID 3424 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 3425 #endif /* CONFIG_IP_ROUTE_CLASSID */ 3426 3427 int __init ip_rt_init(void) 3428 { 3429 int cpu; 3430 3431 ip_idents = kmalloc_array(IP_IDENTS_SZ, sizeof(*ip_idents), 3432 GFP_KERNEL); 3433 if (!ip_idents) 3434 panic("IP: failed to allocate ip_idents\n"); 3435 3436 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents)); 3437 3438 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL); 3439 if (!ip_tstamps) 3440 panic("IP: failed to allocate ip_tstamps\n"); 3441 3442 for_each_possible_cpu(cpu) { 3443 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 3444 3445 INIT_LIST_HEAD(&ul->head); 3446 spin_lock_init(&ul->lock); 3447 } 3448 #ifdef CONFIG_IP_ROUTE_CLASSID 3449 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 3450 if (!ip_rt_acct) 3451 panic("IP: failed to allocate ip_rt_acct\n"); 3452 #endif 3453 3454 ipv4_dst_ops.kmem_cachep = 3455 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0, 3456 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 3457 3458 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 3459 3460 if (dst_entries_init(&ipv4_dst_ops) < 0) 3461 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 3462 3463 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 3464 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 3465 3466 ipv4_dst_ops.gc_thresh = ~0; 3467 ip_rt_max_size = INT_MAX; 3468 3469 devinet_init(); 3470 ip_fib_init(); 3471 3472 if (ip_rt_proc_init()) 3473 pr_err("Unable to create route proc files\n"); 3474 #ifdef CONFIG_XFRM 3475 xfrm_init(); 3476 xfrm4_init(); 3477 #endif 3478 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, 3479 RTNL_FLAG_DOIT_UNLOCKED); 3480 3481 #ifdef CONFIG_SYSCTL 3482 register_pernet_subsys(&sysctl_route_ops); 3483 #endif 3484 register_pernet_subsys(&rt_genid_ops); 3485 register_pernet_subsys(&ipv4_inetpeer_ops); 3486 return 0; 3487 } 3488 3489 #ifdef CONFIG_SYSCTL 3490 /* 3491 * We really need to sanitize the damn ipv4 init order, then all 3492 * this nonsense will go away. 3493 */ 3494 void __init ip_static_sysctl_init(void) 3495 { 3496 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table); 3497 } 3498 #endif 3499