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 bool confirm_neigh); 144 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, 145 struct sk_buff *skb); 146 static void ipv4_dst_destroy(struct dst_entry *dst); 147 148 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old) 149 { 150 WARN_ON(1); 151 return NULL; 152 } 153 154 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 155 struct sk_buff *skb, 156 const void *daddr); 157 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr); 158 159 static struct dst_ops ipv4_dst_ops = { 160 .family = AF_INET, 161 .check = ipv4_dst_check, 162 .default_advmss = ipv4_default_advmss, 163 .mtu = ipv4_mtu, 164 .cow_metrics = ipv4_cow_metrics, 165 .destroy = ipv4_dst_destroy, 166 .negative_advice = ipv4_negative_advice, 167 .link_failure = ipv4_link_failure, 168 .update_pmtu = ip_rt_update_pmtu, 169 .redirect = ip_do_redirect, 170 .local_out = __ip_local_out, 171 .neigh_lookup = ipv4_neigh_lookup, 172 .confirm_neigh = ipv4_confirm_neigh, 173 }; 174 175 #define ECN_OR_COST(class) TC_PRIO_##class 176 177 const __u8 ip_tos2prio[16] = { 178 TC_PRIO_BESTEFFORT, 179 ECN_OR_COST(BESTEFFORT), 180 TC_PRIO_BESTEFFORT, 181 ECN_OR_COST(BESTEFFORT), 182 TC_PRIO_BULK, 183 ECN_OR_COST(BULK), 184 TC_PRIO_BULK, 185 ECN_OR_COST(BULK), 186 TC_PRIO_INTERACTIVE, 187 ECN_OR_COST(INTERACTIVE), 188 TC_PRIO_INTERACTIVE, 189 ECN_OR_COST(INTERACTIVE), 190 TC_PRIO_INTERACTIVE_BULK, 191 ECN_OR_COST(INTERACTIVE_BULK), 192 TC_PRIO_INTERACTIVE_BULK, 193 ECN_OR_COST(INTERACTIVE_BULK) 194 }; 195 EXPORT_SYMBOL(ip_tos2prio); 196 197 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat); 198 #define RT_CACHE_STAT_INC(field) raw_cpu_inc(rt_cache_stat.field) 199 200 #ifdef CONFIG_PROC_FS 201 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos) 202 { 203 if (*pos) 204 return NULL; 205 return SEQ_START_TOKEN; 206 } 207 208 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos) 209 { 210 ++*pos; 211 return NULL; 212 } 213 214 static void rt_cache_seq_stop(struct seq_file *seq, void *v) 215 { 216 } 217 218 static int rt_cache_seq_show(struct seq_file *seq, void *v) 219 { 220 if (v == SEQ_START_TOKEN) 221 seq_printf(seq, "%-127s\n", 222 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t" 223 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t" 224 "HHUptod\tSpecDst"); 225 return 0; 226 } 227 228 static const struct seq_operations rt_cache_seq_ops = { 229 .start = rt_cache_seq_start, 230 .next = rt_cache_seq_next, 231 .stop = rt_cache_seq_stop, 232 .show = rt_cache_seq_show, 233 }; 234 235 static int rt_cache_seq_open(struct inode *inode, struct file *file) 236 { 237 return seq_open(file, &rt_cache_seq_ops); 238 } 239 240 static const struct proc_ops rt_cache_proc_ops = { 241 .proc_open = rt_cache_seq_open, 242 .proc_read = seq_read, 243 .proc_lseek = seq_lseek, 244 .proc_release = seq_release, 245 }; 246 247 248 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos) 249 { 250 int cpu; 251 252 if (*pos == 0) 253 return SEQ_START_TOKEN; 254 255 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) { 256 if (!cpu_possible(cpu)) 257 continue; 258 *pos = cpu+1; 259 return &per_cpu(rt_cache_stat, cpu); 260 } 261 return NULL; 262 } 263 264 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos) 265 { 266 int cpu; 267 268 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) { 269 if (!cpu_possible(cpu)) 270 continue; 271 *pos = cpu+1; 272 return &per_cpu(rt_cache_stat, cpu); 273 } 274 (*pos)++; 275 return NULL; 276 277 } 278 279 static void rt_cpu_seq_stop(struct seq_file *seq, void *v) 280 { 281 282 } 283 284 static int rt_cpu_seq_show(struct seq_file *seq, void *v) 285 { 286 struct rt_cache_stat *st = v; 287 288 if (v == SEQ_START_TOKEN) { 289 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"); 290 return 0; 291 } 292 293 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x " 294 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n", 295 dst_entries_get_slow(&ipv4_dst_ops), 296 0, /* st->in_hit */ 297 st->in_slow_tot, 298 st->in_slow_mc, 299 st->in_no_route, 300 st->in_brd, 301 st->in_martian_dst, 302 st->in_martian_src, 303 304 0, /* st->out_hit */ 305 st->out_slow_tot, 306 st->out_slow_mc, 307 308 0, /* st->gc_total */ 309 0, /* st->gc_ignored */ 310 0, /* st->gc_goal_miss */ 311 0, /* st->gc_dst_overflow */ 312 0, /* st->in_hlist_search */ 313 0 /* st->out_hlist_search */ 314 ); 315 return 0; 316 } 317 318 static const struct seq_operations rt_cpu_seq_ops = { 319 .start = rt_cpu_seq_start, 320 .next = rt_cpu_seq_next, 321 .stop = rt_cpu_seq_stop, 322 .show = rt_cpu_seq_show, 323 }; 324 325 326 static int rt_cpu_seq_open(struct inode *inode, struct file *file) 327 { 328 return seq_open(file, &rt_cpu_seq_ops); 329 } 330 331 static const struct proc_ops rt_cpu_proc_ops = { 332 .proc_open = rt_cpu_seq_open, 333 .proc_read = seq_read, 334 .proc_lseek = seq_lseek, 335 .proc_release = seq_release, 336 }; 337 338 #ifdef CONFIG_IP_ROUTE_CLASSID 339 static int rt_acct_proc_show(struct seq_file *m, void *v) 340 { 341 struct ip_rt_acct *dst, *src; 342 unsigned int i, j; 343 344 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL); 345 if (!dst) 346 return -ENOMEM; 347 348 for_each_possible_cpu(i) { 349 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i); 350 for (j = 0; j < 256; j++) { 351 dst[j].o_bytes += src[j].o_bytes; 352 dst[j].o_packets += src[j].o_packets; 353 dst[j].i_bytes += src[j].i_bytes; 354 dst[j].i_packets += src[j].i_packets; 355 } 356 } 357 358 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct)); 359 kfree(dst); 360 return 0; 361 } 362 #endif 363 364 static int __net_init ip_rt_do_proc_init(struct net *net) 365 { 366 struct proc_dir_entry *pde; 367 368 pde = proc_create("rt_cache", 0444, net->proc_net, 369 &rt_cache_proc_ops); 370 if (!pde) 371 goto err1; 372 373 pde = proc_create("rt_cache", 0444, 374 net->proc_net_stat, &rt_cpu_proc_ops); 375 if (!pde) 376 goto err2; 377 378 #ifdef CONFIG_IP_ROUTE_CLASSID 379 pde = proc_create_single("rt_acct", 0, net->proc_net, 380 rt_acct_proc_show); 381 if (!pde) 382 goto err3; 383 #endif 384 return 0; 385 386 #ifdef CONFIG_IP_ROUTE_CLASSID 387 err3: 388 remove_proc_entry("rt_cache", net->proc_net_stat); 389 #endif 390 err2: 391 remove_proc_entry("rt_cache", net->proc_net); 392 err1: 393 return -ENOMEM; 394 } 395 396 static void __net_exit ip_rt_do_proc_exit(struct net *net) 397 { 398 remove_proc_entry("rt_cache", net->proc_net_stat); 399 remove_proc_entry("rt_cache", net->proc_net); 400 #ifdef CONFIG_IP_ROUTE_CLASSID 401 remove_proc_entry("rt_acct", net->proc_net); 402 #endif 403 } 404 405 static struct pernet_operations ip_rt_proc_ops __net_initdata = { 406 .init = ip_rt_do_proc_init, 407 .exit = ip_rt_do_proc_exit, 408 }; 409 410 static int __init ip_rt_proc_init(void) 411 { 412 return register_pernet_subsys(&ip_rt_proc_ops); 413 } 414 415 #else 416 static inline int ip_rt_proc_init(void) 417 { 418 return 0; 419 } 420 #endif /* CONFIG_PROC_FS */ 421 422 static inline bool rt_is_expired(const struct rtable *rth) 423 { 424 return rth->rt_genid != rt_genid_ipv4(dev_net(rth->dst.dev)); 425 } 426 427 void rt_cache_flush(struct net *net) 428 { 429 rt_genid_bump_ipv4(net); 430 } 431 432 static struct neighbour *ipv4_neigh_lookup(const struct dst_entry *dst, 433 struct sk_buff *skb, 434 const void *daddr) 435 { 436 const struct rtable *rt = container_of(dst, struct rtable, dst); 437 struct net_device *dev = dst->dev; 438 struct neighbour *n; 439 440 rcu_read_lock_bh(); 441 442 if (likely(rt->rt_gw_family == AF_INET)) { 443 n = ip_neigh_gw4(dev, rt->rt_gw4); 444 } else if (rt->rt_gw_family == AF_INET6) { 445 n = ip_neigh_gw6(dev, &rt->rt_gw6); 446 } else { 447 __be32 pkey; 448 449 pkey = skb ? ip_hdr(skb)->daddr : *((__be32 *) daddr); 450 n = ip_neigh_gw4(dev, pkey); 451 } 452 453 if (!IS_ERR(n) && !refcount_inc_not_zero(&n->refcnt)) 454 n = NULL; 455 456 rcu_read_unlock_bh(); 457 458 return n; 459 } 460 461 static void ipv4_confirm_neigh(const struct dst_entry *dst, const void *daddr) 462 { 463 const struct rtable *rt = container_of(dst, struct rtable, dst); 464 struct net_device *dev = dst->dev; 465 const __be32 *pkey = daddr; 466 467 if (rt->rt_gw_family == AF_INET) { 468 pkey = (const __be32 *)&rt->rt_gw4; 469 } else if (rt->rt_gw_family == AF_INET6) { 470 return __ipv6_confirm_neigh_stub(dev, &rt->rt_gw6); 471 } else if (!daddr || 472 (rt->rt_flags & 473 (RTCF_MULTICAST | RTCF_BROADCAST | RTCF_LOCAL))) { 474 return; 475 } 476 __ipv4_confirm_neigh(dev, *(__force u32 *)pkey); 477 } 478 479 #define IP_IDENTS_SZ 2048u 480 481 static atomic_t *ip_idents __read_mostly; 482 static u32 *ip_tstamps __read_mostly; 483 484 /* In order to protect privacy, we add a perturbation to identifiers 485 * if one generator is seldom used. This makes hard for an attacker 486 * to infer how many packets were sent between two points in time. 487 */ 488 u32 ip_idents_reserve(u32 hash, int segs) 489 { 490 u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ; 491 atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ; 492 u32 old = READ_ONCE(*p_tstamp); 493 u32 now = (u32)jiffies; 494 u32 new, delta = 0; 495 496 if (old != now && cmpxchg(p_tstamp, old, now) == old) 497 delta = prandom_u32_max(now - old); 498 499 /* Do not use atomic_add_return() as it makes UBSAN unhappy */ 500 do { 501 old = (u32)atomic_read(p_id); 502 new = old + delta + segs; 503 } while (atomic_cmpxchg(p_id, old, new) != old); 504 505 return new - segs; 506 } 507 EXPORT_SYMBOL(ip_idents_reserve); 508 509 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs) 510 { 511 u32 hash, id; 512 513 /* Note the following code is not safe, but this is okay. */ 514 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key))) 515 get_random_bytes(&net->ipv4.ip_id_key, 516 sizeof(net->ipv4.ip_id_key)); 517 518 hash = siphash_3u32((__force u32)iph->daddr, 519 (__force u32)iph->saddr, 520 iph->protocol, 521 &net->ipv4.ip_id_key); 522 id = ip_idents_reserve(hash, segs); 523 iph->id = htons(id); 524 } 525 EXPORT_SYMBOL(__ip_select_ident); 526 527 static void __build_flow_key(const struct net *net, struct flowi4 *fl4, 528 const struct sock *sk, 529 const struct iphdr *iph, 530 int oif, u8 tos, 531 u8 prot, u32 mark, int flow_flags) 532 { 533 if (sk) { 534 const struct inet_sock *inet = inet_sk(sk); 535 536 oif = sk->sk_bound_dev_if; 537 mark = sk->sk_mark; 538 tos = RT_CONN_FLAGS(sk); 539 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol; 540 } 541 flowi4_init_output(fl4, oif, mark, tos, 542 RT_SCOPE_UNIVERSE, prot, 543 flow_flags, 544 iph->daddr, iph->saddr, 0, 0, 545 sock_net_uid(net, sk)); 546 } 547 548 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb, 549 const struct sock *sk) 550 { 551 const struct net *net = dev_net(skb->dev); 552 const struct iphdr *iph = ip_hdr(skb); 553 int oif = skb->dev->ifindex; 554 u8 tos = RT_TOS(iph->tos); 555 u8 prot = iph->protocol; 556 u32 mark = skb->mark; 557 558 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0); 559 } 560 561 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk) 562 { 563 const struct inet_sock *inet = inet_sk(sk); 564 const struct ip_options_rcu *inet_opt; 565 __be32 daddr = inet->inet_daddr; 566 567 rcu_read_lock(); 568 inet_opt = rcu_dereference(inet->inet_opt); 569 if (inet_opt && inet_opt->opt.srr) 570 daddr = inet_opt->opt.faddr; 571 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark, 572 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 573 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol, 574 inet_sk_flowi_flags(sk), 575 daddr, inet->inet_saddr, 0, 0, sk->sk_uid); 576 rcu_read_unlock(); 577 } 578 579 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk, 580 const struct sk_buff *skb) 581 { 582 if (skb) 583 build_skb_flow_key(fl4, skb, sk); 584 else 585 build_sk_flow_key(fl4, sk); 586 } 587 588 static DEFINE_SPINLOCK(fnhe_lock); 589 590 static void fnhe_flush_routes(struct fib_nh_exception *fnhe) 591 { 592 struct rtable *rt; 593 594 rt = rcu_dereference(fnhe->fnhe_rth_input); 595 if (rt) { 596 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL); 597 dst_dev_put(&rt->dst); 598 dst_release(&rt->dst); 599 } 600 rt = rcu_dereference(fnhe->fnhe_rth_output); 601 if (rt) { 602 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL); 603 dst_dev_put(&rt->dst); 604 dst_release(&rt->dst); 605 } 606 } 607 608 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash) 609 { 610 struct fib_nh_exception *fnhe, *oldest; 611 612 oldest = rcu_dereference(hash->chain); 613 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe; 614 fnhe = rcu_dereference(fnhe->fnhe_next)) { 615 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) 616 oldest = fnhe; 617 } 618 fnhe_flush_routes(oldest); 619 return oldest; 620 } 621 622 static inline u32 fnhe_hashfun(__be32 daddr) 623 { 624 static u32 fnhe_hashrnd __read_mostly; 625 u32 hval; 626 627 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd)); 628 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd); 629 return hash_32(hval, FNHE_HASH_SHIFT); 630 } 631 632 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe) 633 { 634 rt->rt_pmtu = fnhe->fnhe_pmtu; 635 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked; 636 rt->dst.expires = fnhe->fnhe_expires; 637 638 if (fnhe->fnhe_gw) { 639 rt->rt_flags |= RTCF_REDIRECTED; 640 rt->rt_uses_gateway = 1; 641 rt->rt_gw_family = AF_INET; 642 rt->rt_gw4 = fnhe->fnhe_gw; 643 } 644 } 645 646 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr, 647 __be32 gw, u32 pmtu, bool lock, 648 unsigned long expires) 649 { 650 struct fnhe_hash_bucket *hash; 651 struct fib_nh_exception *fnhe; 652 struct rtable *rt; 653 u32 genid, hval; 654 unsigned int i; 655 int depth; 656 657 genid = fnhe_genid(dev_net(nhc->nhc_dev)); 658 hval = fnhe_hashfun(daddr); 659 660 spin_lock_bh(&fnhe_lock); 661 662 hash = rcu_dereference(nhc->nhc_exceptions); 663 if (!hash) { 664 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC); 665 if (!hash) 666 goto out_unlock; 667 rcu_assign_pointer(nhc->nhc_exceptions, hash); 668 } 669 670 hash += hval; 671 672 depth = 0; 673 for (fnhe = rcu_dereference(hash->chain); fnhe; 674 fnhe = rcu_dereference(fnhe->fnhe_next)) { 675 if (fnhe->fnhe_daddr == daddr) 676 break; 677 depth++; 678 } 679 680 if (fnhe) { 681 if (fnhe->fnhe_genid != genid) 682 fnhe->fnhe_genid = genid; 683 if (gw) 684 fnhe->fnhe_gw = gw; 685 if (pmtu) { 686 fnhe->fnhe_pmtu = pmtu; 687 fnhe->fnhe_mtu_locked = lock; 688 } 689 fnhe->fnhe_expires = max(1UL, expires); 690 /* Update all cached dsts too */ 691 rt = rcu_dereference(fnhe->fnhe_rth_input); 692 if (rt) 693 fill_route_from_fnhe(rt, fnhe); 694 rt = rcu_dereference(fnhe->fnhe_rth_output); 695 if (rt) 696 fill_route_from_fnhe(rt, fnhe); 697 } else { 698 if (depth > FNHE_RECLAIM_DEPTH) 699 fnhe = fnhe_oldest(hash); 700 else { 701 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC); 702 if (!fnhe) 703 goto out_unlock; 704 705 fnhe->fnhe_next = hash->chain; 706 rcu_assign_pointer(hash->chain, fnhe); 707 } 708 fnhe->fnhe_genid = genid; 709 fnhe->fnhe_daddr = daddr; 710 fnhe->fnhe_gw = gw; 711 fnhe->fnhe_pmtu = pmtu; 712 fnhe->fnhe_mtu_locked = lock; 713 fnhe->fnhe_expires = max(1UL, expires); 714 715 /* Exception created; mark the cached routes for the nexthop 716 * stale, so anyone caching it rechecks if this exception 717 * applies to them. 718 */ 719 rt = rcu_dereference(nhc->nhc_rth_input); 720 if (rt) 721 rt->dst.obsolete = DST_OBSOLETE_KILL; 722 723 for_each_possible_cpu(i) { 724 struct rtable __rcu **prt; 725 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i); 726 rt = rcu_dereference(*prt); 727 if (rt) 728 rt->dst.obsolete = DST_OBSOLETE_KILL; 729 } 730 } 731 732 fnhe->fnhe_stamp = jiffies; 733 734 out_unlock: 735 spin_unlock_bh(&fnhe_lock); 736 } 737 738 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4, 739 bool kill_route) 740 { 741 __be32 new_gw = icmp_hdr(skb)->un.gateway; 742 __be32 old_gw = ip_hdr(skb)->saddr; 743 struct net_device *dev = skb->dev; 744 struct in_device *in_dev; 745 struct fib_result res; 746 struct neighbour *n; 747 struct net *net; 748 749 switch (icmp_hdr(skb)->code & 7) { 750 case ICMP_REDIR_NET: 751 case ICMP_REDIR_NETTOS: 752 case ICMP_REDIR_HOST: 753 case ICMP_REDIR_HOSTTOS: 754 break; 755 756 default: 757 return; 758 } 759 760 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw) 761 return; 762 763 in_dev = __in_dev_get_rcu(dev); 764 if (!in_dev) 765 return; 766 767 net = dev_net(dev); 768 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) || 769 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) || 770 ipv4_is_zeronet(new_gw)) 771 goto reject_redirect; 772 773 if (!IN_DEV_SHARED_MEDIA(in_dev)) { 774 if (!inet_addr_onlink(in_dev, new_gw, old_gw)) 775 goto reject_redirect; 776 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) 777 goto reject_redirect; 778 } else { 779 if (inet_addr_type(net, new_gw) != RTN_UNICAST) 780 goto reject_redirect; 781 } 782 783 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw); 784 if (!n) 785 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev); 786 if (!IS_ERR(n)) { 787 if (!(n->nud_state & NUD_VALID)) { 788 neigh_event_send(n, NULL); 789 } else { 790 if (fib_lookup(net, fl4, &res, 0) == 0) { 791 struct fib_nh_common *nhc = FIB_RES_NHC(res); 792 793 update_or_create_fnhe(nhc, fl4->daddr, new_gw, 794 0, false, 795 jiffies + ip_rt_gc_timeout); 796 } 797 if (kill_route) 798 rt->dst.obsolete = DST_OBSOLETE_KILL; 799 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n); 800 } 801 neigh_release(n); 802 } 803 return; 804 805 reject_redirect: 806 #ifdef CONFIG_IP_ROUTE_VERBOSE 807 if (IN_DEV_LOG_MARTIANS(in_dev)) { 808 const struct iphdr *iph = (const struct iphdr *) skb->data; 809 __be32 daddr = iph->daddr; 810 __be32 saddr = iph->saddr; 811 812 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n" 813 " Advised path = %pI4 -> %pI4\n", 814 &old_gw, dev->name, &new_gw, 815 &saddr, &daddr); 816 } 817 #endif 818 ; 819 } 820 821 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb) 822 { 823 struct rtable *rt; 824 struct flowi4 fl4; 825 const struct iphdr *iph = (const struct iphdr *) skb->data; 826 struct net *net = dev_net(skb->dev); 827 int oif = skb->dev->ifindex; 828 u8 tos = RT_TOS(iph->tos); 829 u8 prot = iph->protocol; 830 u32 mark = skb->mark; 831 832 rt = (struct rtable *) dst; 833 834 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0); 835 __ip_do_redirect(rt, skb, &fl4, true); 836 } 837 838 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst) 839 { 840 struct rtable *rt = (struct rtable *)dst; 841 struct dst_entry *ret = dst; 842 843 if (rt) { 844 if (dst->obsolete > 0) { 845 ip_rt_put(rt); 846 ret = NULL; 847 } else if ((rt->rt_flags & RTCF_REDIRECTED) || 848 rt->dst.expires) { 849 ip_rt_put(rt); 850 ret = NULL; 851 } 852 } 853 return ret; 854 } 855 856 /* 857 * Algorithm: 858 * 1. The first ip_rt_redirect_number redirects are sent 859 * with exponential backoff, then we stop sending them at all, 860 * assuming that the host ignores our redirects. 861 * 2. If we did not see packets requiring redirects 862 * during ip_rt_redirect_silence, we assume that the host 863 * forgot redirected route and start to send redirects again. 864 * 865 * This algorithm is much cheaper and more intelligent than dumb load limiting 866 * in icmp.c. 867 * 868 * NOTE. Do not forget to inhibit load limiting for redirects (redundant) 869 * and "frag. need" (breaks PMTU discovery) in icmp.c. 870 */ 871 872 void ip_rt_send_redirect(struct sk_buff *skb) 873 { 874 struct rtable *rt = skb_rtable(skb); 875 struct in_device *in_dev; 876 struct inet_peer *peer; 877 struct net *net; 878 int log_martians; 879 int vif; 880 881 rcu_read_lock(); 882 in_dev = __in_dev_get_rcu(rt->dst.dev); 883 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) { 884 rcu_read_unlock(); 885 return; 886 } 887 log_martians = IN_DEV_LOG_MARTIANS(in_dev); 888 vif = l3mdev_master_ifindex_rcu(rt->dst.dev); 889 rcu_read_unlock(); 890 891 net = dev_net(rt->dst.dev); 892 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1); 893 if (!peer) { 894 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, 895 rt_nexthop(rt, ip_hdr(skb)->daddr)); 896 return; 897 } 898 899 /* No redirected packets during ip_rt_redirect_silence; 900 * reset the algorithm. 901 */ 902 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) { 903 peer->rate_tokens = 0; 904 peer->n_redirects = 0; 905 } 906 907 /* Too many ignored redirects; do not send anything 908 * set dst.rate_last to the last seen redirected packet. 909 */ 910 if (peer->n_redirects >= ip_rt_redirect_number) { 911 peer->rate_last = jiffies; 912 goto out_put_peer; 913 } 914 915 /* Check for load limit; set rate_last to the latest sent 916 * redirect. 917 */ 918 if (peer->rate_tokens == 0 || 919 time_after(jiffies, 920 (peer->rate_last + 921 (ip_rt_redirect_load << peer->n_redirects)))) { 922 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr); 923 924 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw); 925 peer->rate_last = jiffies; 926 ++peer->n_redirects; 927 #ifdef CONFIG_IP_ROUTE_VERBOSE 928 if (log_martians && 929 peer->n_redirects == ip_rt_redirect_number) 930 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n", 931 &ip_hdr(skb)->saddr, inet_iif(skb), 932 &ip_hdr(skb)->daddr, &gw); 933 #endif 934 } 935 out_put_peer: 936 inet_putpeer(peer); 937 } 938 939 static int ip_error(struct sk_buff *skb) 940 { 941 struct rtable *rt = skb_rtable(skb); 942 struct net_device *dev = skb->dev; 943 struct in_device *in_dev; 944 struct inet_peer *peer; 945 unsigned long now; 946 struct net *net; 947 bool send; 948 int code; 949 950 if (netif_is_l3_master(skb->dev)) { 951 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif); 952 if (!dev) 953 goto out; 954 } 955 956 in_dev = __in_dev_get_rcu(dev); 957 958 /* IP on this device is disabled. */ 959 if (!in_dev) 960 goto out; 961 962 net = dev_net(rt->dst.dev); 963 if (!IN_DEV_FORWARD(in_dev)) { 964 switch (rt->dst.error) { 965 case EHOSTUNREACH: 966 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS); 967 break; 968 969 case ENETUNREACH: 970 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 971 break; 972 } 973 goto out; 974 } 975 976 switch (rt->dst.error) { 977 case EINVAL: 978 default: 979 goto out; 980 case EHOSTUNREACH: 981 code = ICMP_HOST_UNREACH; 982 break; 983 case ENETUNREACH: 984 code = ICMP_NET_UNREACH; 985 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 986 break; 987 case EACCES: 988 code = ICMP_PKT_FILTERED; 989 break; 990 } 991 992 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 993 l3mdev_master_ifindex(skb->dev), 1); 994 995 send = true; 996 if (peer) { 997 now = jiffies; 998 peer->rate_tokens += now - peer->rate_last; 999 if (peer->rate_tokens > ip_rt_error_burst) 1000 peer->rate_tokens = ip_rt_error_burst; 1001 peer->rate_last = now; 1002 if (peer->rate_tokens >= ip_rt_error_cost) 1003 peer->rate_tokens -= ip_rt_error_cost; 1004 else 1005 send = false; 1006 inet_putpeer(peer); 1007 } 1008 if (send) 1009 icmp_send(skb, ICMP_DEST_UNREACH, code, 0); 1010 1011 out: kfree_skb(skb); 1012 return 0; 1013 } 1014 1015 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu) 1016 { 1017 struct dst_entry *dst = &rt->dst; 1018 u32 old_mtu = ipv4_mtu(dst); 1019 struct fib_result res; 1020 bool lock = false; 1021 1022 if (ip_mtu_locked(dst)) 1023 return; 1024 1025 if (old_mtu < mtu) 1026 return; 1027 1028 if (mtu < ip_rt_min_pmtu) { 1029 lock = true; 1030 mtu = min(old_mtu, ip_rt_min_pmtu); 1031 } 1032 1033 if (rt->rt_pmtu == mtu && !lock && 1034 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2)) 1035 return; 1036 1037 rcu_read_lock(); 1038 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) { 1039 struct fib_nh_common *nhc = FIB_RES_NHC(res); 1040 1041 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1042 jiffies + ip_rt_mtu_expires); 1043 } 1044 rcu_read_unlock(); 1045 } 1046 1047 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 1048 struct sk_buff *skb, u32 mtu, 1049 bool confirm_neigh) 1050 { 1051 struct rtable *rt = (struct rtable *) dst; 1052 struct flowi4 fl4; 1053 1054 ip_rt_build_flow_key(&fl4, sk, skb); 1055 __ip_rt_update_pmtu(rt, &fl4, mtu); 1056 } 1057 1058 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, 1059 int oif, u8 protocol) 1060 { 1061 const struct iphdr *iph = (const struct iphdr *) skb->data; 1062 struct flowi4 fl4; 1063 struct rtable *rt; 1064 u32 mark = IP4_REPLY_MARK(net, skb->mark); 1065 1066 __build_flow_key(net, &fl4, NULL, iph, oif, 1067 RT_TOS(iph->tos), protocol, mark, 0); 1068 rt = __ip_route_output_key(net, &fl4); 1069 if (!IS_ERR(rt)) { 1070 __ip_rt_update_pmtu(rt, &fl4, mtu); 1071 ip_rt_put(rt); 1072 } 1073 } 1074 EXPORT_SYMBOL_GPL(ipv4_update_pmtu); 1075 1076 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1077 { 1078 const struct iphdr *iph = (const struct iphdr *) skb->data; 1079 struct flowi4 fl4; 1080 struct rtable *rt; 1081 1082 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0); 1083 1084 if (!fl4.flowi4_mark) 1085 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark); 1086 1087 rt = __ip_route_output_key(sock_net(sk), &fl4); 1088 if (!IS_ERR(rt)) { 1089 __ip_rt_update_pmtu(rt, &fl4, mtu); 1090 ip_rt_put(rt); 1091 } 1092 } 1093 1094 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1095 { 1096 const struct iphdr *iph = (const struct iphdr *) skb->data; 1097 struct flowi4 fl4; 1098 struct rtable *rt; 1099 struct dst_entry *odst = NULL; 1100 bool new = false; 1101 struct net *net = sock_net(sk); 1102 1103 bh_lock_sock(sk); 1104 1105 if (!ip_sk_accept_pmtu(sk)) 1106 goto out; 1107 1108 odst = sk_dst_get(sk); 1109 1110 if (sock_owned_by_user(sk) || !odst) { 1111 __ipv4_sk_update_pmtu(skb, sk, mtu); 1112 goto out; 1113 } 1114 1115 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1116 1117 rt = (struct rtable *)odst; 1118 if (odst->obsolete && !odst->ops->check(odst, 0)) { 1119 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1120 if (IS_ERR(rt)) 1121 goto out; 1122 1123 new = true; 1124 } 1125 1126 __ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu); 1127 1128 if (!dst_check(&rt->dst, 0)) { 1129 if (new) 1130 dst_release(&rt->dst); 1131 1132 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1133 if (IS_ERR(rt)) 1134 goto out; 1135 1136 new = true; 1137 } 1138 1139 if (new) 1140 sk_dst_set(sk, &rt->dst); 1141 1142 out: 1143 bh_unlock_sock(sk); 1144 dst_release(odst); 1145 } 1146 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu); 1147 1148 void ipv4_redirect(struct sk_buff *skb, struct net *net, 1149 int oif, u8 protocol) 1150 { 1151 const struct iphdr *iph = (const struct iphdr *) skb->data; 1152 struct flowi4 fl4; 1153 struct rtable *rt; 1154 1155 __build_flow_key(net, &fl4, NULL, iph, oif, 1156 RT_TOS(iph->tos), protocol, 0, 0); 1157 rt = __ip_route_output_key(net, &fl4); 1158 if (!IS_ERR(rt)) { 1159 __ip_do_redirect(rt, skb, &fl4, false); 1160 ip_rt_put(rt); 1161 } 1162 } 1163 EXPORT_SYMBOL_GPL(ipv4_redirect); 1164 1165 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk) 1166 { 1167 const struct iphdr *iph = (const struct iphdr *) skb->data; 1168 struct flowi4 fl4; 1169 struct rtable *rt; 1170 struct net *net = sock_net(sk); 1171 1172 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1173 rt = __ip_route_output_key(net, &fl4); 1174 if (!IS_ERR(rt)) { 1175 __ip_do_redirect(rt, skb, &fl4, false); 1176 ip_rt_put(rt); 1177 } 1178 } 1179 EXPORT_SYMBOL_GPL(ipv4_sk_redirect); 1180 1181 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie) 1182 { 1183 struct rtable *rt = (struct rtable *) dst; 1184 1185 /* All IPV4 dsts are created with ->obsolete set to the value 1186 * DST_OBSOLETE_FORCE_CHK which forces validation calls down 1187 * into this function always. 1188 * 1189 * When a PMTU/redirect information update invalidates a route, 1190 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or 1191 * DST_OBSOLETE_DEAD. 1192 */ 1193 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt)) 1194 return NULL; 1195 return dst; 1196 } 1197 1198 static void ipv4_send_dest_unreach(struct sk_buff *skb) 1199 { 1200 struct ip_options opt; 1201 int res; 1202 1203 /* Recompile ip options since IPCB may not be valid anymore. 1204 * Also check we have a reasonable ipv4 header. 1205 */ 1206 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) || 1207 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5) 1208 return; 1209 1210 memset(&opt, 0, sizeof(opt)); 1211 if (ip_hdr(skb)->ihl > 5) { 1212 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4)) 1213 return; 1214 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr); 1215 1216 rcu_read_lock(); 1217 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL); 1218 rcu_read_unlock(); 1219 1220 if (res) 1221 return; 1222 } 1223 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt); 1224 } 1225 1226 static void ipv4_link_failure(struct sk_buff *skb) 1227 { 1228 struct rtable *rt; 1229 1230 ipv4_send_dest_unreach(skb); 1231 1232 rt = skb_rtable(skb); 1233 if (rt) 1234 dst_set_expires(&rt->dst, 0); 1235 } 1236 1237 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb) 1238 { 1239 pr_debug("%s: %pI4 -> %pI4, %s\n", 1240 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1241 skb->dev ? skb->dev->name : "?"); 1242 kfree_skb(skb); 1243 WARN_ON(1); 1244 return 0; 1245 } 1246 1247 /* 1248 We do not cache source address of outgoing interface, 1249 because it is used only by IP RR, TS and SRR options, 1250 so that it out of fast path. 1251 1252 BTW remember: "addr" is allowed to be not aligned 1253 in IP options! 1254 */ 1255 1256 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt) 1257 { 1258 __be32 src; 1259 1260 if (rt_is_output_route(rt)) 1261 src = ip_hdr(skb)->saddr; 1262 else { 1263 struct fib_result res; 1264 struct iphdr *iph = ip_hdr(skb); 1265 struct flowi4 fl4 = { 1266 .daddr = iph->daddr, 1267 .saddr = iph->saddr, 1268 .flowi4_tos = RT_TOS(iph->tos), 1269 .flowi4_oif = rt->dst.dev->ifindex, 1270 .flowi4_iif = skb->dev->ifindex, 1271 .flowi4_mark = skb->mark, 1272 }; 1273 1274 rcu_read_lock(); 1275 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0) 1276 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res); 1277 else 1278 src = inet_select_addr(rt->dst.dev, 1279 rt_nexthop(rt, iph->daddr), 1280 RT_SCOPE_UNIVERSE); 1281 rcu_read_unlock(); 1282 } 1283 memcpy(addr, &src, 4); 1284 } 1285 1286 #ifdef CONFIG_IP_ROUTE_CLASSID 1287 static void set_class_tag(struct rtable *rt, u32 tag) 1288 { 1289 if (!(rt->dst.tclassid & 0xFFFF)) 1290 rt->dst.tclassid |= tag & 0xFFFF; 1291 if (!(rt->dst.tclassid & 0xFFFF0000)) 1292 rt->dst.tclassid |= tag & 0xFFFF0000; 1293 } 1294 #endif 1295 1296 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1297 { 1298 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr); 1299 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size, 1300 ip_rt_min_advmss); 1301 1302 return min(advmss, IPV4_MAX_PMTU - header_size); 1303 } 1304 1305 static unsigned int ipv4_mtu(const struct dst_entry *dst) 1306 { 1307 const struct rtable *rt = (const struct rtable *) dst; 1308 unsigned int mtu = rt->rt_pmtu; 1309 1310 if (!mtu || time_after_eq(jiffies, rt->dst.expires)) 1311 mtu = dst_metric_raw(dst, RTAX_MTU); 1312 1313 if (mtu) 1314 return mtu; 1315 1316 mtu = READ_ONCE(dst->dev->mtu); 1317 1318 if (unlikely(ip_mtu_locked(dst))) { 1319 if (rt->rt_uses_gateway && mtu > 576) 1320 mtu = 576; 1321 } 1322 1323 mtu = min_t(unsigned int, mtu, IP_MAX_MTU); 1324 1325 return mtu - lwtunnel_headroom(dst->lwtstate, mtu); 1326 } 1327 1328 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr) 1329 { 1330 struct fnhe_hash_bucket *hash; 1331 struct fib_nh_exception *fnhe, __rcu **fnhe_p; 1332 u32 hval = fnhe_hashfun(daddr); 1333 1334 spin_lock_bh(&fnhe_lock); 1335 1336 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1337 lockdep_is_held(&fnhe_lock)); 1338 hash += hval; 1339 1340 fnhe_p = &hash->chain; 1341 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock)); 1342 while (fnhe) { 1343 if (fnhe->fnhe_daddr == daddr) { 1344 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected( 1345 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock))); 1346 /* set fnhe_daddr to 0 to ensure it won't bind with 1347 * new dsts in rt_bind_exception(). 1348 */ 1349 fnhe->fnhe_daddr = 0; 1350 fnhe_flush_routes(fnhe); 1351 kfree_rcu(fnhe, rcu); 1352 break; 1353 } 1354 fnhe_p = &fnhe->fnhe_next; 1355 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1356 lockdep_is_held(&fnhe_lock)); 1357 } 1358 1359 spin_unlock_bh(&fnhe_lock); 1360 } 1361 1362 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc, 1363 __be32 daddr) 1364 { 1365 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions); 1366 struct fib_nh_exception *fnhe; 1367 u32 hval; 1368 1369 if (!hash) 1370 return NULL; 1371 1372 hval = fnhe_hashfun(daddr); 1373 1374 for (fnhe = rcu_dereference(hash[hval].chain); fnhe; 1375 fnhe = rcu_dereference(fnhe->fnhe_next)) { 1376 if (fnhe->fnhe_daddr == daddr) { 1377 if (fnhe->fnhe_expires && 1378 time_after(jiffies, fnhe->fnhe_expires)) { 1379 ip_del_fnhe(nhc, daddr); 1380 break; 1381 } 1382 return fnhe; 1383 } 1384 } 1385 return NULL; 1386 } 1387 1388 /* MTU selection: 1389 * 1. mtu on route is locked - use it 1390 * 2. mtu from nexthop exception 1391 * 3. mtu from egress device 1392 */ 1393 1394 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr) 1395 { 1396 struct fib_nh_common *nhc = res->nhc; 1397 struct net_device *dev = nhc->nhc_dev; 1398 struct fib_info *fi = res->fi; 1399 u32 mtu = 0; 1400 1401 if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu || 1402 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU)) 1403 mtu = fi->fib_mtu; 1404 1405 if (likely(!mtu)) { 1406 struct fib_nh_exception *fnhe; 1407 1408 fnhe = find_exception(nhc, daddr); 1409 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires)) 1410 mtu = fnhe->fnhe_pmtu; 1411 } 1412 1413 if (likely(!mtu)) 1414 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU); 1415 1416 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu); 1417 } 1418 1419 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, 1420 __be32 daddr, const bool do_cache) 1421 { 1422 bool ret = false; 1423 1424 spin_lock_bh(&fnhe_lock); 1425 1426 if (daddr == fnhe->fnhe_daddr) { 1427 struct rtable __rcu **porig; 1428 struct rtable *orig; 1429 int genid = fnhe_genid(dev_net(rt->dst.dev)); 1430 1431 if (rt_is_input_route(rt)) 1432 porig = &fnhe->fnhe_rth_input; 1433 else 1434 porig = &fnhe->fnhe_rth_output; 1435 orig = rcu_dereference(*porig); 1436 1437 if (fnhe->fnhe_genid != genid) { 1438 fnhe->fnhe_genid = genid; 1439 fnhe->fnhe_gw = 0; 1440 fnhe->fnhe_pmtu = 0; 1441 fnhe->fnhe_expires = 0; 1442 fnhe->fnhe_mtu_locked = false; 1443 fnhe_flush_routes(fnhe); 1444 orig = NULL; 1445 } 1446 fill_route_from_fnhe(rt, fnhe); 1447 if (!rt->rt_gw4) { 1448 rt->rt_gw4 = daddr; 1449 rt->rt_gw_family = AF_INET; 1450 } 1451 1452 if (do_cache) { 1453 dst_hold(&rt->dst); 1454 rcu_assign_pointer(*porig, rt); 1455 if (orig) { 1456 dst_dev_put(&orig->dst); 1457 dst_release(&orig->dst); 1458 } 1459 ret = true; 1460 } 1461 1462 fnhe->fnhe_stamp = jiffies; 1463 } 1464 spin_unlock_bh(&fnhe_lock); 1465 1466 return ret; 1467 } 1468 1469 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt) 1470 { 1471 struct rtable *orig, *prev, **p; 1472 bool ret = true; 1473 1474 if (rt_is_input_route(rt)) { 1475 p = (struct rtable **)&nhc->nhc_rth_input; 1476 } else { 1477 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 1478 } 1479 orig = *p; 1480 1481 /* hold dst before doing cmpxchg() to avoid race condition 1482 * on this dst 1483 */ 1484 dst_hold(&rt->dst); 1485 prev = cmpxchg(p, orig, rt); 1486 if (prev == orig) { 1487 if (orig) { 1488 rt_add_uncached_list(orig); 1489 dst_release(&orig->dst); 1490 } 1491 } else { 1492 dst_release(&rt->dst); 1493 ret = false; 1494 } 1495 1496 return ret; 1497 } 1498 1499 struct uncached_list { 1500 spinlock_t lock; 1501 struct list_head head; 1502 }; 1503 1504 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list); 1505 1506 void rt_add_uncached_list(struct rtable *rt) 1507 { 1508 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list); 1509 1510 rt->rt_uncached_list = ul; 1511 1512 spin_lock_bh(&ul->lock); 1513 list_add_tail(&rt->rt_uncached, &ul->head); 1514 spin_unlock_bh(&ul->lock); 1515 } 1516 1517 void rt_del_uncached_list(struct rtable *rt) 1518 { 1519 if (!list_empty(&rt->rt_uncached)) { 1520 struct uncached_list *ul = rt->rt_uncached_list; 1521 1522 spin_lock_bh(&ul->lock); 1523 list_del(&rt->rt_uncached); 1524 spin_unlock_bh(&ul->lock); 1525 } 1526 } 1527 1528 static void ipv4_dst_destroy(struct dst_entry *dst) 1529 { 1530 struct rtable *rt = (struct rtable *)dst; 1531 1532 ip_dst_metrics_put(dst); 1533 rt_del_uncached_list(rt); 1534 } 1535 1536 void rt_flush_dev(struct net_device *dev) 1537 { 1538 struct rtable *rt; 1539 int cpu; 1540 1541 for_each_possible_cpu(cpu) { 1542 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 1543 1544 spin_lock_bh(&ul->lock); 1545 list_for_each_entry(rt, &ul->head, rt_uncached) { 1546 if (rt->dst.dev != dev) 1547 continue; 1548 rt->dst.dev = blackhole_netdev; 1549 dev_hold(rt->dst.dev); 1550 dev_put(dev); 1551 } 1552 spin_unlock_bh(&ul->lock); 1553 } 1554 } 1555 1556 static bool rt_cache_valid(const struct rtable *rt) 1557 { 1558 return rt && 1559 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK && 1560 !rt_is_expired(rt); 1561 } 1562 1563 static void rt_set_nexthop(struct rtable *rt, __be32 daddr, 1564 const struct fib_result *res, 1565 struct fib_nh_exception *fnhe, 1566 struct fib_info *fi, u16 type, u32 itag, 1567 const bool do_cache) 1568 { 1569 bool cached = false; 1570 1571 if (fi) { 1572 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1573 1574 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) { 1575 rt->rt_uses_gateway = 1; 1576 rt->rt_gw_family = nhc->nhc_gw_family; 1577 /* only INET and INET6 are supported */ 1578 if (likely(nhc->nhc_gw_family == AF_INET)) 1579 rt->rt_gw4 = nhc->nhc_gw.ipv4; 1580 else 1581 rt->rt_gw6 = nhc->nhc_gw.ipv6; 1582 } 1583 1584 ip_dst_init_metrics(&rt->dst, fi->fib_metrics); 1585 1586 #ifdef CONFIG_IP_ROUTE_CLASSID 1587 if (nhc->nhc_family == AF_INET) { 1588 struct fib_nh *nh; 1589 1590 nh = container_of(nhc, struct fib_nh, nh_common); 1591 rt->dst.tclassid = nh->nh_tclassid; 1592 } 1593 #endif 1594 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 1595 if (unlikely(fnhe)) 1596 cached = rt_bind_exception(rt, fnhe, daddr, do_cache); 1597 else if (do_cache) 1598 cached = rt_cache_route(nhc, rt); 1599 if (unlikely(!cached)) { 1600 /* Routes we intend to cache in nexthop exception or 1601 * FIB nexthop have the DST_NOCACHE bit clear. 1602 * However, if we are unsuccessful at storing this 1603 * route into the cache we really need to set it. 1604 */ 1605 if (!rt->rt_gw4) { 1606 rt->rt_gw_family = AF_INET; 1607 rt->rt_gw4 = daddr; 1608 } 1609 rt_add_uncached_list(rt); 1610 } 1611 } else 1612 rt_add_uncached_list(rt); 1613 1614 #ifdef CONFIG_IP_ROUTE_CLASSID 1615 #ifdef CONFIG_IP_MULTIPLE_TABLES 1616 set_class_tag(rt, res->tclassid); 1617 #endif 1618 set_class_tag(rt, itag); 1619 #endif 1620 } 1621 1622 struct rtable *rt_dst_alloc(struct net_device *dev, 1623 unsigned int flags, u16 type, 1624 bool nopolicy, bool noxfrm) 1625 { 1626 struct rtable *rt; 1627 1628 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1629 (nopolicy ? DST_NOPOLICY : 0) | 1630 (noxfrm ? DST_NOXFRM : 0)); 1631 1632 if (rt) { 1633 rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1634 rt->rt_flags = flags; 1635 rt->rt_type = type; 1636 rt->rt_is_input = 0; 1637 rt->rt_iif = 0; 1638 rt->rt_pmtu = 0; 1639 rt->rt_mtu_locked = 0; 1640 rt->rt_uses_gateway = 0; 1641 rt->rt_gw_family = 0; 1642 rt->rt_gw4 = 0; 1643 INIT_LIST_HEAD(&rt->rt_uncached); 1644 1645 rt->dst.output = ip_output; 1646 if (flags & RTCF_LOCAL) 1647 rt->dst.input = ip_local_deliver; 1648 } 1649 1650 return rt; 1651 } 1652 EXPORT_SYMBOL(rt_dst_alloc); 1653 1654 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt) 1655 { 1656 struct rtable *new_rt; 1657 1658 new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1659 rt->dst.flags); 1660 1661 if (new_rt) { 1662 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1663 new_rt->rt_flags = rt->rt_flags; 1664 new_rt->rt_type = rt->rt_type; 1665 new_rt->rt_is_input = rt->rt_is_input; 1666 new_rt->rt_iif = rt->rt_iif; 1667 new_rt->rt_pmtu = rt->rt_pmtu; 1668 new_rt->rt_mtu_locked = rt->rt_mtu_locked; 1669 new_rt->rt_gw_family = rt->rt_gw_family; 1670 if (rt->rt_gw_family == AF_INET) 1671 new_rt->rt_gw4 = rt->rt_gw4; 1672 else if (rt->rt_gw_family == AF_INET6) 1673 new_rt->rt_gw6 = rt->rt_gw6; 1674 INIT_LIST_HEAD(&new_rt->rt_uncached); 1675 1676 new_rt->dst.input = rt->dst.input; 1677 new_rt->dst.output = rt->dst.output; 1678 new_rt->dst.error = rt->dst.error; 1679 new_rt->dst.lastuse = jiffies; 1680 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate); 1681 } 1682 return new_rt; 1683 } 1684 EXPORT_SYMBOL(rt_dst_clone); 1685 1686 /* called in rcu_read_lock() section */ 1687 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1688 u8 tos, struct net_device *dev, 1689 struct in_device *in_dev, u32 *itag) 1690 { 1691 int err; 1692 1693 /* Primary sanity checks. */ 1694 if (!in_dev) 1695 return -EINVAL; 1696 1697 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) || 1698 skb->protocol != htons(ETH_P_IP)) 1699 return -EINVAL; 1700 1701 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev)) 1702 return -EINVAL; 1703 1704 if (ipv4_is_zeronet(saddr)) { 1705 if (!ipv4_is_local_multicast(daddr) && 1706 ip_hdr(skb)->protocol != IPPROTO_IGMP) 1707 return -EINVAL; 1708 } else { 1709 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 1710 in_dev, itag); 1711 if (err < 0) 1712 return err; 1713 } 1714 return 0; 1715 } 1716 1717 /* called in rcu_read_lock() section */ 1718 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1719 u8 tos, struct net_device *dev, int our) 1720 { 1721 struct in_device *in_dev = __in_dev_get_rcu(dev); 1722 unsigned int flags = RTCF_MULTICAST; 1723 struct rtable *rth; 1724 u32 itag = 0; 1725 int err; 1726 1727 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag); 1728 if (err) 1729 return err; 1730 1731 if (our) 1732 flags |= RTCF_LOCAL; 1733 1734 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST, 1735 IN_DEV_CONF_GET(in_dev, NOPOLICY), false); 1736 if (!rth) 1737 return -ENOBUFS; 1738 1739 #ifdef CONFIG_IP_ROUTE_CLASSID 1740 rth->dst.tclassid = itag; 1741 #endif 1742 rth->dst.output = ip_rt_bug; 1743 rth->rt_is_input= 1; 1744 1745 #ifdef CONFIG_IP_MROUTE 1746 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1747 rth->dst.input = ip_mr_input; 1748 #endif 1749 RT_CACHE_STAT_INC(in_slow_mc); 1750 1751 skb_dst_set(skb, &rth->dst); 1752 return 0; 1753 } 1754 1755 1756 static void ip_handle_martian_source(struct net_device *dev, 1757 struct in_device *in_dev, 1758 struct sk_buff *skb, 1759 __be32 daddr, 1760 __be32 saddr) 1761 { 1762 RT_CACHE_STAT_INC(in_martian_src); 1763 #ifdef CONFIG_IP_ROUTE_VERBOSE 1764 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1765 /* 1766 * RFC1812 recommendation, if source is martian, 1767 * the only hint is MAC header. 1768 */ 1769 pr_warn("martian source %pI4 from %pI4, on dev %s\n", 1770 &daddr, &saddr, dev->name); 1771 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1772 print_hex_dump(KERN_WARNING, "ll header: ", 1773 DUMP_PREFIX_OFFSET, 16, 1, 1774 skb_mac_header(skb), 1775 dev->hard_header_len, false); 1776 } 1777 } 1778 #endif 1779 } 1780 1781 /* called in rcu_read_lock() section */ 1782 static int __mkroute_input(struct sk_buff *skb, 1783 const struct fib_result *res, 1784 struct in_device *in_dev, 1785 __be32 daddr, __be32 saddr, u32 tos) 1786 { 1787 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1788 struct net_device *dev = nhc->nhc_dev; 1789 struct fib_nh_exception *fnhe; 1790 struct rtable *rth; 1791 int err; 1792 struct in_device *out_dev; 1793 bool do_cache; 1794 u32 itag = 0; 1795 1796 /* get a working reference to the output device */ 1797 out_dev = __in_dev_get_rcu(dev); 1798 if (!out_dev) { 1799 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n"); 1800 return -EINVAL; 1801 } 1802 1803 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res), 1804 in_dev->dev, in_dev, &itag); 1805 if (err < 0) { 1806 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1807 saddr); 1808 1809 goto cleanup; 1810 } 1811 1812 do_cache = res->fi && !itag; 1813 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) && 1814 skb->protocol == htons(ETH_P_IP)) { 1815 __be32 gw; 1816 1817 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0; 1818 if (IN_DEV_SHARED_MEDIA(out_dev) || 1819 inet_addr_onlink(out_dev, saddr, gw)) 1820 IPCB(skb)->flags |= IPSKB_DOREDIRECT; 1821 } 1822 1823 if (skb->protocol != htons(ETH_P_IP)) { 1824 /* Not IP (i.e. ARP). Do not create route, if it is 1825 * invalid for proxy arp. DNAT routes are always valid. 1826 * 1827 * Proxy arp feature have been extended to allow, ARP 1828 * replies back to the same interface, to support 1829 * Private VLAN switch technologies. See arp.c. 1830 */ 1831 if (out_dev == in_dev && 1832 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 1833 err = -EINVAL; 1834 goto cleanup; 1835 } 1836 } 1837 1838 fnhe = find_exception(nhc, daddr); 1839 if (do_cache) { 1840 if (fnhe) 1841 rth = rcu_dereference(fnhe->fnhe_rth_input); 1842 else 1843 rth = rcu_dereference(nhc->nhc_rth_input); 1844 if (rt_cache_valid(rth)) { 1845 skb_dst_set_noref(skb, &rth->dst); 1846 goto out; 1847 } 1848 } 1849 1850 rth = rt_dst_alloc(out_dev->dev, 0, res->type, 1851 IN_DEV_CONF_GET(in_dev, NOPOLICY), 1852 IN_DEV_CONF_GET(out_dev, NOXFRM)); 1853 if (!rth) { 1854 err = -ENOBUFS; 1855 goto cleanup; 1856 } 1857 1858 rth->rt_is_input = 1; 1859 RT_CACHE_STAT_INC(in_slow_tot); 1860 1861 rth->dst.input = ip_forward; 1862 1863 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag, 1864 do_cache); 1865 lwtunnel_set_redirect(&rth->dst); 1866 skb_dst_set(skb, &rth->dst); 1867 out: 1868 err = 0; 1869 cleanup: 1870 return err; 1871 } 1872 1873 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1874 /* To make ICMP packets follow the right flow, the multipath hash is 1875 * calculated from the inner IP addresses. 1876 */ 1877 static void ip_multipath_l3_keys(const struct sk_buff *skb, 1878 struct flow_keys *hash_keys) 1879 { 1880 const struct iphdr *outer_iph = ip_hdr(skb); 1881 const struct iphdr *key_iph = outer_iph; 1882 const struct iphdr *inner_iph; 1883 const struct icmphdr *icmph; 1884 struct iphdr _inner_iph; 1885 struct icmphdr _icmph; 1886 1887 if (likely(outer_iph->protocol != IPPROTO_ICMP)) 1888 goto out; 1889 1890 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0)) 1891 goto out; 1892 1893 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph), 1894 &_icmph); 1895 if (!icmph) 1896 goto out; 1897 1898 if (!icmp_is_err(icmph->type)) 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 /* skb is currently provided only when forwarding */ 1968 if (skb) { 1969 struct flow_keys keys; 1970 1971 skb_flow_dissect_flow_keys(skb, &keys, 0); 1972 /* Inner can be v4 or v6 */ 1973 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 1974 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1975 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 1976 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 1977 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 1978 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 1979 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 1980 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 1981 hash_keys.tags.flow_label = keys.tags.flow_label; 1982 hash_keys.basic.ip_proto = keys.basic.ip_proto; 1983 } else { 1984 /* Same as case 0 */ 1985 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1986 ip_multipath_l3_keys(skb, &hash_keys); 1987 } 1988 } else { 1989 /* Same as case 0 */ 1990 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1991 hash_keys.addrs.v4addrs.src = fl4->saddr; 1992 hash_keys.addrs.v4addrs.dst = fl4->daddr; 1993 } 1994 break; 1995 } 1996 mhash = flow_hash_from_keys(&hash_keys); 1997 1998 if (multipath_hash) 1999 mhash = jhash_2words(mhash, multipath_hash, 0); 2000 2001 return mhash >> 1; 2002 } 2003 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 2004 2005 static int ip_mkroute_input(struct sk_buff *skb, 2006 struct fib_result *res, 2007 struct in_device *in_dev, 2008 __be32 daddr, __be32 saddr, u32 tos, 2009 struct flow_keys *hkeys) 2010 { 2011 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2012 if (res->fi && fib_info_num_path(res->fi) > 1) { 2013 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys); 2014 2015 fib_select_multipath(res, h); 2016 } 2017 #endif 2018 2019 /* create a routing cache entry */ 2020 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos); 2021 } 2022 2023 /* Implements all the saddr-related checks as ip_route_input_slow(), 2024 * assuming daddr is valid and the destination is not a local broadcast one. 2025 * Uses the provided hint instead of performing a route lookup. 2026 */ 2027 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2028 u8 tos, struct net_device *dev, 2029 const struct sk_buff *hint) 2030 { 2031 struct in_device *in_dev = __in_dev_get_rcu(dev); 2032 struct rtable *rt = (struct rtable *)hint; 2033 struct net *net = dev_net(dev); 2034 int err = -EINVAL; 2035 u32 tag = 0; 2036 2037 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 2038 goto martian_source; 2039 2040 if (ipv4_is_zeronet(saddr)) 2041 goto martian_source; 2042 2043 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2044 goto martian_source; 2045 2046 if (rt->rt_type != RTN_LOCAL) 2047 goto skip_validate_source; 2048 2049 tos &= IPTOS_RT_MASK; 2050 err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag); 2051 if (err < 0) 2052 goto martian_source; 2053 2054 skip_validate_source: 2055 skb_dst_copy(skb, hint); 2056 return 0; 2057 2058 martian_source: 2059 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2060 return err; 2061 } 2062 2063 /* 2064 * NOTE. We drop all the packets that has local source 2065 * addresses, because every properly looped back packet 2066 * must have correct destination already attached by output routine. 2067 * Changes in the enforced policies must be applied also to 2068 * ip_route_use_hint(). 2069 * 2070 * Such approach solves two big problems: 2071 * 1. Not simplex devices are handled properly. 2072 * 2. IP spoofing attempts are filtered with 100% of guarantee. 2073 * called with rcu_read_lock() 2074 */ 2075 2076 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2077 u8 tos, struct net_device *dev, 2078 struct fib_result *res) 2079 { 2080 struct in_device *in_dev = __in_dev_get_rcu(dev); 2081 struct flow_keys *flkeys = NULL, _flkeys; 2082 struct net *net = dev_net(dev); 2083 struct ip_tunnel_info *tun_info; 2084 int err = -EINVAL; 2085 unsigned int flags = 0; 2086 u32 itag = 0; 2087 struct rtable *rth; 2088 struct flowi4 fl4; 2089 bool do_cache = true; 2090 2091 /* IP on this device is disabled. */ 2092 2093 if (!in_dev) 2094 goto out; 2095 2096 /* Check for the most weird martians, which can be not detected 2097 by fib_lookup. 2098 */ 2099 2100 tun_info = skb_tunnel_info(skb); 2101 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX)) 2102 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id; 2103 else 2104 fl4.flowi4_tun_key.tun_id = 0; 2105 skb_dst_drop(skb); 2106 2107 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 2108 goto martian_source; 2109 2110 res->fi = NULL; 2111 res->table = NULL; 2112 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 2113 goto brd_input; 2114 2115 /* Accept zero addresses only to limited broadcast; 2116 * I even do not know to fix it or not. Waiting for complains :-) 2117 */ 2118 if (ipv4_is_zeronet(saddr)) 2119 goto martian_source; 2120 2121 if (ipv4_is_zeronet(daddr)) 2122 goto martian_destination; 2123 2124 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(), 2125 * and call it once if daddr or/and saddr are loopback addresses 2126 */ 2127 if (ipv4_is_loopback(daddr)) { 2128 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2129 goto martian_destination; 2130 } else if (ipv4_is_loopback(saddr)) { 2131 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2132 goto martian_source; 2133 } 2134 2135 /* 2136 * Now we are ready to route packet. 2137 */ 2138 fl4.flowi4_oif = 0; 2139 fl4.flowi4_iif = dev->ifindex; 2140 fl4.flowi4_mark = skb->mark; 2141 fl4.flowi4_tos = tos; 2142 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 2143 fl4.flowi4_flags = 0; 2144 fl4.daddr = daddr; 2145 fl4.saddr = saddr; 2146 fl4.flowi4_uid = sock_net_uid(net, NULL); 2147 2148 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) { 2149 flkeys = &_flkeys; 2150 } else { 2151 fl4.flowi4_proto = 0; 2152 fl4.fl4_sport = 0; 2153 fl4.fl4_dport = 0; 2154 } 2155 2156 err = fib_lookup(net, &fl4, res, 0); 2157 if (err != 0) { 2158 if (!IN_DEV_FORWARD(in_dev)) 2159 err = -EHOSTUNREACH; 2160 goto no_route; 2161 } 2162 2163 if (res->type == RTN_BROADCAST) { 2164 if (IN_DEV_BFORWARD(in_dev)) 2165 goto make_route; 2166 /* not do cache if bc_forwarding is enabled */ 2167 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING)) 2168 do_cache = false; 2169 goto brd_input; 2170 } 2171 2172 if (res->type == RTN_LOCAL) { 2173 err = fib_validate_source(skb, saddr, daddr, tos, 2174 0, dev, in_dev, &itag); 2175 if (err < 0) 2176 goto martian_source; 2177 goto local_input; 2178 } 2179 2180 if (!IN_DEV_FORWARD(in_dev)) { 2181 err = -EHOSTUNREACH; 2182 goto no_route; 2183 } 2184 if (res->type != RTN_UNICAST) 2185 goto martian_destination; 2186 2187 make_route: 2188 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys); 2189 out: return err; 2190 2191 brd_input: 2192 if (skb->protocol != htons(ETH_P_IP)) 2193 goto e_inval; 2194 2195 if (!ipv4_is_zeronet(saddr)) { 2196 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 2197 in_dev, &itag); 2198 if (err < 0) 2199 goto martian_source; 2200 } 2201 flags |= RTCF_BROADCAST; 2202 res->type = RTN_BROADCAST; 2203 RT_CACHE_STAT_INC(in_brd); 2204 2205 local_input: 2206 do_cache &= res->fi && !itag; 2207 if (do_cache) { 2208 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2209 2210 rth = rcu_dereference(nhc->nhc_rth_input); 2211 if (rt_cache_valid(rth)) { 2212 skb_dst_set_noref(skb, &rth->dst); 2213 err = 0; 2214 goto out; 2215 } 2216 } 2217 2218 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev, 2219 flags | RTCF_LOCAL, res->type, 2220 IN_DEV_CONF_GET(in_dev, NOPOLICY), false); 2221 if (!rth) 2222 goto e_nobufs; 2223 2224 rth->dst.output= ip_rt_bug; 2225 #ifdef CONFIG_IP_ROUTE_CLASSID 2226 rth->dst.tclassid = itag; 2227 #endif 2228 rth->rt_is_input = 1; 2229 2230 RT_CACHE_STAT_INC(in_slow_tot); 2231 if (res->type == RTN_UNREACHABLE) { 2232 rth->dst.input= ip_error; 2233 rth->dst.error= -err; 2234 rth->rt_flags &= ~RTCF_LOCAL; 2235 } 2236 2237 if (do_cache) { 2238 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2239 2240 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 2241 if (lwtunnel_input_redirect(rth->dst.lwtstate)) { 2242 WARN_ON(rth->dst.input == lwtunnel_input); 2243 rth->dst.lwtstate->orig_input = rth->dst.input; 2244 rth->dst.input = lwtunnel_input; 2245 } 2246 2247 if (unlikely(!rt_cache_route(nhc, rth))) 2248 rt_add_uncached_list(rth); 2249 } 2250 skb_dst_set(skb, &rth->dst); 2251 err = 0; 2252 goto out; 2253 2254 no_route: 2255 RT_CACHE_STAT_INC(in_no_route); 2256 res->type = RTN_UNREACHABLE; 2257 res->fi = NULL; 2258 res->table = NULL; 2259 goto local_input; 2260 2261 /* 2262 * Do not cache martian addresses: they should be logged (RFC1812) 2263 */ 2264 martian_destination: 2265 RT_CACHE_STAT_INC(in_martian_dst); 2266 #ifdef CONFIG_IP_ROUTE_VERBOSE 2267 if (IN_DEV_LOG_MARTIANS(in_dev)) 2268 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n", 2269 &daddr, &saddr, dev->name); 2270 #endif 2271 2272 e_inval: 2273 err = -EINVAL; 2274 goto out; 2275 2276 e_nobufs: 2277 err = -ENOBUFS; 2278 goto out; 2279 2280 martian_source: 2281 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2282 goto out; 2283 } 2284 2285 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2286 u8 tos, struct net_device *dev) 2287 { 2288 struct fib_result res; 2289 int err; 2290 2291 tos &= IPTOS_RT_MASK; 2292 rcu_read_lock(); 2293 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res); 2294 rcu_read_unlock(); 2295 2296 return err; 2297 } 2298 EXPORT_SYMBOL(ip_route_input_noref); 2299 2300 /* called with rcu_read_lock held */ 2301 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2302 u8 tos, struct net_device *dev, struct fib_result *res) 2303 { 2304 /* Multicast recognition logic is moved from route cache to here. 2305 The problem was that too many Ethernet cards have broken/missing 2306 hardware multicast filters :-( As result the host on multicasting 2307 network acquires a lot of useless route cache entries, sort of 2308 SDR messages from all the world. Now we try to get rid of them. 2309 Really, provided software IP multicast filter is organized 2310 reasonably (at least, hashed), it does not result in a slowdown 2311 comparing with route cache reject entries. 2312 Note, that multicast routers are not affected, because 2313 route cache entry is created eventually. 2314 */ 2315 if (ipv4_is_multicast(daddr)) { 2316 struct in_device *in_dev = __in_dev_get_rcu(dev); 2317 int our = 0; 2318 int err = -EINVAL; 2319 2320 if (!in_dev) 2321 return err; 2322 our = ip_check_mc_rcu(in_dev, daddr, saddr, 2323 ip_hdr(skb)->protocol); 2324 2325 /* check l3 master if no match yet */ 2326 if (!our && netif_is_l3_slave(dev)) { 2327 struct in_device *l3_in_dev; 2328 2329 l3_in_dev = __in_dev_get_rcu(skb->dev); 2330 if (l3_in_dev) 2331 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr, 2332 ip_hdr(skb)->protocol); 2333 } 2334 2335 if (our 2336 #ifdef CONFIG_IP_MROUTE 2337 || 2338 (!ipv4_is_local_multicast(daddr) && 2339 IN_DEV_MFORWARD(in_dev)) 2340 #endif 2341 ) { 2342 err = ip_route_input_mc(skb, daddr, saddr, 2343 tos, dev, our); 2344 } 2345 return err; 2346 } 2347 2348 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res); 2349 } 2350 2351 /* called with rcu_read_lock() */ 2352 static struct rtable *__mkroute_output(const struct fib_result *res, 2353 const struct flowi4 *fl4, int orig_oif, 2354 struct net_device *dev_out, 2355 unsigned int flags) 2356 { 2357 struct fib_info *fi = res->fi; 2358 struct fib_nh_exception *fnhe; 2359 struct in_device *in_dev; 2360 u16 type = res->type; 2361 struct rtable *rth; 2362 bool do_cache; 2363 2364 in_dev = __in_dev_get_rcu(dev_out); 2365 if (!in_dev) 2366 return ERR_PTR(-EINVAL); 2367 2368 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 2369 if (ipv4_is_loopback(fl4->saddr) && 2370 !(dev_out->flags & IFF_LOOPBACK) && 2371 !netif_is_l3_master(dev_out)) 2372 return ERR_PTR(-EINVAL); 2373 2374 if (ipv4_is_lbcast(fl4->daddr)) 2375 type = RTN_BROADCAST; 2376 else if (ipv4_is_multicast(fl4->daddr)) 2377 type = RTN_MULTICAST; 2378 else if (ipv4_is_zeronet(fl4->daddr)) 2379 return ERR_PTR(-EINVAL); 2380 2381 if (dev_out->flags & IFF_LOOPBACK) 2382 flags |= RTCF_LOCAL; 2383 2384 do_cache = true; 2385 if (type == RTN_BROADCAST) { 2386 flags |= RTCF_BROADCAST | RTCF_LOCAL; 2387 fi = NULL; 2388 } else if (type == RTN_MULTICAST) { 2389 flags |= RTCF_MULTICAST | RTCF_LOCAL; 2390 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr, 2391 fl4->flowi4_proto)) 2392 flags &= ~RTCF_LOCAL; 2393 else 2394 do_cache = false; 2395 /* If multicast route do not exist use 2396 * default one, but do not gateway in this case. 2397 * Yes, it is hack. 2398 */ 2399 if (fi && res->prefixlen < 4) 2400 fi = NULL; 2401 } else if ((type == RTN_LOCAL) && (orig_oif != 0) && 2402 (orig_oif != dev_out->ifindex)) { 2403 /* For local routes that require a particular output interface 2404 * we do not want to cache the result. Caching the result 2405 * causes incorrect behaviour when there are multiple source 2406 * addresses on the interface, the end result being that if the 2407 * intended recipient is waiting on that interface for the 2408 * packet he won't receive it because it will be delivered on 2409 * the loopback interface and the IP_PKTINFO ipi_ifindex will 2410 * be set to the loopback interface as well. 2411 */ 2412 do_cache = false; 2413 } 2414 2415 fnhe = NULL; 2416 do_cache &= fi != NULL; 2417 if (fi) { 2418 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2419 struct rtable __rcu **prth; 2420 2421 fnhe = find_exception(nhc, fl4->daddr); 2422 if (!do_cache) 2423 goto add; 2424 if (fnhe) { 2425 prth = &fnhe->fnhe_rth_output; 2426 } else { 2427 if (unlikely(fl4->flowi4_flags & 2428 FLOWI_FLAG_KNOWN_NH && 2429 !(nhc->nhc_gw_family && 2430 nhc->nhc_scope == RT_SCOPE_LINK))) { 2431 do_cache = false; 2432 goto add; 2433 } 2434 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 2435 } 2436 rth = rcu_dereference(*prth); 2437 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst)) 2438 return rth; 2439 } 2440 2441 add: 2442 rth = rt_dst_alloc(dev_out, flags, type, 2443 IN_DEV_CONF_GET(in_dev, NOPOLICY), 2444 IN_DEV_CONF_GET(in_dev, NOXFRM)); 2445 if (!rth) 2446 return ERR_PTR(-ENOBUFS); 2447 2448 rth->rt_iif = orig_oif; 2449 2450 RT_CACHE_STAT_INC(out_slow_tot); 2451 2452 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 2453 if (flags & RTCF_LOCAL && 2454 !(dev_out->flags & IFF_LOOPBACK)) { 2455 rth->dst.output = ip_mc_output; 2456 RT_CACHE_STAT_INC(out_slow_mc); 2457 } 2458 #ifdef CONFIG_IP_MROUTE 2459 if (type == RTN_MULTICAST) { 2460 if (IN_DEV_MFORWARD(in_dev) && 2461 !ipv4_is_local_multicast(fl4->daddr)) { 2462 rth->dst.input = ip_mr_input; 2463 rth->dst.output = ip_mc_output; 2464 } 2465 } 2466 #endif 2467 } 2468 2469 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache); 2470 lwtunnel_set_redirect(&rth->dst); 2471 2472 return rth; 2473 } 2474 2475 /* 2476 * Major route resolver routine. 2477 */ 2478 2479 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4, 2480 const struct sk_buff *skb) 2481 { 2482 __u8 tos = RT_FL_TOS(fl4); 2483 struct fib_result res = { 2484 .type = RTN_UNSPEC, 2485 .fi = NULL, 2486 .table = NULL, 2487 .tclassid = 0, 2488 }; 2489 struct rtable *rth; 2490 2491 fl4->flowi4_iif = LOOPBACK_IFINDEX; 2492 fl4->flowi4_tos = tos & IPTOS_RT_MASK; 2493 fl4->flowi4_scope = ((tos & RTO_ONLINK) ? 2494 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE); 2495 2496 rcu_read_lock(); 2497 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb); 2498 rcu_read_unlock(); 2499 2500 return rth; 2501 } 2502 EXPORT_SYMBOL_GPL(ip_route_output_key_hash); 2503 2504 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4, 2505 struct fib_result *res, 2506 const struct sk_buff *skb) 2507 { 2508 struct net_device *dev_out = NULL; 2509 int orig_oif = fl4->flowi4_oif; 2510 unsigned int flags = 0; 2511 struct rtable *rth; 2512 int err; 2513 2514 if (fl4->saddr) { 2515 if (ipv4_is_multicast(fl4->saddr) || 2516 ipv4_is_lbcast(fl4->saddr) || 2517 ipv4_is_zeronet(fl4->saddr)) { 2518 rth = ERR_PTR(-EINVAL); 2519 goto out; 2520 } 2521 2522 rth = ERR_PTR(-ENETUNREACH); 2523 2524 /* I removed check for oif == dev_out->oif here. 2525 It was wrong for two reasons: 2526 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 2527 is assigned to multiple interfaces. 2528 2. Moreover, we are allowed to send packets with saddr 2529 of another iface. --ANK 2530 */ 2531 2532 if (fl4->flowi4_oif == 0 && 2533 (ipv4_is_multicast(fl4->daddr) || 2534 ipv4_is_lbcast(fl4->daddr))) { 2535 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2536 dev_out = __ip_dev_find(net, fl4->saddr, false); 2537 if (!dev_out) 2538 goto out; 2539 2540 /* Special hack: user can direct multicasts 2541 and limited broadcast via necessary interface 2542 without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 2543 This hack is not just for fun, it allows 2544 vic,vat and friends to work. 2545 They bind socket to loopback, set ttl to zero 2546 and expect that it will work. 2547 From the viewpoint of routing cache they are broken, 2548 because we are not allowed to build multicast path 2549 with loopback source addr (look, routing cache 2550 cannot know, that ttl is zero, so that packet 2551 will not leave this host and route is valid). 2552 Luckily, this hack is good workaround. 2553 */ 2554 2555 fl4->flowi4_oif = dev_out->ifindex; 2556 goto make_route; 2557 } 2558 2559 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 2560 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2561 if (!__ip_dev_find(net, fl4->saddr, false)) 2562 goto out; 2563 } 2564 } 2565 2566 2567 if (fl4->flowi4_oif) { 2568 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif); 2569 rth = ERR_PTR(-ENODEV); 2570 if (!dev_out) 2571 goto out; 2572 2573 /* RACE: Check return value of inet_select_addr instead. */ 2574 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 2575 rth = ERR_PTR(-ENETUNREACH); 2576 goto out; 2577 } 2578 if (ipv4_is_local_multicast(fl4->daddr) || 2579 ipv4_is_lbcast(fl4->daddr) || 2580 fl4->flowi4_proto == IPPROTO_IGMP) { 2581 if (!fl4->saddr) 2582 fl4->saddr = inet_select_addr(dev_out, 0, 2583 RT_SCOPE_LINK); 2584 goto make_route; 2585 } 2586 if (!fl4->saddr) { 2587 if (ipv4_is_multicast(fl4->daddr)) 2588 fl4->saddr = inet_select_addr(dev_out, 0, 2589 fl4->flowi4_scope); 2590 else if (!fl4->daddr) 2591 fl4->saddr = inet_select_addr(dev_out, 0, 2592 RT_SCOPE_HOST); 2593 } 2594 } 2595 2596 if (!fl4->daddr) { 2597 fl4->daddr = fl4->saddr; 2598 if (!fl4->daddr) 2599 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK); 2600 dev_out = net->loopback_dev; 2601 fl4->flowi4_oif = LOOPBACK_IFINDEX; 2602 res->type = RTN_LOCAL; 2603 flags |= RTCF_LOCAL; 2604 goto make_route; 2605 } 2606 2607 err = fib_lookup(net, fl4, res, 0); 2608 if (err) { 2609 res->fi = NULL; 2610 res->table = NULL; 2611 if (fl4->flowi4_oif && 2612 (ipv4_is_multicast(fl4->daddr) || 2613 !netif_index_is_l3_master(net, fl4->flowi4_oif))) { 2614 /* Apparently, routing tables are wrong. Assume, 2615 that the destination is on link. 2616 2617 WHY? DW. 2618 Because we are allowed to send to iface 2619 even if it has NO routes and NO assigned 2620 addresses. When oif is specified, routing 2621 tables are looked up with only one purpose: 2622 to catch if destination is gatewayed, rather than 2623 direct. Moreover, if MSG_DONTROUTE is set, 2624 we send packet, ignoring both routing tables 2625 and ifaddr state. --ANK 2626 2627 2628 We could make it even if oif is unknown, 2629 likely IPv6, but we do not. 2630 */ 2631 2632 if (fl4->saddr == 0) 2633 fl4->saddr = inet_select_addr(dev_out, 0, 2634 RT_SCOPE_LINK); 2635 res->type = RTN_UNICAST; 2636 goto make_route; 2637 } 2638 rth = ERR_PTR(err); 2639 goto out; 2640 } 2641 2642 if (res->type == RTN_LOCAL) { 2643 if (!fl4->saddr) { 2644 if (res->fi->fib_prefsrc) 2645 fl4->saddr = res->fi->fib_prefsrc; 2646 else 2647 fl4->saddr = fl4->daddr; 2648 } 2649 2650 /* L3 master device is the loopback for that domain */ 2651 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? : 2652 net->loopback_dev; 2653 2654 /* make sure orig_oif points to fib result device even 2655 * though packet rx/tx happens over loopback or l3mdev 2656 */ 2657 orig_oif = FIB_RES_OIF(*res); 2658 2659 fl4->flowi4_oif = dev_out->ifindex; 2660 flags |= RTCF_LOCAL; 2661 goto make_route; 2662 } 2663 2664 fib_select_path(net, res, fl4, skb); 2665 2666 dev_out = FIB_RES_DEV(*res); 2667 fl4->flowi4_oif = dev_out->ifindex; 2668 2669 2670 make_route: 2671 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags); 2672 2673 out: 2674 return rth; 2675 } 2676 2677 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie) 2678 { 2679 return NULL; 2680 } 2681 2682 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst) 2683 { 2684 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU); 2685 2686 return mtu ? : dst->dev->mtu; 2687 } 2688 2689 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk, 2690 struct sk_buff *skb, u32 mtu, 2691 bool confirm_neigh) 2692 { 2693 } 2694 2695 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk, 2696 struct sk_buff *skb) 2697 { 2698 } 2699 2700 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst, 2701 unsigned long old) 2702 { 2703 return NULL; 2704 } 2705 2706 static struct dst_ops ipv4_dst_blackhole_ops = { 2707 .family = AF_INET, 2708 .check = ipv4_blackhole_dst_check, 2709 .mtu = ipv4_blackhole_mtu, 2710 .default_advmss = ipv4_default_advmss, 2711 .update_pmtu = ipv4_rt_blackhole_update_pmtu, 2712 .redirect = ipv4_rt_blackhole_redirect, 2713 .cow_metrics = ipv4_rt_blackhole_cow_metrics, 2714 .neigh_lookup = ipv4_neigh_lookup, 2715 }; 2716 2717 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2718 { 2719 struct rtable *ort = (struct rtable *) dst_orig; 2720 struct rtable *rt; 2721 2722 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0); 2723 if (rt) { 2724 struct dst_entry *new = &rt->dst; 2725 2726 new->__use = 1; 2727 new->input = dst_discard; 2728 new->output = dst_discard_out; 2729 2730 new->dev = net->loopback_dev; 2731 if (new->dev) 2732 dev_hold(new->dev); 2733 2734 rt->rt_is_input = ort->rt_is_input; 2735 rt->rt_iif = ort->rt_iif; 2736 rt->rt_pmtu = ort->rt_pmtu; 2737 rt->rt_mtu_locked = ort->rt_mtu_locked; 2738 2739 rt->rt_genid = rt_genid_ipv4(net); 2740 rt->rt_flags = ort->rt_flags; 2741 rt->rt_type = ort->rt_type; 2742 rt->rt_uses_gateway = ort->rt_uses_gateway; 2743 rt->rt_gw_family = ort->rt_gw_family; 2744 if (rt->rt_gw_family == AF_INET) 2745 rt->rt_gw4 = ort->rt_gw4; 2746 else if (rt->rt_gw_family == AF_INET6) 2747 rt->rt_gw6 = ort->rt_gw6; 2748 2749 INIT_LIST_HEAD(&rt->rt_uncached); 2750 } 2751 2752 dst_release(dst_orig); 2753 2754 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2755 } 2756 2757 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2758 const struct sock *sk) 2759 { 2760 struct rtable *rt = __ip_route_output_key(net, flp4); 2761 2762 if (IS_ERR(rt)) 2763 return rt; 2764 2765 if (flp4->flowi4_proto) 2766 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst, 2767 flowi4_to_flowi(flp4), 2768 sk, 0); 2769 2770 return rt; 2771 } 2772 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2773 2774 struct rtable *ip_route_output_tunnel(struct sk_buff *skb, 2775 struct net_device *dev, 2776 struct net *net, __be32 *saddr, 2777 const struct ip_tunnel_info *info, 2778 u8 protocol, bool use_cache) 2779 { 2780 #ifdef CONFIG_DST_CACHE 2781 struct dst_cache *dst_cache; 2782 #endif 2783 struct rtable *rt = NULL; 2784 struct flowi4 fl4; 2785 __u8 tos; 2786 2787 #ifdef CONFIG_DST_CACHE 2788 dst_cache = (struct dst_cache *)&info->dst_cache; 2789 if (use_cache) { 2790 rt = dst_cache_get_ip4(dst_cache, saddr); 2791 if (rt) 2792 return rt; 2793 } 2794 #endif 2795 memset(&fl4, 0, sizeof(fl4)); 2796 fl4.flowi4_mark = skb->mark; 2797 fl4.flowi4_proto = protocol; 2798 fl4.daddr = info->key.u.ipv4.dst; 2799 fl4.saddr = info->key.u.ipv4.src; 2800 tos = info->key.tos; 2801 fl4.flowi4_tos = RT_TOS(tos); 2802 2803 rt = ip_route_output_key(net, &fl4); 2804 if (IS_ERR(rt)) { 2805 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr); 2806 return ERR_PTR(-ENETUNREACH); 2807 } 2808 if (rt->dst.dev == dev) { /* is this necessary? */ 2809 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr); 2810 ip_rt_put(rt); 2811 return ERR_PTR(-ELOOP); 2812 } 2813 #ifdef CONFIG_DST_CACHE 2814 if (use_cache) 2815 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr); 2816 #endif 2817 *saddr = fl4.saddr; 2818 return rt; 2819 } 2820 EXPORT_SYMBOL_GPL(ip_route_output_tunnel); 2821 2822 /* called with rcu_read_lock held */ 2823 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, 2824 struct rtable *rt, u32 table_id, struct flowi4 *fl4, 2825 struct sk_buff *skb, u32 portid, u32 seq, 2826 unsigned int flags) 2827 { 2828 struct rtmsg *r; 2829 struct nlmsghdr *nlh; 2830 unsigned long expires = 0; 2831 u32 error; 2832 u32 metrics[RTAX_MAX]; 2833 2834 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags); 2835 if (!nlh) 2836 return -EMSGSIZE; 2837 2838 r = nlmsg_data(nlh); 2839 r->rtm_family = AF_INET; 2840 r->rtm_dst_len = 32; 2841 r->rtm_src_len = 0; 2842 r->rtm_tos = fl4 ? fl4->flowi4_tos : 0; 2843 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT; 2844 if (nla_put_u32(skb, RTA_TABLE, table_id)) 2845 goto nla_put_failure; 2846 r->rtm_type = rt->rt_type; 2847 r->rtm_scope = RT_SCOPE_UNIVERSE; 2848 r->rtm_protocol = RTPROT_UNSPEC; 2849 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 2850 if (rt->rt_flags & RTCF_NOTIFY) 2851 r->rtm_flags |= RTM_F_NOTIFY; 2852 if (IPCB(skb)->flags & IPSKB_DOREDIRECT) 2853 r->rtm_flags |= RTCF_DOREDIRECT; 2854 2855 if (nla_put_in_addr(skb, RTA_DST, dst)) 2856 goto nla_put_failure; 2857 if (src) { 2858 r->rtm_src_len = 32; 2859 if (nla_put_in_addr(skb, RTA_SRC, src)) 2860 goto nla_put_failure; 2861 } 2862 if (rt->dst.dev && 2863 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex)) 2864 goto nla_put_failure; 2865 #ifdef CONFIG_IP_ROUTE_CLASSID 2866 if (rt->dst.tclassid && 2867 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid)) 2868 goto nla_put_failure; 2869 #endif 2870 if (fl4 && !rt_is_input_route(rt) && 2871 fl4->saddr != src) { 2872 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr)) 2873 goto nla_put_failure; 2874 } 2875 if (rt->rt_uses_gateway) { 2876 if (rt->rt_gw_family == AF_INET && 2877 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) { 2878 goto nla_put_failure; 2879 } else if (rt->rt_gw_family == AF_INET6) { 2880 int alen = sizeof(struct in6_addr); 2881 struct nlattr *nla; 2882 struct rtvia *via; 2883 2884 nla = nla_reserve(skb, RTA_VIA, alen + 2); 2885 if (!nla) 2886 goto nla_put_failure; 2887 2888 via = nla_data(nla); 2889 via->rtvia_family = AF_INET6; 2890 memcpy(via->rtvia_addr, &rt->rt_gw6, alen); 2891 } 2892 } 2893 2894 expires = rt->dst.expires; 2895 if (expires) { 2896 unsigned long now = jiffies; 2897 2898 if (time_before(now, expires)) 2899 expires -= now; 2900 else 2901 expires = 0; 2902 } 2903 2904 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics)); 2905 if (rt->rt_pmtu && expires) 2906 metrics[RTAX_MTU - 1] = rt->rt_pmtu; 2907 if (rt->rt_mtu_locked && expires) 2908 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU); 2909 if (rtnetlink_put_metrics(skb, metrics) < 0) 2910 goto nla_put_failure; 2911 2912 if (fl4) { 2913 if (fl4->flowi4_mark && 2914 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark)) 2915 goto nla_put_failure; 2916 2917 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) && 2918 nla_put_u32(skb, RTA_UID, 2919 from_kuid_munged(current_user_ns(), 2920 fl4->flowi4_uid))) 2921 goto nla_put_failure; 2922 2923 if (rt_is_input_route(rt)) { 2924 #ifdef CONFIG_IP_MROUTE 2925 if (ipv4_is_multicast(dst) && 2926 !ipv4_is_local_multicast(dst) && 2927 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) { 2928 int err = ipmr_get_route(net, skb, 2929 fl4->saddr, fl4->daddr, 2930 r, portid); 2931 2932 if (err <= 0) { 2933 if (err == 0) 2934 return 0; 2935 goto nla_put_failure; 2936 } 2937 } else 2938 #endif 2939 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif)) 2940 goto nla_put_failure; 2941 } 2942 } 2943 2944 error = rt->dst.error; 2945 2946 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0) 2947 goto nla_put_failure; 2948 2949 nlmsg_end(skb, nlh); 2950 return 0; 2951 2952 nla_put_failure: 2953 nlmsg_cancel(skb, nlh); 2954 return -EMSGSIZE; 2955 } 2956 2957 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb, 2958 struct netlink_callback *cb, u32 table_id, 2959 struct fnhe_hash_bucket *bucket, int genid, 2960 int *fa_index, int fa_start, unsigned int flags) 2961 { 2962 int i; 2963 2964 for (i = 0; i < FNHE_HASH_SIZE; i++) { 2965 struct fib_nh_exception *fnhe; 2966 2967 for (fnhe = rcu_dereference(bucket[i].chain); fnhe; 2968 fnhe = rcu_dereference(fnhe->fnhe_next)) { 2969 struct rtable *rt; 2970 int err; 2971 2972 if (*fa_index < fa_start) 2973 goto next; 2974 2975 if (fnhe->fnhe_genid != genid) 2976 goto next; 2977 2978 if (fnhe->fnhe_expires && 2979 time_after(jiffies, fnhe->fnhe_expires)) 2980 goto next; 2981 2982 rt = rcu_dereference(fnhe->fnhe_rth_input); 2983 if (!rt) 2984 rt = rcu_dereference(fnhe->fnhe_rth_output); 2985 if (!rt) 2986 goto next; 2987 2988 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt, 2989 table_id, NULL, skb, 2990 NETLINK_CB(cb->skb).portid, 2991 cb->nlh->nlmsg_seq, flags); 2992 if (err) 2993 return err; 2994 next: 2995 (*fa_index)++; 2996 } 2997 } 2998 2999 return 0; 3000 } 3001 3002 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb, 3003 u32 table_id, struct fib_info *fi, 3004 int *fa_index, int fa_start, unsigned int flags) 3005 { 3006 struct net *net = sock_net(cb->skb->sk); 3007 int nhsel, genid = fnhe_genid(net); 3008 3009 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) { 3010 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel); 3011 struct fnhe_hash_bucket *bucket; 3012 int err; 3013 3014 if (nhc->nhc_flags & RTNH_F_DEAD) 3015 continue; 3016 3017 rcu_read_lock(); 3018 bucket = rcu_dereference(nhc->nhc_exceptions); 3019 err = 0; 3020 if (bucket) 3021 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket, 3022 genid, fa_index, fa_start, 3023 flags); 3024 rcu_read_unlock(); 3025 if (err) 3026 return err; 3027 } 3028 3029 return 0; 3030 } 3031 3032 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst, 3033 u8 ip_proto, __be16 sport, 3034 __be16 dport) 3035 { 3036 struct sk_buff *skb; 3037 struct iphdr *iph; 3038 3039 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 3040 if (!skb) 3041 return NULL; 3042 3043 /* Reserve room for dummy headers, this skb can pass 3044 * through good chunk of routing engine. 3045 */ 3046 skb_reset_mac_header(skb); 3047 skb_reset_network_header(skb); 3048 skb->protocol = htons(ETH_P_IP); 3049 iph = skb_put(skb, sizeof(struct iphdr)); 3050 iph->protocol = ip_proto; 3051 iph->saddr = src; 3052 iph->daddr = dst; 3053 iph->version = 0x4; 3054 iph->frag_off = 0; 3055 iph->ihl = 0x5; 3056 skb_set_transport_header(skb, skb->len); 3057 3058 switch (iph->protocol) { 3059 case IPPROTO_UDP: { 3060 struct udphdr *udph; 3061 3062 udph = skb_put_zero(skb, sizeof(struct udphdr)); 3063 udph->source = sport; 3064 udph->dest = dport; 3065 udph->len = sizeof(struct udphdr); 3066 udph->check = 0; 3067 break; 3068 } 3069 case IPPROTO_TCP: { 3070 struct tcphdr *tcph; 3071 3072 tcph = skb_put_zero(skb, sizeof(struct tcphdr)); 3073 tcph->source = sport; 3074 tcph->dest = dport; 3075 tcph->doff = sizeof(struct tcphdr) / 4; 3076 tcph->rst = 1; 3077 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr), 3078 src, dst, 0); 3079 break; 3080 } 3081 case IPPROTO_ICMP: { 3082 struct icmphdr *icmph; 3083 3084 icmph = skb_put_zero(skb, sizeof(struct icmphdr)); 3085 icmph->type = ICMP_ECHO; 3086 icmph->code = 0; 3087 } 3088 } 3089 3090 return skb; 3091 } 3092 3093 static int inet_rtm_valid_getroute_req(struct sk_buff *skb, 3094 const struct nlmsghdr *nlh, 3095 struct nlattr **tb, 3096 struct netlink_ext_ack *extack) 3097 { 3098 struct rtmsg *rtm; 3099 int i, err; 3100 3101 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) { 3102 NL_SET_ERR_MSG(extack, 3103 "ipv4: Invalid header for route get request"); 3104 return -EINVAL; 3105 } 3106 3107 if (!netlink_strict_get_check(skb)) 3108 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX, 3109 rtm_ipv4_policy, extack); 3110 3111 rtm = nlmsg_data(nlh); 3112 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) || 3113 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) || 3114 rtm->rtm_table || rtm->rtm_protocol || 3115 rtm->rtm_scope || rtm->rtm_type) { 3116 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request"); 3117 return -EINVAL; 3118 } 3119 3120 if (rtm->rtm_flags & ~(RTM_F_NOTIFY | 3121 RTM_F_LOOKUP_TABLE | 3122 RTM_F_FIB_MATCH)) { 3123 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request"); 3124 return -EINVAL; 3125 } 3126 3127 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX, 3128 rtm_ipv4_policy, extack); 3129 if (err) 3130 return err; 3131 3132 if ((tb[RTA_SRC] && !rtm->rtm_src_len) || 3133 (tb[RTA_DST] && !rtm->rtm_dst_len)) { 3134 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4"); 3135 return -EINVAL; 3136 } 3137 3138 for (i = 0; i <= RTA_MAX; i++) { 3139 if (!tb[i]) 3140 continue; 3141 3142 switch (i) { 3143 case RTA_IIF: 3144 case RTA_OIF: 3145 case RTA_SRC: 3146 case RTA_DST: 3147 case RTA_IP_PROTO: 3148 case RTA_SPORT: 3149 case RTA_DPORT: 3150 case RTA_MARK: 3151 case RTA_UID: 3152 break; 3153 default: 3154 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request"); 3155 return -EINVAL; 3156 } 3157 } 3158 3159 return 0; 3160 } 3161 3162 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, 3163 struct netlink_ext_ack *extack) 3164 { 3165 struct net *net = sock_net(in_skb->sk); 3166 struct nlattr *tb[RTA_MAX+1]; 3167 u32 table_id = RT_TABLE_MAIN; 3168 __be16 sport = 0, dport = 0; 3169 struct fib_result res = {}; 3170 u8 ip_proto = IPPROTO_UDP; 3171 struct rtable *rt = NULL; 3172 struct sk_buff *skb; 3173 struct rtmsg *rtm; 3174 struct flowi4 fl4 = {}; 3175 __be32 dst = 0; 3176 __be32 src = 0; 3177 kuid_t uid; 3178 u32 iif; 3179 int err; 3180 int mark; 3181 3182 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack); 3183 if (err < 0) 3184 return err; 3185 3186 rtm = nlmsg_data(nlh); 3187 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0; 3188 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0; 3189 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0; 3190 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0; 3191 if (tb[RTA_UID]) 3192 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID])); 3193 else 3194 uid = (iif ? INVALID_UID : current_uid()); 3195 3196 if (tb[RTA_IP_PROTO]) { 3197 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO], 3198 &ip_proto, AF_INET, extack); 3199 if (err) 3200 return err; 3201 } 3202 3203 if (tb[RTA_SPORT]) 3204 sport = nla_get_be16(tb[RTA_SPORT]); 3205 3206 if (tb[RTA_DPORT]) 3207 dport = nla_get_be16(tb[RTA_DPORT]); 3208 3209 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport); 3210 if (!skb) 3211 return -ENOBUFS; 3212 3213 fl4.daddr = dst; 3214 fl4.saddr = src; 3215 fl4.flowi4_tos = rtm->rtm_tos; 3216 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0; 3217 fl4.flowi4_mark = mark; 3218 fl4.flowi4_uid = uid; 3219 if (sport) 3220 fl4.fl4_sport = sport; 3221 if (dport) 3222 fl4.fl4_dport = dport; 3223 fl4.flowi4_proto = ip_proto; 3224 3225 rcu_read_lock(); 3226 3227 if (iif) { 3228 struct net_device *dev; 3229 3230 dev = dev_get_by_index_rcu(net, iif); 3231 if (!dev) { 3232 err = -ENODEV; 3233 goto errout_rcu; 3234 } 3235 3236 fl4.flowi4_iif = iif; /* for rt_fill_info */ 3237 skb->dev = dev; 3238 skb->mark = mark; 3239 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos, 3240 dev, &res); 3241 3242 rt = skb_rtable(skb); 3243 if (err == 0 && rt->dst.error) 3244 err = -rt->dst.error; 3245 } else { 3246 fl4.flowi4_iif = LOOPBACK_IFINDEX; 3247 skb->dev = net->loopback_dev; 3248 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb); 3249 err = 0; 3250 if (IS_ERR(rt)) 3251 err = PTR_ERR(rt); 3252 else 3253 skb_dst_set(skb, &rt->dst); 3254 } 3255 3256 if (err) 3257 goto errout_rcu; 3258 3259 if (rtm->rtm_flags & RTM_F_NOTIFY) 3260 rt->rt_flags |= RTCF_NOTIFY; 3261 3262 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE) 3263 table_id = res.table ? res.table->tb_id : 0; 3264 3265 /* reset skb for netlink reply msg */ 3266 skb_trim(skb, 0); 3267 skb_reset_network_header(skb); 3268 skb_reset_transport_header(skb); 3269 skb_reset_mac_header(skb); 3270 3271 if (rtm->rtm_flags & RTM_F_FIB_MATCH) { 3272 struct fib_rt_info fri; 3273 3274 if (!res.fi) { 3275 err = fib_props[res.type].error; 3276 if (!err) 3277 err = -EHOSTUNREACH; 3278 goto errout_rcu; 3279 } 3280 fri.fi = res.fi; 3281 fri.tb_id = table_id; 3282 fri.dst = res.prefix; 3283 fri.dst_len = res.prefixlen; 3284 fri.tos = fl4.flowi4_tos; 3285 fri.type = rt->rt_type; 3286 fri.offload = 0; 3287 fri.trap = 0; 3288 if (res.fa_head) { 3289 struct fib_alias *fa; 3290 3291 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) { 3292 u8 slen = 32 - fri.dst_len; 3293 3294 if (fa->fa_slen == slen && 3295 fa->tb_id == fri.tb_id && 3296 fa->fa_tos == fri.tos && 3297 fa->fa_info == res.fi && 3298 fa->fa_type == fri.type) { 3299 fri.offload = fa->offload; 3300 fri.trap = fa->trap; 3301 break; 3302 } 3303 } 3304 } 3305 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid, 3306 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0); 3307 } else { 3308 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb, 3309 NETLINK_CB(in_skb).portid, 3310 nlh->nlmsg_seq, 0); 3311 } 3312 if (err < 0) 3313 goto errout_rcu; 3314 3315 rcu_read_unlock(); 3316 3317 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 3318 3319 errout_free: 3320 return err; 3321 errout_rcu: 3322 rcu_read_unlock(); 3323 kfree_skb(skb); 3324 goto errout_free; 3325 } 3326 3327 void ip_rt_multicast_event(struct in_device *in_dev) 3328 { 3329 rt_cache_flush(dev_net(in_dev->dev)); 3330 } 3331 3332 #ifdef CONFIG_SYSCTL 3333 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 3334 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 3335 static int ip_rt_gc_elasticity __read_mostly = 8; 3336 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU; 3337 3338 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write, 3339 void __user *buffer, 3340 size_t *lenp, loff_t *ppos) 3341 { 3342 struct net *net = (struct net *)__ctl->extra1; 3343 3344 if (write) { 3345 rt_cache_flush(net); 3346 fnhe_genid_bump(net); 3347 return 0; 3348 } 3349 3350 return -EINVAL; 3351 } 3352 3353 static struct ctl_table ipv4_route_table[] = { 3354 { 3355 .procname = "gc_thresh", 3356 .data = &ipv4_dst_ops.gc_thresh, 3357 .maxlen = sizeof(int), 3358 .mode = 0644, 3359 .proc_handler = proc_dointvec, 3360 }, 3361 { 3362 .procname = "max_size", 3363 .data = &ip_rt_max_size, 3364 .maxlen = sizeof(int), 3365 .mode = 0644, 3366 .proc_handler = proc_dointvec, 3367 }, 3368 { 3369 /* Deprecated. Use gc_min_interval_ms */ 3370 3371 .procname = "gc_min_interval", 3372 .data = &ip_rt_gc_min_interval, 3373 .maxlen = sizeof(int), 3374 .mode = 0644, 3375 .proc_handler = proc_dointvec_jiffies, 3376 }, 3377 { 3378 .procname = "gc_min_interval_ms", 3379 .data = &ip_rt_gc_min_interval, 3380 .maxlen = sizeof(int), 3381 .mode = 0644, 3382 .proc_handler = proc_dointvec_ms_jiffies, 3383 }, 3384 { 3385 .procname = "gc_timeout", 3386 .data = &ip_rt_gc_timeout, 3387 .maxlen = sizeof(int), 3388 .mode = 0644, 3389 .proc_handler = proc_dointvec_jiffies, 3390 }, 3391 { 3392 .procname = "gc_interval", 3393 .data = &ip_rt_gc_interval, 3394 .maxlen = sizeof(int), 3395 .mode = 0644, 3396 .proc_handler = proc_dointvec_jiffies, 3397 }, 3398 { 3399 .procname = "redirect_load", 3400 .data = &ip_rt_redirect_load, 3401 .maxlen = sizeof(int), 3402 .mode = 0644, 3403 .proc_handler = proc_dointvec, 3404 }, 3405 { 3406 .procname = "redirect_number", 3407 .data = &ip_rt_redirect_number, 3408 .maxlen = sizeof(int), 3409 .mode = 0644, 3410 .proc_handler = proc_dointvec, 3411 }, 3412 { 3413 .procname = "redirect_silence", 3414 .data = &ip_rt_redirect_silence, 3415 .maxlen = sizeof(int), 3416 .mode = 0644, 3417 .proc_handler = proc_dointvec, 3418 }, 3419 { 3420 .procname = "error_cost", 3421 .data = &ip_rt_error_cost, 3422 .maxlen = sizeof(int), 3423 .mode = 0644, 3424 .proc_handler = proc_dointvec, 3425 }, 3426 { 3427 .procname = "error_burst", 3428 .data = &ip_rt_error_burst, 3429 .maxlen = sizeof(int), 3430 .mode = 0644, 3431 .proc_handler = proc_dointvec, 3432 }, 3433 { 3434 .procname = "gc_elasticity", 3435 .data = &ip_rt_gc_elasticity, 3436 .maxlen = sizeof(int), 3437 .mode = 0644, 3438 .proc_handler = proc_dointvec, 3439 }, 3440 { 3441 .procname = "mtu_expires", 3442 .data = &ip_rt_mtu_expires, 3443 .maxlen = sizeof(int), 3444 .mode = 0644, 3445 .proc_handler = proc_dointvec_jiffies, 3446 }, 3447 { 3448 .procname = "min_pmtu", 3449 .data = &ip_rt_min_pmtu, 3450 .maxlen = sizeof(int), 3451 .mode = 0644, 3452 .proc_handler = proc_dointvec_minmax, 3453 .extra1 = &ip_min_valid_pmtu, 3454 }, 3455 { 3456 .procname = "min_adv_mss", 3457 .data = &ip_rt_min_advmss, 3458 .maxlen = sizeof(int), 3459 .mode = 0644, 3460 .proc_handler = proc_dointvec, 3461 }, 3462 { } 3463 }; 3464 3465 static const char ipv4_route_flush_procname[] = "flush"; 3466 3467 static struct ctl_table ipv4_route_flush_table[] = { 3468 { 3469 .procname = ipv4_route_flush_procname, 3470 .maxlen = sizeof(int), 3471 .mode = 0200, 3472 .proc_handler = ipv4_sysctl_rtcache_flush, 3473 }, 3474 { }, 3475 }; 3476 3477 static __net_init int sysctl_route_net_init(struct net *net) 3478 { 3479 struct ctl_table *tbl; 3480 3481 tbl = ipv4_route_flush_table; 3482 if (!net_eq(net, &init_net)) { 3483 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL); 3484 if (!tbl) 3485 goto err_dup; 3486 3487 /* Don't export non-whitelisted sysctls to unprivileged users */ 3488 if (net->user_ns != &init_user_ns) { 3489 if (tbl[0].procname != ipv4_route_flush_procname) 3490 tbl[0].procname = NULL; 3491 } 3492 } 3493 tbl[0].extra1 = net; 3494 3495 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl); 3496 if (!net->ipv4.route_hdr) 3497 goto err_reg; 3498 return 0; 3499 3500 err_reg: 3501 if (tbl != ipv4_route_flush_table) 3502 kfree(tbl); 3503 err_dup: 3504 return -ENOMEM; 3505 } 3506 3507 static __net_exit void sysctl_route_net_exit(struct net *net) 3508 { 3509 struct ctl_table *tbl; 3510 3511 tbl = net->ipv4.route_hdr->ctl_table_arg; 3512 unregister_net_sysctl_table(net->ipv4.route_hdr); 3513 BUG_ON(tbl == ipv4_route_flush_table); 3514 kfree(tbl); 3515 } 3516 3517 static __net_initdata struct pernet_operations sysctl_route_ops = { 3518 .init = sysctl_route_net_init, 3519 .exit = sysctl_route_net_exit, 3520 }; 3521 #endif 3522 3523 static __net_init int rt_genid_init(struct net *net) 3524 { 3525 atomic_set(&net->ipv4.rt_genid, 0); 3526 atomic_set(&net->fnhe_genid, 0); 3527 atomic_set(&net->ipv4.dev_addr_genid, get_random_int()); 3528 return 0; 3529 } 3530 3531 static __net_initdata struct pernet_operations rt_genid_ops = { 3532 .init = rt_genid_init, 3533 }; 3534 3535 static int __net_init ipv4_inetpeer_init(struct net *net) 3536 { 3537 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL); 3538 3539 if (!bp) 3540 return -ENOMEM; 3541 inet_peer_base_init(bp); 3542 net->ipv4.peers = bp; 3543 return 0; 3544 } 3545 3546 static void __net_exit ipv4_inetpeer_exit(struct net *net) 3547 { 3548 struct inet_peer_base *bp = net->ipv4.peers; 3549 3550 net->ipv4.peers = NULL; 3551 inetpeer_invalidate_tree(bp); 3552 kfree(bp); 3553 } 3554 3555 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = { 3556 .init = ipv4_inetpeer_init, 3557 .exit = ipv4_inetpeer_exit, 3558 }; 3559 3560 #ifdef CONFIG_IP_ROUTE_CLASSID 3561 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 3562 #endif /* CONFIG_IP_ROUTE_CLASSID */ 3563 3564 int __init ip_rt_init(void) 3565 { 3566 int cpu; 3567 3568 ip_idents = kmalloc_array(IP_IDENTS_SZ, sizeof(*ip_idents), 3569 GFP_KERNEL); 3570 if (!ip_idents) 3571 panic("IP: failed to allocate ip_idents\n"); 3572 3573 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents)); 3574 3575 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL); 3576 if (!ip_tstamps) 3577 panic("IP: failed to allocate ip_tstamps\n"); 3578 3579 for_each_possible_cpu(cpu) { 3580 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 3581 3582 INIT_LIST_HEAD(&ul->head); 3583 spin_lock_init(&ul->lock); 3584 } 3585 #ifdef CONFIG_IP_ROUTE_CLASSID 3586 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 3587 if (!ip_rt_acct) 3588 panic("IP: failed to allocate ip_rt_acct\n"); 3589 #endif 3590 3591 ipv4_dst_ops.kmem_cachep = 3592 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0, 3593 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 3594 3595 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 3596 3597 if (dst_entries_init(&ipv4_dst_ops) < 0) 3598 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 3599 3600 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 3601 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 3602 3603 ipv4_dst_ops.gc_thresh = ~0; 3604 ip_rt_max_size = INT_MAX; 3605 3606 devinet_init(); 3607 ip_fib_init(); 3608 3609 if (ip_rt_proc_init()) 3610 pr_err("Unable to create route proc files\n"); 3611 #ifdef CONFIG_XFRM 3612 xfrm_init(); 3613 xfrm4_init(); 3614 #endif 3615 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, 3616 RTNL_FLAG_DOIT_UNLOCKED); 3617 3618 #ifdef CONFIG_SYSCTL 3619 register_pernet_subsys(&sysctl_route_ops); 3620 #endif 3621 register_pernet_subsys(&rt_genid_ops); 3622 register_pernet_subsys(&ipv4_inetpeer_ops); 3623 return 0; 3624 } 3625 3626 #ifdef CONFIG_SYSCTL 3627 /* 3628 * We really need to sanitize the damn ipv4 init order, then all 3629 * this nonsense will go away. 3630 */ 3631 void __init ip_static_sysctl_init(void) 3632 { 3633 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table); 3634 } 3635 #endif 3636