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 delta = 0; 495 496 if (old != now && cmpxchg(p_tstamp, old, now) == old) 497 delta = prandom_u32_max(now - old); 498 499 /* If UBSAN reports an error there, please make sure your compiler 500 * supports -fno-strict-overflow before reporting it that was a bug 501 * in UBSAN, and it has been fixed in GCC-8. 502 */ 503 return atomic_add_return(segs + delta, p_id) - segs; 504 } 505 EXPORT_SYMBOL(ip_idents_reserve); 506 507 void __ip_select_ident(struct net *net, struct iphdr *iph, int segs) 508 { 509 u32 hash, id; 510 511 /* Note the following code is not safe, but this is okay. */ 512 if (unlikely(siphash_key_is_zero(&net->ipv4.ip_id_key))) 513 get_random_bytes(&net->ipv4.ip_id_key, 514 sizeof(net->ipv4.ip_id_key)); 515 516 hash = siphash_3u32((__force u32)iph->daddr, 517 (__force u32)iph->saddr, 518 iph->protocol, 519 &net->ipv4.ip_id_key); 520 id = ip_idents_reserve(hash, segs); 521 iph->id = htons(id); 522 } 523 EXPORT_SYMBOL(__ip_select_ident); 524 525 static void __build_flow_key(const struct net *net, struct flowi4 *fl4, 526 const struct sock *sk, 527 const struct iphdr *iph, 528 int oif, u8 tos, 529 u8 prot, u32 mark, int flow_flags) 530 { 531 if (sk) { 532 const struct inet_sock *inet = inet_sk(sk); 533 534 oif = sk->sk_bound_dev_if; 535 mark = sk->sk_mark; 536 tos = RT_CONN_FLAGS(sk); 537 prot = inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol; 538 } 539 flowi4_init_output(fl4, oif, mark, tos, 540 RT_SCOPE_UNIVERSE, prot, 541 flow_flags, 542 iph->daddr, iph->saddr, 0, 0, 543 sock_net_uid(net, sk)); 544 } 545 546 static void build_skb_flow_key(struct flowi4 *fl4, const struct sk_buff *skb, 547 const struct sock *sk) 548 { 549 const struct net *net = dev_net(skb->dev); 550 const struct iphdr *iph = ip_hdr(skb); 551 int oif = skb->dev->ifindex; 552 u8 tos = RT_TOS(iph->tos); 553 u8 prot = iph->protocol; 554 u32 mark = skb->mark; 555 556 __build_flow_key(net, fl4, sk, iph, oif, tos, prot, mark, 0); 557 } 558 559 static void build_sk_flow_key(struct flowi4 *fl4, const struct sock *sk) 560 { 561 const struct inet_sock *inet = inet_sk(sk); 562 const struct ip_options_rcu *inet_opt; 563 __be32 daddr = inet->inet_daddr; 564 565 rcu_read_lock(); 566 inet_opt = rcu_dereference(inet->inet_opt); 567 if (inet_opt && inet_opt->opt.srr) 568 daddr = inet_opt->opt.faddr; 569 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark, 570 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 571 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol, 572 inet_sk_flowi_flags(sk), 573 daddr, inet->inet_saddr, 0, 0, sk->sk_uid); 574 rcu_read_unlock(); 575 } 576 577 static void ip_rt_build_flow_key(struct flowi4 *fl4, const struct sock *sk, 578 const struct sk_buff *skb) 579 { 580 if (skb) 581 build_skb_flow_key(fl4, skb, sk); 582 else 583 build_sk_flow_key(fl4, sk); 584 } 585 586 static DEFINE_SPINLOCK(fnhe_lock); 587 588 static void fnhe_flush_routes(struct fib_nh_exception *fnhe) 589 { 590 struct rtable *rt; 591 592 rt = rcu_dereference(fnhe->fnhe_rth_input); 593 if (rt) { 594 RCU_INIT_POINTER(fnhe->fnhe_rth_input, NULL); 595 dst_dev_put(&rt->dst); 596 dst_release(&rt->dst); 597 } 598 rt = rcu_dereference(fnhe->fnhe_rth_output); 599 if (rt) { 600 RCU_INIT_POINTER(fnhe->fnhe_rth_output, NULL); 601 dst_dev_put(&rt->dst); 602 dst_release(&rt->dst); 603 } 604 } 605 606 static struct fib_nh_exception *fnhe_oldest(struct fnhe_hash_bucket *hash) 607 { 608 struct fib_nh_exception *fnhe, *oldest; 609 610 oldest = rcu_dereference(hash->chain); 611 for (fnhe = rcu_dereference(oldest->fnhe_next); fnhe; 612 fnhe = rcu_dereference(fnhe->fnhe_next)) { 613 if (time_before(fnhe->fnhe_stamp, oldest->fnhe_stamp)) 614 oldest = fnhe; 615 } 616 fnhe_flush_routes(oldest); 617 return oldest; 618 } 619 620 static inline u32 fnhe_hashfun(__be32 daddr) 621 { 622 static u32 fnhe_hashrnd __read_mostly; 623 u32 hval; 624 625 net_get_random_once(&fnhe_hashrnd, sizeof(fnhe_hashrnd)); 626 hval = jhash_1word((__force u32) daddr, fnhe_hashrnd); 627 return hash_32(hval, FNHE_HASH_SHIFT); 628 } 629 630 static void fill_route_from_fnhe(struct rtable *rt, struct fib_nh_exception *fnhe) 631 { 632 rt->rt_pmtu = fnhe->fnhe_pmtu; 633 rt->rt_mtu_locked = fnhe->fnhe_mtu_locked; 634 rt->dst.expires = fnhe->fnhe_expires; 635 636 if (fnhe->fnhe_gw) { 637 rt->rt_flags |= RTCF_REDIRECTED; 638 rt->rt_uses_gateway = 1; 639 rt->rt_gw_family = AF_INET; 640 rt->rt_gw4 = fnhe->fnhe_gw; 641 } 642 } 643 644 static void update_or_create_fnhe(struct fib_nh_common *nhc, __be32 daddr, 645 __be32 gw, u32 pmtu, bool lock, 646 unsigned long expires) 647 { 648 struct fnhe_hash_bucket *hash; 649 struct fib_nh_exception *fnhe; 650 struct rtable *rt; 651 u32 genid, hval; 652 unsigned int i; 653 int depth; 654 655 genid = fnhe_genid(dev_net(nhc->nhc_dev)); 656 hval = fnhe_hashfun(daddr); 657 658 spin_lock_bh(&fnhe_lock); 659 660 hash = rcu_dereference(nhc->nhc_exceptions); 661 if (!hash) { 662 hash = kcalloc(FNHE_HASH_SIZE, sizeof(*hash), GFP_ATOMIC); 663 if (!hash) 664 goto out_unlock; 665 rcu_assign_pointer(nhc->nhc_exceptions, hash); 666 } 667 668 hash += hval; 669 670 depth = 0; 671 for (fnhe = rcu_dereference(hash->chain); fnhe; 672 fnhe = rcu_dereference(fnhe->fnhe_next)) { 673 if (fnhe->fnhe_daddr == daddr) 674 break; 675 depth++; 676 } 677 678 if (fnhe) { 679 if (fnhe->fnhe_genid != genid) 680 fnhe->fnhe_genid = genid; 681 if (gw) 682 fnhe->fnhe_gw = gw; 683 if (pmtu) { 684 fnhe->fnhe_pmtu = pmtu; 685 fnhe->fnhe_mtu_locked = lock; 686 } 687 fnhe->fnhe_expires = max(1UL, expires); 688 /* Update all cached dsts too */ 689 rt = rcu_dereference(fnhe->fnhe_rth_input); 690 if (rt) 691 fill_route_from_fnhe(rt, fnhe); 692 rt = rcu_dereference(fnhe->fnhe_rth_output); 693 if (rt) 694 fill_route_from_fnhe(rt, fnhe); 695 } else { 696 if (depth > FNHE_RECLAIM_DEPTH) 697 fnhe = fnhe_oldest(hash); 698 else { 699 fnhe = kzalloc(sizeof(*fnhe), GFP_ATOMIC); 700 if (!fnhe) 701 goto out_unlock; 702 703 fnhe->fnhe_next = hash->chain; 704 rcu_assign_pointer(hash->chain, fnhe); 705 } 706 fnhe->fnhe_genid = genid; 707 fnhe->fnhe_daddr = daddr; 708 fnhe->fnhe_gw = gw; 709 fnhe->fnhe_pmtu = pmtu; 710 fnhe->fnhe_mtu_locked = lock; 711 fnhe->fnhe_expires = max(1UL, expires); 712 713 /* Exception created; mark the cached routes for the nexthop 714 * stale, so anyone caching it rechecks if this exception 715 * applies to them. 716 */ 717 rt = rcu_dereference(nhc->nhc_rth_input); 718 if (rt) 719 rt->dst.obsolete = DST_OBSOLETE_KILL; 720 721 for_each_possible_cpu(i) { 722 struct rtable __rcu **prt; 723 prt = per_cpu_ptr(nhc->nhc_pcpu_rth_output, i); 724 rt = rcu_dereference(*prt); 725 if (rt) 726 rt->dst.obsolete = DST_OBSOLETE_KILL; 727 } 728 } 729 730 fnhe->fnhe_stamp = jiffies; 731 732 out_unlock: 733 spin_unlock_bh(&fnhe_lock); 734 } 735 736 static void __ip_do_redirect(struct rtable *rt, struct sk_buff *skb, struct flowi4 *fl4, 737 bool kill_route) 738 { 739 __be32 new_gw = icmp_hdr(skb)->un.gateway; 740 __be32 old_gw = ip_hdr(skb)->saddr; 741 struct net_device *dev = skb->dev; 742 struct in_device *in_dev; 743 struct fib_result res; 744 struct neighbour *n; 745 struct net *net; 746 747 switch (icmp_hdr(skb)->code & 7) { 748 case ICMP_REDIR_NET: 749 case ICMP_REDIR_NETTOS: 750 case ICMP_REDIR_HOST: 751 case ICMP_REDIR_HOSTTOS: 752 break; 753 754 default: 755 return; 756 } 757 758 if (rt->rt_gw_family != AF_INET || rt->rt_gw4 != old_gw) 759 return; 760 761 in_dev = __in_dev_get_rcu(dev); 762 if (!in_dev) 763 return; 764 765 net = dev_net(dev); 766 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) || 767 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) || 768 ipv4_is_zeronet(new_gw)) 769 goto reject_redirect; 770 771 if (!IN_DEV_SHARED_MEDIA(in_dev)) { 772 if (!inet_addr_onlink(in_dev, new_gw, old_gw)) 773 goto reject_redirect; 774 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) 775 goto reject_redirect; 776 } else { 777 if (inet_addr_type(net, new_gw) != RTN_UNICAST) 778 goto reject_redirect; 779 } 780 781 n = __ipv4_neigh_lookup(rt->dst.dev, new_gw); 782 if (!n) 783 n = neigh_create(&arp_tbl, &new_gw, rt->dst.dev); 784 if (!IS_ERR(n)) { 785 if (!(n->nud_state & NUD_VALID)) { 786 neigh_event_send(n, NULL); 787 } else { 788 if (fib_lookup(net, fl4, &res, 0) == 0) { 789 struct fib_nh_common *nhc = FIB_RES_NHC(res); 790 791 update_or_create_fnhe(nhc, fl4->daddr, new_gw, 792 0, false, 793 jiffies + ip_rt_gc_timeout); 794 } 795 if (kill_route) 796 rt->dst.obsolete = DST_OBSOLETE_KILL; 797 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, n); 798 } 799 neigh_release(n); 800 } 801 return; 802 803 reject_redirect: 804 #ifdef CONFIG_IP_ROUTE_VERBOSE 805 if (IN_DEV_LOG_MARTIANS(in_dev)) { 806 const struct iphdr *iph = (const struct iphdr *) skb->data; 807 __be32 daddr = iph->daddr; 808 __be32 saddr = iph->saddr; 809 810 net_info_ratelimited("Redirect from %pI4 on %s about %pI4 ignored\n" 811 " Advised path = %pI4 -> %pI4\n", 812 &old_gw, dev->name, &new_gw, 813 &saddr, &daddr); 814 } 815 #endif 816 ; 817 } 818 819 static void ip_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb) 820 { 821 struct rtable *rt; 822 struct flowi4 fl4; 823 const struct iphdr *iph = (const struct iphdr *) skb->data; 824 struct net *net = dev_net(skb->dev); 825 int oif = skb->dev->ifindex; 826 u8 tos = RT_TOS(iph->tos); 827 u8 prot = iph->protocol; 828 u32 mark = skb->mark; 829 830 rt = (struct rtable *) dst; 831 832 __build_flow_key(net, &fl4, sk, iph, oif, tos, prot, mark, 0); 833 __ip_do_redirect(rt, skb, &fl4, true); 834 } 835 836 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst) 837 { 838 struct rtable *rt = (struct rtable *)dst; 839 struct dst_entry *ret = dst; 840 841 if (rt) { 842 if (dst->obsolete > 0) { 843 ip_rt_put(rt); 844 ret = NULL; 845 } else if ((rt->rt_flags & RTCF_REDIRECTED) || 846 rt->dst.expires) { 847 ip_rt_put(rt); 848 ret = NULL; 849 } 850 } 851 return ret; 852 } 853 854 /* 855 * Algorithm: 856 * 1. The first ip_rt_redirect_number redirects are sent 857 * with exponential backoff, then we stop sending them at all, 858 * assuming that the host ignores our redirects. 859 * 2. If we did not see packets requiring redirects 860 * during ip_rt_redirect_silence, we assume that the host 861 * forgot redirected route and start to send redirects again. 862 * 863 * This algorithm is much cheaper and more intelligent than dumb load limiting 864 * in icmp.c. 865 * 866 * NOTE. Do not forget to inhibit load limiting for redirects (redundant) 867 * and "frag. need" (breaks PMTU discovery) in icmp.c. 868 */ 869 870 void ip_rt_send_redirect(struct sk_buff *skb) 871 { 872 struct rtable *rt = skb_rtable(skb); 873 struct in_device *in_dev; 874 struct inet_peer *peer; 875 struct net *net; 876 int log_martians; 877 int vif; 878 879 rcu_read_lock(); 880 in_dev = __in_dev_get_rcu(rt->dst.dev); 881 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) { 882 rcu_read_unlock(); 883 return; 884 } 885 log_martians = IN_DEV_LOG_MARTIANS(in_dev); 886 vif = l3mdev_master_ifindex_rcu(rt->dst.dev); 887 rcu_read_unlock(); 888 889 net = dev_net(rt->dst.dev); 890 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, vif, 1); 891 if (!peer) { 892 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, 893 rt_nexthop(rt, ip_hdr(skb)->daddr)); 894 return; 895 } 896 897 /* No redirected packets during ip_rt_redirect_silence; 898 * reset the algorithm. 899 */ 900 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) { 901 peer->rate_tokens = 0; 902 peer->n_redirects = 0; 903 } 904 905 /* Too many ignored redirects; do not send anything 906 * set dst.rate_last to the last seen redirected packet. 907 */ 908 if (peer->n_redirects >= ip_rt_redirect_number) { 909 peer->rate_last = jiffies; 910 goto out_put_peer; 911 } 912 913 /* Check for load limit; set rate_last to the latest sent 914 * redirect. 915 */ 916 if (peer->n_redirects == 0 || 917 time_after(jiffies, 918 (peer->rate_last + 919 (ip_rt_redirect_load << peer->n_redirects)))) { 920 __be32 gw = rt_nexthop(rt, ip_hdr(skb)->daddr); 921 922 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, gw); 923 peer->rate_last = jiffies; 924 ++peer->n_redirects; 925 #ifdef CONFIG_IP_ROUTE_VERBOSE 926 if (log_martians && 927 peer->n_redirects == ip_rt_redirect_number) 928 net_warn_ratelimited("host %pI4/if%d ignores redirects for %pI4 to %pI4\n", 929 &ip_hdr(skb)->saddr, inet_iif(skb), 930 &ip_hdr(skb)->daddr, &gw); 931 #endif 932 } 933 out_put_peer: 934 inet_putpeer(peer); 935 } 936 937 static int ip_error(struct sk_buff *skb) 938 { 939 struct rtable *rt = skb_rtable(skb); 940 struct net_device *dev = skb->dev; 941 struct in_device *in_dev; 942 struct inet_peer *peer; 943 unsigned long now; 944 struct net *net; 945 bool send; 946 int code; 947 948 if (netif_is_l3_master(skb->dev)) { 949 dev = __dev_get_by_index(dev_net(skb->dev), IPCB(skb)->iif); 950 if (!dev) 951 goto out; 952 } 953 954 in_dev = __in_dev_get_rcu(dev); 955 956 /* IP on this device is disabled. */ 957 if (!in_dev) 958 goto out; 959 960 net = dev_net(rt->dst.dev); 961 if (!IN_DEV_FORWARD(in_dev)) { 962 switch (rt->dst.error) { 963 case EHOSTUNREACH: 964 __IP_INC_STATS(net, IPSTATS_MIB_INADDRERRORS); 965 break; 966 967 case ENETUNREACH: 968 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 969 break; 970 } 971 goto out; 972 } 973 974 switch (rt->dst.error) { 975 case EINVAL: 976 default: 977 goto out; 978 case EHOSTUNREACH: 979 code = ICMP_HOST_UNREACH; 980 break; 981 case ENETUNREACH: 982 code = ICMP_NET_UNREACH; 983 __IP_INC_STATS(net, IPSTATS_MIB_INNOROUTES); 984 break; 985 case EACCES: 986 code = ICMP_PKT_FILTERED; 987 break; 988 } 989 990 peer = inet_getpeer_v4(net->ipv4.peers, ip_hdr(skb)->saddr, 991 l3mdev_master_ifindex(skb->dev), 1); 992 993 send = true; 994 if (peer) { 995 now = jiffies; 996 peer->rate_tokens += now - peer->rate_last; 997 if (peer->rate_tokens > ip_rt_error_burst) 998 peer->rate_tokens = ip_rt_error_burst; 999 peer->rate_last = now; 1000 if (peer->rate_tokens >= ip_rt_error_cost) 1001 peer->rate_tokens -= ip_rt_error_cost; 1002 else 1003 send = false; 1004 inet_putpeer(peer); 1005 } 1006 if (send) 1007 icmp_send(skb, ICMP_DEST_UNREACH, code, 0); 1008 1009 out: kfree_skb(skb); 1010 return 0; 1011 } 1012 1013 static void __ip_rt_update_pmtu(struct rtable *rt, struct flowi4 *fl4, u32 mtu) 1014 { 1015 struct dst_entry *dst = &rt->dst; 1016 u32 old_mtu = ipv4_mtu(dst); 1017 struct fib_result res; 1018 bool lock = false; 1019 1020 if (ip_mtu_locked(dst)) 1021 return; 1022 1023 if (old_mtu < mtu) 1024 return; 1025 1026 if (mtu < ip_rt_min_pmtu) { 1027 lock = true; 1028 mtu = min(old_mtu, ip_rt_min_pmtu); 1029 } 1030 1031 if (rt->rt_pmtu == mtu && !lock && 1032 time_before(jiffies, dst->expires - ip_rt_mtu_expires / 2)) 1033 return; 1034 1035 rcu_read_lock(); 1036 if (fib_lookup(dev_net(dst->dev), fl4, &res, 0) == 0) { 1037 struct fib_nh_common *nhc = FIB_RES_NHC(res); 1038 1039 update_or_create_fnhe(nhc, fl4->daddr, 0, mtu, lock, 1040 jiffies + ip_rt_mtu_expires); 1041 } 1042 rcu_read_unlock(); 1043 } 1044 1045 static void ip_rt_update_pmtu(struct dst_entry *dst, struct sock *sk, 1046 struct sk_buff *skb, u32 mtu, 1047 bool confirm_neigh) 1048 { 1049 struct rtable *rt = (struct rtable *) dst; 1050 struct flowi4 fl4; 1051 1052 ip_rt_build_flow_key(&fl4, sk, skb); 1053 __ip_rt_update_pmtu(rt, &fl4, mtu); 1054 } 1055 1056 void ipv4_update_pmtu(struct sk_buff *skb, struct net *net, u32 mtu, 1057 int oif, u8 protocol) 1058 { 1059 const struct iphdr *iph = (const struct iphdr *) skb->data; 1060 struct flowi4 fl4; 1061 struct rtable *rt; 1062 u32 mark = IP4_REPLY_MARK(net, skb->mark); 1063 1064 __build_flow_key(net, &fl4, NULL, iph, oif, 1065 RT_TOS(iph->tos), protocol, mark, 0); 1066 rt = __ip_route_output_key(net, &fl4); 1067 if (!IS_ERR(rt)) { 1068 __ip_rt_update_pmtu(rt, &fl4, mtu); 1069 ip_rt_put(rt); 1070 } 1071 } 1072 EXPORT_SYMBOL_GPL(ipv4_update_pmtu); 1073 1074 static void __ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1075 { 1076 const struct iphdr *iph = (const struct iphdr *) skb->data; 1077 struct flowi4 fl4; 1078 struct rtable *rt; 1079 1080 __build_flow_key(sock_net(sk), &fl4, sk, iph, 0, 0, 0, 0, 0); 1081 1082 if (!fl4.flowi4_mark) 1083 fl4.flowi4_mark = IP4_REPLY_MARK(sock_net(sk), skb->mark); 1084 1085 rt = __ip_route_output_key(sock_net(sk), &fl4); 1086 if (!IS_ERR(rt)) { 1087 __ip_rt_update_pmtu(rt, &fl4, mtu); 1088 ip_rt_put(rt); 1089 } 1090 } 1091 1092 void ipv4_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, u32 mtu) 1093 { 1094 const struct iphdr *iph = (const struct iphdr *) skb->data; 1095 struct flowi4 fl4; 1096 struct rtable *rt; 1097 struct dst_entry *odst = NULL; 1098 bool new = false; 1099 struct net *net = sock_net(sk); 1100 1101 bh_lock_sock(sk); 1102 1103 if (!ip_sk_accept_pmtu(sk)) 1104 goto out; 1105 1106 odst = sk_dst_get(sk); 1107 1108 if (sock_owned_by_user(sk) || !odst) { 1109 __ipv4_sk_update_pmtu(skb, sk, mtu); 1110 goto out; 1111 } 1112 1113 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1114 1115 rt = (struct rtable *)odst; 1116 if (odst->obsolete && !odst->ops->check(odst, 0)) { 1117 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1118 if (IS_ERR(rt)) 1119 goto out; 1120 1121 new = true; 1122 } 1123 1124 __ip_rt_update_pmtu((struct rtable *) xfrm_dst_path(&rt->dst), &fl4, mtu); 1125 1126 if (!dst_check(&rt->dst, 0)) { 1127 if (new) 1128 dst_release(&rt->dst); 1129 1130 rt = ip_route_output_flow(sock_net(sk), &fl4, sk); 1131 if (IS_ERR(rt)) 1132 goto out; 1133 1134 new = true; 1135 } 1136 1137 if (new) 1138 sk_dst_set(sk, &rt->dst); 1139 1140 out: 1141 bh_unlock_sock(sk); 1142 dst_release(odst); 1143 } 1144 EXPORT_SYMBOL_GPL(ipv4_sk_update_pmtu); 1145 1146 void ipv4_redirect(struct sk_buff *skb, struct net *net, 1147 int oif, u8 protocol) 1148 { 1149 const struct iphdr *iph = (const struct iphdr *) skb->data; 1150 struct flowi4 fl4; 1151 struct rtable *rt; 1152 1153 __build_flow_key(net, &fl4, NULL, iph, oif, 1154 RT_TOS(iph->tos), protocol, 0, 0); 1155 rt = __ip_route_output_key(net, &fl4); 1156 if (!IS_ERR(rt)) { 1157 __ip_do_redirect(rt, skb, &fl4, false); 1158 ip_rt_put(rt); 1159 } 1160 } 1161 EXPORT_SYMBOL_GPL(ipv4_redirect); 1162 1163 void ipv4_sk_redirect(struct sk_buff *skb, struct sock *sk) 1164 { 1165 const struct iphdr *iph = (const struct iphdr *) skb->data; 1166 struct flowi4 fl4; 1167 struct rtable *rt; 1168 struct net *net = sock_net(sk); 1169 1170 __build_flow_key(net, &fl4, sk, iph, 0, 0, 0, 0, 0); 1171 rt = __ip_route_output_key(net, &fl4); 1172 if (!IS_ERR(rt)) { 1173 __ip_do_redirect(rt, skb, &fl4, false); 1174 ip_rt_put(rt); 1175 } 1176 } 1177 EXPORT_SYMBOL_GPL(ipv4_sk_redirect); 1178 1179 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie) 1180 { 1181 struct rtable *rt = (struct rtable *) dst; 1182 1183 /* All IPV4 dsts are created with ->obsolete set to the value 1184 * DST_OBSOLETE_FORCE_CHK which forces validation calls down 1185 * into this function always. 1186 * 1187 * When a PMTU/redirect information update invalidates a route, 1188 * this is indicated by setting obsolete to DST_OBSOLETE_KILL or 1189 * DST_OBSOLETE_DEAD. 1190 */ 1191 if (dst->obsolete != DST_OBSOLETE_FORCE_CHK || rt_is_expired(rt)) 1192 return NULL; 1193 return dst; 1194 } 1195 1196 static void ipv4_send_dest_unreach(struct sk_buff *skb) 1197 { 1198 struct ip_options opt; 1199 int res; 1200 1201 /* Recompile ip options since IPCB may not be valid anymore. 1202 * Also check we have a reasonable ipv4 header. 1203 */ 1204 if (!pskb_network_may_pull(skb, sizeof(struct iphdr)) || 1205 ip_hdr(skb)->version != 4 || ip_hdr(skb)->ihl < 5) 1206 return; 1207 1208 memset(&opt, 0, sizeof(opt)); 1209 if (ip_hdr(skb)->ihl > 5) { 1210 if (!pskb_network_may_pull(skb, ip_hdr(skb)->ihl * 4)) 1211 return; 1212 opt.optlen = ip_hdr(skb)->ihl * 4 - sizeof(struct iphdr); 1213 1214 rcu_read_lock(); 1215 res = __ip_options_compile(dev_net(skb->dev), &opt, skb, NULL); 1216 rcu_read_unlock(); 1217 1218 if (res) 1219 return; 1220 } 1221 __icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0, &opt); 1222 } 1223 1224 static void ipv4_link_failure(struct sk_buff *skb) 1225 { 1226 struct rtable *rt; 1227 1228 ipv4_send_dest_unreach(skb); 1229 1230 rt = skb_rtable(skb); 1231 if (rt) 1232 dst_set_expires(&rt->dst, 0); 1233 } 1234 1235 static int ip_rt_bug(struct net *net, struct sock *sk, struct sk_buff *skb) 1236 { 1237 pr_debug("%s: %pI4 -> %pI4, %s\n", 1238 __func__, &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1239 skb->dev ? skb->dev->name : "?"); 1240 kfree_skb(skb); 1241 WARN_ON(1); 1242 return 0; 1243 } 1244 1245 /* 1246 We do not cache source address of outgoing interface, 1247 because it is used only by IP RR, TS and SRR options, 1248 so that it out of fast path. 1249 1250 BTW remember: "addr" is allowed to be not aligned 1251 in IP options! 1252 */ 1253 1254 void ip_rt_get_source(u8 *addr, struct sk_buff *skb, struct rtable *rt) 1255 { 1256 __be32 src; 1257 1258 if (rt_is_output_route(rt)) 1259 src = ip_hdr(skb)->saddr; 1260 else { 1261 struct fib_result res; 1262 struct iphdr *iph = ip_hdr(skb); 1263 struct flowi4 fl4 = { 1264 .daddr = iph->daddr, 1265 .saddr = iph->saddr, 1266 .flowi4_tos = RT_TOS(iph->tos), 1267 .flowi4_oif = rt->dst.dev->ifindex, 1268 .flowi4_iif = skb->dev->ifindex, 1269 .flowi4_mark = skb->mark, 1270 }; 1271 1272 rcu_read_lock(); 1273 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res, 0) == 0) 1274 src = fib_result_prefsrc(dev_net(rt->dst.dev), &res); 1275 else 1276 src = inet_select_addr(rt->dst.dev, 1277 rt_nexthop(rt, iph->daddr), 1278 RT_SCOPE_UNIVERSE); 1279 rcu_read_unlock(); 1280 } 1281 memcpy(addr, &src, 4); 1282 } 1283 1284 #ifdef CONFIG_IP_ROUTE_CLASSID 1285 static void set_class_tag(struct rtable *rt, u32 tag) 1286 { 1287 if (!(rt->dst.tclassid & 0xFFFF)) 1288 rt->dst.tclassid |= tag & 0xFFFF; 1289 if (!(rt->dst.tclassid & 0xFFFF0000)) 1290 rt->dst.tclassid |= tag & 0xFFFF0000; 1291 } 1292 #endif 1293 1294 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1295 { 1296 unsigned int header_size = sizeof(struct tcphdr) + sizeof(struct iphdr); 1297 unsigned int advmss = max_t(unsigned int, ipv4_mtu(dst) - header_size, 1298 ip_rt_min_advmss); 1299 1300 return min(advmss, IPV4_MAX_PMTU - header_size); 1301 } 1302 1303 static unsigned int ipv4_mtu(const struct dst_entry *dst) 1304 { 1305 const struct rtable *rt = (const struct rtable *) dst; 1306 unsigned int mtu = rt->rt_pmtu; 1307 1308 if (!mtu || time_after_eq(jiffies, rt->dst.expires)) 1309 mtu = dst_metric_raw(dst, RTAX_MTU); 1310 1311 if (mtu) 1312 return mtu; 1313 1314 mtu = READ_ONCE(dst->dev->mtu); 1315 1316 if (unlikely(ip_mtu_locked(dst))) { 1317 if (rt->rt_uses_gateway && mtu > 576) 1318 mtu = 576; 1319 } 1320 1321 mtu = min_t(unsigned int, mtu, IP_MAX_MTU); 1322 1323 return mtu - lwtunnel_headroom(dst->lwtstate, mtu); 1324 } 1325 1326 static void ip_del_fnhe(struct fib_nh_common *nhc, __be32 daddr) 1327 { 1328 struct fnhe_hash_bucket *hash; 1329 struct fib_nh_exception *fnhe, __rcu **fnhe_p; 1330 u32 hval = fnhe_hashfun(daddr); 1331 1332 spin_lock_bh(&fnhe_lock); 1333 1334 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1335 lockdep_is_held(&fnhe_lock)); 1336 hash += hval; 1337 1338 fnhe_p = &hash->chain; 1339 fnhe = rcu_dereference_protected(*fnhe_p, lockdep_is_held(&fnhe_lock)); 1340 while (fnhe) { 1341 if (fnhe->fnhe_daddr == daddr) { 1342 rcu_assign_pointer(*fnhe_p, rcu_dereference_protected( 1343 fnhe->fnhe_next, lockdep_is_held(&fnhe_lock))); 1344 /* set fnhe_daddr to 0 to ensure it won't bind with 1345 * new dsts in rt_bind_exception(). 1346 */ 1347 fnhe->fnhe_daddr = 0; 1348 fnhe_flush_routes(fnhe); 1349 kfree_rcu(fnhe, rcu); 1350 break; 1351 } 1352 fnhe_p = &fnhe->fnhe_next; 1353 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1354 lockdep_is_held(&fnhe_lock)); 1355 } 1356 1357 spin_unlock_bh(&fnhe_lock); 1358 } 1359 1360 static struct fib_nh_exception *find_exception(struct fib_nh_common *nhc, 1361 __be32 daddr) 1362 { 1363 struct fnhe_hash_bucket *hash = rcu_dereference(nhc->nhc_exceptions); 1364 struct fib_nh_exception *fnhe; 1365 u32 hval; 1366 1367 if (!hash) 1368 return NULL; 1369 1370 hval = fnhe_hashfun(daddr); 1371 1372 for (fnhe = rcu_dereference(hash[hval].chain); fnhe; 1373 fnhe = rcu_dereference(fnhe->fnhe_next)) { 1374 if (fnhe->fnhe_daddr == daddr) { 1375 if (fnhe->fnhe_expires && 1376 time_after(jiffies, fnhe->fnhe_expires)) { 1377 ip_del_fnhe(nhc, daddr); 1378 break; 1379 } 1380 return fnhe; 1381 } 1382 } 1383 return NULL; 1384 } 1385 1386 /* MTU selection: 1387 * 1. mtu on route is locked - use it 1388 * 2. mtu from nexthop exception 1389 * 3. mtu from egress device 1390 */ 1391 1392 u32 ip_mtu_from_fib_result(struct fib_result *res, __be32 daddr) 1393 { 1394 struct fib_nh_common *nhc = res->nhc; 1395 struct net_device *dev = nhc->nhc_dev; 1396 struct fib_info *fi = res->fi; 1397 u32 mtu = 0; 1398 1399 if (dev_net(dev)->ipv4.sysctl_ip_fwd_use_pmtu || 1400 fi->fib_metrics->metrics[RTAX_LOCK - 1] & (1 << RTAX_MTU)) 1401 mtu = fi->fib_mtu; 1402 1403 if (likely(!mtu)) { 1404 struct fib_nh_exception *fnhe; 1405 1406 fnhe = find_exception(nhc, daddr); 1407 if (fnhe && !time_after_eq(jiffies, fnhe->fnhe_expires)) 1408 mtu = fnhe->fnhe_pmtu; 1409 } 1410 1411 if (likely(!mtu)) 1412 mtu = min(READ_ONCE(dev->mtu), IP_MAX_MTU); 1413 1414 return mtu - lwtunnel_headroom(nhc->nhc_lwtstate, mtu); 1415 } 1416 1417 static bool rt_bind_exception(struct rtable *rt, struct fib_nh_exception *fnhe, 1418 __be32 daddr, const bool do_cache) 1419 { 1420 bool ret = false; 1421 1422 spin_lock_bh(&fnhe_lock); 1423 1424 if (daddr == fnhe->fnhe_daddr) { 1425 struct rtable __rcu **porig; 1426 struct rtable *orig; 1427 int genid = fnhe_genid(dev_net(rt->dst.dev)); 1428 1429 if (rt_is_input_route(rt)) 1430 porig = &fnhe->fnhe_rth_input; 1431 else 1432 porig = &fnhe->fnhe_rth_output; 1433 orig = rcu_dereference(*porig); 1434 1435 if (fnhe->fnhe_genid != genid) { 1436 fnhe->fnhe_genid = genid; 1437 fnhe->fnhe_gw = 0; 1438 fnhe->fnhe_pmtu = 0; 1439 fnhe->fnhe_expires = 0; 1440 fnhe->fnhe_mtu_locked = false; 1441 fnhe_flush_routes(fnhe); 1442 orig = NULL; 1443 } 1444 fill_route_from_fnhe(rt, fnhe); 1445 if (!rt->rt_gw4) { 1446 rt->rt_gw4 = daddr; 1447 rt->rt_gw_family = AF_INET; 1448 } 1449 1450 if (do_cache) { 1451 dst_hold(&rt->dst); 1452 rcu_assign_pointer(*porig, rt); 1453 if (orig) { 1454 dst_dev_put(&orig->dst); 1455 dst_release(&orig->dst); 1456 } 1457 ret = true; 1458 } 1459 1460 fnhe->fnhe_stamp = jiffies; 1461 } 1462 spin_unlock_bh(&fnhe_lock); 1463 1464 return ret; 1465 } 1466 1467 static bool rt_cache_route(struct fib_nh_common *nhc, struct rtable *rt) 1468 { 1469 struct rtable *orig, *prev, **p; 1470 bool ret = true; 1471 1472 if (rt_is_input_route(rt)) { 1473 p = (struct rtable **)&nhc->nhc_rth_input; 1474 } else { 1475 p = (struct rtable **)raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 1476 } 1477 orig = *p; 1478 1479 /* hold dst before doing cmpxchg() to avoid race condition 1480 * on this dst 1481 */ 1482 dst_hold(&rt->dst); 1483 prev = cmpxchg(p, orig, rt); 1484 if (prev == orig) { 1485 if (orig) { 1486 rt_add_uncached_list(orig); 1487 dst_release(&orig->dst); 1488 } 1489 } else { 1490 dst_release(&rt->dst); 1491 ret = false; 1492 } 1493 1494 return ret; 1495 } 1496 1497 struct uncached_list { 1498 spinlock_t lock; 1499 struct list_head head; 1500 }; 1501 1502 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt_uncached_list); 1503 1504 void rt_add_uncached_list(struct rtable *rt) 1505 { 1506 struct uncached_list *ul = raw_cpu_ptr(&rt_uncached_list); 1507 1508 rt->rt_uncached_list = ul; 1509 1510 spin_lock_bh(&ul->lock); 1511 list_add_tail(&rt->rt_uncached, &ul->head); 1512 spin_unlock_bh(&ul->lock); 1513 } 1514 1515 void rt_del_uncached_list(struct rtable *rt) 1516 { 1517 if (!list_empty(&rt->rt_uncached)) { 1518 struct uncached_list *ul = rt->rt_uncached_list; 1519 1520 spin_lock_bh(&ul->lock); 1521 list_del(&rt->rt_uncached); 1522 spin_unlock_bh(&ul->lock); 1523 } 1524 } 1525 1526 static void ipv4_dst_destroy(struct dst_entry *dst) 1527 { 1528 struct rtable *rt = (struct rtable *)dst; 1529 1530 ip_dst_metrics_put(dst); 1531 rt_del_uncached_list(rt); 1532 } 1533 1534 void rt_flush_dev(struct net_device *dev) 1535 { 1536 struct rtable *rt; 1537 int cpu; 1538 1539 for_each_possible_cpu(cpu) { 1540 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 1541 1542 spin_lock_bh(&ul->lock); 1543 list_for_each_entry(rt, &ul->head, rt_uncached) { 1544 if (rt->dst.dev != dev) 1545 continue; 1546 rt->dst.dev = blackhole_netdev; 1547 dev_hold(rt->dst.dev); 1548 dev_put(dev); 1549 } 1550 spin_unlock_bh(&ul->lock); 1551 } 1552 } 1553 1554 static bool rt_cache_valid(const struct rtable *rt) 1555 { 1556 return rt && 1557 rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK && 1558 !rt_is_expired(rt); 1559 } 1560 1561 static void rt_set_nexthop(struct rtable *rt, __be32 daddr, 1562 const struct fib_result *res, 1563 struct fib_nh_exception *fnhe, 1564 struct fib_info *fi, u16 type, u32 itag, 1565 const bool do_cache) 1566 { 1567 bool cached = false; 1568 1569 if (fi) { 1570 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1571 1572 if (nhc->nhc_gw_family && nhc->nhc_scope == RT_SCOPE_LINK) { 1573 rt->rt_uses_gateway = 1; 1574 rt->rt_gw_family = nhc->nhc_gw_family; 1575 /* only INET and INET6 are supported */ 1576 if (likely(nhc->nhc_gw_family == AF_INET)) 1577 rt->rt_gw4 = nhc->nhc_gw.ipv4; 1578 else 1579 rt->rt_gw6 = nhc->nhc_gw.ipv6; 1580 } 1581 1582 ip_dst_init_metrics(&rt->dst, fi->fib_metrics); 1583 1584 #ifdef CONFIG_IP_ROUTE_CLASSID 1585 if (nhc->nhc_family == AF_INET) { 1586 struct fib_nh *nh; 1587 1588 nh = container_of(nhc, struct fib_nh, nh_common); 1589 rt->dst.tclassid = nh->nh_tclassid; 1590 } 1591 #endif 1592 rt->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 1593 if (unlikely(fnhe)) 1594 cached = rt_bind_exception(rt, fnhe, daddr, do_cache); 1595 else if (do_cache) 1596 cached = rt_cache_route(nhc, rt); 1597 if (unlikely(!cached)) { 1598 /* Routes we intend to cache in nexthop exception or 1599 * FIB nexthop have the DST_NOCACHE bit clear. 1600 * However, if we are unsuccessful at storing this 1601 * route into the cache we really need to set it. 1602 */ 1603 if (!rt->rt_gw4) { 1604 rt->rt_gw_family = AF_INET; 1605 rt->rt_gw4 = daddr; 1606 } 1607 rt_add_uncached_list(rt); 1608 } 1609 } else 1610 rt_add_uncached_list(rt); 1611 1612 #ifdef CONFIG_IP_ROUTE_CLASSID 1613 #ifdef CONFIG_IP_MULTIPLE_TABLES 1614 set_class_tag(rt, res->tclassid); 1615 #endif 1616 set_class_tag(rt, itag); 1617 #endif 1618 } 1619 1620 struct rtable *rt_dst_alloc(struct net_device *dev, 1621 unsigned int flags, u16 type, 1622 bool nopolicy, bool noxfrm) 1623 { 1624 struct rtable *rt; 1625 1626 rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1627 (nopolicy ? DST_NOPOLICY : 0) | 1628 (noxfrm ? DST_NOXFRM : 0)); 1629 1630 if (rt) { 1631 rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1632 rt->rt_flags = flags; 1633 rt->rt_type = type; 1634 rt->rt_is_input = 0; 1635 rt->rt_iif = 0; 1636 rt->rt_pmtu = 0; 1637 rt->rt_mtu_locked = 0; 1638 rt->rt_uses_gateway = 0; 1639 rt->rt_gw_family = 0; 1640 rt->rt_gw4 = 0; 1641 INIT_LIST_HEAD(&rt->rt_uncached); 1642 1643 rt->dst.output = ip_output; 1644 if (flags & RTCF_LOCAL) 1645 rt->dst.input = ip_local_deliver; 1646 } 1647 1648 return rt; 1649 } 1650 EXPORT_SYMBOL(rt_dst_alloc); 1651 1652 struct rtable *rt_dst_clone(struct net_device *dev, struct rtable *rt) 1653 { 1654 struct rtable *new_rt; 1655 1656 new_rt = dst_alloc(&ipv4_dst_ops, dev, 1, DST_OBSOLETE_FORCE_CHK, 1657 rt->dst.flags); 1658 1659 if (new_rt) { 1660 new_rt->rt_genid = rt_genid_ipv4(dev_net(dev)); 1661 new_rt->rt_flags = rt->rt_flags; 1662 new_rt->rt_type = rt->rt_type; 1663 new_rt->rt_is_input = rt->rt_is_input; 1664 new_rt->rt_iif = rt->rt_iif; 1665 new_rt->rt_pmtu = rt->rt_pmtu; 1666 new_rt->rt_mtu_locked = rt->rt_mtu_locked; 1667 new_rt->rt_gw_family = rt->rt_gw_family; 1668 if (rt->rt_gw_family == AF_INET) 1669 new_rt->rt_gw4 = rt->rt_gw4; 1670 else if (rt->rt_gw_family == AF_INET6) 1671 new_rt->rt_gw6 = rt->rt_gw6; 1672 INIT_LIST_HEAD(&new_rt->rt_uncached); 1673 1674 new_rt->dst.input = rt->dst.input; 1675 new_rt->dst.output = rt->dst.output; 1676 new_rt->dst.error = rt->dst.error; 1677 new_rt->dst.lastuse = jiffies; 1678 new_rt->dst.lwtstate = lwtstate_get(rt->dst.lwtstate); 1679 } 1680 return new_rt; 1681 } 1682 EXPORT_SYMBOL(rt_dst_clone); 1683 1684 /* called in rcu_read_lock() section */ 1685 int ip_mc_validate_source(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1686 u8 tos, struct net_device *dev, 1687 struct in_device *in_dev, u32 *itag) 1688 { 1689 int err; 1690 1691 /* Primary sanity checks. */ 1692 if (!in_dev) 1693 return -EINVAL; 1694 1695 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) || 1696 skb->protocol != htons(ETH_P_IP)) 1697 return -EINVAL; 1698 1699 if (ipv4_is_loopback(saddr) && !IN_DEV_ROUTE_LOCALNET(in_dev)) 1700 return -EINVAL; 1701 1702 if (ipv4_is_zeronet(saddr)) { 1703 if (!ipv4_is_local_multicast(daddr) && 1704 ip_hdr(skb)->protocol != IPPROTO_IGMP) 1705 return -EINVAL; 1706 } else { 1707 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 1708 in_dev, itag); 1709 if (err < 0) 1710 return err; 1711 } 1712 return 0; 1713 } 1714 1715 /* called in rcu_read_lock() section */ 1716 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1717 u8 tos, struct net_device *dev, int our) 1718 { 1719 struct in_device *in_dev = __in_dev_get_rcu(dev); 1720 unsigned int flags = RTCF_MULTICAST; 1721 struct rtable *rth; 1722 u32 itag = 0; 1723 int err; 1724 1725 err = ip_mc_validate_source(skb, daddr, saddr, tos, dev, in_dev, &itag); 1726 if (err) 1727 return err; 1728 1729 if (our) 1730 flags |= RTCF_LOCAL; 1731 1732 rth = rt_dst_alloc(dev_net(dev)->loopback_dev, flags, RTN_MULTICAST, 1733 IN_DEV_CONF_GET(in_dev, NOPOLICY), false); 1734 if (!rth) 1735 return -ENOBUFS; 1736 1737 #ifdef CONFIG_IP_ROUTE_CLASSID 1738 rth->dst.tclassid = itag; 1739 #endif 1740 rth->dst.output = ip_rt_bug; 1741 rth->rt_is_input= 1; 1742 1743 #ifdef CONFIG_IP_MROUTE 1744 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1745 rth->dst.input = ip_mr_input; 1746 #endif 1747 RT_CACHE_STAT_INC(in_slow_mc); 1748 1749 skb_dst_set(skb, &rth->dst); 1750 return 0; 1751 } 1752 1753 1754 static void ip_handle_martian_source(struct net_device *dev, 1755 struct in_device *in_dev, 1756 struct sk_buff *skb, 1757 __be32 daddr, 1758 __be32 saddr) 1759 { 1760 RT_CACHE_STAT_INC(in_martian_src); 1761 #ifdef CONFIG_IP_ROUTE_VERBOSE 1762 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1763 /* 1764 * RFC1812 recommendation, if source is martian, 1765 * the only hint is MAC header. 1766 */ 1767 pr_warn("martian source %pI4 from %pI4, on dev %s\n", 1768 &daddr, &saddr, dev->name); 1769 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1770 print_hex_dump(KERN_WARNING, "ll header: ", 1771 DUMP_PREFIX_OFFSET, 16, 1, 1772 skb_mac_header(skb), 1773 dev->hard_header_len, false); 1774 } 1775 } 1776 #endif 1777 } 1778 1779 /* called in rcu_read_lock() section */ 1780 static int __mkroute_input(struct sk_buff *skb, 1781 const struct fib_result *res, 1782 struct in_device *in_dev, 1783 __be32 daddr, __be32 saddr, u32 tos) 1784 { 1785 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 1786 struct net_device *dev = nhc->nhc_dev; 1787 struct fib_nh_exception *fnhe; 1788 struct rtable *rth; 1789 int err; 1790 struct in_device *out_dev; 1791 bool do_cache; 1792 u32 itag = 0; 1793 1794 /* get a working reference to the output device */ 1795 out_dev = __in_dev_get_rcu(dev); 1796 if (!out_dev) { 1797 net_crit_ratelimited("Bug in ip_route_input_slow(). Please report.\n"); 1798 return -EINVAL; 1799 } 1800 1801 err = fib_validate_source(skb, saddr, daddr, tos, FIB_RES_OIF(*res), 1802 in_dev->dev, in_dev, &itag); 1803 if (err < 0) { 1804 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1805 saddr); 1806 1807 goto cleanup; 1808 } 1809 1810 do_cache = res->fi && !itag; 1811 if (out_dev == in_dev && err && IN_DEV_TX_REDIRECTS(out_dev) && 1812 skb->protocol == htons(ETH_P_IP)) { 1813 __be32 gw; 1814 1815 gw = nhc->nhc_gw_family == AF_INET ? nhc->nhc_gw.ipv4 : 0; 1816 if (IN_DEV_SHARED_MEDIA(out_dev) || 1817 inet_addr_onlink(out_dev, saddr, gw)) 1818 IPCB(skb)->flags |= IPSKB_DOREDIRECT; 1819 } 1820 1821 if (skb->protocol != htons(ETH_P_IP)) { 1822 /* Not IP (i.e. ARP). Do not create route, if it is 1823 * invalid for proxy arp. DNAT routes are always valid. 1824 * 1825 * Proxy arp feature have been extended to allow, ARP 1826 * replies back to the same interface, to support 1827 * Private VLAN switch technologies. See arp.c. 1828 */ 1829 if (out_dev == in_dev && 1830 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 1831 err = -EINVAL; 1832 goto cleanup; 1833 } 1834 } 1835 1836 fnhe = find_exception(nhc, daddr); 1837 if (do_cache) { 1838 if (fnhe) 1839 rth = rcu_dereference(fnhe->fnhe_rth_input); 1840 else 1841 rth = rcu_dereference(nhc->nhc_rth_input); 1842 if (rt_cache_valid(rth)) { 1843 skb_dst_set_noref(skb, &rth->dst); 1844 goto out; 1845 } 1846 } 1847 1848 rth = rt_dst_alloc(out_dev->dev, 0, res->type, 1849 IN_DEV_CONF_GET(in_dev, NOPOLICY), 1850 IN_DEV_CONF_GET(out_dev, NOXFRM)); 1851 if (!rth) { 1852 err = -ENOBUFS; 1853 goto cleanup; 1854 } 1855 1856 rth->rt_is_input = 1; 1857 RT_CACHE_STAT_INC(in_slow_tot); 1858 1859 rth->dst.input = ip_forward; 1860 1861 rt_set_nexthop(rth, daddr, res, fnhe, res->fi, res->type, itag, 1862 do_cache); 1863 lwtunnel_set_redirect(&rth->dst); 1864 skb_dst_set(skb, &rth->dst); 1865 out: 1866 err = 0; 1867 cleanup: 1868 return err; 1869 } 1870 1871 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1872 /* To make ICMP packets follow the right flow, the multipath hash is 1873 * calculated from the inner IP addresses. 1874 */ 1875 static void ip_multipath_l3_keys(const struct sk_buff *skb, 1876 struct flow_keys *hash_keys) 1877 { 1878 const struct iphdr *outer_iph = ip_hdr(skb); 1879 const struct iphdr *key_iph = outer_iph; 1880 const struct iphdr *inner_iph; 1881 const struct icmphdr *icmph; 1882 struct iphdr _inner_iph; 1883 struct icmphdr _icmph; 1884 1885 if (likely(outer_iph->protocol != IPPROTO_ICMP)) 1886 goto out; 1887 1888 if (unlikely((outer_iph->frag_off & htons(IP_OFFSET)) != 0)) 1889 goto out; 1890 1891 icmph = skb_header_pointer(skb, outer_iph->ihl * 4, sizeof(_icmph), 1892 &_icmph); 1893 if (!icmph) 1894 goto out; 1895 1896 if (!icmp_is_err(icmph->type)) 1897 goto out; 1898 1899 inner_iph = skb_header_pointer(skb, 1900 outer_iph->ihl * 4 + sizeof(_icmph), 1901 sizeof(_inner_iph), &_inner_iph); 1902 if (!inner_iph) 1903 goto out; 1904 1905 key_iph = inner_iph; 1906 out: 1907 hash_keys->addrs.v4addrs.src = key_iph->saddr; 1908 hash_keys->addrs.v4addrs.dst = key_iph->daddr; 1909 } 1910 1911 /* if skb is set it will be used and fl4 can be NULL */ 1912 int fib_multipath_hash(const struct net *net, const struct flowi4 *fl4, 1913 const struct sk_buff *skb, struct flow_keys *flkeys) 1914 { 1915 u32 multipath_hash = fl4 ? fl4->flowi4_multipath_hash : 0; 1916 struct flow_keys hash_keys; 1917 u32 mhash; 1918 1919 switch (net->ipv4.sysctl_fib_multipath_hash_policy) { 1920 case 0: 1921 memset(&hash_keys, 0, sizeof(hash_keys)); 1922 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1923 if (skb) { 1924 ip_multipath_l3_keys(skb, &hash_keys); 1925 } else { 1926 hash_keys.addrs.v4addrs.src = fl4->saddr; 1927 hash_keys.addrs.v4addrs.dst = fl4->daddr; 1928 } 1929 break; 1930 case 1: 1931 /* skb is currently provided only when forwarding */ 1932 if (skb) { 1933 unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP; 1934 struct flow_keys keys; 1935 1936 /* short-circuit if we already have L4 hash present */ 1937 if (skb->l4_hash) 1938 return skb_get_hash_raw(skb) >> 1; 1939 1940 memset(&hash_keys, 0, sizeof(hash_keys)); 1941 1942 if (!flkeys) { 1943 skb_flow_dissect_flow_keys(skb, &keys, flag); 1944 flkeys = &keys; 1945 } 1946 1947 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1948 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src; 1949 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst; 1950 hash_keys.ports.src = flkeys->ports.src; 1951 hash_keys.ports.dst = flkeys->ports.dst; 1952 hash_keys.basic.ip_proto = flkeys->basic.ip_proto; 1953 } else { 1954 memset(&hash_keys, 0, sizeof(hash_keys)); 1955 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1956 hash_keys.addrs.v4addrs.src = fl4->saddr; 1957 hash_keys.addrs.v4addrs.dst = fl4->daddr; 1958 hash_keys.ports.src = fl4->fl4_sport; 1959 hash_keys.ports.dst = fl4->fl4_dport; 1960 hash_keys.basic.ip_proto = fl4->flowi4_proto; 1961 } 1962 break; 1963 case 2: 1964 memset(&hash_keys, 0, sizeof(hash_keys)); 1965 /* skb is currently provided only when forwarding */ 1966 if (skb) { 1967 struct flow_keys keys; 1968 1969 skb_flow_dissect_flow_keys(skb, &keys, 0); 1970 /* Inner can be v4 or v6 */ 1971 if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) { 1972 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1973 hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 1974 hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst; 1975 } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) { 1976 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS; 1977 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 1978 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst; 1979 hash_keys.tags.flow_label = keys.tags.flow_label; 1980 hash_keys.basic.ip_proto = keys.basic.ip_proto; 1981 } else { 1982 /* Same as case 0 */ 1983 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1984 ip_multipath_l3_keys(skb, &hash_keys); 1985 } 1986 } else { 1987 /* Same as case 0 */ 1988 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS; 1989 hash_keys.addrs.v4addrs.src = fl4->saddr; 1990 hash_keys.addrs.v4addrs.dst = fl4->daddr; 1991 } 1992 break; 1993 } 1994 mhash = flow_hash_from_keys(&hash_keys); 1995 1996 if (multipath_hash) 1997 mhash = jhash_2words(mhash, multipath_hash, 0); 1998 1999 return mhash >> 1; 2000 } 2001 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 2002 2003 static int ip_mkroute_input(struct sk_buff *skb, 2004 struct fib_result *res, 2005 struct in_device *in_dev, 2006 __be32 daddr, __be32 saddr, u32 tos, 2007 struct flow_keys *hkeys) 2008 { 2009 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2010 if (res->fi && fib_info_num_path(res->fi) > 1) { 2011 int h = fib_multipath_hash(res->fi->fib_net, NULL, skb, hkeys); 2012 2013 fib_select_multipath(res, h); 2014 } 2015 #endif 2016 2017 /* create a routing cache entry */ 2018 return __mkroute_input(skb, res, in_dev, daddr, saddr, tos); 2019 } 2020 2021 /* Implements all the saddr-related checks as ip_route_input_slow(), 2022 * assuming daddr is valid and the destination is not a local broadcast one. 2023 * Uses the provided hint instead of performing a route lookup. 2024 */ 2025 int ip_route_use_hint(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2026 u8 tos, struct net_device *dev, 2027 const struct sk_buff *hint) 2028 { 2029 struct in_device *in_dev = __in_dev_get_rcu(dev); 2030 struct rtable *rt = skb_rtable(hint); 2031 struct net *net = dev_net(dev); 2032 int err = -EINVAL; 2033 u32 tag = 0; 2034 2035 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 2036 goto martian_source; 2037 2038 if (ipv4_is_zeronet(saddr)) 2039 goto martian_source; 2040 2041 if (ipv4_is_loopback(saddr) && !IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2042 goto martian_source; 2043 2044 if (rt->rt_type != RTN_LOCAL) 2045 goto skip_validate_source; 2046 2047 tos &= IPTOS_RT_MASK; 2048 err = fib_validate_source(skb, saddr, daddr, tos, 0, dev, in_dev, &tag); 2049 if (err < 0) 2050 goto martian_source; 2051 2052 skip_validate_source: 2053 skb_dst_copy(skb, hint); 2054 return 0; 2055 2056 martian_source: 2057 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2058 return err; 2059 } 2060 2061 /* 2062 * NOTE. We drop all the packets that has local source 2063 * addresses, because every properly looped back packet 2064 * must have correct destination already attached by output routine. 2065 * Changes in the enforced policies must be applied also to 2066 * ip_route_use_hint(). 2067 * 2068 * Such approach solves two big problems: 2069 * 1. Not simplex devices are handled properly. 2070 * 2. IP spoofing attempts are filtered with 100% of guarantee. 2071 * called with rcu_read_lock() 2072 */ 2073 2074 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2075 u8 tos, struct net_device *dev, 2076 struct fib_result *res) 2077 { 2078 struct in_device *in_dev = __in_dev_get_rcu(dev); 2079 struct flow_keys *flkeys = NULL, _flkeys; 2080 struct net *net = dev_net(dev); 2081 struct ip_tunnel_info *tun_info; 2082 int err = -EINVAL; 2083 unsigned int flags = 0; 2084 u32 itag = 0; 2085 struct rtable *rth; 2086 struct flowi4 fl4; 2087 bool do_cache = true; 2088 2089 /* IP on this device is disabled. */ 2090 2091 if (!in_dev) 2092 goto out; 2093 2094 /* Check for the most weird martians, which can be not detected 2095 by fib_lookup. 2096 */ 2097 2098 tun_info = skb_tunnel_info(skb); 2099 if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX)) 2100 fl4.flowi4_tun_key.tun_id = tun_info->key.tun_id; 2101 else 2102 fl4.flowi4_tun_key.tun_id = 0; 2103 skb_dst_drop(skb); 2104 2105 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr)) 2106 goto martian_source; 2107 2108 res->fi = NULL; 2109 res->table = NULL; 2110 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 2111 goto brd_input; 2112 2113 /* Accept zero addresses only to limited broadcast; 2114 * I even do not know to fix it or not. Waiting for complains :-) 2115 */ 2116 if (ipv4_is_zeronet(saddr)) 2117 goto martian_source; 2118 2119 if (ipv4_is_zeronet(daddr)) 2120 goto martian_destination; 2121 2122 /* Following code try to avoid calling IN_DEV_NET_ROUTE_LOCALNET(), 2123 * and call it once if daddr or/and saddr are loopback addresses 2124 */ 2125 if (ipv4_is_loopback(daddr)) { 2126 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2127 goto martian_destination; 2128 } else if (ipv4_is_loopback(saddr)) { 2129 if (!IN_DEV_NET_ROUTE_LOCALNET(in_dev, net)) 2130 goto martian_source; 2131 } 2132 2133 /* 2134 * Now we are ready to route packet. 2135 */ 2136 fl4.flowi4_oif = 0; 2137 fl4.flowi4_iif = dev->ifindex; 2138 fl4.flowi4_mark = skb->mark; 2139 fl4.flowi4_tos = tos; 2140 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 2141 fl4.flowi4_flags = 0; 2142 fl4.daddr = daddr; 2143 fl4.saddr = saddr; 2144 fl4.flowi4_uid = sock_net_uid(net, NULL); 2145 2146 if (fib4_rules_early_flow_dissect(net, skb, &fl4, &_flkeys)) { 2147 flkeys = &_flkeys; 2148 } else { 2149 fl4.flowi4_proto = 0; 2150 fl4.fl4_sport = 0; 2151 fl4.fl4_dport = 0; 2152 } 2153 2154 err = fib_lookup(net, &fl4, res, 0); 2155 if (err != 0) { 2156 if (!IN_DEV_FORWARD(in_dev)) 2157 err = -EHOSTUNREACH; 2158 goto no_route; 2159 } 2160 2161 if (res->type == RTN_BROADCAST) { 2162 if (IN_DEV_BFORWARD(in_dev)) 2163 goto make_route; 2164 /* not do cache if bc_forwarding is enabled */ 2165 if (IPV4_DEVCONF_ALL(net, BC_FORWARDING)) 2166 do_cache = false; 2167 goto brd_input; 2168 } 2169 2170 if (res->type == RTN_LOCAL) { 2171 err = fib_validate_source(skb, saddr, daddr, tos, 2172 0, dev, in_dev, &itag); 2173 if (err < 0) 2174 goto martian_source; 2175 goto local_input; 2176 } 2177 2178 if (!IN_DEV_FORWARD(in_dev)) { 2179 err = -EHOSTUNREACH; 2180 goto no_route; 2181 } 2182 if (res->type != RTN_UNICAST) 2183 goto martian_destination; 2184 2185 make_route: 2186 err = ip_mkroute_input(skb, res, in_dev, daddr, saddr, tos, flkeys); 2187 out: return err; 2188 2189 brd_input: 2190 if (skb->protocol != htons(ETH_P_IP)) 2191 goto e_inval; 2192 2193 if (!ipv4_is_zeronet(saddr)) { 2194 err = fib_validate_source(skb, saddr, 0, tos, 0, dev, 2195 in_dev, &itag); 2196 if (err < 0) 2197 goto martian_source; 2198 } 2199 flags |= RTCF_BROADCAST; 2200 res->type = RTN_BROADCAST; 2201 RT_CACHE_STAT_INC(in_brd); 2202 2203 local_input: 2204 do_cache &= res->fi && !itag; 2205 if (do_cache) { 2206 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2207 2208 rth = rcu_dereference(nhc->nhc_rth_input); 2209 if (rt_cache_valid(rth)) { 2210 skb_dst_set_noref(skb, &rth->dst); 2211 err = 0; 2212 goto out; 2213 } 2214 } 2215 2216 rth = rt_dst_alloc(l3mdev_master_dev_rcu(dev) ? : net->loopback_dev, 2217 flags | RTCF_LOCAL, res->type, 2218 IN_DEV_CONF_GET(in_dev, NOPOLICY), false); 2219 if (!rth) 2220 goto e_nobufs; 2221 2222 rth->dst.output= ip_rt_bug; 2223 #ifdef CONFIG_IP_ROUTE_CLASSID 2224 rth->dst.tclassid = itag; 2225 #endif 2226 rth->rt_is_input = 1; 2227 2228 RT_CACHE_STAT_INC(in_slow_tot); 2229 if (res->type == RTN_UNREACHABLE) { 2230 rth->dst.input= ip_error; 2231 rth->dst.error= -err; 2232 rth->rt_flags &= ~RTCF_LOCAL; 2233 } 2234 2235 if (do_cache) { 2236 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2237 2238 rth->dst.lwtstate = lwtstate_get(nhc->nhc_lwtstate); 2239 if (lwtunnel_input_redirect(rth->dst.lwtstate)) { 2240 WARN_ON(rth->dst.input == lwtunnel_input); 2241 rth->dst.lwtstate->orig_input = rth->dst.input; 2242 rth->dst.input = lwtunnel_input; 2243 } 2244 2245 if (unlikely(!rt_cache_route(nhc, rth))) 2246 rt_add_uncached_list(rth); 2247 } 2248 skb_dst_set(skb, &rth->dst); 2249 err = 0; 2250 goto out; 2251 2252 no_route: 2253 RT_CACHE_STAT_INC(in_no_route); 2254 res->type = RTN_UNREACHABLE; 2255 res->fi = NULL; 2256 res->table = NULL; 2257 goto local_input; 2258 2259 /* 2260 * Do not cache martian addresses: they should be logged (RFC1812) 2261 */ 2262 martian_destination: 2263 RT_CACHE_STAT_INC(in_martian_dst); 2264 #ifdef CONFIG_IP_ROUTE_VERBOSE 2265 if (IN_DEV_LOG_MARTIANS(in_dev)) 2266 net_warn_ratelimited("martian destination %pI4 from %pI4, dev %s\n", 2267 &daddr, &saddr, dev->name); 2268 #endif 2269 2270 e_inval: 2271 err = -EINVAL; 2272 goto out; 2273 2274 e_nobufs: 2275 err = -ENOBUFS; 2276 goto out; 2277 2278 martian_source: 2279 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2280 goto out; 2281 } 2282 2283 int ip_route_input_noref(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2284 u8 tos, struct net_device *dev) 2285 { 2286 struct fib_result res; 2287 int err; 2288 2289 tos &= IPTOS_RT_MASK; 2290 rcu_read_lock(); 2291 err = ip_route_input_rcu(skb, daddr, saddr, tos, dev, &res); 2292 rcu_read_unlock(); 2293 2294 return err; 2295 } 2296 EXPORT_SYMBOL(ip_route_input_noref); 2297 2298 /* called with rcu_read_lock held */ 2299 int ip_route_input_rcu(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2300 u8 tos, struct net_device *dev, struct fib_result *res) 2301 { 2302 /* Multicast recognition logic is moved from route cache to here. 2303 The problem was that too many Ethernet cards have broken/missing 2304 hardware multicast filters :-( As result the host on multicasting 2305 network acquires a lot of useless route cache entries, sort of 2306 SDR messages from all the world. Now we try to get rid of them. 2307 Really, provided software IP multicast filter is organized 2308 reasonably (at least, hashed), it does not result in a slowdown 2309 comparing with route cache reject entries. 2310 Note, that multicast routers are not affected, because 2311 route cache entry is created eventually. 2312 */ 2313 if (ipv4_is_multicast(daddr)) { 2314 struct in_device *in_dev = __in_dev_get_rcu(dev); 2315 int our = 0; 2316 int err = -EINVAL; 2317 2318 if (!in_dev) 2319 return err; 2320 our = ip_check_mc_rcu(in_dev, daddr, saddr, 2321 ip_hdr(skb)->protocol); 2322 2323 /* check l3 master if no match yet */ 2324 if (!our && netif_is_l3_slave(dev)) { 2325 struct in_device *l3_in_dev; 2326 2327 l3_in_dev = __in_dev_get_rcu(skb->dev); 2328 if (l3_in_dev) 2329 our = ip_check_mc_rcu(l3_in_dev, daddr, saddr, 2330 ip_hdr(skb)->protocol); 2331 } 2332 2333 if (our 2334 #ifdef CONFIG_IP_MROUTE 2335 || 2336 (!ipv4_is_local_multicast(daddr) && 2337 IN_DEV_MFORWARD(in_dev)) 2338 #endif 2339 ) { 2340 err = ip_route_input_mc(skb, daddr, saddr, 2341 tos, dev, our); 2342 } 2343 return err; 2344 } 2345 2346 return ip_route_input_slow(skb, daddr, saddr, tos, dev, res); 2347 } 2348 2349 /* called with rcu_read_lock() */ 2350 static struct rtable *__mkroute_output(const struct fib_result *res, 2351 const struct flowi4 *fl4, int orig_oif, 2352 struct net_device *dev_out, 2353 unsigned int flags) 2354 { 2355 struct fib_info *fi = res->fi; 2356 struct fib_nh_exception *fnhe; 2357 struct in_device *in_dev; 2358 u16 type = res->type; 2359 struct rtable *rth; 2360 bool do_cache; 2361 2362 in_dev = __in_dev_get_rcu(dev_out); 2363 if (!in_dev) 2364 return ERR_PTR(-EINVAL); 2365 2366 if (likely(!IN_DEV_ROUTE_LOCALNET(in_dev))) 2367 if (ipv4_is_loopback(fl4->saddr) && 2368 !(dev_out->flags & IFF_LOOPBACK) && 2369 !netif_is_l3_master(dev_out)) 2370 return ERR_PTR(-EINVAL); 2371 2372 if (ipv4_is_lbcast(fl4->daddr)) 2373 type = RTN_BROADCAST; 2374 else if (ipv4_is_multicast(fl4->daddr)) 2375 type = RTN_MULTICAST; 2376 else if (ipv4_is_zeronet(fl4->daddr)) 2377 return ERR_PTR(-EINVAL); 2378 2379 if (dev_out->flags & IFF_LOOPBACK) 2380 flags |= RTCF_LOCAL; 2381 2382 do_cache = true; 2383 if (type == RTN_BROADCAST) { 2384 flags |= RTCF_BROADCAST | RTCF_LOCAL; 2385 fi = NULL; 2386 } else if (type == RTN_MULTICAST) { 2387 flags |= RTCF_MULTICAST | RTCF_LOCAL; 2388 if (!ip_check_mc_rcu(in_dev, fl4->daddr, fl4->saddr, 2389 fl4->flowi4_proto)) 2390 flags &= ~RTCF_LOCAL; 2391 else 2392 do_cache = false; 2393 /* If multicast route do not exist use 2394 * default one, but do not gateway in this case. 2395 * Yes, it is hack. 2396 */ 2397 if (fi && res->prefixlen < 4) 2398 fi = NULL; 2399 } else if ((type == RTN_LOCAL) && (orig_oif != 0) && 2400 (orig_oif != dev_out->ifindex)) { 2401 /* For local routes that require a particular output interface 2402 * we do not want to cache the result. Caching the result 2403 * causes incorrect behaviour when there are multiple source 2404 * addresses on the interface, the end result being that if the 2405 * intended recipient is waiting on that interface for the 2406 * packet he won't receive it because it will be delivered on 2407 * the loopback interface and the IP_PKTINFO ipi_ifindex will 2408 * be set to the loopback interface as well. 2409 */ 2410 do_cache = false; 2411 } 2412 2413 fnhe = NULL; 2414 do_cache &= fi != NULL; 2415 if (fi) { 2416 struct fib_nh_common *nhc = FIB_RES_NHC(*res); 2417 struct rtable __rcu **prth; 2418 2419 fnhe = find_exception(nhc, fl4->daddr); 2420 if (!do_cache) 2421 goto add; 2422 if (fnhe) { 2423 prth = &fnhe->fnhe_rth_output; 2424 } else { 2425 if (unlikely(fl4->flowi4_flags & 2426 FLOWI_FLAG_KNOWN_NH && 2427 !(nhc->nhc_gw_family && 2428 nhc->nhc_scope == RT_SCOPE_LINK))) { 2429 do_cache = false; 2430 goto add; 2431 } 2432 prth = raw_cpu_ptr(nhc->nhc_pcpu_rth_output); 2433 } 2434 rth = rcu_dereference(*prth); 2435 if (rt_cache_valid(rth) && dst_hold_safe(&rth->dst)) 2436 return rth; 2437 } 2438 2439 add: 2440 rth = rt_dst_alloc(dev_out, flags, type, 2441 IN_DEV_CONF_GET(in_dev, NOPOLICY), 2442 IN_DEV_CONF_GET(in_dev, NOXFRM)); 2443 if (!rth) 2444 return ERR_PTR(-ENOBUFS); 2445 2446 rth->rt_iif = orig_oif; 2447 2448 RT_CACHE_STAT_INC(out_slow_tot); 2449 2450 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 2451 if (flags & RTCF_LOCAL && 2452 !(dev_out->flags & IFF_LOOPBACK)) { 2453 rth->dst.output = ip_mc_output; 2454 RT_CACHE_STAT_INC(out_slow_mc); 2455 } 2456 #ifdef CONFIG_IP_MROUTE 2457 if (type == RTN_MULTICAST) { 2458 if (IN_DEV_MFORWARD(in_dev) && 2459 !ipv4_is_local_multicast(fl4->daddr)) { 2460 rth->dst.input = ip_mr_input; 2461 rth->dst.output = ip_mc_output; 2462 } 2463 } 2464 #endif 2465 } 2466 2467 rt_set_nexthop(rth, fl4->daddr, res, fnhe, fi, type, 0, do_cache); 2468 lwtunnel_set_redirect(&rth->dst); 2469 2470 return rth; 2471 } 2472 2473 /* 2474 * Major route resolver routine. 2475 */ 2476 2477 struct rtable *ip_route_output_key_hash(struct net *net, struct flowi4 *fl4, 2478 const struct sk_buff *skb) 2479 { 2480 __u8 tos = RT_FL_TOS(fl4); 2481 struct fib_result res = { 2482 .type = RTN_UNSPEC, 2483 .fi = NULL, 2484 .table = NULL, 2485 .tclassid = 0, 2486 }; 2487 struct rtable *rth; 2488 2489 fl4->flowi4_iif = LOOPBACK_IFINDEX; 2490 fl4->flowi4_tos = tos & IPTOS_RT_MASK; 2491 fl4->flowi4_scope = ((tos & RTO_ONLINK) ? 2492 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE); 2493 2494 rcu_read_lock(); 2495 rth = ip_route_output_key_hash_rcu(net, fl4, &res, skb); 2496 rcu_read_unlock(); 2497 2498 return rth; 2499 } 2500 EXPORT_SYMBOL_GPL(ip_route_output_key_hash); 2501 2502 struct rtable *ip_route_output_key_hash_rcu(struct net *net, struct flowi4 *fl4, 2503 struct fib_result *res, 2504 const struct sk_buff *skb) 2505 { 2506 struct net_device *dev_out = NULL; 2507 int orig_oif = fl4->flowi4_oif; 2508 unsigned int flags = 0; 2509 struct rtable *rth; 2510 int err; 2511 2512 if (fl4->saddr) { 2513 if (ipv4_is_multicast(fl4->saddr) || 2514 ipv4_is_lbcast(fl4->saddr) || 2515 ipv4_is_zeronet(fl4->saddr)) { 2516 rth = ERR_PTR(-EINVAL); 2517 goto out; 2518 } 2519 2520 rth = ERR_PTR(-ENETUNREACH); 2521 2522 /* I removed check for oif == dev_out->oif here. 2523 It was wrong for two reasons: 2524 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 2525 is assigned to multiple interfaces. 2526 2. Moreover, we are allowed to send packets with saddr 2527 of another iface. --ANK 2528 */ 2529 2530 if (fl4->flowi4_oif == 0 && 2531 (ipv4_is_multicast(fl4->daddr) || 2532 ipv4_is_lbcast(fl4->daddr))) { 2533 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2534 dev_out = __ip_dev_find(net, fl4->saddr, false); 2535 if (!dev_out) 2536 goto out; 2537 2538 /* Special hack: user can direct multicasts 2539 and limited broadcast via necessary interface 2540 without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 2541 This hack is not just for fun, it allows 2542 vic,vat and friends to work. 2543 They bind socket to loopback, set ttl to zero 2544 and expect that it will work. 2545 From the viewpoint of routing cache they are broken, 2546 because we are not allowed to build multicast path 2547 with loopback source addr (look, routing cache 2548 cannot know, that ttl is zero, so that packet 2549 will not leave this host and route is valid). 2550 Luckily, this hack is good workaround. 2551 */ 2552 2553 fl4->flowi4_oif = dev_out->ifindex; 2554 goto make_route; 2555 } 2556 2557 if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 2558 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2559 if (!__ip_dev_find(net, fl4->saddr, false)) 2560 goto out; 2561 } 2562 } 2563 2564 2565 if (fl4->flowi4_oif) { 2566 dev_out = dev_get_by_index_rcu(net, fl4->flowi4_oif); 2567 rth = ERR_PTR(-ENODEV); 2568 if (!dev_out) 2569 goto out; 2570 2571 /* RACE: Check return value of inet_select_addr instead. */ 2572 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 2573 rth = ERR_PTR(-ENETUNREACH); 2574 goto out; 2575 } 2576 if (ipv4_is_local_multicast(fl4->daddr) || 2577 ipv4_is_lbcast(fl4->daddr) || 2578 fl4->flowi4_proto == IPPROTO_IGMP) { 2579 if (!fl4->saddr) 2580 fl4->saddr = inet_select_addr(dev_out, 0, 2581 RT_SCOPE_LINK); 2582 goto make_route; 2583 } 2584 if (!fl4->saddr) { 2585 if (ipv4_is_multicast(fl4->daddr)) 2586 fl4->saddr = inet_select_addr(dev_out, 0, 2587 fl4->flowi4_scope); 2588 else if (!fl4->daddr) 2589 fl4->saddr = inet_select_addr(dev_out, 0, 2590 RT_SCOPE_HOST); 2591 } 2592 } 2593 2594 if (!fl4->daddr) { 2595 fl4->daddr = fl4->saddr; 2596 if (!fl4->daddr) 2597 fl4->daddr = fl4->saddr = htonl(INADDR_LOOPBACK); 2598 dev_out = net->loopback_dev; 2599 fl4->flowi4_oif = LOOPBACK_IFINDEX; 2600 res->type = RTN_LOCAL; 2601 flags |= RTCF_LOCAL; 2602 goto make_route; 2603 } 2604 2605 err = fib_lookup(net, fl4, res, 0); 2606 if (err) { 2607 res->fi = NULL; 2608 res->table = NULL; 2609 if (fl4->flowi4_oif && 2610 (ipv4_is_multicast(fl4->daddr) || 2611 !netif_index_is_l3_master(net, fl4->flowi4_oif))) { 2612 /* Apparently, routing tables are wrong. Assume, 2613 that the destination is on link. 2614 2615 WHY? DW. 2616 Because we are allowed to send to iface 2617 even if it has NO routes and NO assigned 2618 addresses. When oif is specified, routing 2619 tables are looked up with only one purpose: 2620 to catch if destination is gatewayed, rather than 2621 direct. Moreover, if MSG_DONTROUTE is set, 2622 we send packet, ignoring both routing tables 2623 and ifaddr state. --ANK 2624 2625 2626 We could make it even if oif is unknown, 2627 likely IPv6, but we do not. 2628 */ 2629 2630 if (fl4->saddr == 0) 2631 fl4->saddr = inet_select_addr(dev_out, 0, 2632 RT_SCOPE_LINK); 2633 res->type = RTN_UNICAST; 2634 goto make_route; 2635 } 2636 rth = ERR_PTR(err); 2637 goto out; 2638 } 2639 2640 if (res->type == RTN_LOCAL) { 2641 if (!fl4->saddr) { 2642 if (res->fi->fib_prefsrc) 2643 fl4->saddr = res->fi->fib_prefsrc; 2644 else 2645 fl4->saddr = fl4->daddr; 2646 } 2647 2648 /* L3 master device is the loopback for that domain */ 2649 dev_out = l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) ? : 2650 net->loopback_dev; 2651 2652 /* make sure orig_oif points to fib result device even 2653 * though packet rx/tx happens over loopback or l3mdev 2654 */ 2655 orig_oif = FIB_RES_OIF(*res); 2656 2657 fl4->flowi4_oif = dev_out->ifindex; 2658 flags |= RTCF_LOCAL; 2659 goto make_route; 2660 } 2661 2662 fib_select_path(net, res, fl4, skb); 2663 2664 dev_out = FIB_RES_DEV(*res); 2665 fl4->flowi4_oif = dev_out->ifindex; 2666 2667 2668 make_route: 2669 rth = __mkroute_output(res, fl4, orig_oif, dev_out, flags); 2670 2671 out: 2672 return rth; 2673 } 2674 2675 static struct dst_entry *ipv4_blackhole_dst_check(struct dst_entry *dst, u32 cookie) 2676 { 2677 return NULL; 2678 } 2679 2680 static unsigned int ipv4_blackhole_mtu(const struct dst_entry *dst) 2681 { 2682 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU); 2683 2684 return mtu ? : dst->dev->mtu; 2685 } 2686 2687 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk, 2688 struct sk_buff *skb, u32 mtu, 2689 bool confirm_neigh) 2690 { 2691 } 2692 2693 static void ipv4_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk, 2694 struct sk_buff *skb) 2695 { 2696 } 2697 2698 static u32 *ipv4_rt_blackhole_cow_metrics(struct dst_entry *dst, 2699 unsigned long old) 2700 { 2701 return NULL; 2702 } 2703 2704 static struct dst_ops ipv4_dst_blackhole_ops = { 2705 .family = AF_INET, 2706 .check = ipv4_blackhole_dst_check, 2707 .mtu = ipv4_blackhole_mtu, 2708 .default_advmss = ipv4_default_advmss, 2709 .update_pmtu = ipv4_rt_blackhole_update_pmtu, 2710 .redirect = ipv4_rt_blackhole_redirect, 2711 .cow_metrics = ipv4_rt_blackhole_cow_metrics, 2712 .neigh_lookup = ipv4_neigh_lookup, 2713 }; 2714 2715 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2716 { 2717 struct rtable *ort = (struct rtable *) dst_orig; 2718 struct rtable *rt; 2719 2720 rt = dst_alloc(&ipv4_dst_blackhole_ops, NULL, 1, DST_OBSOLETE_DEAD, 0); 2721 if (rt) { 2722 struct dst_entry *new = &rt->dst; 2723 2724 new->__use = 1; 2725 new->input = dst_discard; 2726 new->output = dst_discard_out; 2727 2728 new->dev = net->loopback_dev; 2729 if (new->dev) 2730 dev_hold(new->dev); 2731 2732 rt->rt_is_input = ort->rt_is_input; 2733 rt->rt_iif = ort->rt_iif; 2734 rt->rt_pmtu = ort->rt_pmtu; 2735 rt->rt_mtu_locked = ort->rt_mtu_locked; 2736 2737 rt->rt_genid = rt_genid_ipv4(net); 2738 rt->rt_flags = ort->rt_flags; 2739 rt->rt_type = ort->rt_type; 2740 rt->rt_uses_gateway = ort->rt_uses_gateway; 2741 rt->rt_gw_family = ort->rt_gw_family; 2742 if (rt->rt_gw_family == AF_INET) 2743 rt->rt_gw4 = ort->rt_gw4; 2744 else if (rt->rt_gw_family == AF_INET6) 2745 rt->rt_gw6 = ort->rt_gw6; 2746 2747 INIT_LIST_HEAD(&rt->rt_uncached); 2748 } 2749 2750 dst_release(dst_orig); 2751 2752 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2753 } 2754 2755 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2756 const struct sock *sk) 2757 { 2758 struct rtable *rt = __ip_route_output_key(net, flp4); 2759 2760 if (IS_ERR(rt)) 2761 return rt; 2762 2763 if (flp4->flowi4_proto) 2764 rt = (struct rtable *)xfrm_lookup_route(net, &rt->dst, 2765 flowi4_to_flowi(flp4), 2766 sk, 0); 2767 2768 return rt; 2769 } 2770 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2771 2772 struct rtable *ip_route_output_tunnel(struct sk_buff *skb, 2773 struct net_device *dev, 2774 struct net *net, __be32 *saddr, 2775 const struct ip_tunnel_info *info, 2776 u8 protocol, bool use_cache) 2777 { 2778 #ifdef CONFIG_DST_CACHE 2779 struct dst_cache *dst_cache; 2780 #endif 2781 struct rtable *rt = NULL; 2782 struct flowi4 fl4; 2783 __u8 tos; 2784 2785 #ifdef CONFIG_DST_CACHE 2786 dst_cache = (struct dst_cache *)&info->dst_cache; 2787 if (use_cache) { 2788 rt = dst_cache_get_ip4(dst_cache, saddr); 2789 if (rt) 2790 return rt; 2791 } 2792 #endif 2793 memset(&fl4, 0, sizeof(fl4)); 2794 fl4.flowi4_mark = skb->mark; 2795 fl4.flowi4_proto = protocol; 2796 fl4.daddr = info->key.u.ipv4.dst; 2797 fl4.saddr = info->key.u.ipv4.src; 2798 tos = info->key.tos; 2799 fl4.flowi4_tos = RT_TOS(tos); 2800 2801 rt = ip_route_output_key(net, &fl4); 2802 if (IS_ERR(rt)) { 2803 netdev_dbg(dev, "no route to %pI4\n", &fl4.daddr); 2804 return ERR_PTR(-ENETUNREACH); 2805 } 2806 if (rt->dst.dev == dev) { /* is this necessary? */ 2807 netdev_dbg(dev, "circular route to %pI4\n", &fl4.daddr); 2808 ip_rt_put(rt); 2809 return ERR_PTR(-ELOOP); 2810 } 2811 #ifdef CONFIG_DST_CACHE 2812 if (use_cache) 2813 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr); 2814 #endif 2815 *saddr = fl4.saddr; 2816 return rt; 2817 } 2818 EXPORT_SYMBOL_GPL(ip_route_output_tunnel); 2819 2820 /* called with rcu_read_lock held */ 2821 static int rt_fill_info(struct net *net, __be32 dst, __be32 src, 2822 struct rtable *rt, u32 table_id, struct flowi4 *fl4, 2823 struct sk_buff *skb, u32 portid, u32 seq, 2824 unsigned int flags) 2825 { 2826 struct rtmsg *r; 2827 struct nlmsghdr *nlh; 2828 unsigned long expires = 0; 2829 u32 error; 2830 u32 metrics[RTAX_MAX]; 2831 2832 nlh = nlmsg_put(skb, portid, seq, RTM_NEWROUTE, sizeof(*r), flags); 2833 if (!nlh) 2834 return -EMSGSIZE; 2835 2836 r = nlmsg_data(nlh); 2837 r->rtm_family = AF_INET; 2838 r->rtm_dst_len = 32; 2839 r->rtm_src_len = 0; 2840 r->rtm_tos = fl4 ? fl4->flowi4_tos : 0; 2841 r->rtm_table = table_id < 256 ? table_id : RT_TABLE_COMPAT; 2842 if (nla_put_u32(skb, RTA_TABLE, table_id)) 2843 goto nla_put_failure; 2844 r->rtm_type = rt->rt_type; 2845 r->rtm_scope = RT_SCOPE_UNIVERSE; 2846 r->rtm_protocol = RTPROT_UNSPEC; 2847 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 2848 if (rt->rt_flags & RTCF_NOTIFY) 2849 r->rtm_flags |= RTM_F_NOTIFY; 2850 if (IPCB(skb)->flags & IPSKB_DOREDIRECT) 2851 r->rtm_flags |= RTCF_DOREDIRECT; 2852 2853 if (nla_put_in_addr(skb, RTA_DST, dst)) 2854 goto nla_put_failure; 2855 if (src) { 2856 r->rtm_src_len = 32; 2857 if (nla_put_in_addr(skb, RTA_SRC, src)) 2858 goto nla_put_failure; 2859 } 2860 if (rt->dst.dev && 2861 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex)) 2862 goto nla_put_failure; 2863 #ifdef CONFIG_IP_ROUTE_CLASSID 2864 if (rt->dst.tclassid && 2865 nla_put_u32(skb, RTA_FLOW, rt->dst.tclassid)) 2866 goto nla_put_failure; 2867 #endif 2868 if (fl4 && !rt_is_input_route(rt) && 2869 fl4->saddr != src) { 2870 if (nla_put_in_addr(skb, RTA_PREFSRC, fl4->saddr)) 2871 goto nla_put_failure; 2872 } 2873 if (rt->rt_uses_gateway) { 2874 if (rt->rt_gw_family == AF_INET && 2875 nla_put_in_addr(skb, RTA_GATEWAY, rt->rt_gw4)) { 2876 goto nla_put_failure; 2877 } else if (rt->rt_gw_family == AF_INET6) { 2878 int alen = sizeof(struct in6_addr); 2879 struct nlattr *nla; 2880 struct rtvia *via; 2881 2882 nla = nla_reserve(skb, RTA_VIA, alen + 2); 2883 if (!nla) 2884 goto nla_put_failure; 2885 2886 via = nla_data(nla); 2887 via->rtvia_family = AF_INET6; 2888 memcpy(via->rtvia_addr, &rt->rt_gw6, alen); 2889 } 2890 } 2891 2892 expires = rt->dst.expires; 2893 if (expires) { 2894 unsigned long now = jiffies; 2895 2896 if (time_before(now, expires)) 2897 expires -= now; 2898 else 2899 expires = 0; 2900 } 2901 2902 memcpy(metrics, dst_metrics_ptr(&rt->dst), sizeof(metrics)); 2903 if (rt->rt_pmtu && expires) 2904 metrics[RTAX_MTU - 1] = rt->rt_pmtu; 2905 if (rt->rt_mtu_locked && expires) 2906 metrics[RTAX_LOCK - 1] |= BIT(RTAX_MTU); 2907 if (rtnetlink_put_metrics(skb, metrics) < 0) 2908 goto nla_put_failure; 2909 2910 if (fl4) { 2911 if (fl4->flowi4_mark && 2912 nla_put_u32(skb, RTA_MARK, fl4->flowi4_mark)) 2913 goto nla_put_failure; 2914 2915 if (!uid_eq(fl4->flowi4_uid, INVALID_UID) && 2916 nla_put_u32(skb, RTA_UID, 2917 from_kuid_munged(current_user_ns(), 2918 fl4->flowi4_uid))) 2919 goto nla_put_failure; 2920 2921 if (rt_is_input_route(rt)) { 2922 #ifdef CONFIG_IP_MROUTE 2923 if (ipv4_is_multicast(dst) && 2924 !ipv4_is_local_multicast(dst) && 2925 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) { 2926 int err = ipmr_get_route(net, skb, 2927 fl4->saddr, fl4->daddr, 2928 r, portid); 2929 2930 if (err <= 0) { 2931 if (err == 0) 2932 return 0; 2933 goto nla_put_failure; 2934 } 2935 } else 2936 #endif 2937 if (nla_put_u32(skb, RTA_IIF, fl4->flowi4_iif)) 2938 goto nla_put_failure; 2939 } 2940 } 2941 2942 error = rt->dst.error; 2943 2944 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires, error) < 0) 2945 goto nla_put_failure; 2946 2947 nlmsg_end(skb, nlh); 2948 return 0; 2949 2950 nla_put_failure: 2951 nlmsg_cancel(skb, nlh); 2952 return -EMSGSIZE; 2953 } 2954 2955 static int fnhe_dump_bucket(struct net *net, struct sk_buff *skb, 2956 struct netlink_callback *cb, u32 table_id, 2957 struct fnhe_hash_bucket *bucket, int genid, 2958 int *fa_index, int fa_start, unsigned int flags) 2959 { 2960 int i; 2961 2962 for (i = 0; i < FNHE_HASH_SIZE; i++) { 2963 struct fib_nh_exception *fnhe; 2964 2965 for (fnhe = rcu_dereference(bucket[i].chain); fnhe; 2966 fnhe = rcu_dereference(fnhe->fnhe_next)) { 2967 struct rtable *rt; 2968 int err; 2969 2970 if (*fa_index < fa_start) 2971 goto next; 2972 2973 if (fnhe->fnhe_genid != genid) 2974 goto next; 2975 2976 if (fnhe->fnhe_expires && 2977 time_after(jiffies, fnhe->fnhe_expires)) 2978 goto next; 2979 2980 rt = rcu_dereference(fnhe->fnhe_rth_input); 2981 if (!rt) 2982 rt = rcu_dereference(fnhe->fnhe_rth_output); 2983 if (!rt) 2984 goto next; 2985 2986 err = rt_fill_info(net, fnhe->fnhe_daddr, 0, rt, 2987 table_id, NULL, skb, 2988 NETLINK_CB(cb->skb).portid, 2989 cb->nlh->nlmsg_seq, flags); 2990 if (err) 2991 return err; 2992 next: 2993 (*fa_index)++; 2994 } 2995 } 2996 2997 return 0; 2998 } 2999 3000 int fib_dump_info_fnhe(struct sk_buff *skb, struct netlink_callback *cb, 3001 u32 table_id, struct fib_info *fi, 3002 int *fa_index, int fa_start, unsigned int flags) 3003 { 3004 struct net *net = sock_net(cb->skb->sk); 3005 int nhsel, genid = fnhe_genid(net); 3006 3007 for (nhsel = 0; nhsel < fib_info_num_path(fi); nhsel++) { 3008 struct fib_nh_common *nhc = fib_info_nhc(fi, nhsel); 3009 struct fnhe_hash_bucket *bucket; 3010 int err; 3011 3012 if (nhc->nhc_flags & RTNH_F_DEAD) 3013 continue; 3014 3015 rcu_read_lock(); 3016 bucket = rcu_dereference(nhc->nhc_exceptions); 3017 err = 0; 3018 if (bucket) 3019 err = fnhe_dump_bucket(net, skb, cb, table_id, bucket, 3020 genid, fa_index, fa_start, 3021 flags); 3022 rcu_read_unlock(); 3023 if (err) 3024 return err; 3025 } 3026 3027 return 0; 3028 } 3029 3030 static struct sk_buff *inet_rtm_getroute_build_skb(__be32 src, __be32 dst, 3031 u8 ip_proto, __be16 sport, 3032 __be16 dport) 3033 { 3034 struct sk_buff *skb; 3035 struct iphdr *iph; 3036 3037 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 3038 if (!skb) 3039 return NULL; 3040 3041 /* Reserve room for dummy headers, this skb can pass 3042 * through good chunk of routing engine. 3043 */ 3044 skb_reset_mac_header(skb); 3045 skb_reset_network_header(skb); 3046 skb->protocol = htons(ETH_P_IP); 3047 iph = skb_put(skb, sizeof(struct iphdr)); 3048 iph->protocol = ip_proto; 3049 iph->saddr = src; 3050 iph->daddr = dst; 3051 iph->version = 0x4; 3052 iph->frag_off = 0; 3053 iph->ihl = 0x5; 3054 skb_set_transport_header(skb, skb->len); 3055 3056 switch (iph->protocol) { 3057 case IPPROTO_UDP: { 3058 struct udphdr *udph; 3059 3060 udph = skb_put_zero(skb, sizeof(struct udphdr)); 3061 udph->source = sport; 3062 udph->dest = dport; 3063 udph->len = sizeof(struct udphdr); 3064 udph->check = 0; 3065 break; 3066 } 3067 case IPPROTO_TCP: { 3068 struct tcphdr *tcph; 3069 3070 tcph = skb_put_zero(skb, sizeof(struct tcphdr)); 3071 tcph->source = sport; 3072 tcph->dest = dport; 3073 tcph->doff = sizeof(struct tcphdr) / 4; 3074 tcph->rst = 1; 3075 tcph->check = ~tcp_v4_check(sizeof(struct tcphdr), 3076 src, dst, 0); 3077 break; 3078 } 3079 case IPPROTO_ICMP: { 3080 struct icmphdr *icmph; 3081 3082 icmph = skb_put_zero(skb, sizeof(struct icmphdr)); 3083 icmph->type = ICMP_ECHO; 3084 icmph->code = 0; 3085 } 3086 } 3087 3088 return skb; 3089 } 3090 3091 static int inet_rtm_valid_getroute_req(struct sk_buff *skb, 3092 const struct nlmsghdr *nlh, 3093 struct nlattr **tb, 3094 struct netlink_ext_ack *extack) 3095 { 3096 struct rtmsg *rtm; 3097 int i, err; 3098 3099 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) { 3100 NL_SET_ERR_MSG(extack, 3101 "ipv4: Invalid header for route get request"); 3102 return -EINVAL; 3103 } 3104 3105 if (!netlink_strict_get_check(skb)) 3106 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX, 3107 rtm_ipv4_policy, extack); 3108 3109 rtm = nlmsg_data(nlh); 3110 if ((rtm->rtm_src_len && rtm->rtm_src_len != 32) || 3111 (rtm->rtm_dst_len && rtm->rtm_dst_len != 32) || 3112 rtm->rtm_table || rtm->rtm_protocol || 3113 rtm->rtm_scope || rtm->rtm_type) { 3114 NL_SET_ERR_MSG(extack, "ipv4: Invalid values in header for route get request"); 3115 return -EINVAL; 3116 } 3117 3118 if (rtm->rtm_flags & ~(RTM_F_NOTIFY | 3119 RTM_F_LOOKUP_TABLE | 3120 RTM_F_FIB_MATCH)) { 3121 NL_SET_ERR_MSG(extack, "ipv4: Unsupported rtm_flags for route get request"); 3122 return -EINVAL; 3123 } 3124 3125 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX, 3126 rtm_ipv4_policy, extack); 3127 if (err) 3128 return err; 3129 3130 if ((tb[RTA_SRC] && !rtm->rtm_src_len) || 3131 (tb[RTA_DST] && !rtm->rtm_dst_len)) { 3132 NL_SET_ERR_MSG(extack, "ipv4: rtm_src_len and rtm_dst_len must be 32 for IPv4"); 3133 return -EINVAL; 3134 } 3135 3136 for (i = 0; i <= RTA_MAX; i++) { 3137 if (!tb[i]) 3138 continue; 3139 3140 switch (i) { 3141 case RTA_IIF: 3142 case RTA_OIF: 3143 case RTA_SRC: 3144 case RTA_DST: 3145 case RTA_IP_PROTO: 3146 case RTA_SPORT: 3147 case RTA_DPORT: 3148 case RTA_MARK: 3149 case RTA_UID: 3150 break; 3151 default: 3152 NL_SET_ERR_MSG(extack, "ipv4: Unsupported attribute in route get request"); 3153 return -EINVAL; 3154 } 3155 } 3156 3157 return 0; 3158 } 3159 3160 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, 3161 struct netlink_ext_ack *extack) 3162 { 3163 struct net *net = sock_net(in_skb->sk); 3164 struct nlattr *tb[RTA_MAX+1]; 3165 u32 table_id = RT_TABLE_MAIN; 3166 __be16 sport = 0, dport = 0; 3167 struct fib_result res = {}; 3168 u8 ip_proto = IPPROTO_UDP; 3169 struct rtable *rt = NULL; 3170 struct sk_buff *skb; 3171 struct rtmsg *rtm; 3172 struct flowi4 fl4 = {}; 3173 __be32 dst = 0; 3174 __be32 src = 0; 3175 kuid_t uid; 3176 u32 iif; 3177 int err; 3178 int mark; 3179 3180 err = inet_rtm_valid_getroute_req(in_skb, nlh, tb, extack); 3181 if (err < 0) 3182 return err; 3183 3184 rtm = nlmsg_data(nlh); 3185 src = tb[RTA_SRC] ? nla_get_in_addr(tb[RTA_SRC]) : 0; 3186 dst = tb[RTA_DST] ? nla_get_in_addr(tb[RTA_DST]) : 0; 3187 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0; 3188 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0; 3189 if (tb[RTA_UID]) 3190 uid = make_kuid(current_user_ns(), nla_get_u32(tb[RTA_UID])); 3191 else 3192 uid = (iif ? INVALID_UID : current_uid()); 3193 3194 if (tb[RTA_IP_PROTO]) { 3195 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO], 3196 &ip_proto, AF_INET, extack); 3197 if (err) 3198 return err; 3199 } 3200 3201 if (tb[RTA_SPORT]) 3202 sport = nla_get_be16(tb[RTA_SPORT]); 3203 3204 if (tb[RTA_DPORT]) 3205 dport = nla_get_be16(tb[RTA_DPORT]); 3206 3207 skb = inet_rtm_getroute_build_skb(src, dst, ip_proto, sport, dport); 3208 if (!skb) 3209 return -ENOBUFS; 3210 3211 fl4.daddr = dst; 3212 fl4.saddr = src; 3213 fl4.flowi4_tos = rtm->rtm_tos; 3214 fl4.flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0; 3215 fl4.flowi4_mark = mark; 3216 fl4.flowi4_uid = uid; 3217 if (sport) 3218 fl4.fl4_sport = sport; 3219 if (dport) 3220 fl4.fl4_dport = dport; 3221 fl4.flowi4_proto = ip_proto; 3222 3223 rcu_read_lock(); 3224 3225 if (iif) { 3226 struct net_device *dev; 3227 3228 dev = dev_get_by_index_rcu(net, iif); 3229 if (!dev) { 3230 err = -ENODEV; 3231 goto errout_rcu; 3232 } 3233 3234 fl4.flowi4_iif = iif; /* for rt_fill_info */ 3235 skb->dev = dev; 3236 skb->mark = mark; 3237 err = ip_route_input_rcu(skb, dst, src, rtm->rtm_tos, 3238 dev, &res); 3239 3240 rt = skb_rtable(skb); 3241 if (err == 0 && rt->dst.error) 3242 err = -rt->dst.error; 3243 } else { 3244 fl4.flowi4_iif = LOOPBACK_IFINDEX; 3245 skb->dev = net->loopback_dev; 3246 rt = ip_route_output_key_hash_rcu(net, &fl4, &res, skb); 3247 err = 0; 3248 if (IS_ERR(rt)) 3249 err = PTR_ERR(rt); 3250 else 3251 skb_dst_set(skb, &rt->dst); 3252 } 3253 3254 if (err) 3255 goto errout_rcu; 3256 3257 if (rtm->rtm_flags & RTM_F_NOTIFY) 3258 rt->rt_flags |= RTCF_NOTIFY; 3259 3260 if (rtm->rtm_flags & RTM_F_LOOKUP_TABLE) 3261 table_id = res.table ? res.table->tb_id : 0; 3262 3263 /* reset skb for netlink reply msg */ 3264 skb_trim(skb, 0); 3265 skb_reset_network_header(skb); 3266 skb_reset_transport_header(skb); 3267 skb_reset_mac_header(skb); 3268 3269 if (rtm->rtm_flags & RTM_F_FIB_MATCH) { 3270 struct fib_rt_info fri; 3271 3272 if (!res.fi) { 3273 err = fib_props[res.type].error; 3274 if (!err) 3275 err = -EHOSTUNREACH; 3276 goto errout_rcu; 3277 } 3278 fri.fi = res.fi; 3279 fri.tb_id = table_id; 3280 fri.dst = res.prefix; 3281 fri.dst_len = res.prefixlen; 3282 fri.tos = fl4.flowi4_tos; 3283 fri.type = rt->rt_type; 3284 fri.offload = 0; 3285 fri.trap = 0; 3286 if (res.fa_head) { 3287 struct fib_alias *fa; 3288 3289 hlist_for_each_entry_rcu(fa, res.fa_head, fa_list) { 3290 u8 slen = 32 - fri.dst_len; 3291 3292 if (fa->fa_slen == slen && 3293 fa->tb_id == fri.tb_id && 3294 fa->fa_tos == fri.tos && 3295 fa->fa_info == res.fi && 3296 fa->fa_type == fri.type) { 3297 fri.offload = fa->offload; 3298 fri.trap = fa->trap; 3299 break; 3300 } 3301 } 3302 } 3303 err = fib_dump_info(skb, NETLINK_CB(in_skb).portid, 3304 nlh->nlmsg_seq, RTM_NEWROUTE, &fri, 0); 3305 } else { 3306 err = rt_fill_info(net, dst, src, rt, table_id, &fl4, skb, 3307 NETLINK_CB(in_skb).portid, 3308 nlh->nlmsg_seq, 0); 3309 } 3310 if (err < 0) 3311 goto errout_rcu; 3312 3313 rcu_read_unlock(); 3314 3315 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 3316 3317 errout_free: 3318 return err; 3319 errout_rcu: 3320 rcu_read_unlock(); 3321 kfree_skb(skb); 3322 goto errout_free; 3323 } 3324 3325 void ip_rt_multicast_event(struct in_device *in_dev) 3326 { 3327 rt_cache_flush(dev_net(in_dev->dev)); 3328 } 3329 3330 #ifdef CONFIG_SYSCTL 3331 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 3332 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 3333 static int ip_rt_gc_elasticity __read_mostly = 8; 3334 static int ip_min_valid_pmtu __read_mostly = IPV4_MIN_MTU; 3335 3336 static int ipv4_sysctl_rtcache_flush(struct ctl_table *__ctl, int write, 3337 void *buffer, size_t *lenp, loff_t *ppos) 3338 { 3339 struct net *net = (struct net *)__ctl->extra1; 3340 3341 if (write) { 3342 rt_cache_flush(net); 3343 fnhe_genid_bump(net); 3344 return 0; 3345 } 3346 3347 return -EINVAL; 3348 } 3349 3350 static struct ctl_table ipv4_route_table[] = { 3351 { 3352 .procname = "gc_thresh", 3353 .data = &ipv4_dst_ops.gc_thresh, 3354 .maxlen = sizeof(int), 3355 .mode = 0644, 3356 .proc_handler = proc_dointvec, 3357 }, 3358 { 3359 .procname = "max_size", 3360 .data = &ip_rt_max_size, 3361 .maxlen = sizeof(int), 3362 .mode = 0644, 3363 .proc_handler = proc_dointvec, 3364 }, 3365 { 3366 /* Deprecated. Use gc_min_interval_ms */ 3367 3368 .procname = "gc_min_interval", 3369 .data = &ip_rt_gc_min_interval, 3370 .maxlen = sizeof(int), 3371 .mode = 0644, 3372 .proc_handler = proc_dointvec_jiffies, 3373 }, 3374 { 3375 .procname = "gc_min_interval_ms", 3376 .data = &ip_rt_gc_min_interval, 3377 .maxlen = sizeof(int), 3378 .mode = 0644, 3379 .proc_handler = proc_dointvec_ms_jiffies, 3380 }, 3381 { 3382 .procname = "gc_timeout", 3383 .data = &ip_rt_gc_timeout, 3384 .maxlen = sizeof(int), 3385 .mode = 0644, 3386 .proc_handler = proc_dointvec_jiffies, 3387 }, 3388 { 3389 .procname = "gc_interval", 3390 .data = &ip_rt_gc_interval, 3391 .maxlen = sizeof(int), 3392 .mode = 0644, 3393 .proc_handler = proc_dointvec_jiffies, 3394 }, 3395 { 3396 .procname = "redirect_load", 3397 .data = &ip_rt_redirect_load, 3398 .maxlen = sizeof(int), 3399 .mode = 0644, 3400 .proc_handler = proc_dointvec, 3401 }, 3402 { 3403 .procname = "redirect_number", 3404 .data = &ip_rt_redirect_number, 3405 .maxlen = sizeof(int), 3406 .mode = 0644, 3407 .proc_handler = proc_dointvec, 3408 }, 3409 { 3410 .procname = "redirect_silence", 3411 .data = &ip_rt_redirect_silence, 3412 .maxlen = sizeof(int), 3413 .mode = 0644, 3414 .proc_handler = proc_dointvec, 3415 }, 3416 { 3417 .procname = "error_cost", 3418 .data = &ip_rt_error_cost, 3419 .maxlen = sizeof(int), 3420 .mode = 0644, 3421 .proc_handler = proc_dointvec, 3422 }, 3423 { 3424 .procname = "error_burst", 3425 .data = &ip_rt_error_burst, 3426 .maxlen = sizeof(int), 3427 .mode = 0644, 3428 .proc_handler = proc_dointvec, 3429 }, 3430 { 3431 .procname = "gc_elasticity", 3432 .data = &ip_rt_gc_elasticity, 3433 .maxlen = sizeof(int), 3434 .mode = 0644, 3435 .proc_handler = proc_dointvec, 3436 }, 3437 { 3438 .procname = "mtu_expires", 3439 .data = &ip_rt_mtu_expires, 3440 .maxlen = sizeof(int), 3441 .mode = 0644, 3442 .proc_handler = proc_dointvec_jiffies, 3443 }, 3444 { 3445 .procname = "min_pmtu", 3446 .data = &ip_rt_min_pmtu, 3447 .maxlen = sizeof(int), 3448 .mode = 0644, 3449 .proc_handler = proc_dointvec_minmax, 3450 .extra1 = &ip_min_valid_pmtu, 3451 }, 3452 { 3453 .procname = "min_adv_mss", 3454 .data = &ip_rt_min_advmss, 3455 .maxlen = sizeof(int), 3456 .mode = 0644, 3457 .proc_handler = proc_dointvec, 3458 }, 3459 { } 3460 }; 3461 3462 static const char ipv4_route_flush_procname[] = "flush"; 3463 3464 static struct ctl_table ipv4_route_flush_table[] = { 3465 { 3466 .procname = ipv4_route_flush_procname, 3467 .maxlen = sizeof(int), 3468 .mode = 0200, 3469 .proc_handler = ipv4_sysctl_rtcache_flush, 3470 }, 3471 { }, 3472 }; 3473 3474 static __net_init int sysctl_route_net_init(struct net *net) 3475 { 3476 struct ctl_table *tbl; 3477 3478 tbl = ipv4_route_flush_table; 3479 if (!net_eq(net, &init_net)) { 3480 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL); 3481 if (!tbl) 3482 goto err_dup; 3483 3484 /* Don't export non-whitelisted sysctls to unprivileged users */ 3485 if (net->user_ns != &init_user_ns) { 3486 if (tbl[0].procname != ipv4_route_flush_procname) 3487 tbl[0].procname = NULL; 3488 } 3489 } 3490 tbl[0].extra1 = net; 3491 3492 net->ipv4.route_hdr = register_net_sysctl(net, "net/ipv4/route", tbl); 3493 if (!net->ipv4.route_hdr) 3494 goto err_reg; 3495 return 0; 3496 3497 err_reg: 3498 if (tbl != ipv4_route_flush_table) 3499 kfree(tbl); 3500 err_dup: 3501 return -ENOMEM; 3502 } 3503 3504 static __net_exit void sysctl_route_net_exit(struct net *net) 3505 { 3506 struct ctl_table *tbl; 3507 3508 tbl = net->ipv4.route_hdr->ctl_table_arg; 3509 unregister_net_sysctl_table(net->ipv4.route_hdr); 3510 BUG_ON(tbl == ipv4_route_flush_table); 3511 kfree(tbl); 3512 } 3513 3514 static __net_initdata struct pernet_operations sysctl_route_ops = { 3515 .init = sysctl_route_net_init, 3516 .exit = sysctl_route_net_exit, 3517 }; 3518 #endif 3519 3520 static __net_init int rt_genid_init(struct net *net) 3521 { 3522 atomic_set(&net->ipv4.rt_genid, 0); 3523 atomic_set(&net->fnhe_genid, 0); 3524 atomic_set(&net->ipv4.dev_addr_genid, get_random_int()); 3525 return 0; 3526 } 3527 3528 static __net_initdata struct pernet_operations rt_genid_ops = { 3529 .init = rt_genid_init, 3530 }; 3531 3532 static int __net_init ipv4_inetpeer_init(struct net *net) 3533 { 3534 struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL); 3535 3536 if (!bp) 3537 return -ENOMEM; 3538 inet_peer_base_init(bp); 3539 net->ipv4.peers = bp; 3540 return 0; 3541 } 3542 3543 static void __net_exit ipv4_inetpeer_exit(struct net *net) 3544 { 3545 struct inet_peer_base *bp = net->ipv4.peers; 3546 3547 net->ipv4.peers = NULL; 3548 inetpeer_invalidate_tree(bp); 3549 kfree(bp); 3550 } 3551 3552 static __net_initdata struct pernet_operations ipv4_inetpeer_ops = { 3553 .init = ipv4_inetpeer_init, 3554 .exit = ipv4_inetpeer_exit, 3555 }; 3556 3557 #ifdef CONFIG_IP_ROUTE_CLASSID 3558 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 3559 #endif /* CONFIG_IP_ROUTE_CLASSID */ 3560 3561 int __init ip_rt_init(void) 3562 { 3563 int cpu; 3564 3565 ip_idents = kmalloc_array(IP_IDENTS_SZ, sizeof(*ip_idents), 3566 GFP_KERNEL); 3567 if (!ip_idents) 3568 panic("IP: failed to allocate ip_idents\n"); 3569 3570 prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents)); 3571 3572 ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL); 3573 if (!ip_tstamps) 3574 panic("IP: failed to allocate ip_tstamps\n"); 3575 3576 for_each_possible_cpu(cpu) { 3577 struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu); 3578 3579 INIT_LIST_HEAD(&ul->head); 3580 spin_lock_init(&ul->lock); 3581 } 3582 #ifdef CONFIG_IP_ROUTE_CLASSID 3583 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 3584 if (!ip_rt_acct) 3585 panic("IP: failed to allocate ip_rt_acct\n"); 3586 #endif 3587 3588 ipv4_dst_ops.kmem_cachep = 3589 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0, 3590 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 3591 3592 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 3593 3594 if (dst_entries_init(&ipv4_dst_ops) < 0) 3595 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 3596 3597 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 3598 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 3599 3600 ipv4_dst_ops.gc_thresh = ~0; 3601 ip_rt_max_size = INT_MAX; 3602 3603 devinet_init(); 3604 ip_fib_init(); 3605 3606 if (ip_rt_proc_init()) 3607 pr_err("Unable to create route proc files\n"); 3608 #ifdef CONFIG_XFRM 3609 xfrm_init(); 3610 xfrm4_init(); 3611 #endif 3612 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL, 3613 RTNL_FLAG_DOIT_UNLOCKED); 3614 3615 #ifdef CONFIG_SYSCTL 3616 register_pernet_subsys(&sysctl_route_ops); 3617 #endif 3618 register_pernet_subsys(&rt_genid_ops); 3619 register_pernet_subsys(&ipv4_inetpeer_ops); 3620 return 0; 3621 } 3622 3623 #ifdef CONFIG_SYSCTL 3624 /* 3625 * We really need to sanitize the damn ipv4 init order, then all 3626 * this nonsense will go away. 3627 */ 3628 void __init ip_static_sysctl_init(void) 3629 { 3630 register_net_sysctl(&init_net, "net/ipv4/route", ipv4_route_table); 3631 } 3632 #endif 3633