1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * ROUTE - implementation of the IP router. 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Alan Cox, <gw4pts@gw4pts.ampr.org> 11 * Linus Torvalds, <Linus.Torvalds@helsinki.fi> 12 * Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 13 * 14 * Fixes: 15 * Alan Cox : Verify area fixes. 16 * Alan Cox : cli() protects routing changes 17 * Rui Oliveira : ICMP routing table updates 18 * (rco@di.uminho.pt) Routing table insertion and update 19 * Linus Torvalds : Rewrote bits to be sensible 20 * Alan Cox : Added BSD route gw semantics 21 * Alan Cox : Super /proc >4K 22 * Alan Cox : MTU in route table 23 * Alan Cox : MSS actually. Also added the window 24 * clamper. 25 * Sam Lantinga : Fixed route matching in rt_del() 26 * Alan Cox : Routing cache support. 27 * Alan Cox : Removed compatibility cruft. 28 * Alan Cox : RTF_REJECT support. 29 * Alan Cox : TCP irtt support. 30 * Jonathan Naylor : Added Metric support. 31 * Miquel van Smoorenburg : BSD API fixes. 32 * Miquel van Smoorenburg : Metrics. 33 * Alan Cox : Use __u32 properly 34 * Alan Cox : Aligned routing errors more closely with BSD 35 * our system is still very different. 36 * Alan Cox : Faster /proc handling 37 * Alexey Kuznetsov : Massive rework to support tree based routing, 38 * routing caches and better behaviour. 39 * 40 * Olaf Erb : irtt wasn't being copied right. 41 * Bjorn Ekwall : Kerneld route support. 42 * Alan Cox : Multicast fixed (I hope) 43 * Pavel Krauz : Limited broadcast fixed 44 * Mike McLagan : Routing by source 45 * Alexey Kuznetsov : End of old history. Split to fib.c and 46 * route.c and rewritten from scratch. 47 * Andi Kleen : Load-limit warning messages. 48 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 49 * Vitaly E. Lavrov : Race condition in ip_route_input_slow. 50 * Tobias Ringstrom : Uninitialized res.type in ip_route_output_slow. 51 * Vladimir V. Ivanov : IP rule info (flowid) is really useful. 52 * Marc Boucher : routing by fwmark 53 * Robert Olsson : Added rt_cache statistics 54 * Arnaldo C. Melo : Convert proc stuff to seq_file 55 * Eric Dumazet : hashed spinlocks and rt_check_expire() fixes. 56 * Ilia Sotnikov : Ignore TOS on PMTUD and Redirect 57 * Ilia Sotnikov : Removed TOS from hash calculations 58 * 59 * This program is free software; you can redistribute it and/or 60 * modify it under the terms of the GNU General Public License 61 * as published by the Free Software Foundation; either version 62 * 2 of the License, or (at your option) any later version. 63 */ 64 65 #include <linux/module.h> 66 #include <asm/uaccess.h> 67 #include <asm/system.h> 68 #include <linux/bitops.h> 69 #include <linux/types.h> 70 #include <linux/kernel.h> 71 #include <linux/mm.h> 72 #include <linux/bootmem.h> 73 #include <linux/string.h> 74 #include <linux/socket.h> 75 #include <linux/sockios.h> 76 #include <linux/errno.h> 77 #include <linux/in.h> 78 #include <linux/inet.h> 79 #include <linux/netdevice.h> 80 #include <linux/proc_fs.h> 81 #include <linux/init.h> 82 #include <linux/workqueue.h> 83 #include <linux/skbuff.h> 84 #include <linux/inetdevice.h> 85 #include <linux/igmp.h> 86 #include <linux/pkt_sched.h> 87 #include <linux/mroute.h> 88 #include <linux/netfilter_ipv4.h> 89 #include <linux/random.h> 90 #include <linux/jhash.h> 91 #include <linux/rcupdate.h> 92 #include <linux/times.h> 93 #include <linux/slab.h> 94 #include <net/dst.h> 95 #include <net/net_namespace.h> 96 #include <net/protocol.h> 97 #include <net/ip.h> 98 #include <net/route.h> 99 #include <net/inetpeer.h> 100 #include <net/sock.h> 101 #include <net/ip_fib.h> 102 #include <net/arp.h> 103 #include <net/tcp.h> 104 #include <net/icmp.h> 105 #include <net/xfrm.h> 106 #include <net/netevent.h> 107 #include <net/rtnetlink.h> 108 #ifdef CONFIG_SYSCTL 109 #include <linux/sysctl.h> 110 #endif 111 112 #define RT_FL_TOS(oldflp4) \ 113 ((u32)(oldflp4->flowi4_tos & (IPTOS_RT_MASK | RTO_ONLINK))) 114 115 #define IP_MAX_MTU 0xFFF0 116 117 #define RT_GC_TIMEOUT (300*HZ) 118 119 static int ip_rt_max_size; 120 static int ip_rt_gc_timeout __read_mostly = RT_GC_TIMEOUT; 121 static int ip_rt_gc_interval __read_mostly = 60 * HZ; 122 static int ip_rt_gc_min_interval __read_mostly = HZ / 2; 123 static int ip_rt_redirect_number __read_mostly = 9; 124 static int ip_rt_redirect_load __read_mostly = HZ / 50; 125 static int ip_rt_redirect_silence __read_mostly = ((HZ / 50) << (9 + 1)); 126 static int ip_rt_error_cost __read_mostly = HZ; 127 static int ip_rt_error_burst __read_mostly = 5 * HZ; 128 static int ip_rt_gc_elasticity __read_mostly = 8; 129 static int ip_rt_mtu_expires __read_mostly = 10 * 60 * HZ; 130 static int ip_rt_min_pmtu __read_mostly = 512 + 20 + 20; 131 static int ip_rt_min_advmss __read_mostly = 256; 132 static int rt_chain_length_max __read_mostly = 20; 133 134 /* 135 * Interface to generic destination cache. 136 */ 137 138 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie); 139 static unsigned int ipv4_default_advmss(const struct dst_entry *dst); 140 static unsigned int ipv4_default_mtu(const struct dst_entry *dst); 141 static void ipv4_dst_destroy(struct dst_entry *dst); 142 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst); 143 static void ipv4_link_failure(struct sk_buff *skb); 144 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu); 145 static int rt_garbage_collect(struct dst_ops *ops); 146 147 static void ipv4_dst_ifdown(struct dst_entry *dst, struct net_device *dev, 148 int how) 149 { 150 } 151 152 static u32 *ipv4_cow_metrics(struct dst_entry *dst, unsigned long old) 153 { 154 struct rtable *rt = (struct rtable *) dst; 155 struct inet_peer *peer; 156 u32 *p = NULL; 157 158 if (!rt->peer) 159 rt_bind_peer(rt, 1); 160 161 peer = rt->peer; 162 if (peer) { 163 u32 *old_p = __DST_METRICS_PTR(old); 164 unsigned long prev, new; 165 166 p = peer->metrics; 167 if (inet_metrics_new(peer)) 168 memcpy(p, old_p, sizeof(u32) * RTAX_MAX); 169 170 new = (unsigned long) p; 171 prev = cmpxchg(&dst->_metrics, old, new); 172 173 if (prev != old) { 174 p = __DST_METRICS_PTR(prev); 175 if (prev & DST_METRICS_READ_ONLY) 176 p = NULL; 177 } else { 178 if (rt->fi) { 179 fib_info_put(rt->fi); 180 rt->fi = NULL; 181 } 182 } 183 } 184 return p; 185 } 186 187 static struct dst_ops ipv4_dst_ops = { 188 .family = AF_INET, 189 .protocol = cpu_to_be16(ETH_P_IP), 190 .gc = rt_garbage_collect, 191 .check = ipv4_dst_check, 192 .default_advmss = ipv4_default_advmss, 193 .default_mtu = ipv4_default_mtu, 194 .cow_metrics = ipv4_cow_metrics, 195 .destroy = ipv4_dst_destroy, 196 .ifdown = ipv4_dst_ifdown, 197 .negative_advice = ipv4_negative_advice, 198 .link_failure = ipv4_link_failure, 199 .update_pmtu = ip_rt_update_pmtu, 200 .local_out = __ip_local_out, 201 }; 202 203 #define ECN_OR_COST(class) TC_PRIO_##class 204 205 const __u8 ip_tos2prio[16] = { 206 TC_PRIO_BESTEFFORT, 207 ECN_OR_COST(BESTEFFORT), 208 TC_PRIO_BESTEFFORT, 209 ECN_OR_COST(BESTEFFORT), 210 TC_PRIO_BULK, 211 ECN_OR_COST(BULK), 212 TC_PRIO_BULK, 213 ECN_OR_COST(BULK), 214 TC_PRIO_INTERACTIVE, 215 ECN_OR_COST(INTERACTIVE), 216 TC_PRIO_INTERACTIVE, 217 ECN_OR_COST(INTERACTIVE), 218 TC_PRIO_INTERACTIVE_BULK, 219 ECN_OR_COST(INTERACTIVE_BULK), 220 TC_PRIO_INTERACTIVE_BULK, 221 ECN_OR_COST(INTERACTIVE_BULK) 222 }; 223 224 225 /* 226 * Route cache. 227 */ 228 229 /* The locking scheme is rather straight forward: 230 * 231 * 1) Read-Copy Update protects the buckets of the central route hash. 232 * 2) Only writers remove entries, and they hold the lock 233 * as they look at rtable reference counts. 234 * 3) Only readers acquire references to rtable entries, 235 * they do so with atomic increments and with the 236 * lock held. 237 */ 238 239 struct rt_hash_bucket { 240 struct rtable __rcu *chain; 241 }; 242 243 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \ 244 defined(CONFIG_PROVE_LOCKING) 245 /* 246 * Instead of using one spinlock for each rt_hash_bucket, we use a table of spinlocks 247 * The size of this table is a power of two and depends on the number of CPUS. 248 * (on lockdep we have a quite big spinlock_t, so keep the size down there) 249 */ 250 #ifdef CONFIG_LOCKDEP 251 # define RT_HASH_LOCK_SZ 256 252 #else 253 # if NR_CPUS >= 32 254 # define RT_HASH_LOCK_SZ 4096 255 # elif NR_CPUS >= 16 256 # define RT_HASH_LOCK_SZ 2048 257 # elif NR_CPUS >= 8 258 # define RT_HASH_LOCK_SZ 1024 259 # elif NR_CPUS >= 4 260 # define RT_HASH_LOCK_SZ 512 261 # else 262 # define RT_HASH_LOCK_SZ 256 263 # endif 264 #endif 265 266 static spinlock_t *rt_hash_locks; 267 # define rt_hash_lock_addr(slot) &rt_hash_locks[(slot) & (RT_HASH_LOCK_SZ - 1)] 268 269 static __init void rt_hash_lock_init(void) 270 { 271 int i; 272 273 rt_hash_locks = kmalloc(sizeof(spinlock_t) * RT_HASH_LOCK_SZ, 274 GFP_KERNEL); 275 if (!rt_hash_locks) 276 panic("IP: failed to allocate rt_hash_locks\n"); 277 278 for (i = 0; i < RT_HASH_LOCK_SZ; i++) 279 spin_lock_init(&rt_hash_locks[i]); 280 } 281 #else 282 # define rt_hash_lock_addr(slot) NULL 283 284 static inline void rt_hash_lock_init(void) 285 { 286 } 287 #endif 288 289 static struct rt_hash_bucket *rt_hash_table __read_mostly; 290 static unsigned rt_hash_mask __read_mostly; 291 static unsigned int rt_hash_log __read_mostly; 292 293 static DEFINE_PER_CPU(struct rt_cache_stat, rt_cache_stat); 294 #define RT_CACHE_STAT_INC(field) __this_cpu_inc(rt_cache_stat.field) 295 296 static inline unsigned int rt_hash(__be32 daddr, __be32 saddr, int idx, 297 int genid) 298 { 299 return jhash_3words((__force u32)daddr, (__force u32)saddr, 300 idx, genid) 301 & rt_hash_mask; 302 } 303 304 static inline int rt_genid(struct net *net) 305 { 306 return atomic_read(&net->ipv4.rt_genid); 307 } 308 309 #ifdef CONFIG_PROC_FS 310 struct rt_cache_iter_state { 311 struct seq_net_private p; 312 int bucket; 313 int genid; 314 }; 315 316 static struct rtable *rt_cache_get_first(struct seq_file *seq) 317 { 318 struct rt_cache_iter_state *st = seq->private; 319 struct rtable *r = NULL; 320 321 for (st->bucket = rt_hash_mask; st->bucket >= 0; --st->bucket) { 322 if (!rcu_dereference_raw(rt_hash_table[st->bucket].chain)) 323 continue; 324 rcu_read_lock_bh(); 325 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain); 326 while (r) { 327 if (dev_net(r->dst.dev) == seq_file_net(seq) && 328 r->rt_genid == st->genid) 329 return r; 330 r = rcu_dereference_bh(r->dst.rt_next); 331 } 332 rcu_read_unlock_bh(); 333 } 334 return r; 335 } 336 337 static struct rtable *__rt_cache_get_next(struct seq_file *seq, 338 struct rtable *r) 339 { 340 struct rt_cache_iter_state *st = seq->private; 341 342 r = rcu_dereference_bh(r->dst.rt_next); 343 while (!r) { 344 rcu_read_unlock_bh(); 345 do { 346 if (--st->bucket < 0) 347 return NULL; 348 } while (!rcu_dereference_raw(rt_hash_table[st->bucket].chain)); 349 rcu_read_lock_bh(); 350 r = rcu_dereference_bh(rt_hash_table[st->bucket].chain); 351 } 352 return r; 353 } 354 355 static struct rtable *rt_cache_get_next(struct seq_file *seq, 356 struct rtable *r) 357 { 358 struct rt_cache_iter_state *st = seq->private; 359 while ((r = __rt_cache_get_next(seq, r)) != NULL) { 360 if (dev_net(r->dst.dev) != seq_file_net(seq)) 361 continue; 362 if (r->rt_genid == st->genid) 363 break; 364 } 365 return r; 366 } 367 368 static struct rtable *rt_cache_get_idx(struct seq_file *seq, loff_t pos) 369 { 370 struct rtable *r = rt_cache_get_first(seq); 371 372 if (r) 373 while (pos && (r = rt_cache_get_next(seq, r))) 374 --pos; 375 return pos ? NULL : r; 376 } 377 378 static void *rt_cache_seq_start(struct seq_file *seq, loff_t *pos) 379 { 380 struct rt_cache_iter_state *st = seq->private; 381 if (*pos) 382 return rt_cache_get_idx(seq, *pos - 1); 383 st->genid = rt_genid(seq_file_net(seq)); 384 return SEQ_START_TOKEN; 385 } 386 387 static void *rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos) 388 { 389 struct rtable *r; 390 391 if (v == SEQ_START_TOKEN) 392 r = rt_cache_get_first(seq); 393 else 394 r = rt_cache_get_next(seq, v); 395 ++*pos; 396 return r; 397 } 398 399 static void rt_cache_seq_stop(struct seq_file *seq, void *v) 400 { 401 if (v && v != SEQ_START_TOKEN) 402 rcu_read_unlock_bh(); 403 } 404 405 static int rt_cache_seq_show(struct seq_file *seq, void *v) 406 { 407 if (v == SEQ_START_TOKEN) 408 seq_printf(seq, "%-127s\n", 409 "Iface\tDestination\tGateway \tFlags\t\tRefCnt\tUse\t" 410 "Metric\tSource\t\tMTU\tWindow\tIRTT\tTOS\tHHRef\t" 411 "HHUptod\tSpecDst"); 412 else { 413 struct rtable *r = v; 414 int len; 415 416 seq_printf(seq, "%s\t%08X\t%08X\t%8X\t%d\t%u\t%d\t" 417 "%08X\t%d\t%u\t%u\t%02X\t%d\t%1d\t%08X%n", 418 r->dst.dev ? r->dst.dev->name : "*", 419 (__force u32)r->rt_dst, 420 (__force u32)r->rt_gateway, 421 r->rt_flags, atomic_read(&r->dst.__refcnt), 422 r->dst.__use, 0, (__force u32)r->rt_src, 423 dst_metric_advmss(&r->dst) + 40, 424 dst_metric(&r->dst, RTAX_WINDOW), 425 (int)((dst_metric(&r->dst, RTAX_RTT) >> 3) + 426 dst_metric(&r->dst, RTAX_RTTVAR)), 427 r->rt_tos, 428 r->dst.hh ? atomic_read(&r->dst.hh->hh_refcnt) : -1, 429 r->dst.hh ? (r->dst.hh->hh_output == 430 dev_queue_xmit) : 0, 431 r->rt_spec_dst, &len); 432 433 seq_printf(seq, "%*s\n", 127 - len, ""); 434 } 435 return 0; 436 } 437 438 static const struct seq_operations rt_cache_seq_ops = { 439 .start = rt_cache_seq_start, 440 .next = rt_cache_seq_next, 441 .stop = rt_cache_seq_stop, 442 .show = rt_cache_seq_show, 443 }; 444 445 static int rt_cache_seq_open(struct inode *inode, struct file *file) 446 { 447 return seq_open_net(inode, file, &rt_cache_seq_ops, 448 sizeof(struct rt_cache_iter_state)); 449 } 450 451 static const struct file_operations rt_cache_seq_fops = { 452 .owner = THIS_MODULE, 453 .open = rt_cache_seq_open, 454 .read = seq_read, 455 .llseek = seq_lseek, 456 .release = seq_release_net, 457 }; 458 459 460 static void *rt_cpu_seq_start(struct seq_file *seq, loff_t *pos) 461 { 462 int cpu; 463 464 if (*pos == 0) 465 return SEQ_START_TOKEN; 466 467 for (cpu = *pos-1; cpu < nr_cpu_ids; ++cpu) { 468 if (!cpu_possible(cpu)) 469 continue; 470 *pos = cpu+1; 471 return &per_cpu(rt_cache_stat, cpu); 472 } 473 return NULL; 474 } 475 476 static void *rt_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos) 477 { 478 int cpu; 479 480 for (cpu = *pos; cpu < nr_cpu_ids; ++cpu) { 481 if (!cpu_possible(cpu)) 482 continue; 483 *pos = cpu+1; 484 return &per_cpu(rt_cache_stat, cpu); 485 } 486 return NULL; 487 488 } 489 490 static void rt_cpu_seq_stop(struct seq_file *seq, void *v) 491 { 492 493 } 494 495 static int rt_cpu_seq_show(struct seq_file *seq, void *v) 496 { 497 struct rt_cache_stat *st = v; 498 499 if (v == SEQ_START_TOKEN) { 500 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"); 501 return 0; 502 } 503 504 seq_printf(seq,"%08x %08x %08x %08x %08x %08x %08x %08x " 505 " %08x %08x %08x %08x %08x %08x %08x %08x %08x \n", 506 dst_entries_get_slow(&ipv4_dst_ops), 507 st->in_hit, 508 st->in_slow_tot, 509 st->in_slow_mc, 510 st->in_no_route, 511 st->in_brd, 512 st->in_martian_dst, 513 st->in_martian_src, 514 515 st->out_hit, 516 st->out_slow_tot, 517 st->out_slow_mc, 518 519 st->gc_total, 520 st->gc_ignored, 521 st->gc_goal_miss, 522 st->gc_dst_overflow, 523 st->in_hlist_search, 524 st->out_hlist_search 525 ); 526 return 0; 527 } 528 529 static const struct seq_operations rt_cpu_seq_ops = { 530 .start = rt_cpu_seq_start, 531 .next = rt_cpu_seq_next, 532 .stop = rt_cpu_seq_stop, 533 .show = rt_cpu_seq_show, 534 }; 535 536 537 static int rt_cpu_seq_open(struct inode *inode, struct file *file) 538 { 539 return seq_open(file, &rt_cpu_seq_ops); 540 } 541 542 static const struct file_operations rt_cpu_seq_fops = { 543 .owner = THIS_MODULE, 544 .open = rt_cpu_seq_open, 545 .read = seq_read, 546 .llseek = seq_lseek, 547 .release = seq_release, 548 }; 549 550 #ifdef CONFIG_IP_ROUTE_CLASSID 551 static int rt_acct_proc_show(struct seq_file *m, void *v) 552 { 553 struct ip_rt_acct *dst, *src; 554 unsigned int i, j; 555 556 dst = kcalloc(256, sizeof(struct ip_rt_acct), GFP_KERNEL); 557 if (!dst) 558 return -ENOMEM; 559 560 for_each_possible_cpu(i) { 561 src = (struct ip_rt_acct *)per_cpu_ptr(ip_rt_acct, i); 562 for (j = 0; j < 256; j++) { 563 dst[j].o_bytes += src[j].o_bytes; 564 dst[j].o_packets += src[j].o_packets; 565 dst[j].i_bytes += src[j].i_bytes; 566 dst[j].i_packets += src[j].i_packets; 567 } 568 } 569 570 seq_write(m, dst, 256 * sizeof(struct ip_rt_acct)); 571 kfree(dst); 572 return 0; 573 } 574 575 static int rt_acct_proc_open(struct inode *inode, struct file *file) 576 { 577 return single_open(file, rt_acct_proc_show, NULL); 578 } 579 580 static const struct file_operations rt_acct_proc_fops = { 581 .owner = THIS_MODULE, 582 .open = rt_acct_proc_open, 583 .read = seq_read, 584 .llseek = seq_lseek, 585 .release = single_release, 586 }; 587 #endif 588 589 static int __net_init ip_rt_do_proc_init(struct net *net) 590 { 591 struct proc_dir_entry *pde; 592 593 pde = proc_net_fops_create(net, "rt_cache", S_IRUGO, 594 &rt_cache_seq_fops); 595 if (!pde) 596 goto err1; 597 598 pde = proc_create("rt_cache", S_IRUGO, 599 net->proc_net_stat, &rt_cpu_seq_fops); 600 if (!pde) 601 goto err2; 602 603 #ifdef CONFIG_IP_ROUTE_CLASSID 604 pde = proc_create("rt_acct", 0, net->proc_net, &rt_acct_proc_fops); 605 if (!pde) 606 goto err3; 607 #endif 608 return 0; 609 610 #ifdef CONFIG_IP_ROUTE_CLASSID 611 err3: 612 remove_proc_entry("rt_cache", net->proc_net_stat); 613 #endif 614 err2: 615 remove_proc_entry("rt_cache", net->proc_net); 616 err1: 617 return -ENOMEM; 618 } 619 620 static void __net_exit ip_rt_do_proc_exit(struct net *net) 621 { 622 remove_proc_entry("rt_cache", net->proc_net_stat); 623 remove_proc_entry("rt_cache", net->proc_net); 624 #ifdef CONFIG_IP_ROUTE_CLASSID 625 remove_proc_entry("rt_acct", net->proc_net); 626 #endif 627 } 628 629 static struct pernet_operations ip_rt_proc_ops __net_initdata = { 630 .init = ip_rt_do_proc_init, 631 .exit = ip_rt_do_proc_exit, 632 }; 633 634 static int __init ip_rt_proc_init(void) 635 { 636 return register_pernet_subsys(&ip_rt_proc_ops); 637 } 638 639 #else 640 static inline int ip_rt_proc_init(void) 641 { 642 return 0; 643 } 644 #endif /* CONFIG_PROC_FS */ 645 646 static inline void rt_free(struct rtable *rt) 647 { 648 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free); 649 } 650 651 static inline void rt_drop(struct rtable *rt) 652 { 653 ip_rt_put(rt); 654 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free); 655 } 656 657 static inline int rt_fast_clean(struct rtable *rth) 658 { 659 /* Kill broadcast/multicast entries very aggresively, if they 660 collide in hash table with more useful entries */ 661 return (rth->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) && 662 rt_is_input_route(rth) && rth->dst.rt_next; 663 } 664 665 static inline int rt_valuable(struct rtable *rth) 666 { 667 return (rth->rt_flags & (RTCF_REDIRECTED | RTCF_NOTIFY)) || 668 (rth->peer && rth->peer->pmtu_expires); 669 } 670 671 static int rt_may_expire(struct rtable *rth, unsigned long tmo1, unsigned long tmo2) 672 { 673 unsigned long age; 674 int ret = 0; 675 676 if (atomic_read(&rth->dst.__refcnt)) 677 goto out; 678 679 age = jiffies - rth->dst.lastuse; 680 if ((age <= tmo1 && !rt_fast_clean(rth)) || 681 (age <= tmo2 && rt_valuable(rth))) 682 goto out; 683 ret = 1; 684 out: return ret; 685 } 686 687 /* Bits of score are: 688 * 31: very valuable 689 * 30: not quite useless 690 * 29..0: usage counter 691 */ 692 static inline u32 rt_score(struct rtable *rt) 693 { 694 u32 score = jiffies - rt->dst.lastuse; 695 696 score = ~score & ~(3<<30); 697 698 if (rt_valuable(rt)) 699 score |= (1<<31); 700 701 if (rt_is_output_route(rt) || 702 !(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST|RTCF_LOCAL))) 703 score |= (1<<30); 704 705 return score; 706 } 707 708 static inline bool rt_caching(const struct net *net) 709 { 710 return net->ipv4.current_rt_cache_rebuild_count <= 711 net->ipv4.sysctl_rt_cache_rebuild_count; 712 } 713 714 static inline bool compare_hash_inputs(const struct rtable *rt1, 715 const struct rtable *rt2) 716 { 717 return ((((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) | 718 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) | 719 (rt1->rt_iif ^ rt2->rt_iif)) == 0); 720 } 721 722 static inline int compare_keys(struct rtable *rt1, struct rtable *rt2) 723 { 724 return (((__force u32)rt1->rt_key_dst ^ (__force u32)rt2->rt_key_dst) | 725 ((__force u32)rt1->rt_key_src ^ (__force u32)rt2->rt_key_src) | 726 (rt1->rt_mark ^ rt2->rt_mark) | 727 (rt1->rt_tos ^ rt2->rt_tos) | 728 (rt1->rt_oif ^ rt2->rt_oif) | 729 (rt1->rt_iif ^ rt2->rt_iif)) == 0; 730 } 731 732 static inline int compare_netns(struct rtable *rt1, struct rtable *rt2) 733 { 734 return net_eq(dev_net(rt1->dst.dev), dev_net(rt2->dst.dev)); 735 } 736 737 static inline int rt_is_expired(struct rtable *rth) 738 { 739 return rth->rt_genid != rt_genid(dev_net(rth->dst.dev)); 740 } 741 742 /* 743 * Perform a full scan of hash table and free all entries. 744 * Can be called by a softirq or a process. 745 * In the later case, we want to be reschedule if necessary 746 */ 747 static void rt_do_flush(struct net *net, int process_context) 748 { 749 unsigned int i; 750 struct rtable *rth, *next; 751 752 for (i = 0; i <= rt_hash_mask; i++) { 753 struct rtable __rcu **pprev; 754 struct rtable *list; 755 756 if (process_context && need_resched()) 757 cond_resched(); 758 rth = rcu_dereference_raw(rt_hash_table[i].chain); 759 if (!rth) 760 continue; 761 762 spin_lock_bh(rt_hash_lock_addr(i)); 763 764 list = NULL; 765 pprev = &rt_hash_table[i].chain; 766 rth = rcu_dereference_protected(*pprev, 767 lockdep_is_held(rt_hash_lock_addr(i))); 768 769 while (rth) { 770 next = rcu_dereference_protected(rth->dst.rt_next, 771 lockdep_is_held(rt_hash_lock_addr(i))); 772 773 if (!net || 774 net_eq(dev_net(rth->dst.dev), net)) { 775 rcu_assign_pointer(*pprev, next); 776 rcu_assign_pointer(rth->dst.rt_next, list); 777 list = rth; 778 } else { 779 pprev = &rth->dst.rt_next; 780 } 781 rth = next; 782 } 783 784 spin_unlock_bh(rt_hash_lock_addr(i)); 785 786 for (; list; list = next) { 787 next = rcu_dereference_protected(list->dst.rt_next, 1); 788 rt_free(list); 789 } 790 } 791 } 792 793 /* 794 * While freeing expired entries, we compute average chain length 795 * and standard deviation, using fixed-point arithmetic. 796 * This to have an estimation of rt_chain_length_max 797 * rt_chain_length_max = max(elasticity, AVG + 4*SD) 798 * We use 3 bits for frational part, and 29 (or 61) for magnitude. 799 */ 800 801 #define FRACT_BITS 3 802 #define ONE (1UL << FRACT_BITS) 803 804 /* 805 * Given a hash chain and an item in this hash chain, 806 * find if a previous entry has the same hash_inputs 807 * (but differs on tos, mark or oif) 808 * Returns 0 if an alias is found. 809 * Returns ONE if rth has no alias before itself. 810 */ 811 static int has_noalias(const struct rtable *head, const struct rtable *rth) 812 { 813 const struct rtable *aux = head; 814 815 while (aux != rth) { 816 if (compare_hash_inputs(aux, rth)) 817 return 0; 818 aux = rcu_dereference_protected(aux->dst.rt_next, 1); 819 } 820 return ONE; 821 } 822 823 /* 824 * Perturbation of rt_genid by a small quantity [1..256] 825 * Using 8 bits of shuffling ensure we can call rt_cache_invalidate() 826 * many times (2^24) without giving recent rt_genid. 827 * Jenkins hash is strong enough that litle changes of rt_genid are OK. 828 */ 829 static void rt_cache_invalidate(struct net *net) 830 { 831 unsigned char shuffle; 832 833 get_random_bytes(&shuffle, sizeof(shuffle)); 834 atomic_add(shuffle + 1U, &net->ipv4.rt_genid); 835 } 836 837 /* 838 * delay < 0 : invalidate cache (fast : entries will be deleted later) 839 * delay >= 0 : invalidate & flush cache (can be long) 840 */ 841 void rt_cache_flush(struct net *net, int delay) 842 { 843 rt_cache_invalidate(net); 844 if (delay >= 0) 845 rt_do_flush(net, !in_softirq()); 846 } 847 848 /* Flush previous cache invalidated entries from the cache */ 849 void rt_cache_flush_batch(struct net *net) 850 { 851 rt_do_flush(net, !in_softirq()); 852 } 853 854 static void rt_emergency_hash_rebuild(struct net *net) 855 { 856 if (net_ratelimit()) 857 printk(KERN_WARNING "Route hash chain too long!\n"); 858 rt_cache_invalidate(net); 859 } 860 861 /* 862 Short description of GC goals. 863 864 We want to build algorithm, which will keep routing cache 865 at some equilibrium point, when number of aged off entries 866 is kept approximately equal to newly generated ones. 867 868 Current expiration strength is variable "expire". 869 We try to adjust it dynamically, so that if networking 870 is idle expires is large enough to keep enough of warm entries, 871 and when load increases it reduces to limit cache size. 872 */ 873 874 static int rt_garbage_collect(struct dst_ops *ops) 875 { 876 static unsigned long expire = RT_GC_TIMEOUT; 877 static unsigned long last_gc; 878 static int rover; 879 static int equilibrium; 880 struct rtable *rth; 881 struct rtable __rcu **rthp; 882 unsigned long now = jiffies; 883 int goal; 884 int entries = dst_entries_get_fast(&ipv4_dst_ops); 885 886 /* 887 * Garbage collection is pretty expensive, 888 * do not make it too frequently. 889 */ 890 891 RT_CACHE_STAT_INC(gc_total); 892 893 if (now - last_gc < ip_rt_gc_min_interval && 894 entries < ip_rt_max_size) { 895 RT_CACHE_STAT_INC(gc_ignored); 896 goto out; 897 } 898 899 entries = dst_entries_get_slow(&ipv4_dst_ops); 900 /* Calculate number of entries, which we want to expire now. */ 901 goal = entries - (ip_rt_gc_elasticity << rt_hash_log); 902 if (goal <= 0) { 903 if (equilibrium < ipv4_dst_ops.gc_thresh) 904 equilibrium = ipv4_dst_ops.gc_thresh; 905 goal = entries - equilibrium; 906 if (goal > 0) { 907 equilibrium += min_t(unsigned int, goal >> 1, rt_hash_mask + 1); 908 goal = entries - equilibrium; 909 } 910 } else { 911 /* We are in dangerous area. Try to reduce cache really 912 * aggressively. 913 */ 914 goal = max_t(unsigned int, goal >> 1, rt_hash_mask + 1); 915 equilibrium = entries - goal; 916 } 917 918 if (now - last_gc >= ip_rt_gc_min_interval) 919 last_gc = now; 920 921 if (goal <= 0) { 922 equilibrium += goal; 923 goto work_done; 924 } 925 926 do { 927 int i, k; 928 929 for (i = rt_hash_mask, k = rover; i >= 0; i--) { 930 unsigned long tmo = expire; 931 932 k = (k + 1) & rt_hash_mask; 933 rthp = &rt_hash_table[k].chain; 934 spin_lock_bh(rt_hash_lock_addr(k)); 935 while ((rth = rcu_dereference_protected(*rthp, 936 lockdep_is_held(rt_hash_lock_addr(k)))) != NULL) { 937 if (!rt_is_expired(rth) && 938 !rt_may_expire(rth, tmo, expire)) { 939 tmo >>= 1; 940 rthp = &rth->dst.rt_next; 941 continue; 942 } 943 *rthp = rth->dst.rt_next; 944 rt_free(rth); 945 goal--; 946 } 947 spin_unlock_bh(rt_hash_lock_addr(k)); 948 if (goal <= 0) 949 break; 950 } 951 rover = k; 952 953 if (goal <= 0) 954 goto work_done; 955 956 /* Goal is not achieved. We stop process if: 957 958 - if expire reduced to zero. Otherwise, expire is halfed. 959 - if table is not full. 960 - if we are called from interrupt. 961 - jiffies check is just fallback/debug loop breaker. 962 We will not spin here for long time in any case. 963 */ 964 965 RT_CACHE_STAT_INC(gc_goal_miss); 966 967 if (expire == 0) 968 break; 969 970 expire >>= 1; 971 #if RT_CACHE_DEBUG >= 2 972 printk(KERN_DEBUG "expire>> %u %d %d %d\n", expire, 973 dst_entries_get_fast(&ipv4_dst_ops), goal, i); 974 #endif 975 976 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size) 977 goto out; 978 } while (!in_softirq() && time_before_eq(jiffies, now)); 979 980 if (dst_entries_get_fast(&ipv4_dst_ops) < ip_rt_max_size) 981 goto out; 982 if (dst_entries_get_slow(&ipv4_dst_ops) < ip_rt_max_size) 983 goto out; 984 if (net_ratelimit()) 985 printk(KERN_WARNING "dst cache overflow\n"); 986 RT_CACHE_STAT_INC(gc_dst_overflow); 987 return 1; 988 989 work_done: 990 expire += ip_rt_gc_min_interval; 991 if (expire > ip_rt_gc_timeout || 992 dst_entries_get_fast(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh || 993 dst_entries_get_slow(&ipv4_dst_ops) < ipv4_dst_ops.gc_thresh) 994 expire = ip_rt_gc_timeout; 995 #if RT_CACHE_DEBUG >= 2 996 printk(KERN_DEBUG "expire++ %u %d %d %d\n", expire, 997 dst_entries_get_fast(&ipv4_dst_ops), goal, rover); 998 #endif 999 out: return 0; 1000 } 1001 1002 /* 1003 * Returns number of entries in a hash chain that have different hash_inputs 1004 */ 1005 static int slow_chain_length(const struct rtable *head) 1006 { 1007 int length = 0; 1008 const struct rtable *rth = head; 1009 1010 while (rth) { 1011 length += has_noalias(head, rth); 1012 rth = rcu_dereference_protected(rth->dst.rt_next, 1); 1013 } 1014 return length >> FRACT_BITS; 1015 } 1016 1017 static struct rtable *rt_intern_hash(unsigned hash, struct rtable *rt, 1018 struct sk_buff *skb, int ifindex) 1019 { 1020 struct rtable *rth, *cand; 1021 struct rtable __rcu **rthp, **candp; 1022 unsigned long now; 1023 u32 min_score; 1024 int chain_length; 1025 int attempts = !in_softirq(); 1026 1027 restart: 1028 chain_length = 0; 1029 min_score = ~(u32)0; 1030 cand = NULL; 1031 candp = NULL; 1032 now = jiffies; 1033 1034 if (!rt_caching(dev_net(rt->dst.dev))) { 1035 /* 1036 * If we're not caching, just tell the caller we 1037 * were successful and don't touch the route. The 1038 * caller hold the sole reference to the cache entry, and 1039 * it will be released when the caller is done with it. 1040 * If we drop it here, the callers have no way to resolve routes 1041 * when we're not caching. Instead, just point *rp at rt, so 1042 * the caller gets a single use out of the route 1043 * Note that we do rt_free on this new route entry, so that 1044 * once its refcount hits zero, we are still able to reap it 1045 * (Thanks Alexey) 1046 * Note: To avoid expensive rcu stuff for this uncached dst, 1047 * we set DST_NOCACHE so that dst_release() can free dst without 1048 * waiting a grace period. 1049 */ 1050 1051 rt->dst.flags |= DST_NOCACHE; 1052 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) { 1053 int err = arp_bind_neighbour(&rt->dst); 1054 if (err) { 1055 if (net_ratelimit()) 1056 printk(KERN_WARNING 1057 "Neighbour table failure & not caching routes.\n"); 1058 ip_rt_put(rt); 1059 return ERR_PTR(err); 1060 } 1061 } 1062 1063 goto skip_hashing; 1064 } 1065 1066 rthp = &rt_hash_table[hash].chain; 1067 1068 spin_lock_bh(rt_hash_lock_addr(hash)); 1069 while ((rth = rcu_dereference_protected(*rthp, 1070 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) { 1071 if (rt_is_expired(rth)) { 1072 *rthp = rth->dst.rt_next; 1073 rt_free(rth); 1074 continue; 1075 } 1076 if (compare_keys(rth, rt) && compare_netns(rth, rt)) { 1077 /* Put it first */ 1078 *rthp = rth->dst.rt_next; 1079 /* 1080 * Since lookup is lockfree, the deletion 1081 * must be visible to another weakly ordered CPU before 1082 * the insertion at the start of the hash chain. 1083 */ 1084 rcu_assign_pointer(rth->dst.rt_next, 1085 rt_hash_table[hash].chain); 1086 /* 1087 * Since lookup is lockfree, the update writes 1088 * must be ordered for consistency on SMP. 1089 */ 1090 rcu_assign_pointer(rt_hash_table[hash].chain, rth); 1091 1092 dst_use(&rth->dst, now); 1093 spin_unlock_bh(rt_hash_lock_addr(hash)); 1094 1095 rt_drop(rt); 1096 if (skb) 1097 skb_dst_set(skb, &rth->dst); 1098 return rth; 1099 } 1100 1101 if (!atomic_read(&rth->dst.__refcnt)) { 1102 u32 score = rt_score(rth); 1103 1104 if (score <= min_score) { 1105 cand = rth; 1106 candp = rthp; 1107 min_score = score; 1108 } 1109 } 1110 1111 chain_length++; 1112 1113 rthp = &rth->dst.rt_next; 1114 } 1115 1116 if (cand) { 1117 /* ip_rt_gc_elasticity used to be average length of chain 1118 * length, when exceeded gc becomes really aggressive. 1119 * 1120 * The second limit is less certain. At the moment it allows 1121 * only 2 entries per bucket. We will see. 1122 */ 1123 if (chain_length > ip_rt_gc_elasticity) { 1124 *candp = cand->dst.rt_next; 1125 rt_free(cand); 1126 } 1127 } else { 1128 if (chain_length > rt_chain_length_max && 1129 slow_chain_length(rt_hash_table[hash].chain) > rt_chain_length_max) { 1130 struct net *net = dev_net(rt->dst.dev); 1131 int num = ++net->ipv4.current_rt_cache_rebuild_count; 1132 if (!rt_caching(net)) { 1133 printk(KERN_WARNING "%s: %d rebuilds is over limit, route caching disabled\n", 1134 rt->dst.dev->name, num); 1135 } 1136 rt_emergency_hash_rebuild(net); 1137 spin_unlock_bh(rt_hash_lock_addr(hash)); 1138 1139 hash = rt_hash(rt->rt_key_dst, rt->rt_key_src, 1140 ifindex, rt_genid(net)); 1141 goto restart; 1142 } 1143 } 1144 1145 /* Try to bind route to arp only if it is output 1146 route or unicast forwarding path. 1147 */ 1148 if (rt->rt_type == RTN_UNICAST || rt_is_output_route(rt)) { 1149 int err = arp_bind_neighbour(&rt->dst); 1150 if (err) { 1151 spin_unlock_bh(rt_hash_lock_addr(hash)); 1152 1153 if (err != -ENOBUFS) { 1154 rt_drop(rt); 1155 return ERR_PTR(err); 1156 } 1157 1158 /* Neighbour tables are full and nothing 1159 can be released. Try to shrink route cache, 1160 it is most likely it holds some neighbour records. 1161 */ 1162 if (attempts-- > 0) { 1163 int saved_elasticity = ip_rt_gc_elasticity; 1164 int saved_int = ip_rt_gc_min_interval; 1165 ip_rt_gc_elasticity = 1; 1166 ip_rt_gc_min_interval = 0; 1167 rt_garbage_collect(&ipv4_dst_ops); 1168 ip_rt_gc_min_interval = saved_int; 1169 ip_rt_gc_elasticity = saved_elasticity; 1170 goto restart; 1171 } 1172 1173 if (net_ratelimit()) 1174 printk(KERN_WARNING "ipv4: Neighbour table overflow.\n"); 1175 rt_drop(rt); 1176 return ERR_PTR(-ENOBUFS); 1177 } 1178 } 1179 1180 rt->dst.rt_next = rt_hash_table[hash].chain; 1181 1182 #if RT_CACHE_DEBUG >= 2 1183 if (rt->dst.rt_next) { 1184 struct rtable *trt; 1185 printk(KERN_DEBUG "rt_cache @%02x: %pI4", 1186 hash, &rt->rt_dst); 1187 for (trt = rt->dst.rt_next; trt; trt = trt->dst.rt_next) 1188 printk(" . %pI4", &trt->rt_dst); 1189 printk("\n"); 1190 } 1191 #endif 1192 /* 1193 * Since lookup is lockfree, we must make sure 1194 * previous writes to rt are committed to memory 1195 * before making rt visible to other CPUS. 1196 */ 1197 rcu_assign_pointer(rt_hash_table[hash].chain, rt); 1198 1199 spin_unlock_bh(rt_hash_lock_addr(hash)); 1200 1201 skip_hashing: 1202 if (skb) 1203 skb_dst_set(skb, &rt->dst); 1204 return rt; 1205 } 1206 1207 static atomic_t __rt_peer_genid = ATOMIC_INIT(0); 1208 1209 static u32 rt_peer_genid(void) 1210 { 1211 return atomic_read(&__rt_peer_genid); 1212 } 1213 1214 void rt_bind_peer(struct rtable *rt, int create) 1215 { 1216 struct inet_peer *peer; 1217 1218 peer = inet_getpeer_v4(rt->rt_dst, create); 1219 1220 if (peer && cmpxchg(&rt->peer, NULL, peer) != NULL) 1221 inet_putpeer(peer); 1222 else 1223 rt->rt_peer_genid = rt_peer_genid(); 1224 } 1225 1226 /* 1227 * Peer allocation may fail only in serious out-of-memory conditions. However 1228 * we still can generate some output. 1229 * Random ID selection looks a bit dangerous because we have no chances to 1230 * select ID being unique in a reasonable period of time. 1231 * But broken packet identifier may be better than no packet at all. 1232 */ 1233 static void ip_select_fb_ident(struct iphdr *iph) 1234 { 1235 static DEFINE_SPINLOCK(ip_fb_id_lock); 1236 static u32 ip_fallback_id; 1237 u32 salt; 1238 1239 spin_lock_bh(&ip_fb_id_lock); 1240 salt = secure_ip_id((__force __be32)ip_fallback_id ^ iph->daddr); 1241 iph->id = htons(salt & 0xFFFF); 1242 ip_fallback_id = salt; 1243 spin_unlock_bh(&ip_fb_id_lock); 1244 } 1245 1246 void __ip_select_ident(struct iphdr *iph, struct dst_entry *dst, int more) 1247 { 1248 struct rtable *rt = (struct rtable *) dst; 1249 1250 if (rt) { 1251 if (rt->peer == NULL) 1252 rt_bind_peer(rt, 1); 1253 1254 /* If peer is attached to destination, it is never detached, 1255 so that we need not to grab a lock to dereference it. 1256 */ 1257 if (rt->peer) { 1258 iph->id = htons(inet_getid(rt->peer, more)); 1259 return; 1260 } 1261 } else 1262 printk(KERN_DEBUG "rt_bind_peer(0) @%p\n", 1263 __builtin_return_address(0)); 1264 1265 ip_select_fb_ident(iph); 1266 } 1267 EXPORT_SYMBOL(__ip_select_ident); 1268 1269 static void rt_del(unsigned hash, struct rtable *rt) 1270 { 1271 struct rtable __rcu **rthp; 1272 struct rtable *aux; 1273 1274 rthp = &rt_hash_table[hash].chain; 1275 spin_lock_bh(rt_hash_lock_addr(hash)); 1276 ip_rt_put(rt); 1277 while ((aux = rcu_dereference_protected(*rthp, 1278 lockdep_is_held(rt_hash_lock_addr(hash)))) != NULL) { 1279 if (aux == rt || rt_is_expired(aux)) { 1280 *rthp = aux->dst.rt_next; 1281 rt_free(aux); 1282 continue; 1283 } 1284 rthp = &aux->dst.rt_next; 1285 } 1286 spin_unlock_bh(rt_hash_lock_addr(hash)); 1287 } 1288 1289 /* called in rcu_read_lock() section */ 1290 void ip_rt_redirect(__be32 old_gw, __be32 daddr, __be32 new_gw, 1291 __be32 saddr, struct net_device *dev) 1292 { 1293 struct in_device *in_dev = __in_dev_get_rcu(dev); 1294 struct inet_peer *peer; 1295 struct net *net; 1296 1297 if (!in_dev) 1298 return; 1299 1300 net = dev_net(dev); 1301 if (new_gw == old_gw || !IN_DEV_RX_REDIRECTS(in_dev) || 1302 ipv4_is_multicast(new_gw) || ipv4_is_lbcast(new_gw) || 1303 ipv4_is_zeronet(new_gw)) 1304 goto reject_redirect; 1305 1306 if (!IN_DEV_SHARED_MEDIA(in_dev)) { 1307 if (!inet_addr_onlink(in_dev, new_gw, old_gw)) 1308 goto reject_redirect; 1309 if (IN_DEV_SEC_REDIRECTS(in_dev) && ip_fib_check_default(new_gw, dev)) 1310 goto reject_redirect; 1311 } else { 1312 if (inet_addr_type(net, new_gw) != RTN_UNICAST) 1313 goto reject_redirect; 1314 } 1315 1316 peer = inet_getpeer_v4(daddr, 1); 1317 if (peer) { 1318 peer->redirect_learned.a4 = new_gw; 1319 1320 inet_putpeer(peer); 1321 1322 atomic_inc(&__rt_peer_genid); 1323 } 1324 return; 1325 1326 reject_redirect: 1327 #ifdef CONFIG_IP_ROUTE_VERBOSE 1328 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) 1329 printk(KERN_INFO "Redirect from %pI4 on %s about %pI4 ignored.\n" 1330 " Advised path = %pI4 -> %pI4\n", 1331 &old_gw, dev->name, &new_gw, 1332 &saddr, &daddr); 1333 #endif 1334 ; 1335 } 1336 1337 static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst) 1338 { 1339 struct rtable *rt = (struct rtable *)dst; 1340 struct dst_entry *ret = dst; 1341 1342 if (rt) { 1343 if (dst->obsolete > 0) { 1344 ip_rt_put(rt); 1345 ret = NULL; 1346 } else if (rt->rt_flags & RTCF_REDIRECTED) { 1347 unsigned hash = rt_hash(rt->rt_key_dst, rt->rt_key_src, 1348 rt->rt_oif, 1349 rt_genid(dev_net(dst->dev))); 1350 #if RT_CACHE_DEBUG >= 1 1351 printk(KERN_DEBUG "ipv4_negative_advice: redirect to %pI4/%02x dropped\n", 1352 &rt->rt_dst, rt->rt_tos); 1353 #endif 1354 rt_del(hash, rt); 1355 ret = NULL; 1356 } else if (rt->peer && 1357 rt->peer->pmtu_expires && 1358 time_after_eq(jiffies, rt->peer->pmtu_expires)) { 1359 unsigned long orig = rt->peer->pmtu_expires; 1360 1361 if (cmpxchg(&rt->peer->pmtu_expires, orig, 0) == orig) 1362 dst_metric_set(dst, RTAX_MTU, 1363 rt->peer->pmtu_orig); 1364 } 1365 } 1366 return ret; 1367 } 1368 1369 /* 1370 * Algorithm: 1371 * 1. The first ip_rt_redirect_number redirects are sent 1372 * with exponential backoff, then we stop sending them at all, 1373 * assuming that the host ignores our redirects. 1374 * 2. If we did not see packets requiring redirects 1375 * during ip_rt_redirect_silence, we assume that the host 1376 * forgot redirected route and start to send redirects again. 1377 * 1378 * This algorithm is much cheaper and more intelligent than dumb load limiting 1379 * in icmp.c. 1380 * 1381 * NOTE. Do not forget to inhibit load limiting for redirects (redundant) 1382 * and "frag. need" (breaks PMTU discovery) in icmp.c. 1383 */ 1384 1385 void ip_rt_send_redirect(struct sk_buff *skb) 1386 { 1387 struct rtable *rt = skb_rtable(skb); 1388 struct in_device *in_dev; 1389 struct inet_peer *peer; 1390 int log_martians; 1391 1392 rcu_read_lock(); 1393 in_dev = __in_dev_get_rcu(rt->dst.dev); 1394 if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) { 1395 rcu_read_unlock(); 1396 return; 1397 } 1398 log_martians = IN_DEV_LOG_MARTIANS(in_dev); 1399 rcu_read_unlock(); 1400 1401 if (!rt->peer) 1402 rt_bind_peer(rt, 1); 1403 peer = rt->peer; 1404 if (!peer) { 1405 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway); 1406 return; 1407 } 1408 1409 /* No redirected packets during ip_rt_redirect_silence; 1410 * reset the algorithm. 1411 */ 1412 if (time_after(jiffies, peer->rate_last + ip_rt_redirect_silence)) 1413 peer->rate_tokens = 0; 1414 1415 /* Too many ignored redirects; do not send anything 1416 * set dst.rate_last to the last seen redirected packet. 1417 */ 1418 if (peer->rate_tokens >= ip_rt_redirect_number) { 1419 peer->rate_last = jiffies; 1420 return; 1421 } 1422 1423 /* Check for load limit; set rate_last to the latest sent 1424 * redirect. 1425 */ 1426 if (peer->rate_tokens == 0 || 1427 time_after(jiffies, 1428 (peer->rate_last + 1429 (ip_rt_redirect_load << peer->rate_tokens)))) { 1430 icmp_send(skb, ICMP_REDIRECT, ICMP_REDIR_HOST, rt->rt_gateway); 1431 peer->rate_last = jiffies; 1432 ++peer->rate_tokens; 1433 #ifdef CONFIG_IP_ROUTE_VERBOSE 1434 if (log_martians && 1435 peer->rate_tokens == ip_rt_redirect_number && 1436 net_ratelimit()) 1437 printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n", 1438 &rt->rt_src, rt->rt_iif, 1439 &rt->rt_dst, &rt->rt_gateway); 1440 #endif 1441 } 1442 } 1443 1444 static int ip_error(struct sk_buff *skb) 1445 { 1446 struct rtable *rt = skb_rtable(skb); 1447 struct inet_peer *peer; 1448 unsigned long now; 1449 bool send; 1450 int code; 1451 1452 switch (rt->dst.error) { 1453 case EINVAL: 1454 default: 1455 goto out; 1456 case EHOSTUNREACH: 1457 code = ICMP_HOST_UNREACH; 1458 break; 1459 case ENETUNREACH: 1460 code = ICMP_NET_UNREACH; 1461 IP_INC_STATS_BH(dev_net(rt->dst.dev), 1462 IPSTATS_MIB_INNOROUTES); 1463 break; 1464 case EACCES: 1465 code = ICMP_PKT_FILTERED; 1466 break; 1467 } 1468 1469 if (!rt->peer) 1470 rt_bind_peer(rt, 1); 1471 peer = rt->peer; 1472 1473 send = true; 1474 if (peer) { 1475 now = jiffies; 1476 peer->rate_tokens += now - peer->rate_last; 1477 if (peer->rate_tokens > ip_rt_error_burst) 1478 peer->rate_tokens = ip_rt_error_burst; 1479 peer->rate_last = now; 1480 if (peer->rate_tokens >= ip_rt_error_cost) 1481 peer->rate_tokens -= ip_rt_error_cost; 1482 else 1483 send = false; 1484 } 1485 if (send) 1486 icmp_send(skb, ICMP_DEST_UNREACH, code, 0); 1487 1488 out: kfree_skb(skb); 1489 return 0; 1490 } 1491 1492 /* 1493 * The last two values are not from the RFC but 1494 * are needed for AMPRnet AX.25 paths. 1495 */ 1496 1497 static const unsigned short mtu_plateau[] = 1498 {32000, 17914, 8166, 4352, 2002, 1492, 576, 296, 216, 128 }; 1499 1500 static inline unsigned short guess_mtu(unsigned short old_mtu) 1501 { 1502 int i; 1503 1504 for (i = 0; i < ARRAY_SIZE(mtu_plateau); i++) 1505 if (old_mtu > mtu_plateau[i]) 1506 return mtu_plateau[i]; 1507 return 68; 1508 } 1509 1510 unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph, 1511 unsigned short new_mtu, 1512 struct net_device *dev) 1513 { 1514 unsigned short old_mtu = ntohs(iph->tot_len); 1515 unsigned short est_mtu = 0; 1516 struct inet_peer *peer; 1517 1518 peer = inet_getpeer_v4(iph->daddr, 1); 1519 if (peer) { 1520 unsigned short mtu = new_mtu; 1521 1522 if (new_mtu < 68 || new_mtu >= old_mtu) { 1523 /* BSD 4.2 derived systems incorrectly adjust 1524 * tot_len by the IP header length, and report 1525 * a zero MTU in the ICMP message. 1526 */ 1527 if (mtu == 0 && 1528 old_mtu >= 68 + (iph->ihl << 2)) 1529 old_mtu -= iph->ihl << 2; 1530 mtu = guess_mtu(old_mtu); 1531 } 1532 1533 if (mtu < ip_rt_min_pmtu) 1534 mtu = ip_rt_min_pmtu; 1535 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) { 1536 unsigned long pmtu_expires; 1537 1538 pmtu_expires = jiffies + ip_rt_mtu_expires; 1539 if (!pmtu_expires) 1540 pmtu_expires = 1UL; 1541 1542 est_mtu = mtu; 1543 peer->pmtu_learned = mtu; 1544 peer->pmtu_expires = pmtu_expires; 1545 } 1546 1547 inet_putpeer(peer); 1548 1549 atomic_inc(&__rt_peer_genid); 1550 } 1551 return est_mtu ? : new_mtu; 1552 } 1553 1554 static void check_peer_pmtu(struct dst_entry *dst, struct inet_peer *peer) 1555 { 1556 unsigned long expires = peer->pmtu_expires; 1557 1558 if (time_before(jiffies, expires)) { 1559 u32 orig_dst_mtu = dst_mtu(dst); 1560 if (peer->pmtu_learned < orig_dst_mtu) { 1561 if (!peer->pmtu_orig) 1562 peer->pmtu_orig = dst_metric_raw(dst, RTAX_MTU); 1563 dst_metric_set(dst, RTAX_MTU, peer->pmtu_learned); 1564 } 1565 } else if (cmpxchg(&peer->pmtu_expires, expires, 0) == expires) 1566 dst_metric_set(dst, RTAX_MTU, peer->pmtu_orig); 1567 } 1568 1569 static void ip_rt_update_pmtu(struct dst_entry *dst, u32 mtu) 1570 { 1571 struct rtable *rt = (struct rtable *) dst; 1572 struct inet_peer *peer; 1573 1574 dst_confirm(dst); 1575 1576 if (!rt->peer) 1577 rt_bind_peer(rt, 1); 1578 peer = rt->peer; 1579 if (peer) { 1580 if (mtu < ip_rt_min_pmtu) 1581 mtu = ip_rt_min_pmtu; 1582 if (!peer->pmtu_expires || mtu < peer->pmtu_learned) { 1583 unsigned long pmtu_expires; 1584 1585 pmtu_expires = jiffies + ip_rt_mtu_expires; 1586 if (!pmtu_expires) 1587 pmtu_expires = 1UL; 1588 1589 peer->pmtu_learned = mtu; 1590 peer->pmtu_expires = pmtu_expires; 1591 1592 atomic_inc(&__rt_peer_genid); 1593 rt->rt_peer_genid = rt_peer_genid(); 1594 } 1595 check_peer_pmtu(dst, peer); 1596 } 1597 } 1598 1599 static int check_peer_redir(struct dst_entry *dst, struct inet_peer *peer) 1600 { 1601 struct rtable *rt = (struct rtable *) dst; 1602 __be32 orig_gw = rt->rt_gateway; 1603 1604 dst_confirm(&rt->dst); 1605 1606 neigh_release(rt->dst.neighbour); 1607 rt->dst.neighbour = NULL; 1608 1609 rt->rt_gateway = peer->redirect_learned.a4; 1610 if (arp_bind_neighbour(&rt->dst) || 1611 !(rt->dst.neighbour->nud_state & NUD_VALID)) { 1612 if (rt->dst.neighbour) 1613 neigh_event_send(rt->dst.neighbour, NULL); 1614 rt->rt_gateway = orig_gw; 1615 return -EAGAIN; 1616 } else { 1617 rt->rt_flags |= RTCF_REDIRECTED; 1618 call_netevent_notifiers(NETEVENT_NEIGH_UPDATE, 1619 rt->dst.neighbour); 1620 } 1621 return 0; 1622 } 1623 1624 static struct dst_entry *ipv4_dst_check(struct dst_entry *dst, u32 cookie) 1625 { 1626 struct rtable *rt = (struct rtable *) dst; 1627 1628 if (rt_is_expired(rt)) 1629 return NULL; 1630 if (rt->rt_peer_genid != rt_peer_genid()) { 1631 struct inet_peer *peer; 1632 1633 if (!rt->peer) 1634 rt_bind_peer(rt, 0); 1635 1636 peer = rt->peer; 1637 if (peer && peer->pmtu_expires) 1638 check_peer_pmtu(dst, peer); 1639 1640 if (peer && peer->redirect_learned.a4 && 1641 peer->redirect_learned.a4 != rt->rt_gateway) { 1642 if (check_peer_redir(dst, peer)) 1643 return NULL; 1644 } 1645 1646 rt->rt_peer_genid = rt_peer_genid(); 1647 } 1648 return dst; 1649 } 1650 1651 static void ipv4_dst_destroy(struct dst_entry *dst) 1652 { 1653 struct rtable *rt = (struct rtable *) dst; 1654 struct inet_peer *peer = rt->peer; 1655 1656 if (rt->fi) { 1657 fib_info_put(rt->fi); 1658 rt->fi = NULL; 1659 } 1660 if (peer) { 1661 rt->peer = NULL; 1662 inet_putpeer(peer); 1663 } 1664 } 1665 1666 1667 static void ipv4_link_failure(struct sk_buff *skb) 1668 { 1669 struct rtable *rt; 1670 1671 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0); 1672 1673 rt = skb_rtable(skb); 1674 if (rt && 1675 rt->peer && 1676 rt->peer->pmtu_expires) { 1677 unsigned long orig = rt->peer->pmtu_expires; 1678 1679 if (cmpxchg(&rt->peer->pmtu_expires, orig, 0) == orig) 1680 dst_metric_set(&rt->dst, RTAX_MTU, rt->peer->pmtu_orig); 1681 } 1682 } 1683 1684 static int ip_rt_bug(struct sk_buff *skb) 1685 { 1686 printk(KERN_DEBUG "ip_rt_bug: %pI4 -> %pI4, %s\n", 1687 &ip_hdr(skb)->saddr, &ip_hdr(skb)->daddr, 1688 skb->dev ? skb->dev->name : "?"); 1689 kfree_skb(skb); 1690 return 0; 1691 } 1692 1693 /* 1694 We do not cache source address of outgoing interface, 1695 because it is used only by IP RR, TS and SRR options, 1696 so that it out of fast path. 1697 1698 BTW remember: "addr" is allowed to be not aligned 1699 in IP options! 1700 */ 1701 1702 void ip_rt_get_source(u8 *addr, struct rtable *rt) 1703 { 1704 __be32 src; 1705 struct fib_result res; 1706 1707 if (rt_is_output_route(rt)) 1708 src = rt->rt_src; 1709 else { 1710 struct flowi4 fl4 = { 1711 .daddr = rt->rt_key_dst, 1712 .saddr = rt->rt_key_src, 1713 .flowi4_tos = rt->rt_tos, 1714 .flowi4_oif = rt->rt_oif, 1715 .flowi4_iif = rt->rt_iif, 1716 .flowi4_mark = rt->rt_mark, 1717 }; 1718 1719 rcu_read_lock(); 1720 if (fib_lookup(dev_net(rt->dst.dev), &fl4, &res) == 0) 1721 src = FIB_RES_PREFSRC(dev_net(rt->dst.dev), res); 1722 else 1723 src = inet_select_addr(rt->dst.dev, rt->rt_gateway, 1724 RT_SCOPE_UNIVERSE); 1725 rcu_read_unlock(); 1726 } 1727 memcpy(addr, &src, 4); 1728 } 1729 1730 #ifdef CONFIG_IP_ROUTE_CLASSID 1731 static void set_class_tag(struct rtable *rt, u32 tag) 1732 { 1733 if (!(rt->dst.tclassid & 0xFFFF)) 1734 rt->dst.tclassid |= tag & 0xFFFF; 1735 if (!(rt->dst.tclassid & 0xFFFF0000)) 1736 rt->dst.tclassid |= tag & 0xFFFF0000; 1737 } 1738 #endif 1739 1740 static unsigned int ipv4_default_advmss(const struct dst_entry *dst) 1741 { 1742 unsigned int advmss = dst_metric_raw(dst, RTAX_ADVMSS); 1743 1744 if (advmss == 0) { 1745 advmss = max_t(unsigned int, dst->dev->mtu - 40, 1746 ip_rt_min_advmss); 1747 if (advmss > 65535 - 40) 1748 advmss = 65535 - 40; 1749 } 1750 return advmss; 1751 } 1752 1753 static unsigned int ipv4_default_mtu(const struct dst_entry *dst) 1754 { 1755 unsigned int mtu = dst->dev->mtu; 1756 1757 if (unlikely(dst_metric_locked(dst, RTAX_MTU))) { 1758 const struct rtable *rt = (const struct rtable *) dst; 1759 1760 if (rt->rt_gateway != rt->rt_dst && mtu > 576) 1761 mtu = 576; 1762 } 1763 1764 if (mtu > IP_MAX_MTU) 1765 mtu = IP_MAX_MTU; 1766 1767 return mtu; 1768 } 1769 1770 static void rt_init_metrics(struct rtable *rt, const struct flowi4 *oldflp4, 1771 struct fib_info *fi) 1772 { 1773 struct inet_peer *peer; 1774 int create = 0; 1775 1776 /* If a peer entry exists for this destination, we must hook 1777 * it up in order to get at cached metrics. 1778 */ 1779 if (oldflp4 && (oldflp4->flowi4_flags & FLOWI_FLAG_PRECOW_METRICS)) 1780 create = 1; 1781 1782 rt->peer = peer = inet_getpeer_v4(rt->rt_dst, create); 1783 if (peer) { 1784 rt->rt_peer_genid = rt_peer_genid(); 1785 if (inet_metrics_new(peer)) 1786 memcpy(peer->metrics, fi->fib_metrics, 1787 sizeof(u32) * RTAX_MAX); 1788 dst_init_metrics(&rt->dst, peer->metrics, false); 1789 1790 if (peer->pmtu_expires) 1791 check_peer_pmtu(&rt->dst, peer); 1792 if (peer->redirect_learned.a4 && 1793 peer->redirect_learned.a4 != rt->rt_gateway) { 1794 rt->rt_gateway = peer->redirect_learned.a4; 1795 rt->rt_flags |= RTCF_REDIRECTED; 1796 } 1797 } else { 1798 if (fi->fib_metrics != (u32 *) dst_default_metrics) { 1799 rt->fi = fi; 1800 atomic_inc(&fi->fib_clntref); 1801 } 1802 dst_init_metrics(&rt->dst, fi->fib_metrics, true); 1803 } 1804 } 1805 1806 static void rt_set_nexthop(struct rtable *rt, const struct flowi4 *oldflp4, 1807 const struct fib_result *res, 1808 struct fib_info *fi, u16 type, u32 itag) 1809 { 1810 struct dst_entry *dst = &rt->dst; 1811 1812 if (fi) { 1813 if (FIB_RES_GW(*res) && 1814 FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK) 1815 rt->rt_gateway = FIB_RES_GW(*res); 1816 rt_init_metrics(rt, oldflp4, fi); 1817 #ifdef CONFIG_IP_ROUTE_CLASSID 1818 dst->tclassid = FIB_RES_NH(*res).nh_tclassid; 1819 #endif 1820 } 1821 1822 if (dst_mtu(dst) > IP_MAX_MTU) 1823 dst_metric_set(dst, RTAX_MTU, IP_MAX_MTU); 1824 if (dst_metric_raw(dst, RTAX_ADVMSS) > 65535 - 40) 1825 dst_metric_set(dst, RTAX_ADVMSS, 65535 - 40); 1826 1827 #ifdef CONFIG_IP_ROUTE_CLASSID 1828 #ifdef CONFIG_IP_MULTIPLE_TABLES 1829 set_class_tag(rt, fib_rules_tclass(res)); 1830 #endif 1831 set_class_tag(rt, itag); 1832 #endif 1833 rt->rt_type = type; 1834 } 1835 1836 static struct rtable *rt_dst_alloc(bool nopolicy, bool noxfrm) 1837 { 1838 struct rtable *rt = dst_alloc(&ipv4_dst_ops, 1); 1839 if (rt) { 1840 rt->dst.obsolete = -1; 1841 1842 rt->dst.flags = DST_HOST | 1843 (nopolicy ? DST_NOPOLICY : 0) | 1844 (noxfrm ? DST_NOXFRM : 0); 1845 } 1846 return rt; 1847 } 1848 1849 /* called in rcu_read_lock() section */ 1850 static int ip_route_input_mc(struct sk_buff *skb, __be32 daddr, __be32 saddr, 1851 u8 tos, struct net_device *dev, int our) 1852 { 1853 unsigned int hash; 1854 struct rtable *rth; 1855 __be32 spec_dst; 1856 struct in_device *in_dev = __in_dev_get_rcu(dev); 1857 u32 itag = 0; 1858 int err; 1859 1860 /* Primary sanity checks. */ 1861 1862 if (in_dev == NULL) 1863 return -EINVAL; 1864 1865 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) || 1866 ipv4_is_loopback(saddr) || skb->protocol != htons(ETH_P_IP)) 1867 goto e_inval; 1868 1869 if (ipv4_is_zeronet(saddr)) { 1870 if (!ipv4_is_local_multicast(daddr)) 1871 goto e_inval; 1872 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK); 1873 } else { 1874 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst, 1875 &itag, 0); 1876 if (err < 0) 1877 goto e_err; 1878 } 1879 rth = rt_dst_alloc(IN_DEV_CONF_GET(in_dev, NOPOLICY), false); 1880 if (!rth) 1881 goto e_nobufs; 1882 1883 rth->dst.output = ip_rt_bug; 1884 1885 rth->rt_key_dst = daddr; 1886 rth->rt_dst = daddr; 1887 rth->rt_tos = tos; 1888 rth->rt_mark = skb->mark; 1889 rth->rt_key_src = saddr; 1890 rth->rt_src = saddr; 1891 #ifdef CONFIG_IP_ROUTE_CLASSID 1892 rth->dst.tclassid = itag; 1893 #endif 1894 rth->rt_iif = dev->ifindex; 1895 rth->dst.dev = init_net.loopback_dev; 1896 dev_hold(rth->dst.dev); 1897 rth->rt_oif = 0; 1898 rth->rt_gateway = daddr; 1899 rth->rt_spec_dst= spec_dst; 1900 rth->rt_genid = rt_genid(dev_net(dev)); 1901 rth->rt_flags = RTCF_MULTICAST; 1902 rth->rt_type = RTN_MULTICAST; 1903 if (our) { 1904 rth->dst.input= ip_local_deliver; 1905 rth->rt_flags |= RTCF_LOCAL; 1906 } 1907 1908 #ifdef CONFIG_IP_MROUTE 1909 if (!ipv4_is_local_multicast(daddr) && IN_DEV_MFORWARD(in_dev)) 1910 rth->dst.input = ip_mr_input; 1911 #endif 1912 RT_CACHE_STAT_INC(in_slow_mc); 1913 1914 hash = rt_hash(daddr, saddr, dev->ifindex, rt_genid(dev_net(dev))); 1915 rth = rt_intern_hash(hash, rth, skb, dev->ifindex); 1916 err = 0; 1917 if (IS_ERR(rth)) 1918 err = PTR_ERR(rth); 1919 1920 e_nobufs: 1921 return -ENOBUFS; 1922 e_inval: 1923 return -EINVAL; 1924 e_err: 1925 return err; 1926 } 1927 1928 1929 static void ip_handle_martian_source(struct net_device *dev, 1930 struct in_device *in_dev, 1931 struct sk_buff *skb, 1932 __be32 daddr, 1933 __be32 saddr) 1934 { 1935 RT_CACHE_STAT_INC(in_martian_src); 1936 #ifdef CONFIG_IP_ROUTE_VERBOSE 1937 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) { 1938 /* 1939 * RFC1812 recommendation, if source is martian, 1940 * the only hint is MAC header. 1941 */ 1942 printk(KERN_WARNING "martian source %pI4 from %pI4, on dev %s\n", 1943 &daddr, &saddr, dev->name); 1944 if (dev->hard_header_len && skb_mac_header_was_set(skb)) { 1945 int i; 1946 const unsigned char *p = skb_mac_header(skb); 1947 printk(KERN_WARNING "ll header: "); 1948 for (i = 0; i < dev->hard_header_len; i++, p++) { 1949 printk("%02x", *p); 1950 if (i < (dev->hard_header_len - 1)) 1951 printk(":"); 1952 } 1953 printk("\n"); 1954 } 1955 } 1956 #endif 1957 } 1958 1959 /* called in rcu_read_lock() section */ 1960 static int __mkroute_input(struct sk_buff *skb, 1961 const struct fib_result *res, 1962 struct in_device *in_dev, 1963 __be32 daddr, __be32 saddr, u32 tos, 1964 struct rtable **result) 1965 { 1966 struct rtable *rth; 1967 int err; 1968 struct in_device *out_dev; 1969 unsigned int flags = 0; 1970 __be32 spec_dst; 1971 u32 itag; 1972 1973 /* get a working reference to the output device */ 1974 out_dev = __in_dev_get_rcu(FIB_RES_DEV(*res)); 1975 if (out_dev == NULL) { 1976 if (net_ratelimit()) 1977 printk(KERN_CRIT "Bug in ip_route_input" \ 1978 "_slow(). Please, report\n"); 1979 return -EINVAL; 1980 } 1981 1982 1983 err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res), 1984 in_dev->dev, &spec_dst, &itag, skb->mark); 1985 if (err < 0) { 1986 ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr, 1987 saddr); 1988 1989 goto cleanup; 1990 } 1991 1992 if (err) 1993 flags |= RTCF_DIRECTSRC; 1994 1995 if (out_dev == in_dev && err && 1996 (IN_DEV_SHARED_MEDIA(out_dev) || 1997 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res)))) 1998 flags |= RTCF_DOREDIRECT; 1999 2000 if (skb->protocol != htons(ETH_P_IP)) { 2001 /* Not IP (i.e. ARP). Do not create route, if it is 2002 * invalid for proxy arp. DNAT routes are always valid. 2003 * 2004 * Proxy arp feature have been extended to allow, ARP 2005 * replies back to the same interface, to support 2006 * Private VLAN switch technologies. See arp.c. 2007 */ 2008 if (out_dev == in_dev && 2009 IN_DEV_PROXY_ARP_PVLAN(in_dev) == 0) { 2010 err = -EINVAL; 2011 goto cleanup; 2012 } 2013 } 2014 2015 rth = rt_dst_alloc(IN_DEV_CONF_GET(in_dev, NOPOLICY), 2016 IN_DEV_CONF_GET(out_dev, NOXFRM)); 2017 if (!rth) { 2018 err = -ENOBUFS; 2019 goto cleanup; 2020 } 2021 2022 rth->rt_key_dst = daddr; 2023 rth->rt_dst = daddr; 2024 rth->rt_tos = tos; 2025 rth->rt_mark = skb->mark; 2026 rth->rt_key_src = saddr; 2027 rth->rt_src = saddr; 2028 rth->rt_gateway = daddr; 2029 rth->rt_iif = in_dev->dev->ifindex; 2030 rth->dst.dev = (out_dev)->dev; 2031 dev_hold(rth->dst.dev); 2032 rth->rt_oif = 0; 2033 rth->rt_spec_dst= spec_dst; 2034 2035 rth->dst.input = ip_forward; 2036 rth->dst.output = ip_output; 2037 rth->rt_genid = rt_genid(dev_net(rth->dst.dev)); 2038 2039 rt_set_nexthop(rth, NULL, res, res->fi, res->type, itag); 2040 2041 rth->rt_flags = flags; 2042 2043 *result = rth; 2044 err = 0; 2045 cleanup: 2046 return err; 2047 } 2048 2049 static int ip_mkroute_input(struct sk_buff *skb, 2050 struct fib_result *res, 2051 const struct flowi4 *fl4, 2052 struct in_device *in_dev, 2053 __be32 daddr, __be32 saddr, u32 tos) 2054 { 2055 struct rtable* rth = NULL; 2056 int err; 2057 unsigned hash; 2058 2059 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2060 if (res->fi && res->fi->fib_nhs > 1) 2061 fib_select_multipath(res); 2062 #endif 2063 2064 /* create a routing cache entry */ 2065 err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth); 2066 if (err) 2067 return err; 2068 2069 /* put it into the cache */ 2070 hash = rt_hash(daddr, saddr, fl4->flowi4_iif, 2071 rt_genid(dev_net(rth->dst.dev))); 2072 rth = rt_intern_hash(hash, rth, skb, fl4->flowi4_iif); 2073 if (IS_ERR(rth)) 2074 return PTR_ERR(rth); 2075 return 0; 2076 } 2077 2078 /* 2079 * NOTE. We drop all the packets that has local source 2080 * addresses, because every properly looped back packet 2081 * must have correct destination already attached by output routine. 2082 * 2083 * Such approach solves two big problems: 2084 * 1. Not simplex devices are handled properly. 2085 * 2. IP spoofing attempts are filtered with 100% of guarantee. 2086 * called with rcu_read_lock() 2087 */ 2088 2089 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2090 u8 tos, struct net_device *dev) 2091 { 2092 struct fib_result res; 2093 struct in_device *in_dev = __in_dev_get_rcu(dev); 2094 struct flowi4 fl4; 2095 unsigned flags = 0; 2096 u32 itag = 0; 2097 struct rtable * rth; 2098 unsigned hash; 2099 __be32 spec_dst; 2100 int err = -EINVAL; 2101 struct net * net = dev_net(dev); 2102 2103 /* IP on this device is disabled. */ 2104 2105 if (!in_dev) 2106 goto out; 2107 2108 /* Check for the most weird martians, which can be not detected 2109 by fib_lookup. 2110 */ 2111 2112 if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) || 2113 ipv4_is_loopback(saddr)) 2114 goto martian_source; 2115 2116 if (ipv4_is_lbcast(daddr) || (saddr == 0 && daddr == 0)) 2117 goto brd_input; 2118 2119 /* Accept zero addresses only to limited broadcast; 2120 * I even do not know to fix it or not. Waiting for complains :-) 2121 */ 2122 if (ipv4_is_zeronet(saddr)) 2123 goto martian_source; 2124 2125 if (ipv4_is_zeronet(daddr) || ipv4_is_loopback(daddr)) 2126 goto martian_destination; 2127 2128 /* 2129 * Now we are ready to route packet. 2130 */ 2131 fl4.flowi4_oif = 0; 2132 fl4.flowi4_iif = dev->ifindex; 2133 fl4.flowi4_mark = skb->mark; 2134 fl4.flowi4_tos = tos; 2135 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 2136 fl4.daddr = daddr; 2137 fl4.saddr = saddr; 2138 err = fib_lookup(net, &fl4, &res); 2139 if (err != 0) { 2140 if (!IN_DEV_FORWARD(in_dev)) 2141 goto e_hostunreach; 2142 goto no_route; 2143 } 2144 2145 RT_CACHE_STAT_INC(in_slow_tot); 2146 2147 if (res.type == RTN_BROADCAST) 2148 goto brd_input; 2149 2150 if (res.type == RTN_LOCAL) { 2151 err = fib_validate_source(saddr, daddr, tos, 2152 net->loopback_dev->ifindex, 2153 dev, &spec_dst, &itag, skb->mark); 2154 if (err < 0) 2155 goto martian_source_keep_err; 2156 if (err) 2157 flags |= RTCF_DIRECTSRC; 2158 spec_dst = daddr; 2159 goto local_input; 2160 } 2161 2162 if (!IN_DEV_FORWARD(in_dev)) 2163 goto e_hostunreach; 2164 if (res.type != RTN_UNICAST) 2165 goto martian_destination; 2166 2167 err = ip_mkroute_input(skb, &res, &fl4, in_dev, daddr, saddr, tos); 2168 out: return err; 2169 2170 brd_input: 2171 if (skb->protocol != htons(ETH_P_IP)) 2172 goto e_inval; 2173 2174 if (ipv4_is_zeronet(saddr)) 2175 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK); 2176 else { 2177 err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst, 2178 &itag, skb->mark); 2179 if (err < 0) 2180 goto martian_source_keep_err; 2181 if (err) 2182 flags |= RTCF_DIRECTSRC; 2183 } 2184 flags |= RTCF_BROADCAST; 2185 res.type = RTN_BROADCAST; 2186 RT_CACHE_STAT_INC(in_brd); 2187 2188 local_input: 2189 rth = rt_dst_alloc(IN_DEV_CONF_GET(in_dev, NOPOLICY), false); 2190 if (!rth) 2191 goto e_nobufs; 2192 2193 rth->dst.output= ip_rt_bug; 2194 rth->rt_genid = rt_genid(net); 2195 2196 rth->rt_key_dst = daddr; 2197 rth->rt_dst = daddr; 2198 rth->rt_tos = tos; 2199 rth->rt_mark = skb->mark; 2200 rth->rt_key_src = saddr; 2201 rth->rt_src = saddr; 2202 #ifdef CONFIG_IP_ROUTE_CLASSID 2203 rth->dst.tclassid = itag; 2204 #endif 2205 rth->rt_iif = dev->ifindex; 2206 rth->dst.dev = net->loopback_dev; 2207 dev_hold(rth->dst.dev); 2208 rth->rt_gateway = daddr; 2209 rth->rt_spec_dst= spec_dst; 2210 rth->dst.input= ip_local_deliver; 2211 rth->rt_flags = flags|RTCF_LOCAL; 2212 if (res.type == RTN_UNREACHABLE) { 2213 rth->dst.input= ip_error; 2214 rth->dst.error= -err; 2215 rth->rt_flags &= ~RTCF_LOCAL; 2216 } 2217 rth->rt_type = res.type; 2218 hash = rt_hash(daddr, saddr, fl4.flowi4_iif, rt_genid(net)); 2219 rth = rt_intern_hash(hash, rth, skb, fl4.flowi4_iif); 2220 err = 0; 2221 if (IS_ERR(rth)) 2222 err = PTR_ERR(rth); 2223 goto out; 2224 2225 no_route: 2226 RT_CACHE_STAT_INC(in_no_route); 2227 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE); 2228 res.type = RTN_UNREACHABLE; 2229 if (err == -ESRCH) 2230 err = -ENETUNREACH; 2231 goto local_input; 2232 2233 /* 2234 * Do not cache martian addresses: they should be logged (RFC1812) 2235 */ 2236 martian_destination: 2237 RT_CACHE_STAT_INC(in_martian_dst); 2238 #ifdef CONFIG_IP_ROUTE_VERBOSE 2239 if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit()) 2240 printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n", 2241 &daddr, &saddr, dev->name); 2242 #endif 2243 2244 e_hostunreach: 2245 err = -EHOSTUNREACH; 2246 goto out; 2247 2248 e_inval: 2249 err = -EINVAL; 2250 goto out; 2251 2252 e_nobufs: 2253 err = -ENOBUFS; 2254 goto out; 2255 2256 martian_source: 2257 err = -EINVAL; 2258 martian_source_keep_err: 2259 ip_handle_martian_source(dev, in_dev, skb, daddr, saddr); 2260 goto out; 2261 } 2262 2263 int ip_route_input_common(struct sk_buff *skb, __be32 daddr, __be32 saddr, 2264 u8 tos, struct net_device *dev, bool noref) 2265 { 2266 struct rtable * rth; 2267 unsigned hash; 2268 int iif = dev->ifindex; 2269 struct net *net; 2270 int res; 2271 2272 net = dev_net(dev); 2273 2274 rcu_read_lock(); 2275 2276 if (!rt_caching(net)) 2277 goto skip_cache; 2278 2279 tos &= IPTOS_RT_MASK; 2280 hash = rt_hash(daddr, saddr, iif, rt_genid(net)); 2281 2282 for (rth = rcu_dereference(rt_hash_table[hash].chain); rth; 2283 rth = rcu_dereference(rth->dst.rt_next)) { 2284 if ((((__force u32)rth->rt_key_dst ^ (__force u32)daddr) | 2285 ((__force u32)rth->rt_key_src ^ (__force u32)saddr) | 2286 (rth->rt_iif ^ iif) | 2287 rth->rt_oif | 2288 (rth->rt_tos ^ tos)) == 0 && 2289 rth->rt_mark == skb->mark && 2290 net_eq(dev_net(rth->dst.dev), net) && 2291 !rt_is_expired(rth)) { 2292 if (noref) { 2293 dst_use_noref(&rth->dst, jiffies); 2294 skb_dst_set_noref(skb, &rth->dst); 2295 } else { 2296 dst_use(&rth->dst, jiffies); 2297 skb_dst_set(skb, &rth->dst); 2298 } 2299 RT_CACHE_STAT_INC(in_hit); 2300 rcu_read_unlock(); 2301 return 0; 2302 } 2303 RT_CACHE_STAT_INC(in_hlist_search); 2304 } 2305 2306 skip_cache: 2307 /* Multicast recognition logic is moved from route cache to here. 2308 The problem was that too many Ethernet cards have broken/missing 2309 hardware multicast filters :-( As result the host on multicasting 2310 network acquires a lot of useless route cache entries, sort of 2311 SDR messages from all the world. Now we try to get rid of them. 2312 Really, provided software IP multicast filter is organized 2313 reasonably (at least, hashed), it does not result in a slowdown 2314 comparing with route cache reject entries. 2315 Note, that multicast routers are not affected, because 2316 route cache entry is created eventually. 2317 */ 2318 if (ipv4_is_multicast(daddr)) { 2319 struct in_device *in_dev = __in_dev_get_rcu(dev); 2320 2321 if (in_dev) { 2322 int our = ip_check_mc_rcu(in_dev, daddr, saddr, 2323 ip_hdr(skb)->protocol); 2324 if (our 2325 #ifdef CONFIG_IP_MROUTE 2326 || 2327 (!ipv4_is_local_multicast(daddr) && 2328 IN_DEV_MFORWARD(in_dev)) 2329 #endif 2330 ) { 2331 int res = ip_route_input_mc(skb, daddr, saddr, 2332 tos, dev, our); 2333 rcu_read_unlock(); 2334 return res; 2335 } 2336 } 2337 rcu_read_unlock(); 2338 return -EINVAL; 2339 } 2340 res = ip_route_input_slow(skb, daddr, saddr, tos, dev); 2341 rcu_read_unlock(); 2342 return res; 2343 } 2344 EXPORT_SYMBOL(ip_route_input_common); 2345 2346 /* called with rcu_read_lock() */ 2347 static struct rtable *__mkroute_output(const struct fib_result *res, 2348 const struct flowi4 *fl4, 2349 const struct flowi4 *oldflp4, 2350 struct net_device *dev_out, 2351 unsigned int flags) 2352 { 2353 struct fib_info *fi = res->fi; 2354 u32 tos = RT_FL_TOS(oldflp4); 2355 struct in_device *in_dev; 2356 u16 type = res->type; 2357 struct rtable *rth; 2358 2359 if (ipv4_is_loopback(fl4->saddr) && !(dev_out->flags & IFF_LOOPBACK)) 2360 return ERR_PTR(-EINVAL); 2361 2362 if (ipv4_is_lbcast(fl4->daddr)) 2363 type = RTN_BROADCAST; 2364 else if (ipv4_is_multicast(fl4->daddr)) 2365 type = RTN_MULTICAST; 2366 else if (ipv4_is_zeronet(fl4->daddr)) 2367 return ERR_PTR(-EINVAL); 2368 2369 if (dev_out->flags & IFF_LOOPBACK) 2370 flags |= RTCF_LOCAL; 2371 2372 in_dev = __in_dev_get_rcu(dev_out); 2373 if (!in_dev) 2374 return ERR_PTR(-EINVAL); 2375 2376 if (type == RTN_BROADCAST) { 2377 flags |= RTCF_BROADCAST | RTCF_LOCAL; 2378 fi = NULL; 2379 } else if (type == RTN_MULTICAST) { 2380 flags |= RTCF_MULTICAST | RTCF_LOCAL; 2381 if (!ip_check_mc_rcu(in_dev, oldflp4->daddr, oldflp4->saddr, 2382 oldflp4->flowi4_proto)) 2383 flags &= ~RTCF_LOCAL; 2384 /* If multicast route do not exist use 2385 * default one, but do not gateway in this case. 2386 * Yes, it is hack. 2387 */ 2388 if (fi && res->prefixlen < 4) 2389 fi = NULL; 2390 } 2391 2392 rth = rt_dst_alloc(IN_DEV_CONF_GET(in_dev, NOPOLICY), 2393 IN_DEV_CONF_GET(in_dev, NOXFRM)); 2394 if (!rth) 2395 return ERR_PTR(-ENOBUFS); 2396 2397 rth->rt_key_dst = oldflp4->daddr; 2398 rth->rt_tos = tos; 2399 rth->rt_key_src = oldflp4->saddr; 2400 rth->rt_oif = oldflp4->flowi4_oif; 2401 rth->rt_mark = oldflp4->flowi4_mark; 2402 rth->rt_dst = fl4->daddr; 2403 rth->rt_src = fl4->saddr; 2404 rth->rt_iif = 0; 2405 /* get references to the devices that are to be hold by the routing 2406 cache entry */ 2407 rth->dst.dev = dev_out; 2408 dev_hold(dev_out); 2409 rth->rt_gateway = fl4->daddr; 2410 rth->rt_spec_dst= fl4->saddr; 2411 2412 rth->dst.output=ip_output; 2413 rth->rt_genid = rt_genid(dev_net(dev_out)); 2414 2415 RT_CACHE_STAT_INC(out_slow_tot); 2416 2417 if (flags & RTCF_LOCAL) { 2418 rth->dst.input = ip_local_deliver; 2419 rth->rt_spec_dst = fl4->daddr; 2420 } 2421 if (flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 2422 rth->rt_spec_dst = fl4->saddr; 2423 if (flags & RTCF_LOCAL && 2424 !(dev_out->flags & IFF_LOOPBACK)) { 2425 rth->dst.output = ip_mc_output; 2426 RT_CACHE_STAT_INC(out_slow_mc); 2427 } 2428 #ifdef CONFIG_IP_MROUTE 2429 if (type == RTN_MULTICAST) { 2430 if (IN_DEV_MFORWARD(in_dev) && 2431 !ipv4_is_local_multicast(oldflp4->daddr)) { 2432 rth->dst.input = ip_mr_input; 2433 rth->dst.output = ip_mc_output; 2434 } 2435 } 2436 #endif 2437 } 2438 2439 rt_set_nexthop(rth, oldflp4, res, fi, type, 0); 2440 2441 rth->rt_flags = flags; 2442 return rth; 2443 } 2444 2445 /* 2446 * Major route resolver routine. 2447 * called with rcu_read_lock(); 2448 */ 2449 2450 static struct rtable *ip_route_output_slow(struct net *net, 2451 const struct flowi4 *oldflp4) 2452 { 2453 u32 tos = RT_FL_TOS(oldflp4); 2454 struct flowi4 fl4; 2455 struct fib_result res; 2456 unsigned int flags = 0; 2457 struct net_device *dev_out = NULL; 2458 struct rtable *rth; 2459 2460 res.fi = NULL; 2461 #ifdef CONFIG_IP_MULTIPLE_TABLES 2462 res.r = NULL; 2463 #endif 2464 2465 fl4.flowi4_oif = oldflp4->flowi4_oif; 2466 fl4.flowi4_iif = net->loopback_dev->ifindex; 2467 fl4.flowi4_mark = oldflp4->flowi4_mark; 2468 fl4.daddr = oldflp4->daddr; 2469 fl4.saddr = oldflp4->saddr; 2470 fl4.flowi4_tos = tos & IPTOS_RT_MASK; 2471 fl4.flowi4_scope = ((tos & RTO_ONLINK) ? 2472 RT_SCOPE_LINK : RT_SCOPE_UNIVERSE); 2473 2474 rcu_read_lock(); 2475 if (oldflp4->saddr) { 2476 rth = ERR_PTR(-EINVAL); 2477 if (ipv4_is_multicast(oldflp4->saddr) || 2478 ipv4_is_lbcast(oldflp4->saddr) || 2479 ipv4_is_zeronet(oldflp4->saddr)) 2480 goto out; 2481 2482 /* I removed check for oif == dev_out->oif here. 2483 It was wrong for two reasons: 2484 1. ip_dev_find(net, saddr) can return wrong iface, if saddr 2485 is assigned to multiple interfaces. 2486 2. Moreover, we are allowed to send packets with saddr 2487 of another iface. --ANK 2488 */ 2489 2490 if (oldflp4->flowi4_oif == 0 && 2491 (ipv4_is_multicast(oldflp4->daddr) || 2492 ipv4_is_lbcast(oldflp4->daddr))) { 2493 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2494 dev_out = __ip_dev_find(net, oldflp4->saddr, false); 2495 if (dev_out == NULL) 2496 goto out; 2497 2498 /* Special hack: user can direct multicasts 2499 and limited broadcast via necessary interface 2500 without fiddling with IP_MULTICAST_IF or IP_PKTINFO. 2501 This hack is not just for fun, it allows 2502 vic,vat and friends to work. 2503 They bind socket to loopback, set ttl to zero 2504 and expect that it will work. 2505 From the viewpoint of routing cache they are broken, 2506 because we are not allowed to build multicast path 2507 with loopback source addr (look, routing cache 2508 cannot know, that ttl is zero, so that packet 2509 will not leave this host and route is valid). 2510 Luckily, this hack is good workaround. 2511 */ 2512 2513 fl4.flowi4_oif = dev_out->ifindex; 2514 goto make_route; 2515 } 2516 2517 if (!(oldflp4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { 2518 /* It is equivalent to inet_addr_type(saddr) == RTN_LOCAL */ 2519 if (!__ip_dev_find(net, oldflp4->saddr, false)) 2520 goto out; 2521 } 2522 } 2523 2524 2525 if (oldflp4->flowi4_oif) { 2526 dev_out = dev_get_by_index_rcu(net, oldflp4->flowi4_oif); 2527 rth = ERR_PTR(-ENODEV); 2528 if (dev_out == NULL) 2529 goto out; 2530 2531 /* RACE: Check return value of inet_select_addr instead. */ 2532 if (!(dev_out->flags & IFF_UP) || !__in_dev_get_rcu(dev_out)) { 2533 rth = ERR_PTR(-ENETUNREACH); 2534 goto out; 2535 } 2536 if (ipv4_is_local_multicast(oldflp4->daddr) || 2537 ipv4_is_lbcast(oldflp4->daddr)) { 2538 if (!fl4.saddr) 2539 fl4.saddr = inet_select_addr(dev_out, 0, 2540 RT_SCOPE_LINK); 2541 goto make_route; 2542 } 2543 if (!fl4.saddr) { 2544 if (ipv4_is_multicast(oldflp4->daddr)) 2545 fl4.saddr = inet_select_addr(dev_out, 0, 2546 fl4.flowi4_scope); 2547 else if (!oldflp4->daddr) 2548 fl4.saddr = inet_select_addr(dev_out, 0, 2549 RT_SCOPE_HOST); 2550 } 2551 } 2552 2553 if (!fl4.daddr) { 2554 fl4.daddr = fl4.saddr; 2555 if (!fl4.daddr) 2556 fl4.daddr = fl4.saddr = htonl(INADDR_LOOPBACK); 2557 dev_out = net->loopback_dev; 2558 fl4.flowi4_oif = net->loopback_dev->ifindex; 2559 res.type = RTN_LOCAL; 2560 flags |= RTCF_LOCAL; 2561 goto make_route; 2562 } 2563 2564 if (fib_lookup(net, &fl4, &res)) { 2565 res.fi = NULL; 2566 if (oldflp4->flowi4_oif) { 2567 /* Apparently, routing tables are wrong. Assume, 2568 that the destination is on link. 2569 2570 WHY? DW. 2571 Because we are allowed to send to iface 2572 even if it has NO routes and NO assigned 2573 addresses. When oif is specified, routing 2574 tables are looked up with only one purpose: 2575 to catch if destination is gatewayed, rather than 2576 direct. Moreover, if MSG_DONTROUTE is set, 2577 we send packet, ignoring both routing tables 2578 and ifaddr state. --ANK 2579 2580 2581 We could make it even if oif is unknown, 2582 likely IPv6, but we do not. 2583 */ 2584 2585 if (fl4.saddr == 0) 2586 fl4.saddr = inet_select_addr(dev_out, 0, 2587 RT_SCOPE_LINK); 2588 res.type = RTN_UNICAST; 2589 goto make_route; 2590 } 2591 rth = ERR_PTR(-ENETUNREACH); 2592 goto out; 2593 } 2594 2595 if (res.type == RTN_LOCAL) { 2596 if (!fl4.saddr) { 2597 if (res.fi->fib_prefsrc) 2598 fl4.saddr = res.fi->fib_prefsrc; 2599 else 2600 fl4.saddr = fl4.daddr; 2601 } 2602 dev_out = net->loopback_dev; 2603 fl4.flowi4_oif = dev_out->ifindex; 2604 res.fi = NULL; 2605 flags |= RTCF_LOCAL; 2606 goto make_route; 2607 } 2608 2609 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2610 if (res.fi->fib_nhs > 1 && fl4.flowi4_oif == 0) 2611 fib_select_multipath(&res); 2612 else 2613 #endif 2614 if (!res.prefixlen && res.type == RTN_UNICAST && !fl4.flowi4_oif) 2615 fib_select_default(&res); 2616 2617 if (!fl4.saddr) 2618 fl4.saddr = FIB_RES_PREFSRC(net, res); 2619 2620 dev_out = FIB_RES_DEV(res); 2621 fl4.flowi4_oif = dev_out->ifindex; 2622 2623 2624 make_route: 2625 rth = __mkroute_output(&res, &fl4, oldflp4, dev_out, flags); 2626 if (!IS_ERR(rth)) { 2627 unsigned int hash; 2628 2629 hash = rt_hash(oldflp4->daddr, oldflp4->saddr, oldflp4->flowi4_oif, 2630 rt_genid(dev_net(dev_out))); 2631 rth = rt_intern_hash(hash, rth, NULL, oldflp4->flowi4_oif); 2632 } 2633 2634 out: 2635 rcu_read_unlock(); 2636 return rth; 2637 } 2638 2639 struct rtable *__ip_route_output_key(struct net *net, const struct flowi4 *flp4) 2640 { 2641 struct rtable *rth; 2642 unsigned int hash; 2643 2644 if (!rt_caching(net)) 2645 goto slow_output; 2646 2647 hash = rt_hash(flp4->daddr, flp4->saddr, flp4->flowi4_oif, rt_genid(net)); 2648 2649 rcu_read_lock_bh(); 2650 for (rth = rcu_dereference_bh(rt_hash_table[hash].chain); rth; 2651 rth = rcu_dereference_bh(rth->dst.rt_next)) { 2652 if (rth->rt_key_dst == flp4->daddr && 2653 rth->rt_key_src == flp4->saddr && 2654 rt_is_output_route(rth) && 2655 rth->rt_oif == flp4->flowi4_oif && 2656 rth->rt_mark == flp4->flowi4_mark && 2657 !((rth->rt_tos ^ flp4->flowi4_tos) & 2658 (IPTOS_RT_MASK | RTO_ONLINK)) && 2659 net_eq(dev_net(rth->dst.dev), net) && 2660 !rt_is_expired(rth)) { 2661 dst_use(&rth->dst, jiffies); 2662 RT_CACHE_STAT_INC(out_hit); 2663 rcu_read_unlock_bh(); 2664 return rth; 2665 } 2666 RT_CACHE_STAT_INC(out_hlist_search); 2667 } 2668 rcu_read_unlock_bh(); 2669 2670 slow_output: 2671 return ip_route_output_slow(net, flp4); 2672 } 2673 EXPORT_SYMBOL_GPL(__ip_route_output_key); 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_default_mtu(const struct dst_entry *dst) 2681 { 2682 return 0; 2683 } 2684 2685 static void ipv4_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu) 2686 { 2687 } 2688 2689 static struct dst_ops ipv4_dst_blackhole_ops = { 2690 .family = AF_INET, 2691 .protocol = cpu_to_be16(ETH_P_IP), 2692 .destroy = ipv4_dst_destroy, 2693 .check = ipv4_blackhole_dst_check, 2694 .default_mtu = ipv4_blackhole_default_mtu, 2695 .default_advmss = ipv4_default_advmss, 2696 .update_pmtu = ipv4_rt_blackhole_update_pmtu, 2697 }; 2698 2699 struct dst_entry *ipv4_blackhole_route(struct net *net, struct dst_entry *dst_orig) 2700 { 2701 struct rtable *rt = dst_alloc(&ipv4_dst_blackhole_ops, 1); 2702 struct rtable *ort = (struct rtable *) dst_orig; 2703 2704 if (rt) { 2705 struct dst_entry *new = &rt->dst; 2706 2707 new->__use = 1; 2708 new->input = dst_discard; 2709 new->output = dst_discard; 2710 dst_copy_metrics(new, &ort->dst); 2711 2712 new->dev = ort->dst.dev; 2713 if (new->dev) 2714 dev_hold(new->dev); 2715 2716 rt->rt_key_dst = ort->rt_key_dst; 2717 rt->rt_key_src = ort->rt_key_src; 2718 rt->rt_tos = ort->rt_tos; 2719 rt->rt_iif = ort->rt_iif; 2720 rt->rt_oif = ort->rt_oif; 2721 rt->rt_mark = ort->rt_mark; 2722 2723 rt->rt_genid = rt_genid(net); 2724 rt->rt_flags = ort->rt_flags; 2725 rt->rt_type = ort->rt_type; 2726 rt->rt_dst = ort->rt_dst; 2727 rt->rt_src = ort->rt_src; 2728 rt->rt_iif = ort->rt_iif; 2729 rt->rt_gateway = ort->rt_gateway; 2730 rt->rt_spec_dst = ort->rt_spec_dst; 2731 rt->peer = ort->peer; 2732 if (rt->peer) 2733 atomic_inc(&rt->peer->refcnt); 2734 rt->fi = ort->fi; 2735 if (rt->fi) 2736 atomic_inc(&rt->fi->fib_clntref); 2737 2738 dst_free(new); 2739 } 2740 2741 dst_release(dst_orig); 2742 2743 return rt ? &rt->dst : ERR_PTR(-ENOMEM); 2744 } 2745 2746 struct rtable *ip_route_output_flow(struct net *net, struct flowi4 *flp4, 2747 struct sock *sk) 2748 { 2749 struct rtable *rt = __ip_route_output_key(net, flp4); 2750 2751 if (IS_ERR(rt)) 2752 return rt; 2753 2754 if (flp4->flowi4_proto) { 2755 if (!flp4->saddr) 2756 flp4->saddr = rt->rt_src; 2757 if (!flp4->daddr) 2758 flp4->daddr = rt->rt_dst; 2759 rt = (struct rtable *) xfrm_lookup(net, &rt->dst, 2760 flowi4_to_flowi(flp4), 2761 sk, 0); 2762 } 2763 2764 return rt; 2765 } 2766 EXPORT_SYMBOL_GPL(ip_route_output_flow); 2767 2768 static int rt_fill_info(struct net *net, 2769 struct sk_buff *skb, u32 pid, u32 seq, int event, 2770 int nowait, unsigned int flags) 2771 { 2772 struct rtable *rt = skb_rtable(skb); 2773 struct rtmsg *r; 2774 struct nlmsghdr *nlh; 2775 long expires; 2776 u32 id = 0, ts = 0, tsage = 0, error; 2777 2778 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*r), flags); 2779 if (nlh == NULL) 2780 return -EMSGSIZE; 2781 2782 r = nlmsg_data(nlh); 2783 r->rtm_family = AF_INET; 2784 r->rtm_dst_len = 32; 2785 r->rtm_src_len = 0; 2786 r->rtm_tos = rt->rt_tos; 2787 r->rtm_table = RT_TABLE_MAIN; 2788 NLA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN); 2789 r->rtm_type = rt->rt_type; 2790 r->rtm_scope = RT_SCOPE_UNIVERSE; 2791 r->rtm_protocol = RTPROT_UNSPEC; 2792 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED; 2793 if (rt->rt_flags & RTCF_NOTIFY) 2794 r->rtm_flags |= RTM_F_NOTIFY; 2795 2796 NLA_PUT_BE32(skb, RTA_DST, rt->rt_dst); 2797 2798 if (rt->rt_key_src) { 2799 r->rtm_src_len = 32; 2800 NLA_PUT_BE32(skb, RTA_SRC, rt->rt_key_src); 2801 } 2802 if (rt->dst.dev) 2803 NLA_PUT_U32(skb, RTA_OIF, rt->dst.dev->ifindex); 2804 #ifdef CONFIG_IP_ROUTE_CLASSID 2805 if (rt->dst.tclassid) 2806 NLA_PUT_U32(skb, RTA_FLOW, rt->dst.tclassid); 2807 #endif 2808 if (rt_is_input_route(rt)) 2809 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_spec_dst); 2810 else if (rt->rt_src != rt->rt_key_src) 2811 NLA_PUT_BE32(skb, RTA_PREFSRC, rt->rt_src); 2812 2813 if (rt->rt_dst != rt->rt_gateway) 2814 NLA_PUT_BE32(skb, RTA_GATEWAY, rt->rt_gateway); 2815 2816 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0) 2817 goto nla_put_failure; 2818 2819 if (rt->rt_mark) 2820 NLA_PUT_BE32(skb, RTA_MARK, rt->rt_mark); 2821 2822 error = rt->dst.error; 2823 expires = (rt->peer && rt->peer->pmtu_expires) ? 2824 rt->peer->pmtu_expires - jiffies : 0; 2825 if (rt->peer) { 2826 inet_peer_refcheck(rt->peer); 2827 id = atomic_read(&rt->peer->ip_id_count) & 0xffff; 2828 if (rt->peer->tcp_ts_stamp) { 2829 ts = rt->peer->tcp_ts; 2830 tsage = get_seconds() - rt->peer->tcp_ts_stamp; 2831 } 2832 } 2833 2834 if (rt_is_input_route(rt)) { 2835 #ifdef CONFIG_IP_MROUTE 2836 __be32 dst = rt->rt_dst; 2837 2838 if (ipv4_is_multicast(dst) && !ipv4_is_local_multicast(dst) && 2839 IPV4_DEVCONF_ALL(net, MC_FORWARDING)) { 2840 int err = ipmr_get_route(net, skb, r, nowait); 2841 if (err <= 0) { 2842 if (!nowait) { 2843 if (err == 0) 2844 return 0; 2845 goto nla_put_failure; 2846 } else { 2847 if (err == -EMSGSIZE) 2848 goto nla_put_failure; 2849 error = err; 2850 } 2851 } 2852 } else 2853 #endif 2854 NLA_PUT_U32(skb, RTA_IIF, rt->rt_iif); 2855 } 2856 2857 if (rtnl_put_cacheinfo(skb, &rt->dst, id, ts, tsage, 2858 expires, error) < 0) 2859 goto nla_put_failure; 2860 2861 return nlmsg_end(skb, nlh); 2862 2863 nla_put_failure: 2864 nlmsg_cancel(skb, nlh); 2865 return -EMSGSIZE; 2866 } 2867 2868 static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg) 2869 { 2870 struct net *net = sock_net(in_skb->sk); 2871 struct rtmsg *rtm; 2872 struct nlattr *tb[RTA_MAX+1]; 2873 struct rtable *rt = NULL; 2874 __be32 dst = 0; 2875 __be32 src = 0; 2876 u32 iif; 2877 int err; 2878 int mark; 2879 struct sk_buff *skb; 2880 2881 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv4_policy); 2882 if (err < 0) 2883 goto errout; 2884 2885 rtm = nlmsg_data(nlh); 2886 2887 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 2888 if (skb == NULL) { 2889 err = -ENOBUFS; 2890 goto errout; 2891 } 2892 2893 /* Reserve room for dummy headers, this skb can pass 2894 through good chunk of routing engine. 2895 */ 2896 skb_reset_mac_header(skb); 2897 skb_reset_network_header(skb); 2898 2899 /* Bugfix: need to give ip_route_input enough of an IP header to not gag. */ 2900 ip_hdr(skb)->protocol = IPPROTO_ICMP; 2901 skb_reserve(skb, MAX_HEADER + sizeof(struct iphdr)); 2902 2903 src = tb[RTA_SRC] ? nla_get_be32(tb[RTA_SRC]) : 0; 2904 dst = tb[RTA_DST] ? nla_get_be32(tb[RTA_DST]) : 0; 2905 iif = tb[RTA_IIF] ? nla_get_u32(tb[RTA_IIF]) : 0; 2906 mark = tb[RTA_MARK] ? nla_get_u32(tb[RTA_MARK]) : 0; 2907 2908 if (iif) { 2909 struct net_device *dev; 2910 2911 dev = __dev_get_by_index(net, iif); 2912 if (dev == NULL) { 2913 err = -ENODEV; 2914 goto errout_free; 2915 } 2916 2917 skb->protocol = htons(ETH_P_IP); 2918 skb->dev = dev; 2919 skb->mark = mark; 2920 local_bh_disable(); 2921 err = ip_route_input(skb, dst, src, rtm->rtm_tos, dev); 2922 local_bh_enable(); 2923 2924 rt = skb_rtable(skb); 2925 if (err == 0 && rt->dst.error) 2926 err = -rt->dst.error; 2927 } else { 2928 struct flowi4 fl4 = { 2929 .daddr = dst, 2930 .saddr = src, 2931 .flowi4_tos = rtm->rtm_tos, 2932 .flowi4_oif = tb[RTA_OIF] ? nla_get_u32(tb[RTA_OIF]) : 0, 2933 .flowi4_mark = mark, 2934 }; 2935 rt = ip_route_output_key(net, &fl4); 2936 2937 err = 0; 2938 if (IS_ERR(rt)) 2939 err = PTR_ERR(rt); 2940 } 2941 2942 if (err) 2943 goto errout_free; 2944 2945 skb_dst_set(skb, &rt->dst); 2946 if (rtm->rtm_flags & RTM_F_NOTIFY) 2947 rt->rt_flags |= RTCF_NOTIFY; 2948 2949 err = rt_fill_info(net, skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, 2950 RTM_NEWROUTE, 0, 0); 2951 if (err <= 0) 2952 goto errout_free; 2953 2954 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid); 2955 errout: 2956 return err; 2957 2958 errout_free: 2959 kfree_skb(skb); 2960 goto errout; 2961 } 2962 2963 int ip_rt_dump(struct sk_buff *skb, struct netlink_callback *cb) 2964 { 2965 struct rtable *rt; 2966 int h, s_h; 2967 int idx, s_idx; 2968 struct net *net; 2969 2970 net = sock_net(skb->sk); 2971 2972 s_h = cb->args[0]; 2973 if (s_h < 0) 2974 s_h = 0; 2975 s_idx = idx = cb->args[1]; 2976 for (h = s_h; h <= rt_hash_mask; h++, s_idx = 0) { 2977 if (!rt_hash_table[h].chain) 2978 continue; 2979 rcu_read_lock_bh(); 2980 for (rt = rcu_dereference_bh(rt_hash_table[h].chain), idx = 0; rt; 2981 rt = rcu_dereference_bh(rt->dst.rt_next), idx++) { 2982 if (!net_eq(dev_net(rt->dst.dev), net) || idx < s_idx) 2983 continue; 2984 if (rt_is_expired(rt)) 2985 continue; 2986 skb_dst_set_noref(skb, &rt->dst); 2987 if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid, 2988 cb->nlh->nlmsg_seq, RTM_NEWROUTE, 2989 1, NLM_F_MULTI) <= 0) { 2990 skb_dst_drop(skb); 2991 rcu_read_unlock_bh(); 2992 goto done; 2993 } 2994 skb_dst_drop(skb); 2995 } 2996 rcu_read_unlock_bh(); 2997 } 2998 2999 done: 3000 cb->args[0] = h; 3001 cb->args[1] = idx; 3002 return skb->len; 3003 } 3004 3005 void ip_rt_multicast_event(struct in_device *in_dev) 3006 { 3007 rt_cache_flush(dev_net(in_dev->dev), 0); 3008 } 3009 3010 #ifdef CONFIG_SYSCTL 3011 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write, 3012 void __user *buffer, 3013 size_t *lenp, loff_t *ppos) 3014 { 3015 if (write) { 3016 int flush_delay; 3017 ctl_table ctl; 3018 struct net *net; 3019 3020 memcpy(&ctl, __ctl, sizeof(ctl)); 3021 ctl.data = &flush_delay; 3022 proc_dointvec(&ctl, write, buffer, lenp, ppos); 3023 3024 net = (struct net *)__ctl->extra1; 3025 rt_cache_flush(net, flush_delay); 3026 return 0; 3027 } 3028 3029 return -EINVAL; 3030 } 3031 3032 static ctl_table ipv4_route_table[] = { 3033 { 3034 .procname = "gc_thresh", 3035 .data = &ipv4_dst_ops.gc_thresh, 3036 .maxlen = sizeof(int), 3037 .mode = 0644, 3038 .proc_handler = proc_dointvec, 3039 }, 3040 { 3041 .procname = "max_size", 3042 .data = &ip_rt_max_size, 3043 .maxlen = sizeof(int), 3044 .mode = 0644, 3045 .proc_handler = proc_dointvec, 3046 }, 3047 { 3048 /* Deprecated. Use gc_min_interval_ms */ 3049 3050 .procname = "gc_min_interval", 3051 .data = &ip_rt_gc_min_interval, 3052 .maxlen = sizeof(int), 3053 .mode = 0644, 3054 .proc_handler = proc_dointvec_jiffies, 3055 }, 3056 { 3057 .procname = "gc_min_interval_ms", 3058 .data = &ip_rt_gc_min_interval, 3059 .maxlen = sizeof(int), 3060 .mode = 0644, 3061 .proc_handler = proc_dointvec_ms_jiffies, 3062 }, 3063 { 3064 .procname = "gc_timeout", 3065 .data = &ip_rt_gc_timeout, 3066 .maxlen = sizeof(int), 3067 .mode = 0644, 3068 .proc_handler = proc_dointvec_jiffies, 3069 }, 3070 { 3071 .procname = "gc_interval", 3072 .data = &ip_rt_gc_interval, 3073 .maxlen = sizeof(int), 3074 .mode = 0644, 3075 .proc_handler = proc_dointvec_jiffies, 3076 }, 3077 { 3078 .procname = "redirect_load", 3079 .data = &ip_rt_redirect_load, 3080 .maxlen = sizeof(int), 3081 .mode = 0644, 3082 .proc_handler = proc_dointvec, 3083 }, 3084 { 3085 .procname = "redirect_number", 3086 .data = &ip_rt_redirect_number, 3087 .maxlen = sizeof(int), 3088 .mode = 0644, 3089 .proc_handler = proc_dointvec, 3090 }, 3091 { 3092 .procname = "redirect_silence", 3093 .data = &ip_rt_redirect_silence, 3094 .maxlen = sizeof(int), 3095 .mode = 0644, 3096 .proc_handler = proc_dointvec, 3097 }, 3098 { 3099 .procname = "error_cost", 3100 .data = &ip_rt_error_cost, 3101 .maxlen = sizeof(int), 3102 .mode = 0644, 3103 .proc_handler = proc_dointvec, 3104 }, 3105 { 3106 .procname = "error_burst", 3107 .data = &ip_rt_error_burst, 3108 .maxlen = sizeof(int), 3109 .mode = 0644, 3110 .proc_handler = proc_dointvec, 3111 }, 3112 { 3113 .procname = "gc_elasticity", 3114 .data = &ip_rt_gc_elasticity, 3115 .maxlen = sizeof(int), 3116 .mode = 0644, 3117 .proc_handler = proc_dointvec, 3118 }, 3119 { 3120 .procname = "mtu_expires", 3121 .data = &ip_rt_mtu_expires, 3122 .maxlen = sizeof(int), 3123 .mode = 0644, 3124 .proc_handler = proc_dointvec_jiffies, 3125 }, 3126 { 3127 .procname = "min_pmtu", 3128 .data = &ip_rt_min_pmtu, 3129 .maxlen = sizeof(int), 3130 .mode = 0644, 3131 .proc_handler = proc_dointvec, 3132 }, 3133 { 3134 .procname = "min_adv_mss", 3135 .data = &ip_rt_min_advmss, 3136 .maxlen = sizeof(int), 3137 .mode = 0644, 3138 .proc_handler = proc_dointvec, 3139 }, 3140 { } 3141 }; 3142 3143 static struct ctl_table empty[1]; 3144 3145 static struct ctl_table ipv4_skeleton[] = 3146 { 3147 { .procname = "route", 3148 .mode = 0555, .child = ipv4_route_table}, 3149 { .procname = "neigh", 3150 .mode = 0555, .child = empty}, 3151 { } 3152 }; 3153 3154 static __net_initdata struct ctl_path ipv4_path[] = { 3155 { .procname = "net", }, 3156 { .procname = "ipv4", }, 3157 { }, 3158 }; 3159 3160 static struct ctl_table ipv4_route_flush_table[] = { 3161 { 3162 .procname = "flush", 3163 .maxlen = sizeof(int), 3164 .mode = 0200, 3165 .proc_handler = ipv4_sysctl_rtcache_flush, 3166 }, 3167 { }, 3168 }; 3169 3170 static __net_initdata struct ctl_path ipv4_route_path[] = { 3171 { .procname = "net", }, 3172 { .procname = "ipv4", }, 3173 { .procname = "route", }, 3174 { }, 3175 }; 3176 3177 static __net_init int sysctl_route_net_init(struct net *net) 3178 { 3179 struct ctl_table *tbl; 3180 3181 tbl = ipv4_route_flush_table; 3182 if (!net_eq(net, &init_net)) { 3183 tbl = kmemdup(tbl, sizeof(ipv4_route_flush_table), GFP_KERNEL); 3184 if (tbl == NULL) 3185 goto err_dup; 3186 } 3187 tbl[0].extra1 = net; 3188 3189 net->ipv4.route_hdr = 3190 register_net_sysctl_table(net, ipv4_route_path, tbl); 3191 if (net->ipv4.route_hdr == NULL) 3192 goto err_reg; 3193 return 0; 3194 3195 err_reg: 3196 if (tbl != ipv4_route_flush_table) 3197 kfree(tbl); 3198 err_dup: 3199 return -ENOMEM; 3200 } 3201 3202 static __net_exit void sysctl_route_net_exit(struct net *net) 3203 { 3204 struct ctl_table *tbl; 3205 3206 tbl = net->ipv4.route_hdr->ctl_table_arg; 3207 unregister_net_sysctl_table(net->ipv4.route_hdr); 3208 BUG_ON(tbl == ipv4_route_flush_table); 3209 kfree(tbl); 3210 } 3211 3212 static __net_initdata struct pernet_operations sysctl_route_ops = { 3213 .init = sysctl_route_net_init, 3214 .exit = sysctl_route_net_exit, 3215 }; 3216 #endif 3217 3218 static __net_init int rt_genid_init(struct net *net) 3219 { 3220 get_random_bytes(&net->ipv4.rt_genid, 3221 sizeof(net->ipv4.rt_genid)); 3222 get_random_bytes(&net->ipv4.dev_addr_genid, 3223 sizeof(net->ipv4.dev_addr_genid)); 3224 return 0; 3225 } 3226 3227 static __net_initdata struct pernet_operations rt_genid_ops = { 3228 .init = rt_genid_init, 3229 }; 3230 3231 3232 #ifdef CONFIG_IP_ROUTE_CLASSID 3233 struct ip_rt_acct __percpu *ip_rt_acct __read_mostly; 3234 #endif /* CONFIG_IP_ROUTE_CLASSID */ 3235 3236 static __initdata unsigned long rhash_entries; 3237 static int __init set_rhash_entries(char *str) 3238 { 3239 if (!str) 3240 return 0; 3241 rhash_entries = simple_strtoul(str, &str, 0); 3242 return 1; 3243 } 3244 __setup("rhash_entries=", set_rhash_entries); 3245 3246 int __init ip_rt_init(void) 3247 { 3248 int rc = 0; 3249 3250 #ifdef CONFIG_IP_ROUTE_CLASSID 3251 ip_rt_acct = __alloc_percpu(256 * sizeof(struct ip_rt_acct), __alignof__(struct ip_rt_acct)); 3252 if (!ip_rt_acct) 3253 panic("IP: failed to allocate ip_rt_acct\n"); 3254 #endif 3255 3256 ipv4_dst_ops.kmem_cachep = 3257 kmem_cache_create("ip_dst_cache", sizeof(struct rtable), 0, 3258 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 3259 3260 ipv4_dst_blackhole_ops.kmem_cachep = ipv4_dst_ops.kmem_cachep; 3261 3262 if (dst_entries_init(&ipv4_dst_ops) < 0) 3263 panic("IP: failed to allocate ipv4_dst_ops counter\n"); 3264 3265 if (dst_entries_init(&ipv4_dst_blackhole_ops) < 0) 3266 panic("IP: failed to allocate ipv4_dst_blackhole_ops counter\n"); 3267 3268 rt_hash_table = (struct rt_hash_bucket *) 3269 alloc_large_system_hash("IP route cache", 3270 sizeof(struct rt_hash_bucket), 3271 rhash_entries, 3272 (totalram_pages >= 128 * 1024) ? 3273 15 : 17, 3274 0, 3275 &rt_hash_log, 3276 &rt_hash_mask, 3277 rhash_entries ? 0 : 512 * 1024); 3278 memset(rt_hash_table, 0, (rt_hash_mask + 1) * sizeof(struct rt_hash_bucket)); 3279 rt_hash_lock_init(); 3280 3281 ipv4_dst_ops.gc_thresh = (rt_hash_mask + 1); 3282 ip_rt_max_size = (rt_hash_mask + 1) * 16; 3283 3284 devinet_init(); 3285 ip_fib_init(); 3286 3287 if (ip_rt_proc_init()) 3288 printk(KERN_ERR "Unable to create route proc files\n"); 3289 #ifdef CONFIG_XFRM 3290 xfrm_init(); 3291 xfrm4_init(ip_rt_max_size); 3292 #endif 3293 rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL); 3294 3295 #ifdef CONFIG_SYSCTL 3296 register_pernet_subsys(&sysctl_route_ops); 3297 #endif 3298 register_pernet_subsys(&rt_genid_ops); 3299 return rc; 3300 } 3301 3302 #ifdef CONFIG_SYSCTL 3303 /* 3304 * We really need to sanitize the damn ipv4 init order, then all 3305 * this nonsense will go away. 3306 */ 3307 void __init ip_static_sysctl_init(void) 3308 { 3309 register_sysctl_paths(ipv4_path, ipv4_skeleton); 3310 } 3311 #endif 3312