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 * IPv4 Forwarding Information Base: semantics. 7 * 8 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 9 * 10 * This program is free software; you can redistribute it and/or 11 * modify it under the terms of the GNU General Public License 12 * as published by the Free Software Foundation; either version 13 * 2 of the License, or (at your option) any later version. 14 */ 15 16#include <asm/uaccess.h> 17#include <linux/bitops.h> 18#include <linux/types.h> 19#include <linux/kernel.h> 20#include <linux/jiffies.h> 21#include <linux/mm.h> 22#include <linux/string.h> 23#include <linux/socket.h> 24#include <linux/sockios.h> 25#include <linux/errno.h> 26#include <linux/in.h> 27#include <linux/inet.h> 28#include <linux/inetdevice.h> 29#include <linux/netdevice.h> 30#include <linux/if_arp.h> 31#include <linux/proc_fs.h> 32#include <linux/skbuff.h> 33#include <linux/init.h> 34#include <linux/slab.h> 35 36#include <net/arp.h> 37#include <net/ip.h> 38#include <net/protocol.h> 39#include <net/route.h> 40#include <net/tcp.h> 41#include <net/sock.h> 42#include <net/ip_fib.h> 43#include <net/netlink.h> 44#include <net/nexthop.h> 45 46#include "fib_lookup.h" 47 48static DEFINE_SPINLOCK(fib_info_lock); 49static struct hlist_head *fib_info_hash; 50static struct hlist_head *fib_info_laddrhash; 51static unsigned int fib_info_hash_size; 52static unsigned int fib_info_cnt; 53 54#define DEVINDEX_HASHBITS 8 55#define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS) 56static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE]; 57 58#ifdef CONFIG_IP_ROUTE_MULTIPATH 59 60static DEFINE_SPINLOCK(fib_multipath_lock); 61 62#define for_nexthops(fi) { \ 63 int nhsel; const struct fib_nh *nh; \ 64 for (nhsel = 0, nh = (fi)->fib_nh; \ 65 nhsel < (fi)->fib_nhs; \ 66 nh++, nhsel++) 67 68#define change_nexthops(fi) { \ 69 int nhsel; struct fib_nh *nexthop_nh; \ 70 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \ 71 nhsel < (fi)->fib_nhs; \ 72 nexthop_nh++, nhsel++) 73 74#else /* CONFIG_IP_ROUTE_MULTIPATH */ 75 76/* Hope, that gcc will optimize it to get rid of dummy loop */ 77 78#define for_nexthops(fi) { \ 79 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \ 80 for (nhsel = 0; nhsel < 1; nhsel++) 81 82#define change_nexthops(fi) { \ 83 int nhsel; \ 84 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \ 85 for (nhsel = 0; nhsel < 1; nhsel++) 86 87#endif /* CONFIG_IP_ROUTE_MULTIPATH */ 88 89#define endfor_nexthops(fi) } 90 91 92const struct fib_prop fib_props[RTN_MAX + 1] = { 93 [RTN_UNSPEC] = { 94 .error = 0, 95 .scope = RT_SCOPE_NOWHERE, 96 }, 97 [RTN_UNICAST] = { 98 .error = 0, 99 .scope = RT_SCOPE_UNIVERSE, 100 }, 101 [RTN_LOCAL] = { 102 .error = 0, 103 .scope = RT_SCOPE_HOST, 104 }, 105 [RTN_BROADCAST] = { 106 .error = 0, 107 .scope = RT_SCOPE_LINK, 108 }, 109 [RTN_ANYCAST] = { 110 .error = 0, 111 .scope = RT_SCOPE_LINK, 112 }, 113 [RTN_MULTICAST] = { 114 .error = 0, 115 .scope = RT_SCOPE_UNIVERSE, 116 }, 117 [RTN_BLACKHOLE] = { 118 .error = -EINVAL, 119 .scope = RT_SCOPE_UNIVERSE, 120 }, 121 [RTN_UNREACHABLE] = { 122 .error = -EHOSTUNREACH, 123 .scope = RT_SCOPE_UNIVERSE, 124 }, 125 [RTN_PROHIBIT] = { 126 .error = -EACCES, 127 .scope = RT_SCOPE_UNIVERSE, 128 }, 129 [RTN_THROW] = { 130 .error = -EAGAIN, 131 .scope = RT_SCOPE_UNIVERSE, 132 }, 133 [RTN_NAT] = { 134 .error = -EINVAL, 135 .scope = RT_SCOPE_NOWHERE, 136 }, 137 [RTN_XRESOLVE] = { 138 .error = -EINVAL, 139 .scope = RT_SCOPE_NOWHERE, 140 }, 141}; 142 143static void rt_fibinfo_free(struct rtable __rcu **rtp) 144{ 145 struct rtable *rt = rcu_dereference_protected(*rtp, 1); 146 147 if (!rt) 148 return; 149 150 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL); 151 * because we waited an RCU grace period before calling 152 * free_fib_info_rcu() 153 */ 154 155 dst_free(&rt->dst); 156} 157 158static void free_nh_exceptions(struct fib_nh *nh) 159{ 160 struct fnhe_hash_bucket *hash; 161 int i; 162 163 hash = rcu_dereference_protected(nh->nh_exceptions, 1); 164 if (!hash) 165 return; 166 for (i = 0; i < FNHE_HASH_SIZE; i++) { 167 struct fib_nh_exception *fnhe; 168 169 fnhe = rcu_dereference_protected(hash[i].chain, 1); 170 while (fnhe) { 171 struct fib_nh_exception *next; 172 173 next = rcu_dereference_protected(fnhe->fnhe_next, 1); 174 175 rt_fibinfo_free(&fnhe->fnhe_rth_input); 176 rt_fibinfo_free(&fnhe->fnhe_rth_output); 177 178 kfree(fnhe); 179 180 fnhe = next; 181 } 182 } 183 kfree(hash); 184} 185 186static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp) 187{ 188 int cpu; 189 190 if (!rtp) 191 return; 192 193 for_each_possible_cpu(cpu) { 194 struct rtable *rt; 195 196 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1); 197 if (rt) 198 dst_free(&rt->dst); 199 } 200 free_percpu(rtp); 201} 202 203/* Release a nexthop info record */ 204static void free_fib_info_rcu(struct rcu_head *head) 205{ 206 struct fib_info *fi = container_of(head, struct fib_info, rcu); 207 208 change_nexthops(fi) { 209 if (nexthop_nh->nh_dev) 210 dev_put(nexthop_nh->nh_dev); 211 free_nh_exceptions(nexthop_nh); 212 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output); 213 rt_fibinfo_free(&nexthop_nh->nh_rth_input); 214 } endfor_nexthops(fi); 215 216 release_net(fi->fib_net); 217 if (fi->fib_metrics != (u32 *) dst_default_metrics) 218 kfree(fi->fib_metrics); 219 kfree(fi); 220} 221 222void free_fib_info(struct fib_info *fi) 223{ 224 if (fi->fib_dead == 0) { 225 pr_warn("Freeing alive fib_info %p\n", fi); 226 return; 227 } 228 fib_info_cnt--; 229#ifdef CONFIG_IP_ROUTE_CLASSID 230 change_nexthops(fi) { 231 if (nexthop_nh->nh_tclassid) 232 fi->fib_net->ipv4.fib_num_tclassid_users--; 233 } endfor_nexthops(fi); 234#endif 235 call_rcu(&fi->rcu, free_fib_info_rcu); 236} 237 238void fib_release_info(struct fib_info *fi) 239{ 240 spin_lock_bh(&fib_info_lock); 241 if (fi && --fi->fib_treeref == 0) { 242 hlist_del(&fi->fib_hash); 243 if (fi->fib_prefsrc) 244 hlist_del(&fi->fib_lhash); 245 change_nexthops(fi) { 246 if (!nexthop_nh->nh_dev) 247 continue; 248 hlist_del(&nexthop_nh->nh_hash); 249 } endfor_nexthops(fi) 250 fi->fib_dead = 1; 251 fib_info_put(fi); 252 } 253 spin_unlock_bh(&fib_info_lock); 254} 255 256static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi) 257{ 258 const struct fib_nh *onh = ofi->fib_nh; 259 260 for_nexthops(fi) { 261 if (nh->nh_oif != onh->nh_oif || 262 nh->nh_gw != onh->nh_gw || 263 nh->nh_scope != onh->nh_scope || 264#ifdef CONFIG_IP_ROUTE_MULTIPATH 265 nh->nh_weight != onh->nh_weight || 266#endif 267#ifdef CONFIG_IP_ROUTE_CLASSID 268 nh->nh_tclassid != onh->nh_tclassid || 269#endif 270 ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD)) 271 return -1; 272 onh++; 273 } endfor_nexthops(fi); 274 return 0; 275} 276 277static inline unsigned int fib_devindex_hashfn(unsigned int val) 278{ 279 unsigned int mask = DEVINDEX_HASHSIZE - 1; 280 281 return (val ^ 282 (val >> DEVINDEX_HASHBITS) ^ 283 (val >> (DEVINDEX_HASHBITS * 2))) & mask; 284} 285 286static inline unsigned int fib_info_hashfn(const struct fib_info *fi) 287{ 288 unsigned int mask = (fib_info_hash_size - 1); 289 unsigned int val = fi->fib_nhs; 290 291 val ^= (fi->fib_protocol << 8) | fi->fib_scope; 292 val ^= (__force u32)fi->fib_prefsrc; 293 val ^= fi->fib_priority; 294 for_nexthops(fi) { 295 val ^= fib_devindex_hashfn(nh->nh_oif); 296 } endfor_nexthops(fi) 297 298 return (val ^ (val >> 7) ^ (val >> 12)) & mask; 299} 300 301static struct fib_info *fib_find_info(const struct fib_info *nfi) 302{ 303 struct hlist_head *head; 304 struct fib_info *fi; 305 unsigned int hash; 306 307 hash = fib_info_hashfn(nfi); 308 head = &fib_info_hash[hash]; 309 310 hlist_for_each_entry(fi, head, fib_hash) { 311 if (!net_eq(fi->fib_net, nfi->fib_net)) 312 continue; 313 if (fi->fib_nhs != nfi->fib_nhs) 314 continue; 315 if (nfi->fib_protocol == fi->fib_protocol && 316 nfi->fib_scope == fi->fib_scope && 317 nfi->fib_prefsrc == fi->fib_prefsrc && 318 nfi->fib_priority == fi->fib_priority && 319 nfi->fib_type == fi->fib_type && 320 memcmp(nfi->fib_metrics, fi->fib_metrics, 321 sizeof(u32) * RTAX_MAX) == 0 && 322 ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 && 323 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0)) 324 return fi; 325 } 326 327 return NULL; 328} 329 330/* Check, that the gateway is already configured. 331 * Used only by redirect accept routine. 332 */ 333int ip_fib_check_default(__be32 gw, struct net_device *dev) 334{ 335 struct hlist_head *head; 336 struct fib_nh *nh; 337 unsigned int hash; 338 339 spin_lock(&fib_info_lock); 340 341 hash = fib_devindex_hashfn(dev->ifindex); 342 head = &fib_info_devhash[hash]; 343 hlist_for_each_entry(nh, head, nh_hash) { 344 if (nh->nh_dev == dev && 345 nh->nh_gw == gw && 346 !(nh->nh_flags & RTNH_F_DEAD)) { 347 spin_unlock(&fib_info_lock); 348 return 0; 349 } 350 } 351 352 spin_unlock(&fib_info_lock); 353 354 return -1; 355} 356 357static inline size_t fib_nlmsg_size(struct fib_info *fi) 358{ 359 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg)) 360 + nla_total_size(4) /* RTA_TABLE */ 361 + nla_total_size(4) /* RTA_DST */ 362 + nla_total_size(4) /* RTA_PRIORITY */ 363 + nla_total_size(4); /* RTA_PREFSRC */ 364 365 /* space for nested metrics */ 366 payload += nla_total_size((RTAX_MAX * nla_total_size(4))); 367 368 if (fi->fib_nhs) { 369 /* Also handles the special case fib_nhs == 1 */ 370 371 /* each nexthop is packed in an attribute */ 372 size_t nhsize = nla_total_size(sizeof(struct rtnexthop)); 373 374 /* may contain flow and gateway attribute */ 375 nhsize += 2 * nla_total_size(4); 376 377 /* all nexthops are packed in a nested attribute */ 378 payload += nla_total_size(fi->fib_nhs * nhsize); 379 } 380 381 return payload; 382} 383 384void rtmsg_fib(int event, __be32 key, struct fib_alias *fa, 385 int dst_len, u32 tb_id, const struct nl_info *info, 386 unsigned int nlm_flags) 387{ 388 struct sk_buff *skb; 389 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0; 390 int err = -ENOBUFS; 391 392 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL); 393 if (skb == NULL) 394 goto errout; 395 396 err = fib_dump_info(skb, info->portid, seq, event, tb_id, 397 fa->fa_type, key, dst_len, 398 fa->fa_tos, fa->fa_info, nlm_flags); 399 if (err < 0) { 400 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */ 401 WARN_ON(err == -EMSGSIZE); 402 kfree_skb(skb); 403 goto errout; 404 } 405 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE, 406 info->nlh, GFP_KERNEL); 407 return; 408errout: 409 if (err < 0) 410 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err); 411} 412 413/* Return the first fib alias matching TOS with 414 * priority less than or equal to PRIO. 415 */ 416struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio) 417{ 418 if (fah) { 419 struct fib_alias *fa; 420 list_for_each_entry(fa, fah, fa_list) { 421 if (fa->fa_tos > tos) 422 continue; 423 if (fa->fa_info->fib_priority >= prio || 424 fa->fa_tos < tos) 425 return fa; 426 } 427 } 428 return NULL; 429} 430 431static int fib_detect_death(struct fib_info *fi, int order, 432 struct fib_info **last_resort, int *last_idx, 433 int dflt) 434{ 435 struct neighbour *n; 436 int state = NUD_NONE; 437 438 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev); 439 if (n) { 440 state = n->nud_state; 441 neigh_release(n); 442 } 443 if (state == NUD_REACHABLE) 444 return 0; 445 if ((state & NUD_VALID) && order != dflt) 446 return 0; 447 if ((state & NUD_VALID) || 448 (*last_idx < 0 && order > dflt)) { 449 *last_resort = fi; 450 *last_idx = order; 451 } 452 return 1; 453} 454 455#ifdef CONFIG_IP_ROUTE_MULTIPATH 456 457static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining) 458{ 459 int nhs = 0; 460 461 while (rtnh_ok(rtnh, remaining)) { 462 nhs++; 463 rtnh = rtnh_next(rtnh, &remaining); 464 } 465 466 /* leftover implies invalid nexthop configuration, discard it */ 467 return remaining > 0 ? 0 : nhs; 468} 469 470static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh, 471 int remaining, struct fib_config *cfg) 472{ 473 change_nexthops(fi) { 474 int attrlen; 475 476 if (!rtnh_ok(rtnh, remaining)) 477 return -EINVAL; 478 479 nexthop_nh->nh_flags = 480 (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags; 481 nexthop_nh->nh_oif = rtnh->rtnh_ifindex; 482 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1; 483 484 attrlen = rtnh_attrlen(rtnh); 485 if (attrlen > 0) { 486 struct nlattr *nla, *attrs = rtnh_attrs(rtnh); 487 488 nla = nla_find(attrs, attrlen, RTA_GATEWAY); 489 nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0; 490#ifdef CONFIG_IP_ROUTE_CLASSID 491 nla = nla_find(attrs, attrlen, RTA_FLOW); 492 nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0; 493 if (nexthop_nh->nh_tclassid) 494 fi->fib_net->ipv4.fib_num_tclassid_users++; 495#endif 496 } 497 498 rtnh = rtnh_next(rtnh, &remaining); 499 } endfor_nexthops(fi); 500 501 return 0; 502} 503 504#endif 505 506int fib_nh_match(struct fib_config *cfg, struct fib_info *fi) 507{ 508#ifdef CONFIG_IP_ROUTE_MULTIPATH 509 struct rtnexthop *rtnh; 510 int remaining; 511#endif 512 513 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority) 514 return 1; 515 516 if (cfg->fc_oif || cfg->fc_gw) { 517 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) && 518 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw)) 519 return 0; 520 return 1; 521 } 522 523#ifdef CONFIG_IP_ROUTE_MULTIPATH 524 if (cfg->fc_mp == NULL) 525 return 0; 526 527 rtnh = cfg->fc_mp; 528 remaining = cfg->fc_mp_len; 529 530 for_nexthops(fi) { 531 int attrlen; 532 533 if (!rtnh_ok(rtnh, remaining)) 534 return -EINVAL; 535 536 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif) 537 return 1; 538 539 attrlen = rtnh_attrlen(rtnh); 540 if (attrlen > 0) { 541 struct nlattr *nla, *attrs = rtnh_attrs(rtnh); 542 543 nla = nla_find(attrs, attrlen, RTA_GATEWAY); 544 if (nla && nla_get_be32(nla) != nh->nh_gw) 545 return 1; 546#ifdef CONFIG_IP_ROUTE_CLASSID 547 nla = nla_find(attrs, attrlen, RTA_FLOW); 548 if (nla && nla_get_u32(nla) != nh->nh_tclassid) 549 return 1; 550#endif 551 } 552 553 rtnh = rtnh_next(rtnh, &remaining); 554 } endfor_nexthops(fi); 555#endif 556 return 0; 557} 558 559 560/* 561 * Picture 562 * ------- 563 * 564 * Semantics of nexthop is very messy by historical reasons. 565 * We have to take into account, that: 566 * a) gateway can be actually local interface address, 567 * so that gatewayed route is direct. 568 * b) gateway must be on-link address, possibly 569 * described not by an ifaddr, but also by a direct route. 570 * c) If both gateway and interface are specified, they should not 571 * contradict. 572 * d) If we use tunnel routes, gateway could be not on-link. 573 * 574 * Attempt to reconcile all of these (alas, self-contradictory) conditions 575 * results in pretty ugly and hairy code with obscure logic. 576 * 577 * I chose to generalized it instead, so that the size 578 * of code does not increase practically, but it becomes 579 * much more general. 580 * Every prefix is assigned a "scope" value: "host" is local address, 581 * "link" is direct route, 582 * [ ... "site" ... "interior" ... ] 583 * and "universe" is true gateway route with global meaning. 584 * 585 * Every prefix refers to a set of "nexthop"s (gw, oif), 586 * where gw must have narrower scope. This recursion stops 587 * when gw has LOCAL scope or if "nexthop" is declared ONLINK, 588 * which means that gw is forced to be on link. 589 * 590 * Code is still hairy, but now it is apparently logically 591 * consistent and very flexible. F.e. as by-product it allows 592 * to co-exists in peace independent exterior and interior 593 * routing processes. 594 * 595 * Normally it looks as following. 596 * 597 * {universe prefix} -> (gw, oif) [scope link] 598 * | 599 * |-> {link prefix} -> (gw, oif) [scope local] 600 * | 601 * |-> {local prefix} (terminal node) 602 */ 603static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi, 604 struct fib_nh *nh) 605{ 606 int err; 607 struct net *net; 608 struct net_device *dev; 609 610 net = cfg->fc_nlinfo.nl_net; 611 if (nh->nh_gw) { 612 struct fib_result res; 613 614 if (nh->nh_flags & RTNH_F_ONLINK) { 615 616 if (cfg->fc_scope >= RT_SCOPE_LINK) 617 return -EINVAL; 618 if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST) 619 return -EINVAL; 620 dev = __dev_get_by_index(net, nh->nh_oif); 621 if (!dev) 622 return -ENODEV; 623 if (!(dev->flags & IFF_UP)) 624 return -ENETDOWN; 625 nh->nh_dev = dev; 626 dev_hold(dev); 627 nh->nh_scope = RT_SCOPE_LINK; 628 return 0; 629 } 630 rcu_read_lock(); 631 { 632 struct flowi4 fl4 = { 633 .daddr = nh->nh_gw, 634 .flowi4_scope = cfg->fc_scope + 1, 635 .flowi4_oif = nh->nh_oif, 636 .flowi4_iif = LOOPBACK_IFINDEX, 637 }; 638 639 /* It is not necessary, but requires a bit of thinking */ 640 if (fl4.flowi4_scope < RT_SCOPE_LINK) 641 fl4.flowi4_scope = RT_SCOPE_LINK; 642 err = fib_lookup(net, &fl4, &res); 643 if (err) { 644 rcu_read_unlock(); 645 return err; 646 } 647 } 648 err = -EINVAL; 649 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) 650 goto out; 651 nh->nh_scope = res.scope; 652 nh->nh_oif = FIB_RES_OIF(res); 653 nh->nh_dev = dev = FIB_RES_DEV(res); 654 if (!dev) 655 goto out; 656 dev_hold(dev); 657 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN; 658 } else { 659 struct in_device *in_dev; 660 661 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) 662 return -EINVAL; 663 664 rcu_read_lock(); 665 err = -ENODEV; 666 in_dev = inetdev_by_index(net, nh->nh_oif); 667 if (in_dev == NULL) 668 goto out; 669 err = -ENETDOWN; 670 if (!(in_dev->dev->flags & IFF_UP)) 671 goto out; 672 nh->nh_dev = in_dev->dev; 673 dev_hold(nh->nh_dev); 674 nh->nh_scope = RT_SCOPE_HOST; 675 err = 0; 676 } 677out: 678 rcu_read_unlock(); 679 return err; 680} 681 682static inline unsigned int fib_laddr_hashfn(__be32 val) 683{ 684 unsigned int mask = (fib_info_hash_size - 1); 685 686 return ((__force u32)val ^ 687 ((__force u32)val >> 7) ^ 688 ((__force u32)val >> 14)) & mask; 689} 690 691static struct hlist_head *fib_info_hash_alloc(int bytes) 692{ 693 if (bytes <= PAGE_SIZE) 694 return kzalloc(bytes, GFP_KERNEL); 695 else 696 return (struct hlist_head *) 697 __get_free_pages(GFP_KERNEL | __GFP_ZERO, 698 get_order(bytes)); 699} 700 701static void fib_info_hash_free(struct hlist_head *hash, int bytes) 702{ 703 if (!hash) 704 return; 705 706 if (bytes <= PAGE_SIZE) 707 kfree(hash); 708 else 709 free_pages((unsigned long) hash, get_order(bytes)); 710} 711 712static void fib_info_hash_move(struct hlist_head *new_info_hash, 713 struct hlist_head *new_laddrhash, 714 unsigned int new_size) 715{ 716 struct hlist_head *old_info_hash, *old_laddrhash; 717 unsigned int old_size = fib_info_hash_size; 718 unsigned int i, bytes; 719 720 spin_lock_bh(&fib_info_lock); 721 old_info_hash = fib_info_hash; 722 old_laddrhash = fib_info_laddrhash; 723 fib_info_hash_size = new_size; 724 725 for (i = 0; i < old_size; i++) { 726 struct hlist_head *head = &fib_info_hash[i]; 727 struct hlist_node *n; 728 struct fib_info *fi; 729 730 hlist_for_each_entry_safe(fi, n, head, fib_hash) { 731 struct hlist_head *dest; 732 unsigned int new_hash; 733 734 hlist_del(&fi->fib_hash); 735 736 new_hash = fib_info_hashfn(fi); 737 dest = &new_info_hash[new_hash]; 738 hlist_add_head(&fi->fib_hash, dest); 739 } 740 } 741 fib_info_hash = new_info_hash; 742 743 for (i = 0; i < old_size; i++) { 744 struct hlist_head *lhead = &fib_info_laddrhash[i]; 745 struct hlist_node *n; 746 struct fib_info *fi; 747 748 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) { 749 struct hlist_head *ldest; 750 unsigned int new_hash; 751 752 hlist_del(&fi->fib_lhash); 753 754 new_hash = fib_laddr_hashfn(fi->fib_prefsrc); 755 ldest = &new_laddrhash[new_hash]; 756 hlist_add_head(&fi->fib_lhash, ldest); 757 } 758 } 759 fib_info_laddrhash = new_laddrhash; 760 761 spin_unlock_bh(&fib_info_lock); 762 763 bytes = old_size * sizeof(struct hlist_head *); 764 fib_info_hash_free(old_info_hash, bytes); 765 fib_info_hash_free(old_laddrhash, bytes); 766} 767 768__be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh) 769{ 770 nh->nh_saddr = inet_select_addr(nh->nh_dev, 771 nh->nh_gw, 772 nh->nh_parent->fib_scope); 773 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid); 774 775 return nh->nh_saddr; 776} 777 778struct fib_info *fib_create_info(struct fib_config *cfg) 779{ 780 int err; 781 struct fib_info *fi = NULL; 782 struct fib_info *ofi; 783 int nhs = 1; 784 struct net *net = cfg->fc_nlinfo.nl_net; 785 786 if (cfg->fc_type > RTN_MAX) 787 goto err_inval; 788 789 /* Fast check to catch the most weird cases */ 790 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) 791 goto err_inval; 792 793#ifdef CONFIG_IP_ROUTE_MULTIPATH 794 if (cfg->fc_mp) { 795 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len); 796 if (nhs == 0) 797 goto err_inval; 798 } 799#endif 800 801 err = -ENOBUFS; 802 if (fib_info_cnt >= fib_info_hash_size) { 803 unsigned int new_size = fib_info_hash_size << 1; 804 struct hlist_head *new_info_hash; 805 struct hlist_head *new_laddrhash; 806 unsigned int bytes; 807 808 if (!new_size) 809 new_size = 16; 810 bytes = new_size * sizeof(struct hlist_head *); 811 new_info_hash = fib_info_hash_alloc(bytes); 812 new_laddrhash = fib_info_hash_alloc(bytes); 813 if (!new_info_hash || !new_laddrhash) { 814 fib_info_hash_free(new_info_hash, bytes); 815 fib_info_hash_free(new_laddrhash, bytes); 816 } else 817 fib_info_hash_move(new_info_hash, new_laddrhash, new_size); 818 819 if (!fib_info_hash_size) 820 goto failure; 821 } 822 823 fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL); 824 if (fi == NULL) 825 goto failure; 826 fib_info_cnt++; 827 if (cfg->fc_mx) { 828 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL); 829 if (!fi->fib_metrics) 830 goto failure; 831 } else 832 fi->fib_metrics = (u32 *) dst_default_metrics; 833 834 fi->fib_net = hold_net(net); 835 fi->fib_protocol = cfg->fc_protocol; 836 fi->fib_scope = cfg->fc_scope; 837 fi->fib_flags = cfg->fc_flags; 838 fi->fib_priority = cfg->fc_priority; 839 fi->fib_prefsrc = cfg->fc_prefsrc; 840 fi->fib_type = cfg->fc_type; 841 842 fi->fib_nhs = nhs; 843 change_nexthops(fi) { 844 nexthop_nh->nh_parent = fi; 845 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *); 846 if (!nexthop_nh->nh_pcpu_rth_output) 847 goto failure; 848 } endfor_nexthops(fi) 849 850 if (cfg->fc_mx) { 851 struct nlattr *nla; 852 int remaining; 853 854 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) { 855 int type = nla_type(nla); 856 857 if (type) { 858 u32 val; 859 860 if (type > RTAX_MAX) 861 goto err_inval; 862 val = nla_get_u32(nla); 863 if (type == RTAX_ADVMSS && val > 65535 - 40) 864 val = 65535 - 40; 865 if (type == RTAX_MTU && val > 65535 - 15) 866 val = 65535 - 15; 867 fi->fib_metrics[type - 1] = val; 868 } 869 } 870 } 871 872 if (cfg->fc_mp) { 873#ifdef CONFIG_IP_ROUTE_MULTIPATH 874 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg); 875 if (err != 0) 876 goto failure; 877 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif) 878 goto err_inval; 879 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw) 880 goto err_inval; 881#ifdef CONFIG_IP_ROUTE_CLASSID 882 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) 883 goto err_inval; 884#endif 885#else 886 goto err_inval; 887#endif 888 } else { 889 struct fib_nh *nh = fi->fib_nh; 890 891 nh->nh_oif = cfg->fc_oif; 892 nh->nh_gw = cfg->fc_gw; 893 nh->nh_flags = cfg->fc_flags; 894#ifdef CONFIG_IP_ROUTE_CLASSID 895 nh->nh_tclassid = cfg->fc_flow; 896 if (nh->nh_tclassid) 897 fi->fib_net->ipv4.fib_num_tclassid_users++; 898#endif 899#ifdef CONFIG_IP_ROUTE_MULTIPATH 900 nh->nh_weight = 1; 901#endif 902 } 903 904 if (fib_props[cfg->fc_type].error) { 905 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp) 906 goto err_inval; 907 goto link_it; 908 } else { 909 switch (cfg->fc_type) { 910 case RTN_UNICAST: 911 case RTN_LOCAL: 912 case RTN_BROADCAST: 913 case RTN_ANYCAST: 914 case RTN_MULTICAST: 915 break; 916 default: 917 goto err_inval; 918 } 919 } 920 921 if (cfg->fc_scope > RT_SCOPE_HOST) 922 goto err_inval; 923 924 if (cfg->fc_scope == RT_SCOPE_HOST) { 925 struct fib_nh *nh = fi->fib_nh; 926 927 /* Local address is added. */ 928 if (nhs != 1 || nh->nh_gw) 929 goto err_inval; 930 nh->nh_scope = RT_SCOPE_NOWHERE; 931 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif); 932 err = -ENODEV; 933 if (nh->nh_dev == NULL) 934 goto failure; 935 } else { 936 change_nexthops(fi) { 937 err = fib_check_nh(cfg, fi, nexthop_nh); 938 if (err != 0) 939 goto failure; 940 } endfor_nexthops(fi) 941 } 942 943 if (fi->fib_prefsrc) { 944 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst || 945 fi->fib_prefsrc != cfg->fc_dst) 946 if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL) 947 goto err_inval; 948 } 949 950 change_nexthops(fi) { 951 fib_info_update_nh_saddr(net, nexthop_nh); 952 } endfor_nexthops(fi) 953 954link_it: 955 ofi = fib_find_info(fi); 956 if (ofi) { 957 fi->fib_dead = 1; 958 free_fib_info(fi); 959 ofi->fib_treeref++; 960 return ofi; 961 } 962 963 fi->fib_treeref++; 964 atomic_inc(&fi->fib_clntref); 965 spin_lock_bh(&fib_info_lock); 966 hlist_add_head(&fi->fib_hash, 967 &fib_info_hash[fib_info_hashfn(fi)]); 968 if (fi->fib_prefsrc) { 969 struct hlist_head *head; 970 971 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)]; 972 hlist_add_head(&fi->fib_lhash, head); 973 } 974 change_nexthops(fi) { 975 struct hlist_head *head; 976 unsigned int hash; 977 978 if (!nexthop_nh->nh_dev) 979 continue; 980 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex); 981 head = &fib_info_devhash[hash]; 982 hlist_add_head(&nexthop_nh->nh_hash, head); 983 } endfor_nexthops(fi) 984 spin_unlock_bh(&fib_info_lock); 985 return fi; 986 987err_inval: 988 err = -EINVAL; 989 990failure: 991 if (fi) { 992 fi->fib_dead = 1; 993 free_fib_info(fi); 994 } 995 996 return ERR_PTR(err); 997} 998 999int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event, 1000 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos, 1001 struct fib_info *fi, unsigned int flags) 1002{ 1003 struct nlmsghdr *nlh; 1004 struct rtmsg *rtm; 1005 1006 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags); 1007 if (nlh == NULL) 1008 return -EMSGSIZE; 1009 1010 rtm = nlmsg_data(nlh); 1011 rtm->rtm_family = AF_INET; 1012 rtm->rtm_dst_len = dst_len; 1013 rtm->rtm_src_len = 0; 1014 rtm->rtm_tos = tos; 1015 if (tb_id < 256) 1016 rtm->rtm_table = tb_id; 1017 else 1018 rtm->rtm_table = RT_TABLE_COMPAT; 1019 if (nla_put_u32(skb, RTA_TABLE, tb_id)) 1020 goto nla_put_failure; 1021 rtm->rtm_type = type; 1022 rtm->rtm_flags = fi->fib_flags; 1023 rtm->rtm_scope = fi->fib_scope; 1024 rtm->rtm_protocol = fi->fib_protocol; 1025 1026 if (rtm->rtm_dst_len && 1027 nla_put_be32(skb, RTA_DST, dst)) 1028 goto nla_put_failure; 1029 if (fi->fib_priority && 1030 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority)) 1031 goto nla_put_failure; 1032 if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0) 1033 goto nla_put_failure; 1034 1035 if (fi->fib_prefsrc && 1036 nla_put_be32(skb, RTA_PREFSRC, fi->fib_prefsrc)) 1037 goto nla_put_failure; 1038 if (fi->fib_nhs == 1) { 1039 if (fi->fib_nh->nh_gw && 1040 nla_put_be32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw)) 1041 goto nla_put_failure; 1042 if (fi->fib_nh->nh_oif && 1043 nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif)) 1044 goto nla_put_failure; 1045#ifdef CONFIG_IP_ROUTE_CLASSID 1046 if (fi->fib_nh[0].nh_tclassid && 1047 nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid)) 1048 goto nla_put_failure; 1049#endif 1050 } 1051#ifdef CONFIG_IP_ROUTE_MULTIPATH 1052 if (fi->fib_nhs > 1) { 1053 struct rtnexthop *rtnh; 1054 struct nlattr *mp; 1055 1056 mp = nla_nest_start(skb, RTA_MULTIPATH); 1057 if (mp == NULL) 1058 goto nla_put_failure; 1059 1060 for_nexthops(fi) { 1061 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh)); 1062 if (rtnh == NULL) 1063 goto nla_put_failure; 1064 1065 rtnh->rtnh_flags = nh->nh_flags & 0xFF; 1066 rtnh->rtnh_hops = nh->nh_weight - 1; 1067 rtnh->rtnh_ifindex = nh->nh_oif; 1068 1069 if (nh->nh_gw && 1070 nla_put_be32(skb, RTA_GATEWAY, nh->nh_gw)) 1071 goto nla_put_failure; 1072#ifdef CONFIG_IP_ROUTE_CLASSID 1073 if (nh->nh_tclassid && 1074 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid)) 1075 goto nla_put_failure; 1076#endif 1077 /* length of rtnetlink header + attributes */ 1078 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh; 1079 } endfor_nexthops(fi); 1080 1081 nla_nest_end(skb, mp); 1082 } 1083#endif 1084 return nlmsg_end(skb, nlh); 1085 1086nla_put_failure: 1087 nlmsg_cancel(skb, nlh); 1088 return -EMSGSIZE; 1089} 1090 1091/* 1092 * Update FIB if: 1093 * - local address disappeared -> we must delete all the entries 1094 * referring to it. 1095 * - device went down -> we must shutdown all nexthops going via it. 1096 */ 1097int fib_sync_down_addr(struct net *net, __be32 local) 1098{ 1099 int ret = 0; 1100 unsigned int hash = fib_laddr_hashfn(local); 1101 struct hlist_head *head = &fib_info_laddrhash[hash]; 1102 struct fib_info *fi; 1103 1104 if (fib_info_laddrhash == NULL || local == 0) 1105 return 0; 1106 1107 hlist_for_each_entry(fi, head, fib_lhash) { 1108 if (!net_eq(fi->fib_net, net)) 1109 continue; 1110 if (fi->fib_prefsrc == local) { 1111 fi->fib_flags |= RTNH_F_DEAD; 1112 ret++; 1113 } 1114 } 1115 return ret; 1116} 1117 1118int fib_sync_down_dev(struct net_device *dev, int force) 1119{ 1120 int ret = 0; 1121 int scope = RT_SCOPE_NOWHERE; 1122 struct fib_info *prev_fi = NULL; 1123 unsigned int hash = fib_devindex_hashfn(dev->ifindex); 1124 struct hlist_head *head = &fib_info_devhash[hash]; 1125 struct fib_nh *nh; 1126 1127 if (force) 1128 scope = -1; 1129 1130 hlist_for_each_entry(nh, head, nh_hash) { 1131 struct fib_info *fi = nh->nh_parent; 1132 int dead; 1133 1134 BUG_ON(!fi->fib_nhs); 1135 if (nh->nh_dev != dev || fi == prev_fi) 1136 continue; 1137 prev_fi = fi; 1138 dead = 0; 1139 change_nexthops(fi) { 1140 if (nexthop_nh->nh_flags & RTNH_F_DEAD) 1141 dead++; 1142 else if (nexthop_nh->nh_dev == dev && 1143 nexthop_nh->nh_scope != scope) { 1144 nexthop_nh->nh_flags |= RTNH_F_DEAD; 1145#ifdef CONFIG_IP_ROUTE_MULTIPATH 1146 spin_lock_bh(&fib_multipath_lock); 1147 fi->fib_power -= nexthop_nh->nh_power; 1148 nexthop_nh->nh_power = 0; 1149 spin_unlock_bh(&fib_multipath_lock); 1150#endif 1151 dead++; 1152 } 1153#ifdef CONFIG_IP_ROUTE_MULTIPATH 1154 if (force > 1 && nexthop_nh->nh_dev == dev) { 1155 dead = fi->fib_nhs; 1156 break; 1157 } 1158#endif 1159 } endfor_nexthops(fi) 1160 if (dead == fi->fib_nhs) { 1161 fi->fib_flags |= RTNH_F_DEAD; 1162 ret++; 1163 } 1164 } 1165 1166 return ret; 1167} 1168 1169/* Must be invoked inside of an RCU protected region. */ 1170void fib_select_default(struct fib_result *res) 1171{ 1172 struct fib_info *fi = NULL, *last_resort = NULL; 1173 struct list_head *fa_head = res->fa_head; 1174 struct fib_table *tb = res->table; 1175 int order = -1, last_idx = -1; 1176 struct fib_alias *fa; 1177 1178 list_for_each_entry_rcu(fa, fa_head, fa_list) { 1179 struct fib_info *next_fi = fa->fa_info; 1180 1181 if (next_fi->fib_scope != res->scope || 1182 fa->fa_type != RTN_UNICAST) 1183 continue; 1184 1185 if (next_fi->fib_priority > res->fi->fib_priority) 1186 break; 1187 if (!next_fi->fib_nh[0].nh_gw || 1188 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK) 1189 continue; 1190 1191 fib_alias_accessed(fa); 1192 1193 if (fi == NULL) { 1194 if (next_fi != res->fi) 1195 break; 1196 } else if (!fib_detect_death(fi, order, &last_resort, 1197 &last_idx, tb->tb_default)) { 1198 fib_result_assign(res, fi); 1199 tb->tb_default = order; 1200 goto out; 1201 } 1202 fi = next_fi; 1203 order++; 1204 } 1205 1206 if (order <= 0 || fi == NULL) { 1207 tb->tb_default = -1; 1208 goto out; 1209 } 1210 1211 if (!fib_detect_death(fi, order, &last_resort, &last_idx, 1212 tb->tb_default)) { 1213 fib_result_assign(res, fi); 1214 tb->tb_default = order; 1215 goto out; 1216 } 1217 1218 if (last_idx >= 0) 1219 fib_result_assign(res, last_resort); 1220 tb->tb_default = last_idx; 1221out: 1222 return; 1223} 1224 1225#ifdef CONFIG_IP_ROUTE_MULTIPATH 1226 1227/* 1228 * Dead device goes up. We wake up dead nexthops. 1229 * It takes sense only on multipath routes. 1230 */ 1231int fib_sync_up(struct net_device *dev) 1232{ 1233 struct fib_info *prev_fi; 1234 unsigned int hash; 1235 struct hlist_head *head; 1236 struct fib_nh *nh; 1237 int ret; 1238 1239 if (!(dev->flags & IFF_UP)) 1240 return 0; 1241 1242 prev_fi = NULL; 1243 hash = fib_devindex_hashfn(dev->ifindex); 1244 head = &fib_info_devhash[hash]; 1245 ret = 0; 1246 1247 hlist_for_each_entry(nh, head, nh_hash) { 1248 struct fib_info *fi = nh->nh_parent; 1249 int alive; 1250 1251 BUG_ON(!fi->fib_nhs); 1252 if (nh->nh_dev != dev || fi == prev_fi) 1253 continue; 1254 1255 prev_fi = fi; 1256 alive = 0; 1257 change_nexthops(fi) { 1258 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) { 1259 alive++; 1260 continue; 1261 } 1262 if (nexthop_nh->nh_dev == NULL || 1263 !(nexthop_nh->nh_dev->flags & IFF_UP)) 1264 continue; 1265 if (nexthop_nh->nh_dev != dev || 1266 !__in_dev_get_rtnl(dev)) 1267 continue; 1268 alive++; 1269 spin_lock_bh(&fib_multipath_lock); 1270 nexthop_nh->nh_power = 0; 1271 nexthop_nh->nh_flags &= ~RTNH_F_DEAD; 1272 spin_unlock_bh(&fib_multipath_lock); 1273 } endfor_nexthops(fi) 1274 1275 if (alive > 0) { 1276 fi->fib_flags &= ~RTNH_F_DEAD; 1277 ret++; 1278 } 1279 } 1280 1281 return ret; 1282} 1283 1284/* 1285 * The algorithm is suboptimal, but it provides really 1286 * fair weighted route distribution. 1287 */ 1288void fib_select_multipath(struct fib_result *res) 1289{ 1290 struct fib_info *fi = res->fi; 1291 int w; 1292 1293 spin_lock_bh(&fib_multipath_lock); 1294 if (fi->fib_power <= 0) { 1295 int power = 0; 1296 change_nexthops(fi) { 1297 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) { 1298 power += nexthop_nh->nh_weight; 1299 nexthop_nh->nh_power = nexthop_nh->nh_weight; 1300 } 1301 } endfor_nexthops(fi); 1302 fi->fib_power = power; 1303 if (power <= 0) { 1304 spin_unlock_bh(&fib_multipath_lock); 1305 /* Race condition: route has just become dead. */ 1306 res->nh_sel = 0; 1307 return; 1308 } 1309 } 1310 1311 1312 /* w should be random number [0..fi->fib_power-1], 1313 * it is pretty bad approximation. 1314 */ 1315 1316 w = jiffies % fi->fib_power; 1317 1318 change_nexthops(fi) { 1319 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) && 1320 nexthop_nh->nh_power) { 1321 w -= nexthop_nh->nh_power; 1322 if (w <= 0) { 1323 nexthop_nh->nh_power--; 1324 fi->fib_power--; 1325 res->nh_sel = nhsel; 1326 spin_unlock_bh(&fib_multipath_lock); 1327 return; 1328 } 1329 } 1330 } endfor_nexthops(fi); 1331 1332 /* Race condition: route has just become dead. */ 1333 res->nh_sel = 0; 1334 spin_unlock_bh(&fib_multipath_lock); 1335} 1336#endif 1337