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