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; 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 } 185 186 static 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 */ 204 static 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 } 221 222 void 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 } 237 238 void 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 } 255 256 static 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 } 276 277 static 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 } 285 286 static 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 } 300 301 static 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 */ 333 int 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 } 356 357 static 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 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */ 365 366 /* space for nested metrics */ 367 payload += nla_total_size((RTAX_MAX * nla_total_size(4))); 368 369 if (fi->fib_nhs) { 370 /* Also handles the special case fib_nhs == 1 */ 371 372 /* each nexthop is packed in an attribute */ 373 size_t nhsize = nla_total_size(sizeof(struct rtnexthop)); 374 375 /* may contain flow and gateway attribute */ 376 nhsize += 2 * nla_total_size(4); 377 378 /* all nexthops are packed in a nested attribute */ 379 payload += nla_total_size(fi->fib_nhs * nhsize); 380 } 381 382 return payload; 383 } 384 385 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa, 386 int dst_len, u32 tb_id, const struct nl_info *info, 387 unsigned int nlm_flags) 388 { 389 struct sk_buff *skb; 390 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0; 391 int err = -ENOBUFS; 392 393 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL); 394 if (skb == NULL) 395 goto errout; 396 397 err = fib_dump_info(skb, info->portid, seq, event, tb_id, 398 fa->fa_type, key, dst_len, 399 fa->fa_tos, fa->fa_info, nlm_flags); 400 if (err < 0) { 401 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */ 402 WARN_ON(err == -EMSGSIZE); 403 kfree_skb(skb); 404 goto errout; 405 } 406 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE, 407 info->nlh, GFP_KERNEL); 408 return; 409 errout: 410 if (err < 0) 411 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err); 412 } 413 414 static int fib_detect_death(struct fib_info *fi, int order, 415 struct fib_info **last_resort, int *last_idx, 416 int dflt) 417 { 418 struct neighbour *n; 419 int state = NUD_NONE; 420 421 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev); 422 if (n) { 423 state = n->nud_state; 424 neigh_release(n); 425 } 426 if (state == NUD_REACHABLE) 427 return 0; 428 if ((state & NUD_VALID) && order != dflt) 429 return 0; 430 if ((state & NUD_VALID) || 431 (*last_idx < 0 && order > dflt)) { 432 *last_resort = fi; 433 *last_idx = order; 434 } 435 return 1; 436 } 437 438 #ifdef CONFIG_IP_ROUTE_MULTIPATH 439 440 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining) 441 { 442 int nhs = 0; 443 444 while (rtnh_ok(rtnh, remaining)) { 445 nhs++; 446 rtnh = rtnh_next(rtnh, &remaining); 447 } 448 449 /* leftover implies invalid nexthop configuration, discard it */ 450 return remaining > 0 ? 0 : nhs; 451 } 452 453 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh, 454 int remaining, struct fib_config *cfg) 455 { 456 change_nexthops(fi) { 457 int attrlen; 458 459 if (!rtnh_ok(rtnh, remaining)) 460 return -EINVAL; 461 462 nexthop_nh->nh_flags = 463 (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags; 464 nexthop_nh->nh_oif = rtnh->rtnh_ifindex; 465 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1; 466 467 attrlen = rtnh_attrlen(rtnh); 468 if (attrlen > 0) { 469 struct nlattr *nla, *attrs = rtnh_attrs(rtnh); 470 471 nla = nla_find(attrs, attrlen, RTA_GATEWAY); 472 nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0; 473 #ifdef CONFIG_IP_ROUTE_CLASSID 474 nla = nla_find(attrs, attrlen, RTA_FLOW); 475 nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0; 476 if (nexthop_nh->nh_tclassid) 477 fi->fib_net->ipv4.fib_num_tclassid_users++; 478 #endif 479 } 480 481 rtnh = rtnh_next(rtnh, &remaining); 482 } endfor_nexthops(fi); 483 484 return 0; 485 } 486 487 #endif 488 489 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi) 490 { 491 #ifdef CONFIG_IP_ROUTE_MULTIPATH 492 struct rtnexthop *rtnh; 493 int remaining; 494 #endif 495 496 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority) 497 return 1; 498 499 if (cfg->fc_oif || cfg->fc_gw) { 500 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) && 501 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw)) 502 return 0; 503 return 1; 504 } 505 506 #ifdef CONFIG_IP_ROUTE_MULTIPATH 507 if (cfg->fc_mp == NULL) 508 return 0; 509 510 rtnh = cfg->fc_mp; 511 remaining = cfg->fc_mp_len; 512 513 for_nexthops(fi) { 514 int attrlen; 515 516 if (!rtnh_ok(rtnh, remaining)) 517 return -EINVAL; 518 519 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif) 520 return 1; 521 522 attrlen = rtnh_attrlen(rtnh); 523 if (attrlen > 0) { 524 struct nlattr *nla, *attrs = rtnh_attrs(rtnh); 525 526 nla = nla_find(attrs, attrlen, RTA_GATEWAY); 527 if (nla && nla_get_be32(nla) != nh->nh_gw) 528 return 1; 529 #ifdef CONFIG_IP_ROUTE_CLASSID 530 nla = nla_find(attrs, attrlen, RTA_FLOW); 531 if (nla && nla_get_u32(nla) != nh->nh_tclassid) 532 return 1; 533 #endif 534 } 535 536 rtnh = rtnh_next(rtnh, &remaining); 537 } endfor_nexthops(fi); 538 #endif 539 return 0; 540 } 541 542 543 /* 544 * Picture 545 * ------- 546 * 547 * Semantics of nexthop is very messy by historical reasons. 548 * We have to take into account, that: 549 * a) gateway can be actually local interface address, 550 * so that gatewayed route is direct. 551 * b) gateway must be on-link address, possibly 552 * described not by an ifaddr, but also by a direct route. 553 * c) If both gateway and interface are specified, they should not 554 * contradict. 555 * d) If we use tunnel routes, gateway could be not on-link. 556 * 557 * Attempt to reconcile all of these (alas, self-contradictory) conditions 558 * results in pretty ugly and hairy code with obscure logic. 559 * 560 * I chose to generalized it instead, so that the size 561 * of code does not increase practically, but it becomes 562 * much more general. 563 * Every prefix is assigned a "scope" value: "host" is local address, 564 * "link" is direct route, 565 * [ ... "site" ... "interior" ... ] 566 * and "universe" is true gateway route with global meaning. 567 * 568 * Every prefix refers to a set of "nexthop"s (gw, oif), 569 * where gw must have narrower scope. This recursion stops 570 * when gw has LOCAL scope or if "nexthop" is declared ONLINK, 571 * which means that gw is forced to be on link. 572 * 573 * Code is still hairy, but now it is apparently logically 574 * consistent and very flexible. F.e. as by-product it allows 575 * to co-exists in peace independent exterior and interior 576 * routing processes. 577 * 578 * Normally it looks as following. 579 * 580 * {universe prefix} -> (gw, oif) [scope link] 581 * | 582 * |-> {link prefix} -> (gw, oif) [scope local] 583 * | 584 * |-> {local prefix} (terminal node) 585 */ 586 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi, 587 struct fib_nh *nh) 588 { 589 int err; 590 struct net *net; 591 struct net_device *dev; 592 593 net = cfg->fc_nlinfo.nl_net; 594 if (nh->nh_gw) { 595 struct fib_result res; 596 597 if (nh->nh_flags & RTNH_F_ONLINK) { 598 599 if (cfg->fc_scope >= RT_SCOPE_LINK) 600 return -EINVAL; 601 if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST) 602 return -EINVAL; 603 dev = __dev_get_by_index(net, nh->nh_oif); 604 if (!dev) 605 return -ENODEV; 606 if (!(dev->flags & IFF_UP)) 607 return -ENETDOWN; 608 nh->nh_dev = dev; 609 dev_hold(dev); 610 nh->nh_scope = RT_SCOPE_LINK; 611 return 0; 612 } 613 rcu_read_lock(); 614 { 615 struct flowi4 fl4 = { 616 .daddr = nh->nh_gw, 617 .flowi4_scope = cfg->fc_scope + 1, 618 .flowi4_oif = nh->nh_oif, 619 .flowi4_iif = LOOPBACK_IFINDEX, 620 }; 621 622 /* It is not necessary, but requires a bit of thinking */ 623 if (fl4.flowi4_scope < RT_SCOPE_LINK) 624 fl4.flowi4_scope = RT_SCOPE_LINK; 625 err = fib_lookup(net, &fl4, &res); 626 if (err) { 627 rcu_read_unlock(); 628 return err; 629 } 630 } 631 err = -EINVAL; 632 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) 633 goto out; 634 nh->nh_scope = res.scope; 635 nh->nh_oif = FIB_RES_OIF(res); 636 nh->nh_dev = dev = FIB_RES_DEV(res); 637 if (!dev) 638 goto out; 639 dev_hold(dev); 640 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN; 641 } else { 642 struct in_device *in_dev; 643 644 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) 645 return -EINVAL; 646 647 rcu_read_lock(); 648 err = -ENODEV; 649 in_dev = inetdev_by_index(net, nh->nh_oif); 650 if (in_dev == NULL) 651 goto out; 652 err = -ENETDOWN; 653 if (!(in_dev->dev->flags & IFF_UP)) 654 goto out; 655 nh->nh_dev = in_dev->dev; 656 dev_hold(nh->nh_dev); 657 nh->nh_scope = RT_SCOPE_HOST; 658 err = 0; 659 } 660 out: 661 rcu_read_unlock(); 662 return err; 663 } 664 665 static inline unsigned int fib_laddr_hashfn(__be32 val) 666 { 667 unsigned int mask = (fib_info_hash_size - 1); 668 669 return ((__force u32)val ^ 670 ((__force u32)val >> 7) ^ 671 ((__force u32)val >> 14)) & mask; 672 } 673 674 static struct hlist_head *fib_info_hash_alloc(int bytes) 675 { 676 if (bytes <= PAGE_SIZE) 677 return kzalloc(bytes, GFP_KERNEL); 678 else 679 return (struct hlist_head *) 680 __get_free_pages(GFP_KERNEL | __GFP_ZERO, 681 get_order(bytes)); 682 } 683 684 static void fib_info_hash_free(struct hlist_head *hash, int bytes) 685 { 686 if (!hash) 687 return; 688 689 if (bytes <= PAGE_SIZE) 690 kfree(hash); 691 else 692 free_pages((unsigned long) hash, get_order(bytes)); 693 } 694 695 static void fib_info_hash_move(struct hlist_head *new_info_hash, 696 struct hlist_head *new_laddrhash, 697 unsigned int new_size) 698 { 699 struct hlist_head *old_info_hash, *old_laddrhash; 700 unsigned int old_size = fib_info_hash_size; 701 unsigned int i, bytes; 702 703 spin_lock_bh(&fib_info_lock); 704 old_info_hash = fib_info_hash; 705 old_laddrhash = fib_info_laddrhash; 706 fib_info_hash_size = new_size; 707 708 for (i = 0; i < old_size; i++) { 709 struct hlist_head *head = &fib_info_hash[i]; 710 struct hlist_node *n; 711 struct fib_info *fi; 712 713 hlist_for_each_entry_safe(fi, n, head, fib_hash) { 714 struct hlist_head *dest; 715 unsigned int new_hash; 716 717 hlist_del(&fi->fib_hash); 718 719 new_hash = fib_info_hashfn(fi); 720 dest = &new_info_hash[new_hash]; 721 hlist_add_head(&fi->fib_hash, dest); 722 } 723 } 724 fib_info_hash = new_info_hash; 725 726 for (i = 0; i < old_size; i++) { 727 struct hlist_head *lhead = &fib_info_laddrhash[i]; 728 struct hlist_node *n; 729 struct fib_info *fi; 730 731 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) { 732 struct hlist_head *ldest; 733 unsigned int new_hash; 734 735 hlist_del(&fi->fib_lhash); 736 737 new_hash = fib_laddr_hashfn(fi->fib_prefsrc); 738 ldest = &new_laddrhash[new_hash]; 739 hlist_add_head(&fi->fib_lhash, ldest); 740 } 741 } 742 fib_info_laddrhash = new_laddrhash; 743 744 spin_unlock_bh(&fib_info_lock); 745 746 bytes = old_size * sizeof(struct hlist_head *); 747 fib_info_hash_free(old_info_hash, bytes); 748 fib_info_hash_free(old_laddrhash, bytes); 749 } 750 751 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh) 752 { 753 nh->nh_saddr = inet_select_addr(nh->nh_dev, 754 nh->nh_gw, 755 nh->nh_parent->fib_scope); 756 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid); 757 758 return nh->nh_saddr; 759 } 760 761 struct fib_info *fib_create_info(struct fib_config *cfg) 762 { 763 int err; 764 struct fib_info *fi = NULL; 765 struct fib_info *ofi; 766 int nhs = 1; 767 struct net *net = cfg->fc_nlinfo.nl_net; 768 769 if (cfg->fc_type > RTN_MAX) 770 goto err_inval; 771 772 /* Fast check to catch the most weird cases */ 773 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) 774 goto err_inval; 775 776 #ifdef CONFIG_IP_ROUTE_MULTIPATH 777 if (cfg->fc_mp) { 778 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len); 779 if (nhs == 0) 780 goto err_inval; 781 } 782 #endif 783 784 err = -ENOBUFS; 785 if (fib_info_cnt >= fib_info_hash_size) { 786 unsigned int new_size = fib_info_hash_size << 1; 787 struct hlist_head *new_info_hash; 788 struct hlist_head *new_laddrhash; 789 unsigned int bytes; 790 791 if (!new_size) 792 new_size = 16; 793 bytes = new_size * sizeof(struct hlist_head *); 794 new_info_hash = fib_info_hash_alloc(bytes); 795 new_laddrhash = fib_info_hash_alloc(bytes); 796 if (!new_info_hash || !new_laddrhash) { 797 fib_info_hash_free(new_info_hash, bytes); 798 fib_info_hash_free(new_laddrhash, bytes); 799 } else 800 fib_info_hash_move(new_info_hash, new_laddrhash, new_size); 801 802 if (!fib_info_hash_size) 803 goto failure; 804 } 805 806 fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL); 807 if (fi == NULL) 808 goto failure; 809 fib_info_cnt++; 810 if (cfg->fc_mx) { 811 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL); 812 if (!fi->fib_metrics) 813 goto failure; 814 } else 815 fi->fib_metrics = (u32 *) dst_default_metrics; 816 817 fi->fib_net = hold_net(net); 818 fi->fib_protocol = cfg->fc_protocol; 819 fi->fib_scope = cfg->fc_scope; 820 fi->fib_flags = cfg->fc_flags; 821 fi->fib_priority = cfg->fc_priority; 822 fi->fib_prefsrc = cfg->fc_prefsrc; 823 fi->fib_type = cfg->fc_type; 824 825 fi->fib_nhs = nhs; 826 change_nexthops(fi) { 827 nexthop_nh->nh_parent = fi; 828 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *); 829 if (!nexthop_nh->nh_pcpu_rth_output) 830 goto failure; 831 } endfor_nexthops(fi) 832 833 if (cfg->fc_mx) { 834 struct nlattr *nla; 835 int remaining; 836 837 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) { 838 int type = nla_type(nla); 839 840 if (type) { 841 u32 val; 842 843 if (type > RTAX_MAX) 844 goto err_inval; 845 if (type == RTAX_CC_ALGO) { 846 char tmp[TCP_CA_NAME_MAX]; 847 848 nla_strlcpy(tmp, nla, sizeof(tmp)); 849 val = tcp_ca_get_key_by_name(tmp); 850 if (val == TCP_CA_UNSPEC) 851 goto err_inval; 852 } else { 853 val = nla_get_u32(nla); 854 } 855 if (type == RTAX_ADVMSS && val > 65535 - 40) 856 val = 65535 - 40; 857 if (type == RTAX_MTU && val > 65535 - 15) 858 val = 65535 - 15; 859 fi->fib_metrics[type - 1] = val; 860 } 861 } 862 } 863 864 if (cfg->fc_mp) { 865 #ifdef CONFIG_IP_ROUTE_MULTIPATH 866 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg); 867 if (err != 0) 868 goto failure; 869 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif) 870 goto err_inval; 871 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw) 872 goto err_inval; 873 #ifdef CONFIG_IP_ROUTE_CLASSID 874 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) 875 goto err_inval; 876 #endif 877 #else 878 goto err_inval; 879 #endif 880 } else { 881 struct fib_nh *nh = fi->fib_nh; 882 883 nh->nh_oif = cfg->fc_oif; 884 nh->nh_gw = cfg->fc_gw; 885 nh->nh_flags = cfg->fc_flags; 886 #ifdef CONFIG_IP_ROUTE_CLASSID 887 nh->nh_tclassid = cfg->fc_flow; 888 if (nh->nh_tclassid) 889 fi->fib_net->ipv4.fib_num_tclassid_users++; 890 #endif 891 #ifdef CONFIG_IP_ROUTE_MULTIPATH 892 nh->nh_weight = 1; 893 #endif 894 } 895 896 if (fib_props[cfg->fc_type].error) { 897 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp) 898 goto err_inval; 899 goto link_it; 900 } else { 901 switch (cfg->fc_type) { 902 case RTN_UNICAST: 903 case RTN_LOCAL: 904 case RTN_BROADCAST: 905 case RTN_ANYCAST: 906 case RTN_MULTICAST: 907 break; 908 default: 909 goto err_inval; 910 } 911 } 912 913 if (cfg->fc_scope > RT_SCOPE_HOST) 914 goto err_inval; 915 916 if (cfg->fc_scope == RT_SCOPE_HOST) { 917 struct fib_nh *nh = fi->fib_nh; 918 919 /* Local address is added. */ 920 if (nhs != 1 || nh->nh_gw) 921 goto err_inval; 922 nh->nh_scope = RT_SCOPE_NOWHERE; 923 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif); 924 err = -ENODEV; 925 if (nh->nh_dev == NULL) 926 goto failure; 927 } else { 928 change_nexthops(fi) { 929 err = fib_check_nh(cfg, fi, nexthop_nh); 930 if (err != 0) 931 goto failure; 932 } endfor_nexthops(fi) 933 } 934 935 if (fi->fib_prefsrc) { 936 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst || 937 fi->fib_prefsrc != cfg->fc_dst) 938 if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL) 939 goto err_inval; 940 } 941 942 change_nexthops(fi) { 943 fib_info_update_nh_saddr(net, nexthop_nh); 944 } endfor_nexthops(fi) 945 946 link_it: 947 ofi = fib_find_info(fi); 948 if (ofi) { 949 fi->fib_dead = 1; 950 free_fib_info(fi); 951 ofi->fib_treeref++; 952 return ofi; 953 } 954 955 fi->fib_treeref++; 956 atomic_inc(&fi->fib_clntref); 957 spin_lock_bh(&fib_info_lock); 958 hlist_add_head(&fi->fib_hash, 959 &fib_info_hash[fib_info_hashfn(fi)]); 960 if (fi->fib_prefsrc) { 961 struct hlist_head *head; 962 963 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)]; 964 hlist_add_head(&fi->fib_lhash, head); 965 } 966 change_nexthops(fi) { 967 struct hlist_head *head; 968 unsigned int hash; 969 970 if (!nexthop_nh->nh_dev) 971 continue; 972 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex); 973 head = &fib_info_devhash[hash]; 974 hlist_add_head(&nexthop_nh->nh_hash, head); 975 } endfor_nexthops(fi) 976 spin_unlock_bh(&fib_info_lock); 977 return fi; 978 979 err_inval: 980 err = -EINVAL; 981 982 failure: 983 if (fi) { 984 fi->fib_dead = 1; 985 free_fib_info(fi); 986 } 987 988 return ERR_PTR(err); 989 } 990 991 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event, 992 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos, 993 struct fib_info *fi, unsigned int flags) 994 { 995 struct nlmsghdr *nlh; 996 struct rtmsg *rtm; 997 998 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags); 999 if (nlh == NULL) 1000 return -EMSGSIZE; 1001 1002 rtm = nlmsg_data(nlh); 1003 rtm->rtm_family = AF_INET; 1004 rtm->rtm_dst_len = dst_len; 1005 rtm->rtm_src_len = 0; 1006 rtm->rtm_tos = tos; 1007 if (tb_id < 256) 1008 rtm->rtm_table = tb_id; 1009 else 1010 rtm->rtm_table = RT_TABLE_COMPAT; 1011 if (nla_put_u32(skb, RTA_TABLE, tb_id)) 1012 goto nla_put_failure; 1013 rtm->rtm_type = type; 1014 rtm->rtm_flags = fi->fib_flags; 1015 rtm->rtm_scope = fi->fib_scope; 1016 rtm->rtm_protocol = fi->fib_protocol; 1017 1018 if (rtm->rtm_dst_len && 1019 nla_put_be32(skb, RTA_DST, dst)) 1020 goto nla_put_failure; 1021 if (fi->fib_priority && 1022 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority)) 1023 goto nla_put_failure; 1024 if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0) 1025 goto nla_put_failure; 1026 1027 if (fi->fib_prefsrc && 1028 nla_put_be32(skb, RTA_PREFSRC, fi->fib_prefsrc)) 1029 goto nla_put_failure; 1030 if (fi->fib_nhs == 1) { 1031 if (fi->fib_nh->nh_gw && 1032 nla_put_be32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw)) 1033 goto nla_put_failure; 1034 if (fi->fib_nh->nh_oif && 1035 nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif)) 1036 goto nla_put_failure; 1037 #ifdef CONFIG_IP_ROUTE_CLASSID 1038 if (fi->fib_nh[0].nh_tclassid && 1039 nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid)) 1040 goto nla_put_failure; 1041 #endif 1042 } 1043 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1044 if (fi->fib_nhs > 1) { 1045 struct rtnexthop *rtnh; 1046 struct nlattr *mp; 1047 1048 mp = nla_nest_start(skb, RTA_MULTIPATH); 1049 if (mp == NULL) 1050 goto nla_put_failure; 1051 1052 for_nexthops(fi) { 1053 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh)); 1054 if (rtnh == NULL) 1055 goto nla_put_failure; 1056 1057 rtnh->rtnh_flags = nh->nh_flags & 0xFF; 1058 rtnh->rtnh_hops = nh->nh_weight - 1; 1059 rtnh->rtnh_ifindex = nh->nh_oif; 1060 1061 if (nh->nh_gw && 1062 nla_put_be32(skb, RTA_GATEWAY, nh->nh_gw)) 1063 goto nla_put_failure; 1064 #ifdef CONFIG_IP_ROUTE_CLASSID 1065 if (nh->nh_tclassid && 1066 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid)) 1067 goto nla_put_failure; 1068 #endif 1069 /* length of rtnetlink header + attributes */ 1070 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh; 1071 } endfor_nexthops(fi); 1072 1073 nla_nest_end(skb, mp); 1074 } 1075 #endif 1076 nlmsg_end(skb, nlh); 1077 return 0; 1078 1079 nla_put_failure: 1080 nlmsg_cancel(skb, nlh); 1081 return -EMSGSIZE; 1082 } 1083 1084 /* 1085 * Update FIB if: 1086 * - local address disappeared -> we must delete all the entries 1087 * referring to it. 1088 * - device went down -> we must shutdown all nexthops going via it. 1089 */ 1090 int fib_sync_down_addr(struct net *net, __be32 local) 1091 { 1092 int ret = 0; 1093 unsigned int hash = fib_laddr_hashfn(local); 1094 struct hlist_head *head = &fib_info_laddrhash[hash]; 1095 struct fib_info *fi; 1096 1097 if (fib_info_laddrhash == NULL || local == 0) 1098 return 0; 1099 1100 hlist_for_each_entry(fi, head, fib_lhash) { 1101 if (!net_eq(fi->fib_net, net)) 1102 continue; 1103 if (fi->fib_prefsrc == local) { 1104 fi->fib_flags |= RTNH_F_DEAD; 1105 ret++; 1106 } 1107 } 1108 return ret; 1109 } 1110 1111 int fib_sync_down_dev(struct net_device *dev, int force) 1112 { 1113 int ret = 0; 1114 int scope = RT_SCOPE_NOWHERE; 1115 struct fib_info *prev_fi = NULL; 1116 unsigned int hash = fib_devindex_hashfn(dev->ifindex); 1117 struct hlist_head *head = &fib_info_devhash[hash]; 1118 struct fib_nh *nh; 1119 1120 if (force) 1121 scope = -1; 1122 1123 hlist_for_each_entry(nh, head, nh_hash) { 1124 struct fib_info *fi = nh->nh_parent; 1125 int dead; 1126 1127 BUG_ON(!fi->fib_nhs); 1128 if (nh->nh_dev != dev || fi == prev_fi) 1129 continue; 1130 prev_fi = fi; 1131 dead = 0; 1132 change_nexthops(fi) { 1133 if (nexthop_nh->nh_flags & RTNH_F_DEAD) 1134 dead++; 1135 else if (nexthop_nh->nh_dev == dev && 1136 nexthop_nh->nh_scope != scope) { 1137 nexthop_nh->nh_flags |= RTNH_F_DEAD; 1138 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1139 spin_lock_bh(&fib_multipath_lock); 1140 fi->fib_power -= nexthop_nh->nh_power; 1141 nexthop_nh->nh_power = 0; 1142 spin_unlock_bh(&fib_multipath_lock); 1143 #endif 1144 dead++; 1145 } 1146 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1147 if (force > 1 && nexthop_nh->nh_dev == dev) { 1148 dead = fi->fib_nhs; 1149 break; 1150 } 1151 #endif 1152 } endfor_nexthops(fi) 1153 if (dead == fi->fib_nhs) { 1154 fi->fib_flags |= RTNH_F_DEAD; 1155 ret++; 1156 } 1157 } 1158 1159 return ret; 1160 } 1161 1162 /* Must be invoked inside of an RCU protected region. */ 1163 void fib_select_default(struct fib_result *res) 1164 { 1165 struct fib_info *fi = NULL, *last_resort = NULL; 1166 struct list_head *fa_head = res->fa_head; 1167 struct fib_table *tb = res->table; 1168 int order = -1, last_idx = -1; 1169 struct fib_alias *fa; 1170 1171 list_for_each_entry_rcu(fa, fa_head, fa_list) { 1172 struct fib_info *next_fi = fa->fa_info; 1173 1174 if (next_fi->fib_scope != res->scope || 1175 fa->fa_type != RTN_UNICAST) 1176 continue; 1177 1178 if (next_fi->fib_priority > res->fi->fib_priority) 1179 break; 1180 if (!next_fi->fib_nh[0].nh_gw || 1181 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK) 1182 continue; 1183 1184 fib_alias_accessed(fa); 1185 1186 if (fi == NULL) { 1187 if (next_fi != res->fi) 1188 break; 1189 } else if (!fib_detect_death(fi, order, &last_resort, 1190 &last_idx, tb->tb_default)) { 1191 fib_result_assign(res, fi); 1192 tb->tb_default = order; 1193 goto out; 1194 } 1195 fi = next_fi; 1196 order++; 1197 } 1198 1199 if (order <= 0 || fi == NULL) { 1200 tb->tb_default = -1; 1201 goto out; 1202 } 1203 1204 if (!fib_detect_death(fi, order, &last_resort, &last_idx, 1205 tb->tb_default)) { 1206 fib_result_assign(res, fi); 1207 tb->tb_default = order; 1208 goto out; 1209 } 1210 1211 if (last_idx >= 0) 1212 fib_result_assign(res, last_resort); 1213 tb->tb_default = last_idx; 1214 out: 1215 return; 1216 } 1217 1218 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1219 1220 /* 1221 * Dead device goes up. We wake up dead nexthops. 1222 * It takes sense only on multipath routes. 1223 */ 1224 int fib_sync_up(struct net_device *dev) 1225 { 1226 struct fib_info *prev_fi; 1227 unsigned int hash; 1228 struct hlist_head *head; 1229 struct fib_nh *nh; 1230 int ret; 1231 1232 if (!(dev->flags & IFF_UP)) 1233 return 0; 1234 1235 prev_fi = NULL; 1236 hash = fib_devindex_hashfn(dev->ifindex); 1237 head = &fib_info_devhash[hash]; 1238 ret = 0; 1239 1240 hlist_for_each_entry(nh, head, nh_hash) { 1241 struct fib_info *fi = nh->nh_parent; 1242 int alive; 1243 1244 BUG_ON(!fi->fib_nhs); 1245 if (nh->nh_dev != dev || fi == prev_fi) 1246 continue; 1247 1248 prev_fi = fi; 1249 alive = 0; 1250 change_nexthops(fi) { 1251 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) { 1252 alive++; 1253 continue; 1254 } 1255 if (nexthop_nh->nh_dev == NULL || 1256 !(nexthop_nh->nh_dev->flags & IFF_UP)) 1257 continue; 1258 if (nexthop_nh->nh_dev != dev || 1259 !__in_dev_get_rtnl(dev)) 1260 continue; 1261 alive++; 1262 spin_lock_bh(&fib_multipath_lock); 1263 nexthop_nh->nh_power = 0; 1264 nexthop_nh->nh_flags &= ~RTNH_F_DEAD; 1265 spin_unlock_bh(&fib_multipath_lock); 1266 } endfor_nexthops(fi) 1267 1268 if (alive > 0) { 1269 fi->fib_flags &= ~RTNH_F_DEAD; 1270 ret++; 1271 } 1272 } 1273 1274 return ret; 1275 } 1276 1277 /* 1278 * The algorithm is suboptimal, but it provides really 1279 * fair weighted route distribution. 1280 */ 1281 void fib_select_multipath(struct fib_result *res) 1282 { 1283 struct fib_info *fi = res->fi; 1284 int w; 1285 1286 spin_lock_bh(&fib_multipath_lock); 1287 if (fi->fib_power <= 0) { 1288 int power = 0; 1289 change_nexthops(fi) { 1290 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) { 1291 power += nexthop_nh->nh_weight; 1292 nexthop_nh->nh_power = nexthop_nh->nh_weight; 1293 } 1294 } endfor_nexthops(fi); 1295 fi->fib_power = power; 1296 if (power <= 0) { 1297 spin_unlock_bh(&fib_multipath_lock); 1298 /* Race condition: route has just become dead. */ 1299 res->nh_sel = 0; 1300 return; 1301 } 1302 } 1303 1304 1305 /* w should be random number [0..fi->fib_power-1], 1306 * it is pretty bad approximation. 1307 */ 1308 1309 w = jiffies % fi->fib_power; 1310 1311 change_nexthops(fi) { 1312 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) && 1313 nexthop_nh->nh_power) { 1314 w -= nexthop_nh->nh_power; 1315 if (w <= 0) { 1316 nexthop_nh->nh_power--; 1317 fi->fib_power--; 1318 res->nh_sel = nhsel; 1319 spin_unlock_bh(&fib_multipath_lock); 1320 return; 1321 } 1322 } 1323 } endfor_nexthops(fi); 1324 1325 /* Race condition: route has just become dead. */ 1326 res->nh_sel = 0; 1327 spin_unlock_bh(&fib_multipath_lock); 1328 } 1329 #endif 1330