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 <linux/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 #include <linux/netlink.h> 36 37 #include <net/arp.h> 38 #include <net/ip.h> 39 #include <net/protocol.h> 40 #include <net/route.h> 41 #include <net/tcp.h> 42 #include <net/sock.h> 43 #include <net/ip_fib.h> 44 #include <net/netlink.h> 45 #include <net/nexthop.h> 46 #include <net/lwtunnel.h> 47 #include <net/fib_notifier.h> 48 49 #include "fib_lookup.h" 50 51 static DEFINE_SPINLOCK(fib_info_lock); 52 static struct hlist_head *fib_info_hash; 53 static struct hlist_head *fib_info_laddrhash; 54 static unsigned int fib_info_hash_size; 55 static unsigned int fib_info_cnt; 56 57 #define DEVINDEX_HASHBITS 8 58 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS) 59 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE]; 60 61 #ifdef CONFIG_IP_ROUTE_MULTIPATH 62 63 #define for_nexthops(fi) { \ 64 int nhsel; const struct fib_nh *nh; \ 65 for (nhsel = 0, nh = (fi)->fib_nh; \ 66 nhsel < (fi)->fib_nhs; \ 67 nh++, nhsel++) 68 69 #define change_nexthops(fi) { \ 70 int nhsel; struct fib_nh *nexthop_nh; \ 71 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \ 72 nhsel < (fi)->fib_nhs; \ 73 nexthop_nh++, nhsel++) 74 75 #else /* CONFIG_IP_ROUTE_MULTIPATH */ 76 77 /* Hope, that gcc will optimize it to get rid of dummy loop */ 78 79 #define for_nexthops(fi) { \ 80 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \ 81 for (nhsel = 0; nhsel < 1; nhsel++) 82 83 #define change_nexthops(fi) { \ 84 int nhsel; \ 85 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \ 86 for (nhsel = 0; nhsel < 1; nhsel++) 87 88 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 89 90 #define endfor_nexthops(fi) } 91 92 93 const struct fib_prop fib_props[RTN_MAX + 1] = { 94 [RTN_UNSPEC] = { 95 .error = 0, 96 .scope = RT_SCOPE_NOWHERE, 97 }, 98 [RTN_UNICAST] = { 99 .error = 0, 100 .scope = RT_SCOPE_UNIVERSE, 101 }, 102 [RTN_LOCAL] = { 103 .error = 0, 104 .scope = RT_SCOPE_HOST, 105 }, 106 [RTN_BROADCAST] = { 107 .error = 0, 108 .scope = RT_SCOPE_LINK, 109 }, 110 [RTN_ANYCAST] = { 111 .error = 0, 112 .scope = RT_SCOPE_LINK, 113 }, 114 [RTN_MULTICAST] = { 115 .error = 0, 116 .scope = RT_SCOPE_UNIVERSE, 117 }, 118 [RTN_BLACKHOLE] = { 119 .error = -EINVAL, 120 .scope = RT_SCOPE_UNIVERSE, 121 }, 122 [RTN_UNREACHABLE] = { 123 .error = -EHOSTUNREACH, 124 .scope = RT_SCOPE_UNIVERSE, 125 }, 126 [RTN_PROHIBIT] = { 127 .error = -EACCES, 128 .scope = RT_SCOPE_UNIVERSE, 129 }, 130 [RTN_THROW] = { 131 .error = -EAGAIN, 132 .scope = RT_SCOPE_UNIVERSE, 133 }, 134 [RTN_NAT] = { 135 .error = -EINVAL, 136 .scope = RT_SCOPE_NOWHERE, 137 }, 138 [RTN_XRESOLVE] = { 139 .error = -EINVAL, 140 .scope = RT_SCOPE_NOWHERE, 141 }, 142 }; 143 144 static void rt_fibinfo_free(struct rtable __rcu **rtp) 145 { 146 struct rtable *rt = rcu_dereference_protected(*rtp, 1); 147 148 if (!rt) 149 return; 150 151 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL); 152 * because we waited an RCU grace period before calling 153 * free_fib_info_rcu() 154 */ 155 156 dst_dev_put(&rt->dst); 157 dst_release_immediate(&rt->dst); 158 } 159 160 static void free_nh_exceptions(struct fib_nh *nh) 161 { 162 struct fnhe_hash_bucket *hash; 163 int i; 164 165 hash = rcu_dereference_protected(nh->nh_exceptions, 1); 166 if (!hash) 167 return; 168 for (i = 0; i < FNHE_HASH_SIZE; i++) { 169 struct fib_nh_exception *fnhe; 170 171 fnhe = rcu_dereference_protected(hash[i].chain, 1); 172 while (fnhe) { 173 struct fib_nh_exception *next; 174 175 next = rcu_dereference_protected(fnhe->fnhe_next, 1); 176 177 rt_fibinfo_free(&fnhe->fnhe_rth_input); 178 rt_fibinfo_free(&fnhe->fnhe_rth_output); 179 180 kfree(fnhe); 181 182 fnhe = next; 183 } 184 } 185 kfree(hash); 186 } 187 188 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp) 189 { 190 int cpu; 191 192 if (!rtp) 193 return; 194 195 for_each_possible_cpu(cpu) { 196 struct rtable *rt; 197 198 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1); 199 if (rt) { 200 dst_dev_put(&rt->dst); 201 dst_release_immediate(&rt->dst); 202 } 203 } 204 free_percpu(rtp); 205 } 206 207 /* Release a nexthop info record */ 208 static void free_fib_info_rcu(struct rcu_head *head) 209 { 210 struct fib_info *fi = container_of(head, struct fib_info, rcu); 211 struct dst_metrics *m; 212 213 change_nexthops(fi) { 214 if (nexthop_nh->nh_dev) 215 dev_put(nexthop_nh->nh_dev); 216 lwtstate_put(nexthop_nh->nh_lwtstate); 217 free_nh_exceptions(nexthop_nh); 218 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output); 219 rt_fibinfo_free(&nexthop_nh->nh_rth_input); 220 } endfor_nexthops(fi); 221 222 m = fi->fib_metrics; 223 if (m != &dst_default_metrics && refcount_dec_and_test(&m->refcnt)) 224 kfree(m); 225 kfree(fi); 226 } 227 228 void free_fib_info(struct fib_info *fi) 229 { 230 if (fi->fib_dead == 0) { 231 pr_warn("Freeing alive fib_info %p\n", fi); 232 return; 233 } 234 fib_info_cnt--; 235 #ifdef CONFIG_IP_ROUTE_CLASSID 236 change_nexthops(fi) { 237 if (nexthop_nh->nh_tclassid) 238 fi->fib_net->ipv4.fib_num_tclassid_users--; 239 } endfor_nexthops(fi); 240 #endif 241 call_rcu(&fi->rcu, free_fib_info_rcu); 242 } 243 EXPORT_SYMBOL_GPL(free_fib_info); 244 245 void fib_release_info(struct fib_info *fi) 246 { 247 spin_lock_bh(&fib_info_lock); 248 if (fi && --fi->fib_treeref == 0) { 249 hlist_del(&fi->fib_hash); 250 if (fi->fib_prefsrc) 251 hlist_del(&fi->fib_lhash); 252 change_nexthops(fi) { 253 if (!nexthop_nh->nh_dev) 254 continue; 255 hlist_del(&nexthop_nh->nh_hash); 256 } endfor_nexthops(fi) 257 fi->fib_dead = 1; 258 fib_info_put(fi); 259 } 260 spin_unlock_bh(&fib_info_lock); 261 } 262 263 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi) 264 { 265 const struct fib_nh *onh = ofi->fib_nh; 266 267 for_nexthops(fi) { 268 if (nh->nh_oif != onh->nh_oif || 269 nh->nh_gw != onh->nh_gw || 270 nh->nh_scope != onh->nh_scope || 271 #ifdef CONFIG_IP_ROUTE_MULTIPATH 272 nh->nh_weight != onh->nh_weight || 273 #endif 274 #ifdef CONFIG_IP_ROUTE_CLASSID 275 nh->nh_tclassid != onh->nh_tclassid || 276 #endif 277 lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) || 278 ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK)) 279 return -1; 280 onh++; 281 } endfor_nexthops(fi); 282 return 0; 283 } 284 285 static inline unsigned int fib_devindex_hashfn(unsigned int val) 286 { 287 unsigned int mask = DEVINDEX_HASHSIZE - 1; 288 289 return (val ^ 290 (val >> DEVINDEX_HASHBITS) ^ 291 (val >> (DEVINDEX_HASHBITS * 2))) & mask; 292 } 293 294 static inline unsigned int fib_info_hashfn(const struct fib_info *fi) 295 { 296 unsigned int mask = (fib_info_hash_size - 1); 297 unsigned int val = fi->fib_nhs; 298 299 val ^= (fi->fib_protocol << 8) | fi->fib_scope; 300 val ^= (__force u32)fi->fib_prefsrc; 301 val ^= fi->fib_priority; 302 for_nexthops(fi) { 303 val ^= fib_devindex_hashfn(nh->nh_oif); 304 } endfor_nexthops(fi) 305 306 return (val ^ (val >> 7) ^ (val >> 12)) & mask; 307 } 308 309 static struct fib_info *fib_find_info(const struct fib_info *nfi) 310 { 311 struct hlist_head *head; 312 struct fib_info *fi; 313 unsigned int hash; 314 315 hash = fib_info_hashfn(nfi); 316 head = &fib_info_hash[hash]; 317 318 hlist_for_each_entry(fi, head, fib_hash) { 319 if (!net_eq(fi->fib_net, nfi->fib_net)) 320 continue; 321 if (fi->fib_nhs != nfi->fib_nhs) 322 continue; 323 if (nfi->fib_protocol == fi->fib_protocol && 324 nfi->fib_scope == fi->fib_scope && 325 nfi->fib_prefsrc == fi->fib_prefsrc && 326 nfi->fib_priority == fi->fib_priority && 327 nfi->fib_type == fi->fib_type && 328 memcmp(nfi->fib_metrics, fi->fib_metrics, 329 sizeof(u32) * RTAX_MAX) == 0 && 330 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) && 331 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0)) 332 return fi; 333 } 334 335 return NULL; 336 } 337 338 /* Check, that the gateway is already configured. 339 * Used only by redirect accept routine. 340 */ 341 int ip_fib_check_default(__be32 gw, struct net_device *dev) 342 { 343 struct hlist_head *head; 344 struct fib_nh *nh; 345 unsigned int hash; 346 347 spin_lock(&fib_info_lock); 348 349 hash = fib_devindex_hashfn(dev->ifindex); 350 head = &fib_info_devhash[hash]; 351 hlist_for_each_entry(nh, head, nh_hash) { 352 if (nh->nh_dev == dev && 353 nh->nh_gw == gw && 354 !(nh->nh_flags & RTNH_F_DEAD)) { 355 spin_unlock(&fib_info_lock); 356 return 0; 357 } 358 } 359 360 spin_unlock(&fib_info_lock); 361 362 return -1; 363 } 364 365 static inline size_t fib_nlmsg_size(struct fib_info *fi) 366 { 367 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg)) 368 + nla_total_size(4) /* RTA_TABLE */ 369 + nla_total_size(4) /* RTA_DST */ 370 + nla_total_size(4) /* RTA_PRIORITY */ 371 + nla_total_size(4) /* RTA_PREFSRC */ 372 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */ 373 374 /* space for nested metrics */ 375 payload += nla_total_size((RTAX_MAX * nla_total_size(4))); 376 377 if (fi->fib_nhs) { 378 size_t nh_encapsize = 0; 379 /* Also handles the special case fib_nhs == 1 */ 380 381 /* each nexthop is packed in an attribute */ 382 size_t nhsize = nla_total_size(sizeof(struct rtnexthop)); 383 384 /* may contain flow and gateway attribute */ 385 nhsize += 2 * nla_total_size(4); 386 387 /* grab encap info */ 388 for_nexthops(fi) { 389 if (nh->nh_lwtstate) { 390 /* RTA_ENCAP_TYPE */ 391 nh_encapsize += lwtunnel_get_encap_size( 392 nh->nh_lwtstate); 393 /* RTA_ENCAP */ 394 nh_encapsize += nla_total_size(2); 395 } 396 } endfor_nexthops(fi); 397 398 /* all nexthops are packed in a nested attribute */ 399 payload += nla_total_size((fi->fib_nhs * nhsize) + 400 nh_encapsize); 401 402 } 403 404 return payload; 405 } 406 407 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa, 408 int dst_len, u32 tb_id, const struct nl_info *info, 409 unsigned int nlm_flags) 410 { 411 struct sk_buff *skb; 412 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0; 413 int err = -ENOBUFS; 414 415 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL); 416 if (!skb) 417 goto errout; 418 419 err = fib_dump_info(skb, info->portid, seq, event, tb_id, 420 fa->fa_type, key, dst_len, 421 fa->fa_tos, fa->fa_info, nlm_flags); 422 if (err < 0) { 423 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */ 424 WARN_ON(err == -EMSGSIZE); 425 kfree_skb(skb); 426 goto errout; 427 } 428 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE, 429 info->nlh, GFP_KERNEL); 430 return; 431 errout: 432 if (err < 0) 433 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err); 434 } 435 436 static int fib_detect_death(struct fib_info *fi, int order, 437 struct fib_info **last_resort, int *last_idx, 438 int dflt) 439 { 440 struct neighbour *n; 441 int state = NUD_NONE; 442 443 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev); 444 if (n) { 445 state = n->nud_state; 446 neigh_release(n); 447 } else { 448 return 0; 449 } 450 if (state == NUD_REACHABLE) 451 return 0; 452 if ((state & NUD_VALID) && order != dflt) 453 return 0; 454 if ((state & NUD_VALID) || 455 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) { 456 *last_resort = fi; 457 *last_idx = order; 458 } 459 return 1; 460 } 461 462 #ifdef CONFIG_IP_ROUTE_MULTIPATH 463 464 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining, 465 struct netlink_ext_ack *extack) 466 { 467 int nhs = 0; 468 469 while (rtnh_ok(rtnh, remaining)) { 470 nhs++; 471 rtnh = rtnh_next(rtnh, &remaining); 472 } 473 474 /* leftover implies invalid nexthop configuration, discard it */ 475 if (remaining > 0) { 476 NL_SET_ERR_MSG(extack, 477 "Invalid nexthop configuration - extra data after nexthops"); 478 nhs = 0; 479 } 480 481 return nhs; 482 } 483 484 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh, 485 int remaining, struct fib_config *cfg, 486 struct netlink_ext_ack *extack) 487 { 488 int ret; 489 490 change_nexthops(fi) { 491 int attrlen; 492 493 if (!rtnh_ok(rtnh, remaining)) { 494 NL_SET_ERR_MSG(extack, 495 "Invalid nexthop configuration - extra data after nexthop"); 496 return -EINVAL; 497 } 498 499 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) { 500 NL_SET_ERR_MSG(extack, 501 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN"); 502 return -EINVAL; 503 } 504 505 nexthop_nh->nh_flags = 506 (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags; 507 nexthop_nh->nh_oif = rtnh->rtnh_ifindex; 508 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1; 509 510 attrlen = rtnh_attrlen(rtnh); 511 if (attrlen > 0) { 512 struct nlattr *nla, *attrs = rtnh_attrs(rtnh); 513 514 nla = nla_find(attrs, attrlen, RTA_GATEWAY); 515 nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0; 516 #ifdef CONFIG_IP_ROUTE_CLASSID 517 nla = nla_find(attrs, attrlen, RTA_FLOW); 518 nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0; 519 if (nexthop_nh->nh_tclassid) 520 fi->fib_net->ipv4.fib_num_tclassid_users++; 521 #endif 522 nla = nla_find(attrs, attrlen, RTA_ENCAP); 523 if (nla) { 524 struct lwtunnel_state *lwtstate; 525 struct nlattr *nla_entype; 526 527 nla_entype = nla_find(attrs, attrlen, 528 RTA_ENCAP_TYPE); 529 if (!nla_entype) { 530 NL_SET_BAD_ATTR(extack, nla); 531 NL_SET_ERR_MSG(extack, 532 "Encap type is missing"); 533 goto err_inval; 534 } 535 536 ret = lwtunnel_build_state(nla_get_u16( 537 nla_entype), 538 nla, AF_INET, cfg, 539 &lwtstate, extack); 540 if (ret) 541 goto errout; 542 nexthop_nh->nh_lwtstate = 543 lwtstate_get(lwtstate); 544 } 545 } 546 547 rtnh = rtnh_next(rtnh, &remaining); 548 } endfor_nexthops(fi); 549 550 return 0; 551 552 err_inval: 553 ret = -EINVAL; 554 555 errout: 556 return ret; 557 } 558 559 static void fib_rebalance(struct fib_info *fi) 560 { 561 int total; 562 int w; 563 struct in_device *in_dev; 564 565 if (fi->fib_nhs < 2) 566 return; 567 568 total = 0; 569 for_nexthops(fi) { 570 if (nh->nh_flags & RTNH_F_DEAD) 571 continue; 572 573 in_dev = __in_dev_get_rtnl(nh->nh_dev); 574 575 if (in_dev && 576 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) && 577 nh->nh_flags & RTNH_F_LINKDOWN) 578 continue; 579 580 total += nh->nh_weight; 581 } endfor_nexthops(fi); 582 583 w = 0; 584 change_nexthops(fi) { 585 int upper_bound; 586 587 in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev); 588 589 if (nexthop_nh->nh_flags & RTNH_F_DEAD) { 590 upper_bound = -1; 591 } else if (in_dev && 592 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) && 593 nexthop_nh->nh_flags & RTNH_F_LINKDOWN) { 594 upper_bound = -1; 595 } else { 596 w += nexthop_nh->nh_weight; 597 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31, 598 total) - 1; 599 } 600 601 atomic_set(&nexthop_nh->nh_upper_bound, upper_bound); 602 } endfor_nexthops(fi); 603 } 604 #else /* CONFIG_IP_ROUTE_MULTIPATH */ 605 606 #define fib_rebalance(fi) do { } while (0) 607 608 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 609 610 static int fib_encap_match(u16 encap_type, 611 struct nlattr *encap, 612 const struct fib_nh *nh, 613 const struct fib_config *cfg, 614 struct netlink_ext_ack *extack) 615 { 616 struct lwtunnel_state *lwtstate; 617 int ret, result = 0; 618 619 if (encap_type == LWTUNNEL_ENCAP_NONE) 620 return 0; 621 622 ret = lwtunnel_build_state(encap_type, encap, AF_INET, 623 cfg, &lwtstate, extack); 624 if (!ret) { 625 result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate); 626 lwtstate_free(lwtstate); 627 } 628 629 return result; 630 } 631 632 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi, 633 struct netlink_ext_ack *extack) 634 { 635 #ifdef CONFIG_IP_ROUTE_MULTIPATH 636 struct rtnexthop *rtnh; 637 int remaining; 638 #endif 639 640 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority) 641 return 1; 642 643 if (cfg->fc_oif || cfg->fc_gw) { 644 if (cfg->fc_encap) { 645 if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap, 646 fi->fib_nh, cfg, extack)) 647 return 1; 648 } 649 #ifdef CONFIG_IP_ROUTE_CLASSID 650 if (cfg->fc_flow && 651 cfg->fc_flow != fi->fib_nh->nh_tclassid) 652 return 1; 653 #endif 654 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) && 655 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw)) 656 return 0; 657 return 1; 658 } 659 660 #ifdef CONFIG_IP_ROUTE_MULTIPATH 661 if (!cfg->fc_mp) 662 return 0; 663 664 rtnh = cfg->fc_mp; 665 remaining = cfg->fc_mp_len; 666 667 for_nexthops(fi) { 668 int attrlen; 669 670 if (!rtnh_ok(rtnh, remaining)) 671 return -EINVAL; 672 673 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif) 674 return 1; 675 676 attrlen = rtnh_attrlen(rtnh); 677 if (attrlen > 0) { 678 struct nlattr *nla, *attrs = rtnh_attrs(rtnh); 679 680 nla = nla_find(attrs, attrlen, RTA_GATEWAY); 681 if (nla && nla_get_in_addr(nla) != nh->nh_gw) 682 return 1; 683 #ifdef CONFIG_IP_ROUTE_CLASSID 684 nla = nla_find(attrs, attrlen, RTA_FLOW); 685 if (nla && nla_get_u32(nla) != nh->nh_tclassid) 686 return 1; 687 #endif 688 } 689 690 rtnh = rtnh_next(rtnh, &remaining); 691 } endfor_nexthops(fi); 692 #endif 693 return 0; 694 } 695 696 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi) 697 { 698 struct nlattr *nla; 699 int remaining; 700 701 if (!cfg->fc_mx) 702 return true; 703 704 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) { 705 int type = nla_type(nla); 706 u32 fi_val, val; 707 708 if (!type) 709 continue; 710 if (type > RTAX_MAX) 711 return false; 712 713 if (type == RTAX_CC_ALGO) { 714 char tmp[TCP_CA_NAME_MAX]; 715 bool ecn_ca = false; 716 717 nla_strlcpy(tmp, nla, sizeof(tmp)); 718 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca); 719 } else { 720 if (nla_len(nla) != sizeof(u32)) 721 return false; 722 val = nla_get_u32(nla); 723 } 724 725 fi_val = fi->fib_metrics->metrics[type - 1]; 726 if (type == RTAX_FEATURES) 727 fi_val &= ~DST_FEATURE_ECN_CA; 728 729 if (fi_val != val) 730 return false; 731 } 732 733 return true; 734 } 735 736 737 /* 738 * Picture 739 * ------- 740 * 741 * Semantics of nexthop is very messy by historical reasons. 742 * We have to take into account, that: 743 * a) gateway can be actually local interface address, 744 * so that gatewayed route is direct. 745 * b) gateway must be on-link address, possibly 746 * described not by an ifaddr, but also by a direct route. 747 * c) If both gateway and interface are specified, they should not 748 * contradict. 749 * d) If we use tunnel routes, gateway could be not on-link. 750 * 751 * Attempt to reconcile all of these (alas, self-contradictory) conditions 752 * results in pretty ugly and hairy code with obscure logic. 753 * 754 * I chose to generalized it instead, so that the size 755 * of code does not increase practically, but it becomes 756 * much more general. 757 * Every prefix is assigned a "scope" value: "host" is local address, 758 * "link" is direct route, 759 * [ ... "site" ... "interior" ... ] 760 * and "universe" is true gateway route with global meaning. 761 * 762 * Every prefix refers to a set of "nexthop"s (gw, oif), 763 * where gw must have narrower scope. This recursion stops 764 * when gw has LOCAL scope or if "nexthop" is declared ONLINK, 765 * which means that gw is forced to be on link. 766 * 767 * Code is still hairy, but now it is apparently logically 768 * consistent and very flexible. F.e. as by-product it allows 769 * to co-exists in peace independent exterior and interior 770 * routing processes. 771 * 772 * Normally it looks as following. 773 * 774 * {universe prefix} -> (gw, oif) [scope link] 775 * | 776 * |-> {link prefix} -> (gw, oif) [scope local] 777 * | 778 * |-> {local prefix} (terminal node) 779 */ 780 static int fib_check_nh(struct fib_config *cfg, struct fib_nh *nh, 781 struct netlink_ext_ack *extack) 782 { 783 int err = 0; 784 struct net *net; 785 struct net_device *dev; 786 787 net = cfg->fc_nlinfo.nl_net; 788 if (nh->nh_gw) { 789 struct fib_result res; 790 791 if (nh->nh_flags & RTNH_F_ONLINK) { 792 unsigned int addr_type; 793 794 if (cfg->fc_scope >= RT_SCOPE_LINK) { 795 NL_SET_ERR_MSG(extack, 796 "Nexthop has invalid scope"); 797 return -EINVAL; 798 } 799 dev = __dev_get_by_index(net, nh->nh_oif); 800 if (!dev) { 801 NL_SET_ERR_MSG(extack, "Nexthop device required for onlink"); 802 return -ENODEV; 803 } 804 if (!(dev->flags & IFF_UP)) { 805 NL_SET_ERR_MSG(extack, 806 "Nexthop device is not up"); 807 return -ENETDOWN; 808 } 809 addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw); 810 if (addr_type != RTN_UNICAST) { 811 NL_SET_ERR_MSG(extack, 812 "Nexthop has invalid gateway"); 813 return -EINVAL; 814 } 815 if (!netif_carrier_ok(dev)) 816 nh->nh_flags |= RTNH_F_LINKDOWN; 817 nh->nh_dev = dev; 818 dev_hold(dev); 819 nh->nh_scope = RT_SCOPE_LINK; 820 return 0; 821 } 822 rcu_read_lock(); 823 { 824 struct fib_table *tbl = NULL; 825 struct flowi4 fl4 = { 826 .daddr = nh->nh_gw, 827 .flowi4_scope = cfg->fc_scope + 1, 828 .flowi4_oif = nh->nh_oif, 829 .flowi4_iif = LOOPBACK_IFINDEX, 830 }; 831 832 /* It is not necessary, but requires a bit of thinking */ 833 if (fl4.flowi4_scope < RT_SCOPE_LINK) 834 fl4.flowi4_scope = RT_SCOPE_LINK; 835 836 if (cfg->fc_table) 837 tbl = fib_get_table(net, cfg->fc_table); 838 839 if (tbl) 840 err = fib_table_lookup(tbl, &fl4, &res, 841 FIB_LOOKUP_IGNORE_LINKSTATE | 842 FIB_LOOKUP_NOREF); 843 844 /* on error or if no table given do full lookup. This 845 * is needed for example when nexthops are in the local 846 * table rather than the given table 847 */ 848 if (!tbl || err) { 849 err = fib_lookup(net, &fl4, &res, 850 FIB_LOOKUP_IGNORE_LINKSTATE); 851 } 852 853 if (err) { 854 NL_SET_ERR_MSG(extack, 855 "Nexthop has invalid gateway"); 856 rcu_read_unlock(); 857 return err; 858 } 859 } 860 err = -EINVAL; 861 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) { 862 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway"); 863 goto out; 864 } 865 nh->nh_scope = res.scope; 866 nh->nh_oif = FIB_RES_OIF(res); 867 nh->nh_dev = dev = FIB_RES_DEV(res); 868 if (!dev) { 869 NL_SET_ERR_MSG(extack, 870 "No egress device for nexthop gateway"); 871 goto out; 872 } 873 dev_hold(dev); 874 if (!netif_carrier_ok(dev)) 875 nh->nh_flags |= RTNH_F_LINKDOWN; 876 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN; 877 } else { 878 struct in_device *in_dev; 879 880 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) { 881 NL_SET_ERR_MSG(extack, 882 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set"); 883 return -EINVAL; 884 } 885 rcu_read_lock(); 886 err = -ENODEV; 887 in_dev = inetdev_by_index(net, nh->nh_oif); 888 if (!in_dev) 889 goto out; 890 err = -ENETDOWN; 891 if (!(in_dev->dev->flags & IFF_UP)) { 892 NL_SET_ERR_MSG(extack, "Device for nexthop is not up"); 893 goto out; 894 } 895 nh->nh_dev = in_dev->dev; 896 dev_hold(nh->nh_dev); 897 nh->nh_scope = RT_SCOPE_HOST; 898 if (!netif_carrier_ok(nh->nh_dev)) 899 nh->nh_flags |= RTNH_F_LINKDOWN; 900 err = 0; 901 } 902 out: 903 rcu_read_unlock(); 904 return err; 905 } 906 907 static inline unsigned int fib_laddr_hashfn(__be32 val) 908 { 909 unsigned int mask = (fib_info_hash_size - 1); 910 911 return ((__force u32)val ^ 912 ((__force u32)val >> 7) ^ 913 ((__force u32)val >> 14)) & mask; 914 } 915 916 static struct hlist_head *fib_info_hash_alloc(int bytes) 917 { 918 if (bytes <= PAGE_SIZE) 919 return kzalloc(bytes, GFP_KERNEL); 920 else 921 return (struct hlist_head *) 922 __get_free_pages(GFP_KERNEL | __GFP_ZERO, 923 get_order(bytes)); 924 } 925 926 static void fib_info_hash_free(struct hlist_head *hash, int bytes) 927 { 928 if (!hash) 929 return; 930 931 if (bytes <= PAGE_SIZE) 932 kfree(hash); 933 else 934 free_pages((unsigned long) hash, get_order(bytes)); 935 } 936 937 static void fib_info_hash_move(struct hlist_head *new_info_hash, 938 struct hlist_head *new_laddrhash, 939 unsigned int new_size) 940 { 941 struct hlist_head *old_info_hash, *old_laddrhash; 942 unsigned int old_size = fib_info_hash_size; 943 unsigned int i, bytes; 944 945 spin_lock_bh(&fib_info_lock); 946 old_info_hash = fib_info_hash; 947 old_laddrhash = fib_info_laddrhash; 948 fib_info_hash_size = new_size; 949 950 for (i = 0; i < old_size; i++) { 951 struct hlist_head *head = &fib_info_hash[i]; 952 struct hlist_node *n; 953 struct fib_info *fi; 954 955 hlist_for_each_entry_safe(fi, n, head, fib_hash) { 956 struct hlist_head *dest; 957 unsigned int new_hash; 958 959 new_hash = fib_info_hashfn(fi); 960 dest = &new_info_hash[new_hash]; 961 hlist_add_head(&fi->fib_hash, dest); 962 } 963 } 964 fib_info_hash = new_info_hash; 965 966 for (i = 0; i < old_size; i++) { 967 struct hlist_head *lhead = &fib_info_laddrhash[i]; 968 struct hlist_node *n; 969 struct fib_info *fi; 970 971 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) { 972 struct hlist_head *ldest; 973 unsigned int new_hash; 974 975 new_hash = fib_laddr_hashfn(fi->fib_prefsrc); 976 ldest = &new_laddrhash[new_hash]; 977 hlist_add_head(&fi->fib_lhash, ldest); 978 } 979 } 980 fib_info_laddrhash = new_laddrhash; 981 982 spin_unlock_bh(&fib_info_lock); 983 984 bytes = old_size * sizeof(struct hlist_head *); 985 fib_info_hash_free(old_info_hash, bytes); 986 fib_info_hash_free(old_laddrhash, bytes); 987 } 988 989 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh) 990 { 991 nh->nh_saddr = inet_select_addr(nh->nh_dev, 992 nh->nh_gw, 993 nh->nh_parent->fib_scope); 994 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid); 995 996 return nh->nh_saddr; 997 } 998 999 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc) 1000 { 1001 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst || 1002 fib_prefsrc != cfg->fc_dst) { 1003 u32 tb_id = cfg->fc_table; 1004 int rc; 1005 1006 if (tb_id == RT_TABLE_MAIN) 1007 tb_id = RT_TABLE_LOCAL; 1008 1009 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net, 1010 fib_prefsrc, tb_id); 1011 1012 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) { 1013 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net, 1014 fib_prefsrc, RT_TABLE_LOCAL); 1015 } 1016 1017 if (rc != RTN_LOCAL) 1018 return false; 1019 } 1020 return true; 1021 } 1022 1023 static int 1024 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg) 1025 { 1026 return ip_metrics_convert(fi->fib_net, cfg->fc_mx, cfg->fc_mx_len, 1027 fi->fib_metrics->metrics); 1028 } 1029 1030 struct fib_info *fib_create_info(struct fib_config *cfg, 1031 struct netlink_ext_ack *extack) 1032 { 1033 int err; 1034 struct fib_info *fi = NULL; 1035 struct fib_info *ofi; 1036 int nhs = 1; 1037 struct net *net = cfg->fc_nlinfo.nl_net; 1038 1039 if (cfg->fc_type > RTN_MAX) 1040 goto err_inval; 1041 1042 /* Fast check to catch the most weird cases */ 1043 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) { 1044 NL_SET_ERR_MSG(extack, "Invalid scope"); 1045 goto err_inval; 1046 } 1047 1048 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) { 1049 NL_SET_ERR_MSG(extack, 1050 "Invalid rtm_flags - can not contain DEAD or LINKDOWN"); 1051 goto err_inval; 1052 } 1053 1054 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1055 if (cfg->fc_mp) { 1056 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack); 1057 if (nhs == 0) 1058 goto err_inval; 1059 } 1060 #endif 1061 1062 err = -ENOBUFS; 1063 if (fib_info_cnt >= fib_info_hash_size) { 1064 unsigned int new_size = fib_info_hash_size << 1; 1065 struct hlist_head *new_info_hash; 1066 struct hlist_head *new_laddrhash; 1067 unsigned int bytes; 1068 1069 if (!new_size) 1070 new_size = 16; 1071 bytes = new_size * sizeof(struct hlist_head *); 1072 new_info_hash = fib_info_hash_alloc(bytes); 1073 new_laddrhash = fib_info_hash_alloc(bytes); 1074 if (!new_info_hash || !new_laddrhash) { 1075 fib_info_hash_free(new_info_hash, bytes); 1076 fib_info_hash_free(new_laddrhash, bytes); 1077 } else 1078 fib_info_hash_move(new_info_hash, new_laddrhash, new_size); 1079 1080 if (!fib_info_hash_size) 1081 goto failure; 1082 } 1083 1084 fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL); 1085 if (!fi) 1086 goto failure; 1087 if (cfg->fc_mx) { 1088 fi->fib_metrics = kzalloc(sizeof(*fi->fib_metrics), GFP_KERNEL); 1089 if (unlikely(!fi->fib_metrics)) { 1090 kfree(fi); 1091 return ERR_PTR(err); 1092 } 1093 refcount_set(&fi->fib_metrics->refcnt, 1); 1094 } else { 1095 fi->fib_metrics = (struct dst_metrics *)&dst_default_metrics; 1096 } 1097 fib_info_cnt++; 1098 fi->fib_net = net; 1099 fi->fib_protocol = cfg->fc_protocol; 1100 fi->fib_scope = cfg->fc_scope; 1101 fi->fib_flags = cfg->fc_flags; 1102 fi->fib_priority = cfg->fc_priority; 1103 fi->fib_prefsrc = cfg->fc_prefsrc; 1104 fi->fib_type = cfg->fc_type; 1105 fi->fib_tb_id = cfg->fc_table; 1106 1107 fi->fib_nhs = nhs; 1108 change_nexthops(fi) { 1109 nexthop_nh->nh_parent = fi; 1110 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *); 1111 if (!nexthop_nh->nh_pcpu_rth_output) 1112 goto failure; 1113 } endfor_nexthops(fi) 1114 1115 err = fib_convert_metrics(fi, cfg); 1116 if (err) 1117 goto failure; 1118 1119 if (cfg->fc_mp) { 1120 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1121 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack); 1122 if (err != 0) 1123 goto failure; 1124 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif) { 1125 NL_SET_ERR_MSG(extack, 1126 "Nexthop device index does not match RTA_OIF"); 1127 goto err_inval; 1128 } 1129 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw) { 1130 NL_SET_ERR_MSG(extack, 1131 "Nexthop gateway does not match RTA_GATEWAY"); 1132 goto err_inval; 1133 } 1134 #ifdef CONFIG_IP_ROUTE_CLASSID 1135 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) { 1136 NL_SET_ERR_MSG(extack, 1137 "Nexthop class id does not match RTA_FLOW"); 1138 goto err_inval; 1139 } 1140 #endif 1141 #else 1142 NL_SET_ERR_MSG(extack, 1143 "Multipath support not enabled in kernel"); 1144 goto err_inval; 1145 #endif 1146 } else { 1147 struct fib_nh *nh = fi->fib_nh; 1148 1149 if (cfg->fc_encap) { 1150 struct lwtunnel_state *lwtstate; 1151 1152 if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE) { 1153 NL_SET_ERR_MSG(extack, 1154 "LWT encap type not specified"); 1155 goto err_inval; 1156 } 1157 err = lwtunnel_build_state(cfg->fc_encap_type, 1158 cfg->fc_encap, AF_INET, cfg, 1159 &lwtstate, extack); 1160 if (err) 1161 goto failure; 1162 1163 nh->nh_lwtstate = lwtstate_get(lwtstate); 1164 } 1165 nh->nh_oif = cfg->fc_oif; 1166 nh->nh_gw = cfg->fc_gw; 1167 nh->nh_flags = cfg->fc_flags; 1168 #ifdef CONFIG_IP_ROUTE_CLASSID 1169 nh->nh_tclassid = cfg->fc_flow; 1170 if (nh->nh_tclassid) 1171 fi->fib_net->ipv4.fib_num_tclassid_users++; 1172 #endif 1173 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1174 nh->nh_weight = 1; 1175 #endif 1176 } 1177 1178 if (fib_props[cfg->fc_type].error) { 1179 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp) { 1180 NL_SET_ERR_MSG(extack, 1181 "Gateway, device and multipath can not be specified for this route type"); 1182 goto err_inval; 1183 } 1184 goto link_it; 1185 } else { 1186 switch (cfg->fc_type) { 1187 case RTN_UNICAST: 1188 case RTN_LOCAL: 1189 case RTN_BROADCAST: 1190 case RTN_ANYCAST: 1191 case RTN_MULTICAST: 1192 break; 1193 default: 1194 NL_SET_ERR_MSG(extack, "Invalid route type"); 1195 goto err_inval; 1196 } 1197 } 1198 1199 if (cfg->fc_scope > RT_SCOPE_HOST) { 1200 NL_SET_ERR_MSG(extack, "Invalid scope"); 1201 goto err_inval; 1202 } 1203 1204 if (cfg->fc_scope == RT_SCOPE_HOST) { 1205 struct fib_nh *nh = fi->fib_nh; 1206 1207 /* Local address is added. */ 1208 if (nhs != 1) { 1209 NL_SET_ERR_MSG(extack, 1210 "Route with host scope can not have multiple nexthops"); 1211 goto err_inval; 1212 } 1213 if (nh->nh_gw) { 1214 NL_SET_ERR_MSG(extack, 1215 "Route with host scope can not have a gateway"); 1216 goto err_inval; 1217 } 1218 nh->nh_scope = RT_SCOPE_NOWHERE; 1219 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif); 1220 err = -ENODEV; 1221 if (!nh->nh_dev) 1222 goto failure; 1223 } else { 1224 int linkdown = 0; 1225 1226 change_nexthops(fi) { 1227 err = fib_check_nh(cfg, nexthop_nh, extack); 1228 if (err != 0) 1229 goto failure; 1230 if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN) 1231 linkdown++; 1232 } endfor_nexthops(fi) 1233 if (linkdown == fi->fib_nhs) 1234 fi->fib_flags |= RTNH_F_LINKDOWN; 1235 } 1236 1237 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) { 1238 NL_SET_ERR_MSG(extack, "Invalid prefsrc address"); 1239 goto err_inval; 1240 } 1241 1242 change_nexthops(fi) { 1243 fib_info_update_nh_saddr(net, nexthop_nh); 1244 } endfor_nexthops(fi) 1245 1246 fib_rebalance(fi); 1247 1248 link_it: 1249 ofi = fib_find_info(fi); 1250 if (ofi) { 1251 fi->fib_dead = 1; 1252 free_fib_info(fi); 1253 ofi->fib_treeref++; 1254 return ofi; 1255 } 1256 1257 fi->fib_treeref++; 1258 refcount_set(&fi->fib_clntref, 1); 1259 spin_lock_bh(&fib_info_lock); 1260 hlist_add_head(&fi->fib_hash, 1261 &fib_info_hash[fib_info_hashfn(fi)]); 1262 if (fi->fib_prefsrc) { 1263 struct hlist_head *head; 1264 1265 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)]; 1266 hlist_add_head(&fi->fib_lhash, head); 1267 } 1268 change_nexthops(fi) { 1269 struct hlist_head *head; 1270 unsigned int hash; 1271 1272 if (!nexthop_nh->nh_dev) 1273 continue; 1274 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex); 1275 head = &fib_info_devhash[hash]; 1276 hlist_add_head(&nexthop_nh->nh_hash, head); 1277 } endfor_nexthops(fi) 1278 spin_unlock_bh(&fib_info_lock); 1279 return fi; 1280 1281 err_inval: 1282 err = -EINVAL; 1283 1284 failure: 1285 if (fi) { 1286 fi->fib_dead = 1; 1287 free_fib_info(fi); 1288 } 1289 1290 return ERR_PTR(err); 1291 } 1292 1293 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event, 1294 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos, 1295 struct fib_info *fi, unsigned int flags) 1296 { 1297 struct nlmsghdr *nlh; 1298 struct rtmsg *rtm; 1299 1300 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags); 1301 if (!nlh) 1302 return -EMSGSIZE; 1303 1304 rtm = nlmsg_data(nlh); 1305 rtm->rtm_family = AF_INET; 1306 rtm->rtm_dst_len = dst_len; 1307 rtm->rtm_src_len = 0; 1308 rtm->rtm_tos = tos; 1309 if (tb_id < 256) 1310 rtm->rtm_table = tb_id; 1311 else 1312 rtm->rtm_table = RT_TABLE_COMPAT; 1313 if (nla_put_u32(skb, RTA_TABLE, tb_id)) 1314 goto nla_put_failure; 1315 rtm->rtm_type = type; 1316 rtm->rtm_flags = fi->fib_flags; 1317 rtm->rtm_scope = fi->fib_scope; 1318 rtm->rtm_protocol = fi->fib_protocol; 1319 1320 if (rtm->rtm_dst_len && 1321 nla_put_in_addr(skb, RTA_DST, dst)) 1322 goto nla_put_failure; 1323 if (fi->fib_priority && 1324 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority)) 1325 goto nla_put_failure; 1326 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0) 1327 goto nla_put_failure; 1328 1329 if (fi->fib_prefsrc && 1330 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc)) 1331 goto nla_put_failure; 1332 if (fi->fib_nhs == 1) { 1333 if (fi->fib_nh->nh_gw && 1334 nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw)) 1335 goto nla_put_failure; 1336 if (fi->fib_nh->nh_oif && 1337 nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif)) 1338 goto nla_put_failure; 1339 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) { 1340 struct in_device *in_dev; 1341 1342 rcu_read_lock(); 1343 in_dev = __in_dev_get_rcu(fi->fib_nh->nh_dev); 1344 if (in_dev && 1345 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev)) 1346 rtm->rtm_flags |= RTNH_F_DEAD; 1347 rcu_read_unlock(); 1348 } 1349 if (fi->fib_nh->nh_flags & RTNH_F_OFFLOAD) 1350 rtm->rtm_flags |= RTNH_F_OFFLOAD; 1351 #ifdef CONFIG_IP_ROUTE_CLASSID 1352 if (fi->fib_nh[0].nh_tclassid && 1353 nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid)) 1354 goto nla_put_failure; 1355 #endif 1356 if (fi->fib_nh->nh_lwtstate && 1357 lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0) 1358 goto nla_put_failure; 1359 } 1360 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1361 if (fi->fib_nhs > 1) { 1362 struct rtnexthop *rtnh; 1363 struct nlattr *mp; 1364 1365 mp = nla_nest_start(skb, RTA_MULTIPATH); 1366 if (!mp) 1367 goto nla_put_failure; 1368 1369 for_nexthops(fi) { 1370 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh)); 1371 if (!rtnh) 1372 goto nla_put_failure; 1373 1374 rtnh->rtnh_flags = nh->nh_flags & 0xFF; 1375 if (nh->nh_flags & RTNH_F_LINKDOWN) { 1376 struct in_device *in_dev; 1377 1378 rcu_read_lock(); 1379 in_dev = __in_dev_get_rcu(nh->nh_dev); 1380 if (in_dev && 1381 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev)) 1382 rtnh->rtnh_flags |= RTNH_F_DEAD; 1383 rcu_read_unlock(); 1384 } 1385 rtnh->rtnh_hops = nh->nh_weight - 1; 1386 rtnh->rtnh_ifindex = nh->nh_oif; 1387 1388 if (nh->nh_gw && 1389 nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw)) 1390 goto nla_put_failure; 1391 #ifdef CONFIG_IP_ROUTE_CLASSID 1392 if (nh->nh_tclassid && 1393 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid)) 1394 goto nla_put_failure; 1395 #endif 1396 if (nh->nh_lwtstate && 1397 lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0) 1398 goto nla_put_failure; 1399 1400 /* length of rtnetlink header + attributes */ 1401 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh; 1402 } endfor_nexthops(fi); 1403 1404 nla_nest_end(skb, mp); 1405 } 1406 #endif 1407 nlmsg_end(skb, nlh); 1408 return 0; 1409 1410 nla_put_failure: 1411 nlmsg_cancel(skb, nlh); 1412 return -EMSGSIZE; 1413 } 1414 1415 /* 1416 * Update FIB if: 1417 * - local address disappeared -> we must delete all the entries 1418 * referring to it. 1419 * - device went down -> we must shutdown all nexthops going via it. 1420 */ 1421 int fib_sync_down_addr(struct net_device *dev, __be32 local) 1422 { 1423 int ret = 0; 1424 unsigned int hash = fib_laddr_hashfn(local); 1425 struct hlist_head *head = &fib_info_laddrhash[hash]; 1426 struct net *net = dev_net(dev); 1427 int tb_id = l3mdev_fib_table(dev); 1428 struct fib_info *fi; 1429 1430 if (!fib_info_laddrhash || local == 0) 1431 return 0; 1432 1433 hlist_for_each_entry(fi, head, fib_lhash) { 1434 if (!net_eq(fi->fib_net, net) || 1435 fi->fib_tb_id != tb_id) 1436 continue; 1437 if (fi->fib_prefsrc == local) { 1438 fi->fib_flags |= RTNH_F_DEAD; 1439 ret++; 1440 } 1441 } 1442 return ret; 1443 } 1444 1445 static int call_fib_nh_notifiers(struct fib_nh *fib_nh, 1446 enum fib_event_type event_type) 1447 { 1448 struct in_device *in_dev = __in_dev_get_rtnl(fib_nh->nh_dev); 1449 struct fib_nh_notifier_info info = { 1450 .fib_nh = fib_nh, 1451 }; 1452 1453 switch (event_type) { 1454 case FIB_EVENT_NH_ADD: 1455 if (fib_nh->nh_flags & RTNH_F_DEAD) 1456 break; 1457 if (IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) && 1458 fib_nh->nh_flags & RTNH_F_LINKDOWN) 1459 break; 1460 return call_fib4_notifiers(dev_net(fib_nh->nh_dev), event_type, 1461 &info.info); 1462 case FIB_EVENT_NH_DEL: 1463 if ((in_dev && IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) && 1464 fib_nh->nh_flags & RTNH_F_LINKDOWN) || 1465 (fib_nh->nh_flags & RTNH_F_DEAD)) 1466 return call_fib4_notifiers(dev_net(fib_nh->nh_dev), 1467 event_type, &info.info); 1468 default: 1469 break; 1470 } 1471 1472 return NOTIFY_DONE; 1473 } 1474 1475 /* Event force Flags Description 1476 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host 1477 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host 1478 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed 1479 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed 1480 */ 1481 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force) 1482 { 1483 int ret = 0; 1484 int scope = RT_SCOPE_NOWHERE; 1485 struct fib_info *prev_fi = NULL; 1486 unsigned int hash = fib_devindex_hashfn(dev->ifindex); 1487 struct hlist_head *head = &fib_info_devhash[hash]; 1488 struct fib_nh *nh; 1489 1490 if (force) 1491 scope = -1; 1492 1493 hlist_for_each_entry(nh, head, nh_hash) { 1494 struct fib_info *fi = nh->nh_parent; 1495 int dead; 1496 1497 BUG_ON(!fi->fib_nhs); 1498 if (nh->nh_dev != dev || fi == prev_fi) 1499 continue; 1500 prev_fi = fi; 1501 dead = 0; 1502 change_nexthops(fi) { 1503 if (nexthop_nh->nh_flags & RTNH_F_DEAD) 1504 dead++; 1505 else if (nexthop_nh->nh_dev == dev && 1506 nexthop_nh->nh_scope != scope) { 1507 switch (event) { 1508 case NETDEV_DOWN: 1509 case NETDEV_UNREGISTER: 1510 nexthop_nh->nh_flags |= RTNH_F_DEAD; 1511 /* fall through */ 1512 case NETDEV_CHANGE: 1513 nexthop_nh->nh_flags |= RTNH_F_LINKDOWN; 1514 break; 1515 } 1516 call_fib_nh_notifiers(nexthop_nh, 1517 FIB_EVENT_NH_DEL); 1518 dead++; 1519 } 1520 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1521 if (event == NETDEV_UNREGISTER && 1522 nexthop_nh->nh_dev == dev) { 1523 dead = fi->fib_nhs; 1524 break; 1525 } 1526 #endif 1527 } endfor_nexthops(fi) 1528 if (dead == fi->fib_nhs) { 1529 switch (event) { 1530 case NETDEV_DOWN: 1531 case NETDEV_UNREGISTER: 1532 fi->fib_flags |= RTNH_F_DEAD; 1533 /* fall through */ 1534 case NETDEV_CHANGE: 1535 fi->fib_flags |= RTNH_F_LINKDOWN; 1536 break; 1537 } 1538 ret++; 1539 } 1540 1541 fib_rebalance(fi); 1542 } 1543 1544 return ret; 1545 } 1546 1547 /* Must be invoked inside of an RCU protected region. */ 1548 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res) 1549 { 1550 struct fib_info *fi = NULL, *last_resort = NULL; 1551 struct hlist_head *fa_head = res->fa_head; 1552 struct fib_table *tb = res->table; 1553 u8 slen = 32 - res->prefixlen; 1554 int order = -1, last_idx = -1; 1555 struct fib_alias *fa, *fa1 = NULL; 1556 u32 last_prio = res->fi->fib_priority; 1557 u8 last_tos = 0; 1558 1559 hlist_for_each_entry_rcu(fa, fa_head, fa_list) { 1560 struct fib_info *next_fi = fa->fa_info; 1561 1562 if (fa->fa_slen != slen) 1563 continue; 1564 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos) 1565 continue; 1566 if (fa->tb_id != tb->tb_id) 1567 continue; 1568 if (next_fi->fib_priority > last_prio && 1569 fa->fa_tos == last_tos) { 1570 if (last_tos) 1571 continue; 1572 break; 1573 } 1574 if (next_fi->fib_flags & RTNH_F_DEAD) 1575 continue; 1576 last_tos = fa->fa_tos; 1577 last_prio = next_fi->fib_priority; 1578 1579 if (next_fi->fib_scope != res->scope || 1580 fa->fa_type != RTN_UNICAST) 1581 continue; 1582 if (!next_fi->fib_nh[0].nh_gw || 1583 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK) 1584 continue; 1585 1586 fib_alias_accessed(fa); 1587 1588 if (!fi) { 1589 if (next_fi != res->fi) 1590 break; 1591 fa1 = fa; 1592 } else if (!fib_detect_death(fi, order, &last_resort, 1593 &last_idx, fa1->fa_default)) { 1594 fib_result_assign(res, fi); 1595 fa1->fa_default = order; 1596 goto out; 1597 } 1598 fi = next_fi; 1599 order++; 1600 } 1601 1602 if (order <= 0 || !fi) { 1603 if (fa1) 1604 fa1->fa_default = -1; 1605 goto out; 1606 } 1607 1608 if (!fib_detect_death(fi, order, &last_resort, &last_idx, 1609 fa1->fa_default)) { 1610 fib_result_assign(res, fi); 1611 fa1->fa_default = order; 1612 goto out; 1613 } 1614 1615 if (last_idx >= 0) 1616 fib_result_assign(res, last_resort); 1617 fa1->fa_default = last_idx; 1618 out: 1619 return; 1620 } 1621 1622 /* 1623 * Dead device goes up. We wake up dead nexthops. 1624 * It takes sense only on multipath routes. 1625 */ 1626 int fib_sync_up(struct net_device *dev, unsigned int nh_flags) 1627 { 1628 struct fib_info *prev_fi; 1629 unsigned int hash; 1630 struct hlist_head *head; 1631 struct fib_nh *nh; 1632 int ret; 1633 1634 if (!(dev->flags & IFF_UP)) 1635 return 0; 1636 1637 if (nh_flags & RTNH_F_DEAD) { 1638 unsigned int flags = dev_get_flags(dev); 1639 1640 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 1641 nh_flags |= RTNH_F_LINKDOWN; 1642 } 1643 1644 prev_fi = NULL; 1645 hash = fib_devindex_hashfn(dev->ifindex); 1646 head = &fib_info_devhash[hash]; 1647 ret = 0; 1648 1649 hlist_for_each_entry(nh, head, nh_hash) { 1650 struct fib_info *fi = nh->nh_parent; 1651 int alive; 1652 1653 BUG_ON(!fi->fib_nhs); 1654 if (nh->nh_dev != dev || fi == prev_fi) 1655 continue; 1656 1657 prev_fi = fi; 1658 alive = 0; 1659 change_nexthops(fi) { 1660 if (!(nexthop_nh->nh_flags & nh_flags)) { 1661 alive++; 1662 continue; 1663 } 1664 if (!nexthop_nh->nh_dev || 1665 !(nexthop_nh->nh_dev->flags & IFF_UP)) 1666 continue; 1667 if (nexthop_nh->nh_dev != dev || 1668 !__in_dev_get_rtnl(dev)) 1669 continue; 1670 alive++; 1671 nexthop_nh->nh_flags &= ~nh_flags; 1672 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD); 1673 } endfor_nexthops(fi) 1674 1675 if (alive > 0) { 1676 fi->fib_flags &= ~nh_flags; 1677 ret++; 1678 } 1679 1680 fib_rebalance(fi); 1681 } 1682 1683 return ret; 1684 } 1685 1686 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1687 static bool fib_good_nh(const struct fib_nh *nh) 1688 { 1689 int state = NUD_REACHABLE; 1690 1691 if (nh->nh_scope == RT_SCOPE_LINK) { 1692 struct neighbour *n; 1693 1694 rcu_read_lock_bh(); 1695 1696 n = __ipv4_neigh_lookup_noref(nh->nh_dev, 1697 (__force u32)nh->nh_gw); 1698 if (n) 1699 state = n->nud_state; 1700 1701 rcu_read_unlock_bh(); 1702 } 1703 1704 return !!(state & NUD_VALID); 1705 } 1706 1707 void fib_select_multipath(struct fib_result *res, int hash) 1708 { 1709 struct fib_info *fi = res->fi; 1710 struct net *net = fi->fib_net; 1711 bool first = false; 1712 1713 for_nexthops(fi) { 1714 if (net->ipv4.sysctl_fib_multipath_use_neigh) { 1715 if (!fib_good_nh(nh)) 1716 continue; 1717 if (!first) { 1718 res->nh_sel = nhsel; 1719 first = true; 1720 } 1721 } 1722 1723 if (hash > atomic_read(&nh->nh_upper_bound)) 1724 continue; 1725 1726 res->nh_sel = nhsel; 1727 return; 1728 } endfor_nexthops(fi); 1729 } 1730 #endif 1731 1732 void fib_select_path(struct net *net, struct fib_result *res, 1733 struct flowi4 *fl4, const struct sk_buff *skb) 1734 { 1735 if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) 1736 goto check_saddr; 1737 1738 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1739 if (res->fi->fib_nhs > 1) { 1740 int h = fib_multipath_hash(net, fl4, skb, NULL); 1741 1742 fib_select_multipath(res, h); 1743 } 1744 else 1745 #endif 1746 if (!res->prefixlen && 1747 res->table->tb_num_default > 1 && 1748 res->type == RTN_UNICAST) 1749 fib_select_default(fl4, res); 1750 1751 check_saddr: 1752 if (!fl4->saddr) 1753 fl4->saddr = FIB_RES_PREFSRC(net, *res); 1754 } 1755