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 #include <net/addrconf.h> 49 50 #include "fib_lookup.h" 51 52 static DEFINE_SPINLOCK(fib_info_lock); 53 static struct hlist_head *fib_info_hash; 54 static struct hlist_head *fib_info_laddrhash; 55 static unsigned int fib_info_hash_size; 56 static unsigned int fib_info_cnt; 57 58 #define DEVINDEX_HASHBITS 8 59 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS) 60 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE]; 61 62 #ifdef CONFIG_IP_ROUTE_MULTIPATH 63 64 #define for_nexthops(fi) { \ 65 int nhsel; const struct fib_nh *nh; \ 66 for (nhsel = 0, nh = (fi)->fib_nh; \ 67 nhsel < (fi)->fib_nhs; \ 68 nh++, nhsel++) 69 70 #define change_nexthops(fi) { \ 71 int nhsel; struct fib_nh *nexthop_nh; \ 72 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \ 73 nhsel < (fi)->fib_nhs; \ 74 nexthop_nh++, nhsel++) 75 76 #else /* CONFIG_IP_ROUTE_MULTIPATH */ 77 78 /* Hope, that gcc will optimize it to get rid of dummy loop */ 79 80 #define for_nexthops(fi) { \ 81 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \ 82 for (nhsel = 0; nhsel < 1; nhsel++) 83 84 #define change_nexthops(fi) { \ 85 int nhsel; \ 86 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \ 87 for (nhsel = 0; nhsel < 1; nhsel++) 88 89 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 90 91 #define endfor_nexthops(fi) } 92 93 94 const struct fib_prop fib_props[RTN_MAX + 1] = { 95 [RTN_UNSPEC] = { 96 .error = 0, 97 .scope = RT_SCOPE_NOWHERE, 98 }, 99 [RTN_UNICAST] = { 100 .error = 0, 101 .scope = RT_SCOPE_UNIVERSE, 102 }, 103 [RTN_LOCAL] = { 104 .error = 0, 105 .scope = RT_SCOPE_HOST, 106 }, 107 [RTN_BROADCAST] = { 108 .error = 0, 109 .scope = RT_SCOPE_LINK, 110 }, 111 [RTN_ANYCAST] = { 112 .error = 0, 113 .scope = RT_SCOPE_LINK, 114 }, 115 [RTN_MULTICAST] = { 116 .error = 0, 117 .scope = RT_SCOPE_UNIVERSE, 118 }, 119 [RTN_BLACKHOLE] = { 120 .error = -EINVAL, 121 .scope = RT_SCOPE_UNIVERSE, 122 }, 123 [RTN_UNREACHABLE] = { 124 .error = -EHOSTUNREACH, 125 .scope = RT_SCOPE_UNIVERSE, 126 }, 127 [RTN_PROHIBIT] = { 128 .error = -EACCES, 129 .scope = RT_SCOPE_UNIVERSE, 130 }, 131 [RTN_THROW] = { 132 .error = -EAGAIN, 133 .scope = RT_SCOPE_UNIVERSE, 134 }, 135 [RTN_NAT] = { 136 .error = -EINVAL, 137 .scope = RT_SCOPE_NOWHERE, 138 }, 139 [RTN_XRESOLVE] = { 140 .error = -EINVAL, 141 .scope = RT_SCOPE_NOWHERE, 142 }, 143 }; 144 145 static void rt_fibinfo_free(struct rtable __rcu **rtp) 146 { 147 struct rtable *rt = rcu_dereference_protected(*rtp, 1); 148 149 if (!rt) 150 return; 151 152 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL); 153 * because we waited an RCU grace period before calling 154 * free_fib_info_rcu() 155 */ 156 157 dst_dev_put(&rt->dst); 158 dst_release_immediate(&rt->dst); 159 } 160 161 static void free_nh_exceptions(struct fib_nh *nh) 162 { 163 struct fnhe_hash_bucket *hash; 164 int i; 165 166 hash = rcu_dereference_protected(nh->nh_exceptions, 1); 167 if (!hash) 168 return; 169 for (i = 0; i < FNHE_HASH_SIZE; i++) { 170 struct fib_nh_exception *fnhe; 171 172 fnhe = rcu_dereference_protected(hash[i].chain, 1); 173 while (fnhe) { 174 struct fib_nh_exception *next; 175 176 next = rcu_dereference_protected(fnhe->fnhe_next, 1); 177 178 rt_fibinfo_free(&fnhe->fnhe_rth_input); 179 rt_fibinfo_free(&fnhe->fnhe_rth_output); 180 181 kfree(fnhe); 182 183 fnhe = next; 184 } 185 } 186 kfree(hash); 187 } 188 189 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp) 190 { 191 int cpu; 192 193 if (!rtp) 194 return; 195 196 for_each_possible_cpu(cpu) { 197 struct rtable *rt; 198 199 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1); 200 if (rt) { 201 dst_dev_put(&rt->dst); 202 dst_release_immediate(&rt->dst); 203 } 204 } 205 free_percpu(rtp); 206 } 207 208 void fib_nh_common_release(struct fib_nh_common *nhc) 209 { 210 if (nhc->nhc_dev) 211 dev_put(nhc->nhc_dev); 212 213 lwtstate_put(nhc->nhc_lwtstate); 214 } 215 EXPORT_SYMBOL_GPL(fib_nh_common_release); 216 217 void fib_nh_release(struct net *net, struct fib_nh *fib_nh) 218 { 219 #ifdef CONFIG_IP_ROUTE_CLASSID 220 if (fib_nh->nh_tclassid) 221 net->ipv4.fib_num_tclassid_users--; 222 #endif 223 fib_nh_common_release(&fib_nh->nh_common); 224 free_nh_exceptions(fib_nh); 225 rt_fibinfo_free_cpus(fib_nh->nh_pcpu_rth_output); 226 rt_fibinfo_free(&fib_nh->nh_rth_input); 227 } 228 229 /* Release a nexthop info record */ 230 static void free_fib_info_rcu(struct rcu_head *head) 231 { 232 struct fib_info *fi = container_of(head, struct fib_info, rcu); 233 234 change_nexthops(fi) { 235 fib_nh_release(fi->fib_net, nexthop_nh); 236 } endfor_nexthops(fi); 237 238 ip_fib_metrics_put(fi->fib_metrics); 239 240 kfree(fi); 241 } 242 243 void free_fib_info(struct fib_info *fi) 244 { 245 if (fi->fib_dead == 0) { 246 pr_warn("Freeing alive fib_info %p\n", fi); 247 return; 248 } 249 fib_info_cnt--; 250 251 call_rcu(&fi->rcu, free_fib_info_rcu); 252 } 253 EXPORT_SYMBOL_GPL(free_fib_info); 254 255 void fib_release_info(struct fib_info *fi) 256 { 257 spin_lock_bh(&fib_info_lock); 258 if (fi && --fi->fib_treeref == 0) { 259 hlist_del(&fi->fib_hash); 260 if (fi->fib_prefsrc) 261 hlist_del(&fi->fib_lhash); 262 change_nexthops(fi) { 263 if (!nexthop_nh->fib_nh_dev) 264 continue; 265 hlist_del(&nexthop_nh->nh_hash); 266 } endfor_nexthops(fi) 267 fi->fib_dead = 1; 268 fib_info_put(fi); 269 } 270 spin_unlock_bh(&fib_info_lock); 271 } 272 273 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi) 274 { 275 const struct fib_nh *onh = ofi->fib_nh; 276 277 for_nexthops(fi) { 278 if (nh->fib_nh_oif != onh->fib_nh_oif || 279 nh->fib_nh_gw4 != onh->fib_nh_gw4 || 280 nh->fib_nh_scope != onh->fib_nh_scope || 281 #ifdef CONFIG_IP_ROUTE_MULTIPATH 282 nh->fib_nh_weight != onh->fib_nh_weight || 283 #endif 284 #ifdef CONFIG_IP_ROUTE_CLASSID 285 nh->nh_tclassid != onh->nh_tclassid || 286 #endif 287 lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) || 288 ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK)) 289 return -1; 290 onh++; 291 } endfor_nexthops(fi); 292 return 0; 293 } 294 295 static inline unsigned int fib_devindex_hashfn(unsigned int val) 296 { 297 unsigned int mask = DEVINDEX_HASHSIZE - 1; 298 299 return (val ^ 300 (val >> DEVINDEX_HASHBITS) ^ 301 (val >> (DEVINDEX_HASHBITS * 2))) & mask; 302 } 303 304 static inline unsigned int fib_info_hashfn(const struct fib_info *fi) 305 { 306 unsigned int mask = (fib_info_hash_size - 1); 307 unsigned int val = fi->fib_nhs; 308 309 val ^= (fi->fib_protocol << 8) | fi->fib_scope; 310 val ^= (__force u32)fi->fib_prefsrc; 311 val ^= fi->fib_priority; 312 for_nexthops(fi) { 313 val ^= fib_devindex_hashfn(nh->fib_nh_oif); 314 } endfor_nexthops(fi) 315 316 return (val ^ (val >> 7) ^ (val >> 12)) & mask; 317 } 318 319 static struct fib_info *fib_find_info(const struct fib_info *nfi) 320 { 321 struct hlist_head *head; 322 struct fib_info *fi; 323 unsigned int hash; 324 325 hash = fib_info_hashfn(nfi); 326 head = &fib_info_hash[hash]; 327 328 hlist_for_each_entry(fi, head, fib_hash) { 329 if (!net_eq(fi->fib_net, nfi->fib_net)) 330 continue; 331 if (fi->fib_nhs != nfi->fib_nhs) 332 continue; 333 if (nfi->fib_protocol == fi->fib_protocol && 334 nfi->fib_scope == fi->fib_scope && 335 nfi->fib_prefsrc == fi->fib_prefsrc && 336 nfi->fib_priority == fi->fib_priority && 337 nfi->fib_type == fi->fib_type && 338 memcmp(nfi->fib_metrics, fi->fib_metrics, 339 sizeof(u32) * RTAX_MAX) == 0 && 340 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) && 341 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0)) 342 return fi; 343 } 344 345 return NULL; 346 } 347 348 /* Check, that the gateway is already configured. 349 * Used only by redirect accept routine. 350 */ 351 int ip_fib_check_default(__be32 gw, struct net_device *dev) 352 { 353 struct hlist_head *head; 354 struct fib_nh *nh; 355 unsigned int hash; 356 357 spin_lock(&fib_info_lock); 358 359 hash = fib_devindex_hashfn(dev->ifindex); 360 head = &fib_info_devhash[hash]; 361 hlist_for_each_entry(nh, head, nh_hash) { 362 if (nh->fib_nh_dev == dev && 363 nh->fib_nh_gw4 == gw && 364 !(nh->fib_nh_flags & RTNH_F_DEAD)) { 365 spin_unlock(&fib_info_lock); 366 return 0; 367 } 368 } 369 370 spin_unlock(&fib_info_lock); 371 372 return -1; 373 } 374 375 static inline size_t fib_nlmsg_size(struct fib_info *fi) 376 { 377 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg)) 378 + nla_total_size(4) /* RTA_TABLE */ 379 + nla_total_size(4) /* RTA_DST */ 380 + nla_total_size(4) /* RTA_PRIORITY */ 381 + nla_total_size(4) /* RTA_PREFSRC */ 382 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */ 383 384 /* space for nested metrics */ 385 payload += nla_total_size((RTAX_MAX * nla_total_size(4))); 386 387 if (fi->fib_nhs) { 388 size_t nh_encapsize = 0; 389 /* Also handles the special case fib_nhs == 1 */ 390 391 /* each nexthop is packed in an attribute */ 392 size_t nhsize = nla_total_size(sizeof(struct rtnexthop)); 393 394 /* may contain flow and gateway attribute */ 395 nhsize += 2 * nla_total_size(4); 396 397 /* grab encap info */ 398 for_nexthops(fi) { 399 if (nh->fib_nh_lws) { 400 /* RTA_ENCAP_TYPE */ 401 nh_encapsize += lwtunnel_get_encap_size( 402 nh->fib_nh_lws); 403 /* RTA_ENCAP */ 404 nh_encapsize += nla_total_size(2); 405 } 406 } endfor_nexthops(fi); 407 408 /* all nexthops are packed in a nested attribute */ 409 payload += nla_total_size((fi->fib_nhs * nhsize) + 410 nh_encapsize); 411 412 } 413 414 return payload; 415 } 416 417 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa, 418 int dst_len, u32 tb_id, const struct nl_info *info, 419 unsigned int nlm_flags) 420 { 421 struct sk_buff *skb; 422 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0; 423 int err = -ENOBUFS; 424 425 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL); 426 if (!skb) 427 goto errout; 428 429 err = fib_dump_info(skb, info->portid, seq, event, tb_id, 430 fa->fa_type, key, dst_len, 431 fa->fa_tos, fa->fa_info, nlm_flags); 432 if (err < 0) { 433 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */ 434 WARN_ON(err == -EMSGSIZE); 435 kfree_skb(skb); 436 goto errout; 437 } 438 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE, 439 info->nlh, GFP_KERNEL); 440 return; 441 errout: 442 if (err < 0) 443 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err); 444 } 445 446 static int fib_detect_death(struct fib_info *fi, int order, 447 struct fib_info **last_resort, int *last_idx, 448 int dflt) 449 { 450 struct neighbour *n; 451 int state = NUD_NONE; 452 453 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].fib_nh_gw4, fi->fib_dev); 454 if (n) { 455 state = n->nud_state; 456 neigh_release(n); 457 } else { 458 return 0; 459 } 460 if (state == NUD_REACHABLE) 461 return 0; 462 if ((state & NUD_VALID) && order != dflt) 463 return 0; 464 if ((state & NUD_VALID) || 465 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) { 466 *last_resort = fi; 467 *last_idx = order; 468 } 469 return 1; 470 } 471 472 int fib_nh_common_init(struct fib_nh_common *nhc, struct nlattr *encap, 473 u16 encap_type, void *cfg, gfp_t gfp_flags, 474 struct netlink_ext_ack *extack) 475 { 476 if (encap) { 477 struct lwtunnel_state *lwtstate; 478 int err; 479 480 if (encap_type == LWTUNNEL_ENCAP_NONE) { 481 NL_SET_ERR_MSG(extack, "LWT encap type not specified"); 482 return -EINVAL; 483 } 484 err = lwtunnel_build_state(encap_type, encap, nhc->nhc_family, 485 cfg, &lwtstate, extack); 486 if (err) 487 return err; 488 489 nhc->nhc_lwtstate = lwtstate_get(lwtstate); 490 } 491 492 return 0; 493 } 494 EXPORT_SYMBOL_GPL(fib_nh_common_init); 495 496 int fib_nh_init(struct net *net, struct fib_nh *nh, 497 struct fib_config *cfg, int nh_weight, 498 struct netlink_ext_ack *extack) 499 { 500 int err = -ENOMEM; 501 502 nh->fib_nh_family = AF_INET; 503 504 nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *); 505 if (!nh->nh_pcpu_rth_output) 506 goto err_out; 507 508 err = fib_nh_common_init(&nh->nh_common, cfg->fc_encap, 509 cfg->fc_encap_type, cfg, GFP_KERNEL, extack); 510 if (err) 511 goto init_failure; 512 513 nh->fib_nh_oif = cfg->fc_oif; 514 if (cfg->fc_gw) { 515 nh->fib_nh_gw4 = cfg->fc_gw; 516 nh->fib_nh_has_gw = 1; 517 } 518 nh->fib_nh_flags = cfg->fc_flags; 519 520 #ifdef CONFIG_IP_ROUTE_CLASSID 521 nh->nh_tclassid = cfg->fc_flow; 522 if (nh->nh_tclassid) 523 net->ipv4.fib_num_tclassid_users++; 524 #endif 525 #ifdef CONFIG_IP_ROUTE_MULTIPATH 526 nh->fib_nh_weight = nh_weight; 527 #endif 528 return 0; 529 530 init_failure: 531 rt_fibinfo_free_cpus(nh->nh_pcpu_rth_output); 532 nh->nh_pcpu_rth_output = NULL; 533 err_out: 534 return err; 535 } 536 537 #ifdef CONFIG_IP_ROUTE_MULTIPATH 538 539 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining, 540 struct netlink_ext_ack *extack) 541 { 542 int nhs = 0; 543 544 while (rtnh_ok(rtnh, remaining)) { 545 nhs++; 546 rtnh = rtnh_next(rtnh, &remaining); 547 } 548 549 /* leftover implies invalid nexthop configuration, discard it */ 550 if (remaining > 0) { 551 NL_SET_ERR_MSG(extack, 552 "Invalid nexthop configuration - extra data after nexthops"); 553 nhs = 0; 554 } 555 556 return nhs; 557 } 558 559 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh, 560 int remaining, struct fib_config *cfg, 561 struct netlink_ext_ack *extack) 562 { 563 struct net *net = fi->fib_net; 564 struct fib_config fib_cfg; 565 int ret; 566 567 change_nexthops(fi) { 568 int attrlen; 569 570 memset(&fib_cfg, 0, sizeof(fib_cfg)); 571 572 if (!rtnh_ok(rtnh, remaining)) { 573 NL_SET_ERR_MSG(extack, 574 "Invalid nexthop configuration - extra data after nexthop"); 575 return -EINVAL; 576 } 577 578 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) { 579 NL_SET_ERR_MSG(extack, 580 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN"); 581 return -EINVAL; 582 } 583 584 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags; 585 fib_cfg.fc_oif = rtnh->rtnh_ifindex; 586 587 attrlen = rtnh_attrlen(rtnh); 588 if (attrlen > 0) { 589 struct nlattr *nla, *attrs = rtnh_attrs(rtnh); 590 591 nla = nla_find(attrs, attrlen, RTA_GATEWAY); 592 if (nla) 593 fib_cfg.fc_gw = nla_get_in_addr(nla); 594 595 nla = nla_find(attrs, attrlen, RTA_FLOW); 596 if (nla) 597 fib_cfg.fc_flow = nla_get_u32(nla); 598 599 fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP); 600 nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE); 601 if (nla) 602 fib_cfg.fc_encap_type = nla_get_u16(nla); 603 } 604 605 ret = fib_nh_init(net, nexthop_nh, &fib_cfg, 606 rtnh->rtnh_hops + 1, extack); 607 if (ret) 608 goto errout; 609 610 rtnh = rtnh_next(rtnh, &remaining); 611 } endfor_nexthops(fi); 612 613 ret = -EINVAL; 614 if (cfg->fc_oif && fi->fib_nh->fib_nh_oif != cfg->fc_oif) { 615 NL_SET_ERR_MSG(extack, 616 "Nexthop device index does not match RTA_OIF"); 617 goto errout; 618 } 619 if (cfg->fc_gw && fi->fib_nh->fib_nh_gw4 != cfg->fc_gw) { 620 NL_SET_ERR_MSG(extack, 621 "Nexthop gateway does not match RTA_GATEWAY"); 622 goto errout; 623 } 624 #ifdef CONFIG_IP_ROUTE_CLASSID 625 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) { 626 NL_SET_ERR_MSG(extack, 627 "Nexthop class id does not match RTA_FLOW"); 628 goto errout; 629 } 630 #endif 631 ret = 0; 632 errout: 633 return ret; 634 } 635 636 static void fib_rebalance(struct fib_info *fi) 637 { 638 int total; 639 int w; 640 641 if (fi->fib_nhs < 2) 642 return; 643 644 total = 0; 645 for_nexthops(fi) { 646 if (nh->fib_nh_flags & RTNH_F_DEAD) 647 continue; 648 649 if (ip_ignore_linkdown(nh->fib_nh_dev) && 650 nh->fib_nh_flags & RTNH_F_LINKDOWN) 651 continue; 652 653 total += nh->fib_nh_weight; 654 } endfor_nexthops(fi); 655 656 w = 0; 657 change_nexthops(fi) { 658 int upper_bound; 659 660 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) { 661 upper_bound = -1; 662 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) && 663 nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) { 664 upper_bound = -1; 665 } else { 666 w += nexthop_nh->fib_nh_weight; 667 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31, 668 total) - 1; 669 } 670 671 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound); 672 } endfor_nexthops(fi); 673 } 674 #else /* CONFIG_IP_ROUTE_MULTIPATH */ 675 676 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh, 677 int remaining, struct fib_config *cfg, 678 struct netlink_ext_ack *extack) 679 { 680 NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel"); 681 682 return -EINVAL; 683 } 684 685 #define fib_rebalance(fi) do { } while (0) 686 687 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 688 689 static int fib_encap_match(u16 encap_type, 690 struct nlattr *encap, 691 const struct fib_nh *nh, 692 const struct fib_config *cfg, 693 struct netlink_ext_ack *extack) 694 { 695 struct lwtunnel_state *lwtstate; 696 int ret, result = 0; 697 698 if (encap_type == LWTUNNEL_ENCAP_NONE) 699 return 0; 700 701 ret = lwtunnel_build_state(encap_type, encap, AF_INET, 702 cfg, &lwtstate, extack); 703 if (!ret) { 704 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws); 705 lwtstate_free(lwtstate); 706 } 707 708 return result; 709 } 710 711 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi, 712 struct netlink_ext_ack *extack) 713 { 714 #ifdef CONFIG_IP_ROUTE_MULTIPATH 715 struct rtnexthop *rtnh; 716 int remaining; 717 #endif 718 719 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority) 720 return 1; 721 722 if (cfg->fc_oif || cfg->fc_gw) { 723 if (cfg->fc_encap) { 724 if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap, 725 fi->fib_nh, cfg, extack)) 726 return 1; 727 } 728 #ifdef CONFIG_IP_ROUTE_CLASSID 729 if (cfg->fc_flow && 730 cfg->fc_flow != fi->fib_nh->nh_tclassid) 731 return 1; 732 #endif 733 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->fib_nh_oif) && 734 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->fib_nh_gw4)) 735 return 0; 736 return 1; 737 } 738 739 #ifdef CONFIG_IP_ROUTE_MULTIPATH 740 if (!cfg->fc_mp) 741 return 0; 742 743 rtnh = cfg->fc_mp; 744 remaining = cfg->fc_mp_len; 745 746 for_nexthops(fi) { 747 int attrlen; 748 749 if (!rtnh_ok(rtnh, remaining)) 750 return -EINVAL; 751 752 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif) 753 return 1; 754 755 attrlen = rtnh_attrlen(rtnh); 756 if (attrlen > 0) { 757 struct nlattr *nla, *attrs = rtnh_attrs(rtnh); 758 759 nla = nla_find(attrs, attrlen, RTA_GATEWAY); 760 if (nla && nla_get_in_addr(nla) != nh->fib_nh_gw4) 761 return 1; 762 #ifdef CONFIG_IP_ROUTE_CLASSID 763 nla = nla_find(attrs, attrlen, RTA_FLOW); 764 if (nla && nla_get_u32(nla) != nh->nh_tclassid) 765 return 1; 766 #endif 767 } 768 769 rtnh = rtnh_next(rtnh, &remaining); 770 } endfor_nexthops(fi); 771 #endif 772 return 0; 773 } 774 775 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi) 776 { 777 struct nlattr *nla; 778 int remaining; 779 780 if (!cfg->fc_mx) 781 return true; 782 783 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) { 784 int type = nla_type(nla); 785 u32 fi_val, val; 786 787 if (!type) 788 continue; 789 if (type > RTAX_MAX) 790 return false; 791 792 if (type == RTAX_CC_ALGO) { 793 char tmp[TCP_CA_NAME_MAX]; 794 bool ecn_ca = false; 795 796 nla_strlcpy(tmp, nla, sizeof(tmp)); 797 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca); 798 } else { 799 if (nla_len(nla) != sizeof(u32)) 800 return false; 801 val = nla_get_u32(nla); 802 } 803 804 fi_val = fi->fib_metrics->metrics[type - 1]; 805 if (type == RTAX_FEATURES) 806 fi_val &= ~DST_FEATURE_ECN_CA; 807 808 if (fi_val != val) 809 return false; 810 } 811 812 return true; 813 } 814 815 816 /* 817 * Picture 818 * ------- 819 * 820 * Semantics of nexthop is very messy by historical reasons. 821 * We have to take into account, that: 822 * a) gateway can be actually local interface address, 823 * so that gatewayed route is direct. 824 * b) gateway must be on-link address, possibly 825 * described not by an ifaddr, but also by a direct route. 826 * c) If both gateway and interface are specified, they should not 827 * contradict. 828 * d) If we use tunnel routes, gateway could be not on-link. 829 * 830 * Attempt to reconcile all of these (alas, self-contradictory) conditions 831 * results in pretty ugly and hairy code with obscure logic. 832 * 833 * I chose to generalized it instead, so that the size 834 * of code does not increase practically, but it becomes 835 * much more general. 836 * Every prefix is assigned a "scope" value: "host" is local address, 837 * "link" is direct route, 838 * [ ... "site" ... "interior" ... ] 839 * and "universe" is true gateway route with global meaning. 840 * 841 * Every prefix refers to a set of "nexthop"s (gw, oif), 842 * where gw must have narrower scope. This recursion stops 843 * when gw has LOCAL scope or if "nexthop" is declared ONLINK, 844 * which means that gw is forced to be on link. 845 * 846 * Code is still hairy, but now it is apparently logically 847 * consistent and very flexible. F.e. as by-product it allows 848 * to co-exists in peace independent exterior and interior 849 * routing processes. 850 * 851 * Normally it looks as following. 852 * 853 * {universe prefix} -> (gw, oif) [scope link] 854 * | 855 * |-> {link prefix} -> (gw, oif) [scope local] 856 * | 857 * |-> {local prefix} (terminal node) 858 */ 859 static int fib_check_nh(struct fib_config *cfg, struct fib_nh *nh, 860 struct netlink_ext_ack *extack) 861 { 862 int err = 0; 863 struct net *net; 864 struct net_device *dev; 865 866 net = cfg->fc_nlinfo.nl_net; 867 if (nh->fib_nh_gw4) { 868 struct fib_result res; 869 870 if (nh->fib_nh_flags & RTNH_F_ONLINK) { 871 unsigned int addr_type; 872 873 if (cfg->fc_scope >= RT_SCOPE_LINK) { 874 NL_SET_ERR_MSG(extack, 875 "Nexthop has invalid scope"); 876 return -EINVAL; 877 } 878 dev = __dev_get_by_index(net, nh->fib_nh_oif); 879 if (!dev) { 880 NL_SET_ERR_MSG(extack, "Nexthop device required for onlink"); 881 return -ENODEV; 882 } 883 if (!(dev->flags & IFF_UP)) { 884 NL_SET_ERR_MSG(extack, 885 "Nexthop device is not up"); 886 return -ENETDOWN; 887 } 888 addr_type = inet_addr_type_dev_table(net, dev, 889 nh->fib_nh_gw4); 890 if (addr_type != RTN_UNICAST) { 891 NL_SET_ERR_MSG(extack, 892 "Nexthop has invalid gateway"); 893 return -EINVAL; 894 } 895 if (!netif_carrier_ok(dev)) 896 nh->fib_nh_flags |= RTNH_F_LINKDOWN; 897 nh->fib_nh_dev = dev; 898 dev_hold(dev); 899 nh->fib_nh_scope = RT_SCOPE_LINK; 900 return 0; 901 } 902 rcu_read_lock(); 903 { 904 struct fib_table *tbl = NULL; 905 struct flowi4 fl4 = { 906 .daddr = nh->fib_nh_gw4, 907 .flowi4_scope = cfg->fc_scope + 1, 908 .flowi4_oif = nh->fib_nh_oif, 909 .flowi4_iif = LOOPBACK_IFINDEX, 910 }; 911 912 /* It is not necessary, but requires a bit of thinking */ 913 if (fl4.flowi4_scope < RT_SCOPE_LINK) 914 fl4.flowi4_scope = RT_SCOPE_LINK; 915 916 if (cfg->fc_table) 917 tbl = fib_get_table(net, cfg->fc_table); 918 919 if (tbl) 920 err = fib_table_lookup(tbl, &fl4, &res, 921 FIB_LOOKUP_IGNORE_LINKSTATE | 922 FIB_LOOKUP_NOREF); 923 924 /* on error or if no table given do full lookup. This 925 * is needed for example when nexthops are in the local 926 * table rather than the given table 927 */ 928 if (!tbl || err) { 929 err = fib_lookup(net, &fl4, &res, 930 FIB_LOOKUP_IGNORE_LINKSTATE); 931 } 932 933 if (err) { 934 NL_SET_ERR_MSG(extack, 935 "Nexthop has invalid gateway"); 936 rcu_read_unlock(); 937 return err; 938 } 939 } 940 err = -EINVAL; 941 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) { 942 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway"); 943 goto out; 944 } 945 nh->fib_nh_scope = res.scope; 946 nh->fib_nh_oif = FIB_RES_OIF(res); 947 nh->fib_nh_dev = dev = FIB_RES_DEV(res); 948 if (!dev) { 949 NL_SET_ERR_MSG(extack, 950 "No egress device for nexthop gateway"); 951 goto out; 952 } 953 dev_hold(dev); 954 if (!netif_carrier_ok(dev)) 955 nh->fib_nh_flags |= RTNH_F_LINKDOWN; 956 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN; 957 } else { 958 struct in_device *in_dev; 959 960 if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) { 961 NL_SET_ERR_MSG(extack, 962 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set"); 963 return -EINVAL; 964 } 965 rcu_read_lock(); 966 err = -ENODEV; 967 in_dev = inetdev_by_index(net, nh->fib_nh_oif); 968 if (!in_dev) 969 goto out; 970 err = -ENETDOWN; 971 if (!(in_dev->dev->flags & IFF_UP)) { 972 NL_SET_ERR_MSG(extack, "Device for nexthop is not up"); 973 goto out; 974 } 975 nh->fib_nh_dev = in_dev->dev; 976 dev_hold(nh->fib_nh_dev); 977 nh->fib_nh_scope = RT_SCOPE_HOST; 978 if (!netif_carrier_ok(nh->fib_nh_dev)) 979 nh->fib_nh_flags |= RTNH_F_LINKDOWN; 980 err = 0; 981 } 982 out: 983 rcu_read_unlock(); 984 return err; 985 } 986 987 static inline unsigned int fib_laddr_hashfn(__be32 val) 988 { 989 unsigned int mask = (fib_info_hash_size - 1); 990 991 return ((__force u32)val ^ 992 ((__force u32)val >> 7) ^ 993 ((__force u32)val >> 14)) & mask; 994 } 995 996 static struct hlist_head *fib_info_hash_alloc(int bytes) 997 { 998 if (bytes <= PAGE_SIZE) 999 return kzalloc(bytes, GFP_KERNEL); 1000 else 1001 return (struct hlist_head *) 1002 __get_free_pages(GFP_KERNEL | __GFP_ZERO, 1003 get_order(bytes)); 1004 } 1005 1006 static void fib_info_hash_free(struct hlist_head *hash, int bytes) 1007 { 1008 if (!hash) 1009 return; 1010 1011 if (bytes <= PAGE_SIZE) 1012 kfree(hash); 1013 else 1014 free_pages((unsigned long) hash, get_order(bytes)); 1015 } 1016 1017 static void fib_info_hash_move(struct hlist_head *new_info_hash, 1018 struct hlist_head *new_laddrhash, 1019 unsigned int new_size) 1020 { 1021 struct hlist_head *old_info_hash, *old_laddrhash; 1022 unsigned int old_size = fib_info_hash_size; 1023 unsigned int i, bytes; 1024 1025 spin_lock_bh(&fib_info_lock); 1026 old_info_hash = fib_info_hash; 1027 old_laddrhash = fib_info_laddrhash; 1028 fib_info_hash_size = new_size; 1029 1030 for (i = 0; i < old_size; i++) { 1031 struct hlist_head *head = &fib_info_hash[i]; 1032 struct hlist_node *n; 1033 struct fib_info *fi; 1034 1035 hlist_for_each_entry_safe(fi, n, head, fib_hash) { 1036 struct hlist_head *dest; 1037 unsigned int new_hash; 1038 1039 new_hash = fib_info_hashfn(fi); 1040 dest = &new_info_hash[new_hash]; 1041 hlist_add_head(&fi->fib_hash, dest); 1042 } 1043 } 1044 fib_info_hash = new_info_hash; 1045 1046 for (i = 0; i < old_size; i++) { 1047 struct hlist_head *lhead = &fib_info_laddrhash[i]; 1048 struct hlist_node *n; 1049 struct fib_info *fi; 1050 1051 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) { 1052 struct hlist_head *ldest; 1053 unsigned int new_hash; 1054 1055 new_hash = fib_laddr_hashfn(fi->fib_prefsrc); 1056 ldest = &new_laddrhash[new_hash]; 1057 hlist_add_head(&fi->fib_lhash, ldest); 1058 } 1059 } 1060 fib_info_laddrhash = new_laddrhash; 1061 1062 spin_unlock_bh(&fib_info_lock); 1063 1064 bytes = old_size * sizeof(struct hlist_head *); 1065 fib_info_hash_free(old_info_hash, bytes); 1066 fib_info_hash_free(old_laddrhash, bytes); 1067 } 1068 1069 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh) 1070 { 1071 nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, 1072 nh->fib_nh_gw4, 1073 nh->nh_parent->fib_scope); 1074 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid); 1075 1076 return nh->nh_saddr; 1077 } 1078 1079 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res) 1080 { 1081 struct fib_nh_common *nhc = res->nhc; 1082 struct fib_nh *nh; 1083 1084 if (res->fi->fib_prefsrc) 1085 return res->fi->fib_prefsrc; 1086 1087 nh = container_of(nhc, struct fib_nh, nh_common); 1088 if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid)) 1089 return nh->nh_saddr; 1090 1091 return fib_info_update_nh_saddr(net, nh); 1092 } 1093 1094 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc) 1095 { 1096 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst || 1097 fib_prefsrc != cfg->fc_dst) { 1098 u32 tb_id = cfg->fc_table; 1099 int rc; 1100 1101 if (tb_id == RT_TABLE_MAIN) 1102 tb_id = RT_TABLE_LOCAL; 1103 1104 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net, 1105 fib_prefsrc, tb_id); 1106 1107 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) { 1108 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net, 1109 fib_prefsrc, RT_TABLE_LOCAL); 1110 } 1111 1112 if (rc != RTN_LOCAL) 1113 return false; 1114 } 1115 return true; 1116 } 1117 1118 struct fib_info *fib_create_info(struct fib_config *cfg, 1119 struct netlink_ext_ack *extack) 1120 { 1121 int err; 1122 struct fib_info *fi = NULL; 1123 struct fib_info *ofi; 1124 int nhs = 1; 1125 struct net *net = cfg->fc_nlinfo.nl_net; 1126 1127 if (cfg->fc_type > RTN_MAX) 1128 goto err_inval; 1129 1130 /* Fast check to catch the most weird cases */ 1131 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) { 1132 NL_SET_ERR_MSG(extack, "Invalid scope"); 1133 goto err_inval; 1134 } 1135 1136 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) { 1137 NL_SET_ERR_MSG(extack, 1138 "Invalid rtm_flags - can not contain DEAD or LINKDOWN"); 1139 goto err_inval; 1140 } 1141 1142 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1143 if (cfg->fc_mp) { 1144 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack); 1145 if (nhs == 0) 1146 goto err_inval; 1147 } 1148 #endif 1149 1150 err = -ENOBUFS; 1151 if (fib_info_cnt >= fib_info_hash_size) { 1152 unsigned int new_size = fib_info_hash_size << 1; 1153 struct hlist_head *new_info_hash; 1154 struct hlist_head *new_laddrhash; 1155 unsigned int bytes; 1156 1157 if (!new_size) 1158 new_size = 16; 1159 bytes = new_size * sizeof(struct hlist_head *); 1160 new_info_hash = fib_info_hash_alloc(bytes); 1161 new_laddrhash = fib_info_hash_alloc(bytes); 1162 if (!new_info_hash || !new_laddrhash) { 1163 fib_info_hash_free(new_info_hash, bytes); 1164 fib_info_hash_free(new_laddrhash, bytes); 1165 } else 1166 fib_info_hash_move(new_info_hash, new_laddrhash, new_size); 1167 1168 if (!fib_info_hash_size) 1169 goto failure; 1170 } 1171 1172 fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL); 1173 if (!fi) 1174 goto failure; 1175 fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx, 1176 cfg->fc_mx_len, extack); 1177 if (unlikely(IS_ERR(fi->fib_metrics))) { 1178 err = PTR_ERR(fi->fib_metrics); 1179 kfree(fi); 1180 return ERR_PTR(err); 1181 } 1182 1183 fib_info_cnt++; 1184 fi->fib_net = net; 1185 fi->fib_protocol = cfg->fc_protocol; 1186 fi->fib_scope = cfg->fc_scope; 1187 fi->fib_flags = cfg->fc_flags; 1188 fi->fib_priority = cfg->fc_priority; 1189 fi->fib_prefsrc = cfg->fc_prefsrc; 1190 fi->fib_type = cfg->fc_type; 1191 fi->fib_tb_id = cfg->fc_table; 1192 1193 fi->fib_nhs = nhs; 1194 change_nexthops(fi) { 1195 nexthop_nh->nh_parent = fi; 1196 } endfor_nexthops(fi) 1197 1198 if (cfg->fc_mp) 1199 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack); 1200 else 1201 err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack); 1202 1203 if (err != 0) 1204 goto failure; 1205 1206 if (fib_props[cfg->fc_type].error) { 1207 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp) { 1208 NL_SET_ERR_MSG(extack, 1209 "Gateway, device and multipath can not be specified for this route type"); 1210 goto err_inval; 1211 } 1212 goto link_it; 1213 } else { 1214 switch (cfg->fc_type) { 1215 case RTN_UNICAST: 1216 case RTN_LOCAL: 1217 case RTN_BROADCAST: 1218 case RTN_ANYCAST: 1219 case RTN_MULTICAST: 1220 break; 1221 default: 1222 NL_SET_ERR_MSG(extack, "Invalid route type"); 1223 goto err_inval; 1224 } 1225 } 1226 1227 if (cfg->fc_scope > RT_SCOPE_HOST) { 1228 NL_SET_ERR_MSG(extack, "Invalid scope"); 1229 goto err_inval; 1230 } 1231 1232 if (cfg->fc_scope == RT_SCOPE_HOST) { 1233 struct fib_nh *nh = fi->fib_nh; 1234 1235 /* Local address is added. */ 1236 if (nhs != 1) { 1237 NL_SET_ERR_MSG(extack, 1238 "Route with host scope can not have multiple nexthops"); 1239 goto err_inval; 1240 } 1241 if (nh->fib_nh_gw4) { 1242 NL_SET_ERR_MSG(extack, 1243 "Route with host scope can not have a gateway"); 1244 goto err_inval; 1245 } 1246 nh->fib_nh_scope = RT_SCOPE_NOWHERE; 1247 nh->fib_nh_dev = dev_get_by_index(net, fi->fib_nh->fib_nh_oif); 1248 err = -ENODEV; 1249 if (!nh->fib_nh_dev) 1250 goto failure; 1251 } else { 1252 int linkdown = 0; 1253 1254 change_nexthops(fi) { 1255 err = fib_check_nh(cfg, nexthop_nh, extack); 1256 if (err != 0) 1257 goto failure; 1258 if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) 1259 linkdown++; 1260 } endfor_nexthops(fi) 1261 if (linkdown == fi->fib_nhs) 1262 fi->fib_flags |= RTNH_F_LINKDOWN; 1263 } 1264 1265 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) { 1266 NL_SET_ERR_MSG(extack, "Invalid prefsrc address"); 1267 goto err_inval; 1268 } 1269 1270 change_nexthops(fi) { 1271 fib_info_update_nh_saddr(net, nexthop_nh); 1272 } endfor_nexthops(fi) 1273 1274 fib_rebalance(fi); 1275 1276 link_it: 1277 ofi = fib_find_info(fi); 1278 if (ofi) { 1279 fi->fib_dead = 1; 1280 free_fib_info(fi); 1281 ofi->fib_treeref++; 1282 return ofi; 1283 } 1284 1285 fi->fib_treeref++; 1286 refcount_set(&fi->fib_clntref, 1); 1287 spin_lock_bh(&fib_info_lock); 1288 hlist_add_head(&fi->fib_hash, 1289 &fib_info_hash[fib_info_hashfn(fi)]); 1290 if (fi->fib_prefsrc) { 1291 struct hlist_head *head; 1292 1293 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)]; 1294 hlist_add_head(&fi->fib_lhash, head); 1295 } 1296 change_nexthops(fi) { 1297 struct hlist_head *head; 1298 unsigned int hash; 1299 1300 if (!nexthop_nh->fib_nh_dev) 1301 continue; 1302 hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex); 1303 head = &fib_info_devhash[hash]; 1304 hlist_add_head(&nexthop_nh->nh_hash, head); 1305 } endfor_nexthops(fi) 1306 spin_unlock_bh(&fib_info_lock); 1307 return fi; 1308 1309 err_inval: 1310 err = -EINVAL; 1311 1312 failure: 1313 if (fi) { 1314 fi->fib_dead = 1; 1315 free_fib_info(fi); 1316 } 1317 1318 return ERR_PTR(err); 1319 } 1320 1321 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc, 1322 unsigned int *flags, bool skip_oif) 1323 { 1324 if (nhc->nhc_flags & RTNH_F_DEAD) 1325 *flags |= RTNH_F_DEAD; 1326 1327 if (nhc->nhc_flags & RTNH_F_LINKDOWN) { 1328 *flags |= RTNH_F_LINKDOWN; 1329 1330 rcu_read_lock(); 1331 switch (nhc->nhc_family) { 1332 case AF_INET: 1333 if (ip_ignore_linkdown(nhc->nhc_dev)) 1334 *flags |= RTNH_F_DEAD; 1335 break; 1336 case AF_INET6: 1337 if (ip6_ignore_linkdown(nhc->nhc_dev)) 1338 *flags |= RTNH_F_DEAD; 1339 break; 1340 } 1341 rcu_read_unlock(); 1342 } 1343 1344 if (nhc->nhc_has_gw) { 1345 switch (nhc->nhc_family) { 1346 case AF_INET: 1347 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4)) 1348 goto nla_put_failure; 1349 break; 1350 case AF_INET6: 1351 if (nla_put_in6_addr(skb, RTA_GATEWAY, 1352 &nhc->nhc_gw.ipv6) < 0) 1353 goto nla_put_failure; 1354 break; 1355 } 1356 } 1357 1358 *flags |= (nhc->nhc_flags & RTNH_F_ONLINK); 1359 if (nhc->nhc_flags & RTNH_F_OFFLOAD) 1360 *flags |= RTNH_F_OFFLOAD; 1361 1362 if (!skip_oif && nhc->nhc_dev && 1363 nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex)) 1364 goto nla_put_failure; 1365 1366 if (nhc->nhc_lwtstate && 1367 lwtunnel_fill_encap(skb, nhc->nhc_lwtstate) < 0) 1368 goto nla_put_failure; 1369 1370 return 0; 1371 1372 nla_put_failure: 1373 return -EMSGSIZE; 1374 } 1375 EXPORT_SYMBOL_GPL(fib_nexthop_info); 1376 1377 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6) 1378 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc, 1379 int nh_weight) 1380 { 1381 const struct net_device *dev = nhc->nhc_dev; 1382 struct rtnexthop *rtnh; 1383 unsigned int flags = 0; 1384 1385 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh)); 1386 if (!rtnh) 1387 goto nla_put_failure; 1388 1389 rtnh->rtnh_hops = nh_weight - 1; 1390 rtnh->rtnh_ifindex = dev ? dev->ifindex : 0; 1391 1392 if (fib_nexthop_info(skb, nhc, &flags, true) < 0) 1393 goto nla_put_failure; 1394 1395 rtnh->rtnh_flags = flags; 1396 1397 /* length of rtnetlink header + attributes */ 1398 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh; 1399 1400 return 0; 1401 1402 nla_put_failure: 1403 return -EMSGSIZE; 1404 } 1405 EXPORT_SYMBOL_GPL(fib_add_nexthop); 1406 #endif 1407 1408 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1409 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi) 1410 { 1411 struct nlattr *mp; 1412 1413 mp = nla_nest_start(skb, RTA_MULTIPATH); 1414 if (!mp) 1415 goto nla_put_failure; 1416 1417 for_nexthops(fi) { 1418 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight) < 0) 1419 goto nla_put_failure; 1420 #ifdef CONFIG_IP_ROUTE_CLASSID 1421 if (nh->nh_tclassid && 1422 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid)) 1423 goto nla_put_failure; 1424 #endif 1425 } endfor_nexthops(fi); 1426 1427 nla_nest_end(skb, mp); 1428 1429 return 0; 1430 1431 nla_put_failure: 1432 return -EMSGSIZE; 1433 } 1434 #else 1435 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi) 1436 { 1437 return 0; 1438 } 1439 #endif 1440 1441 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event, 1442 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos, 1443 struct fib_info *fi, unsigned int flags) 1444 { 1445 struct nlmsghdr *nlh; 1446 struct rtmsg *rtm; 1447 1448 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags); 1449 if (!nlh) 1450 return -EMSGSIZE; 1451 1452 rtm = nlmsg_data(nlh); 1453 rtm->rtm_family = AF_INET; 1454 rtm->rtm_dst_len = dst_len; 1455 rtm->rtm_src_len = 0; 1456 rtm->rtm_tos = tos; 1457 if (tb_id < 256) 1458 rtm->rtm_table = tb_id; 1459 else 1460 rtm->rtm_table = RT_TABLE_COMPAT; 1461 if (nla_put_u32(skb, RTA_TABLE, tb_id)) 1462 goto nla_put_failure; 1463 rtm->rtm_type = type; 1464 rtm->rtm_flags = fi->fib_flags; 1465 rtm->rtm_scope = fi->fib_scope; 1466 rtm->rtm_protocol = fi->fib_protocol; 1467 1468 if (rtm->rtm_dst_len && 1469 nla_put_in_addr(skb, RTA_DST, dst)) 1470 goto nla_put_failure; 1471 if (fi->fib_priority && 1472 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority)) 1473 goto nla_put_failure; 1474 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0) 1475 goto nla_put_failure; 1476 1477 if (fi->fib_prefsrc && 1478 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc)) 1479 goto nla_put_failure; 1480 if (fi->fib_nhs == 1) { 1481 struct fib_nh *nh = &fi->fib_nh[0]; 1482 unsigned int flags = 0; 1483 1484 if (fib_nexthop_info(skb, &nh->nh_common, &flags, false) < 0) 1485 goto nla_put_failure; 1486 1487 rtm->rtm_flags = flags; 1488 #ifdef CONFIG_IP_ROUTE_CLASSID 1489 if (nh->nh_tclassid && 1490 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid)) 1491 goto nla_put_failure; 1492 #endif 1493 } else { 1494 if (fib_add_multipath(skb, fi) < 0) 1495 goto nla_put_failure; 1496 } 1497 1498 nlmsg_end(skb, nlh); 1499 return 0; 1500 1501 nla_put_failure: 1502 nlmsg_cancel(skb, nlh); 1503 return -EMSGSIZE; 1504 } 1505 1506 /* 1507 * Update FIB if: 1508 * - local address disappeared -> we must delete all the entries 1509 * referring to it. 1510 * - device went down -> we must shutdown all nexthops going via it. 1511 */ 1512 int fib_sync_down_addr(struct net_device *dev, __be32 local) 1513 { 1514 int ret = 0; 1515 unsigned int hash = fib_laddr_hashfn(local); 1516 struct hlist_head *head = &fib_info_laddrhash[hash]; 1517 struct net *net = dev_net(dev); 1518 int tb_id = l3mdev_fib_table(dev); 1519 struct fib_info *fi; 1520 1521 if (!fib_info_laddrhash || local == 0) 1522 return 0; 1523 1524 hlist_for_each_entry(fi, head, fib_lhash) { 1525 if (!net_eq(fi->fib_net, net) || 1526 fi->fib_tb_id != tb_id) 1527 continue; 1528 if (fi->fib_prefsrc == local) { 1529 fi->fib_flags |= RTNH_F_DEAD; 1530 ret++; 1531 } 1532 } 1533 return ret; 1534 } 1535 1536 static int call_fib_nh_notifiers(struct fib_nh *nh, 1537 enum fib_event_type event_type) 1538 { 1539 bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev); 1540 struct fib_nh_notifier_info info = { 1541 .fib_nh = nh, 1542 }; 1543 1544 switch (event_type) { 1545 case FIB_EVENT_NH_ADD: 1546 if (nh->fib_nh_flags & RTNH_F_DEAD) 1547 break; 1548 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) 1549 break; 1550 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type, 1551 &info.info); 1552 case FIB_EVENT_NH_DEL: 1553 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) || 1554 (nh->fib_nh_flags & RTNH_F_DEAD)) 1555 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), 1556 event_type, &info.info); 1557 default: 1558 break; 1559 } 1560 1561 return NOTIFY_DONE; 1562 } 1563 1564 /* Update the PMTU of exceptions when: 1565 * - the new MTU of the first hop becomes smaller than the PMTU 1566 * - the old MTU was the same as the PMTU, and it limited discovery of 1567 * larger MTUs on the path. With that limit raised, we can now 1568 * discover larger MTUs 1569 * A special case is locked exceptions, for which the PMTU is smaller 1570 * than the minimal accepted PMTU: 1571 * - if the new MTU is greater than the PMTU, don't make any change 1572 * - otherwise, unlock and set PMTU 1573 */ 1574 static void nh_update_mtu(struct fib_nh *nh, u32 new, u32 orig) 1575 { 1576 struct fnhe_hash_bucket *bucket; 1577 int i; 1578 1579 bucket = rcu_dereference_protected(nh->nh_exceptions, 1); 1580 if (!bucket) 1581 return; 1582 1583 for (i = 0; i < FNHE_HASH_SIZE; i++) { 1584 struct fib_nh_exception *fnhe; 1585 1586 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1); 1587 fnhe; 1588 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) { 1589 if (fnhe->fnhe_mtu_locked) { 1590 if (new <= fnhe->fnhe_pmtu) { 1591 fnhe->fnhe_pmtu = new; 1592 fnhe->fnhe_mtu_locked = false; 1593 } 1594 } else if (new < fnhe->fnhe_pmtu || 1595 orig == fnhe->fnhe_pmtu) { 1596 fnhe->fnhe_pmtu = new; 1597 } 1598 } 1599 } 1600 } 1601 1602 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu) 1603 { 1604 unsigned int hash = fib_devindex_hashfn(dev->ifindex); 1605 struct hlist_head *head = &fib_info_devhash[hash]; 1606 struct fib_nh *nh; 1607 1608 hlist_for_each_entry(nh, head, nh_hash) { 1609 if (nh->fib_nh_dev == dev) 1610 nh_update_mtu(nh, dev->mtu, orig_mtu); 1611 } 1612 } 1613 1614 /* Event force Flags Description 1615 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host 1616 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host 1617 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed 1618 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed 1619 */ 1620 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force) 1621 { 1622 int ret = 0; 1623 int scope = RT_SCOPE_NOWHERE; 1624 struct fib_info *prev_fi = NULL; 1625 unsigned int hash = fib_devindex_hashfn(dev->ifindex); 1626 struct hlist_head *head = &fib_info_devhash[hash]; 1627 struct fib_nh *nh; 1628 1629 if (force) 1630 scope = -1; 1631 1632 hlist_for_each_entry(nh, head, nh_hash) { 1633 struct fib_info *fi = nh->nh_parent; 1634 int dead; 1635 1636 BUG_ON(!fi->fib_nhs); 1637 if (nh->fib_nh_dev != dev || fi == prev_fi) 1638 continue; 1639 prev_fi = fi; 1640 dead = 0; 1641 change_nexthops(fi) { 1642 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) 1643 dead++; 1644 else if (nexthop_nh->fib_nh_dev == dev && 1645 nexthop_nh->fib_nh_scope != scope) { 1646 switch (event) { 1647 case NETDEV_DOWN: 1648 case NETDEV_UNREGISTER: 1649 nexthop_nh->fib_nh_flags |= RTNH_F_DEAD; 1650 /* fall through */ 1651 case NETDEV_CHANGE: 1652 nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN; 1653 break; 1654 } 1655 call_fib_nh_notifiers(nexthop_nh, 1656 FIB_EVENT_NH_DEL); 1657 dead++; 1658 } 1659 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1660 if (event == NETDEV_UNREGISTER && 1661 nexthop_nh->fib_nh_dev == dev) { 1662 dead = fi->fib_nhs; 1663 break; 1664 } 1665 #endif 1666 } endfor_nexthops(fi) 1667 if (dead == fi->fib_nhs) { 1668 switch (event) { 1669 case NETDEV_DOWN: 1670 case NETDEV_UNREGISTER: 1671 fi->fib_flags |= RTNH_F_DEAD; 1672 /* fall through */ 1673 case NETDEV_CHANGE: 1674 fi->fib_flags |= RTNH_F_LINKDOWN; 1675 break; 1676 } 1677 ret++; 1678 } 1679 1680 fib_rebalance(fi); 1681 } 1682 1683 return ret; 1684 } 1685 1686 /* Must be invoked inside of an RCU protected region. */ 1687 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res) 1688 { 1689 struct fib_info *fi = NULL, *last_resort = NULL; 1690 struct hlist_head *fa_head = res->fa_head; 1691 struct fib_table *tb = res->table; 1692 u8 slen = 32 - res->prefixlen; 1693 int order = -1, last_idx = -1; 1694 struct fib_alias *fa, *fa1 = NULL; 1695 u32 last_prio = res->fi->fib_priority; 1696 u8 last_tos = 0; 1697 1698 hlist_for_each_entry_rcu(fa, fa_head, fa_list) { 1699 struct fib_info *next_fi = fa->fa_info; 1700 1701 if (fa->fa_slen != slen) 1702 continue; 1703 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos) 1704 continue; 1705 if (fa->tb_id != tb->tb_id) 1706 continue; 1707 if (next_fi->fib_priority > last_prio && 1708 fa->fa_tos == last_tos) { 1709 if (last_tos) 1710 continue; 1711 break; 1712 } 1713 if (next_fi->fib_flags & RTNH_F_DEAD) 1714 continue; 1715 last_tos = fa->fa_tos; 1716 last_prio = next_fi->fib_priority; 1717 1718 if (next_fi->fib_scope != res->scope || 1719 fa->fa_type != RTN_UNICAST) 1720 continue; 1721 if (!next_fi->fib_nh[0].fib_nh_gw4 || 1722 next_fi->fib_nh[0].fib_nh_scope != RT_SCOPE_LINK) 1723 continue; 1724 1725 fib_alias_accessed(fa); 1726 1727 if (!fi) { 1728 if (next_fi != res->fi) 1729 break; 1730 fa1 = fa; 1731 } else if (!fib_detect_death(fi, order, &last_resort, 1732 &last_idx, fa1->fa_default)) { 1733 fib_result_assign(res, fi); 1734 fa1->fa_default = order; 1735 goto out; 1736 } 1737 fi = next_fi; 1738 order++; 1739 } 1740 1741 if (order <= 0 || !fi) { 1742 if (fa1) 1743 fa1->fa_default = -1; 1744 goto out; 1745 } 1746 1747 if (!fib_detect_death(fi, order, &last_resort, &last_idx, 1748 fa1->fa_default)) { 1749 fib_result_assign(res, fi); 1750 fa1->fa_default = order; 1751 goto out; 1752 } 1753 1754 if (last_idx >= 0) 1755 fib_result_assign(res, last_resort); 1756 fa1->fa_default = last_idx; 1757 out: 1758 return; 1759 } 1760 1761 /* 1762 * Dead device goes up. We wake up dead nexthops. 1763 * It takes sense only on multipath routes. 1764 */ 1765 int fib_sync_up(struct net_device *dev, unsigned int nh_flags) 1766 { 1767 struct fib_info *prev_fi; 1768 unsigned int hash; 1769 struct hlist_head *head; 1770 struct fib_nh *nh; 1771 int ret; 1772 1773 if (!(dev->flags & IFF_UP)) 1774 return 0; 1775 1776 if (nh_flags & RTNH_F_DEAD) { 1777 unsigned int flags = dev_get_flags(dev); 1778 1779 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 1780 nh_flags |= RTNH_F_LINKDOWN; 1781 } 1782 1783 prev_fi = NULL; 1784 hash = fib_devindex_hashfn(dev->ifindex); 1785 head = &fib_info_devhash[hash]; 1786 ret = 0; 1787 1788 hlist_for_each_entry(nh, head, nh_hash) { 1789 struct fib_info *fi = nh->nh_parent; 1790 int alive; 1791 1792 BUG_ON(!fi->fib_nhs); 1793 if (nh->fib_nh_dev != dev || fi == prev_fi) 1794 continue; 1795 1796 prev_fi = fi; 1797 alive = 0; 1798 change_nexthops(fi) { 1799 if (!(nexthop_nh->fib_nh_flags & nh_flags)) { 1800 alive++; 1801 continue; 1802 } 1803 if (!nexthop_nh->fib_nh_dev || 1804 !(nexthop_nh->fib_nh_dev->flags & IFF_UP)) 1805 continue; 1806 if (nexthop_nh->fib_nh_dev != dev || 1807 !__in_dev_get_rtnl(dev)) 1808 continue; 1809 alive++; 1810 nexthop_nh->fib_nh_flags &= ~nh_flags; 1811 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD); 1812 } endfor_nexthops(fi) 1813 1814 if (alive > 0) { 1815 fi->fib_flags &= ~nh_flags; 1816 ret++; 1817 } 1818 1819 fib_rebalance(fi); 1820 } 1821 1822 return ret; 1823 } 1824 1825 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1826 static bool fib_good_nh(const struct fib_nh *nh) 1827 { 1828 int state = NUD_REACHABLE; 1829 1830 if (nh->fib_nh_scope == RT_SCOPE_LINK) { 1831 struct neighbour *n; 1832 1833 rcu_read_lock_bh(); 1834 1835 n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev, 1836 (__force u32)nh->fib_nh_gw4); 1837 if (n) 1838 state = n->nud_state; 1839 1840 rcu_read_unlock_bh(); 1841 } 1842 1843 return !!(state & NUD_VALID); 1844 } 1845 1846 void fib_select_multipath(struct fib_result *res, int hash) 1847 { 1848 struct fib_info *fi = res->fi; 1849 struct net *net = fi->fib_net; 1850 bool first = false; 1851 1852 change_nexthops(fi) { 1853 if (net->ipv4.sysctl_fib_multipath_use_neigh) { 1854 if (!fib_good_nh(nexthop_nh)) 1855 continue; 1856 if (!first) { 1857 res->nh_sel = nhsel; 1858 res->nhc = &nexthop_nh->nh_common; 1859 first = true; 1860 } 1861 } 1862 1863 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound)) 1864 continue; 1865 1866 res->nh_sel = nhsel; 1867 res->nhc = &nexthop_nh->nh_common; 1868 return; 1869 } endfor_nexthops(fi); 1870 } 1871 #endif 1872 1873 void fib_select_path(struct net *net, struct fib_result *res, 1874 struct flowi4 *fl4, const struct sk_buff *skb) 1875 { 1876 if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF)) 1877 goto check_saddr; 1878 1879 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1880 if (res->fi->fib_nhs > 1) { 1881 int h = fib_multipath_hash(net, fl4, skb, NULL); 1882 1883 fib_select_multipath(res, h); 1884 } 1885 else 1886 #endif 1887 if (!res->prefixlen && 1888 res->table->tb_num_default > 1 && 1889 res->type == RTN_UNICAST) 1890 fib_select_default(fl4, res); 1891 1892 check_saddr: 1893 if (!fl4->saddr) 1894 fl4->saddr = fib_result_prefsrc(net, res); 1895 } 1896