1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * IPv4 Forwarding Information Base: semantics. 8 * 9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 10 */ 11 12 #include <linux/uaccess.h> 13 #include <linux/bitops.h> 14 #include <linux/types.h> 15 #include <linux/kernel.h> 16 #include <linux/jiffies.h> 17 #include <linux/mm.h> 18 #include <linux/string.h> 19 #include <linux/socket.h> 20 #include <linux/sockios.h> 21 #include <linux/errno.h> 22 #include <linux/in.h> 23 #include <linux/inet.h> 24 #include <linux/inetdevice.h> 25 #include <linux/netdevice.h> 26 #include <linux/if_arp.h> 27 #include <linux/proc_fs.h> 28 #include <linux/skbuff.h> 29 #include <linux/init.h> 30 #include <linux/slab.h> 31 #include <linux/netlink.h> 32 #include <linux/hash.h> 33 #include <linux/nospec.h> 34 35 #include <net/arp.h> 36 #include <net/inet_dscp.h> 37 #include <net/ip.h> 38 #include <net/protocol.h> 39 #include <net/route.h> 40 #include <net/tcp.h> 41 #include <net/sock.h> 42 #include <net/ip_fib.h> 43 #include <net/ip6_fib.h> 44 #include <net/nexthop.h> 45 #include <net/netlink.h> 46 #include <net/rtnh.h> 47 #include <net/lwtunnel.h> 48 #include <net/fib_notifier.h> 49 #include <net/addrconf.h> 50 51 #include "fib_lookup.h" 52 53 /* for_nexthops and change_nexthops only used when nexthop object 54 * is not set in a fib_info. The logic within can reference fib_nh. 55 */ 56 #ifdef CONFIG_IP_ROUTE_MULTIPATH 57 58 #define for_nexthops(fi) { \ 59 int nhsel; const struct fib_nh *nh; \ 60 for (nhsel = 0, nh = (fi)->fib_nh; \ 61 nhsel < fib_info_num_path((fi)); \ 62 nh++, nhsel++) 63 64 #define change_nexthops(fi) { \ 65 int nhsel; struct fib_nh *nexthop_nh; \ 66 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \ 67 nhsel < fib_info_num_path((fi)); \ 68 nexthop_nh++, nhsel++) 69 70 #else /* CONFIG_IP_ROUTE_MULTIPATH */ 71 72 /* Hope, that gcc will optimize it to get rid of dummy loop */ 73 74 #define for_nexthops(fi) { \ 75 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \ 76 for (nhsel = 0; nhsel < 1; nhsel++) 77 78 #define change_nexthops(fi) { \ 79 int nhsel; \ 80 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \ 81 for (nhsel = 0; nhsel < 1; nhsel++) 82 83 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 84 85 #define endfor_nexthops(fi) } 86 87 88 const struct fib_prop fib_props[RTN_MAX + 1] = { 89 [RTN_UNSPEC] = { 90 .error = 0, 91 .scope = RT_SCOPE_NOWHERE, 92 }, 93 [RTN_UNICAST] = { 94 .error = 0, 95 .scope = RT_SCOPE_UNIVERSE, 96 }, 97 [RTN_LOCAL] = { 98 .error = 0, 99 .scope = RT_SCOPE_HOST, 100 }, 101 [RTN_BROADCAST] = { 102 .error = 0, 103 .scope = RT_SCOPE_LINK, 104 }, 105 [RTN_ANYCAST] = { 106 .error = 0, 107 .scope = RT_SCOPE_LINK, 108 }, 109 [RTN_MULTICAST] = { 110 .error = 0, 111 .scope = RT_SCOPE_UNIVERSE, 112 }, 113 [RTN_BLACKHOLE] = { 114 .error = -EINVAL, 115 .scope = RT_SCOPE_UNIVERSE, 116 }, 117 [RTN_UNREACHABLE] = { 118 .error = -EHOSTUNREACH, 119 .scope = RT_SCOPE_UNIVERSE, 120 }, 121 [RTN_PROHIBIT] = { 122 .error = -EACCES, 123 .scope = RT_SCOPE_UNIVERSE, 124 }, 125 [RTN_THROW] = { 126 .error = -EAGAIN, 127 .scope = RT_SCOPE_UNIVERSE, 128 }, 129 [RTN_NAT] = { 130 .error = -EINVAL, 131 .scope = RT_SCOPE_NOWHERE, 132 }, 133 [RTN_XRESOLVE] = { 134 .error = -EINVAL, 135 .scope = RT_SCOPE_NOWHERE, 136 }, 137 }; 138 139 static void rt_fibinfo_free(struct rtable __rcu **rtp) 140 { 141 struct rtable *rt = rcu_dereference_protected(*rtp, 1); 142 143 if (!rt) 144 return; 145 146 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL); 147 * because we waited an RCU grace period before calling 148 * free_fib_info_rcu() 149 */ 150 151 dst_dev_put(&rt->dst); 152 dst_release_immediate(&rt->dst); 153 } 154 155 static void free_nh_exceptions(struct fib_nh_common *nhc) 156 { 157 struct fnhe_hash_bucket *hash; 158 int i; 159 160 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1); 161 if (!hash) 162 return; 163 for (i = 0; i < FNHE_HASH_SIZE; i++) { 164 struct fib_nh_exception *fnhe; 165 166 fnhe = rcu_dereference_protected(hash[i].chain, 1); 167 while (fnhe) { 168 struct fib_nh_exception *next; 169 170 next = rcu_dereference_protected(fnhe->fnhe_next, 1); 171 172 rt_fibinfo_free(&fnhe->fnhe_rth_input); 173 rt_fibinfo_free(&fnhe->fnhe_rth_output); 174 175 kfree(fnhe); 176 177 fnhe = next; 178 } 179 } 180 kfree(hash); 181 } 182 183 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp) 184 { 185 int cpu; 186 187 if (!rtp) 188 return; 189 190 for_each_possible_cpu(cpu) { 191 struct rtable *rt; 192 193 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1); 194 if (rt) { 195 dst_dev_put(&rt->dst); 196 dst_release_immediate(&rt->dst); 197 } 198 } 199 free_percpu(rtp); 200 } 201 202 void fib_nh_common_release(struct fib_nh_common *nhc) 203 { 204 netdev_put(nhc->nhc_dev, &nhc->nhc_dev_tracker); 205 lwtstate_put(nhc->nhc_lwtstate); 206 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output); 207 rt_fibinfo_free(&nhc->nhc_rth_input); 208 free_nh_exceptions(nhc); 209 } 210 EXPORT_SYMBOL_GPL(fib_nh_common_release); 211 212 void fib_nh_release(struct net *net, struct fib_nh *fib_nh) 213 { 214 #ifdef CONFIG_IP_ROUTE_CLASSID 215 if (fib_nh->nh_tclassid) 216 atomic_dec(&net->ipv4.fib_num_tclassid_users); 217 #endif 218 fib_nh_common_release(&fib_nh->nh_common); 219 } 220 221 /* Release a nexthop info record */ 222 static void free_fib_info_rcu(struct rcu_head *head) 223 { 224 struct fib_info *fi = container_of(head, struct fib_info, rcu); 225 226 if (fi->nh) { 227 nexthop_put(fi->nh); 228 } else { 229 change_nexthops(fi) { 230 fib_nh_release(fi->fib_net, nexthop_nh); 231 } endfor_nexthops(fi); 232 } 233 234 ip_fib_metrics_put(fi->fib_metrics); 235 236 kfree(fi); 237 } 238 239 void free_fib_info(struct fib_info *fi) 240 { 241 if (fi->fib_dead == 0) { 242 pr_warn("Freeing alive fib_info %p\n", fi); 243 return; 244 } 245 246 call_rcu_hurry(&fi->rcu, free_fib_info_rcu); 247 } 248 EXPORT_SYMBOL_GPL(free_fib_info); 249 250 void fib_release_info(struct fib_info *fi) 251 { 252 ASSERT_RTNL(); 253 if (fi && refcount_dec_and_test(&fi->fib_treeref)) { 254 hlist_del(&fi->fib_hash); 255 fi->fib_net->ipv4.fib_info_cnt--; 256 257 if (fi->fib_prefsrc) 258 hlist_del(&fi->fib_lhash); 259 if (fi->nh) { 260 list_del(&fi->nh_list); 261 } else { 262 change_nexthops(fi) { 263 if (!nexthop_nh->fib_nh_dev) 264 continue; 265 hlist_del_rcu(&nexthop_nh->nh_hash); 266 } endfor_nexthops(fi) 267 } 268 /* Paired with READ_ONCE() from fib_table_lookup() */ 269 WRITE_ONCE(fi->fib_dead, 1); 270 fib_info_put(fi); 271 } 272 } 273 274 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi) 275 { 276 const struct fib_nh *onh; 277 278 if (fi->nh || ofi->nh) 279 return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1; 280 281 if (ofi->fib_nhs == 0) 282 return 0; 283 284 for_nexthops(fi) { 285 onh = fib_info_nh(ofi, nhsel); 286 287 if (nh->fib_nh_oif != onh->fib_nh_oif || 288 nh->fib_nh_gw_family != onh->fib_nh_gw_family || 289 nh->fib_nh_scope != onh->fib_nh_scope || 290 #ifdef CONFIG_IP_ROUTE_MULTIPATH 291 nh->fib_nh_weight != onh->fib_nh_weight || 292 #endif 293 #ifdef CONFIG_IP_ROUTE_CLASSID 294 nh->nh_tclassid != onh->nh_tclassid || 295 #endif 296 lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) || 297 ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK)) 298 return -1; 299 300 if (nh->fib_nh_gw_family == AF_INET && 301 nh->fib_nh_gw4 != onh->fib_nh_gw4) 302 return -1; 303 304 if (nh->fib_nh_gw_family == AF_INET6 && 305 ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6)) 306 return -1; 307 } endfor_nexthops(fi); 308 return 0; 309 } 310 311 static struct hlist_head *fib_nh_head(struct net_device *dev) 312 { 313 return &dev->fib_nh_head; 314 } 315 316 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope, 317 u32 prefsrc, u32 priority) 318 { 319 unsigned int val = init_val; 320 321 val ^= (protocol << 8) | scope; 322 val ^= prefsrc; 323 val ^= priority; 324 325 return val; 326 } 327 328 static unsigned int fib_info_hashfn_result(const struct net *net, 329 unsigned int val) 330 { 331 return hash_32(val ^ net_hash_mix(net), net->ipv4.fib_info_hash_bits); 332 } 333 334 static struct hlist_head *fib_info_hash_bucket(struct fib_info *fi) 335 { 336 struct net *net = fi->fib_net; 337 unsigned int val; 338 339 val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol, 340 fi->fib_scope, (__force u32)fi->fib_prefsrc, 341 fi->fib_priority); 342 343 if (fi->nh) { 344 val ^= fi->nh->id; 345 } else { 346 for_nexthops(fi) { 347 val ^= nh->fib_nh_oif; 348 } endfor_nexthops(fi) 349 } 350 351 return &net->ipv4.fib_info_hash[fib_info_hashfn_result(net, val)]; 352 } 353 354 static struct hlist_head *fib_info_laddrhash_bucket(const struct net *net, 355 __be32 val) 356 { 357 unsigned int hash_bits = net->ipv4.fib_info_hash_bits; 358 u32 slot; 359 360 slot = hash_32(net_hash_mix(net) ^ (__force u32)val, hash_bits); 361 362 return &net->ipv4.fib_info_hash[(1 << hash_bits) + slot]; 363 } 364 365 static struct hlist_head *fib_info_hash_alloc(unsigned int hash_bits) 366 { 367 /* The second half is used for prefsrc */ 368 return kvcalloc((1 << hash_bits) * 2, sizeof(struct hlist_head *), 369 GFP_KERNEL); 370 } 371 372 static void fib_info_hash_free(struct hlist_head *head) 373 { 374 kvfree(head); 375 } 376 377 static void fib_info_hash_grow(struct net *net) 378 { 379 unsigned int old_size = 1 << net->ipv4.fib_info_hash_bits; 380 struct hlist_head *new_info_hash, *old_info_hash; 381 unsigned int i; 382 383 if (net->ipv4.fib_info_cnt < old_size) 384 return; 385 386 new_info_hash = fib_info_hash_alloc(net->ipv4.fib_info_hash_bits + 1); 387 if (!new_info_hash) 388 return; 389 390 old_info_hash = net->ipv4.fib_info_hash; 391 net->ipv4.fib_info_hash = new_info_hash; 392 net->ipv4.fib_info_hash_bits += 1; 393 394 for (i = 0; i < old_size; i++) { 395 struct hlist_head *head = &old_info_hash[i]; 396 struct hlist_node *n; 397 struct fib_info *fi; 398 399 hlist_for_each_entry_safe(fi, n, head, fib_hash) 400 hlist_add_head(&fi->fib_hash, fib_info_hash_bucket(fi)); 401 } 402 403 for (i = 0; i < old_size; i++) { 404 struct hlist_head *lhead = &old_info_hash[old_size + i]; 405 struct hlist_node *n; 406 struct fib_info *fi; 407 408 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) 409 hlist_add_head(&fi->fib_lhash, 410 fib_info_laddrhash_bucket(fi->fib_net, 411 fi->fib_prefsrc)); 412 } 413 414 fib_info_hash_free(old_info_hash); 415 } 416 417 /* no metrics, only nexthop id */ 418 static struct fib_info *fib_find_info_nh(struct net *net, 419 const struct fib_config *cfg) 420 { 421 struct hlist_head *head; 422 struct fib_info *fi; 423 unsigned int hash; 424 425 hash = fib_info_hashfn_1(cfg->fc_nh_id, 426 cfg->fc_protocol, cfg->fc_scope, 427 (__force u32)cfg->fc_prefsrc, 428 cfg->fc_priority); 429 hash = fib_info_hashfn_result(net, hash); 430 head = &net->ipv4.fib_info_hash[hash]; 431 432 hlist_for_each_entry(fi, head, fib_hash) { 433 if (!fi->nh || fi->nh->id != cfg->fc_nh_id) 434 continue; 435 436 if (cfg->fc_protocol == fi->fib_protocol && 437 cfg->fc_scope == fi->fib_scope && 438 cfg->fc_prefsrc == fi->fib_prefsrc && 439 cfg->fc_priority == fi->fib_priority && 440 cfg->fc_type == fi->fib_type && 441 cfg->fc_table == fi->fib_tb_id && 442 !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK)) 443 return fi; 444 } 445 446 return NULL; 447 } 448 449 static struct fib_info *fib_find_info(struct fib_info *nfi) 450 { 451 struct hlist_head *head = fib_info_hash_bucket(nfi); 452 struct fib_info *fi; 453 454 hlist_for_each_entry(fi, head, fib_hash) { 455 if (fi->fib_nhs != nfi->fib_nhs) 456 continue; 457 458 if (nfi->fib_protocol == fi->fib_protocol && 459 nfi->fib_scope == fi->fib_scope && 460 nfi->fib_prefsrc == fi->fib_prefsrc && 461 nfi->fib_priority == fi->fib_priority && 462 nfi->fib_type == fi->fib_type && 463 nfi->fib_tb_id == fi->fib_tb_id && 464 memcmp(nfi->fib_metrics, fi->fib_metrics, 465 sizeof(u32) * RTAX_MAX) == 0 && 466 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) && 467 nh_comp(fi, nfi) == 0) 468 return fi; 469 } 470 471 return NULL; 472 } 473 474 /* Check, that the gateway is already configured. 475 * Used only by redirect accept routine, under rcu_read_lock(); 476 */ 477 int ip_fib_check_default(__be32 gw, struct net_device *dev) 478 { 479 struct hlist_head *head; 480 struct fib_nh *nh; 481 482 head = fib_nh_head(dev); 483 484 hlist_for_each_entry_rcu(nh, head, nh_hash) { 485 DEBUG_NET_WARN_ON_ONCE(nh->fib_nh_dev != dev); 486 if (nh->fib_nh_gw4 == gw && 487 !(nh->fib_nh_flags & RTNH_F_DEAD)) { 488 return 0; 489 } 490 } 491 492 return -1; 493 } 494 495 size_t fib_nlmsg_size(struct fib_info *fi) 496 { 497 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg)) 498 + nla_total_size(4) /* RTA_TABLE */ 499 + nla_total_size(4) /* RTA_DST */ 500 + nla_total_size(4) /* RTA_PRIORITY */ 501 + nla_total_size(4) /* RTA_PREFSRC */ 502 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */ 503 unsigned int nhs = fib_info_num_path(fi); 504 505 /* space for nested metrics */ 506 payload += nla_total_size((RTAX_MAX * nla_total_size(4))); 507 508 if (fi->nh) 509 payload += nla_total_size(4); /* RTA_NH_ID */ 510 511 if (nhs) { 512 size_t nh_encapsize = 0; 513 /* Also handles the special case nhs == 1 */ 514 515 /* each nexthop is packed in an attribute */ 516 size_t nhsize = nla_total_size(sizeof(struct rtnexthop)); 517 unsigned int i; 518 519 /* may contain flow and gateway attribute */ 520 nhsize += 2 * nla_total_size(4); 521 522 /* grab encap info */ 523 for (i = 0; i < fib_info_num_path(fi); i++) { 524 struct fib_nh_common *nhc = fib_info_nhc(fi, i); 525 526 if (nhc->nhc_lwtstate) { 527 /* RTA_ENCAP_TYPE */ 528 nh_encapsize += lwtunnel_get_encap_size( 529 nhc->nhc_lwtstate); 530 /* RTA_ENCAP */ 531 nh_encapsize += nla_total_size(2); 532 } 533 } 534 535 /* all nexthops are packed in a nested attribute */ 536 payload += nla_total_size((nhs * nhsize) + nh_encapsize); 537 538 } 539 540 return payload; 541 } 542 543 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa, 544 int dst_len, u32 tb_id, const struct nl_info *info, 545 unsigned int nlm_flags) 546 { 547 struct fib_rt_info fri; 548 struct sk_buff *skb; 549 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0; 550 int err = -ENOBUFS; 551 552 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL); 553 if (!skb) 554 goto errout; 555 556 fri.fi = fa->fa_info; 557 fri.tb_id = tb_id; 558 fri.dst = key; 559 fri.dst_len = dst_len; 560 fri.dscp = fa->fa_dscp; 561 fri.type = fa->fa_type; 562 fri.offload = READ_ONCE(fa->offload); 563 fri.trap = READ_ONCE(fa->trap); 564 fri.offload_failed = READ_ONCE(fa->offload_failed); 565 err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags); 566 if (err < 0) { 567 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */ 568 WARN_ON(err == -EMSGSIZE); 569 kfree_skb(skb); 570 goto errout; 571 } 572 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE, 573 info->nlh, GFP_KERNEL); 574 return; 575 errout: 576 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err); 577 } 578 579 static int fib_detect_death(struct fib_info *fi, int order, 580 struct fib_info **last_resort, int *last_idx, 581 int dflt) 582 { 583 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0); 584 struct neighbour *n; 585 int state = NUD_NONE; 586 587 if (likely(nhc->nhc_gw_family == AF_INET)) 588 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev); 589 else if (nhc->nhc_gw_family == AF_INET6) 590 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6, 591 nhc->nhc_dev); 592 else 593 n = NULL; 594 595 if (n) { 596 state = READ_ONCE(n->nud_state); 597 neigh_release(n); 598 } else { 599 return 0; 600 } 601 if (state == NUD_REACHABLE) 602 return 0; 603 if ((state & NUD_VALID) && order != dflt) 604 return 0; 605 if ((state & NUD_VALID) || 606 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) { 607 *last_resort = fi; 608 *last_idx = order; 609 } 610 return 1; 611 } 612 613 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc, 614 struct nlattr *encap, u16 encap_type, 615 void *cfg, gfp_t gfp_flags, 616 struct netlink_ext_ack *extack) 617 { 618 int err; 619 620 nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *, 621 gfp_flags); 622 if (!nhc->nhc_pcpu_rth_output) 623 return -ENOMEM; 624 625 if (encap) { 626 struct lwtunnel_state *lwtstate; 627 628 if (encap_type == LWTUNNEL_ENCAP_NONE) { 629 NL_SET_ERR_MSG(extack, "LWT encap type not specified"); 630 err = -EINVAL; 631 goto lwt_failure; 632 } 633 err = lwtunnel_build_state(net, encap_type, encap, 634 nhc->nhc_family, cfg, &lwtstate, 635 extack); 636 if (err) 637 goto lwt_failure; 638 639 nhc->nhc_lwtstate = lwtstate_get(lwtstate); 640 } 641 642 return 0; 643 644 lwt_failure: 645 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output); 646 nhc->nhc_pcpu_rth_output = NULL; 647 return err; 648 } 649 EXPORT_SYMBOL_GPL(fib_nh_common_init); 650 651 int fib_nh_init(struct net *net, struct fib_nh *nh, 652 struct fib_config *cfg, int nh_weight, 653 struct netlink_ext_ack *extack) 654 { 655 int err; 656 657 nh->fib_nh_family = AF_INET; 658 659 err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap, 660 cfg->fc_encap_type, cfg, GFP_KERNEL, extack); 661 if (err) 662 return err; 663 664 nh->fib_nh_oif = cfg->fc_oif; 665 nh->fib_nh_gw_family = cfg->fc_gw_family; 666 if (cfg->fc_gw_family == AF_INET) 667 nh->fib_nh_gw4 = cfg->fc_gw4; 668 else if (cfg->fc_gw_family == AF_INET6) 669 nh->fib_nh_gw6 = cfg->fc_gw6; 670 671 nh->fib_nh_flags = cfg->fc_flags; 672 673 #ifdef CONFIG_IP_ROUTE_CLASSID 674 nh->nh_tclassid = cfg->fc_flow; 675 if (nh->nh_tclassid) 676 atomic_inc(&net->ipv4.fib_num_tclassid_users); 677 #endif 678 #ifdef CONFIG_IP_ROUTE_MULTIPATH 679 nh->fib_nh_weight = nh_weight; 680 #endif 681 return 0; 682 } 683 684 #ifdef CONFIG_IP_ROUTE_MULTIPATH 685 686 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining, 687 struct netlink_ext_ack *extack) 688 { 689 int nhs = 0; 690 691 while (rtnh_ok(rtnh, remaining)) { 692 nhs++; 693 rtnh = rtnh_next(rtnh, &remaining); 694 } 695 696 /* leftover implies invalid nexthop configuration, discard it */ 697 if (remaining > 0) { 698 NL_SET_ERR_MSG(extack, 699 "Invalid nexthop configuration - extra data after nexthops"); 700 nhs = 0; 701 } 702 703 return nhs; 704 } 705 706 static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla, 707 struct netlink_ext_ack *extack) 708 { 709 if (nla_len(nla) < sizeof(*gw)) { 710 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY"); 711 return -EINVAL; 712 } 713 714 *gw = nla_get_in_addr(nla); 715 716 return 0; 717 } 718 719 /* only called when fib_nh is integrated into fib_info */ 720 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh, 721 int remaining, struct fib_config *cfg, 722 struct netlink_ext_ack *extack) 723 { 724 struct net *net = fi->fib_net; 725 struct fib_config fib_cfg; 726 struct fib_nh *nh; 727 int ret; 728 729 change_nexthops(fi) { 730 int attrlen; 731 732 memset(&fib_cfg, 0, sizeof(fib_cfg)); 733 734 if (!rtnh_ok(rtnh, remaining)) { 735 NL_SET_ERR_MSG(extack, 736 "Invalid nexthop configuration - extra data after nexthop"); 737 return -EINVAL; 738 } 739 740 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) { 741 NL_SET_ERR_MSG(extack, 742 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN"); 743 return -EINVAL; 744 } 745 746 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags; 747 fib_cfg.fc_oif = rtnh->rtnh_ifindex; 748 749 attrlen = rtnh_attrlen(rtnh); 750 if (attrlen > 0) { 751 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh); 752 753 nla = nla_find(attrs, attrlen, RTA_GATEWAY); 754 nlav = nla_find(attrs, attrlen, RTA_VIA); 755 if (nla && nlav) { 756 NL_SET_ERR_MSG(extack, 757 "Nexthop configuration can not contain both GATEWAY and VIA"); 758 return -EINVAL; 759 } 760 if (nla) { 761 ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla, 762 extack); 763 if (ret) 764 goto errout; 765 766 if (fib_cfg.fc_gw4) 767 fib_cfg.fc_gw_family = AF_INET; 768 } else if (nlav) { 769 ret = fib_gw_from_via(&fib_cfg, nlav, extack); 770 if (ret) 771 goto errout; 772 } 773 774 nla = nla_find(attrs, attrlen, RTA_FLOW); 775 if (nla) { 776 if (nla_len(nla) < sizeof(u32)) { 777 NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW"); 778 return -EINVAL; 779 } 780 fib_cfg.fc_flow = nla_get_u32(nla); 781 } 782 783 fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP); 784 /* RTA_ENCAP_TYPE length checked in 785 * lwtunnel_valid_encap_type_attr 786 */ 787 nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE); 788 if (nla) 789 fib_cfg.fc_encap_type = nla_get_u16(nla); 790 } 791 792 ret = fib_nh_init(net, nexthop_nh, &fib_cfg, 793 rtnh->rtnh_hops + 1, extack); 794 if (ret) 795 goto errout; 796 797 rtnh = rtnh_next(rtnh, &remaining); 798 } endfor_nexthops(fi); 799 800 ret = -EINVAL; 801 nh = fib_info_nh(fi, 0); 802 if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) { 803 NL_SET_ERR_MSG(extack, 804 "Nexthop device index does not match RTA_OIF"); 805 goto errout; 806 } 807 if (cfg->fc_gw_family) { 808 if (cfg->fc_gw_family != nh->fib_nh_gw_family || 809 (cfg->fc_gw_family == AF_INET && 810 nh->fib_nh_gw4 != cfg->fc_gw4) || 811 (cfg->fc_gw_family == AF_INET6 && 812 ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) { 813 NL_SET_ERR_MSG(extack, 814 "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA"); 815 goto errout; 816 } 817 } 818 #ifdef CONFIG_IP_ROUTE_CLASSID 819 if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) { 820 NL_SET_ERR_MSG(extack, 821 "Nexthop class id does not match RTA_FLOW"); 822 goto errout; 823 } 824 #endif 825 ret = 0; 826 errout: 827 return ret; 828 } 829 830 /* only called when fib_nh is integrated into fib_info */ 831 static void fib_rebalance(struct fib_info *fi) 832 { 833 int total; 834 int w; 835 836 if (fib_info_num_path(fi) < 2) 837 return; 838 839 total = 0; 840 for_nexthops(fi) { 841 if (nh->fib_nh_flags & RTNH_F_DEAD) 842 continue; 843 844 if (ip_ignore_linkdown(nh->fib_nh_dev) && 845 nh->fib_nh_flags & RTNH_F_LINKDOWN) 846 continue; 847 848 total += nh->fib_nh_weight; 849 } endfor_nexthops(fi); 850 851 w = 0; 852 change_nexthops(fi) { 853 int upper_bound; 854 855 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) { 856 upper_bound = -1; 857 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) && 858 nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) { 859 upper_bound = -1; 860 } else { 861 w += nexthop_nh->fib_nh_weight; 862 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31, 863 total) - 1; 864 } 865 866 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound); 867 } endfor_nexthops(fi); 868 } 869 #else /* CONFIG_IP_ROUTE_MULTIPATH */ 870 871 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh, 872 int remaining, struct fib_config *cfg, 873 struct netlink_ext_ack *extack) 874 { 875 NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel"); 876 877 return -EINVAL; 878 } 879 880 #define fib_rebalance(fi) do { } while (0) 881 882 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 883 884 static int fib_encap_match(struct net *net, u16 encap_type, 885 struct nlattr *encap, 886 const struct fib_nh *nh, 887 const struct fib_config *cfg, 888 struct netlink_ext_ack *extack) 889 { 890 struct lwtunnel_state *lwtstate; 891 int ret, result = 0; 892 893 if (encap_type == LWTUNNEL_ENCAP_NONE) 894 return 0; 895 896 ret = lwtunnel_build_state(net, encap_type, encap, AF_INET, 897 cfg, &lwtstate, extack); 898 if (!ret) { 899 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws); 900 lwtstate_free(lwtstate); 901 } 902 903 return result; 904 } 905 906 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi, 907 struct netlink_ext_ack *extack) 908 { 909 #ifdef CONFIG_IP_ROUTE_MULTIPATH 910 struct rtnexthop *rtnh; 911 int remaining; 912 #endif 913 914 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority) 915 return 1; 916 917 if (cfg->fc_nh_id) { 918 if (fi->nh && cfg->fc_nh_id == fi->nh->id) 919 return 0; 920 return 1; 921 } 922 923 if (fi->nh) { 924 if (cfg->fc_oif || cfg->fc_gw_family || cfg->fc_mp) 925 return 1; 926 return 0; 927 } 928 929 if (cfg->fc_oif || cfg->fc_gw_family) { 930 struct fib_nh *nh; 931 932 nh = fib_info_nh(fi, 0); 933 if (cfg->fc_encap) { 934 if (fib_encap_match(net, cfg->fc_encap_type, 935 cfg->fc_encap, nh, cfg, extack)) 936 return 1; 937 } 938 #ifdef CONFIG_IP_ROUTE_CLASSID 939 if (cfg->fc_flow && 940 cfg->fc_flow != nh->nh_tclassid) 941 return 1; 942 #endif 943 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) || 944 (cfg->fc_gw_family && 945 cfg->fc_gw_family != nh->fib_nh_gw_family)) 946 return 1; 947 948 if (cfg->fc_gw_family == AF_INET && 949 cfg->fc_gw4 != nh->fib_nh_gw4) 950 return 1; 951 952 if (cfg->fc_gw_family == AF_INET6 && 953 ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6)) 954 return 1; 955 956 return 0; 957 } 958 959 #ifdef CONFIG_IP_ROUTE_MULTIPATH 960 if (!cfg->fc_mp) 961 return 0; 962 963 rtnh = cfg->fc_mp; 964 remaining = cfg->fc_mp_len; 965 966 for_nexthops(fi) { 967 int attrlen; 968 969 if (!rtnh_ok(rtnh, remaining)) 970 return -EINVAL; 971 972 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif) 973 return 1; 974 975 attrlen = rtnh_attrlen(rtnh); 976 if (attrlen > 0) { 977 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh); 978 int err; 979 980 nla = nla_find(attrs, attrlen, RTA_GATEWAY); 981 nlav = nla_find(attrs, attrlen, RTA_VIA); 982 if (nla && nlav) { 983 NL_SET_ERR_MSG(extack, 984 "Nexthop configuration can not contain both GATEWAY and VIA"); 985 return -EINVAL; 986 } 987 988 if (nla) { 989 __be32 gw; 990 991 err = fib_gw_from_attr(&gw, nla, extack); 992 if (err) 993 return err; 994 995 if (nh->fib_nh_gw_family != AF_INET || 996 gw != nh->fib_nh_gw4) 997 return 1; 998 } else if (nlav) { 999 struct fib_config cfg2; 1000 1001 err = fib_gw_from_via(&cfg2, nlav, extack); 1002 if (err) 1003 return err; 1004 1005 switch (nh->fib_nh_gw_family) { 1006 case AF_INET: 1007 if (cfg2.fc_gw_family != AF_INET || 1008 cfg2.fc_gw4 != nh->fib_nh_gw4) 1009 return 1; 1010 break; 1011 case AF_INET6: 1012 if (cfg2.fc_gw_family != AF_INET6 || 1013 ipv6_addr_cmp(&cfg2.fc_gw6, 1014 &nh->fib_nh_gw6)) 1015 return 1; 1016 break; 1017 } 1018 } 1019 1020 #ifdef CONFIG_IP_ROUTE_CLASSID 1021 nla = nla_find(attrs, attrlen, RTA_FLOW); 1022 if (nla) { 1023 if (nla_len(nla) < sizeof(u32)) { 1024 NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW"); 1025 return -EINVAL; 1026 } 1027 if (nla_get_u32(nla) != nh->nh_tclassid) 1028 return 1; 1029 } 1030 #endif 1031 } 1032 1033 rtnh = rtnh_next(rtnh, &remaining); 1034 } endfor_nexthops(fi); 1035 #endif 1036 return 0; 1037 } 1038 1039 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi) 1040 { 1041 struct nlattr *nla; 1042 int remaining; 1043 1044 if (!cfg->fc_mx) 1045 return true; 1046 1047 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) { 1048 int type = nla_type(nla); 1049 u32 fi_val, val; 1050 1051 if (!type) 1052 continue; 1053 if (type > RTAX_MAX) 1054 return false; 1055 1056 type = array_index_nospec(type, RTAX_MAX + 1); 1057 if (type == RTAX_CC_ALGO) { 1058 char tmp[TCP_CA_NAME_MAX]; 1059 bool ecn_ca = false; 1060 1061 nla_strscpy(tmp, nla, sizeof(tmp)); 1062 val = tcp_ca_get_key_by_name(tmp, &ecn_ca); 1063 } else { 1064 if (nla_len(nla) != sizeof(u32)) 1065 return false; 1066 val = nla_get_u32(nla); 1067 } 1068 1069 fi_val = fi->fib_metrics->metrics[type - 1]; 1070 if (type == RTAX_FEATURES) 1071 fi_val &= ~DST_FEATURE_ECN_CA; 1072 1073 if (fi_val != val) 1074 return false; 1075 } 1076 1077 return true; 1078 } 1079 1080 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh, 1081 u32 table, struct netlink_ext_ack *extack) 1082 { 1083 struct fib6_config cfg = { 1084 .fc_table = table, 1085 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY, 1086 .fc_ifindex = nh->fib_nh_oif, 1087 .fc_gateway = nh->fib_nh_gw6, 1088 }; 1089 struct fib6_nh fib6_nh = {}; 1090 int err; 1091 1092 err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack); 1093 if (!err) { 1094 nh->fib_nh_dev = fib6_nh.fib_nh_dev; 1095 netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, 1096 GFP_KERNEL); 1097 nh->fib_nh_oif = nh->fib_nh_dev->ifindex; 1098 nh->fib_nh_scope = RT_SCOPE_LINK; 1099 1100 ipv6_stub->fib6_nh_release(&fib6_nh); 1101 } 1102 1103 return err; 1104 } 1105 1106 /* 1107 * Picture 1108 * ------- 1109 * 1110 * Semantics of nexthop is very messy by historical reasons. 1111 * We have to take into account, that: 1112 * a) gateway can be actually local interface address, 1113 * so that gatewayed route is direct. 1114 * b) gateway must be on-link address, possibly 1115 * described not by an ifaddr, but also by a direct route. 1116 * c) If both gateway and interface are specified, they should not 1117 * contradict. 1118 * d) If we use tunnel routes, gateway could be not on-link. 1119 * 1120 * Attempt to reconcile all of these (alas, self-contradictory) conditions 1121 * results in pretty ugly and hairy code with obscure logic. 1122 * 1123 * I chose to generalized it instead, so that the size 1124 * of code does not increase practically, but it becomes 1125 * much more general. 1126 * Every prefix is assigned a "scope" value: "host" is local address, 1127 * "link" is direct route, 1128 * [ ... "site" ... "interior" ... ] 1129 * and "universe" is true gateway route with global meaning. 1130 * 1131 * Every prefix refers to a set of "nexthop"s (gw, oif), 1132 * where gw must have narrower scope. This recursion stops 1133 * when gw has LOCAL scope or if "nexthop" is declared ONLINK, 1134 * which means that gw is forced to be on link. 1135 * 1136 * Code is still hairy, but now it is apparently logically 1137 * consistent and very flexible. F.e. as by-product it allows 1138 * to co-exists in peace independent exterior and interior 1139 * routing processes. 1140 * 1141 * Normally it looks as following. 1142 * 1143 * {universe prefix} -> (gw, oif) [scope link] 1144 * | 1145 * |-> {link prefix} -> (gw, oif) [scope local] 1146 * | 1147 * |-> {local prefix} (terminal node) 1148 */ 1149 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table, 1150 u8 scope, struct netlink_ext_ack *extack) 1151 { 1152 struct net_device *dev; 1153 struct fib_result res; 1154 int err = 0; 1155 1156 if (nh->fib_nh_flags & RTNH_F_ONLINK) { 1157 unsigned int addr_type; 1158 1159 if (scope >= RT_SCOPE_LINK) { 1160 NL_SET_ERR_MSG(extack, "Nexthop has invalid scope"); 1161 return -EINVAL; 1162 } 1163 dev = __dev_get_by_index(net, nh->fib_nh_oif); 1164 if (!dev) { 1165 NL_SET_ERR_MSG(extack, "Nexthop device required for onlink"); 1166 return -ENODEV; 1167 } 1168 if (!(dev->flags & IFF_UP)) { 1169 NL_SET_ERR_MSG(extack, "Nexthop device is not up"); 1170 return -ENETDOWN; 1171 } 1172 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4); 1173 if (addr_type != RTN_UNICAST) { 1174 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway"); 1175 return -EINVAL; 1176 } 1177 if (!netif_carrier_ok(dev)) 1178 nh->fib_nh_flags |= RTNH_F_LINKDOWN; 1179 nh->fib_nh_dev = dev; 1180 netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC); 1181 nh->fib_nh_scope = RT_SCOPE_LINK; 1182 return 0; 1183 } 1184 rcu_read_lock(); 1185 { 1186 struct fib_table *tbl = NULL; 1187 struct flowi4 fl4 = { 1188 .daddr = nh->fib_nh_gw4, 1189 .flowi4_scope = scope + 1, 1190 .flowi4_oif = nh->fib_nh_oif, 1191 .flowi4_iif = LOOPBACK_IFINDEX, 1192 }; 1193 1194 /* It is not necessary, but requires a bit of thinking */ 1195 if (fl4.flowi4_scope < RT_SCOPE_LINK) 1196 fl4.flowi4_scope = RT_SCOPE_LINK; 1197 1198 if (table && table != RT_TABLE_MAIN) 1199 tbl = fib_get_table(net, table); 1200 1201 if (tbl) 1202 err = fib_table_lookup(tbl, &fl4, &res, 1203 FIB_LOOKUP_IGNORE_LINKSTATE | 1204 FIB_LOOKUP_NOREF); 1205 1206 /* on error or if no table given do full lookup. This 1207 * is needed for example when nexthops are in the local 1208 * table rather than the given table 1209 */ 1210 if (!tbl || err) { 1211 err = fib_lookup(net, &fl4, &res, 1212 FIB_LOOKUP_IGNORE_LINKSTATE); 1213 } 1214 1215 if (err) { 1216 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway"); 1217 goto out; 1218 } 1219 } 1220 1221 err = -EINVAL; 1222 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) { 1223 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway"); 1224 goto out; 1225 } 1226 nh->fib_nh_scope = res.scope; 1227 nh->fib_nh_oif = FIB_RES_OIF(res); 1228 nh->fib_nh_dev = dev = FIB_RES_DEV(res); 1229 if (!dev) { 1230 NL_SET_ERR_MSG(extack, 1231 "No egress device for nexthop gateway"); 1232 goto out; 1233 } 1234 netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC); 1235 if (!netif_carrier_ok(dev)) 1236 nh->fib_nh_flags |= RTNH_F_LINKDOWN; 1237 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN; 1238 out: 1239 rcu_read_unlock(); 1240 return err; 1241 } 1242 1243 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh, 1244 struct netlink_ext_ack *extack) 1245 { 1246 struct in_device *in_dev; 1247 int err; 1248 1249 if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) { 1250 NL_SET_ERR_MSG(extack, 1251 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set"); 1252 return -EINVAL; 1253 } 1254 1255 rcu_read_lock(); 1256 1257 err = -ENODEV; 1258 in_dev = inetdev_by_index(net, nh->fib_nh_oif); 1259 if (!in_dev) 1260 goto out; 1261 err = -ENETDOWN; 1262 if (!(in_dev->dev->flags & IFF_UP)) { 1263 NL_SET_ERR_MSG(extack, "Device for nexthop is not up"); 1264 goto out; 1265 } 1266 1267 nh->fib_nh_dev = in_dev->dev; 1268 netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC); 1269 nh->fib_nh_scope = RT_SCOPE_HOST; 1270 if (!netif_carrier_ok(nh->fib_nh_dev)) 1271 nh->fib_nh_flags |= RTNH_F_LINKDOWN; 1272 err = 0; 1273 out: 1274 rcu_read_unlock(); 1275 return err; 1276 } 1277 1278 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope, 1279 struct netlink_ext_ack *extack) 1280 { 1281 int err; 1282 1283 if (nh->fib_nh_gw_family == AF_INET) 1284 err = fib_check_nh_v4_gw(net, nh, table, scope, extack); 1285 else if (nh->fib_nh_gw_family == AF_INET6) 1286 err = fib_check_nh_v6_gw(net, nh, table, extack); 1287 else 1288 err = fib_check_nh_nongw(net, nh, extack); 1289 1290 return err; 1291 } 1292 1293 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc, 1294 unsigned char scope) 1295 { 1296 struct fib_nh *nh; 1297 __be32 saddr; 1298 1299 if (nhc->nhc_family != AF_INET) 1300 return inet_select_addr(nhc->nhc_dev, 0, scope); 1301 1302 nh = container_of(nhc, struct fib_nh, nh_common); 1303 saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope); 1304 1305 WRITE_ONCE(nh->nh_saddr, saddr); 1306 WRITE_ONCE(nh->nh_saddr_genid, atomic_read(&net->ipv4.dev_addr_genid)); 1307 1308 return saddr; 1309 } 1310 1311 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res) 1312 { 1313 struct fib_nh_common *nhc = res->nhc; 1314 1315 if (res->fi->fib_prefsrc) 1316 return res->fi->fib_prefsrc; 1317 1318 if (nhc->nhc_family == AF_INET) { 1319 struct fib_nh *nh; 1320 1321 nh = container_of(nhc, struct fib_nh, nh_common); 1322 if (READ_ONCE(nh->nh_saddr_genid) == 1323 atomic_read(&net->ipv4.dev_addr_genid)) 1324 return READ_ONCE(nh->nh_saddr); 1325 } 1326 1327 return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope); 1328 } 1329 1330 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc) 1331 { 1332 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst || 1333 fib_prefsrc != cfg->fc_dst) { 1334 u32 tb_id = cfg->fc_table; 1335 int rc; 1336 1337 if (tb_id == RT_TABLE_MAIN) 1338 tb_id = RT_TABLE_LOCAL; 1339 1340 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net, 1341 fib_prefsrc, tb_id); 1342 1343 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) { 1344 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net, 1345 fib_prefsrc, RT_TABLE_LOCAL); 1346 } 1347 1348 if (rc != RTN_LOCAL) 1349 return false; 1350 } 1351 return true; 1352 } 1353 1354 struct fib_info *fib_create_info(struct fib_config *cfg, 1355 struct netlink_ext_ack *extack) 1356 { 1357 int err; 1358 struct fib_info *fi = NULL; 1359 struct nexthop *nh = NULL; 1360 struct fib_info *ofi; 1361 int nhs = 1; 1362 struct net *net = cfg->fc_nlinfo.nl_net; 1363 1364 ASSERT_RTNL(); 1365 if (cfg->fc_type > RTN_MAX) 1366 goto err_inval; 1367 1368 /* Fast check to catch the most weird cases */ 1369 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) { 1370 NL_SET_ERR_MSG(extack, "Invalid scope"); 1371 goto err_inval; 1372 } 1373 1374 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) { 1375 NL_SET_ERR_MSG(extack, 1376 "Invalid rtm_flags - can not contain DEAD or LINKDOWN"); 1377 goto err_inval; 1378 } 1379 1380 if (cfg->fc_nh_id) { 1381 if (!cfg->fc_mx) { 1382 fi = fib_find_info_nh(net, cfg); 1383 if (fi) { 1384 refcount_inc(&fi->fib_treeref); 1385 return fi; 1386 } 1387 } 1388 1389 nh = nexthop_find_by_id(net, cfg->fc_nh_id); 1390 if (!nh) { 1391 NL_SET_ERR_MSG(extack, "Nexthop id does not exist"); 1392 goto err_inval; 1393 } 1394 nhs = 0; 1395 } 1396 1397 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1398 if (cfg->fc_mp) { 1399 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack); 1400 if (nhs == 0) 1401 goto err_inval; 1402 } 1403 #endif 1404 1405 fib_info_hash_grow(net); 1406 1407 fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL); 1408 if (!fi) { 1409 err = -ENOBUFS; 1410 goto failure; 1411 } 1412 1413 fi->fib_metrics = ip_fib_metrics_init(cfg->fc_mx, cfg->fc_mx_len, extack); 1414 if (IS_ERR(fi->fib_metrics)) { 1415 err = PTR_ERR(fi->fib_metrics); 1416 kfree(fi); 1417 return ERR_PTR(err); 1418 } 1419 1420 fi->fib_net = net; 1421 fi->fib_protocol = cfg->fc_protocol; 1422 fi->fib_scope = cfg->fc_scope; 1423 fi->fib_flags = cfg->fc_flags; 1424 fi->fib_priority = cfg->fc_priority; 1425 fi->fib_prefsrc = cfg->fc_prefsrc; 1426 fi->fib_type = cfg->fc_type; 1427 fi->fib_tb_id = cfg->fc_table; 1428 1429 fi->fib_nhs = nhs; 1430 if (nh) { 1431 if (!nexthop_get(nh)) { 1432 NL_SET_ERR_MSG(extack, "Nexthop has been deleted"); 1433 err = -EINVAL; 1434 } else { 1435 err = 0; 1436 fi->nh = nh; 1437 } 1438 } else { 1439 change_nexthops(fi) { 1440 nexthop_nh->nh_parent = fi; 1441 } endfor_nexthops(fi) 1442 1443 if (cfg->fc_mp) 1444 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, 1445 extack); 1446 else 1447 err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack); 1448 } 1449 1450 if (err != 0) 1451 goto failure; 1452 1453 if (fib_props[cfg->fc_type].error) { 1454 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) { 1455 NL_SET_ERR_MSG(extack, 1456 "Gateway, device and multipath can not be specified for this route type"); 1457 goto err_inval; 1458 } 1459 goto link_it; 1460 } else { 1461 switch (cfg->fc_type) { 1462 case RTN_UNICAST: 1463 case RTN_LOCAL: 1464 case RTN_BROADCAST: 1465 case RTN_ANYCAST: 1466 case RTN_MULTICAST: 1467 break; 1468 default: 1469 NL_SET_ERR_MSG(extack, "Invalid route type"); 1470 goto err_inval; 1471 } 1472 } 1473 1474 if (cfg->fc_scope > RT_SCOPE_HOST) { 1475 NL_SET_ERR_MSG(extack, "Invalid scope"); 1476 goto err_inval; 1477 } 1478 1479 if (fi->nh) { 1480 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack); 1481 if (err) 1482 goto failure; 1483 } else if (cfg->fc_scope == RT_SCOPE_HOST) { 1484 struct fib_nh *nh = fi->fib_nh; 1485 1486 /* Local address is added. */ 1487 if (nhs != 1) { 1488 NL_SET_ERR_MSG(extack, 1489 "Route with host scope can not have multiple nexthops"); 1490 goto err_inval; 1491 } 1492 if (nh->fib_nh_gw_family) { 1493 NL_SET_ERR_MSG(extack, 1494 "Route with host scope can not have a gateway"); 1495 goto err_inval; 1496 } 1497 nh->fib_nh_scope = RT_SCOPE_NOWHERE; 1498 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif); 1499 err = -ENODEV; 1500 if (!nh->fib_nh_dev) 1501 goto failure; 1502 netdev_tracker_alloc(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, 1503 GFP_KERNEL); 1504 } else { 1505 int linkdown = 0; 1506 1507 change_nexthops(fi) { 1508 err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh, 1509 cfg->fc_table, cfg->fc_scope, 1510 extack); 1511 if (err != 0) 1512 goto failure; 1513 if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) 1514 linkdown++; 1515 } endfor_nexthops(fi) 1516 if (linkdown == fi->fib_nhs) 1517 fi->fib_flags |= RTNH_F_LINKDOWN; 1518 } 1519 1520 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) { 1521 NL_SET_ERR_MSG(extack, "Invalid prefsrc address"); 1522 goto err_inval; 1523 } 1524 1525 if (!fi->nh) { 1526 change_nexthops(fi) { 1527 fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common, 1528 fi->fib_scope); 1529 if (nexthop_nh->fib_nh_gw_family == AF_INET6) 1530 fi->fib_nh_is_v6 = true; 1531 } endfor_nexthops(fi) 1532 1533 fib_rebalance(fi); 1534 } 1535 1536 link_it: 1537 ofi = fib_find_info(fi); 1538 if (ofi) { 1539 /* fib_table_lookup() should not see @fi yet. */ 1540 fi->fib_dead = 1; 1541 free_fib_info(fi); 1542 refcount_inc(&ofi->fib_treeref); 1543 return ofi; 1544 } 1545 1546 refcount_set(&fi->fib_treeref, 1); 1547 refcount_set(&fi->fib_clntref, 1); 1548 1549 net->ipv4.fib_info_cnt++; 1550 hlist_add_head(&fi->fib_hash, fib_info_hash_bucket(fi)); 1551 1552 if (fi->fib_prefsrc) { 1553 struct hlist_head *head; 1554 1555 head = fib_info_laddrhash_bucket(net, fi->fib_prefsrc); 1556 hlist_add_head(&fi->fib_lhash, head); 1557 } 1558 if (fi->nh) { 1559 list_add(&fi->nh_list, &nh->fi_list); 1560 } else { 1561 change_nexthops(fi) { 1562 struct hlist_head *head; 1563 1564 if (!nexthop_nh->fib_nh_dev) 1565 continue; 1566 head = fib_nh_head(nexthop_nh->fib_nh_dev); 1567 hlist_add_head_rcu(&nexthop_nh->nh_hash, head); 1568 } endfor_nexthops(fi) 1569 } 1570 return fi; 1571 1572 err_inval: 1573 err = -EINVAL; 1574 1575 failure: 1576 if (fi) { 1577 /* fib_table_lookup() should not see @fi yet. */ 1578 fi->fib_dead = 1; 1579 free_fib_info(fi); 1580 } 1581 1582 return ERR_PTR(err); 1583 } 1584 1585 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc, 1586 u8 rt_family, unsigned char *flags, bool skip_oif) 1587 { 1588 if (nhc->nhc_flags & RTNH_F_DEAD) 1589 *flags |= RTNH_F_DEAD; 1590 1591 if (nhc->nhc_flags & RTNH_F_LINKDOWN) { 1592 *flags |= RTNH_F_LINKDOWN; 1593 1594 rcu_read_lock(); 1595 switch (nhc->nhc_family) { 1596 case AF_INET: 1597 if (ip_ignore_linkdown(nhc->nhc_dev)) 1598 *flags |= RTNH_F_DEAD; 1599 break; 1600 case AF_INET6: 1601 if (ip6_ignore_linkdown(nhc->nhc_dev)) 1602 *flags |= RTNH_F_DEAD; 1603 break; 1604 } 1605 rcu_read_unlock(); 1606 } 1607 1608 switch (nhc->nhc_gw_family) { 1609 case AF_INET: 1610 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4)) 1611 goto nla_put_failure; 1612 break; 1613 case AF_INET6: 1614 /* if gateway family does not match nexthop family 1615 * gateway is encoded as RTA_VIA 1616 */ 1617 if (rt_family != nhc->nhc_gw_family) { 1618 int alen = sizeof(struct in6_addr); 1619 struct nlattr *nla; 1620 struct rtvia *via; 1621 1622 nla = nla_reserve(skb, RTA_VIA, alen + 2); 1623 if (!nla) 1624 goto nla_put_failure; 1625 1626 via = nla_data(nla); 1627 via->rtvia_family = AF_INET6; 1628 memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen); 1629 } else if (nla_put_in6_addr(skb, RTA_GATEWAY, 1630 &nhc->nhc_gw.ipv6) < 0) { 1631 goto nla_put_failure; 1632 } 1633 break; 1634 } 1635 1636 *flags |= (nhc->nhc_flags & 1637 (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP)); 1638 1639 if (!skip_oif && nhc->nhc_dev && 1640 nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex)) 1641 goto nla_put_failure; 1642 1643 if (nhc->nhc_lwtstate && 1644 lwtunnel_fill_encap(skb, nhc->nhc_lwtstate, 1645 RTA_ENCAP, RTA_ENCAP_TYPE) < 0) 1646 goto nla_put_failure; 1647 1648 return 0; 1649 1650 nla_put_failure: 1651 return -EMSGSIZE; 1652 } 1653 EXPORT_SYMBOL_GPL(fib_nexthop_info); 1654 1655 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6) 1656 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc, 1657 int nh_weight, u8 rt_family, u32 nh_tclassid) 1658 { 1659 const struct net_device *dev = nhc->nhc_dev; 1660 struct rtnexthop *rtnh; 1661 unsigned char flags = 0; 1662 1663 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh)); 1664 if (!rtnh) 1665 goto nla_put_failure; 1666 1667 rtnh->rtnh_hops = nh_weight - 1; 1668 rtnh->rtnh_ifindex = dev ? dev->ifindex : 0; 1669 1670 if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0) 1671 goto nla_put_failure; 1672 1673 rtnh->rtnh_flags = flags; 1674 1675 if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid)) 1676 goto nla_put_failure; 1677 1678 /* length of rtnetlink header + attributes */ 1679 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh; 1680 1681 return 0; 1682 1683 nla_put_failure: 1684 return -EMSGSIZE; 1685 } 1686 EXPORT_SYMBOL_GPL(fib_add_nexthop); 1687 #endif 1688 1689 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1690 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi) 1691 { 1692 struct nlattr *mp; 1693 1694 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH); 1695 if (!mp) 1696 goto nla_put_failure; 1697 1698 if (unlikely(fi->nh)) { 1699 if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0) 1700 goto nla_put_failure; 1701 goto mp_end; 1702 } 1703 1704 for_nexthops(fi) { 1705 u32 nh_tclassid = 0; 1706 #ifdef CONFIG_IP_ROUTE_CLASSID 1707 nh_tclassid = nh->nh_tclassid; 1708 #endif 1709 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight, 1710 AF_INET, nh_tclassid) < 0) 1711 goto nla_put_failure; 1712 } endfor_nexthops(fi); 1713 1714 mp_end: 1715 nla_nest_end(skb, mp); 1716 1717 return 0; 1718 1719 nla_put_failure: 1720 return -EMSGSIZE; 1721 } 1722 #else 1723 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi) 1724 { 1725 return 0; 1726 } 1727 #endif 1728 1729 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event, 1730 const struct fib_rt_info *fri, unsigned int flags) 1731 { 1732 unsigned int nhs = fib_info_num_path(fri->fi); 1733 struct fib_info *fi = fri->fi; 1734 u32 tb_id = fri->tb_id; 1735 struct nlmsghdr *nlh; 1736 struct rtmsg *rtm; 1737 1738 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags); 1739 if (!nlh) 1740 return -EMSGSIZE; 1741 1742 rtm = nlmsg_data(nlh); 1743 rtm->rtm_family = AF_INET; 1744 rtm->rtm_dst_len = fri->dst_len; 1745 rtm->rtm_src_len = 0; 1746 rtm->rtm_tos = inet_dscp_to_dsfield(fri->dscp); 1747 if (tb_id < 256) 1748 rtm->rtm_table = tb_id; 1749 else 1750 rtm->rtm_table = RT_TABLE_COMPAT; 1751 if (nla_put_u32(skb, RTA_TABLE, tb_id)) 1752 goto nla_put_failure; 1753 rtm->rtm_type = fri->type; 1754 rtm->rtm_flags = fi->fib_flags; 1755 rtm->rtm_scope = fi->fib_scope; 1756 rtm->rtm_protocol = fi->fib_protocol; 1757 1758 if (rtm->rtm_dst_len && 1759 nla_put_in_addr(skb, RTA_DST, fri->dst)) 1760 goto nla_put_failure; 1761 if (fi->fib_priority && 1762 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority)) 1763 goto nla_put_failure; 1764 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0) 1765 goto nla_put_failure; 1766 1767 if (fi->fib_prefsrc && 1768 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc)) 1769 goto nla_put_failure; 1770 1771 if (fi->nh) { 1772 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id)) 1773 goto nla_put_failure; 1774 if (nexthop_is_blackhole(fi->nh)) 1775 rtm->rtm_type = RTN_BLACKHOLE; 1776 if (!READ_ONCE(fi->fib_net->ipv4.sysctl_nexthop_compat_mode)) 1777 goto offload; 1778 } 1779 1780 if (nhs == 1) { 1781 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0); 1782 unsigned char flags = 0; 1783 1784 if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0) 1785 goto nla_put_failure; 1786 1787 rtm->rtm_flags = flags; 1788 #ifdef CONFIG_IP_ROUTE_CLASSID 1789 if (nhc->nhc_family == AF_INET) { 1790 struct fib_nh *nh; 1791 1792 nh = container_of(nhc, struct fib_nh, nh_common); 1793 if (nh->nh_tclassid && 1794 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid)) 1795 goto nla_put_failure; 1796 } 1797 #endif 1798 } else { 1799 if (fib_add_multipath(skb, fi) < 0) 1800 goto nla_put_failure; 1801 } 1802 1803 offload: 1804 if (fri->offload) 1805 rtm->rtm_flags |= RTM_F_OFFLOAD; 1806 if (fri->trap) 1807 rtm->rtm_flags |= RTM_F_TRAP; 1808 if (fri->offload_failed) 1809 rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED; 1810 1811 nlmsg_end(skb, nlh); 1812 return 0; 1813 1814 nla_put_failure: 1815 nlmsg_cancel(skb, nlh); 1816 return -EMSGSIZE; 1817 } 1818 1819 /* 1820 * Update FIB if: 1821 * - local address disappeared -> we must delete all the entries 1822 * referring to it. 1823 * - device went down -> we must shutdown all nexthops going via it. 1824 */ 1825 int fib_sync_down_addr(struct net_device *dev, __be32 local) 1826 { 1827 int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN; 1828 struct net *net = dev_net(dev); 1829 struct hlist_head *head; 1830 struct fib_info *fi; 1831 int ret = 0; 1832 1833 if (!local) 1834 return 0; 1835 1836 head = fib_info_laddrhash_bucket(net, local); 1837 hlist_for_each_entry(fi, head, fib_lhash) { 1838 if (!net_eq(fi->fib_net, net) || 1839 fi->fib_tb_id != tb_id) 1840 continue; 1841 if (fi->fib_prefsrc == local) { 1842 fi->fib_flags |= RTNH_F_DEAD; 1843 fi->pfsrc_removed = true; 1844 ret++; 1845 } 1846 } 1847 return ret; 1848 } 1849 1850 static int call_fib_nh_notifiers(struct fib_nh *nh, 1851 enum fib_event_type event_type) 1852 { 1853 bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev); 1854 struct fib_nh_notifier_info info = { 1855 .fib_nh = nh, 1856 }; 1857 1858 switch (event_type) { 1859 case FIB_EVENT_NH_ADD: 1860 if (nh->fib_nh_flags & RTNH_F_DEAD) 1861 break; 1862 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) 1863 break; 1864 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type, 1865 &info.info); 1866 case FIB_EVENT_NH_DEL: 1867 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) || 1868 (nh->fib_nh_flags & RTNH_F_DEAD)) 1869 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), 1870 event_type, &info.info); 1871 break; 1872 default: 1873 break; 1874 } 1875 1876 return NOTIFY_DONE; 1877 } 1878 1879 /* Update the PMTU of exceptions when: 1880 * - the new MTU of the first hop becomes smaller than the PMTU 1881 * - the old MTU was the same as the PMTU, and it limited discovery of 1882 * larger MTUs on the path. With that limit raised, we can now 1883 * discover larger MTUs 1884 * A special case is locked exceptions, for which the PMTU is smaller 1885 * than the minimal accepted PMTU: 1886 * - if the new MTU is greater than the PMTU, don't make any change 1887 * - otherwise, unlock and set PMTU 1888 */ 1889 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig) 1890 { 1891 struct fnhe_hash_bucket *bucket; 1892 int i; 1893 1894 bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1); 1895 if (!bucket) 1896 return; 1897 1898 for (i = 0; i < FNHE_HASH_SIZE; i++) { 1899 struct fib_nh_exception *fnhe; 1900 1901 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1); 1902 fnhe; 1903 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) { 1904 if (fnhe->fnhe_mtu_locked) { 1905 if (new <= fnhe->fnhe_pmtu) { 1906 fnhe->fnhe_pmtu = new; 1907 fnhe->fnhe_mtu_locked = false; 1908 } 1909 } else if (new < fnhe->fnhe_pmtu || 1910 orig == fnhe->fnhe_pmtu) { 1911 fnhe->fnhe_pmtu = new; 1912 } 1913 } 1914 } 1915 } 1916 1917 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu) 1918 { 1919 struct hlist_head *head = fib_nh_head(dev); 1920 struct fib_nh *nh; 1921 1922 hlist_for_each_entry(nh, head, nh_hash) { 1923 DEBUG_NET_WARN_ON_ONCE(nh->fib_nh_dev != dev); 1924 fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu); 1925 } 1926 } 1927 1928 /* Event force Flags Description 1929 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host 1930 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host 1931 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed 1932 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed 1933 * 1934 * only used when fib_nh is built into fib_info 1935 */ 1936 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force) 1937 { 1938 struct hlist_head *head = fib_nh_head(dev); 1939 struct fib_info *prev_fi = NULL; 1940 int scope = RT_SCOPE_NOWHERE; 1941 struct fib_nh *nh; 1942 int ret = 0; 1943 1944 if (force) 1945 scope = -1; 1946 1947 hlist_for_each_entry(nh, head, nh_hash) { 1948 struct fib_info *fi = nh->nh_parent; 1949 int dead; 1950 1951 BUG_ON(!fi->fib_nhs); 1952 DEBUG_NET_WARN_ON_ONCE(nh->fib_nh_dev != dev); 1953 if (fi == prev_fi) 1954 continue; 1955 prev_fi = fi; 1956 dead = 0; 1957 change_nexthops(fi) { 1958 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) 1959 dead++; 1960 else if (nexthop_nh->fib_nh_dev == dev && 1961 nexthop_nh->fib_nh_scope != scope) { 1962 switch (event) { 1963 case NETDEV_DOWN: 1964 case NETDEV_UNREGISTER: 1965 nexthop_nh->fib_nh_flags |= RTNH_F_DEAD; 1966 fallthrough; 1967 case NETDEV_CHANGE: 1968 nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN; 1969 break; 1970 } 1971 call_fib_nh_notifiers(nexthop_nh, 1972 FIB_EVENT_NH_DEL); 1973 dead++; 1974 } 1975 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1976 if (event == NETDEV_UNREGISTER && 1977 nexthop_nh->fib_nh_dev == dev) { 1978 dead = fi->fib_nhs; 1979 break; 1980 } 1981 #endif 1982 } endfor_nexthops(fi) 1983 if (dead == fi->fib_nhs) { 1984 switch (event) { 1985 case NETDEV_DOWN: 1986 case NETDEV_UNREGISTER: 1987 fi->fib_flags |= RTNH_F_DEAD; 1988 fallthrough; 1989 case NETDEV_CHANGE: 1990 fi->fib_flags |= RTNH_F_LINKDOWN; 1991 break; 1992 } 1993 ret++; 1994 } 1995 1996 fib_rebalance(fi); 1997 } 1998 1999 return ret; 2000 } 2001 2002 /* Must be invoked inside of an RCU protected region. */ 2003 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res) 2004 { 2005 struct fib_info *fi = NULL, *last_resort = NULL; 2006 struct hlist_head *fa_head = res->fa_head; 2007 struct fib_table *tb = res->table; 2008 u8 slen = 32 - res->prefixlen; 2009 int order = -1, last_idx = -1; 2010 struct fib_alias *fa, *fa1 = NULL; 2011 u32 last_prio = res->fi->fib_priority; 2012 dscp_t last_dscp = 0; 2013 2014 hlist_for_each_entry_rcu(fa, fa_head, fa_list) { 2015 struct fib_info *next_fi = fa->fa_info; 2016 struct fib_nh_common *nhc; 2017 2018 if (fa->fa_slen != slen) 2019 continue; 2020 if (fa->fa_dscp && !fib_dscp_masked_match(fa->fa_dscp, flp)) 2021 continue; 2022 if (fa->tb_id != tb->tb_id) 2023 continue; 2024 if (next_fi->fib_priority > last_prio && 2025 fa->fa_dscp == last_dscp) { 2026 if (last_dscp) 2027 continue; 2028 break; 2029 } 2030 if (next_fi->fib_flags & RTNH_F_DEAD) 2031 continue; 2032 last_dscp = fa->fa_dscp; 2033 last_prio = next_fi->fib_priority; 2034 2035 if (next_fi->fib_scope != res->scope || 2036 fa->fa_type != RTN_UNICAST) 2037 continue; 2038 2039 nhc = fib_info_nhc(next_fi, 0); 2040 if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK) 2041 continue; 2042 2043 fib_alias_accessed(fa); 2044 2045 if (!fi) { 2046 if (next_fi != res->fi) 2047 break; 2048 fa1 = fa; 2049 } else if (!fib_detect_death(fi, order, &last_resort, 2050 &last_idx, fa1->fa_default)) { 2051 fib_result_assign(res, fi); 2052 fa1->fa_default = order; 2053 goto out; 2054 } 2055 fi = next_fi; 2056 order++; 2057 } 2058 2059 if (order <= 0 || !fi) { 2060 if (fa1) 2061 fa1->fa_default = -1; 2062 goto out; 2063 } 2064 2065 if (!fib_detect_death(fi, order, &last_resort, &last_idx, 2066 fa1->fa_default)) { 2067 fib_result_assign(res, fi); 2068 fa1->fa_default = order; 2069 goto out; 2070 } 2071 2072 if (last_idx >= 0) 2073 fib_result_assign(res, last_resort); 2074 fa1->fa_default = last_idx; 2075 out: 2076 return; 2077 } 2078 2079 /* 2080 * Dead device goes up. We wake up dead nexthops. 2081 * It takes sense only on multipath routes. 2082 * 2083 * only used when fib_nh is built into fib_info 2084 */ 2085 int fib_sync_up(struct net_device *dev, unsigned char nh_flags) 2086 { 2087 struct fib_info *prev_fi; 2088 struct hlist_head *head; 2089 struct fib_nh *nh; 2090 int ret; 2091 2092 if (!(dev->flags & IFF_UP)) 2093 return 0; 2094 2095 if (nh_flags & RTNH_F_DEAD) { 2096 unsigned int flags = dev_get_flags(dev); 2097 2098 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 2099 nh_flags |= RTNH_F_LINKDOWN; 2100 } 2101 2102 prev_fi = NULL; 2103 head = fib_nh_head(dev); 2104 ret = 0; 2105 2106 hlist_for_each_entry(nh, head, nh_hash) { 2107 struct fib_info *fi = nh->nh_parent; 2108 int alive; 2109 2110 BUG_ON(!fi->fib_nhs); 2111 DEBUG_NET_WARN_ON_ONCE(nh->fib_nh_dev != dev); 2112 if (fi == prev_fi) 2113 continue; 2114 2115 prev_fi = fi; 2116 alive = 0; 2117 change_nexthops(fi) { 2118 if (!(nexthop_nh->fib_nh_flags & nh_flags)) { 2119 alive++; 2120 continue; 2121 } 2122 if (!nexthop_nh->fib_nh_dev || 2123 !(nexthop_nh->fib_nh_dev->flags & IFF_UP)) 2124 continue; 2125 if (nexthop_nh->fib_nh_dev != dev || 2126 !__in_dev_get_rtnl(dev)) 2127 continue; 2128 alive++; 2129 nexthop_nh->fib_nh_flags &= ~nh_flags; 2130 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD); 2131 } endfor_nexthops(fi) 2132 2133 if (alive > 0) { 2134 fi->fib_flags &= ~nh_flags; 2135 ret++; 2136 } 2137 2138 fib_rebalance(fi); 2139 } 2140 2141 return ret; 2142 } 2143 2144 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2145 static bool fib_good_nh(const struct fib_nh *nh) 2146 { 2147 int state = NUD_REACHABLE; 2148 2149 if (nh->fib_nh_scope == RT_SCOPE_LINK) { 2150 struct neighbour *n; 2151 2152 rcu_read_lock(); 2153 2154 if (likely(nh->fib_nh_gw_family == AF_INET)) 2155 n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev, 2156 (__force u32)nh->fib_nh_gw4); 2157 else if (nh->fib_nh_gw_family == AF_INET6) 2158 n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev, 2159 &nh->fib_nh_gw6); 2160 else 2161 n = NULL; 2162 if (n) 2163 state = READ_ONCE(n->nud_state); 2164 2165 rcu_read_unlock(); 2166 } 2167 2168 return !!(state & NUD_VALID); 2169 } 2170 2171 void fib_select_multipath(struct fib_result *res, int hash) 2172 { 2173 struct fib_info *fi = res->fi; 2174 struct net *net = fi->fib_net; 2175 bool first = false; 2176 2177 if (unlikely(res->fi->nh)) { 2178 nexthop_path_fib_result(res, hash); 2179 return; 2180 } 2181 2182 change_nexthops(fi) { 2183 if (READ_ONCE(net->ipv4.sysctl_fib_multipath_use_neigh)) { 2184 if (!fib_good_nh(nexthop_nh)) 2185 continue; 2186 if (!first) { 2187 res->nh_sel = nhsel; 2188 res->nhc = &nexthop_nh->nh_common; 2189 first = true; 2190 } 2191 } 2192 2193 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound)) 2194 continue; 2195 2196 res->nh_sel = nhsel; 2197 res->nhc = &nexthop_nh->nh_common; 2198 return; 2199 } endfor_nexthops(fi); 2200 } 2201 #endif 2202 2203 void fib_select_path(struct net *net, struct fib_result *res, 2204 struct flowi4 *fl4, const struct sk_buff *skb) 2205 { 2206 if (fl4->flowi4_oif) 2207 goto check_saddr; 2208 2209 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2210 if (fib_info_num_path(res->fi) > 1) { 2211 int h = fib_multipath_hash(net, fl4, skb, NULL); 2212 2213 fib_select_multipath(res, h); 2214 } 2215 else 2216 #endif 2217 if (!res->prefixlen && 2218 res->table->tb_num_default > 1 && 2219 res->type == RTN_UNICAST) 2220 fib_select_default(fl4, res); 2221 2222 check_saddr: 2223 if (!fl4->saddr) { 2224 struct net_device *l3mdev; 2225 2226 l3mdev = dev_get_by_index_rcu(net, fl4->flowi4_l3mdev); 2227 2228 if (!l3mdev || 2229 l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) == l3mdev) 2230 fl4->saddr = fib_result_prefsrc(net, res); 2231 else 2232 fl4->saddr = inet_select_addr(l3mdev, 0, RT_SCOPE_LINK); 2233 } 2234 } 2235 2236 int __net_init fib4_semantics_init(struct net *net) 2237 { 2238 unsigned int hash_bits = 4; 2239 2240 net->ipv4.fib_info_hash = fib_info_hash_alloc(hash_bits); 2241 if (!net->ipv4.fib_info_hash) 2242 return -ENOMEM; 2243 2244 net->ipv4.fib_info_hash_bits = hash_bits; 2245 net->ipv4.fib_info_cnt = 0; 2246 2247 return 0; 2248 } 2249 2250 void __net_exit fib4_semantics_exit(struct net *net) 2251 { 2252 fib_info_hash_free(net->ipv4.fib_info_hash); 2253 } 2254