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 } 210 211 void fib_nh_release(struct net *net, struct fib_nh *fib_nh) 212 { 213 #ifdef CONFIG_IP_ROUTE_CLASSID 214 if (fib_nh->nh_tclassid) 215 atomic_dec(&net->ipv4.fib_num_tclassid_users); 216 #endif 217 fib_nh_common_release(&fib_nh->nh_common); 218 } 219 220 /* Release a nexthop info record */ 221 static void free_fib_info_rcu(struct rcu_head *head) 222 { 223 struct fib_info *fi = container_of(head, struct fib_info, rcu); 224 225 if (fi->nh) { 226 nexthop_put(fi->nh); 227 } else { 228 change_nexthops(fi) { 229 fib_nh_release(fi->fib_net, nexthop_nh); 230 } endfor_nexthops(fi); 231 } 232 233 ip_fib_metrics_put(fi->fib_metrics); 234 235 kfree(fi); 236 } 237 238 void free_fib_info(struct fib_info *fi) 239 { 240 if (fi->fib_dead == 0) { 241 pr_warn("Freeing alive fib_info %p\n", fi); 242 return; 243 } 244 245 call_rcu_hurry(&fi->rcu, free_fib_info_rcu); 246 } 247 EXPORT_SYMBOL_GPL(free_fib_info); 248 249 void fib_release_info(struct fib_info *fi) 250 { 251 ASSERT_RTNL(); 252 if (fi && refcount_dec_and_test(&fi->fib_treeref)) { 253 hlist_del(&fi->fib_hash); 254 fi->fib_net->ipv4.fib_info_cnt--; 255 256 if (fi->fib_prefsrc) 257 hlist_del(&fi->fib_lhash); 258 if (fi->nh) { 259 list_del(&fi->nh_list); 260 } else { 261 change_nexthops(fi) { 262 if (!nexthop_nh->fib_nh_dev) 263 continue; 264 hlist_del_rcu(&nexthop_nh->nh_hash); 265 } endfor_nexthops(fi) 266 } 267 /* Paired with READ_ONCE() from fib_table_lookup() */ 268 WRITE_ONCE(fi->fib_dead, 1); 269 fib_info_put(fi); 270 } 271 } 272 273 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi) 274 { 275 const struct fib_nh *onh; 276 277 if (fi->nh || ofi->nh) 278 return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1; 279 280 if (ofi->fib_nhs == 0) 281 return 0; 282 283 for_nexthops(fi) { 284 onh = fib_info_nh(ofi, nhsel); 285 286 if (nh->fib_nh_oif != onh->fib_nh_oif || 287 nh->fib_nh_gw_family != onh->fib_nh_gw_family || 288 nh->fib_nh_scope != onh->fib_nh_scope || 289 #ifdef CONFIG_IP_ROUTE_MULTIPATH 290 nh->fib_nh_weight != onh->fib_nh_weight || 291 #endif 292 #ifdef CONFIG_IP_ROUTE_CLASSID 293 nh->nh_tclassid != onh->nh_tclassid || 294 #endif 295 lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) || 296 ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK)) 297 return -1; 298 299 if (nh->fib_nh_gw_family == AF_INET && 300 nh->fib_nh_gw4 != onh->fib_nh_gw4) 301 return -1; 302 303 if (nh->fib_nh_gw_family == AF_INET6 && 304 ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6)) 305 return -1; 306 } endfor_nexthops(fi); 307 return 0; 308 } 309 310 static struct hlist_head *fib_nh_head(struct net_device *dev) 311 { 312 return &dev->fib_nh_head; 313 } 314 315 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope, 316 u32 prefsrc, u32 priority) 317 { 318 unsigned int val = init_val; 319 320 val ^= (protocol << 8) | scope; 321 val ^= prefsrc; 322 val ^= priority; 323 324 return val; 325 } 326 327 static unsigned int fib_info_hashfn_result(const struct net *net, 328 unsigned int val) 329 { 330 return hash_32(val ^ net_hash_mix(net), net->ipv4.fib_info_hash_bits); 331 } 332 333 static struct hlist_head *fib_info_hash_bucket(struct fib_info *fi) 334 { 335 struct net *net = fi->fib_net; 336 unsigned int val; 337 338 val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol, 339 fi->fib_scope, (__force u32)fi->fib_prefsrc, 340 fi->fib_priority); 341 342 if (fi->nh) { 343 val ^= fi->nh->id; 344 } else { 345 for_nexthops(fi) { 346 val ^= nh->fib_nh_oif; 347 } endfor_nexthops(fi) 348 } 349 350 return &net->ipv4.fib_info_hash[fib_info_hashfn_result(net, val)]; 351 } 352 353 static struct hlist_head *fib_info_laddrhash_bucket(const struct net *net, 354 __be32 val) 355 { 356 unsigned int hash_bits = net->ipv4.fib_info_hash_bits; 357 u32 slot; 358 359 slot = hash_32(net_hash_mix(net) ^ (__force u32)val, hash_bits); 360 361 return &net->ipv4.fib_info_hash[(1 << hash_bits) + slot]; 362 } 363 364 static struct hlist_head *fib_info_hash_alloc(unsigned int hash_bits) 365 { 366 /* The second half is used for prefsrc */ 367 return kvzalloc_objs(struct hlist_head, (1 << hash_bits) * 2); 368 } 369 370 static void fib_info_hash_free(struct hlist_head *head) 371 { 372 kvfree(head); 373 } 374 375 static void fib_info_hash_grow(struct net *net) 376 { 377 unsigned int old_size = 1 << net->ipv4.fib_info_hash_bits; 378 struct hlist_head *new_info_hash, *old_info_hash; 379 unsigned int i; 380 381 if (net->ipv4.fib_info_cnt < old_size) 382 return; 383 384 new_info_hash = fib_info_hash_alloc(net->ipv4.fib_info_hash_bits + 1); 385 if (!new_info_hash) 386 return; 387 388 old_info_hash = net->ipv4.fib_info_hash; 389 net->ipv4.fib_info_hash = new_info_hash; 390 net->ipv4.fib_info_hash_bits += 1; 391 392 for (i = 0; i < old_size; i++) { 393 struct hlist_head *head = &old_info_hash[i]; 394 struct hlist_node *n; 395 struct fib_info *fi; 396 397 hlist_for_each_entry_safe(fi, n, head, fib_hash) 398 hlist_add_head(&fi->fib_hash, fib_info_hash_bucket(fi)); 399 } 400 401 for (i = 0; i < old_size; i++) { 402 struct hlist_head *lhead = &old_info_hash[old_size + i]; 403 struct hlist_node *n; 404 struct fib_info *fi; 405 406 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) 407 hlist_add_head(&fi->fib_lhash, 408 fib_info_laddrhash_bucket(fi->fib_net, 409 fi->fib_prefsrc)); 410 } 411 412 fib_info_hash_free(old_info_hash); 413 } 414 415 /* no metrics, only nexthop id */ 416 static struct fib_info *fib_find_info_nh(struct net *net, 417 const struct fib_config *cfg) 418 { 419 struct hlist_head *head; 420 struct fib_info *fi; 421 unsigned int hash; 422 423 hash = fib_info_hashfn_1(cfg->fc_nh_id, 424 cfg->fc_protocol, cfg->fc_scope, 425 (__force u32)cfg->fc_prefsrc, 426 cfg->fc_priority); 427 hash = fib_info_hashfn_result(net, hash); 428 head = &net->ipv4.fib_info_hash[hash]; 429 430 hlist_for_each_entry(fi, head, fib_hash) { 431 if (!fi->nh || fi->nh->id != cfg->fc_nh_id) 432 continue; 433 434 if (cfg->fc_protocol == fi->fib_protocol && 435 cfg->fc_scope == fi->fib_scope && 436 cfg->fc_prefsrc == fi->fib_prefsrc && 437 cfg->fc_priority == fi->fib_priority && 438 cfg->fc_type == fi->fib_type && 439 cfg->fc_table == fi->fib_tb_id && 440 !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK)) 441 return fi; 442 } 443 444 return NULL; 445 } 446 447 static struct fib_info *fib_find_info(struct fib_info *nfi) 448 { 449 struct hlist_head *head = fib_info_hash_bucket(nfi); 450 struct fib_info *fi; 451 452 hlist_for_each_entry(fi, head, fib_hash) { 453 if (fi->fib_nhs != nfi->fib_nhs) 454 continue; 455 456 if (nfi->fib_protocol == fi->fib_protocol && 457 nfi->fib_scope == fi->fib_scope && 458 nfi->fib_prefsrc == fi->fib_prefsrc && 459 nfi->fib_priority == fi->fib_priority && 460 nfi->fib_type == fi->fib_type && 461 nfi->fib_tb_id == fi->fib_tb_id && 462 memcmp(nfi->fib_metrics, fi->fib_metrics, 463 sizeof(u32) * RTAX_MAX) == 0 && 464 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) && 465 nh_comp(fi, nfi) == 0) 466 return fi; 467 } 468 469 return NULL; 470 } 471 472 /* Check, that the gateway is already configured. 473 * Used only by redirect accept routine, under rcu_read_lock(); 474 */ 475 int ip_fib_check_default(__be32 gw, struct net_device *dev) 476 { 477 struct hlist_head *head; 478 struct fib_nh *nh; 479 480 head = fib_nh_head(dev); 481 482 hlist_for_each_entry_rcu(nh, head, nh_hash) { 483 DEBUG_NET_WARN_ON_ONCE(nh->fib_nh_dev != dev); 484 if (nh->fib_nh_gw4 == gw && 485 !(nh->fib_nh_flags & RTNH_F_DEAD)) { 486 return 0; 487 } 488 } 489 490 return -1; 491 } 492 493 static size_t fib_nexthop_nlmsg_size(const struct fib_nh_common *nhc, 494 bool skip_oif) 495 { 496 size_t nhsize = 0; 497 498 switch (nhc->nhc_gw_family) { 499 case AF_INET: 500 nhsize += nla_total_size(4); /* RTA_GATEWAY */ 501 break; 502 case AF_INET6: 503 nhsize += nla_total_size(sizeof(struct rtvia) + 504 sizeof(struct in6_addr)); 505 break; 506 } 507 508 if (!skip_oif && nhc->nhc_dev) 509 nhsize += nla_total_size(4); /* RTA_OIF */ 510 511 if (nhc->nhc_lwtstate) { 512 /* RTA_ENCAP */ 513 nhsize += lwtunnel_get_encap_size(nhc->nhc_lwtstate); 514 /* RTA_ENCAP_TYPE */ 515 nhsize += nla_total_size(2); 516 } 517 518 return nhsize; 519 } 520 521 size_t fib_nlmsg_size(struct fib_info *fi) 522 { 523 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg)) 524 + nla_total_size(4) /* RTA_TABLE */ 525 + nla_total_size(4) /* RTA_DST */ 526 + nla_total_size(4) /* RTA_PRIORITY */ 527 + nla_total_size(4) /* RTA_PREFSRC */ 528 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */ 529 unsigned int nhs = fib_info_num_path(fi); 530 531 /* space for nested metrics */ 532 payload += nla_total_size((RTAX_MAX * nla_total_size(4))); 533 534 if (fi->nh) 535 payload += nla_total_size(4); /* RTA_NH_ID */ 536 537 if (nhs) { 538 size_t mpsize = 0; 539 unsigned int i; 540 541 for (i = 0; i < fib_info_num_path(fi); i++) { 542 struct fib_nh_common *nhc = fib_info_nhc(fi, i); 543 size_t nhsize; 544 545 if (!nhc) 546 break; 547 548 nhsize = fib_nexthop_nlmsg_size(nhc, nhs != 1); 549 550 if (nhs != 1) 551 nhsize += NLA_ALIGN(sizeof(struct rtnexthop)); 552 553 #ifdef CONFIG_IP_ROUTE_CLASSID 554 if (nhc->nhc_family == AF_INET) { 555 struct fib_nh *nh; 556 557 nh = container_of(nhc, struct fib_nh, nh_common); 558 if (nh->nh_tclassid) 559 nhsize += nla_total_size(4); 560 } 561 #endif 562 if (nhs == 1) 563 payload += nhsize; 564 else 565 mpsize += nhsize; 566 } 567 568 if (nhs != 1) 569 payload += nla_total_size(mpsize); 570 } 571 572 return payload; 573 } 574 575 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa, 576 int dst_len, u32 tb_id, const struct nl_info *info, 577 unsigned int nlm_flags) 578 { 579 struct fib_rt_info fri; 580 struct sk_buff *skb; 581 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0; 582 int err = -ENOBUFS; 583 584 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL); 585 if (!skb) 586 goto errout; 587 588 fri.fi = fa->fa_info; 589 fri.tb_id = tb_id; 590 fri.dst = key; 591 fri.dst_len = dst_len; 592 fri.dscp = fa->fa_dscp; 593 fri.type = fa->fa_type; 594 fri.offload = READ_ONCE(fa->offload); 595 fri.trap = READ_ONCE(fa->trap); 596 fri.offload_failed = READ_ONCE(fa->offload_failed); 597 err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags); 598 if (err < 0) { 599 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */ 600 WARN_ON(err == -EMSGSIZE); 601 kfree_skb(skb); 602 goto errout; 603 } 604 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE, 605 info->nlh, GFP_KERNEL); 606 return; 607 errout: 608 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err); 609 } 610 611 static int fib_detect_death(struct fib_info *fi, int order, 612 struct fib_info **last_resort, int *last_idx, 613 int dflt) 614 { 615 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0); 616 struct neighbour *n; 617 int state = NUD_NONE; 618 619 if (likely(nhc->nhc_gw_family == AF_INET)) 620 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev); 621 else if (IS_ENABLED(CONFIG_IPV6) && nhc->nhc_gw_family == AF_INET6) 622 n = neigh_lookup(&nd_tbl, &nhc->nhc_gw.ipv6, nhc->nhc_dev); 623 else 624 n = NULL; 625 626 if (n) { 627 state = READ_ONCE(n->nud_state); 628 neigh_release(n); 629 } else { 630 return 0; 631 } 632 if (state == NUD_REACHABLE) 633 return 0; 634 if ((state & NUD_VALID) && order != dflt) 635 return 0; 636 if ((state & NUD_VALID) || 637 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) { 638 *last_resort = fi; 639 *last_idx = order; 640 } 641 return 1; 642 } 643 644 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc, 645 struct nlattr *encap, u16 encap_type, 646 void *cfg, gfp_t gfp_flags, 647 struct netlink_ext_ack *extack) 648 { 649 int err; 650 651 nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *, 652 gfp_flags); 653 if (!nhc->nhc_pcpu_rth_output) 654 return -ENOMEM; 655 656 if (encap) { 657 struct lwtunnel_state *lwtstate; 658 659 err = lwtunnel_build_state(net, encap_type, encap, 660 nhc->nhc_family, cfg, &lwtstate, 661 extack); 662 if (err) 663 goto lwt_failure; 664 665 nhc->nhc_lwtstate = lwtstate_get(lwtstate); 666 } 667 668 return 0; 669 670 lwt_failure: 671 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output); 672 nhc->nhc_pcpu_rth_output = NULL; 673 return err; 674 } 675 676 int fib_nh_init(struct net *net, struct fib_nh *nh, 677 struct fib_config *cfg, int nh_weight, 678 struct netlink_ext_ack *extack) 679 { 680 int err; 681 682 nh->fib_nh_family = AF_INET; 683 684 err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap, 685 cfg->fc_encap_type, cfg, GFP_KERNEL, extack); 686 if (err) 687 return err; 688 689 nh->fib_nh_oif = cfg->fc_oif; 690 nh->fib_nh_gw_family = cfg->fc_gw_family; 691 if (cfg->fc_gw_family == AF_INET) 692 nh->fib_nh_gw4 = cfg->fc_gw4; 693 else if (cfg->fc_gw_family == AF_INET6) 694 nh->fib_nh_gw6 = cfg->fc_gw6; 695 696 nh->fib_nh_flags = cfg->fc_flags; 697 698 #ifdef CONFIG_IP_ROUTE_CLASSID 699 nh->nh_tclassid = cfg->fc_flow; 700 if (nh->nh_tclassid) 701 atomic_inc(&net->ipv4.fib_num_tclassid_users); 702 #endif 703 #ifdef CONFIG_IP_ROUTE_MULTIPATH 704 nh->fib_nh_weight = nh_weight; 705 #endif 706 return 0; 707 } 708 709 #ifdef CONFIG_IP_ROUTE_MULTIPATH 710 711 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining, 712 struct netlink_ext_ack *extack) 713 { 714 int nhs = 0; 715 716 while (rtnh_ok(rtnh, remaining)) { 717 nhs++; 718 rtnh = rtnh_next(rtnh, &remaining); 719 } 720 721 /* leftover implies invalid nexthop configuration, discard it */ 722 if (remaining > 0) { 723 NL_SET_ERR_MSG(extack, 724 "Invalid nexthop configuration - extra data after nexthops"); 725 nhs = 0; 726 } 727 728 return nhs; 729 } 730 731 static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla, 732 struct netlink_ext_ack *extack) 733 { 734 if (nla_len(nla) < sizeof(*gw)) { 735 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY"); 736 return -EINVAL; 737 } 738 739 *gw = nla_get_in_addr(nla); 740 741 return 0; 742 } 743 744 /* only called when fib_nh is integrated into fib_info */ 745 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh, 746 int remaining, struct fib_config *cfg, 747 struct netlink_ext_ack *extack) 748 { 749 struct net *net = fi->fib_net; 750 struct fib_config fib_cfg; 751 struct fib_nh *nh; 752 int ret; 753 754 change_nexthops(fi) { 755 int attrlen; 756 757 memset(&fib_cfg, 0, sizeof(fib_cfg)); 758 759 if (!rtnh_ok(rtnh, remaining)) { 760 NL_SET_ERR_MSG(extack, 761 "Invalid nexthop configuration - extra data after nexthop"); 762 return -EINVAL; 763 } 764 765 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) { 766 NL_SET_ERR_MSG(extack, 767 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN"); 768 return -EINVAL; 769 } 770 771 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags; 772 fib_cfg.fc_oif = rtnh->rtnh_ifindex; 773 774 attrlen = rtnh_attrlen(rtnh); 775 if (attrlen > 0) { 776 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh); 777 778 nla = nla_find(attrs, attrlen, RTA_GATEWAY); 779 nlav = nla_find(attrs, attrlen, RTA_VIA); 780 if (nla && nlav) { 781 NL_SET_ERR_MSG(extack, 782 "Nexthop configuration can not contain both GATEWAY and VIA"); 783 return -EINVAL; 784 } 785 if (nla) { 786 ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla, 787 extack); 788 if (ret) 789 goto errout; 790 791 if (fib_cfg.fc_gw4) 792 fib_cfg.fc_gw_family = AF_INET; 793 } else if (nlav) { 794 ret = fib_gw_from_via(&fib_cfg, nlav, extack); 795 if (ret) 796 goto errout; 797 } 798 799 nla = nla_find(attrs, attrlen, RTA_FLOW); 800 if (nla) { 801 if (nla_len(nla) < sizeof(u32)) { 802 NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW"); 803 return -EINVAL; 804 } 805 fib_cfg.fc_flow = nla_get_u32(nla); 806 } 807 808 fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP); 809 /* RTA_ENCAP_TYPE length checked in 810 * lwtunnel_valid_encap_type_attr 811 */ 812 nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE); 813 if (nla) 814 fib_cfg.fc_encap_type = nla_get_u16(nla); 815 } 816 817 ret = fib_nh_init(net, nexthop_nh, &fib_cfg, 818 rtnh->rtnh_hops + 1, extack); 819 if (ret) 820 goto errout; 821 822 rtnh = rtnh_next(rtnh, &remaining); 823 } endfor_nexthops(fi); 824 825 ret = -EINVAL; 826 nh = fib_info_nh(fi, 0); 827 if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) { 828 NL_SET_ERR_MSG(extack, 829 "Nexthop device index does not match RTA_OIF"); 830 goto errout; 831 } 832 if (cfg->fc_gw_family) { 833 if (cfg->fc_gw_family != nh->fib_nh_gw_family || 834 (cfg->fc_gw_family == AF_INET && 835 nh->fib_nh_gw4 != cfg->fc_gw4) || 836 (cfg->fc_gw_family == AF_INET6 && 837 ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) { 838 NL_SET_ERR_MSG(extack, 839 "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA"); 840 goto errout; 841 } 842 } 843 #ifdef CONFIG_IP_ROUTE_CLASSID 844 if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) { 845 NL_SET_ERR_MSG(extack, 846 "Nexthop class id does not match RTA_FLOW"); 847 goto errout; 848 } 849 #endif 850 ret = 0; 851 errout: 852 return ret; 853 } 854 855 /* only called when fib_nh is integrated into fib_info */ 856 static void fib_rebalance(struct fib_info *fi) 857 { 858 int total; 859 int w; 860 861 if (fib_info_num_path(fi) < 2) 862 return; 863 864 total = 0; 865 for_nexthops(fi) { 866 if (nh->fib_nh_flags & RTNH_F_DEAD) 867 continue; 868 869 if (ip_ignore_linkdown(nh->fib_nh_dev) && 870 nh->fib_nh_flags & RTNH_F_LINKDOWN) 871 continue; 872 873 total += nh->fib_nh_weight; 874 } endfor_nexthops(fi); 875 876 w = 0; 877 change_nexthops(fi) { 878 int upper_bound; 879 880 if (!total || nexthop_nh->fib_nh_flags & RTNH_F_DEAD) { 881 upper_bound = -1; 882 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) && 883 nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) { 884 upper_bound = -1; 885 } else { 886 w += nexthop_nh->fib_nh_weight; 887 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31, 888 total) - 1; 889 } 890 891 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound); 892 } endfor_nexthops(fi); 893 } 894 #else /* CONFIG_IP_ROUTE_MULTIPATH */ 895 896 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh, 897 int remaining, struct fib_config *cfg, 898 struct netlink_ext_ack *extack) 899 { 900 NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel"); 901 902 return -EINVAL; 903 } 904 905 #define fib_rebalance(fi) do { } while (0) 906 907 #endif /* CONFIG_IP_ROUTE_MULTIPATH */ 908 909 static int fib_encap_match(struct net *net, u16 encap_type, 910 struct nlattr *encap, 911 const struct fib_nh *nh, 912 const struct fib_config *cfg, 913 struct netlink_ext_ack *extack) 914 { 915 struct lwtunnel_state *lwtstate; 916 int ret, result = 0; 917 918 if (encap_type == LWTUNNEL_ENCAP_NONE) 919 return 0; 920 921 ret = lwtunnel_build_state(net, encap_type, encap, AF_INET, 922 cfg, &lwtstate, extack); 923 if (!ret) { 924 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws); 925 lwtstate_free(lwtstate); 926 } 927 928 return result; 929 } 930 931 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi, 932 struct netlink_ext_ack *extack) 933 { 934 #ifdef CONFIG_IP_ROUTE_MULTIPATH 935 struct rtnexthop *rtnh; 936 int remaining; 937 #endif 938 939 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority) 940 return 1; 941 942 if (cfg->fc_nh_id) { 943 if (fi->nh && cfg->fc_nh_id == fi->nh->id) 944 return 0; 945 return 1; 946 } 947 948 if (fi->nh) { 949 if (cfg->fc_oif || cfg->fc_gw_family || cfg->fc_mp) 950 return 1; 951 return 0; 952 } 953 954 if (cfg->fc_oif || cfg->fc_gw_family) { 955 struct fib_nh *nh; 956 957 nh = fib_info_nh(fi, 0); 958 if (cfg->fc_encap) { 959 if (fib_encap_match(net, cfg->fc_encap_type, 960 cfg->fc_encap, nh, cfg, extack)) 961 return 1; 962 } 963 #ifdef CONFIG_IP_ROUTE_CLASSID 964 if (cfg->fc_flow && 965 cfg->fc_flow != nh->nh_tclassid) 966 return 1; 967 #endif 968 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) || 969 (cfg->fc_gw_family && 970 cfg->fc_gw_family != nh->fib_nh_gw_family)) 971 return 1; 972 973 if (cfg->fc_gw_family == AF_INET && 974 cfg->fc_gw4 != nh->fib_nh_gw4) 975 return 1; 976 977 if (cfg->fc_gw_family == AF_INET6 && 978 ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6)) 979 return 1; 980 981 return 0; 982 } 983 984 #ifdef CONFIG_IP_ROUTE_MULTIPATH 985 if (!cfg->fc_mp) 986 return 0; 987 988 rtnh = cfg->fc_mp; 989 remaining = cfg->fc_mp_len; 990 991 for_nexthops(fi) { 992 int attrlen; 993 994 if (!rtnh_ok(rtnh, remaining)) 995 return -EINVAL; 996 997 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif) 998 return 1; 999 1000 attrlen = rtnh_attrlen(rtnh); 1001 if (attrlen > 0) { 1002 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh); 1003 int err; 1004 1005 nla = nla_find(attrs, attrlen, RTA_GATEWAY); 1006 nlav = nla_find(attrs, attrlen, RTA_VIA); 1007 if (nla && nlav) { 1008 NL_SET_ERR_MSG(extack, 1009 "Nexthop configuration can not contain both GATEWAY and VIA"); 1010 return -EINVAL; 1011 } 1012 1013 if (nla) { 1014 __be32 gw; 1015 1016 err = fib_gw_from_attr(&gw, nla, extack); 1017 if (err) 1018 return err; 1019 1020 if (nh->fib_nh_gw_family != AF_INET || 1021 gw != nh->fib_nh_gw4) 1022 return 1; 1023 } else if (nlav) { 1024 struct fib_config cfg2; 1025 1026 err = fib_gw_from_via(&cfg2, nlav, extack); 1027 if (err) 1028 return err; 1029 1030 switch (nh->fib_nh_gw_family) { 1031 case AF_INET: 1032 if (cfg2.fc_gw_family != AF_INET || 1033 cfg2.fc_gw4 != nh->fib_nh_gw4) 1034 return 1; 1035 break; 1036 case AF_INET6: 1037 if (cfg2.fc_gw_family != AF_INET6 || 1038 ipv6_addr_cmp(&cfg2.fc_gw6, 1039 &nh->fib_nh_gw6)) 1040 return 1; 1041 break; 1042 } 1043 } 1044 1045 #ifdef CONFIG_IP_ROUTE_CLASSID 1046 nla = nla_find(attrs, attrlen, RTA_FLOW); 1047 if (nla) { 1048 if (nla_len(nla) < sizeof(u32)) { 1049 NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW"); 1050 return -EINVAL; 1051 } 1052 if (nla_get_u32(nla) != nh->nh_tclassid) 1053 return 1; 1054 } 1055 #endif 1056 } 1057 1058 rtnh = rtnh_next(rtnh, &remaining); 1059 } endfor_nexthops(fi); 1060 #endif 1061 return 0; 1062 } 1063 1064 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi) 1065 { 1066 struct nlattr *nla; 1067 int remaining; 1068 1069 if (!cfg->fc_mx) 1070 return true; 1071 1072 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) { 1073 int type = nla_type(nla); 1074 u32 fi_val, val; 1075 1076 if (!type) 1077 continue; 1078 if (type > RTAX_MAX) 1079 return false; 1080 1081 type = array_index_nospec(type, RTAX_MAX + 1); 1082 if (type == RTAX_CC_ALGO) { 1083 char tmp[TCP_CA_NAME_MAX]; 1084 bool ecn_ca = false; 1085 1086 nla_strscpy(tmp, nla, sizeof(tmp)); 1087 val = tcp_ca_get_key_by_name(tmp, &ecn_ca); 1088 } else { 1089 if (nla_len(nla) != sizeof(u32)) 1090 return false; 1091 val = nla_get_u32(nla); 1092 } 1093 1094 fi_val = fi->fib_metrics->metrics[type - 1]; 1095 if (type == RTAX_FEATURES) 1096 fi_val &= ~DST_FEATURE_ECN_CA; 1097 1098 if (fi_val != val) 1099 return false; 1100 } 1101 1102 return true; 1103 } 1104 1105 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh, 1106 u32 table, struct netlink_ext_ack *extack) 1107 { 1108 struct fib6_config cfg = { 1109 .fc_table = table, 1110 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY, 1111 .fc_ifindex = nh->fib_nh_oif, 1112 .fc_gateway = nh->fib_nh_gw6, 1113 }; 1114 struct fib6_nh fib6_nh = {}; 1115 int err; 1116 1117 err = fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack); 1118 if (!err) { 1119 nh->fib_nh_dev = fib6_nh.fib_nh_dev; 1120 netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, 1121 GFP_KERNEL); 1122 nh->fib_nh_oif = nh->fib_nh_dev->ifindex; 1123 nh->fib_nh_scope = RT_SCOPE_LINK; 1124 1125 fib6_nh_release(&fib6_nh); 1126 } 1127 1128 return err; 1129 } 1130 1131 /* 1132 * Picture 1133 * ------- 1134 * 1135 * Semantics of nexthop is very messy by historical reasons. 1136 * We have to take into account, that: 1137 * a) gateway can be actually local interface address, 1138 * so that gatewayed route is direct. 1139 * b) gateway must be on-link address, possibly 1140 * described not by an ifaddr, but also by a direct route. 1141 * c) If both gateway and interface are specified, they should not 1142 * contradict. 1143 * d) If we use tunnel routes, gateway could be not on-link. 1144 * 1145 * Attempt to reconcile all of these (alas, self-contradictory) conditions 1146 * results in pretty ugly and hairy code with obscure logic. 1147 * 1148 * I chose to generalized it instead, so that the size 1149 * of code does not increase practically, but it becomes 1150 * much more general. 1151 * Every prefix is assigned a "scope" value: "host" is local address, 1152 * "link" is direct route, 1153 * [ ... "site" ... "interior" ... ] 1154 * and "universe" is true gateway route with global meaning. 1155 * 1156 * Every prefix refers to a set of "nexthop"s (gw, oif), 1157 * where gw must have narrower scope. This recursion stops 1158 * when gw has LOCAL scope or if "nexthop" is declared ONLINK, 1159 * which means that gw is forced to be on link. 1160 * 1161 * Code is still hairy, but now it is apparently logically 1162 * consistent and very flexible. F.e. as by-product it allows 1163 * to co-exists in peace independent exterior and interior 1164 * routing processes. 1165 * 1166 * Normally it looks as following. 1167 * 1168 * {universe prefix} -> (gw, oif) [scope link] 1169 * | 1170 * |-> {link prefix} -> (gw, oif) [scope local] 1171 * | 1172 * |-> {local prefix} (terminal node) 1173 */ 1174 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table, 1175 u8 scope, struct netlink_ext_ack *extack) 1176 { 1177 struct net_device *dev; 1178 struct fib_result res; 1179 int err = 0; 1180 1181 if (nh->fib_nh_flags & RTNH_F_ONLINK) { 1182 unsigned int addr_type; 1183 1184 if (scope >= RT_SCOPE_LINK) { 1185 NL_SET_ERR_MSG(extack, "Nexthop has invalid scope"); 1186 return -EINVAL; 1187 } 1188 dev = __dev_get_by_index(net, nh->fib_nh_oif); 1189 if (!dev) { 1190 NL_SET_ERR_MSG(extack, "Nexthop device required for onlink"); 1191 return -ENODEV; 1192 } 1193 if (!(dev->flags & IFF_UP)) { 1194 NL_SET_ERR_MSG(extack, "Nexthop device is not up"); 1195 return -ENETDOWN; 1196 } 1197 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4); 1198 if (addr_type != RTN_UNICAST) { 1199 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway"); 1200 return -EINVAL; 1201 } 1202 if (!netif_carrier_ok(dev)) 1203 nh->fib_nh_flags |= RTNH_F_LINKDOWN; 1204 nh->fib_nh_dev = dev; 1205 netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC); 1206 nh->fib_nh_scope = RT_SCOPE_LINK; 1207 return 0; 1208 } 1209 rcu_read_lock(); 1210 { 1211 struct fib_table *tbl = NULL; 1212 struct flowi4 fl4 = { 1213 .daddr = nh->fib_nh_gw4, 1214 .flowi4_scope = scope + 1, 1215 .flowi4_oif = nh->fib_nh_oif, 1216 .flowi4_iif = LOOPBACK_IFINDEX, 1217 }; 1218 1219 /* It is not necessary, but requires a bit of thinking */ 1220 if (fl4.flowi4_scope < RT_SCOPE_LINK) 1221 fl4.flowi4_scope = RT_SCOPE_LINK; 1222 1223 if (table && table != RT_TABLE_MAIN) 1224 tbl = fib_get_table(net, table); 1225 1226 if (tbl) 1227 err = fib_table_lookup(tbl, &fl4, &res, 1228 FIB_LOOKUP_IGNORE_LINKSTATE | 1229 FIB_LOOKUP_NOREF); 1230 1231 /* on error or if no table given do full lookup. This 1232 * is needed for example when nexthops are in the local 1233 * table rather than the given table 1234 */ 1235 if (!tbl || err) { 1236 err = fib_lookup(net, &fl4, &res, 1237 FIB_LOOKUP_IGNORE_LINKSTATE); 1238 } 1239 1240 if (err) { 1241 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway"); 1242 goto out; 1243 } 1244 } 1245 1246 err = -EINVAL; 1247 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) { 1248 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway"); 1249 goto out; 1250 } 1251 nh->fib_nh_scope = res.scope; 1252 nh->fib_nh_oif = FIB_RES_OIF(res); 1253 nh->fib_nh_dev = dev = FIB_RES_DEV(res); 1254 if (!dev) { 1255 NL_SET_ERR_MSG(extack, 1256 "No egress device for nexthop gateway"); 1257 goto out; 1258 } 1259 netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC); 1260 if (!netif_carrier_ok(dev)) 1261 nh->fib_nh_flags |= RTNH_F_LINKDOWN; 1262 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN; 1263 out: 1264 rcu_read_unlock(); 1265 return err; 1266 } 1267 1268 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh, 1269 struct netlink_ext_ack *extack) 1270 { 1271 struct in_device *in_dev; 1272 int err; 1273 1274 if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) { 1275 NL_SET_ERR_MSG(extack, 1276 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set"); 1277 return -EINVAL; 1278 } 1279 1280 rcu_read_lock(); 1281 1282 err = -ENODEV; 1283 in_dev = inetdev_by_index(net, nh->fib_nh_oif); 1284 if (!in_dev) 1285 goto out; 1286 err = -ENETDOWN; 1287 if (!(in_dev->dev->flags & IFF_UP)) { 1288 NL_SET_ERR_MSG(extack, "Device for nexthop is not up"); 1289 goto out; 1290 } 1291 1292 nh->fib_nh_dev = in_dev->dev; 1293 netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC); 1294 nh->fib_nh_scope = RT_SCOPE_HOST; 1295 if (!netif_carrier_ok(nh->fib_nh_dev)) 1296 nh->fib_nh_flags |= RTNH_F_LINKDOWN; 1297 err = 0; 1298 out: 1299 rcu_read_unlock(); 1300 return err; 1301 } 1302 1303 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope, 1304 struct netlink_ext_ack *extack) 1305 { 1306 int err; 1307 1308 if (nh->fib_nh_gw_family == AF_INET) 1309 err = fib_check_nh_v4_gw(net, nh, table, scope, extack); 1310 else if (nh->fib_nh_gw_family == AF_INET6) 1311 err = fib_check_nh_v6_gw(net, nh, table, extack); 1312 else 1313 err = fib_check_nh_nongw(net, nh, extack); 1314 1315 return err; 1316 } 1317 1318 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc, 1319 unsigned char scope) 1320 { 1321 struct fib_nh *nh; 1322 __be32 saddr; 1323 1324 if (nhc->nhc_family != AF_INET) 1325 return inet_select_addr(nhc->nhc_dev, 0, scope); 1326 1327 nh = container_of(nhc, struct fib_nh, nh_common); 1328 saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope); 1329 1330 WRITE_ONCE(nh->nh_saddr, saddr); 1331 WRITE_ONCE(nh->nh_saddr_genid, atomic_read(&net->ipv4.dev_addr_genid)); 1332 1333 return saddr; 1334 } 1335 1336 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res) 1337 { 1338 struct fib_nh_common *nhc = res->nhc; 1339 1340 if (res->fi->fib_prefsrc) 1341 return res->fi->fib_prefsrc; 1342 1343 if (nhc->nhc_family == AF_INET) { 1344 struct fib_nh *nh; 1345 1346 nh = container_of(nhc, struct fib_nh, nh_common); 1347 if (READ_ONCE(nh->nh_saddr_genid) == 1348 atomic_read(&net->ipv4.dev_addr_genid)) 1349 return READ_ONCE(nh->nh_saddr); 1350 } 1351 1352 return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope); 1353 } 1354 1355 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc) 1356 { 1357 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst || 1358 fib_prefsrc != cfg->fc_dst) { 1359 u32 tb_id = cfg->fc_table; 1360 int rc; 1361 1362 if (tb_id == RT_TABLE_MAIN) 1363 tb_id = RT_TABLE_LOCAL; 1364 1365 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net, 1366 fib_prefsrc, tb_id); 1367 1368 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) { 1369 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net, 1370 fib_prefsrc, RT_TABLE_LOCAL); 1371 } 1372 1373 if (rc != RTN_LOCAL) 1374 return false; 1375 } 1376 return true; 1377 } 1378 1379 struct fib_info *fib_create_info(struct fib_config *cfg, 1380 struct netlink_ext_ack *extack) 1381 { 1382 int err; 1383 struct fib_info *fi = NULL; 1384 struct nexthop *nh = NULL; 1385 struct fib_info *ofi; 1386 int nhs = 1; 1387 struct net *net = cfg->fc_nlinfo.nl_net; 1388 1389 ASSERT_RTNL(); 1390 if (cfg->fc_type > RTN_MAX) 1391 goto err_inval; 1392 1393 /* Fast check to catch the most weird cases */ 1394 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) { 1395 NL_SET_ERR_MSG(extack, "Invalid scope"); 1396 goto err_inval; 1397 } 1398 1399 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) { 1400 NL_SET_ERR_MSG(extack, 1401 "Invalid rtm_flags - can not contain DEAD or LINKDOWN"); 1402 goto err_inval; 1403 } 1404 1405 if (cfg->fc_nh_id) { 1406 if (!cfg->fc_mx) { 1407 fi = fib_find_info_nh(net, cfg); 1408 if (fi) { 1409 refcount_inc(&fi->fib_treeref); 1410 return fi; 1411 } 1412 } 1413 1414 nh = nexthop_find_by_id(net, cfg->fc_nh_id); 1415 if (!nh) { 1416 NL_SET_ERR_MSG(extack, "Nexthop id does not exist"); 1417 goto err_inval; 1418 } 1419 nhs = 0; 1420 } 1421 1422 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1423 if (cfg->fc_mp) { 1424 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack); 1425 if (nhs == 0) 1426 goto err_inval; 1427 } 1428 #endif 1429 1430 fib_info_hash_grow(net); 1431 1432 fi = kzalloc_flex(*fi, fib_nh, nhs); 1433 if (!fi) { 1434 err = -ENOBUFS; 1435 goto failure; 1436 } 1437 1438 fi->fib_metrics = ip_fib_metrics_init(cfg->fc_mx, cfg->fc_mx_len, extack); 1439 if (IS_ERR(fi->fib_metrics)) { 1440 err = PTR_ERR(fi->fib_metrics); 1441 kfree(fi); 1442 return ERR_PTR(err); 1443 } 1444 1445 fi->fib_net = net; 1446 fi->fib_protocol = cfg->fc_protocol; 1447 fi->fib_scope = cfg->fc_scope; 1448 fi->fib_flags = cfg->fc_flags; 1449 fi->fib_priority = cfg->fc_priority; 1450 fi->fib_prefsrc = cfg->fc_prefsrc; 1451 fi->fib_type = cfg->fc_type; 1452 fi->fib_tb_id = cfg->fc_table; 1453 1454 fi->fib_nhs = nhs; 1455 if (nh) { 1456 if (!nexthop_get(nh)) { 1457 NL_SET_ERR_MSG(extack, "Nexthop has been deleted"); 1458 err = -EINVAL; 1459 } else { 1460 err = 0; 1461 fi->nh = nh; 1462 } 1463 } else { 1464 change_nexthops(fi) { 1465 nexthop_nh->nh_parent = fi; 1466 } endfor_nexthops(fi) 1467 1468 if (cfg->fc_mp) 1469 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, 1470 extack); 1471 else 1472 err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack); 1473 } 1474 1475 if (err != 0) 1476 goto failure; 1477 1478 if (fib_props[cfg->fc_type].error) { 1479 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) { 1480 NL_SET_ERR_MSG(extack, 1481 "Gateway, device and multipath can not be specified for this route type"); 1482 goto err_inval; 1483 } 1484 goto link_it; 1485 } else { 1486 switch (cfg->fc_type) { 1487 case RTN_UNICAST: 1488 case RTN_LOCAL: 1489 case RTN_BROADCAST: 1490 case RTN_ANYCAST: 1491 case RTN_MULTICAST: 1492 break; 1493 default: 1494 NL_SET_ERR_MSG(extack, "Invalid route type"); 1495 goto err_inval; 1496 } 1497 } 1498 1499 if (cfg->fc_scope > RT_SCOPE_HOST) { 1500 NL_SET_ERR_MSG(extack, "Invalid scope"); 1501 goto err_inval; 1502 } 1503 1504 if (fi->nh) { 1505 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack); 1506 if (err) 1507 goto failure; 1508 } else if (cfg->fc_scope == RT_SCOPE_HOST) { 1509 struct fib_nh *nh = fi->fib_nh; 1510 1511 /* Local address is added. */ 1512 if (nhs != 1) { 1513 NL_SET_ERR_MSG(extack, 1514 "Route with host scope can not have multiple nexthops"); 1515 goto err_inval; 1516 } 1517 if (nh->fib_nh_gw_family) { 1518 NL_SET_ERR_MSG(extack, 1519 "Route with host scope can not have a gateway"); 1520 goto err_inval; 1521 } 1522 nh->fib_nh_scope = RT_SCOPE_NOWHERE; 1523 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif); 1524 err = -ENODEV; 1525 if (!nh->fib_nh_dev) 1526 goto failure; 1527 netdev_tracker_alloc(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, 1528 GFP_KERNEL); 1529 } else { 1530 int linkdown = 0; 1531 1532 change_nexthops(fi) { 1533 err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh, 1534 cfg->fc_table, cfg->fc_scope, 1535 extack); 1536 if (err != 0) 1537 goto failure; 1538 if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) 1539 linkdown++; 1540 } endfor_nexthops(fi) 1541 if (linkdown == fi->fib_nhs) 1542 fi->fib_flags |= RTNH_F_LINKDOWN; 1543 } 1544 1545 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) { 1546 NL_SET_ERR_MSG(extack, "Invalid prefsrc address"); 1547 goto err_inval; 1548 } 1549 1550 if (!fi->nh) { 1551 change_nexthops(fi) { 1552 fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common, 1553 fi->fib_scope); 1554 if (nexthop_nh->fib_nh_gw_family == AF_INET6) 1555 fi->fib_nh_is_v6 = true; 1556 } endfor_nexthops(fi) 1557 1558 fib_rebalance(fi); 1559 } 1560 1561 link_it: 1562 ofi = fib_find_info(fi); 1563 if (ofi) { 1564 /* fib_table_lookup() should not see @fi yet. */ 1565 fi->fib_dead = 1; 1566 free_fib_info(fi); 1567 refcount_inc(&ofi->fib_treeref); 1568 return ofi; 1569 } 1570 1571 refcount_set(&fi->fib_treeref, 1); 1572 refcount_set(&fi->fib_clntref, 1); 1573 1574 net->ipv4.fib_info_cnt++; 1575 hlist_add_head(&fi->fib_hash, fib_info_hash_bucket(fi)); 1576 1577 if (fi->fib_prefsrc) { 1578 struct hlist_head *head; 1579 1580 head = fib_info_laddrhash_bucket(net, fi->fib_prefsrc); 1581 hlist_add_head(&fi->fib_lhash, head); 1582 } 1583 if (fi->nh) { 1584 list_add(&fi->nh_list, &nh->fi_list); 1585 } else { 1586 change_nexthops(fi) { 1587 struct hlist_head *head; 1588 1589 if (!nexthop_nh->fib_nh_dev) 1590 continue; 1591 head = fib_nh_head(nexthop_nh->fib_nh_dev); 1592 hlist_add_head_rcu(&nexthop_nh->nh_hash, head); 1593 } endfor_nexthops(fi) 1594 } 1595 return fi; 1596 1597 err_inval: 1598 err = -EINVAL; 1599 1600 failure: 1601 if (fi) { 1602 /* fib_table_lookup() should not see @fi yet. */ 1603 fi->fib_dead = 1; 1604 free_fib_info(fi); 1605 } 1606 1607 return ERR_PTR(err); 1608 } 1609 1610 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc, 1611 u8 rt_family, unsigned char *flags, bool skip_oif) 1612 { 1613 if (nhc->nhc_flags & RTNH_F_DEAD) 1614 *flags |= RTNH_F_DEAD; 1615 1616 if (nhc->nhc_flags & RTNH_F_LINKDOWN) { 1617 *flags |= RTNH_F_LINKDOWN; 1618 1619 rcu_read_lock(); 1620 switch (nhc->nhc_family) { 1621 case AF_INET: 1622 if (ip_ignore_linkdown(nhc->nhc_dev)) 1623 *flags |= RTNH_F_DEAD; 1624 break; 1625 case AF_INET6: 1626 if (ip6_ignore_linkdown(nhc->nhc_dev)) 1627 *flags |= RTNH_F_DEAD; 1628 break; 1629 } 1630 rcu_read_unlock(); 1631 } 1632 1633 switch (nhc->nhc_gw_family) { 1634 case AF_INET: 1635 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4)) 1636 goto nla_put_failure; 1637 break; 1638 case AF_INET6: 1639 /* if gateway family does not match nexthop family 1640 * gateway is encoded as RTA_VIA 1641 */ 1642 if (rt_family != nhc->nhc_gw_family) { 1643 int alen = sizeof(struct in6_addr); 1644 struct nlattr *nla; 1645 struct rtvia *via; 1646 1647 nla = nla_reserve(skb, RTA_VIA, alen + 2); 1648 if (!nla) 1649 goto nla_put_failure; 1650 1651 via = nla_data(nla); 1652 via->rtvia_family = AF_INET6; 1653 memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen); 1654 } else if (nla_put_in6_addr(skb, RTA_GATEWAY, 1655 &nhc->nhc_gw.ipv6) < 0) { 1656 goto nla_put_failure; 1657 } 1658 break; 1659 } 1660 1661 *flags |= (nhc->nhc_flags & 1662 (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP)); 1663 1664 if (!skip_oif && nhc->nhc_dev && 1665 nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex)) 1666 goto nla_put_failure; 1667 1668 if (lwtunnel_fill_encap(skb, nhc->nhc_lwtstate, 1669 RTA_ENCAP, RTA_ENCAP_TYPE) < 0) 1670 goto nla_put_failure; 1671 1672 return 0; 1673 1674 nla_put_failure: 1675 return -EMSGSIZE; 1676 } 1677 1678 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6) 1679 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc, 1680 int nh_weight, u8 rt_family, u32 nh_tclassid) 1681 { 1682 const struct net_device *dev = nhc->nhc_dev; 1683 struct rtnexthop *rtnh; 1684 unsigned char flags = 0; 1685 1686 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh)); 1687 if (!rtnh) 1688 goto nla_put_failure; 1689 1690 rtnh->rtnh_hops = nh_weight - 1; 1691 rtnh->rtnh_ifindex = dev ? dev->ifindex : 0; 1692 1693 if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0) 1694 goto nla_put_failure; 1695 1696 rtnh->rtnh_flags = flags; 1697 1698 if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid)) 1699 goto nla_put_failure; 1700 1701 /* length of rtnetlink header + attributes */ 1702 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh; 1703 1704 return 0; 1705 1706 nla_put_failure: 1707 return -EMSGSIZE; 1708 } 1709 #endif 1710 1711 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1712 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi) 1713 { 1714 struct nlattr *mp; 1715 1716 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH); 1717 if (!mp) 1718 goto nla_put_failure; 1719 1720 if (unlikely(fi->nh)) { 1721 if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0) 1722 goto nla_put_failure; 1723 goto mp_end; 1724 } 1725 1726 for_nexthops(fi) { 1727 u32 nh_tclassid = 0; 1728 #ifdef CONFIG_IP_ROUTE_CLASSID 1729 nh_tclassid = nh->nh_tclassid; 1730 #endif 1731 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight, 1732 AF_INET, nh_tclassid) < 0) 1733 goto nla_put_failure; 1734 } endfor_nexthops(fi); 1735 1736 mp_end: 1737 nla_nest_end(skb, mp); 1738 1739 return 0; 1740 1741 nla_put_failure: 1742 return -EMSGSIZE; 1743 } 1744 #else 1745 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi) 1746 { 1747 return 0; 1748 } 1749 #endif 1750 1751 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event, 1752 const struct fib_rt_info *fri, unsigned int flags) 1753 { 1754 unsigned int nhs = fib_info_num_path(fri->fi); 1755 struct fib_info *fi = fri->fi; 1756 u32 tb_id = fri->tb_id; 1757 struct nlmsghdr *nlh; 1758 struct rtmsg *rtm; 1759 1760 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags); 1761 if (!nlh) 1762 return -EMSGSIZE; 1763 1764 rtm = nlmsg_data(nlh); 1765 rtm->rtm_family = AF_INET; 1766 rtm->rtm_dst_len = fri->dst_len; 1767 rtm->rtm_src_len = 0; 1768 rtm->rtm_tos = inet_dscp_to_dsfield(fri->dscp); 1769 if (tb_id < 256) 1770 rtm->rtm_table = tb_id; 1771 else 1772 rtm->rtm_table = RT_TABLE_COMPAT; 1773 if (nla_put_u32(skb, RTA_TABLE, tb_id)) 1774 goto nla_put_failure; 1775 rtm->rtm_type = fri->type; 1776 rtm->rtm_flags = fi->fib_flags; 1777 rtm->rtm_scope = fi->fib_scope; 1778 rtm->rtm_protocol = fi->fib_protocol; 1779 1780 if (rtm->rtm_dst_len && 1781 nla_put_in_addr(skb, RTA_DST, fri->dst)) 1782 goto nla_put_failure; 1783 if (fi->fib_priority && 1784 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority)) 1785 goto nla_put_failure; 1786 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0) 1787 goto nla_put_failure; 1788 1789 if (fi->fib_prefsrc && 1790 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc)) 1791 goto nla_put_failure; 1792 1793 if (fi->nh) { 1794 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id)) 1795 goto nla_put_failure; 1796 if (nexthop_is_blackhole(fi->nh)) 1797 rtm->rtm_type = RTN_BLACKHOLE; 1798 if (!READ_ONCE(fi->fib_net->ipv4.sysctl_nexthop_compat_mode)) 1799 goto offload; 1800 } 1801 1802 if (nhs == 1) { 1803 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0); 1804 unsigned char flags = 0; 1805 1806 if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0) 1807 goto nla_put_failure; 1808 1809 rtm->rtm_flags = flags; 1810 #ifdef CONFIG_IP_ROUTE_CLASSID 1811 if (nhc->nhc_family == AF_INET) { 1812 struct fib_nh *nh; 1813 1814 nh = container_of(nhc, struct fib_nh, nh_common); 1815 if (nh->nh_tclassid && 1816 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid)) 1817 goto nla_put_failure; 1818 } 1819 #endif 1820 } else { 1821 if (fib_add_multipath(skb, fi) < 0) 1822 goto nla_put_failure; 1823 } 1824 1825 offload: 1826 if (fri->offload) 1827 rtm->rtm_flags |= RTM_F_OFFLOAD; 1828 if (fri->trap) 1829 rtm->rtm_flags |= RTM_F_TRAP; 1830 if (fri->offload_failed) 1831 rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED; 1832 1833 nlmsg_end(skb, nlh); 1834 return 0; 1835 1836 nla_put_failure: 1837 nlmsg_cancel(skb, nlh); 1838 return -EMSGSIZE; 1839 } 1840 1841 /* 1842 * Update FIB if: 1843 * - local address disappeared -> we must delete all the entries 1844 * referring to it. 1845 * - device went down -> we must shutdown all nexthops going via it. 1846 */ 1847 int fib_sync_down_addr(struct net_device *dev, __be32 local) 1848 { 1849 int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN; 1850 struct net *net = dev_net(dev); 1851 struct hlist_head *head; 1852 struct fib_info *fi; 1853 int ret = 0; 1854 1855 if (!local) 1856 return 0; 1857 1858 head = fib_info_laddrhash_bucket(net, local); 1859 hlist_for_each_entry(fi, head, fib_lhash) { 1860 if (!net_eq(fi->fib_net, net) || 1861 fi->fib_tb_id != tb_id) 1862 continue; 1863 if (fi->fib_prefsrc == local) { 1864 fi->fib_flags |= RTNH_F_DEAD; 1865 fi->pfsrc_removed = true; 1866 ret++; 1867 } 1868 } 1869 return ret; 1870 } 1871 1872 static int call_fib_nh_notifiers(struct fib_nh *nh, 1873 enum fib_event_type event_type) 1874 { 1875 bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev); 1876 struct fib_nh_notifier_info info = { 1877 .fib_nh = nh, 1878 }; 1879 1880 switch (event_type) { 1881 case FIB_EVENT_NH_ADD: 1882 if (nh->fib_nh_flags & RTNH_F_DEAD) 1883 break; 1884 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) 1885 break; 1886 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type, 1887 &info.info); 1888 case FIB_EVENT_NH_DEL: 1889 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) || 1890 (nh->fib_nh_flags & RTNH_F_DEAD)) 1891 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), 1892 event_type, &info.info); 1893 break; 1894 default: 1895 break; 1896 } 1897 1898 return NOTIFY_DONE; 1899 } 1900 1901 /* Walk the exceptions of a nexthop after its first hop MTU changed. The 1902 * chain is RCU protected here, while fnhe_update_pmtu() takes fnhe_lock 1903 * for the update of each entry. 1904 */ 1905 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig) 1906 { 1907 struct fnhe_hash_bucket *bucket; 1908 int i; 1909 1910 rcu_read_lock(); 1911 bucket = rcu_dereference(nhc->nhc_exceptions); 1912 if (!bucket) 1913 goto out; 1914 1915 for (i = 0; i < FNHE_HASH_SIZE; i++) { 1916 struct fib_nh_exception *fnhe; 1917 1918 for (fnhe = rcu_dereference(bucket[i].chain); 1919 fnhe; 1920 fnhe = rcu_dereference(fnhe->fnhe_next)) 1921 fnhe_update_pmtu(fnhe, new, orig); 1922 } 1923 out: 1924 rcu_read_unlock(); 1925 } 1926 1927 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu) 1928 { 1929 struct hlist_head *head = fib_nh_head(dev); 1930 struct fib_nh *nh; 1931 1932 hlist_for_each_entry(nh, head, nh_hash) { 1933 DEBUG_NET_WARN_ON_ONCE(nh->fib_nh_dev != dev); 1934 fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu); 1935 } 1936 } 1937 1938 /* Event force Flags Description 1939 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host 1940 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host 1941 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed 1942 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed 1943 * 1944 * only used when fib_nh is built into fib_info 1945 */ 1946 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force) 1947 { 1948 struct hlist_head *head = fib_nh_head(dev); 1949 struct fib_info *prev_fi = NULL; 1950 int scope = RT_SCOPE_NOWHERE; 1951 struct fib_nh *nh; 1952 int ret = 0; 1953 1954 if (force) 1955 scope = -1; 1956 1957 hlist_for_each_entry(nh, head, nh_hash) { 1958 struct fib_info *fi = nh->nh_parent; 1959 int dead; 1960 1961 BUG_ON(!fi->fib_nhs); 1962 DEBUG_NET_WARN_ON_ONCE(nh->fib_nh_dev != dev); 1963 if (fi == prev_fi) 1964 continue; 1965 prev_fi = fi; 1966 dead = 0; 1967 change_nexthops(fi) { 1968 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) 1969 dead++; 1970 else if (nexthop_nh->fib_nh_dev == dev && 1971 nexthop_nh->fib_nh_scope != scope) { 1972 switch (event) { 1973 case NETDEV_DOWN: 1974 case NETDEV_UNREGISTER: 1975 nexthop_nh->fib_nh_flags |= RTNH_F_DEAD; 1976 fallthrough; 1977 case NETDEV_CHANGE: 1978 nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN; 1979 break; 1980 } 1981 call_fib_nh_notifiers(nexthop_nh, 1982 FIB_EVENT_NH_DEL); 1983 dead++; 1984 } 1985 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1986 if (event == NETDEV_UNREGISTER && 1987 nexthop_nh->fib_nh_dev == dev) { 1988 dead = fi->fib_nhs; 1989 break; 1990 } 1991 #endif 1992 } endfor_nexthops(fi) 1993 if (dead == fi->fib_nhs) { 1994 switch (event) { 1995 case NETDEV_DOWN: 1996 case NETDEV_UNREGISTER: 1997 fi->fib_flags |= RTNH_F_DEAD; 1998 fallthrough; 1999 case NETDEV_CHANGE: 2000 fi->fib_flags |= RTNH_F_LINKDOWN; 2001 break; 2002 } 2003 ret++; 2004 } 2005 2006 fib_rebalance(fi); 2007 } 2008 2009 return ret; 2010 } 2011 2012 /* Must be invoked inside of an RCU protected region. */ 2013 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res) 2014 { 2015 struct fib_info *fi = NULL, *last_resort = NULL; 2016 struct hlist_head *fa_head = res->fa_head; 2017 struct fib_table *tb = res->table; 2018 u8 slen = 32 - res->prefixlen; 2019 int order = -1, last_idx = -1; 2020 struct fib_alias *fa, *fa1 = NULL; 2021 u32 last_prio = res->fi->fib_priority; 2022 dscp_t last_dscp = 0; 2023 2024 hlist_for_each_entry_rcu(fa, fa_head, fa_list) { 2025 struct fib_info *next_fi = fa->fa_info; 2026 struct fib_nh_common *nhc; 2027 2028 if (fa->fa_slen != slen) 2029 continue; 2030 if (fa->fa_dscp && !fib_dscp_masked_match(fa->fa_dscp, flp)) 2031 continue; 2032 if (fa->tb_id != tb->tb_id) 2033 continue; 2034 if (next_fi->fib_priority > last_prio && 2035 fa->fa_dscp == last_dscp) { 2036 if (last_dscp) 2037 continue; 2038 break; 2039 } 2040 if (next_fi->fib_flags & RTNH_F_DEAD) 2041 continue; 2042 last_dscp = fa->fa_dscp; 2043 last_prio = next_fi->fib_priority; 2044 2045 if (next_fi->fib_scope != res->scope || 2046 fa->fa_type != RTN_UNICAST) 2047 continue; 2048 2049 nhc = fib_info_nhc(next_fi, 0); 2050 if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK) 2051 continue; 2052 2053 fib_alias_accessed(fa); 2054 2055 if (!fi) { 2056 if (next_fi != res->fi) 2057 break; 2058 fa1 = fa; 2059 } else if (!fib_detect_death(fi, order, &last_resort, 2060 &last_idx, fa1->fa_default)) { 2061 fib_result_assign(res, fi); 2062 fa1->fa_default = order; 2063 goto out; 2064 } 2065 fi = next_fi; 2066 order++; 2067 } 2068 2069 if (order <= 0 || !fi) { 2070 if (fa1) 2071 fa1->fa_default = -1; 2072 goto out; 2073 } 2074 2075 if (!fib_detect_death(fi, order, &last_resort, &last_idx, 2076 fa1->fa_default)) { 2077 fib_result_assign(res, fi); 2078 fa1->fa_default = order; 2079 goto out; 2080 } 2081 2082 if (last_idx >= 0) 2083 fib_result_assign(res, last_resort); 2084 fa1->fa_default = last_idx; 2085 out: 2086 return; 2087 } 2088 2089 /* 2090 * Dead device goes up. We wake up dead nexthops. 2091 * It takes sense only on multipath routes. 2092 * 2093 * only used when fib_nh is built into fib_info 2094 */ 2095 int fib_sync_up(struct net_device *dev, unsigned char nh_flags) 2096 { 2097 struct fib_info *prev_fi; 2098 struct hlist_head *head; 2099 struct fib_nh *nh; 2100 int ret; 2101 2102 if (!(dev->flags & IFF_UP)) 2103 return 0; 2104 2105 if (nh_flags & RTNH_F_DEAD) { 2106 unsigned int flags = netif_get_flags(dev); 2107 2108 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 2109 nh_flags |= RTNH_F_LINKDOWN; 2110 } 2111 2112 prev_fi = NULL; 2113 head = fib_nh_head(dev); 2114 ret = 0; 2115 2116 hlist_for_each_entry(nh, head, nh_hash) { 2117 struct fib_info *fi = nh->nh_parent; 2118 int alive; 2119 2120 BUG_ON(!fi->fib_nhs); 2121 DEBUG_NET_WARN_ON_ONCE(nh->fib_nh_dev != dev); 2122 if (fi == prev_fi) 2123 continue; 2124 2125 prev_fi = fi; 2126 alive = 0; 2127 change_nexthops(fi) { 2128 if (!(nexthop_nh->fib_nh_flags & nh_flags)) { 2129 alive++; 2130 continue; 2131 } 2132 if (!nexthop_nh->fib_nh_dev || 2133 !(nexthop_nh->fib_nh_dev->flags & IFF_UP)) 2134 continue; 2135 if (nexthop_nh->fib_nh_dev != dev || 2136 !__in_dev_get_rtnl(dev)) 2137 continue; 2138 alive++; 2139 nexthop_nh->fib_nh_flags &= ~nh_flags; 2140 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD); 2141 } endfor_nexthops(fi) 2142 2143 if (alive > 0) { 2144 fi->fib_flags &= ~nh_flags; 2145 ret++; 2146 } 2147 2148 fib_rebalance(fi); 2149 } 2150 2151 return ret; 2152 } 2153 2154 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2155 static bool fib_good_nh(const struct fib_nh *nh) 2156 { 2157 int state = NUD_REACHABLE; 2158 2159 if (nh->fib_nh_scope == RT_SCOPE_LINK) { 2160 struct neighbour *n; 2161 2162 rcu_read_lock(); 2163 2164 if (likely(nh->fib_nh_gw_family == AF_INET)) 2165 n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev, 2166 (__force u32)nh->fib_nh_gw4); 2167 else if (IS_ENABLED(CONFIG_IPV6) && 2168 nh->fib_nh_gw_family == AF_INET6) 2169 n = __ipv6_neigh_lookup_noref(nh->fib_nh_dev, 2170 &nh->fib_nh_gw6); 2171 else 2172 n = NULL; 2173 if (n) 2174 state = READ_ONCE(n->nud_state); 2175 2176 rcu_read_unlock(); 2177 } 2178 2179 return !!(state & NUD_VALID); 2180 } 2181 2182 static __be32 fib_nh_saddr(struct net *net, const struct fib_info *fi, 2183 struct fib_nh *nh, int genid) 2184 { 2185 if (READ_ONCE(nh->nh_saddr_genid) == genid) 2186 return READ_ONCE(nh->nh_saddr); 2187 2188 return fib_info_update_nhc_saddr(net, &nh->nh_common, fi->fib_scope); 2189 } 2190 2191 void fib_select_multipath(struct fib_result *res, int hash, 2192 const struct flowi4 *fl4) 2193 { 2194 struct fib_info *fi = res->fi; 2195 struct net *net = fi->fib_net; 2196 bool use_neigh; 2197 int score = -1; 2198 __be32 saddr; 2199 int genid; 2200 2201 if (unlikely(res->fi->nh)) { 2202 nexthop_path_fib_result(res, hash); 2203 return; 2204 } 2205 2206 use_neigh = READ_ONCE(net->ipv4.sysctl_fib_multipath_use_neigh); 2207 saddr = fl4 ? fl4->saddr : 0; 2208 genid = saddr ? atomic_read(&net->ipv4.dev_addr_genid) : 0; 2209 2210 change_nexthops(fi) { 2211 int nh_upper_bound, nh_score = 0; 2212 2213 /* Nexthops without a carrier are assigned an upper bound of 2214 * minus one when "ignore_routes_with_linkdown" is set. 2215 */ 2216 nh_upper_bound = atomic_read(&nexthop_nh->fib_nh_upper_bound); 2217 if (nh_upper_bound == -1 || 2218 (use_neigh && !fib_good_nh(nexthop_nh))) 2219 continue; 2220 2221 if (saddr && fib_nh_saddr(net, fi, nexthop_nh, genid) == saddr) 2222 nh_score += 2; 2223 if (hash <= nh_upper_bound) 2224 nh_score++; 2225 if (score < nh_score) { 2226 res->nh_sel = nhsel; 2227 res->nhc = &nexthop_nh->nh_common; 2228 if (nh_score == 3 || (!saddr && nh_score == 1)) 2229 return; 2230 score = nh_score; 2231 } 2232 2233 } endfor_nexthops(fi); 2234 } 2235 #endif 2236 2237 void fib_select_path(struct net *net, struct fib_result *res, 2238 struct flowi4 *fl4, const struct sk_buff *skb) 2239 { 2240 if (fl4->flowi4_oif) 2241 goto check_saddr; 2242 2243 #ifdef CONFIG_IP_ROUTE_MULTIPATH 2244 if (fib_info_num_path(res->fi) > 1) { 2245 int h = fib_multipath_hash(net, fl4, skb, NULL); 2246 2247 fib_select_multipath(res, h, fl4); 2248 } 2249 else 2250 #endif 2251 if (!res->prefixlen && 2252 res->table->tb_num_default > 1 && 2253 res->type == RTN_UNICAST) 2254 fib_select_default(fl4, res); 2255 2256 check_saddr: 2257 if (!fl4->saddr) { 2258 struct net_device *l3mdev; 2259 2260 l3mdev = dev_get_by_index_rcu(net, fl4->flowi4_l3mdev); 2261 2262 if (!l3mdev || 2263 l3mdev_master_dev_rcu(FIB_RES_DEV(*res)) == l3mdev) 2264 fl4->saddr = fib_result_prefsrc(net, res); 2265 else 2266 fl4->saddr = inet_select_addr(l3mdev, 0, RT_SCOPE_LINK); 2267 } 2268 } 2269 2270 int __net_init fib4_semantics_init(struct net *net) 2271 { 2272 unsigned int hash_bits = 4; 2273 2274 net->ipv4.fib_info_hash = fib_info_hash_alloc(hash_bits); 2275 if (!net->ipv4.fib_info_hash) 2276 return -ENOMEM; 2277 2278 net->ipv4.fib_info_hash_bits = hash_bits; 2279 net->ipv4.fib_info_cnt = 0; 2280 2281 return 0; 2282 } 2283 2284 void __net_exit fib4_semantics_exit(struct net *net) 2285 { 2286 fib_info_hash_free(net->ipv4.fib_info_hash); 2287 } 2288