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