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: FIB frontend. 8 * 9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 10 */ 11 12 #include <linux/module.h> 13 #include <linux/uaccess.h> 14 #include <linux/bitops.h> 15 #include <linux/capability.h> 16 #include <linux/types.h> 17 #include <linux/kernel.h> 18 #include <linux/mm.h> 19 #include <linux/string.h> 20 #include <linux/socket.h> 21 #include <linux/sockios.h> 22 #include <linux/errno.h> 23 #include <linux/in.h> 24 #include <linux/inet.h> 25 #include <linux/inetdevice.h> 26 #include <linux/netdevice.h> 27 #include <linux/if_addr.h> 28 #include <linux/if_arp.h> 29 #include <linux/skbuff.h> 30 #include <linux/cache.h> 31 #include <linux/init.h> 32 #include <linux/list.h> 33 #include <linux/slab.h> 34 35 #include <net/flow.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/arp.h> 43 #include <net/ip_fib.h> 44 #include <net/nexthop.h> 45 #include <net/rtnetlink.h> 46 #include <net/xfrm.h> 47 #include <net/l3mdev.h> 48 #include <net/lwtunnel.h> 49 #include <trace/events/fib.h> 50 51 #ifndef CONFIG_IP_MULTIPLE_TABLES 52 53 static int __net_init fib4_rules_init(struct net *net) 54 { 55 struct fib_table *local_table, *main_table; 56 57 main_table = fib_trie_table(RT_TABLE_MAIN, NULL); 58 if (!main_table) 59 return -ENOMEM; 60 61 local_table = fib_trie_table(RT_TABLE_LOCAL, main_table); 62 if (!local_table) 63 goto fail; 64 65 hlist_add_head_rcu(&local_table->tb_hlist, 66 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]); 67 hlist_add_head_rcu(&main_table->tb_hlist, 68 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]); 69 return 0; 70 71 fail: 72 fib_free_table(main_table); 73 return -ENOMEM; 74 } 75 #else 76 77 struct fib_table *fib_new_table(struct net *net, u32 id) 78 { 79 struct fib_table *tb, *new_tb, *alias = NULL; 80 unsigned int h; 81 82 if (id == 0) 83 id = RT_TABLE_MAIN; 84 tb = fib_get_table(net, id); 85 if (tb) 86 return tb; 87 88 if (!check_net(net)) 89 return NULL; 90 91 if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules) 92 alias = fib_new_table(net, RT_TABLE_MAIN); 93 94 new_tb = fib_trie_table(id, alias); 95 if (!new_tb) 96 return NULL; 97 98 spin_lock(&net->ipv4.fib_table_hash_lock); 99 100 tb = fib_get_table(net, id); 101 if (tb) { 102 spin_unlock(&net->ipv4.fib_table_hash_lock); 103 fib_free_table(new_tb); 104 return tb; 105 } 106 107 tb = new_tb; 108 109 switch (id) { 110 case RT_TABLE_MAIN: 111 rcu_assign_pointer(net->ipv4.fib_main, tb); 112 break; 113 case RT_TABLE_DEFAULT: 114 rcu_assign_pointer(net->ipv4.fib_default, tb); 115 break; 116 default: 117 break; 118 } 119 120 h = id & (FIB_TABLE_HASHSZ - 1); 121 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]); 122 123 spin_unlock(&net->ipv4.fib_table_hash_lock); 124 125 return tb; 126 } 127 EXPORT_SYMBOL_GPL(fib_new_table); 128 129 struct fib_table *fib_get_table(struct net *net, u32 id) 130 { 131 struct fib_table *tb = NULL; 132 struct hlist_head *head; 133 unsigned int h; 134 135 if (id == 0) 136 id = RT_TABLE_MAIN; 137 h = id & (FIB_TABLE_HASHSZ - 1); 138 head = &net->ipv4.fib_table_hash[h]; 139 140 /* fib_table is not destroyed until ip_fib_net_exit() 141 * except for the merged main/local table. 142 * fib_unmerge() is called under RTNL, so other readers 143 * under RTNL (e.g. fib_flush(), fib_info_notify_update()) 144 * can safely traverse the list with rcu_dereference_raw(). 145 */ 146 hlist_for_each_entry_rcu(tb, head, tb_hlist, true) 147 if (tb->tb_id == id) 148 break; 149 150 return tb; 151 } 152 #endif /* CONFIG_IP_MULTIPLE_TABLES */ 153 154 static void fib_replace_table(struct net *net, struct fib_table *old, 155 struct fib_table *new) 156 { 157 #ifdef CONFIG_IP_MULTIPLE_TABLES 158 switch (new->tb_id) { 159 case RT_TABLE_MAIN: 160 rcu_assign_pointer(net->ipv4.fib_main, new); 161 break; 162 case RT_TABLE_DEFAULT: 163 rcu_assign_pointer(net->ipv4.fib_default, new); 164 break; 165 default: 166 break; 167 } 168 169 #endif 170 /* replace the old table in the hlist */ 171 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist); 172 } 173 174 int fib_unmerge(struct net *net) 175 { 176 struct fib_table *old, *new, *main_table; 177 178 /* attempt to fetch local table if it has been allocated */ 179 old = fib_get_table(net, RT_TABLE_LOCAL); 180 if (!old) 181 return 0; 182 183 new = fib_trie_unmerge(old); 184 if (!new) 185 return -ENOMEM; 186 187 /* table is already unmerged */ 188 if (new == old) 189 return 0; 190 191 /* replace merged table with clean table */ 192 fib_replace_table(net, old, new); 193 fib_free_table(old); 194 195 /* attempt to fetch main table if it has been allocated */ 196 main_table = fib_get_table(net, RT_TABLE_MAIN); 197 if (!main_table) 198 return 0; 199 200 /* flush local entries from main table */ 201 fib_table_flush_external(main_table); 202 203 return 0; 204 } 205 206 void fib_flush(struct net *net) 207 { 208 int flushed = 0; 209 unsigned int h; 210 211 for (h = 0; h < FIB_TABLE_HASHSZ; h++) { 212 struct hlist_head *head = &net->ipv4.fib_table_hash[h]; 213 struct fib_table *tb; 214 215 hlist_for_each_entry_rcu(tb, head, tb_hlist, true) 216 flushed += fib_table_flush(net, tb, false); 217 } 218 219 if (flushed) 220 rt_cache_flush(net); 221 } 222 223 /* 224 * Find address type as if only "dev" was present in the system. If 225 * on_dev is NULL then all interfaces are taken into consideration. 226 */ 227 static inline unsigned int __inet_dev_addr_type(struct net *net, 228 const struct net_device *dev, 229 __be32 addr, u32 tb_id) 230 { 231 struct flowi4 fl4 = { .daddr = addr }; 232 struct fib_result res; 233 unsigned int ret = RTN_BROADCAST; 234 struct fib_table *table; 235 236 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr)) 237 return RTN_BROADCAST; 238 if (ipv4_is_multicast(addr)) 239 return RTN_MULTICAST; 240 241 rcu_read_lock(); 242 243 table = fib_get_table(net, tb_id); 244 if (table) { 245 ret = RTN_UNICAST; 246 if (!fib_table_lookup(net, table, &fl4, &res, 247 FIB_LOOKUP_NOREF)) { 248 struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0); 249 250 if (!dev || dev == nhc->nhc_dev) 251 ret = res.type; 252 } 253 } 254 255 rcu_read_unlock(); 256 return ret; 257 } 258 259 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id) 260 { 261 return __inet_dev_addr_type(net, NULL, addr, tb_id); 262 } 263 EXPORT_SYMBOL(inet_addr_type_table); 264 265 unsigned int inet_addr_type(struct net *net, __be32 addr) 266 { 267 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL); 268 } 269 EXPORT_SYMBOL(inet_addr_type); 270 271 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev, 272 __be32 addr) 273 { 274 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL; 275 276 return __inet_dev_addr_type(net, dev, addr, rt_table); 277 } 278 EXPORT_SYMBOL(inet_dev_addr_type); 279 280 /* inet_addr_type with dev == NULL but using the table from a dev 281 * if one is associated 282 */ 283 unsigned int inet_addr_type_dev_table(struct net *net, 284 const struct net_device *dev, 285 __be32 addr) 286 { 287 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL; 288 289 return __inet_dev_addr_type(net, NULL, addr, rt_table); 290 } 291 EXPORT_SYMBOL(inet_addr_type_dev_table); 292 293 __be32 fib_compute_spec_dst(struct sk_buff *skb) 294 { 295 struct net_device *dev = skb->dev; 296 struct in_device *in_dev; 297 struct fib_result res; 298 struct rtable *rt; 299 struct net *net; 300 int scope; 301 302 rt = skb_rtable(skb); 303 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) == 304 RTCF_LOCAL) 305 return ip_hdr(skb)->daddr; 306 307 in_dev = __in_dev_get_rcu(dev); 308 309 net = dev_net(dev); 310 311 scope = RT_SCOPE_UNIVERSE; 312 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) { 313 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev); 314 struct flowi4 fl4 = { 315 .flowi4_iif = LOOPBACK_IFINDEX, 316 .flowi4_l3mdev = l3mdev_master_ifindex_rcu(dev), 317 .daddr = ip_hdr(skb)->saddr, 318 .flowi4_dscp = ip4h_dscp(ip_hdr(skb)), 319 .flowi4_scope = scope, 320 .flowi4_mark = vmark ? skb->mark : 0, 321 }; 322 if (!fib_lookup(net, &fl4, &res, 0)) 323 return fib_result_prefsrc(net, &res); 324 } else { 325 scope = RT_SCOPE_LINK; 326 } 327 328 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope); 329 } 330 331 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev) 332 { 333 bool dev_match = false; 334 #ifdef CONFIG_IP_ROUTE_MULTIPATH 335 if (unlikely(fi->nh)) { 336 dev_match = nexthop_uses_dev(fi->nh, dev); 337 } else { 338 int ret; 339 340 for (ret = 0; ret < fib_info_num_path(fi); ret++) { 341 const struct fib_nh_common *nhc = fib_info_nhc(fi, ret); 342 343 if (nhc_l3mdev_matches_dev(nhc, dev)) { 344 dev_match = true; 345 break; 346 } 347 } 348 } 349 #else 350 if (fib_info_nhc(fi, 0)->nhc_dev == dev) 351 dev_match = true; 352 #endif 353 354 return dev_match; 355 } 356 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev); 357 358 /* Given (packet source, input interface) and optional (dst, oif, tos): 359 * - (main) check, that source is valid i.e. not broadcast or our local 360 * address. 361 * - figure out what "logical" interface this packet arrived 362 * and calculate "specific destination" address. 363 * - check, that packet arrived from expected physical interface. 364 * called with rcu_read_lock() 365 */ 366 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 367 dscp_t dscp, int oif, struct net_device *dev, 368 int rpf, struct in_device *idev, u32 *itag) 369 { 370 struct net *net = dev_net(dev); 371 enum skb_drop_reason reason; 372 struct flow_keys flkeys; 373 int ret, no_addr; 374 struct fib_result res; 375 struct flowi4 fl4; 376 bool dev_match; 377 378 fl4.flowi4_oif = 0; 379 fl4.flowi4_l3mdev = l3mdev_master_ifindex_rcu(dev); 380 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX; 381 fl4.daddr = src; 382 fl4.saddr = dst; 383 fl4.flowi4_dscp = dscp; 384 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 385 fl4.flowi4_tun_key.tun_id = 0; 386 fl4.flowi4_flags = 0; 387 fl4.flowi4_uid = sock_net_uid(net, NULL); 388 fl4.flowi4_multipath_hash = 0; 389 390 no_addr = idev->ifa_list == NULL; 391 392 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0; 393 if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) { 394 fl4.flowi4_proto = 0; 395 fl4.fl4_sport = 0; 396 fl4.fl4_dport = 0; 397 } else { 398 swap(fl4.fl4_sport, fl4.fl4_dport); 399 } 400 401 if (fib_lookup(net, &fl4, &res, 0)) 402 goto last_resort; 403 if (res.type != RTN_UNICAST) { 404 if (res.type != RTN_LOCAL) { 405 reason = SKB_DROP_REASON_IP_INVALID_SOURCE; 406 goto e_inval; 407 } else if (!IN_DEV_ACCEPT_LOCAL(idev)) { 408 reason = SKB_DROP_REASON_IP_LOCAL_SOURCE; 409 goto e_inval; 410 } 411 } 412 fib_combine_itag(itag, &res); 413 414 dev_match = fib_info_nh_uses_dev(res.fi, dev); 415 /* This is not common, loopback packets retain skb_dst so normally they 416 * would not even hit this slow path. 417 */ 418 dev_match = dev_match || (res.type == RTN_LOCAL && 419 dev == net->loopback_dev); 420 if (dev_match) { 421 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST; 422 return ret; 423 } 424 if (no_addr) 425 goto last_resort; 426 if (rpf == 1) 427 goto e_rpf; 428 fl4.flowi4_oif = dev->ifindex; 429 430 ret = 0; 431 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) { 432 if (res.type == RTN_UNICAST) 433 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST; 434 } 435 return ret; 436 437 last_resort: 438 if (rpf) 439 goto e_rpf; 440 *itag = 0; 441 return 0; 442 443 e_inval: 444 return -reason; 445 e_rpf: 446 return -SKB_DROP_REASON_IP_RPFILTER; 447 } 448 449 /* Ignore rp_filter for packets protected by IPsec. */ 450 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 451 dscp_t dscp, int oif, struct net_device *dev, 452 struct in_device *idev, u32 *itag) 453 { 454 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev); 455 struct net *net = dev_net(dev); 456 457 if (!r && !fib_num_tclassid_users(net) && 458 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) { 459 if (IN_DEV_ACCEPT_LOCAL(idev)) 460 goto ok; 461 /* with custom local routes in place, checking local addresses 462 * only will be too optimistic, with custom rules, checking 463 * local addresses only can be too strict, e.g. due to vrf 464 */ 465 if (net->ipv4.fib_has_custom_local_routes || 466 fib4_has_custom_rules(net)) 467 goto full_check; 468 /* Within the same container, it is regarded as a martian source, 469 * and the same host but different containers are not. 470 */ 471 if (inet_lookup_ifaddr_rcu(net, src)) 472 return -SKB_DROP_REASON_IP_LOCAL_SOURCE; 473 474 ok: 475 *itag = 0; 476 return 0; 477 } 478 479 full_check: 480 return __fib_validate_source(skb, src, dst, dscp, oif, dev, r, idev, 481 itag); 482 } 483 484 static inline __be32 sk_extract_addr(struct sockaddr *addr) 485 { 486 return ((struct sockaddr_in *) addr)->sin_addr.s_addr; 487 } 488 489 static int put_rtax(struct nlattr *mx, int len, int type, u32 value) 490 { 491 struct nlattr *nla; 492 493 nla = (struct nlattr *) ((char *) mx + len); 494 nla->nla_type = type; 495 nla->nla_len = nla_attr_size(4); 496 *(u32 *) nla_data(nla) = value; 497 498 return len + nla_total_size(4); 499 } 500 501 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt, 502 struct fib_config *cfg) 503 { 504 __be32 addr; 505 int plen; 506 507 memset(cfg, 0, sizeof(*cfg)); 508 cfg->fc_nlinfo.nl_net = net; 509 510 if (rt->rt_dst.sa_family != AF_INET) 511 return -EAFNOSUPPORT; 512 513 /* 514 * Check mask for validity: 515 * a) it must be contiguous. 516 * b) destination must have all host bits clear. 517 * c) if application forgot to set correct family (AF_INET), 518 * reject request unless it is absolutely clear i.e. 519 * both family and mask are zero. 520 */ 521 plen = 32; 522 addr = sk_extract_addr(&rt->rt_dst); 523 if (!(rt->rt_flags & RTF_HOST)) { 524 __be32 mask = sk_extract_addr(&rt->rt_genmask); 525 526 if (rt->rt_genmask.sa_family != AF_INET) { 527 if (mask || rt->rt_genmask.sa_family) 528 return -EAFNOSUPPORT; 529 } 530 531 if (bad_mask(mask, addr)) 532 return -EINVAL; 533 534 plen = inet_mask_len(mask); 535 } 536 537 cfg->fc_dst_len = plen; 538 cfg->fc_dst = addr; 539 540 if (cmd != SIOCDELRT) { 541 cfg->fc_nlflags = NLM_F_CREATE; 542 cfg->fc_protocol = RTPROT_BOOT; 543 } 544 545 if (rt->rt_metric) 546 cfg->fc_priority = rt->rt_metric - 1; 547 548 if (rt->rt_flags & RTF_REJECT) { 549 cfg->fc_scope = RT_SCOPE_HOST; 550 cfg->fc_type = RTN_UNREACHABLE; 551 return 0; 552 } 553 554 cfg->fc_scope = RT_SCOPE_NOWHERE; 555 cfg->fc_type = RTN_UNICAST; 556 557 if (rt->rt_dev) { 558 char *colon; 559 struct net_device *dev; 560 char devname[IFNAMSIZ]; 561 562 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1)) 563 return -EFAULT; 564 565 devname[IFNAMSIZ-1] = 0; 566 colon = strchr(devname, ':'); 567 if (colon) 568 *colon = 0; 569 dev = __dev_get_by_name(net, devname); 570 if (!dev) 571 return -ENODEV; 572 cfg->fc_oif = dev->ifindex; 573 cfg->fc_table = l3mdev_fib_table(dev); 574 if (colon) { 575 const struct in_ifaddr *ifa; 576 struct in_device *in_dev; 577 578 in_dev = __in_dev_get_rtnl_net(dev); 579 if (!in_dev) 580 return -ENODEV; 581 582 *colon = ':'; 583 584 in_dev_for_each_ifa_rtnl_net(net, ifa, in_dev) { 585 if (strcmp(ifa->ifa_label, devname) == 0) 586 break; 587 } 588 589 if (!ifa) 590 return -ENODEV; 591 cfg->fc_prefsrc = ifa->ifa_local; 592 } 593 } 594 595 addr = sk_extract_addr(&rt->rt_gateway); 596 if (rt->rt_gateway.sa_family == AF_INET && addr) { 597 unsigned int addr_type; 598 599 cfg->fc_gw4 = addr; 600 cfg->fc_gw_family = AF_INET; 601 addr_type = inet_addr_type_table(net, addr, cfg->fc_table); 602 if (rt->rt_flags & RTF_GATEWAY && 603 addr_type == RTN_UNICAST) 604 cfg->fc_scope = RT_SCOPE_UNIVERSE; 605 } 606 607 if (!cfg->fc_table) 608 cfg->fc_table = RT_TABLE_MAIN; 609 610 if (cmd == SIOCDELRT) 611 return 0; 612 613 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family) 614 return -EINVAL; 615 616 if (cfg->fc_scope == RT_SCOPE_NOWHERE) 617 cfg->fc_scope = RT_SCOPE_LINK; 618 619 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) { 620 struct nlattr *mx; 621 int len = 0; 622 623 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL); 624 if (!mx) 625 return -ENOMEM; 626 627 if (rt->rt_flags & RTF_MTU) 628 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40); 629 630 if (rt->rt_flags & RTF_WINDOW) 631 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window); 632 633 if (rt->rt_flags & RTF_IRTT) 634 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3); 635 636 cfg->fc_mx = mx; 637 cfg->fc_mx_len = len; 638 } 639 640 return 0; 641 } 642 643 /* 644 * Handle IP routing ioctl calls. 645 * These are used to manipulate the routing tables 646 */ 647 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt) 648 { 649 struct fib_config cfg; 650 int err; 651 652 switch (cmd) { 653 case SIOCADDRT: /* Add a route */ 654 case SIOCDELRT: /* Delete a route */ 655 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 656 return -EPERM; 657 658 rtnl_net_lock(net); 659 err = rtentry_to_fib_config(net, cmd, rt, &cfg); 660 if (err == 0) { 661 struct fib_table *tb; 662 663 if (cmd == SIOCDELRT) { 664 tb = fib_get_table(net, cfg.fc_table); 665 if (tb) 666 err = fib_table_delete(net, tb, &cfg, 667 NULL); 668 else 669 err = -ESRCH; 670 } else { 671 tb = fib_new_table(net, cfg.fc_table); 672 if (tb) 673 err = fib_table_insert(net, tb, 674 &cfg, NULL); 675 else 676 err = -ENOBUFS; 677 } 678 679 /* allocated by rtentry_to_fib_config() */ 680 kfree(cfg.fc_mx); 681 } 682 rtnl_net_unlock(net); 683 return err; 684 } 685 return -EINVAL; 686 } 687 688 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = { 689 [RTA_UNSPEC] = { .strict_start_type = RTA_DPORT + 1 }, 690 [RTA_DST] = { .type = NLA_U32 }, 691 [RTA_SRC] = { .type = NLA_U32 }, 692 [RTA_IIF] = { .type = NLA_U32 }, 693 [RTA_OIF] = { .type = NLA_U32 }, 694 [RTA_GATEWAY] = { .type = NLA_U32 }, 695 [RTA_PRIORITY] = { .type = NLA_U32 }, 696 [RTA_PREFSRC] = { .type = NLA_U32 }, 697 [RTA_METRICS] = { .type = NLA_NESTED }, 698 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) }, 699 [RTA_FLOW] = { .type = NLA_U32 }, 700 [RTA_ENCAP_TYPE] = { .type = NLA_U16 }, 701 [RTA_ENCAP] = { .type = NLA_NESTED }, 702 [RTA_UID] = { .type = NLA_U32 }, 703 [RTA_MARK] = { .type = NLA_U32 }, 704 [RTA_TABLE] = { .type = NLA_U32 }, 705 [RTA_IP_PROTO] = { .type = NLA_U8 }, 706 [RTA_SPORT] = { .type = NLA_U16 }, 707 [RTA_DPORT] = { .type = NLA_U16 }, 708 [RTA_NH_ID] = { .type = NLA_U32 }, 709 }; 710 711 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla, 712 struct netlink_ext_ack *extack) 713 { 714 struct rtvia *via; 715 int alen; 716 717 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) { 718 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA"); 719 return -EINVAL; 720 } 721 722 via = nla_data(nla); 723 alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr); 724 725 switch (via->rtvia_family) { 726 case AF_INET: 727 if (alen != sizeof(__be32)) { 728 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA"); 729 return -EINVAL; 730 } 731 cfg->fc_gw_family = AF_INET; 732 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr); 733 break; 734 case AF_INET6: 735 #if IS_ENABLED(CONFIG_IPV6) 736 if (alen != sizeof(struct in6_addr)) { 737 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA"); 738 return -EINVAL; 739 } 740 cfg->fc_gw_family = AF_INET6; 741 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr); 742 #else 743 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel"); 744 return -EINVAL; 745 #endif 746 break; 747 default: 748 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA"); 749 return -EINVAL; 750 } 751 752 return 0; 753 } 754 755 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb, 756 struct nlmsghdr *nlh, struct fib_config *cfg, 757 struct netlink_ext_ack *extack) 758 { 759 bool has_gw = false, has_via = false; 760 struct nlattr *attr; 761 int err, remaining; 762 struct rtmsg *rtm; 763 764 err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX, 765 rtm_ipv4_policy, extack); 766 if (err < 0) 767 goto errout; 768 769 memset(cfg, 0, sizeof(*cfg)); 770 771 rtm = nlmsg_data(nlh); 772 773 if (!inet_validate_dscp(rtm->rtm_tos)) { 774 NL_SET_ERR_MSG(extack, 775 "Invalid dsfield (tos): ECN bits must be 0"); 776 err = -EINVAL; 777 goto errout; 778 } 779 cfg->fc_dscp = inet_dsfield_to_dscp(rtm->rtm_tos); 780 781 cfg->fc_dst_len = rtm->rtm_dst_len; 782 cfg->fc_table = rtm->rtm_table; 783 cfg->fc_protocol = rtm->rtm_protocol; 784 cfg->fc_scope = rtm->rtm_scope; 785 cfg->fc_type = rtm->rtm_type; 786 cfg->fc_flags = rtm->rtm_flags; 787 cfg->fc_nlflags = nlh->nlmsg_flags; 788 789 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid; 790 cfg->fc_nlinfo.nlh = nlh; 791 cfg->fc_nlinfo.nl_net = net; 792 793 if (cfg->fc_type > RTN_MAX) { 794 NL_SET_ERR_MSG(extack, "Invalid route type"); 795 err = -EINVAL; 796 goto errout; 797 } 798 799 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) { 800 switch (nla_type(attr)) { 801 case RTA_DST: 802 cfg->fc_dst = nla_get_be32(attr); 803 break; 804 case RTA_OIF: 805 cfg->fc_oif = nla_get_u32(attr); 806 break; 807 case RTA_GATEWAY: 808 has_gw = true; 809 cfg->fc_gw4 = nla_get_be32(attr); 810 if (cfg->fc_gw4) 811 cfg->fc_gw_family = AF_INET; 812 break; 813 case RTA_VIA: 814 has_via = true; 815 err = fib_gw_from_via(cfg, attr, extack); 816 if (err) 817 goto errout; 818 break; 819 case RTA_PRIORITY: 820 cfg->fc_priority = nla_get_u32(attr); 821 break; 822 case RTA_PREFSRC: 823 cfg->fc_prefsrc = nla_get_be32(attr); 824 break; 825 case RTA_METRICS: 826 cfg->fc_mx = nla_data(attr); 827 cfg->fc_mx_len = nla_len(attr); 828 break; 829 case RTA_MULTIPATH: 830 err = lwtunnel_valid_encap_type_attr(nla_data(attr), 831 nla_len(attr), 832 extack); 833 if (err < 0) 834 goto errout; 835 cfg->fc_mp = nla_data(attr); 836 cfg->fc_mp_len = nla_len(attr); 837 break; 838 case RTA_FLOW: 839 cfg->fc_flow = nla_get_u32(attr); 840 break; 841 case RTA_TABLE: 842 cfg->fc_table = nla_get_u32(attr); 843 break; 844 case RTA_ENCAP: 845 cfg->fc_encap = attr; 846 break; 847 case RTA_ENCAP_TYPE: 848 cfg->fc_encap_type = nla_get_u16(attr); 849 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, 850 extack); 851 if (err < 0) 852 goto errout; 853 break; 854 case RTA_NH_ID: 855 cfg->fc_nh_id = nla_get_u32(attr); 856 break; 857 } 858 } 859 860 if (cfg->fc_dst_len > 32) { 861 NL_SET_ERR_MSG(extack, "Invalid prefix length"); 862 err = -EINVAL; 863 goto errout; 864 } 865 866 if (cfg->fc_dst_len < 32 && (ntohl(cfg->fc_dst) << cfg->fc_dst_len)) { 867 NL_SET_ERR_MSG(extack, "Invalid prefix for given prefix length"); 868 err = -EINVAL; 869 goto errout; 870 } 871 872 if (cfg->fc_nh_id) { 873 if (cfg->fc_oif || cfg->fc_gw_family || 874 cfg->fc_encap || cfg->fc_mp) { 875 NL_SET_ERR_MSG(extack, 876 "Nexthop specification and nexthop id are mutually exclusive"); 877 err = -EINVAL; 878 goto errout; 879 } 880 } 881 882 if (has_gw && has_via) { 883 NL_SET_ERR_MSG(extack, 884 "Nexthop configuration can not contain both GATEWAY and VIA"); 885 err = -EINVAL; 886 goto errout; 887 } 888 889 if (!cfg->fc_table) 890 cfg->fc_table = RT_TABLE_MAIN; 891 892 return 0; 893 errout: 894 return err; 895 } 896 897 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, 898 struct netlink_ext_ack *extack) 899 { 900 struct net *net = sock_net(skb->sk); 901 struct fib_config cfg; 902 struct fib_table *tb; 903 int err; 904 905 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack); 906 if (err < 0) 907 goto errout; 908 909 rtnl_net_lock(net); 910 911 if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) { 912 NL_SET_ERR_MSG(extack, "Nexthop id does not exist"); 913 err = -EINVAL; 914 goto unlock; 915 } 916 917 tb = fib_get_table(net, cfg.fc_table); 918 if (!tb) { 919 NL_SET_ERR_MSG(extack, "FIB table does not exist"); 920 err = -ESRCH; 921 goto unlock; 922 } 923 924 err = fib_table_delete(net, tb, &cfg, extack); 925 unlock: 926 rtnl_net_unlock(net); 927 errout: 928 return err; 929 } 930 931 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, 932 struct netlink_ext_ack *extack) 933 { 934 struct net *net = sock_net(skb->sk); 935 struct fib_config cfg; 936 struct fib_table *tb; 937 int err; 938 939 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack); 940 if (err < 0) 941 goto errout; 942 943 rtnl_net_lock(net); 944 945 tb = fib_new_table(net, cfg.fc_table); 946 if (!tb) { 947 err = -ENOBUFS; 948 goto unlock; 949 } 950 951 err = fib_table_insert(net, tb, &cfg, extack); 952 if (!err && cfg.fc_type == RTN_LOCAL) 953 net->ipv4.fib_has_custom_local_routes = true; 954 955 unlock: 956 rtnl_net_unlock(net); 957 errout: 958 return err; 959 } 960 961 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh, 962 struct fib_dump_filter *filter, 963 struct netlink_callback *cb) 964 { 965 struct netlink_ext_ack *extack = cb->extack; 966 struct nlattr *tb[RTA_MAX + 1]; 967 struct rtmsg *rtm; 968 int err, i; 969 970 rtm = nlmsg_payload(nlh, sizeof(*rtm)); 971 if (!rtm) { 972 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request"); 973 return -EINVAL; 974 } 975 976 if (rtm->rtm_dst_len || rtm->rtm_src_len || rtm->rtm_tos || 977 rtm->rtm_scope) { 978 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request"); 979 return -EINVAL; 980 } 981 982 if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) { 983 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request"); 984 return -EINVAL; 985 } 986 if (rtm->rtm_flags & RTM_F_CLONED) 987 filter->dump_routes = false; 988 else 989 filter->dump_exceptions = false; 990 991 filter->flags = rtm->rtm_flags; 992 filter->protocol = rtm->rtm_protocol; 993 filter->rt_type = rtm->rtm_type; 994 filter->table_id = rtm->rtm_table; 995 996 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX, 997 rtm_ipv4_policy, extack); 998 if (err < 0) 999 return err; 1000 1001 for (i = 0; i <= RTA_MAX; ++i) { 1002 int ifindex; 1003 1004 if (!tb[i]) 1005 continue; 1006 1007 switch (i) { 1008 case RTA_TABLE: 1009 filter->table_id = nla_get_u32(tb[i]); 1010 break; 1011 case RTA_OIF: 1012 ifindex = nla_get_u32(tb[i]); 1013 1014 filter->dev = dev_get_by_index_rcu(net, ifindex); 1015 if (!filter->dev) 1016 return -ENODEV; 1017 break; 1018 default: 1019 NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request"); 1020 return -EINVAL; 1021 } 1022 } 1023 1024 if (filter->flags || filter->protocol || filter->rt_type || 1025 filter->table_id || filter->dev) { 1026 filter->filter_set = 1; 1027 cb->answer_flags = NLM_F_DUMP_FILTERED; 1028 } 1029 1030 return 0; 1031 } 1032 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req); 1033 1034 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 1035 { 1036 const struct nlmsghdr *nlh = cb->nlh; 1037 struct net *net = sock_net(skb->sk); 1038 struct fib_dump_filter filter = { 1039 .dump_routes = true, 1040 .dump_exceptions = true, 1041 }; 1042 struct hlist_head *head; 1043 int dumped = 0, err = 0; 1044 struct fib_table *tb; 1045 unsigned int h, s_h; 1046 u32 s_id; 1047 1048 rcu_read_lock(); 1049 if (cb->strict_check) { 1050 err = ip_valid_fib_dump_req(net, nlh, &filter, cb); 1051 if (err < 0) 1052 goto unlock; 1053 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) { 1054 struct rtmsg *rtm = nlmsg_data(nlh); 1055 1056 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED); 1057 } 1058 1059 /* ipv4 does not use prefix flag */ 1060 if (filter.flags & RTM_F_PREFIX) 1061 goto unlock; 1062 1063 if (filter.table_id) { 1064 tb = fib_get_table(net, filter.table_id); 1065 if (!tb) { 1066 if (rtnl_msg_family(cb->nlh) != PF_INET) 1067 goto unlock; 1068 1069 NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist"); 1070 err = -ENOENT; 1071 goto unlock; 1072 } 1073 err = fib_table_dump(tb, skb, cb, &filter); 1074 goto unlock; 1075 } 1076 1077 s_h = cb->args[0]; 1078 s_id = cb->args[1]; 1079 1080 err = 0; 1081 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_id = 0) { 1082 head = &net->ipv4.fib_table_hash[h]; 1083 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 1084 if (s_id && tb->tb_id != s_id) 1085 continue; 1086 1087 s_id = 0; 1088 if (dumped) 1089 memset(&cb->args[2], 0, sizeof(cb->args) - 1090 2 * sizeof(cb->args[0])); 1091 cb->args[1] = tb->tb_id; 1092 err = fib_table_dump(tb, skb, cb, &filter); 1093 if (err < 0) 1094 goto out; 1095 dumped = 1; 1096 } 1097 } 1098 out: 1099 1100 cb->args[0] = h; 1101 1102 unlock: 1103 rcu_read_unlock(); 1104 return err; 1105 } 1106 1107 /* Prepare and feed intra-kernel routing request. 1108 * Really, it should be netlink message, but :-( netlink 1109 * can be not configured, so that we feed it directly 1110 * to fib engine. It is legal, because all events occur 1111 * only when netlink is already locked. 1112 */ 1113 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, 1114 struct in_ifaddr *ifa, u32 rt_priority) 1115 { 1116 struct net *net = dev_net(ifa->ifa_dev->dev); 1117 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev); 1118 struct fib_table *tb; 1119 struct fib_config cfg = { 1120 .fc_protocol = RTPROT_KERNEL, 1121 .fc_type = type, 1122 .fc_dst = dst, 1123 .fc_dst_len = dst_len, 1124 .fc_priority = rt_priority, 1125 .fc_prefsrc = ifa->ifa_local, 1126 .fc_oif = ifa->ifa_dev->dev->ifindex, 1127 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND, 1128 .fc_nlinfo = { 1129 .nl_net = net, 1130 }, 1131 }; 1132 1133 if (!tb_id) 1134 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL; 1135 1136 tb = fib_new_table(net, tb_id); 1137 if (!tb) 1138 return; 1139 1140 cfg.fc_table = tb->tb_id; 1141 1142 if (type != RTN_LOCAL) 1143 cfg.fc_scope = RT_SCOPE_LINK; 1144 else 1145 cfg.fc_scope = RT_SCOPE_HOST; 1146 1147 if (cmd == RTM_NEWROUTE) 1148 fib_table_insert(net, tb, &cfg, NULL); 1149 else 1150 fib_table_delete(net, tb, &cfg, NULL); 1151 } 1152 1153 void fib_add_ifaddr(struct in_ifaddr *ifa) 1154 { 1155 struct in_device *in_dev = ifa->ifa_dev; 1156 struct net_device *dev = in_dev->dev; 1157 struct in_ifaddr *prim = ifa; 1158 __be32 mask = ifa->ifa_mask; 1159 __be32 addr = ifa->ifa_local; 1160 __be32 prefix = ifa->ifa_address & mask; 1161 1162 if (ifa->ifa_flags & IFA_F_SECONDARY) { 1163 prim = inet_ifa_byprefix(in_dev, prefix, mask); 1164 if (!prim) { 1165 pr_warn("%s: bug: prim == NULL\n", __func__); 1166 return; 1167 } 1168 } 1169 1170 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0); 1171 1172 if (!(dev->flags & IFF_UP)) 1173 return; 1174 1175 /* Add broadcast address, if it is explicitly assigned. */ 1176 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) { 1177 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, 1178 prim, 0); 1179 arp_invalidate(dev, ifa->ifa_broadcast, false); 1180 } 1181 1182 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) && 1183 (prefix != addr || ifa->ifa_prefixlen < 32)) { 1184 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 1185 fib_magic(RTM_NEWROUTE, 1186 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 1187 prefix, ifa->ifa_prefixlen, prim, 1188 ifa->ifa_rt_priority); 1189 1190 /* Add the network broadcast address, when it makes sense */ 1191 if (ifa->ifa_prefixlen < 31) { 1192 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask, 1193 32, prim, 0); 1194 arp_invalidate(dev, prefix | ~mask, false); 1195 } 1196 } 1197 } 1198 1199 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric) 1200 { 1201 __be32 prefix = ifa->ifa_address & ifa->ifa_mask; 1202 struct in_device *in_dev = ifa->ifa_dev; 1203 struct net_device *dev = in_dev->dev; 1204 1205 if (!(dev->flags & IFF_UP) || 1206 ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) || 1207 ipv4_is_zeronet(prefix) || 1208 (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32)) 1209 return; 1210 1211 /* add the new */ 1212 fib_magic(RTM_NEWROUTE, 1213 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 1214 prefix, ifa->ifa_prefixlen, ifa, new_metric); 1215 1216 /* delete the old */ 1217 fib_magic(RTM_DELROUTE, 1218 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 1219 prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority); 1220 } 1221 1222 /* Delete primary or secondary address. 1223 * Optionally, on secondary address promotion consider the addresses 1224 * from subnet iprim as deleted, even if they are in device list. 1225 * In this case the secondary ifa can be in device list. 1226 */ 1227 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim) 1228 { 1229 struct in_device *in_dev = ifa->ifa_dev; 1230 struct net_device *dev = in_dev->dev; 1231 struct in_ifaddr *ifa1; 1232 struct in_ifaddr *prim = ifa, *prim1 = NULL; 1233 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask; 1234 __be32 any = ifa->ifa_address & ifa->ifa_mask; 1235 #define LOCAL_OK 1 1236 #define BRD_OK 2 1237 #define BRD0_OK 4 1238 #define BRD1_OK 8 1239 unsigned int ok = 0; 1240 int subnet = 0; /* Primary network */ 1241 int gone = 1; /* Address is missing */ 1242 int same_prefsrc = 0; /* Another primary with same IP */ 1243 1244 if (ifa->ifa_flags & IFA_F_SECONDARY) { 1245 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask); 1246 if (!prim) { 1247 /* if the device has been deleted, we don't perform 1248 * address promotion 1249 */ 1250 if (!in_dev->dead) 1251 pr_warn("%s: bug: prim == NULL\n", __func__); 1252 return; 1253 } 1254 if (iprim && iprim != prim) { 1255 pr_warn("%s: bug: iprim != prim\n", __func__); 1256 return; 1257 } 1258 } else if (!ipv4_is_zeronet(any) && 1259 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) { 1260 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 1261 fib_magic(RTM_DELROUTE, 1262 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 1263 any, ifa->ifa_prefixlen, prim, 0); 1264 subnet = 1; 1265 } 1266 1267 if (in_dev->dead) 1268 goto no_promotions; 1269 1270 /* Deletion is more complicated than add. 1271 * We should take care of not to delete too much :-) 1272 * 1273 * Scan address list to be sure that addresses are really gone. 1274 */ 1275 rcu_read_lock(); 1276 in_dev_for_each_ifa_rcu(ifa1, in_dev) { 1277 if (ifa1 == ifa) { 1278 /* promotion, keep the IP */ 1279 gone = 0; 1280 continue; 1281 } 1282 /* Ignore IFAs from our subnet */ 1283 if (iprim && ifa1->ifa_mask == iprim->ifa_mask && 1284 inet_ifa_match(ifa1->ifa_address, iprim)) 1285 continue; 1286 1287 /* Ignore ifa1 if it uses different primary IP (prefsrc) */ 1288 if (ifa1->ifa_flags & IFA_F_SECONDARY) { 1289 /* Another address from our subnet? */ 1290 if (ifa1->ifa_mask == prim->ifa_mask && 1291 inet_ifa_match(ifa1->ifa_address, prim)) 1292 prim1 = prim; 1293 else { 1294 /* We reached the secondaries, so 1295 * same_prefsrc should be determined. 1296 */ 1297 if (!same_prefsrc) 1298 continue; 1299 /* Search new prim1 if ifa1 is not 1300 * using the current prim1 1301 */ 1302 if (!prim1 || 1303 ifa1->ifa_mask != prim1->ifa_mask || 1304 !inet_ifa_match(ifa1->ifa_address, prim1)) 1305 prim1 = inet_ifa_byprefix(in_dev, 1306 ifa1->ifa_address, 1307 ifa1->ifa_mask); 1308 if (!prim1) 1309 continue; 1310 if (prim1->ifa_local != prim->ifa_local) 1311 continue; 1312 } 1313 } else { 1314 if (prim->ifa_local != ifa1->ifa_local) 1315 continue; 1316 prim1 = ifa1; 1317 if (prim != prim1) 1318 same_prefsrc = 1; 1319 } 1320 if (ifa->ifa_local == ifa1->ifa_local) 1321 ok |= LOCAL_OK; 1322 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 1323 ok |= BRD_OK; 1324 if (brd == ifa1->ifa_broadcast) 1325 ok |= BRD1_OK; 1326 if (any == ifa1->ifa_broadcast) 1327 ok |= BRD0_OK; 1328 /* primary has network specific broadcasts */ 1329 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) { 1330 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask; 1331 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask; 1332 1333 if (!ipv4_is_zeronet(any1)) { 1334 if (ifa->ifa_broadcast == brd1 || 1335 ifa->ifa_broadcast == any1) 1336 ok |= BRD_OK; 1337 if (brd == brd1 || brd == any1) 1338 ok |= BRD1_OK; 1339 if (any == brd1 || any == any1) 1340 ok |= BRD0_OK; 1341 } 1342 } 1343 } 1344 rcu_read_unlock(); 1345 1346 no_promotions: 1347 if (!(ok & BRD_OK)) 1348 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, 1349 prim, 0); 1350 if (subnet && ifa->ifa_prefixlen < 31) { 1351 if (!(ok & BRD1_OK)) 1352 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, 1353 prim, 0); 1354 if (!(ok & BRD0_OK)) 1355 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, 1356 prim, 0); 1357 } 1358 if (!(ok & LOCAL_OK)) { 1359 unsigned int addr_type; 1360 1361 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0); 1362 1363 /* Check, that this local address finally disappeared. */ 1364 addr_type = inet_addr_type_dev_table(dev_net(dev), dev, 1365 ifa->ifa_local); 1366 if (gone && addr_type != RTN_LOCAL) { 1367 /* And the last, but not the least thing. 1368 * We must flush stray FIB entries. 1369 * 1370 * First of all, we scan fib_info list searching 1371 * for stray nexthop entries, then ignite fib_flush. 1372 */ 1373 if (fib_sync_down_addr(dev, ifa->ifa_local)) 1374 fib_flush(dev_net(dev)); 1375 } 1376 } 1377 #undef LOCAL_OK 1378 #undef BRD_OK 1379 #undef BRD0_OK 1380 #undef BRD1_OK 1381 } 1382 1383 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn) 1384 { 1385 1386 struct fib_result res; 1387 struct flowi4 fl4 = { 1388 .flowi4_mark = frn->fl_mark, 1389 .daddr = frn->fl_addr, 1390 .flowi4_dscp = inet_dsfield_to_dscp(frn->fl_tos), 1391 .flowi4_scope = frn->fl_scope, 1392 }; 1393 struct fib_table *tb; 1394 1395 rcu_read_lock(); 1396 1397 tb = fib_get_table(net, frn->tb_id_in); 1398 1399 frn->err = -ENOENT; 1400 if (tb) { 1401 local_bh_disable(); 1402 1403 frn->tb_id = tb->tb_id; 1404 frn->err = fib_table_lookup(net, tb, &fl4, &res, 1405 FIB_LOOKUP_NOREF); 1406 1407 if (!frn->err) { 1408 frn->prefixlen = res.prefixlen; 1409 frn->nh_sel = res.nh_sel; 1410 frn->type = res.type; 1411 frn->scope = res.scope; 1412 } 1413 local_bh_enable(); 1414 } 1415 1416 rcu_read_unlock(); 1417 } 1418 1419 static void nl_fib_input(struct sk_buff *skb) 1420 { 1421 struct net *net; 1422 struct fib_result_nl *frn; 1423 struct nlmsghdr *nlh; 1424 u32 portid; 1425 1426 net = sock_net(skb->sk); 1427 nlh = nlmsg_hdr(skb); 1428 if (skb->len < nlmsg_total_size(sizeof(*frn)) || 1429 skb->len < nlh->nlmsg_len || 1430 nlmsg_len(nlh) < sizeof(*frn)) 1431 return; 1432 1433 skb = netlink_skb_clone(skb, GFP_KERNEL); 1434 if (!skb) 1435 return; 1436 nlh = nlmsg_hdr(skb); 1437 1438 frn = nlmsg_data(nlh); 1439 nl_fib_lookup(net, frn); 1440 1441 portid = NETLINK_CB(skb).portid; /* netlink portid */ 1442 NETLINK_CB(skb).portid = 0; /* from kernel */ 1443 NETLINK_CB(skb).dst_group = 0; /* unicast */ 1444 nlmsg_unicast(net->ipv4.fibnl, skb, portid); 1445 } 1446 1447 static int __net_init nl_fib_lookup_init(struct net *net) 1448 { 1449 struct sock *sk; 1450 struct netlink_kernel_cfg cfg = { 1451 .input = nl_fib_input, 1452 }; 1453 1454 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg); 1455 if (!sk) 1456 return -EAFNOSUPPORT; 1457 net->ipv4.fibnl = sk; 1458 return 0; 1459 } 1460 1461 static void nl_fib_lookup_exit(struct net *net) 1462 { 1463 netlink_kernel_release(net->ipv4.fibnl); 1464 net->ipv4.fibnl = NULL; 1465 } 1466 1467 static void fib_disable_ip(struct net_device *dev, unsigned long event, 1468 bool force) 1469 { 1470 if (fib_sync_down_dev(dev, event, force)) 1471 fib_flush(dev_net(dev)); 1472 else 1473 rt_cache_flush(dev_net(dev)); 1474 arp_ifdown(dev); 1475 } 1476 1477 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 1478 { 1479 struct in_ifaddr *ifa = ptr; 1480 struct net_device *dev = ifa->ifa_dev->dev; 1481 struct net *net = dev_net(dev); 1482 1483 switch (event) { 1484 case NETDEV_UP: 1485 fib_add_ifaddr(ifa); 1486 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1487 fib_sync_up(dev, RTNH_F_DEAD); 1488 #endif 1489 atomic_inc(&net->ipv4.dev_addr_genid); 1490 rt_cache_flush(net); 1491 break; 1492 case NETDEV_DOWN: 1493 fib_del_ifaddr(ifa, NULL); 1494 atomic_inc(&net->ipv4.dev_addr_genid); 1495 if (!ifa->ifa_dev->ifa_list) { 1496 /* Last address was deleted from this interface. 1497 * Disable IP. 1498 */ 1499 fib_disable_ip(dev, event, true); 1500 } else { 1501 rt_cache_flush(net); 1502 } 1503 break; 1504 } 1505 return NOTIFY_DONE; 1506 } 1507 1508 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 1509 { 1510 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1511 struct netdev_notifier_changeupper_info *upper_info = ptr; 1512 struct netdev_notifier_info_ext *info_ext = ptr; 1513 struct in_device *in_dev; 1514 struct net *net = dev_net(dev); 1515 struct in_ifaddr *ifa; 1516 unsigned int flags; 1517 1518 if (event == NETDEV_UNREGISTER) { 1519 fib_disable_ip(dev, event, true); 1520 rt_flush_dev(dev); 1521 return NOTIFY_DONE; 1522 } 1523 1524 in_dev = __in_dev_get_rtnl(dev); 1525 if (!in_dev) 1526 return NOTIFY_DONE; 1527 1528 switch (event) { 1529 case NETDEV_UP: 1530 in_dev_for_each_ifa_rtnl(ifa, in_dev) { 1531 fib_add_ifaddr(ifa); 1532 } 1533 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1534 fib_sync_up(dev, RTNH_F_DEAD); 1535 #endif 1536 atomic_inc(&net->ipv4.dev_addr_genid); 1537 rt_cache_flush(net); 1538 break; 1539 case NETDEV_DOWN: 1540 fib_disable_ip(dev, event, false); 1541 break; 1542 case NETDEV_CHANGE: 1543 flags = netif_get_flags(dev); 1544 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 1545 fib_sync_up(dev, RTNH_F_LINKDOWN); 1546 else 1547 fib_sync_down_dev(dev, event, false); 1548 rt_cache_flush(net); 1549 break; 1550 case NETDEV_CHANGEMTU: 1551 fib_sync_mtu(dev, info_ext->ext.mtu); 1552 rt_cache_flush(net); 1553 break; 1554 case NETDEV_CHANGEUPPER: 1555 upper_info = ptr; 1556 /* flush all routes if dev is linked to or unlinked from 1557 * an L3 master device (e.g., VRF) 1558 */ 1559 if (upper_info->upper_dev && 1560 netif_is_l3_master(upper_info->upper_dev)) 1561 fib_disable_ip(dev, NETDEV_DOWN, true); 1562 break; 1563 } 1564 return NOTIFY_DONE; 1565 } 1566 1567 static struct notifier_block fib_inetaddr_notifier = { 1568 .notifier_call = fib_inetaddr_event, 1569 }; 1570 1571 static struct notifier_block fib_netdev_notifier = { 1572 .notifier_call = fib_netdev_event, 1573 }; 1574 1575 static int __net_init ip_fib_net_init(struct net *net) 1576 { 1577 int err; 1578 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ; 1579 1580 err = fib4_notifier_init(net); 1581 if (err) 1582 return err; 1583 1584 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1585 /* Default to 3-tuple */ 1586 net->ipv4.sysctl_fib_multipath_hash_fields = 1587 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK; 1588 #endif 1589 1590 #ifdef CONFIG_IP_MULTIPLE_TABLES 1591 spin_lock_init(&net->ipv4.fib_table_hash_lock); 1592 #endif 1593 1594 /* Avoid false sharing : Use at least a full cache line */ 1595 size = max_t(size_t, size, L1_CACHE_BYTES); 1596 1597 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL); 1598 if (!net->ipv4.fib_table_hash) { 1599 err = -ENOMEM; 1600 goto err_table_hash_alloc; 1601 } 1602 1603 err = fib4_rules_init(net); 1604 if (err < 0) 1605 goto err_rules_init; 1606 return 0; 1607 1608 err_rules_init: 1609 kfree(net->ipv4.fib_table_hash); 1610 err_table_hash_alloc: 1611 fib4_notifier_exit(net); 1612 return err; 1613 } 1614 1615 static void ip_fib_net_exit(struct net *net) 1616 { 1617 int i; 1618 1619 ASSERT_RTNL_NET(net); 1620 #ifdef CONFIG_IP_MULTIPLE_TABLES 1621 RCU_INIT_POINTER(net->ipv4.fib_main, NULL); 1622 RCU_INIT_POINTER(net->ipv4.fib_default, NULL); 1623 #endif 1624 /* Destroy the tables in reverse order to guarantee that the 1625 * local table, ID 255, is destroyed before the main table, ID 1626 * 254. This is necessary as the local table may contain 1627 * references to data contained in the main table. 1628 */ 1629 for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) { 1630 struct hlist_head *head = &net->ipv4.fib_table_hash[i]; 1631 struct hlist_node *tmp; 1632 struct fib_table *tb; 1633 1634 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) { 1635 hlist_del_rcu(&tb->tb_hlist); 1636 fib_table_flush(net, tb, true); 1637 fib_free_table(tb); 1638 } 1639 } 1640 } 1641 1642 static int __net_init fib_net_init(struct net *net) 1643 { 1644 int error; 1645 1646 #ifdef CONFIG_IP_ROUTE_CLASSID 1647 atomic_set(&net->ipv4.fib_num_tclassid_users, 0); 1648 #endif 1649 error = ip_fib_net_init(net); 1650 if (error < 0) 1651 goto out; 1652 1653 error = fib4_semantics_init(net); 1654 if (error) 1655 goto out_semantics; 1656 1657 error = nl_fib_lookup_init(net); 1658 if (error < 0) 1659 goto out_nlfl; 1660 1661 error = fib_proc_init(net); 1662 if (error < 0) 1663 goto out_proc; 1664 out: 1665 return error; 1666 1667 out_proc: 1668 nl_fib_lookup_exit(net); 1669 out_nlfl: 1670 fib4_semantics_exit(net); 1671 out_semantics: 1672 rtnl_net_lock(net); 1673 ip_fib_net_exit(net); 1674 rtnl_net_unlock(net); 1675 1676 #ifdef CONFIG_IP_MULTIPLE_TABLES 1677 fib4_rules_exit(net); 1678 #endif 1679 kfree(net->ipv4.fib_table_hash); 1680 fib4_notifier_exit(net); 1681 goto out; 1682 } 1683 1684 static void __net_exit fib_net_pre_exit(struct net *net) 1685 { 1686 fib_proc_exit(net); 1687 nl_fib_lookup_exit(net); 1688 } 1689 1690 static void __net_exit fib_net_exit_rtnl(struct net *net, 1691 struct list_head *dev_kill_list) 1692 { 1693 ip_fib_net_exit(net); 1694 } 1695 1696 static void __net_exit fib_net_exit(struct net *net) 1697 { 1698 #ifdef CONFIG_IP_MULTIPLE_TABLES 1699 fib4_rules_exit(net); 1700 #endif 1701 kfree(net->ipv4.fib_table_hash); 1702 fib4_notifier_exit(net); 1703 fib4_semantics_exit(net); 1704 } 1705 1706 static struct pernet_operations fib_net_ops = { 1707 .init = fib_net_init, 1708 .pre_exit = fib_net_pre_exit, 1709 .exit_rtnl = fib_net_exit_rtnl, 1710 .exit = fib_net_exit, 1711 }; 1712 1713 static const struct rtnl_msg_handler fib_rtnl_msg_handlers[] __initconst = { 1714 {.protocol = PF_INET, .msgtype = RTM_NEWROUTE, 1715 .doit = inet_rtm_newroute, .flags = RTNL_FLAG_DOIT_PERNET}, 1716 {.protocol = PF_INET, .msgtype = RTM_DELROUTE, 1717 .doit = inet_rtm_delroute, .flags = RTNL_FLAG_DOIT_PERNET}, 1718 {.protocol = PF_INET, .msgtype = RTM_GETROUTE, .dumpit = inet_dump_fib, 1719 .flags = RTNL_FLAG_DUMP_UNLOCKED | RTNL_FLAG_DUMP_SPLIT_NLM_DONE}, 1720 }; 1721 1722 void __init ip_fib_init(void) 1723 { 1724 fib_trie_init(); 1725 1726 register_pernet_subsys(&fib_net_ops); 1727 1728 register_netdevice_notifier(&fib_netdev_notifier); 1729 register_inetaddr_notifier(&fib_inetaddr_notifier); 1730 1731 rtnl_register_many(fib_rtnl_msg_handlers); 1732 } 1733