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/inet_dscp.h> 36 #include <net/ip.h> 37 #include <net/protocol.h> 38 #include <net/route.h> 39 #include <net/tcp.h> 40 #include <net/sock.h> 41 #include <net/arp.h> 42 #include <net/ip_fib.h> 43 #include <net/nexthop.h> 44 #include <net/rtnetlink.h> 45 #include <net/xfrm.h> 46 #include <net/l3mdev.h> 47 #include <net/lwtunnel.h> 48 #include <trace/events/fib.h> 49 50 #ifndef CONFIG_IP_MULTIPLE_TABLES 51 52 static int __net_init fib4_rules_init(struct net *net) 53 { 54 struct fib_table *local_table, *main_table; 55 56 main_table = fib_trie_table(RT_TABLE_MAIN, NULL); 57 if (!main_table) 58 return -ENOMEM; 59 60 local_table = fib_trie_table(RT_TABLE_LOCAL, main_table); 61 if (!local_table) 62 goto fail; 63 64 hlist_add_head_rcu(&local_table->tb_hlist, 65 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]); 66 hlist_add_head_rcu(&main_table->tb_hlist, 67 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]); 68 return 0; 69 70 fail: 71 fib_free_table(main_table); 72 return -ENOMEM; 73 } 74 #else 75 76 struct fib_table *fib_new_table(struct net *net, u32 id) 77 { 78 struct fib_table *tb, *alias = NULL; 79 unsigned int h; 80 81 if (id == 0) 82 id = RT_TABLE_MAIN; 83 tb = fib_get_table(net, id); 84 if (tb) 85 return tb; 86 87 if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules) 88 alias = fib_new_table(net, RT_TABLE_MAIN); 89 90 tb = fib_trie_table(id, alias); 91 if (!tb) 92 return NULL; 93 94 switch (id) { 95 case RT_TABLE_MAIN: 96 rcu_assign_pointer(net->ipv4.fib_main, tb); 97 break; 98 case RT_TABLE_DEFAULT: 99 rcu_assign_pointer(net->ipv4.fib_default, tb); 100 break; 101 default: 102 break; 103 } 104 105 h = id & (FIB_TABLE_HASHSZ - 1); 106 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]); 107 return tb; 108 } 109 EXPORT_SYMBOL_GPL(fib_new_table); 110 111 /* caller must hold either rtnl or rcu read lock */ 112 struct fib_table *fib_get_table(struct net *net, u32 id) 113 { 114 struct fib_table *tb; 115 struct hlist_head *head; 116 unsigned int h; 117 118 if (id == 0) 119 id = RT_TABLE_MAIN; 120 h = id & (FIB_TABLE_HASHSZ - 1); 121 122 head = &net->ipv4.fib_table_hash[h]; 123 hlist_for_each_entry_rcu(tb, head, tb_hlist, 124 lockdep_rtnl_is_held()) { 125 if (tb->tb_id == id) 126 return tb; 127 } 128 return NULL; 129 } 130 #endif /* CONFIG_IP_MULTIPLE_TABLES */ 131 132 static void fib_replace_table(struct net *net, struct fib_table *old, 133 struct fib_table *new) 134 { 135 #ifdef CONFIG_IP_MULTIPLE_TABLES 136 switch (new->tb_id) { 137 case RT_TABLE_MAIN: 138 rcu_assign_pointer(net->ipv4.fib_main, new); 139 break; 140 case RT_TABLE_DEFAULT: 141 rcu_assign_pointer(net->ipv4.fib_default, new); 142 break; 143 default: 144 break; 145 } 146 147 #endif 148 /* replace the old table in the hlist */ 149 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist); 150 } 151 152 int fib_unmerge(struct net *net) 153 { 154 struct fib_table *old, *new, *main_table; 155 156 /* attempt to fetch local table if it has been allocated */ 157 old = fib_get_table(net, RT_TABLE_LOCAL); 158 if (!old) 159 return 0; 160 161 new = fib_trie_unmerge(old); 162 if (!new) 163 return -ENOMEM; 164 165 /* table is already unmerged */ 166 if (new == old) 167 return 0; 168 169 /* replace merged table with clean table */ 170 fib_replace_table(net, old, new); 171 fib_free_table(old); 172 173 /* attempt to fetch main table if it has been allocated */ 174 main_table = fib_get_table(net, RT_TABLE_MAIN); 175 if (!main_table) 176 return 0; 177 178 /* flush local entries from main table */ 179 fib_table_flush_external(main_table); 180 181 return 0; 182 } 183 184 void fib_flush(struct net *net) 185 { 186 int flushed = 0; 187 unsigned int h; 188 189 for (h = 0; h < FIB_TABLE_HASHSZ; h++) { 190 struct hlist_head *head = &net->ipv4.fib_table_hash[h]; 191 struct hlist_node *tmp; 192 struct fib_table *tb; 193 194 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) 195 flushed += fib_table_flush(net, tb, false); 196 } 197 198 if (flushed) 199 rt_cache_flush(net); 200 } 201 202 /* 203 * Find address type as if only "dev" was present in the system. If 204 * on_dev is NULL then all interfaces are taken into consideration. 205 */ 206 static inline unsigned int __inet_dev_addr_type(struct net *net, 207 const struct net_device *dev, 208 __be32 addr, u32 tb_id) 209 { 210 struct flowi4 fl4 = { .daddr = addr }; 211 struct fib_result res; 212 unsigned int ret = RTN_BROADCAST; 213 struct fib_table *table; 214 215 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr)) 216 return RTN_BROADCAST; 217 if (ipv4_is_multicast(addr)) 218 return RTN_MULTICAST; 219 220 rcu_read_lock(); 221 222 table = fib_get_table(net, tb_id); 223 if (table) { 224 ret = RTN_UNICAST; 225 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) { 226 struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0); 227 228 if (!dev || dev == nhc->nhc_dev) 229 ret = res.type; 230 } 231 } 232 233 rcu_read_unlock(); 234 return ret; 235 } 236 237 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id) 238 { 239 return __inet_dev_addr_type(net, NULL, addr, tb_id); 240 } 241 EXPORT_SYMBOL(inet_addr_type_table); 242 243 unsigned int inet_addr_type(struct net *net, __be32 addr) 244 { 245 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL); 246 } 247 EXPORT_SYMBOL(inet_addr_type); 248 249 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev, 250 __be32 addr) 251 { 252 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL; 253 254 return __inet_dev_addr_type(net, dev, addr, rt_table); 255 } 256 EXPORT_SYMBOL(inet_dev_addr_type); 257 258 /* inet_addr_type with dev == NULL but using the table from a dev 259 * if one is associated 260 */ 261 unsigned int inet_addr_type_dev_table(struct net *net, 262 const struct net_device *dev, 263 __be32 addr) 264 { 265 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL; 266 267 return __inet_dev_addr_type(net, NULL, addr, rt_table); 268 } 269 EXPORT_SYMBOL(inet_addr_type_dev_table); 270 271 __be32 fib_compute_spec_dst(struct sk_buff *skb) 272 { 273 struct net_device *dev = skb->dev; 274 struct in_device *in_dev; 275 struct fib_result res; 276 struct rtable *rt; 277 struct net *net; 278 int scope; 279 280 rt = skb_rtable(skb); 281 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) == 282 RTCF_LOCAL) 283 return ip_hdr(skb)->daddr; 284 285 in_dev = __in_dev_get_rcu(dev); 286 287 net = dev_net(dev); 288 289 scope = RT_SCOPE_UNIVERSE; 290 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) { 291 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev); 292 struct flowi4 fl4 = { 293 .flowi4_iif = LOOPBACK_IFINDEX, 294 .flowi4_l3mdev = l3mdev_master_ifindex_rcu(dev), 295 .daddr = ip_hdr(skb)->saddr, 296 .flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(ip_hdr(skb))), 297 .flowi4_scope = scope, 298 .flowi4_mark = vmark ? skb->mark : 0, 299 }; 300 if (!fib_lookup(net, &fl4, &res, 0)) 301 return fib_result_prefsrc(net, &res); 302 } else { 303 scope = RT_SCOPE_LINK; 304 } 305 306 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope); 307 } 308 309 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev) 310 { 311 bool dev_match = false; 312 #ifdef CONFIG_IP_ROUTE_MULTIPATH 313 if (unlikely(fi->nh)) { 314 dev_match = nexthop_uses_dev(fi->nh, dev); 315 } else { 316 int ret; 317 318 for (ret = 0; ret < fib_info_num_path(fi); ret++) { 319 const struct fib_nh_common *nhc = fib_info_nhc(fi, ret); 320 321 if (nhc_l3mdev_matches_dev(nhc, dev)) { 322 dev_match = true; 323 break; 324 } 325 } 326 } 327 #else 328 if (fib_info_nhc(fi, 0)->nhc_dev == dev) 329 dev_match = true; 330 #endif 331 332 return dev_match; 333 } 334 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev); 335 336 /* Given (packet source, input interface) and optional (dst, oif, tos): 337 * - (main) check, that source is valid i.e. not broadcast or our local 338 * address. 339 * - figure out what "logical" interface this packet arrived 340 * and calculate "specific destination" address. 341 * - check, that packet arrived from expected physical interface. 342 * called with rcu_read_lock() 343 */ 344 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 345 dscp_t dscp, int oif, struct net_device *dev, 346 int rpf, struct in_device *idev, u32 *itag) 347 { 348 struct net *net = dev_net(dev); 349 struct flow_keys flkeys; 350 int ret, no_addr; 351 struct fib_result res; 352 struct flowi4 fl4; 353 bool dev_match; 354 355 fl4.flowi4_oif = 0; 356 fl4.flowi4_l3mdev = l3mdev_master_ifindex_rcu(dev); 357 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX; 358 fl4.daddr = src; 359 fl4.saddr = dst; 360 fl4.flowi4_tos = inet_dscp_to_dsfield(dscp); 361 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 362 fl4.flowi4_tun_key.tun_id = 0; 363 fl4.flowi4_flags = 0; 364 fl4.flowi4_uid = sock_net_uid(net, NULL); 365 fl4.flowi4_multipath_hash = 0; 366 367 no_addr = idev->ifa_list == NULL; 368 369 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0; 370 if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) { 371 fl4.flowi4_proto = 0; 372 fl4.fl4_sport = 0; 373 fl4.fl4_dport = 0; 374 } else { 375 swap(fl4.fl4_sport, fl4.fl4_dport); 376 } 377 378 if (fib_lookup(net, &fl4, &res, 0)) 379 goto last_resort; 380 if (res.type != RTN_UNICAST && 381 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev))) 382 goto e_inval; 383 fib_combine_itag(itag, &res); 384 385 dev_match = fib_info_nh_uses_dev(res.fi, dev); 386 /* This is not common, loopback packets retain skb_dst so normally they 387 * would not even hit this slow path. 388 */ 389 dev_match = dev_match || (res.type == RTN_LOCAL && 390 dev == net->loopback_dev); 391 if (dev_match) { 392 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST; 393 return ret; 394 } 395 if (no_addr) 396 goto last_resort; 397 if (rpf == 1) 398 goto e_rpf; 399 fl4.flowi4_oif = dev->ifindex; 400 401 ret = 0; 402 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) { 403 if (res.type == RTN_UNICAST) 404 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST; 405 } 406 return ret; 407 408 last_resort: 409 if (rpf) 410 goto e_rpf; 411 *itag = 0; 412 return 0; 413 414 e_inval: 415 return -EINVAL; 416 e_rpf: 417 return -EXDEV; 418 } 419 420 /* Ignore rp_filter for packets protected by IPsec. */ 421 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 422 dscp_t dscp, int oif, struct net_device *dev, 423 struct in_device *idev, u32 *itag) 424 { 425 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev); 426 struct net *net = dev_net(dev); 427 428 if (!r && !fib_num_tclassid_users(net) && 429 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) { 430 if (IN_DEV_ACCEPT_LOCAL(idev)) 431 goto ok; 432 /* with custom local routes in place, checking local addresses 433 * only will be too optimistic, with custom rules, checking 434 * local addresses only can be too strict, e.g. due to vrf 435 */ 436 if (net->ipv4.fib_has_custom_local_routes || 437 fib4_has_custom_rules(net)) 438 goto full_check; 439 /* Within the same container, it is regarded as a martian source, 440 * and the same host but different containers are not. 441 */ 442 if (inet_lookup_ifaddr_rcu(net, src)) 443 return -EINVAL; 444 445 ok: 446 *itag = 0; 447 return 0; 448 } 449 450 full_check: 451 return __fib_validate_source(skb, src, dst, dscp, oif, dev, r, idev, 452 itag); 453 } 454 455 static inline __be32 sk_extract_addr(struct sockaddr *addr) 456 { 457 return ((struct sockaddr_in *) addr)->sin_addr.s_addr; 458 } 459 460 static int put_rtax(struct nlattr *mx, int len, int type, u32 value) 461 { 462 struct nlattr *nla; 463 464 nla = (struct nlattr *) ((char *) mx + len); 465 nla->nla_type = type; 466 nla->nla_len = nla_attr_size(4); 467 *(u32 *) nla_data(nla) = value; 468 469 return len + nla_total_size(4); 470 } 471 472 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt, 473 struct fib_config *cfg) 474 { 475 __be32 addr; 476 int plen; 477 478 memset(cfg, 0, sizeof(*cfg)); 479 cfg->fc_nlinfo.nl_net = net; 480 481 if (rt->rt_dst.sa_family != AF_INET) 482 return -EAFNOSUPPORT; 483 484 /* 485 * Check mask for validity: 486 * a) it must be contiguous. 487 * b) destination must have all host bits clear. 488 * c) if application forgot to set correct family (AF_INET), 489 * reject request unless it is absolutely clear i.e. 490 * both family and mask are zero. 491 */ 492 plen = 32; 493 addr = sk_extract_addr(&rt->rt_dst); 494 if (!(rt->rt_flags & RTF_HOST)) { 495 __be32 mask = sk_extract_addr(&rt->rt_genmask); 496 497 if (rt->rt_genmask.sa_family != AF_INET) { 498 if (mask || rt->rt_genmask.sa_family) 499 return -EAFNOSUPPORT; 500 } 501 502 if (bad_mask(mask, addr)) 503 return -EINVAL; 504 505 plen = inet_mask_len(mask); 506 } 507 508 cfg->fc_dst_len = plen; 509 cfg->fc_dst = addr; 510 511 if (cmd != SIOCDELRT) { 512 cfg->fc_nlflags = NLM_F_CREATE; 513 cfg->fc_protocol = RTPROT_BOOT; 514 } 515 516 if (rt->rt_metric) 517 cfg->fc_priority = rt->rt_metric - 1; 518 519 if (rt->rt_flags & RTF_REJECT) { 520 cfg->fc_scope = RT_SCOPE_HOST; 521 cfg->fc_type = RTN_UNREACHABLE; 522 return 0; 523 } 524 525 cfg->fc_scope = RT_SCOPE_NOWHERE; 526 cfg->fc_type = RTN_UNICAST; 527 528 if (rt->rt_dev) { 529 char *colon; 530 struct net_device *dev; 531 char devname[IFNAMSIZ]; 532 533 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1)) 534 return -EFAULT; 535 536 devname[IFNAMSIZ-1] = 0; 537 colon = strchr(devname, ':'); 538 if (colon) 539 *colon = 0; 540 dev = __dev_get_by_name(net, devname); 541 if (!dev) 542 return -ENODEV; 543 cfg->fc_oif = dev->ifindex; 544 cfg->fc_table = l3mdev_fib_table(dev); 545 if (colon) { 546 const struct in_ifaddr *ifa; 547 struct in_device *in_dev; 548 549 in_dev = __in_dev_get_rtnl(dev); 550 if (!in_dev) 551 return -ENODEV; 552 553 *colon = ':'; 554 555 rcu_read_lock(); 556 in_dev_for_each_ifa_rcu(ifa, in_dev) { 557 if (strcmp(ifa->ifa_label, devname) == 0) 558 break; 559 } 560 rcu_read_unlock(); 561 562 if (!ifa) 563 return -ENODEV; 564 cfg->fc_prefsrc = ifa->ifa_local; 565 } 566 } 567 568 addr = sk_extract_addr(&rt->rt_gateway); 569 if (rt->rt_gateway.sa_family == AF_INET && addr) { 570 unsigned int addr_type; 571 572 cfg->fc_gw4 = addr; 573 cfg->fc_gw_family = AF_INET; 574 addr_type = inet_addr_type_table(net, addr, cfg->fc_table); 575 if (rt->rt_flags & RTF_GATEWAY && 576 addr_type == RTN_UNICAST) 577 cfg->fc_scope = RT_SCOPE_UNIVERSE; 578 } 579 580 if (!cfg->fc_table) 581 cfg->fc_table = RT_TABLE_MAIN; 582 583 if (cmd == SIOCDELRT) 584 return 0; 585 586 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family) 587 return -EINVAL; 588 589 if (cfg->fc_scope == RT_SCOPE_NOWHERE) 590 cfg->fc_scope = RT_SCOPE_LINK; 591 592 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) { 593 struct nlattr *mx; 594 int len = 0; 595 596 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL); 597 if (!mx) 598 return -ENOMEM; 599 600 if (rt->rt_flags & RTF_MTU) 601 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40); 602 603 if (rt->rt_flags & RTF_WINDOW) 604 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window); 605 606 if (rt->rt_flags & RTF_IRTT) 607 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3); 608 609 cfg->fc_mx = mx; 610 cfg->fc_mx_len = len; 611 } 612 613 return 0; 614 } 615 616 /* 617 * Handle IP routing ioctl calls. 618 * These are used to manipulate the routing tables 619 */ 620 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt) 621 { 622 struct fib_config cfg; 623 int err; 624 625 switch (cmd) { 626 case SIOCADDRT: /* Add a route */ 627 case SIOCDELRT: /* Delete a route */ 628 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 629 return -EPERM; 630 631 rtnl_lock(); 632 err = rtentry_to_fib_config(net, cmd, rt, &cfg); 633 if (err == 0) { 634 struct fib_table *tb; 635 636 if (cmd == SIOCDELRT) { 637 tb = fib_get_table(net, cfg.fc_table); 638 if (tb) 639 err = fib_table_delete(net, tb, &cfg, 640 NULL); 641 else 642 err = -ESRCH; 643 } else { 644 tb = fib_new_table(net, cfg.fc_table); 645 if (tb) 646 err = fib_table_insert(net, tb, 647 &cfg, NULL); 648 else 649 err = -ENOBUFS; 650 } 651 652 /* allocated by rtentry_to_fib_config() */ 653 kfree(cfg.fc_mx); 654 } 655 rtnl_unlock(); 656 return err; 657 } 658 return -EINVAL; 659 } 660 661 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = { 662 [RTA_UNSPEC] = { .strict_start_type = RTA_DPORT + 1 }, 663 [RTA_DST] = { .type = NLA_U32 }, 664 [RTA_SRC] = { .type = NLA_U32 }, 665 [RTA_IIF] = { .type = NLA_U32 }, 666 [RTA_OIF] = { .type = NLA_U32 }, 667 [RTA_GATEWAY] = { .type = NLA_U32 }, 668 [RTA_PRIORITY] = { .type = NLA_U32 }, 669 [RTA_PREFSRC] = { .type = NLA_U32 }, 670 [RTA_METRICS] = { .type = NLA_NESTED }, 671 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) }, 672 [RTA_FLOW] = { .type = NLA_U32 }, 673 [RTA_ENCAP_TYPE] = { .type = NLA_U16 }, 674 [RTA_ENCAP] = { .type = NLA_NESTED }, 675 [RTA_UID] = { .type = NLA_U32 }, 676 [RTA_MARK] = { .type = NLA_U32 }, 677 [RTA_TABLE] = { .type = NLA_U32 }, 678 [RTA_IP_PROTO] = { .type = NLA_U8 }, 679 [RTA_SPORT] = { .type = NLA_U16 }, 680 [RTA_DPORT] = { .type = NLA_U16 }, 681 [RTA_NH_ID] = { .type = NLA_U32 }, 682 }; 683 684 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla, 685 struct netlink_ext_ack *extack) 686 { 687 struct rtvia *via; 688 int alen; 689 690 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) { 691 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA"); 692 return -EINVAL; 693 } 694 695 via = nla_data(nla); 696 alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr); 697 698 switch (via->rtvia_family) { 699 case AF_INET: 700 if (alen != sizeof(__be32)) { 701 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA"); 702 return -EINVAL; 703 } 704 cfg->fc_gw_family = AF_INET; 705 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr); 706 break; 707 case AF_INET6: 708 #if IS_ENABLED(CONFIG_IPV6) 709 if (alen != sizeof(struct in6_addr)) { 710 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA"); 711 return -EINVAL; 712 } 713 cfg->fc_gw_family = AF_INET6; 714 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr); 715 #else 716 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel"); 717 return -EINVAL; 718 #endif 719 break; 720 default: 721 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA"); 722 return -EINVAL; 723 } 724 725 return 0; 726 } 727 728 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb, 729 struct nlmsghdr *nlh, struct fib_config *cfg, 730 struct netlink_ext_ack *extack) 731 { 732 bool has_gw = false, has_via = false; 733 struct nlattr *attr; 734 int err, remaining; 735 struct rtmsg *rtm; 736 737 err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX, 738 rtm_ipv4_policy, extack); 739 if (err < 0) 740 goto errout; 741 742 memset(cfg, 0, sizeof(*cfg)); 743 744 rtm = nlmsg_data(nlh); 745 746 if (!inet_validate_dscp(rtm->rtm_tos)) { 747 NL_SET_ERR_MSG(extack, 748 "Invalid dsfield (tos): ECN bits must be 0"); 749 err = -EINVAL; 750 goto errout; 751 } 752 cfg->fc_dscp = inet_dsfield_to_dscp(rtm->rtm_tos); 753 754 cfg->fc_dst_len = rtm->rtm_dst_len; 755 cfg->fc_table = rtm->rtm_table; 756 cfg->fc_protocol = rtm->rtm_protocol; 757 cfg->fc_scope = rtm->rtm_scope; 758 cfg->fc_type = rtm->rtm_type; 759 cfg->fc_flags = rtm->rtm_flags; 760 cfg->fc_nlflags = nlh->nlmsg_flags; 761 762 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid; 763 cfg->fc_nlinfo.nlh = nlh; 764 cfg->fc_nlinfo.nl_net = net; 765 766 if (cfg->fc_type > RTN_MAX) { 767 NL_SET_ERR_MSG(extack, "Invalid route type"); 768 err = -EINVAL; 769 goto errout; 770 } 771 772 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) { 773 switch (nla_type(attr)) { 774 case RTA_DST: 775 cfg->fc_dst = nla_get_be32(attr); 776 break; 777 case RTA_OIF: 778 cfg->fc_oif = nla_get_u32(attr); 779 break; 780 case RTA_GATEWAY: 781 has_gw = true; 782 cfg->fc_gw4 = nla_get_be32(attr); 783 if (cfg->fc_gw4) 784 cfg->fc_gw_family = AF_INET; 785 break; 786 case RTA_VIA: 787 has_via = true; 788 err = fib_gw_from_via(cfg, attr, extack); 789 if (err) 790 goto errout; 791 break; 792 case RTA_PRIORITY: 793 cfg->fc_priority = nla_get_u32(attr); 794 break; 795 case RTA_PREFSRC: 796 cfg->fc_prefsrc = nla_get_be32(attr); 797 break; 798 case RTA_METRICS: 799 cfg->fc_mx = nla_data(attr); 800 cfg->fc_mx_len = nla_len(attr); 801 break; 802 case RTA_MULTIPATH: 803 err = lwtunnel_valid_encap_type_attr(nla_data(attr), 804 nla_len(attr), 805 extack); 806 if (err < 0) 807 goto errout; 808 cfg->fc_mp = nla_data(attr); 809 cfg->fc_mp_len = nla_len(attr); 810 break; 811 case RTA_FLOW: 812 cfg->fc_flow = nla_get_u32(attr); 813 break; 814 case RTA_TABLE: 815 cfg->fc_table = nla_get_u32(attr); 816 break; 817 case RTA_ENCAP: 818 cfg->fc_encap = attr; 819 break; 820 case RTA_ENCAP_TYPE: 821 cfg->fc_encap_type = nla_get_u16(attr); 822 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, 823 extack); 824 if (err < 0) 825 goto errout; 826 break; 827 case RTA_NH_ID: 828 cfg->fc_nh_id = nla_get_u32(attr); 829 break; 830 } 831 } 832 833 if (cfg->fc_nh_id) { 834 if (cfg->fc_oif || cfg->fc_gw_family || 835 cfg->fc_encap || cfg->fc_mp) { 836 NL_SET_ERR_MSG(extack, 837 "Nexthop specification and nexthop id are mutually exclusive"); 838 return -EINVAL; 839 } 840 } 841 842 if (has_gw && has_via) { 843 NL_SET_ERR_MSG(extack, 844 "Nexthop configuration can not contain both GATEWAY and VIA"); 845 return -EINVAL; 846 } 847 848 if (!cfg->fc_table) 849 cfg->fc_table = RT_TABLE_MAIN; 850 851 return 0; 852 errout: 853 return err; 854 } 855 856 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, 857 struct netlink_ext_ack *extack) 858 { 859 struct net *net = sock_net(skb->sk); 860 struct fib_config cfg; 861 struct fib_table *tb; 862 int err; 863 864 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack); 865 if (err < 0) 866 goto errout; 867 868 if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) { 869 NL_SET_ERR_MSG(extack, "Nexthop id does not exist"); 870 err = -EINVAL; 871 goto errout; 872 } 873 874 tb = fib_get_table(net, cfg.fc_table); 875 if (!tb) { 876 NL_SET_ERR_MSG(extack, "FIB table does not exist"); 877 err = -ESRCH; 878 goto errout; 879 } 880 881 err = fib_table_delete(net, tb, &cfg, extack); 882 errout: 883 return err; 884 } 885 886 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, 887 struct netlink_ext_ack *extack) 888 { 889 struct net *net = sock_net(skb->sk); 890 struct fib_config cfg; 891 struct fib_table *tb; 892 int err; 893 894 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack); 895 if (err < 0) 896 goto errout; 897 898 tb = fib_new_table(net, cfg.fc_table); 899 if (!tb) { 900 err = -ENOBUFS; 901 goto errout; 902 } 903 904 err = fib_table_insert(net, tb, &cfg, extack); 905 if (!err && cfg.fc_type == RTN_LOCAL) 906 net->ipv4.fib_has_custom_local_routes = true; 907 errout: 908 return err; 909 } 910 911 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh, 912 struct fib_dump_filter *filter, 913 struct netlink_callback *cb) 914 { 915 struct netlink_ext_ack *extack = cb->extack; 916 struct nlattr *tb[RTA_MAX + 1]; 917 struct rtmsg *rtm; 918 int err, i; 919 920 if (filter->rtnl_held) 921 ASSERT_RTNL(); 922 923 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) { 924 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request"); 925 return -EINVAL; 926 } 927 928 rtm = nlmsg_data(nlh); 929 if (rtm->rtm_dst_len || rtm->rtm_src_len || rtm->rtm_tos || 930 rtm->rtm_scope) { 931 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request"); 932 return -EINVAL; 933 } 934 935 if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) { 936 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request"); 937 return -EINVAL; 938 } 939 if (rtm->rtm_flags & RTM_F_CLONED) 940 filter->dump_routes = false; 941 else 942 filter->dump_exceptions = false; 943 944 filter->flags = rtm->rtm_flags; 945 filter->protocol = rtm->rtm_protocol; 946 filter->rt_type = rtm->rtm_type; 947 filter->table_id = rtm->rtm_table; 948 949 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX, 950 rtm_ipv4_policy, extack); 951 if (err < 0) 952 return err; 953 954 for (i = 0; i <= RTA_MAX; ++i) { 955 int ifindex; 956 957 if (!tb[i]) 958 continue; 959 960 switch (i) { 961 case RTA_TABLE: 962 filter->table_id = nla_get_u32(tb[i]); 963 break; 964 case RTA_OIF: 965 ifindex = nla_get_u32(tb[i]); 966 if (filter->rtnl_held) 967 filter->dev = __dev_get_by_index(net, ifindex); 968 else 969 filter->dev = dev_get_by_index_rcu(net, ifindex); 970 if (!filter->dev) 971 return -ENODEV; 972 break; 973 default: 974 NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request"); 975 return -EINVAL; 976 } 977 } 978 979 if (filter->flags || filter->protocol || filter->rt_type || 980 filter->table_id || filter->dev) { 981 filter->filter_set = 1; 982 cb->answer_flags = NLM_F_DUMP_FILTERED; 983 } 984 985 return 0; 986 } 987 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req); 988 989 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 990 { 991 struct fib_dump_filter filter = { 992 .dump_routes = true, 993 .dump_exceptions = true, 994 .rtnl_held = false, 995 }; 996 const struct nlmsghdr *nlh = cb->nlh; 997 struct net *net = sock_net(skb->sk); 998 unsigned int h, s_h; 999 unsigned int e = 0, s_e; 1000 struct fib_table *tb; 1001 struct hlist_head *head; 1002 int dumped = 0, err = 0; 1003 1004 rcu_read_lock(); 1005 if (cb->strict_check) { 1006 err = ip_valid_fib_dump_req(net, nlh, &filter, cb); 1007 if (err < 0) 1008 goto unlock; 1009 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) { 1010 struct rtmsg *rtm = nlmsg_data(nlh); 1011 1012 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED); 1013 } 1014 1015 /* ipv4 does not use prefix flag */ 1016 if (filter.flags & RTM_F_PREFIX) 1017 goto unlock; 1018 1019 if (filter.table_id) { 1020 tb = fib_get_table(net, filter.table_id); 1021 if (!tb) { 1022 if (rtnl_msg_family(cb->nlh) != PF_INET) 1023 goto unlock; 1024 1025 NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist"); 1026 err = -ENOENT; 1027 goto unlock; 1028 } 1029 err = fib_table_dump(tb, skb, cb, &filter); 1030 goto unlock; 1031 } 1032 1033 s_h = cb->args[0]; 1034 s_e = cb->args[1]; 1035 1036 err = 0; 1037 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) { 1038 e = 0; 1039 head = &net->ipv4.fib_table_hash[h]; 1040 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 1041 if (e < s_e) 1042 goto next; 1043 if (dumped) 1044 memset(&cb->args[2], 0, sizeof(cb->args) - 1045 2 * sizeof(cb->args[0])); 1046 err = fib_table_dump(tb, skb, cb, &filter); 1047 if (err < 0) 1048 goto out; 1049 dumped = 1; 1050 next: 1051 e++; 1052 } 1053 } 1054 out: 1055 1056 cb->args[1] = e; 1057 cb->args[0] = h; 1058 1059 unlock: 1060 rcu_read_unlock(); 1061 return err; 1062 } 1063 1064 /* Prepare and feed intra-kernel routing request. 1065 * Really, it should be netlink message, but :-( netlink 1066 * can be not configured, so that we feed it directly 1067 * to fib engine. It is legal, because all events occur 1068 * only when netlink is already locked. 1069 */ 1070 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, 1071 struct in_ifaddr *ifa, u32 rt_priority) 1072 { 1073 struct net *net = dev_net(ifa->ifa_dev->dev); 1074 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev); 1075 struct fib_table *tb; 1076 struct fib_config cfg = { 1077 .fc_protocol = RTPROT_KERNEL, 1078 .fc_type = type, 1079 .fc_dst = dst, 1080 .fc_dst_len = dst_len, 1081 .fc_priority = rt_priority, 1082 .fc_prefsrc = ifa->ifa_local, 1083 .fc_oif = ifa->ifa_dev->dev->ifindex, 1084 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND, 1085 .fc_nlinfo = { 1086 .nl_net = net, 1087 }, 1088 }; 1089 1090 if (!tb_id) 1091 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL; 1092 1093 tb = fib_new_table(net, tb_id); 1094 if (!tb) 1095 return; 1096 1097 cfg.fc_table = tb->tb_id; 1098 1099 if (type != RTN_LOCAL) 1100 cfg.fc_scope = RT_SCOPE_LINK; 1101 else 1102 cfg.fc_scope = RT_SCOPE_HOST; 1103 1104 if (cmd == RTM_NEWROUTE) 1105 fib_table_insert(net, tb, &cfg, NULL); 1106 else 1107 fib_table_delete(net, tb, &cfg, NULL); 1108 } 1109 1110 void fib_add_ifaddr(struct in_ifaddr *ifa) 1111 { 1112 struct in_device *in_dev = ifa->ifa_dev; 1113 struct net_device *dev = in_dev->dev; 1114 struct in_ifaddr *prim = ifa; 1115 __be32 mask = ifa->ifa_mask; 1116 __be32 addr = ifa->ifa_local; 1117 __be32 prefix = ifa->ifa_address & mask; 1118 1119 if (ifa->ifa_flags & IFA_F_SECONDARY) { 1120 prim = inet_ifa_byprefix(in_dev, prefix, mask); 1121 if (!prim) { 1122 pr_warn("%s: bug: prim == NULL\n", __func__); 1123 return; 1124 } 1125 } 1126 1127 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0); 1128 1129 if (!(dev->flags & IFF_UP)) 1130 return; 1131 1132 /* Add broadcast address, if it is explicitly assigned. */ 1133 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) { 1134 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, 1135 prim, 0); 1136 arp_invalidate(dev, ifa->ifa_broadcast, false); 1137 } 1138 1139 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) && 1140 (prefix != addr || ifa->ifa_prefixlen < 32)) { 1141 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 1142 fib_magic(RTM_NEWROUTE, 1143 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 1144 prefix, ifa->ifa_prefixlen, prim, 1145 ifa->ifa_rt_priority); 1146 1147 /* Add the network broadcast address, when it makes sense */ 1148 if (ifa->ifa_prefixlen < 31) { 1149 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask, 1150 32, prim, 0); 1151 arp_invalidate(dev, prefix | ~mask, false); 1152 } 1153 } 1154 } 1155 1156 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric) 1157 { 1158 __be32 prefix = ifa->ifa_address & ifa->ifa_mask; 1159 struct in_device *in_dev = ifa->ifa_dev; 1160 struct net_device *dev = in_dev->dev; 1161 1162 if (!(dev->flags & IFF_UP) || 1163 ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) || 1164 ipv4_is_zeronet(prefix) || 1165 (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32)) 1166 return; 1167 1168 /* add the new */ 1169 fib_magic(RTM_NEWROUTE, 1170 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 1171 prefix, ifa->ifa_prefixlen, ifa, new_metric); 1172 1173 /* delete the old */ 1174 fib_magic(RTM_DELROUTE, 1175 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 1176 prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority); 1177 } 1178 1179 /* Delete primary or secondary address. 1180 * Optionally, on secondary address promotion consider the addresses 1181 * from subnet iprim as deleted, even if they are in device list. 1182 * In this case the secondary ifa can be in device list. 1183 */ 1184 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim) 1185 { 1186 struct in_device *in_dev = ifa->ifa_dev; 1187 struct net_device *dev = in_dev->dev; 1188 struct in_ifaddr *ifa1; 1189 struct in_ifaddr *prim = ifa, *prim1 = NULL; 1190 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask; 1191 __be32 any = ifa->ifa_address & ifa->ifa_mask; 1192 #define LOCAL_OK 1 1193 #define BRD_OK 2 1194 #define BRD0_OK 4 1195 #define BRD1_OK 8 1196 unsigned int ok = 0; 1197 int subnet = 0; /* Primary network */ 1198 int gone = 1; /* Address is missing */ 1199 int same_prefsrc = 0; /* Another primary with same IP */ 1200 1201 if (ifa->ifa_flags & IFA_F_SECONDARY) { 1202 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask); 1203 if (!prim) { 1204 /* if the device has been deleted, we don't perform 1205 * address promotion 1206 */ 1207 if (!in_dev->dead) 1208 pr_warn("%s: bug: prim == NULL\n", __func__); 1209 return; 1210 } 1211 if (iprim && iprim != prim) { 1212 pr_warn("%s: bug: iprim != prim\n", __func__); 1213 return; 1214 } 1215 } else if (!ipv4_is_zeronet(any) && 1216 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) { 1217 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 1218 fib_magic(RTM_DELROUTE, 1219 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 1220 any, ifa->ifa_prefixlen, prim, 0); 1221 subnet = 1; 1222 } 1223 1224 if (in_dev->dead) 1225 goto no_promotions; 1226 1227 /* Deletion is more complicated than add. 1228 * We should take care of not to delete too much :-) 1229 * 1230 * Scan address list to be sure that addresses are really gone. 1231 */ 1232 rcu_read_lock(); 1233 in_dev_for_each_ifa_rcu(ifa1, in_dev) { 1234 if (ifa1 == ifa) { 1235 /* promotion, keep the IP */ 1236 gone = 0; 1237 continue; 1238 } 1239 /* Ignore IFAs from our subnet */ 1240 if (iprim && ifa1->ifa_mask == iprim->ifa_mask && 1241 inet_ifa_match(ifa1->ifa_address, iprim)) 1242 continue; 1243 1244 /* Ignore ifa1 if it uses different primary IP (prefsrc) */ 1245 if (ifa1->ifa_flags & IFA_F_SECONDARY) { 1246 /* Another address from our subnet? */ 1247 if (ifa1->ifa_mask == prim->ifa_mask && 1248 inet_ifa_match(ifa1->ifa_address, prim)) 1249 prim1 = prim; 1250 else { 1251 /* We reached the secondaries, so 1252 * same_prefsrc should be determined. 1253 */ 1254 if (!same_prefsrc) 1255 continue; 1256 /* Search new prim1 if ifa1 is not 1257 * using the current prim1 1258 */ 1259 if (!prim1 || 1260 ifa1->ifa_mask != prim1->ifa_mask || 1261 !inet_ifa_match(ifa1->ifa_address, prim1)) 1262 prim1 = inet_ifa_byprefix(in_dev, 1263 ifa1->ifa_address, 1264 ifa1->ifa_mask); 1265 if (!prim1) 1266 continue; 1267 if (prim1->ifa_local != prim->ifa_local) 1268 continue; 1269 } 1270 } else { 1271 if (prim->ifa_local != ifa1->ifa_local) 1272 continue; 1273 prim1 = ifa1; 1274 if (prim != prim1) 1275 same_prefsrc = 1; 1276 } 1277 if (ifa->ifa_local == ifa1->ifa_local) 1278 ok |= LOCAL_OK; 1279 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 1280 ok |= BRD_OK; 1281 if (brd == ifa1->ifa_broadcast) 1282 ok |= BRD1_OK; 1283 if (any == ifa1->ifa_broadcast) 1284 ok |= BRD0_OK; 1285 /* primary has network specific broadcasts */ 1286 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) { 1287 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask; 1288 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask; 1289 1290 if (!ipv4_is_zeronet(any1)) { 1291 if (ifa->ifa_broadcast == brd1 || 1292 ifa->ifa_broadcast == any1) 1293 ok |= BRD_OK; 1294 if (brd == brd1 || brd == any1) 1295 ok |= BRD1_OK; 1296 if (any == brd1 || any == any1) 1297 ok |= BRD0_OK; 1298 } 1299 } 1300 } 1301 rcu_read_unlock(); 1302 1303 no_promotions: 1304 if (!(ok & BRD_OK)) 1305 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, 1306 prim, 0); 1307 if (subnet && ifa->ifa_prefixlen < 31) { 1308 if (!(ok & BRD1_OK)) 1309 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, 1310 prim, 0); 1311 if (!(ok & BRD0_OK)) 1312 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, 1313 prim, 0); 1314 } 1315 if (!(ok & LOCAL_OK)) { 1316 unsigned int addr_type; 1317 1318 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0); 1319 1320 /* Check, that this local address finally disappeared. */ 1321 addr_type = inet_addr_type_dev_table(dev_net(dev), dev, 1322 ifa->ifa_local); 1323 if (gone && addr_type != RTN_LOCAL) { 1324 /* And the last, but not the least thing. 1325 * We must flush stray FIB entries. 1326 * 1327 * First of all, we scan fib_info list searching 1328 * for stray nexthop entries, then ignite fib_flush. 1329 */ 1330 if (fib_sync_down_addr(dev, ifa->ifa_local)) 1331 fib_flush(dev_net(dev)); 1332 } 1333 } 1334 #undef LOCAL_OK 1335 #undef BRD_OK 1336 #undef BRD0_OK 1337 #undef BRD1_OK 1338 } 1339 1340 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn) 1341 { 1342 1343 struct fib_result res; 1344 struct flowi4 fl4 = { 1345 .flowi4_mark = frn->fl_mark, 1346 .daddr = frn->fl_addr, 1347 .flowi4_tos = frn->fl_tos & INET_DSCP_MASK, 1348 .flowi4_scope = frn->fl_scope, 1349 }; 1350 struct fib_table *tb; 1351 1352 rcu_read_lock(); 1353 1354 tb = fib_get_table(net, frn->tb_id_in); 1355 1356 frn->err = -ENOENT; 1357 if (tb) { 1358 local_bh_disable(); 1359 1360 frn->tb_id = tb->tb_id; 1361 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 1362 1363 if (!frn->err) { 1364 frn->prefixlen = res.prefixlen; 1365 frn->nh_sel = res.nh_sel; 1366 frn->type = res.type; 1367 frn->scope = res.scope; 1368 } 1369 local_bh_enable(); 1370 } 1371 1372 rcu_read_unlock(); 1373 } 1374 1375 static void nl_fib_input(struct sk_buff *skb) 1376 { 1377 struct net *net; 1378 struct fib_result_nl *frn; 1379 struct nlmsghdr *nlh; 1380 u32 portid; 1381 1382 net = sock_net(skb->sk); 1383 nlh = nlmsg_hdr(skb); 1384 if (skb->len < nlmsg_total_size(sizeof(*frn)) || 1385 skb->len < nlh->nlmsg_len || 1386 nlmsg_len(nlh) < sizeof(*frn)) 1387 return; 1388 1389 skb = netlink_skb_clone(skb, GFP_KERNEL); 1390 if (!skb) 1391 return; 1392 nlh = nlmsg_hdr(skb); 1393 1394 frn = nlmsg_data(nlh); 1395 nl_fib_lookup(net, frn); 1396 1397 portid = NETLINK_CB(skb).portid; /* netlink portid */ 1398 NETLINK_CB(skb).portid = 0; /* from kernel */ 1399 NETLINK_CB(skb).dst_group = 0; /* unicast */ 1400 nlmsg_unicast(net->ipv4.fibnl, skb, portid); 1401 } 1402 1403 static int __net_init nl_fib_lookup_init(struct net *net) 1404 { 1405 struct sock *sk; 1406 struct netlink_kernel_cfg cfg = { 1407 .input = nl_fib_input, 1408 }; 1409 1410 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg); 1411 if (!sk) 1412 return -EAFNOSUPPORT; 1413 net->ipv4.fibnl = sk; 1414 return 0; 1415 } 1416 1417 static void nl_fib_lookup_exit(struct net *net) 1418 { 1419 netlink_kernel_release(net->ipv4.fibnl); 1420 net->ipv4.fibnl = NULL; 1421 } 1422 1423 static void fib_disable_ip(struct net_device *dev, unsigned long event, 1424 bool force) 1425 { 1426 if (fib_sync_down_dev(dev, event, force)) 1427 fib_flush(dev_net(dev)); 1428 else 1429 rt_cache_flush(dev_net(dev)); 1430 arp_ifdown(dev); 1431 } 1432 1433 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 1434 { 1435 struct in_ifaddr *ifa = ptr; 1436 struct net_device *dev = ifa->ifa_dev->dev; 1437 struct net *net = dev_net(dev); 1438 1439 switch (event) { 1440 case NETDEV_UP: 1441 fib_add_ifaddr(ifa); 1442 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1443 fib_sync_up(dev, RTNH_F_DEAD); 1444 #endif 1445 atomic_inc(&net->ipv4.dev_addr_genid); 1446 rt_cache_flush(dev_net(dev)); 1447 break; 1448 case NETDEV_DOWN: 1449 fib_del_ifaddr(ifa, NULL); 1450 atomic_inc(&net->ipv4.dev_addr_genid); 1451 if (!ifa->ifa_dev->ifa_list) { 1452 /* Last address was deleted from this interface. 1453 * Disable IP. 1454 */ 1455 fib_disable_ip(dev, event, true); 1456 } else { 1457 rt_cache_flush(dev_net(dev)); 1458 } 1459 break; 1460 } 1461 return NOTIFY_DONE; 1462 } 1463 1464 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 1465 { 1466 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1467 struct netdev_notifier_changeupper_info *upper_info = ptr; 1468 struct netdev_notifier_info_ext *info_ext = ptr; 1469 struct in_device *in_dev; 1470 struct net *net = dev_net(dev); 1471 struct in_ifaddr *ifa; 1472 unsigned int flags; 1473 1474 if (event == NETDEV_UNREGISTER) { 1475 fib_disable_ip(dev, event, true); 1476 rt_flush_dev(dev); 1477 return NOTIFY_DONE; 1478 } 1479 1480 in_dev = __in_dev_get_rtnl(dev); 1481 if (!in_dev) 1482 return NOTIFY_DONE; 1483 1484 switch (event) { 1485 case NETDEV_UP: 1486 in_dev_for_each_ifa_rtnl(ifa, in_dev) { 1487 fib_add_ifaddr(ifa); 1488 } 1489 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1490 fib_sync_up(dev, RTNH_F_DEAD); 1491 #endif 1492 atomic_inc(&net->ipv4.dev_addr_genid); 1493 rt_cache_flush(net); 1494 break; 1495 case NETDEV_DOWN: 1496 fib_disable_ip(dev, event, false); 1497 break; 1498 case NETDEV_CHANGE: 1499 flags = dev_get_flags(dev); 1500 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 1501 fib_sync_up(dev, RTNH_F_LINKDOWN); 1502 else 1503 fib_sync_down_dev(dev, event, false); 1504 rt_cache_flush(net); 1505 break; 1506 case NETDEV_CHANGEMTU: 1507 fib_sync_mtu(dev, info_ext->ext.mtu); 1508 rt_cache_flush(net); 1509 break; 1510 case NETDEV_CHANGEUPPER: 1511 upper_info = ptr; 1512 /* flush all routes if dev is linked to or unlinked from 1513 * an L3 master device (e.g., VRF) 1514 */ 1515 if (upper_info->upper_dev && 1516 netif_is_l3_master(upper_info->upper_dev)) 1517 fib_disable_ip(dev, NETDEV_DOWN, true); 1518 break; 1519 } 1520 return NOTIFY_DONE; 1521 } 1522 1523 static struct notifier_block fib_inetaddr_notifier = { 1524 .notifier_call = fib_inetaddr_event, 1525 }; 1526 1527 static struct notifier_block fib_netdev_notifier = { 1528 .notifier_call = fib_netdev_event, 1529 }; 1530 1531 static int __net_init ip_fib_net_init(struct net *net) 1532 { 1533 int err; 1534 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ; 1535 1536 err = fib4_notifier_init(net); 1537 if (err) 1538 return err; 1539 1540 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1541 /* Default to 3-tuple */ 1542 net->ipv4.sysctl_fib_multipath_hash_fields = 1543 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK; 1544 #endif 1545 1546 /* Avoid false sharing : Use at least a full cache line */ 1547 size = max_t(size_t, size, L1_CACHE_BYTES); 1548 1549 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL); 1550 if (!net->ipv4.fib_table_hash) { 1551 err = -ENOMEM; 1552 goto err_table_hash_alloc; 1553 } 1554 1555 err = fib4_rules_init(net); 1556 if (err < 0) 1557 goto err_rules_init; 1558 return 0; 1559 1560 err_rules_init: 1561 kfree(net->ipv4.fib_table_hash); 1562 err_table_hash_alloc: 1563 fib4_notifier_exit(net); 1564 return err; 1565 } 1566 1567 static void ip_fib_net_exit(struct net *net) 1568 { 1569 int i; 1570 1571 ASSERT_RTNL(); 1572 #ifdef CONFIG_IP_MULTIPLE_TABLES 1573 RCU_INIT_POINTER(net->ipv4.fib_main, NULL); 1574 RCU_INIT_POINTER(net->ipv4.fib_default, NULL); 1575 #endif 1576 /* Destroy the tables in reverse order to guarantee that the 1577 * local table, ID 255, is destroyed before the main table, ID 1578 * 254. This is necessary as the local table may contain 1579 * references to data contained in the main table. 1580 */ 1581 for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) { 1582 struct hlist_head *head = &net->ipv4.fib_table_hash[i]; 1583 struct hlist_node *tmp; 1584 struct fib_table *tb; 1585 1586 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) { 1587 hlist_del(&tb->tb_hlist); 1588 fib_table_flush(net, tb, true); 1589 fib_free_table(tb); 1590 } 1591 } 1592 1593 #ifdef CONFIG_IP_MULTIPLE_TABLES 1594 fib4_rules_exit(net); 1595 #endif 1596 1597 kfree(net->ipv4.fib_table_hash); 1598 fib4_notifier_exit(net); 1599 } 1600 1601 static int __net_init fib_net_init(struct net *net) 1602 { 1603 int error; 1604 1605 #ifdef CONFIG_IP_ROUTE_CLASSID 1606 atomic_set(&net->ipv4.fib_num_tclassid_users, 0); 1607 #endif 1608 error = ip_fib_net_init(net); 1609 if (error < 0) 1610 goto out; 1611 error = nl_fib_lookup_init(net); 1612 if (error < 0) 1613 goto out_nlfl; 1614 error = fib_proc_init(net); 1615 if (error < 0) 1616 goto out_proc; 1617 out: 1618 return error; 1619 1620 out_proc: 1621 nl_fib_lookup_exit(net); 1622 out_nlfl: 1623 rtnl_lock(); 1624 ip_fib_net_exit(net); 1625 rtnl_unlock(); 1626 goto out; 1627 } 1628 1629 static void __net_exit fib_net_exit(struct net *net) 1630 { 1631 fib_proc_exit(net); 1632 nl_fib_lookup_exit(net); 1633 } 1634 1635 static void __net_exit fib_net_exit_batch(struct list_head *net_list) 1636 { 1637 struct net *net; 1638 1639 rtnl_lock(); 1640 list_for_each_entry(net, net_list, exit_list) 1641 ip_fib_net_exit(net); 1642 1643 rtnl_unlock(); 1644 } 1645 1646 static struct pernet_operations fib_net_ops = { 1647 .init = fib_net_init, 1648 .exit = fib_net_exit, 1649 .exit_batch = fib_net_exit_batch, 1650 }; 1651 1652 static const struct rtnl_msg_handler fib_rtnl_msg_handlers[] __initconst = { 1653 {.protocol = PF_INET, .msgtype = RTM_NEWROUTE, 1654 .doit = inet_rtm_newroute}, 1655 {.protocol = PF_INET, .msgtype = RTM_DELROUTE, 1656 .doit = inet_rtm_delroute}, 1657 {.protocol = PF_INET, .msgtype = RTM_GETROUTE, .dumpit = inet_dump_fib, 1658 .flags = RTNL_FLAG_DUMP_UNLOCKED | RTNL_FLAG_DUMP_SPLIT_NLM_DONE}, 1659 }; 1660 1661 void __init ip_fib_init(void) 1662 { 1663 fib_trie_init(); 1664 1665 register_pernet_subsys(&fib_net_ops); 1666 1667 register_netdevice_notifier(&fib_netdev_notifier); 1668 register_inetaddr_notifier(&fib_inetaddr_notifier); 1669 1670 rtnl_register_many(fib_rtnl_msg_handlers); 1671 } 1672