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_oif = l3mdev_master_ifindex_rcu(dev), 295 .daddr = ip_hdr(skb)->saddr, 296 .flowi4_tos = ip_hdr(skb)->tos & IPTOS_RT_MASK, 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 u8 tos, 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_iif = l3mdev_master_ifindex_rcu(dev); 357 if (!fl4.flowi4_iif) 358 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX; 359 fl4.daddr = src; 360 fl4.saddr = dst; 361 fl4.flowi4_tos = tos; 362 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 363 fl4.flowi4_tun_key.tun_id = 0; 364 fl4.flowi4_flags = 0; 365 fl4.flowi4_uid = sock_net_uid(net, NULL); 366 fl4.flowi4_multipath_hash = 0; 367 368 no_addr = idev->ifa_list == NULL; 369 370 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0; 371 if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) { 372 fl4.flowi4_proto = 0; 373 fl4.fl4_sport = 0; 374 fl4.fl4_dport = 0; 375 } else { 376 swap(fl4.fl4_sport, fl4.fl4_dport); 377 } 378 379 if (fib_lookup(net, &fl4, &res, 0)) 380 goto last_resort; 381 if (res.type != RTN_UNICAST && 382 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev))) 383 goto e_inval; 384 fib_combine_itag(itag, &res); 385 386 dev_match = fib_info_nh_uses_dev(res.fi, dev); 387 /* This is not common, loopback packets retain skb_dst so normally they 388 * would not even hit this slow path. 389 */ 390 dev_match = dev_match || (res.type == RTN_LOCAL && 391 dev == net->loopback_dev); 392 if (dev_match) { 393 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST; 394 return ret; 395 } 396 if (no_addr) 397 goto last_resort; 398 if (rpf == 1) 399 goto e_rpf; 400 fl4.flowi4_oif = dev->ifindex; 401 402 ret = 0; 403 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) { 404 if (res.type == RTN_UNICAST) 405 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST; 406 } 407 return ret; 408 409 last_resort: 410 if (rpf) 411 goto e_rpf; 412 *itag = 0; 413 return 0; 414 415 e_inval: 416 return -EINVAL; 417 e_rpf: 418 return -EXDEV; 419 } 420 421 /* Ignore rp_filter for packets protected by IPsec. */ 422 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 423 u8 tos, int oif, struct net_device *dev, 424 struct in_device *idev, u32 *itag) 425 { 426 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev); 427 struct net *net = dev_net(dev); 428 429 if (!r && !fib_num_tclassid_users(net) && 430 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) { 431 if (IN_DEV_ACCEPT_LOCAL(idev)) 432 goto ok; 433 /* with custom local routes in place, checking local addresses 434 * only will be too optimistic, with custom rules, checking 435 * local addresses only can be too strict, e.g. due to vrf 436 */ 437 if (net->ipv4.fib_has_custom_local_routes || 438 fib4_has_custom_rules(net)) 439 goto full_check; 440 /* Within the same container, it is regarded as a martian source, 441 * and the same host but different containers are not. 442 */ 443 if (inet_lookup_ifaddr_rcu(net, src)) 444 return -EINVAL; 445 446 ok: 447 *itag = 0; 448 return 0; 449 } 450 451 full_check: 452 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, 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 (cmd == SIOCDELRT) 581 return 0; 582 583 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family) 584 return -EINVAL; 585 586 if (cfg->fc_scope == RT_SCOPE_NOWHERE) 587 cfg->fc_scope = RT_SCOPE_LINK; 588 589 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) { 590 struct nlattr *mx; 591 int len = 0; 592 593 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL); 594 if (!mx) 595 return -ENOMEM; 596 597 if (rt->rt_flags & RTF_MTU) 598 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40); 599 600 if (rt->rt_flags & RTF_WINDOW) 601 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window); 602 603 if (rt->rt_flags & RTF_IRTT) 604 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3); 605 606 cfg->fc_mx = mx; 607 cfg->fc_mx_len = len; 608 } 609 610 return 0; 611 } 612 613 /* 614 * Handle IP routing ioctl calls. 615 * These are used to manipulate the routing tables 616 */ 617 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt) 618 { 619 struct fib_config cfg; 620 int err; 621 622 switch (cmd) { 623 case SIOCADDRT: /* Add a route */ 624 case SIOCDELRT: /* Delete a route */ 625 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 626 return -EPERM; 627 628 rtnl_lock(); 629 err = rtentry_to_fib_config(net, cmd, rt, &cfg); 630 if (err == 0) { 631 struct fib_table *tb; 632 633 if (cmd == SIOCDELRT) { 634 tb = fib_get_table(net, cfg.fc_table); 635 if (tb) 636 err = fib_table_delete(net, tb, &cfg, 637 NULL); 638 else 639 err = -ESRCH; 640 } else { 641 tb = fib_new_table(net, cfg.fc_table); 642 if (tb) 643 err = fib_table_insert(net, tb, 644 &cfg, NULL); 645 else 646 err = -ENOBUFS; 647 } 648 649 /* allocated by rtentry_to_fib_config() */ 650 kfree(cfg.fc_mx); 651 } 652 rtnl_unlock(); 653 return err; 654 } 655 return -EINVAL; 656 } 657 658 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = { 659 [RTA_UNSPEC] = { .strict_start_type = RTA_DPORT + 1 }, 660 [RTA_DST] = { .type = NLA_U32 }, 661 [RTA_SRC] = { .type = NLA_U32 }, 662 [RTA_IIF] = { .type = NLA_U32 }, 663 [RTA_OIF] = { .type = NLA_U32 }, 664 [RTA_GATEWAY] = { .type = NLA_U32 }, 665 [RTA_PRIORITY] = { .type = NLA_U32 }, 666 [RTA_PREFSRC] = { .type = NLA_U32 }, 667 [RTA_METRICS] = { .type = NLA_NESTED }, 668 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) }, 669 [RTA_FLOW] = { .type = NLA_U32 }, 670 [RTA_ENCAP_TYPE] = { .type = NLA_U16 }, 671 [RTA_ENCAP] = { .type = NLA_NESTED }, 672 [RTA_UID] = { .type = NLA_U32 }, 673 [RTA_MARK] = { .type = NLA_U32 }, 674 [RTA_TABLE] = { .type = NLA_U32 }, 675 [RTA_IP_PROTO] = { .type = NLA_U8 }, 676 [RTA_SPORT] = { .type = NLA_U16 }, 677 [RTA_DPORT] = { .type = NLA_U16 }, 678 [RTA_NH_ID] = { .type = NLA_U32 }, 679 }; 680 681 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla, 682 struct netlink_ext_ack *extack) 683 { 684 struct rtvia *via; 685 int alen; 686 687 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) { 688 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA"); 689 return -EINVAL; 690 } 691 692 via = nla_data(nla); 693 alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr); 694 695 switch (via->rtvia_family) { 696 case AF_INET: 697 if (alen != sizeof(__be32)) { 698 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA"); 699 return -EINVAL; 700 } 701 cfg->fc_gw_family = AF_INET; 702 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr); 703 break; 704 case AF_INET6: 705 #if IS_ENABLED(CONFIG_IPV6) 706 if (alen != sizeof(struct in6_addr)) { 707 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA"); 708 return -EINVAL; 709 } 710 cfg->fc_gw_family = AF_INET6; 711 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr); 712 #else 713 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel"); 714 return -EINVAL; 715 #endif 716 break; 717 default: 718 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA"); 719 return -EINVAL; 720 } 721 722 return 0; 723 } 724 725 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb, 726 struct nlmsghdr *nlh, struct fib_config *cfg, 727 struct netlink_ext_ack *extack) 728 { 729 bool has_gw = false, has_via = false; 730 struct nlattr *attr; 731 int err, remaining; 732 struct rtmsg *rtm; 733 734 err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX, 735 rtm_ipv4_policy, extack); 736 if (err < 0) 737 goto errout; 738 739 memset(cfg, 0, sizeof(*cfg)); 740 741 rtm = nlmsg_data(nlh); 742 743 if (!inet_validate_dscp(rtm->rtm_tos)) { 744 NL_SET_ERR_MSG(extack, 745 "Invalid dsfield (tos): ECN bits must be 0"); 746 err = -EINVAL; 747 goto errout; 748 } 749 cfg->fc_dscp = inet_dsfield_to_dscp(rtm->rtm_tos); 750 751 cfg->fc_dst_len = rtm->rtm_dst_len; 752 cfg->fc_table = rtm->rtm_table; 753 cfg->fc_protocol = rtm->rtm_protocol; 754 cfg->fc_scope = rtm->rtm_scope; 755 cfg->fc_type = rtm->rtm_type; 756 cfg->fc_flags = rtm->rtm_flags; 757 cfg->fc_nlflags = nlh->nlmsg_flags; 758 759 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid; 760 cfg->fc_nlinfo.nlh = nlh; 761 cfg->fc_nlinfo.nl_net = net; 762 763 if (cfg->fc_type > RTN_MAX) { 764 NL_SET_ERR_MSG(extack, "Invalid route type"); 765 err = -EINVAL; 766 goto errout; 767 } 768 769 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) { 770 switch (nla_type(attr)) { 771 case RTA_DST: 772 cfg->fc_dst = nla_get_be32(attr); 773 break; 774 case RTA_OIF: 775 cfg->fc_oif = nla_get_u32(attr); 776 break; 777 case RTA_GATEWAY: 778 has_gw = true; 779 cfg->fc_gw4 = nla_get_be32(attr); 780 if (cfg->fc_gw4) 781 cfg->fc_gw_family = AF_INET; 782 break; 783 case RTA_VIA: 784 has_via = true; 785 err = fib_gw_from_via(cfg, attr, extack); 786 if (err) 787 goto errout; 788 break; 789 case RTA_PRIORITY: 790 cfg->fc_priority = nla_get_u32(attr); 791 break; 792 case RTA_PREFSRC: 793 cfg->fc_prefsrc = nla_get_be32(attr); 794 break; 795 case RTA_METRICS: 796 cfg->fc_mx = nla_data(attr); 797 cfg->fc_mx_len = nla_len(attr); 798 break; 799 case RTA_MULTIPATH: 800 err = lwtunnel_valid_encap_type_attr(nla_data(attr), 801 nla_len(attr), 802 extack); 803 if (err < 0) 804 goto errout; 805 cfg->fc_mp = nla_data(attr); 806 cfg->fc_mp_len = nla_len(attr); 807 break; 808 case RTA_FLOW: 809 cfg->fc_flow = nla_get_u32(attr); 810 break; 811 case RTA_TABLE: 812 cfg->fc_table = nla_get_u32(attr); 813 break; 814 case RTA_ENCAP: 815 cfg->fc_encap = attr; 816 break; 817 case RTA_ENCAP_TYPE: 818 cfg->fc_encap_type = nla_get_u16(attr); 819 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, 820 extack); 821 if (err < 0) 822 goto errout; 823 break; 824 case RTA_NH_ID: 825 cfg->fc_nh_id = nla_get_u32(attr); 826 break; 827 } 828 } 829 830 if (cfg->fc_nh_id) { 831 if (cfg->fc_oif || cfg->fc_gw_family || 832 cfg->fc_encap || cfg->fc_mp) { 833 NL_SET_ERR_MSG(extack, 834 "Nexthop specification and nexthop id are mutually exclusive"); 835 return -EINVAL; 836 } 837 } 838 839 if (has_gw && has_via) { 840 NL_SET_ERR_MSG(extack, 841 "Nexthop configuration can not contain both GATEWAY and VIA"); 842 return -EINVAL; 843 } 844 845 return 0; 846 errout: 847 return err; 848 } 849 850 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, 851 struct netlink_ext_ack *extack) 852 { 853 struct net *net = sock_net(skb->sk); 854 struct fib_config cfg; 855 struct fib_table *tb; 856 int err; 857 858 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack); 859 if (err < 0) 860 goto errout; 861 862 if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) { 863 NL_SET_ERR_MSG(extack, "Nexthop id does not exist"); 864 err = -EINVAL; 865 goto errout; 866 } 867 868 tb = fib_get_table(net, cfg.fc_table); 869 if (!tb) { 870 NL_SET_ERR_MSG(extack, "FIB table does not exist"); 871 err = -ESRCH; 872 goto errout; 873 } 874 875 err = fib_table_delete(net, tb, &cfg, extack); 876 errout: 877 return err; 878 } 879 880 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, 881 struct netlink_ext_ack *extack) 882 { 883 struct net *net = sock_net(skb->sk); 884 struct fib_config cfg; 885 struct fib_table *tb; 886 int err; 887 888 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack); 889 if (err < 0) 890 goto errout; 891 892 tb = fib_new_table(net, cfg.fc_table); 893 if (!tb) { 894 err = -ENOBUFS; 895 goto errout; 896 } 897 898 err = fib_table_insert(net, tb, &cfg, extack); 899 if (!err && cfg.fc_type == RTN_LOCAL) 900 net->ipv4.fib_has_custom_local_routes = true; 901 errout: 902 return err; 903 } 904 905 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh, 906 struct fib_dump_filter *filter, 907 struct netlink_callback *cb) 908 { 909 struct netlink_ext_ack *extack = cb->extack; 910 struct nlattr *tb[RTA_MAX + 1]; 911 struct rtmsg *rtm; 912 int err, i; 913 914 ASSERT_RTNL(); 915 916 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) { 917 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request"); 918 return -EINVAL; 919 } 920 921 rtm = nlmsg_data(nlh); 922 if (rtm->rtm_dst_len || rtm->rtm_src_len || rtm->rtm_tos || 923 rtm->rtm_scope) { 924 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request"); 925 return -EINVAL; 926 } 927 928 if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) { 929 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request"); 930 return -EINVAL; 931 } 932 if (rtm->rtm_flags & RTM_F_CLONED) 933 filter->dump_routes = false; 934 else 935 filter->dump_exceptions = false; 936 937 filter->flags = rtm->rtm_flags; 938 filter->protocol = rtm->rtm_protocol; 939 filter->rt_type = rtm->rtm_type; 940 filter->table_id = rtm->rtm_table; 941 942 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX, 943 rtm_ipv4_policy, extack); 944 if (err < 0) 945 return err; 946 947 for (i = 0; i <= RTA_MAX; ++i) { 948 int ifindex; 949 950 if (!tb[i]) 951 continue; 952 953 switch (i) { 954 case RTA_TABLE: 955 filter->table_id = nla_get_u32(tb[i]); 956 break; 957 case RTA_OIF: 958 ifindex = nla_get_u32(tb[i]); 959 filter->dev = __dev_get_by_index(net, ifindex); 960 if (!filter->dev) 961 return -ENODEV; 962 break; 963 default: 964 NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request"); 965 return -EINVAL; 966 } 967 } 968 969 if (filter->flags || filter->protocol || filter->rt_type || 970 filter->table_id || filter->dev) { 971 filter->filter_set = 1; 972 cb->answer_flags = NLM_F_DUMP_FILTERED; 973 } 974 975 return 0; 976 } 977 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req); 978 979 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 980 { 981 struct fib_dump_filter filter = { .dump_routes = true, 982 .dump_exceptions = true }; 983 const struct nlmsghdr *nlh = cb->nlh; 984 struct net *net = sock_net(skb->sk); 985 unsigned int h, s_h; 986 unsigned int e = 0, s_e; 987 struct fib_table *tb; 988 struct hlist_head *head; 989 int dumped = 0, err; 990 991 if (cb->strict_check) { 992 err = ip_valid_fib_dump_req(net, nlh, &filter, cb); 993 if (err < 0) 994 return err; 995 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) { 996 struct rtmsg *rtm = nlmsg_data(nlh); 997 998 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED); 999 } 1000 1001 /* ipv4 does not use prefix flag */ 1002 if (filter.flags & RTM_F_PREFIX) 1003 return skb->len; 1004 1005 if (filter.table_id) { 1006 tb = fib_get_table(net, filter.table_id); 1007 if (!tb) { 1008 if (rtnl_msg_family(cb->nlh) != PF_INET) 1009 return skb->len; 1010 1011 NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist"); 1012 return -ENOENT; 1013 } 1014 1015 rcu_read_lock(); 1016 err = fib_table_dump(tb, skb, cb, &filter); 1017 rcu_read_unlock(); 1018 return skb->len ? : err; 1019 } 1020 1021 s_h = cb->args[0]; 1022 s_e = cb->args[1]; 1023 1024 rcu_read_lock(); 1025 1026 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) { 1027 e = 0; 1028 head = &net->ipv4.fib_table_hash[h]; 1029 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 1030 if (e < s_e) 1031 goto next; 1032 if (dumped) 1033 memset(&cb->args[2], 0, sizeof(cb->args) - 1034 2 * sizeof(cb->args[0])); 1035 err = fib_table_dump(tb, skb, cb, &filter); 1036 if (err < 0) { 1037 if (likely(skb->len)) 1038 goto out; 1039 1040 goto out_err; 1041 } 1042 dumped = 1; 1043 next: 1044 e++; 1045 } 1046 } 1047 out: 1048 err = skb->len; 1049 out_err: 1050 rcu_read_unlock(); 1051 1052 cb->args[1] = e; 1053 cb->args[0] = h; 1054 1055 return err; 1056 } 1057 1058 /* Prepare and feed intra-kernel routing request. 1059 * Really, it should be netlink message, but :-( netlink 1060 * can be not configured, so that we feed it directly 1061 * to fib engine. It is legal, because all events occur 1062 * only when netlink is already locked. 1063 */ 1064 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, 1065 struct in_ifaddr *ifa, u32 rt_priority) 1066 { 1067 struct net *net = dev_net(ifa->ifa_dev->dev); 1068 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev); 1069 struct fib_table *tb; 1070 struct fib_config cfg = { 1071 .fc_protocol = RTPROT_KERNEL, 1072 .fc_type = type, 1073 .fc_dst = dst, 1074 .fc_dst_len = dst_len, 1075 .fc_priority = rt_priority, 1076 .fc_prefsrc = ifa->ifa_local, 1077 .fc_oif = ifa->ifa_dev->dev->ifindex, 1078 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND, 1079 .fc_nlinfo = { 1080 .nl_net = net, 1081 }, 1082 }; 1083 1084 if (!tb_id) 1085 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL; 1086 1087 tb = fib_new_table(net, tb_id); 1088 if (!tb) 1089 return; 1090 1091 cfg.fc_table = tb->tb_id; 1092 1093 if (type != RTN_LOCAL) 1094 cfg.fc_scope = RT_SCOPE_LINK; 1095 else 1096 cfg.fc_scope = RT_SCOPE_HOST; 1097 1098 if (cmd == RTM_NEWROUTE) 1099 fib_table_insert(net, tb, &cfg, NULL); 1100 else 1101 fib_table_delete(net, tb, &cfg, NULL); 1102 } 1103 1104 void fib_add_ifaddr(struct in_ifaddr *ifa) 1105 { 1106 struct in_device *in_dev = ifa->ifa_dev; 1107 struct net_device *dev = in_dev->dev; 1108 struct in_ifaddr *prim = ifa; 1109 __be32 mask = ifa->ifa_mask; 1110 __be32 addr = ifa->ifa_local; 1111 __be32 prefix = ifa->ifa_address & mask; 1112 1113 if (ifa->ifa_flags & IFA_F_SECONDARY) { 1114 prim = inet_ifa_byprefix(in_dev, prefix, mask); 1115 if (!prim) { 1116 pr_warn("%s: bug: prim == NULL\n", __func__); 1117 return; 1118 } 1119 } 1120 1121 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0); 1122 1123 if (!(dev->flags & IFF_UP)) 1124 return; 1125 1126 /* Add broadcast address, if it is explicitly assigned. */ 1127 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) { 1128 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, 1129 prim, 0); 1130 arp_invalidate(dev, ifa->ifa_broadcast, false); 1131 } 1132 1133 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) && 1134 (prefix != addr || ifa->ifa_prefixlen < 32)) { 1135 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 1136 fib_magic(RTM_NEWROUTE, 1137 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 1138 prefix, ifa->ifa_prefixlen, prim, 1139 ifa->ifa_rt_priority); 1140 1141 /* Add the network broadcast address, when it makes sense */ 1142 if (ifa->ifa_prefixlen < 31) { 1143 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask, 1144 32, prim, 0); 1145 arp_invalidate(dev, prefix | ~mask, false); 1146 } 1147 } 1148 } 1149 1150 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric) 1151 { 1152 __be32 prefix = ifa->ifa_address & ifa->ifa_mask; 1153 struct in_device *in_dev = ifa->ifa_dev; 1154 struct net_device *dev = in_dev->dev; 1155 1156 if (!(dev->flags & IFF_UP) || 1157 ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) || 1158 ipv4_is_zeronet(prefix) || 1159 (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32)) 1160 return; 1161 1162 /* add the new */ 1163 fib_magic(RTM_NEWROUTE, 1164 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 1165 prefix, ifa->ifa_prefixlen, ifa, new_metric); 1166 1167 /* delete the old */ 1168 fib_magic(RTM_DELROUTE, 1169 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 1170 prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority); 1171 } 1172 1173 /* Delete primary or secondary address. 1174 * Optionally, on secondary address promotion consider the addresses 1175 * from subnet iprim as deleted, even if they are in device list. 1176 * In this case the secondary ifa can be in device list. 1177 */ 1178 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim) 1179 { 1180 struct in_device *in_dev = ifa->ifa_dev; 1181 struct net_device *dev = in_dev->dev; 1182 struct in_ifaddr *ifa1; 1183 struct in_ifaddr *prim = ifa, *prim1 = NULL; 1184 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask; 1185 __be32 any = ifa->ifa_address & ifa->ifa_mask; 1186 #define LOCAL_OK 1 1187 #define BRD_OK 2 1188 #define BRD0_OK 4 1189 #define BRD1_OK 8 1190 unsigned int ok = 0; 1191 int subnet = 0; /* Primary network */ 1192 int gone = 1; /* Address is missing */ 1193 int same_prefsrc = 0; /* Another primary with same IP */ 1194 1195 if (ifa->ifa_flags & IFA_F_SECONDARY) { 1196 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask); 1197 if (!prim) { 1198 /* if the device has been deleted, we don't perform 1199 * address promotion 1200 */ 1201 if (!in_dev->dead) 1202 pr_warn("%s: bug: prim == NULL\n", __func__); 1203 return; 1204 } 1205 if (iprim && iprim != prim) { 1206 pr_warn("%s: bug: iprim != prim\n", __func__); 1207 return; 1208 } 1209 } else if (!ipv4_is_zeronet(any) && 1210 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) { 1211 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 1212 fib_magic(RTM_DELROUTE, 1213 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 1214 any, ifa->ifa_prefixlen, prim, 0); 1215 subnet = 1; 1216 } 1217 1218 if (in_dev->dead) 1219 goto no_promotions; 1220 1221 /* Deletion is more complicated than add. 1222 * We should take care of not to delete too much :-) 1223 * 1224 * Scan address list to be sure that addresses are really gone. 1225 */ 1226 rcu_read_lock(); 1227 in_dev_for_each_ifa_rcu(ifa1, in_dev) { 1228 if (ifa1 == ifa) { 1229 /* promotion, keep the IP */ 1230 gone = 0; 1231 continue; 1232 } 1233 /* Ignore IFAs from our subnet */ 1234 if (iprim && ifa1->ifa_mask == iprim->ifa_mask && 1235 inet_ifa_match(ifa1->ifa_address, iprim)) 1236 continue; 1237 1238 /* Ignore ifa1 if it uses different primary IP (prefsrc) */ 1239 if (ifa1->ifa_flags & IFA_F_SECONDARY) { 1240 /* Another address from our subnet? */ 1241 if (ifa1->ifa_mask == prim->ifa_mask && 1242 inet_ifa_match(ifa1->ifa_address, prim)) 1243 prim1 = prim; 1244 else { 1245 /* We reached the secondaries, so 1246 * same_prefsrc should be determined. 1247 */ 1248 if (!same_prefsrc) 1249 continue; 1250 /* Search new prim1 if ifa1 is not 1251 * using the current prim1 1252 */ 1253 if (!prim1 || 1254 ifa1->ifa_mask != prim1->ifa_mask || 1255 !inet_ifa_match(ifa1->ifa_address, prim1)) 1256 prim1 = inet_ifa_byprefix(in_dev, 1257 ifa1->ifa_address, 1258 ifa1->ifa_mask); 1259 if (!prim1) 1260 continue; 1261 if (prim1->ifa_local != prim->ifa_local) 1262 continue; 1263 } 1264 } else { 1265 if (prim->ifa_local != ifa1->ifa_local) 1266 continue; 1267 prim1 = ifa1; 1268 if (prim != prim1) 1269 same_prefsrc = 1; 1270 } 1271 if (ifa->ifa_local == ifa1->ifa_local) 1272 ok |= LOCAL_OK; 1273 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 1274 ok |= BRD_OK; 1275 if (brd == ifa1->ifa_broadcast) 1276 ok |= BRD1_OK; 1277 if (any == ifa1->ifa_broadcast) 1278 ok |= BRD0_OK; 1279 /* primary has network specific broadcasts */ 1280 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) { 1281 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask; 1282 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask; 1283 1284 if (!ipv4_is_zeronet(any1)) { 1285 if (ifa->ifa_broadcast == brd1 || 1286 ifa->ifa_broadcast == any1) 1287 ok |= BRD_OK; 1288 if (brd == brd1 || brd == any1) 1289 ok |= BRD1_OK; 1290 if (any == brd1 || any == any1) 1291 ok |= BRD0_OK; 1292 } 1293 } 1294 } 1295 rcu_read_unlock(); 1296 1297 no_promotions: 1298 if (!(ok & BRD_OK)) 1299 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, 1300 prim, 0); 1301 if (subnet && ifa->ifa_prefixlen < 31) { 1302 if (!(ok & BRD1_OK)) 1303 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, 1304 prim, 0); 1305 if (!(ok & BRD0_OK)) 1306 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, 1307 prim, 0); 1308 } 1309 if (!(ok & LOCAL_OK)) { 1310 unsigned int addr_type; 1311 1312 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0); 1313 1314 /* Check, that this local address finally disappeared. */ 1315 addr_type = inet_addr_type_dev_table(dev_net(dev), dev, 1316 ifa->ifa_local); 1317 if (gone && addr_type != RTN_LOCAL) { 1318 /* And the last, but not the least thing. 1319 * We must flush stray FIB entries. 1320 * 1321 * First of all, we scan fib_info list searching 1322 * for stray nexthop entries, then ignite fib_flush. 1323 */ 1324 if (fib_sync_down_addr(dev, ifa->ifa_local)) 1325 fib_flush(dev_net(dev)); 1326 } 1327 } 1328 #undef LOCAL_OK 1329 #undef BRD_OK 1330 #undef BRD0_OK 1331 #undef BRD1_OK 1332 } 1333 1334 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn) 1335 { 1336 1337 struct fib_result res; 1338 struct flowi4 fl4 = { 1339 .flowi4_mark = frn->fl_mark, 1340 .daddr = frn->fl_addr, 1341 .flowi4_tos = frn->fl_tos, 1342 .flowi4_scope = frn->fl_scope, 1343 }; 1344 struct fib_table *tb; 1345 1346 rcu_read_lock(); 1347 1348 tb = fib_get_table(net, frn->tb_id_in); 1349 1350 frn->err = -ENOENT; 1351 if (tb) { 1352 local_bh_disable(); 1353 1354 frn->tb_id = tb->tb_id; 1355 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 1356 1357 if (!frn->err) { 1358 frn->prefixlen = res.prefixlen; 1359 frn->nh_sel = res.nh_sel; 1360 frn->type = res.type; 1361 frn->scope = res.scope; 1362 } 1363 local_bh_enable(); 1364 } 1365 1366 rcu_read_unlock(); 1367 } 1368 1369 static void nl_fib_input(struct sk_buff *skb) 1370 { 1371 struct net *net; 1372 struct fib_result_nl *frn; 1373 struct nlmsghdr *nlh; 1374 u32 portid; 1375 1376 net = sock_net(skb->sk); 1377 nlh = nlmsg_hdr(skb); 1378 if (skb->len < nlmsg_total_size(sizeof(*frn)) || 1379 skb->len < nlh->nlmsg_len || 1380 nlmsg_len(nlh) < sizeof(*frn)) 1381 return; 1382 1383 skb = netlink_skb_clone(skb, GFP_KERNEL); 1384 if (!skb) 1385 return; 1386 nlh = nlmsg_hdr(skb); 1387 1388 frn = (struct fib_result_nl *) nlmsg_data(nlh); 1389 nl_fib_lookup(net, frn); 1390 1391 portid = NETLINK_CB(skb).portid; /* netlink portid */ 1392 NETLINK_CB(skb).portid = 0; /* from kernel */ 1393 NETLINK_CB(skb).dst_group = 0; /* unicast */ 1394 nlmsg_unicast(net->ipv4.fibnl, skb, portid); 1395 } 1396 1397 static int __net_init nl_fib_lookup_init(struct net *net) 1398 { 1399 struct sock *sk; 1400 struct netlink_kernel_cfg cfg = { 1401 .input = nl_fib_input, 1402 }; 1403 1404 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg); 1405 if (!sk) 1406 return -EAFNOSUPPORT; 1407 net->ipv4.fibnl = sk; 1408 return 0; 1409 } 1410 1411 static void nl_fib_lookup_exit(struct net *net) 1412 { 1413 netlink_kernel_release(net->ipv4.fibnl); 1414 net->ipv4.fibnl = NULL; 1415 } 1416 1417 static void fib_disable_ip(struct net_device *dev, unsigned long event, 1418 bool force) 1419 { 1420 if (fib_sync_down_dev(dev, event, force)) 1421 fib_flush(dev_net(dev)); 1422 else 1423 rt_cache_flush(dev_net(dev)); 1424 arp_ifdown(dev); 1425 } 1426 1427 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 1428 { 1429 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; 1430 struct net_device *dev = ifa->ifa_dev->dev; 1431 struct net *net = dev_net(dev); 1432 1433 switch (event) { 1434 case NETDEV_UP: 1435 fib_add_ifaddr(ifa); 1436 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1437 fib_sync_up(dev, RTNH_F_DEAD); 1438 #endif 1439 atomic_inc(&net->ipv4.dev_addr_genid); 1440 rt_cache_flush(dev_net(dev)); 1441 break; 1442 case NETDEV_DOWN: 1443 fib_del_ifaddr(ifa, NULL); 1444 atomic_inc(&net->ipv4.dev_addr_genid); 1445 if (!ifa->ifa_dev->ifa_list) { 1446 /* Last address was deleted from this interface. 1447 * Disable IP. 1448 */ 1449 fib_disable_ip(dev, event, true); 1450 } else { 1451 rt_cache_flush(dev_net(dev)); 1452 } 1453 break; 1454 } 1455 return NOTIFY_DONE; 1456 } 1457 1458 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 1459 { 1460 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1461 struct netdev_notifier_changeupper_info *upper_info = ptr; 1462 struct netdev_notifier_info_ext *info_ext = ptr; 1463 struct in_device *in_dev; 1464 struct net *net = dev_net(dev); 1465 struct in_ifaddr *ifa; 1466 unsigned int flags; 1467 1468 if (event == NETDEV_UNREGISTER) { 1469 fib_disable_ip(dev, event, true); 1470 rt_flush_dev(dev); 1471 return NOTIFY_DONE; 1472 } 1473 1474 in_dev = __in_dev_get_rtnl(dev); 1475 if (!in_dev) 1476 return NOTIFY_DONE; 1477 1478 switch (event) { 1479 case NETDEV_UP: 1480 in_dev_for_each_ifa_rtnl(ifa, in_dev) { 1481 fib_add_ifaddr(ifa); 1482 } 1483 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1484 fib_sync_up(dev, RTNH_F_DEAD); 1485 #endif 1486 atomic_inc(&net->ipv4.dev_addr_genid); 1487 rt_cache_flush(net); 1488 break; 1489 case NETDEV_DOWN: 1490 fib_disable_ip(dev, event, false); 1491 break; 1492 case NETDEV_CHANGE: 1493 flags = dev_get_flags(dev); 1494 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 1495 fib_sync_up(dev, RTNH_F_LINKDOWN); 1496 else 1497 fib_sync_down_dev(dev, event, false); 1498 rt_cache_flush(net); 1499 break; 1500 case NETDEV_CHANGEMTU: 1501 fib_sync_mtu(dev, info_ext->ext.mtu); 1502 rt_cache_flush(net); 1503 break; 1504 case NETDEV_CHANGEUPPER: 1505 upper_info = ptr; 1506 /* flush all routes if dev is linked to or unlinked from 1507 * an L3 master device (e.g., VRF) 1508 */ 1509 if (upper_info->upper_dev && 1510 netif_is_l3_master(upper_info->upper_dev)) 1511 fib_disable_ip(dev, NETDEV_DOWN, true); 1512 break; 1513 } 1514 return NOTIFY_DONE; 1515 } 1516 1517 static struct notifier_block fib_inetaddr_notifier = { 1518 .notifier_call = fib_inetaddr_event, 1519 }; 1520 1521 static struct notifier_block fib_netdev_notifier = { 1522 .notifier_call = fib_netdev_event, 1523 }; 1524 1525 static int __net_init ip_fib_net_init(struct net *net) 1526 { 1527 int err; 1528 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ; 1529 1530 err = fib4_notifier_init(net); 1531 if (err) 1532 return err; 1533 1534 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1535 /* Default to 3-tuple */ 1536 net->ipv4.sysctl_fib_multipath_hash_fields = 1537 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK; 1538 #endif 1539 1540 /* Avoid false sharing : Use at least a full cache line */ 1541 size = max_t(size_t, size, L1_CACHE_BYTES); 1542 1543 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL); 1544 if (!net->ipv4.fib_table_hash) { 1545 err = -ENOMEM; 1546 goto err_table_hash_alloc; 1547 } 1548 1549 err = fib4_rules_init(net); 1550 if (err < 0) 1551 goto err_rules_init; 1552 return 0; 1553 1554 err_rules_init: 1555 kfree(net->ipv4.fib_table_hash); 1556 err_table_hash_alloc: 1557 fib4_notifier_exit(net); 1558 return err; 1559 } 1560 1561 static void ip_fib_net_exit(struct net *net) 1562 { 1563 int i; 1564 1565 ASSERT_RTNL(); 1566 #ifdef CONFIG_IP_MULTIPLE_TABLES 1567 RCU_INIT_POINTER(net->ipv4.fib_main, NULL); 1568 RCU_INIT_POINTER(net->ipv4.fib_default, NULL); 1569 #endif 1570 /* Destroy the tables in reverse order to guarantee that the 1571 * local table, ID 255, is destroyed before the main table, ID 1572 * 254. This is necessary as the local table may contain 1573 * references to data contained in the main table. 1574 */ 1575 for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) { 1576 struct hlist_head *head = &net->ipv4.fib_table_hash[i]; 1577 struct hlist_node *tmp; 1578 struct fib_table *tb; 1579 1580 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) { 1581 hlist_del(&tb->tb_hlist); 1582 fib_table_flush(net, tb, true); 1583 fib_free_table(tb); 1584 } 1585 } 1586 1587 #ifdef CONFIG_IP_MULTIPLE_TABLES 1588 fib4_rules_exit(net); 1589 #endif 1590 1591 kfree(net->ipv4.fib_table_hash); 1592 fib4_notifier_exit(net); 1593 } 1594 1595 static int __net_init fib_net_init(struct net *net) 1596 { 1597 int error; 1598 1599 #ifdef CONFIG_IP_ROUTE_CLASSID 1600 atomic_set(&net->ipv4.fib_num_tclassid_users, 0); 1601 #endif 1602 error = ip_fib_net_init(net); 1603 if (error < 0) 1604 goto out; 1605 error = nl_fib_lookup_init(net); 1606 if (error < 0) 1607 goto out_nlfl; 1608 error = fib_proc_init(net); 1609 if (error < 0) 1610 goto out_proc; 1611 out: 1612 return error; 1613 1614 out_proc: 1615 nl_fib_lookup_exit(net); 1616 out_nlfl: 1617 rtnl_lock(); 1618 ip_fib_net_exit(net); 1619 rtnl_unlock(); 1620 goto out; 1621 } 1622 1623 static void __net_exit fib_net_exit(struct net *net) 1624 { 1625 fib_proc_exit(net); 1626 nl_fib_lookup_exit(net); 1627 } 1628 1629 static void __net_exit fib_net_exit_batch(struct list_head *net_list) 1630 { 1631 struct net *net; 1632 1633 rtnl_lock(); 1634 list_for_each_entry(net, net_list, exit_list) 1635 ip_fib_net_exit(net); 1636 1637 rtnl_unlock(); 1638 } 1639 1640 static struct pernet_operations fib_net_ops = { 1641 .init = fib_net_init, 1642 .exit = fib_net_exit, 1643 .exit_batch = fib_net_exit_batch, 1644 }; 1645 1646 void __init ip_fib_init(void) 1647 { 1648 fib_trie_init(); 1649 1650 register_pernet_subsys(&fib_net_ops); 1651 1652 register_netdevice_notifier(&fib_netdev_notifier); 1653 register_inetaddr_notifier(&fib_inetaddr_notifier); 1654 1655 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0); 1656 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0); 1657 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0); 1658 } 1659