1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * IPv4 Forwarding Information Base: FIB frontend. 7 * 8 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 9 * 10 * This program is free software; you can redistribute it and/or 11 * modify it under the terms of the GNU General Public License 12 * as published by the Free Software Foundation; either version 13 * 2 of the License, or (at your option) any later version. 14 */ 15 16 #include <linux/module.h> 17 #include <linux/uaccess.h> 18 #include <linux/bitops.h> 19 #include <linux/capability.h> 20 #include <linux/types.h> 21 #include <linux/kernel.h> 22 #include <linux/mm.h> 23 #include <linux/string.h> 24 #include <linux/socket.h> 25 #include <linux/sockios.h> 26 #include <linux/errno.h> 27 #include <linux/in.h> 28 #include <linux/inet.h> 29 #include <linux/inetdevice.h> 30 #include <linux/netdevice.h> 31 #include <linux/if_addr.h> 32 #include <linux/if_arp.h> 33 #include <linux/skbuff.h> 34 #include <linux/cache.h> 35 #include <linux/init.h> 36 #include <linux/list.h> 37 #include <linux/slab.h> 38 39 #include <net/ip.h> 40 #include <net/protocol.h> 41 #include <net/route.h> 42 #include <net/tcp.h> 43 #include <net/sock.h> 44 #include <net/arp.h> 45 #include <net/ip_fib.h> 46 #include <net/rtnetlink.h> 47 #include <net/xfrm.h> 48 #include <net/l3mdev.h> 49 #include <net/lwtunnel.h> 50 #include <trace/events/fib.h> 51 52 #ifndef CONFIG_IP_MULTIPLE_TABLES 53 54 static int __net_init fib4_rules_init(struct net *net) 55 { 56 struct fib_table *local_table, *main_table; 57 58 main_table = fib_trie_table(RT_TABLE_MAIN, NULL); 59 if (!main_table) 60 return -ENOMEM; 61 62 local_table = fib_trie_table(RT_TABLE_LOCAL, main_table); 63 if (!local_table) 64 goto fail; 65 66 hlist_add_head_rcu(&local_table->tb_hlist, 67 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]); 68 hlist_add_head_rcu(&main_table->tb_hlist, 69 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]); 70 return 0; 71 72 fail: 73 fib_free_table(main_table); 74 return -ENOMEM; 75 } 76 77 static bool fib4_has_custom_rules(struct net *net) 78 { 79 return false; 80 } 81 #else 82 83 struct fib_table *fib_new_table(struct net *net, u32 id) 84 { 85 struct fib_table *tb, *alias = NULL; 86 unsigned int h; 87 88 if (id == 0) 89 id = RT_TABLE_MAIN; 90 tb = fib_get_table(net, id); 91 if (tb) 92 return tb; 93 94 if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules) 95 alias = fib_new_table(net, RT_TABLE_MAIN); 96 97 tb = fib_trie_table(id, alias); 98 if (!tb) 99 return NULL; 100 101 switch (id) { 102 case RT_TABLE_MAIN: 103 rcu_assign_pointer(net->ipv4.fib_main, tb); 104 break; 105 case RT_TABLE_DEFAULT: 106 rcu_assign_pointer(net->ipv4.fib_default, tb); 107 break; 108 default: 109 break; 110 } 111 112 h = id & (FIB_TABLE_HASHSZ - 1); 113 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]); 114 return tb; 115 } 116 EXPORT_SYMBOL_GPL(fib_new_table); 117 118 /* caller must hold either rtnl or rcu read lock */ 119 struct fib_table *fib_get_table(struct net *net, u32 id) 120 { 121 struct fib_table *tb; 122 struct hlist_head *head; 123 unsigned int h; 124 125 if (id == 0) 126 id = RT_TABLE_MAIN; 127 h = id & (FIB_TABLE_HASHSZ - 1); 128 129 head = &net->ipv4.fib_table_hash[h]; 130 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 131 if (tb->tb_id == id) 132 return tb; 133 } 134 return NULL; 135 } 136 137 static bool fib4_has_custom_rules(struct net *net) 138 { 139 return net->ipv4.fib_has_custom_rules; 140 } 141 #endif /* CONFIG_IP_MULTIPLE_TABLES */ 142 143 static void fib_replace_table(struct net *net, struct fib_table *old, 144 struct fib_table *new) 145 { 146 #ifdef CONFIG_IP_MULTIPLE_TABLES 147 switch (new->tb_id) { 148 case RT_TABLE_MAIN: 149 rcu_assign_pointer(net->ipv4.fib_main, new); 150 break; 151 case RT_TABLE_DEFAULT: 152 rcu_assign_pointer(net->ipv4.fib_default, new); 153 break; 154 default: 155 break; 156 } 157 158 #endif 159 /* replace the old table in the hlist */ 160 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist); 161 } 162 163 int fib_unmerge(struct net *net) 164 { 165 struct fib_table *old, *new, *main_table; 166 167 /* attempt to fetch local table if it has been allocated */ 168 old = fib_get_table(net, RT_TABLE_LOCAL); 169 if (!old) 170 return 0; 171 172 new = fib_trie_unmerge(old); 173 if (!new) 174 return -ENOMEM; 175 176 /* table is already unmerged */ 177 if (new == old) 178 return 0; 179 180 /* replace merged table with clean table */ 181 fib_replace_table(net, old, new); 182 fib_free_table(old); 183 184 /* attempt to fetch main table if it has been allocated */ 185 main_table = fib_get_table(net, RT_TABLE_MAIN); 186 if (!main_table) 187 return 0; 188 189 /* flush local entries from main table */ 190 fib_table_flush_external(main_table); 191 192 return 0; 193 } 194 195 static void fib_flush(struct net *net) 196 { 197 int flushed = 0; 198 unsigned int h; 199 200 for (h = 0; h < FIB_TABLE_HASHSZ; h++) { 201 struct hlist_head *head = &net->ipv4.fib_table_hash[h]; 202 struct hlist_node *tmp; 203 struct fib_table *tb; 204 205 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) 206 flushed += fib_table_flush(net, tb); 207 } 208 209 if (flushed) 210 rt_cache_flush(net); 211 } 212 213 /* 214 * Find address type as if only "dev" was present in the system. If 215 * on_dev is NULL then all interfaces are taken into consideration. 216 */ 217 static inline unsigned int __inet_dev_addr_type(struct net *net, 218 const struct net_device *dev, 219 __be32 addr, u32 tb_id) 220 { 221 struct flowi4 fl4 = { .daddr = addr }; 222 struct fib_result res; 223 unsigned int ret = RTN_BROADCAST; 224 struct fib_table *table; 225 226 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr)) 227 return RTN_BROADCAST; 228 if (ipv4_is_multicast(addr)) 229 return RTN_MULTICAST; 230 231 rcu_read_lock(); 232 233 table = fib_get_table(net, tb_id); 234 if (table) { 235 ret = RTN_UNICAST; 236 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) { 237 if (!dev || dev == res.fi->fib_dev) 238 ret = res.type; 239 } 240 } 241 242 rcu_read_unlock(); 243 return ret; 244 } 245 246 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id) 247 { 248 return __inet_dev_addr_type(net, NULL, addr, tb_id); 249 } 250 EXPORT_SYMBOL(inet_addr_type_table); 251 252 unsigned int inet_addr_type(struct net *net, __be32 addr) 253 { 254 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL); 255 } 256 EXPORT_SYMBOL(inet_addr_type); 257 258 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev, 259 __be32 addr) 260 { 261 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL; 262 263 return __inet_dev_addr_type(net, dev, addr, rt_table); 264 } 265 EXPORT_SYMBOL(inet_dev_addr_type); 266 267 /* inet_addr_type with dev == NULL but using the table from a dev 268 * if one is associated 269 */ 270 unsigned int inet_addr_type_dev_table(struct net *net, 271 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, NULL, addr, rt_table); 277 } 278 EXPORT_SYMBOL(inet_addr_type_dev_table); 279 280 __be32 fib_compute_spec_dst(struct sk_buff *skb) 281 { 282 struct net_device *dev = skb->dev; 283 struct in_device *in_dev; 284 struct fib_result res; 285 struct rtable *rt; 286 struct net *net; 287 int scope; 288 289 rt = skb_rtable(skb); 290 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) == 291 RTCF_LOCAL) 292 return ip_hdr(skb)->daddr; 293 294 in_dev = __in_dev_get_rcu(dev); 295 296 net = dev_net(dev); 297 298 scope = RT_SCOPE_UNIVERSE; 299 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) { 300 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev); 301 struct flowi4 fl4 = { 302 .flowi4_iif = LOOPBACK_IFINDEX, 303 .flowi4_oif = l3mdev_master_ifindex_rcu(dev), 304 .daddr = ip_hdr(skb)->saddr, 305 .flowi4_tos = RT_TOS(ip_hdr(skb)->tos), 306 .flowi4_scope = scope, 307 .flowi4_mark = vmark ? skb->mark : 0, 308 }; 309 if (!fib_lookup(net, &fl4, &res, 0)) 310 return FIB_RES_PREFSRC(net, res); 311 } else { 312 scope = RT_SCOPE_LINK; 313 } 314 315 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope); 316 } 317 318 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev) 319 { 320 bool dev_match = false; 321 #ifdef CONFIG_IP_ROUTE_MULTIPATH 322 int ret; 323 324 for (ret = 0; ret < fi->fib_nhs; ret++) { 325 struct fib_nh *nh = &fi->fib_nh[ret]; 326 327 if (nh->nh_dev == dev) { 328 dev_match = true; 329 break; 330 } else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) { 331 dev_match = true; 332 break; 333 } 334 } 335 #else 336 if (fi->fib_nh[0].nh_dev == dev) 337 dev_match = true; 338 #endif 339 340 return dev_match; 341 } 342 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev); 343 344 /* Given (packet source, input interface) and optional (dst, oif, tos): 345 * - (main) check, that source is valid i.e. not broadcast or our local 346 * address. 347 * - figure out what "logical" interface this packet arrived 348 * and calculate "specific destination" address. 349 * - check, that packet arrived from expected physical interface. 350 * called with rcu_read_lock() 351 */ 352 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 353 u8 tos, int oif, struct net_device *dev, 354 int rpf, struct in_device *idev, u32 *itag) 355 { 356 struct net *net = dev_net(dev); 357 struct flow_keys flkeys; 358 int ret, no_addr; 359 struct fib_result res; 360 struct flowi4 fl4; 361 bool dev_match; 362 363 fl4.flowi4_oif = 0; 364 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev); 365 if (!fl4.flowi4_iif) 366 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX; 367 fl4.daddr = src; 368 fl4.saddr = dst; 369 fl4.flowi4_tos = tos; 370 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 371 fl4.flowi4_tun_key.tun_id = 0; 372 fl4.flowi4_flags = 0; 373 fl4.flowi4_uid = sock_net_uid(net, NULL); 374 375 no_addr = idev->ifa_list == NULL; 376 377 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0; 378 if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) { 379 fl4.flowi4_proto = 0; 380 fl4.fl4_sport = 0; 381 fl4.fl4_dport = 0; 382 } 383 384 if (fib_lookup(net, &fl4, &res, 0)) 385 goto last_resort; 386 if (res.type != RTN_UNICAST && 387 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev))) 388 goto e_inval; 389 fib_combine_itag(itag, &res); 390 391 dev_match = fib_info_nh_uses_dev(res.fi, dev); 392 if (dev_match) { 393 ret = FIB_RES_NH(res).nh_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_NH(res).nh_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 if (inet_lookup_ifaddr_rcu(net, src)) 441 return -EINVAL; 442 443 ok: 444 *itag = 0; 445 return 0; 446 } 447 448 full_check: 449 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag); 450 } 451 452 static inline __be32 sk_extract_addr(struct sockaddr *addr) 453 { 454 return ((struct sockaddr_in *) addr)->sin_addr.s_addr; 455 } 456 457 static int put_rtax(struct nlattr *mx, int len, int type, u32 value) 458 { 459 struct nlattr *nla; 460 461 nla = (struct nlattr *) ((char *) mx + len); 462 nla->nla_type = type; 463 nla->nla_len = nla_attr_size(4); 464 *(u32 *) nla_data(nla) = value; 465 466 return len + nla_total_size(4); 467 } 468 469 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt, 470 struct fib_config *cfg) 471 { 472 __be32 addr; 473 int plen; 474 475 memset(cfg, 0, sizeof(*cfg)); 476 cfg->fc_nlinfo.nl_net = net; 477 478 if (rt->rt_dst.sa_family != AF_INET) 479 return -EAFNOSUPPORT; 480 481 /* 482 * Check mask for validity: 483 * a) it must be contiguous. 484 * b) destination must have all host bits clear. 485 * c) if application forgot to set correct family (AF_INET), 486 * reject request unless it is absolutely clear i.e. 487 * both family and mask are zero. 488 */ 489 plen = 32; 490 addr = sk_extract_addr(&rt->rt_dst); 491 if (!(rt->rt_flags & RTF_HOST)) { 492 __be32 mask = sk_extract_addr(&rt->rt_genmask); 493 494 if (rt->rt_genmask.sa_family != AF_INET) { 495 if (mask || rt->rt_genmask.sa_family) 496 return -EAFNOSUPPORT; 497 } 498 499 if (bad_mask(mask, addr)) 500 return -EINVAL; 501 502 plen = inet_mask_len(mask); 503 } 504 505 cfg->fc_dst_len = plen; 506 cfg->fc_dst = addr; 507 508 if (cmd != SIOCDELRT) { 509 cfg->fc_nlflags = NLM_F_CREATE; 510 cfg->fc_protocol = RTPROT_BOOT; 511 } 512 513 if (rt->rt_metric) 514 cfg->fc_priority = rt->rt_metric - 1; 515 516 if (rt->rt_flags & RTF_REJECT) { 517 cfg->fc_scope = RT_SCOPE_HOST; 518 cfg->fc_type = RTN_UNREACHABLE; 519 return 0; 520 } 521 522 cfg->fc_scope = RT_SCOPE_NOWHERE; 523 cfg->fc_type = RTN_UNICAST; 524 525 if (rt->rt_dev) { 526 char *colon; 527 struct net_device *dev; 528 char devname[IFNAMSIZ]; 529 530 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1)) 531 return -EFAULT; 532 533 devname[IFNAMSIZ-1] = 0; 534 colon = strchr(devname, ':'); 535 if (colon) 536 *colon = 0; 537 dev = __dev_get_by_name(net, devname); 538 if (!dev) 539 return -ENODEV; 540 cfg->fc_oif = dev->ifindex; 541 cfg->fc_table = l3mdev_fib_table(dev); 542 if (colon) { 543 struct in_ifaddr *ifa; 544 struct in_device *in_dev = __in_dev_get_rtnl(dev); 545 if (!in_dev) 546 return -ENODEV; 547 *colon = ':'; 548 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) 549 if (strcmp(ifa->ifa_label, devname) == 0) 550 break; 551 if (!ifa) 552 return -ENODEV; 553 cfg->fc_prefsrc = ifa->ifa_local; 554 } 555 } 556 557 addr = sk_extract_addr(&rt->rt_gateway); 558 if (rt->rt_gateway.sa_family == AF_INET && addr) { 559 unsigned int addr_type; 560 561 cfg->fc_gw = addr; 562 addr_type = inet_addr_type_table(net, addr, cfg->fc_table); 563 if (rt->rt_flags & RTF_GATEWAY && 564 addr_type == RTN_UNICAST) 565 cfg->fc_scope = RT_SCOPE_UNIVERSE; 566 } 567 568 if (cmd == SIOCDELRT) 569 return 0; 570 571 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw) 572 return -EINVAL; 573 574 if (cfg->fc_scope == RT_SCOPE_NOWHERE) 575 cfg->fc_scope = RT_SCOPE_LINK; 576 577 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) { 578 struct nlattr *mx; 579 int len = 0; 580 581 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL); 582 if (!mx) 583 return -ENOMEM; 584 585 if (rt->rt_flags & RTF_MTU) 586 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40); 587 588 if (rt->rt_flags & RTF_WINDOW) 589 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window); 590 591 if (rt->rt_flags & RTF_IRTT) 592 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3); 593 594 cfg->fc_mx = mx; 595 cfg->fc_mx_len = len; 596 } 597 598 return 0; 599 } 600 601 /* 602 * Handle IP routing ioctl calls. 603 * These are used to manipulate the routing tables 604 */ 605 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt) 606 { 607 struct fib_config cfg; 608 int err; 609 610 switch (cmd) { 611 case SIOCADDRT: /* Add a route */ 612 case SIOCDELRT: /* Delete a route */ 613 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 614 return -EPERM; 615 616 rtnl_lock(); 617 err = rtentry_to_fib_config(net, cmd, rt, &cfg); 618 if (err == 0) { 619 struct fib_table *tb; 620 621 if (cmd == SIOCDELRT) { 622 tb = fib_get_table(net, cfg.fc_table); 623 if (tb) 624 err = fib_table_delete(net, tb, &cfg, 625 NULL); 626 else 627 err = -ESRCH; 628 } else { 629 tb = fib_new_table(net, cfg.fc_table); 630 if (tb) 631 err = fib_table_insert(net, tb, 632 &cfg, NULL); 633 else 634 err = -ENOBUFS; 635 } 636 637 /* allocated by rtentry_to_fib_config() */ 638 kfree(cfg.fc_mx); 639 } 640 rtnl_unlock(); 641 return err; 642 } 643 return -EINVAL; 644 } 645 646 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = { 647 [RTA_DST] = { .type = NLA_U32 }, 648 [RTA_SRC] = { .type = NLA_U32 }, 649 [RTA_IIF] = { .type = NLA_U32 }, 650 [RTA_OIF] = { .type = NLA_U32 }, 651 [RTA_GATEWAY] = { .type = NLA_U32 }, 652 [RTA_PRIORITY] = { .type = NLA_U32 }, 653 [RTA_PREFSRC] = { .type = NLA_U32 }, 654 [RTA_METRICS] = { .type = NLA_NESTED }, 655 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) }, 656 [RTA_FLOW] = { .type = NLA_U32 }, 657 [RTA_ENCAP_TYPE] = { .type = NLA_U16 }, 658 [RTA_ENCAP] = { .type = NLA_NESTED }, 659 [RTA_UID] = { .type = NLA_U32 }, 660 [RTA_MARK] = { .type = NLA_U32 }, 661 [RTA_TABLE] = { .type = NLA_U32 }, 662 [RTA_IP_PROTO] = { .type = NLA_U8 }, 663 [RTA_SPORT] = { .type = NLA_U16 }, 664 [RTA_DPORT] = { .type = NLA_U16 }, 665 }; 666 667 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb, 668 struct nlmsghdr *nlh, struct fib_config *cfg, 669 struct netlink_ext_ack *extack) 670 { 671 struct nlattr *attr; 672 int err, remaining; 673 struct rtmsg *rtm; 674 675 err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy, 676 extack); 677 if (err < 0) 678 goto errout; 679 680 memset(cfg, 0, sizeof(*cfg)); 681 682 rtm = nlmsg_data(nlh); 683 cfg->fc_dst_len = rtm->rtm_dst_len; 684 cfg->fc_tos = rtm->rtm_tos; 685 cfg->fc_table = rtm->rtm_table; 686 cfg->fc_protocol = rtm->rtm_protocol; 687 cfg->fc_scope = rtm->rtm_scope; 688 cfg->fc_type = rtm->rtm_type; 689 cfg->fc_flags = rtm->rtm_flags; 690 cfg->fc_nlflags = nlh->nlmsg_flags; 691 692 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid; 693 cfg->fc_nlinfo.nlh = nlh; 694 cfg->fc_nlinfo.nl_net = net; 695 696 if (cfg->fc_type > RTN_MAX) { 697 NL_SET_ERR_MSG(extack, "Invalid route type"); 698 err = -EINVAL; 699 goto errout; 700 } 701 702 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) { 703 switch (nla_type(attr)) { 704 case RTA_DST: 705 cfg->fc_dst = nla_get_be32(attr); 706 break; 707 case RTA_OIF: 708 cfg->fc_oif = nla_get_u32(attr); 709 break; 710 case RTA_GATEWAY: 711 cfg->fc_gw = nla_get_be32(attr); 712 break; 713 case RTA_PRIORITY: 714 cfg->fc_priority = nla_get_u32(attr); 715 break; 716 case RTA_PREFSRC: 717 cfg->fc_prefsrc = nla_get_be32(attr); 718 break; 719 case RTA_METRICS: 720 cfg->fc_mx = nla_data(attr); 721 cfg->fc_mx_len = nla_len(attr); 722 break; 723 case RTA_MULTIPATH: 724 err = lwtunnel_valid_encap_type_attr(nla_data(attr), 725 nla_len(attr), 726 extack); 727 if (err < 0) 728 goto errout; 729 cfg->fc_mp = nla_data(attr); 730 cfg->fc_mp_len = nla_len(attr); 731 break; 732 case RTA_FLOW: 733 cfg->fc_flow = nla_get_u32(attr); 734 break; 735 case RTA_TABLE: 736 cfg->fc_table = nla_get_u32(attr); 737 break; 738 case RTA_ENCAP: 739 cfg->fc_encap = attr; 740 break; 741 case RTA_ENCAP_TYPE: 742 cfg->fc_encap_type = nla_get_u16(attr); 743 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, 744 extack); 745 if (err < 0) 746 goto errout; 747 break; 748 } 749 } 750 751 return 0; 752 errout: 753 return err; 754 } 755 756 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, 757 struct netlink_ext_ack *extack) 758 { 759 struct net *net = sock_net(skb->sk); 760 struct fib_config cfg; 761 struct fib_table *tb; 762 int err; 763 764 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack); 765 if (err < 0) 766 goto errout; 767 768 tb = fib_get_table(net, cfg.fc_table); 769 if (!tb) { 770 NL_SET_ERR_MSG(extack, "FIB table does not exist"); 771 err = -ESRCH; 772 goto errout; 773 } 774 775 err = fib_table_delete(net, tb, &cfg, extack); 776 errout: 777 return err; 778 } 779 780 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, 781 struct netlink_ext_ack *extack) 782 { 783 struct net *net = sock_net(skb->sk); 784 struct fib_config cfg; 785 struct fib_table *tb; 786 int err; 787 788 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack); 789 if (err < 0) 790 goto errout; 791 792 tb = fib_new_table(net, cfg.fc_table); 793 if (!tb) { 794 err = -ENOBUFS; 795 goto errout; 796 } 797 798 err = fib_table_insert(net, tb, &cfg, extack); 799 if (!err && cfg.fc_type == RTN_LOCAL) 800 net->ipv4.fib_has_custom_local_routes = true; 801 errout: 802 return err; 803 } 804 805 int ip_valid_fib_dump_req(const struct nlmsghdr *nlh, 806 struct netlink_ext_ack *extack) 807 { 808 struct rtmsg *rtm; 809 810 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) { 811 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request"); 812 return -EINVAL; 813 } 814 815 rtm = nlmsg_data(nlh); 816 if (rtm->rtm_dst_len || rtm->rtm_src_len || rtm->rtm_tos || 817 rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope || 818 rtm->rtm_type) { 819 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request"); 820 return -EINVAL; 821 } 822 if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) { 823 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request"); 824 return -EINVAL; 825 } 826 827 if (nlmsg_attrlen(nlh, sizeof(*rtm))) { 828 NL_SET_ERR_MSG(extack, "Invalid data after header in FIB dump request"); 829 return -EINVAL; 830 } 831 832 return 0; 833 } 834 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req); 835 836 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 837 { 838 const struct nlmsghdr *nlh = cb->nlh; 839 struct net *net = sock_net(skb->sk); 840 unsigned int h, s_h; 841 unsigned int e = 0, s_e; 842 struct fib_table *tb; 843 struct hlist_head *head; 844 int dumped = 0, err; 845 846 if (cb->strict_check) { 847 err = ip_valid_fib_dump_req(nlh, cb->extack); 848 if (err < 0) 849 return err; 850 } 851 852 if (nlmsg_len(nlh) >= sizeof(struct rtmsg) && 853 ((struct rtmsg *)nlmsg_data(nlh))->rtm_flags & RTM_F_CLONED) 854 return skb->len; 855 856 s_h = cb->args[0]; 857 s_e = cb->args[1]; 858 859 rcu_read_lock(); 860 861 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) { 862 e = 0; 863 head = &net->ipv4.fib_table_hash[h]; 864 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 865 if (e < s_e) 866 goto next; 867 if (dumped) 868 memset(&cb->args[2], 0, sizeof(cb->args) - 869 2 * sizeof(cb->args[0])); 870 err = fib_table_dump(tb, skb, cb); 871 if (err < 0) { 872 if (likely(skb->len)) 873 goto out; 874 875 goto out_err; 876 } 877 dumped = 1; 878 next: 879 e++; 880 } 881 } 882 out: 883 err = skb->len; 884 out_err: 885 rcu_read_unlock(); 886 887 cb->args[1] = e; 888 cb->args[0] = h; 889 890 return err; 891 } 892 893 /* Prepare and feed intra-kernel routing request. 894 * Really, it should be netlink message, but :-( netlink 895 * can be not configured, so that we feed it directly 896 * to fib engine. It is legal, because all events occur 897 * only when netlink is already locked. 898 */ 899 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, 900 struct in_ifaddr *ifa, u32 rt_priority) 901 { 902 struct net *net = dev_net(ifa->ifa_dev->dev); 903 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev); 904 struct fib_table *tb; 905 struct fib_config cfg = { 906 .fc_protocol = RTPROT_KERNEL, 907 .fc_type = type, 908 .fc_dst = dst, 909 .fc_dst_len = dst_len, 910 .fc_priority = rt_priority, 911 .fc_prefsrc = ifa->ifa_local, 912 .fc_oif = ifa->ifa_dev->dev->ifindex, 913 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND, 914 .fc_nlinfo = { 915 .nl_net = net, 916 }, 917 }; 918 919 if (!tb_id) 920 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL; 921 922 tb = fib_new_table(net, tb_id); 923 if (!tb) 924 return; 925 926 cfg.fc_table = tb->tb_id; 927 928 if (type != RTN_LOCAL) 929 cfg.fc_scope = RT_SCOPE_LINK; 930 else 931 cfg.fc_scope = RT_SCOPE_HOST; 932 933 if (cmd == RTM_NEWROUTE) 934 fib_table_insert(net, tb, &cfg, NULL); 935 else 936 fib_table_delete(net, tb, &cfg, NULL); 937 } 938 939 void fib_add_ifaddr(struct in_ifaddr *ifa) 940 { 941 struct in_device *in_dev = ifa->ifa_dev; 942 struct net_device *dev = in_dev->dev; 943 struct in_ifaddr *prim = ifa; 944 __be32 mask = ifa->ifa_mask; 945 __be32 addr = ifa->ifa_local; 946 __be32 prefix = ifa->ifa_address & mask; 947 948 if (ifa->ifa_flags & IFA_F_SECONDARY) { 949 prim = inet_ifa_byprefix(in_dev, prefix, mask); 950 if (!prim) { 951 pr_warn("%s: bug: prim == NULL\n", __func__); 952 return; 953 } 954 } 955 956 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0); 957 958 if (!(dev->flags & IFF_UP)) 959 return; 960 961 /* Add broadcast address, if it is explicitly assigned. */ 962 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) 963 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, 964 prim, 0); 965 966 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) && 967 (prefix != addr || ifa->ifa_prefixlen < 32)) { 968 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 969 fib_magic(RTM_NEWROUTE, 970 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 971 prefix, ifa->ifa_prefixlen, prim, 972 ifa->ifa_rt_priority); 973 974 /* Add network specific broadcasts, when it takes a sense */ 975 if (ifa->ifa_prefixlen < 31) { 976 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, 977 prim, 0); 978 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask, 979 32, prim, 0); 980 } 981 } 982 } 983 984 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric) 985 { 986 __be32 prefix = ifa->ifa_address & ifa->ifa_mask; 987 struct in_device *in_dev = ifa->ifa_dev; 988 struct net_device *dev = in_dev->dev; 989 990 if (!(dev->flags & IFF_UP) || 991 ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) || 992 ipv4_is_zeronet(prefix) || 993 prefix == ifa->ifa_local || ifa->ifa_prefixlen == 32) 994 return; 995 996 /* add the new */ 997 fib_magic(RTM_NEWROUTE, 998 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 999 prefix, ifa->ifa_prefixlen, ifa, new_metric); 1000 1001 /* delete the old */ 1002 fib_magic(RTM_DELROUTE, 1003 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 1004 prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority); 1005 } 1006 1007 /* Delete primary or secondary address. 1008 * Optionally, on secondary address promotion consider the addresses 1009 * from subnet iprim as deleted, even if they are in device list. 1010 * In this case the secondary ifa can be in device list. 1011 */ 1012 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim) 1013 { 1014 struct in_device *in_dev = ifa->ifa_dev; 1015 struct net_device *dev = in_dev->dev; 1016 struct in_ifaddr *ifa1; 1017 struct in_ifaddr *prim = ifa, *prim1 = NULL; 1018 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask; 1019 __be32 any = ifa->ifa_address & ifa->ifa_mask; 1020 #define LOCAL_OK 1 1021 #define BRD_OK 2 1022 #define BRD0_OK 4 1023 #define BRD1_OK 8 1024 unsigned int ok = 0; 1025 int subnet = 0; /* Primary network */ 1026 int gone = 1; /* Address is missing */ 1027 int same_prefsrc = 0; /* Another primary with same IP */ 1028 1029 if (ifa->ifa_flags & IFA_F_SECONDARY) { 1030 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask); 1031 if (!prim) { 1032 /* if the device has been deleted, we don't perform 1033 * address promotion 1034 */ 1035 if (!in_dev->dead) 1036 pr_warn("%s: bug: prim == NULL\n", __func__); 1037 return; 1038 } 1039 if (iprim && iprim != prim) { 1040 pr_warn("%s: bug: iprim != prim\n", __func__); 1041 return; 1042 } 1043 } else if (!ipv4_is_zeronet(any) && 1044 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) { 1045 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 1046 fib_magic(RTM_DELROUTE, 1047 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 1048 any, ifa->ifa_prefixlen, prim, 0); 1049 subnet = 1; 1050 } 1051 1052 if (in_dev->dead) 1053 goto no_promotions; 1054 1055 /* Deletion is more complicated than add. 1056 * We should take care of not to delete too much :-) 1057 * 1058 * Scan address list to be sure that addresses are really gone. 1059 */ 1060 1061 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) { 1062 if (ifa1 == ifa) { 1063 /* promotion, keep the IP */ 1064 gone = 0; 1065 continue; 1066 } 1067 /* Ignore IFAs from our subnet */ 1068 if (iprim && ifa1->ifa_mask == iprim->ifa_mask && 1069 inet_ifa_match(ifa1->ifa_address, iprim)) 1070 continue; 1071 1072 /* Ignore ifa1 if it uses different primary IP (prefsrc) */ 1073 if (ifa1->ifa_flags & IFA_F_SECONDARY) { 1074 /* Another address from our subnet? */ 1075 if (ifa1->ifa_mask == prim->ifa_mask && 1076 inet_ifa_match(ifa1->ifa_address, prim)) 1077 prim1 = prim; 1078 else { 1079 /* We reached the secondaries, so 1080 * same_prefsrc should be determined. 1081 */ 1082 if (!same_prefsrc) 1083 continue; 1084 /* Search new prim1 if ifa1 is not 1085 * using the current prim1 1086 */ 1087 if (!prim1 || 1088 ifa1->ifa_mask != prim1->ifa_mask || 1089 !inet_ifa_match(ifa1->ifa_address, prim1)) 1090 prim1 = inet_ifa_byprefix(in_dev, 1091 ifa1->ifa_address, 1092 ifa1->ifa_mask); 1093 if (!prim1) 1094 continue; 1095 if (prim1->ifa_local != prim->ifa_local) 1096 continue; 1097 } 1098 } else { 1099 if (prim->ifa_local != ifa1->ifa_local) 1100 continue; 1101 prim1 = ifa1; 1102 if (prim != prim1) 1103 same_prefsrc = 1; 1104 } 1105 if (ifa->ifa_local == ifa1->ifa_local) 1106 ok |= LOCAL_OK; 1107 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 1108 ok |= BRD_OK; 1109 if (brd == ifa1->ifa_broadcast) 1110 ok |= BRD1_OK; 1111 if (any == ifa1->ifa_broadcast) 1112 ok |= BRD0_OK; 1113 /* primary has network specific broadcasts */ 1114 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) { 1115 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask; 1116 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask; 1117 1118 if (!ipv4_is_zeronet(any1)) { 1119 if (ifa->ifa_broadcast == brd1 || 1120 ifa->ifa_broadcast == any1) 1121 ok |= BRD_OK; 1122 if (brd == brd1 || brd == any1) 1123 ok |= BRD1_OK; 1124 if (any == brd1 || any == any1) 1125 ok |= BRD0_OK; 1126 } 1127 } 1128 } 1129 1130 no_promotions: 1131 if (!(ok & BRD_OK)) 1132 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, 1133 prim, 0); 1134 if (subnet && ifa->ifa_prefixlen < 31) { 1135 if (!(ok & BRD1_OK)) 1136 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, 1137 prim, 0); 1138 if (!(ok & BRD0_OK)) 1139 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, 1140 prim, 0); 1141 } 1142 if (!(ok & LOCAL_OK)) { 1143 unsigned int addr_type; 1144 1145 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0); 1146 1147 /* Check, that this local address finally disappeared. */ 1148 addr_type = inet_addr_type_dev_table(dev_net(dev), dev, 1149 ifa->ifa_local); 1150 if (gone && addr_type != RTN_LOCAL) { 1151 /* And the last, but not the least thing. 1152 * We must flush stray FIB entries. 1153 * 1154 * First of all, we scan fib_info list searching 1155 * for stray nexthop entries, then ignite fib_flush. 1156 */ 1157 if (fib_sync_down_addr(dev, ifa->ifa_local)) 1158 fib_flush(dev_net(dev)); 1159 } 1160 } 1161 #undef LOCAL_OK 1162 #undef BRD_OK 1163 #undef BRD0_OK 1164 #undef BRD1_OK 1165 } 1166 1167 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn) 1168 { 1169 1170 struct fib_result res; 1171 struct flowi4 fl4 = { 1172 .flowi4_mark = frn->fl_mark, 1173 .daddr = frn->fl_addr, 1174 .flowi4_tos = frn->fl_tos, 1175 .flowi4_scope = frn->fl_scope, 1176 }; 1177 struct fib_table *tb; 1178 1179 rcu_read_lock(); 1180 1181 tb = fib_get_table(net, frn->tb_id_in); 1182 1183 frn->err = -ENOENT; 1184 if (tb) { 1185 local_bh_disable(); 1186 1187 frn->tb_id = tb->tb_id; 1188 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 1189 1190 if (!frn->err) { 1191 frn->prefixlen = res.prefixlen; 1192 frn->nh_sel = res.nh_sel; 1193 frn->type = res.type; 1194 frn->scope = res.scope; 1195 } 1196 local_bh_enable(); 1197 } 1198 1199 rcu_read_unlock(); 1200 } 1201 1202 static void nl_fib_input(struct sk_buff *skb) 1203 { 1204 struct net *net; 1205 struct fib_result_nl *frn; 1206 struct nlmsghdr *nlh; 1207 u32 portid; 1208 1209 net = sock_net(skb->sk); 1210 nlh = nlmsg_hdr(skb); 1211 if (skb->len < nlmsg_total_size(sizeof(*frn)) || 1212 skb->len < nlh->nlmsg_len || 1213 nlmsg_len(nlh) < sizeof(*frn)) 1214 return; 1215 1216 skb = netlink_skb_clone(skb, GFP_KERNEL); 1217 if (!skb) 1218 return; 1219 nlh = nlmsg_hdr(skb); 1220 1221 frn = (struct fib_result_nl *) nlmsg_data(nlh); 1222 nl_fib_lookup(net, frn); 1223 1224 portid = NETLINK_CB(skb).portid; /* netlink portid */ 1225 NETLINK_CB(skb).portid = 0; /* from kernel */ 1226 NETLINK_CB(skb).dst_group = 0; /* unicast */ 1227 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT); 1228 } 1229 1230 static int __net_init nl_fib_lookup_init(struct net *net) 1231 { 1232 struct sock *sk; 1233 struct netlink_kernel_cfg cfg = { 1234 .input = nl_fib_input, 1235 }; 1236 1237 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg); 1238 if (!sk) 1239 return -EAFNOSUPPORT; 1240 net->ipv4.fibnl = sk; 1241 return 0; 1242 } 1243 1244 static void nl_fib_lookup_exit(struct net *net) 1245 { 1246 netlink_kernel_release(net->ipv4.fibnl); 1247 net->ipv4.fibnl = NULL; 1248 } 1249 1250 static void fib_disable_ip(struct net_device *dev, unsigned long event, 1251 bool force) 1252 { 1253 if (fib_sync_down_dev(dev, event, force)) 1254 fib_flush(dev_net(dev)); 1255 else 1256 rt_cache_flush(dev_net(dev)); 1257 arp_ifdown(dev); 1258 } 1259 1260 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 1261 { 1262 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; 1263 struct net_device *dev = ifa->ifa_dev->dev; 1264 struct net *net = dev_net(dev); 1265 1266 switch (event) { 1267 case NETDEV_UP: 1268 fib_add_ifaddr(ifa); 1269 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1270 fib_sync_up(dev, RTNH_F_DEAD); 1271 #endif 1272 atomic_inc(&net->ipv4.dev_addr_genid); 1273 rt_cache_flush(dev_net(dev)); 1274 break; 1275 case NETDEV_DOWN: 1276 fib_del_ifaddr(ifa, NULL); 1277 atomic_inc(&net->ipv4.dev_addr_genid); 1278 if (!ifa->ifa_dev->ifa_list) { 1279 /* Last address was deleted from this interface. 1280 * Disable IP. 1281 */ 1282 fib_disable_ip(dev, event, true); 1283 } else { 1284 rt_cache_flush(dev_net(dev)); 1285 } 1286 break; 1287 } 1288 return NOTIFY_DONE; 1289 } 1290 1291 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 1292 { 1293 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1294 struct netdev_notifier_changeupper_info *info; 1295 struct in_device *in_dev; 1296 struct net *net = dev_net(dev); 1297 unsigned int flags; 1298 1299 if (event == NETDEV_UNREGISTER) { 1300 fib_disable_ip(dev, event, true); 1301 rt_flush_dev(dev); 1302 return NOTIFY_DONE; 1303 } 1304 1305 in_dev = __in_dev_get_rtnl(dev); 1306 if (!in_dev) 1307 return NOTIFY_DONE; 1308 1309 switch (event) { 1310 case NETDEV_UP: 1311 for_ifa(in_dev) { 1312 fib_add_ifaddr(ifa); 1313 } endfor_ifa(in_dev); 1314 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1315 fib_sync_up(dev, RTNH_F_DEAD); 1316 #endif 1317 atomic_inc(&net->ipv4.dev_addr_genid); 1318 rt_cache_flush(net); 1319 break; 1320 case NETDEV_DOWN: 1321 fib_disable_ip(dev, event, false); 1322 break; 1323 case NETDEV_CHANGE: 1324 flags = dev_get_flags(dev); 1325 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 1326 fib_sync_up(dev, RTNH_F_LINKDOWN); 1327 else 1328 fib_sync_down_dev(dev, event, false); 1329 /* fall through */ 1330 case NETDEV_CHANGEMTU: 1331 rt_cache_flush(net); 1332 break; 1333 case NETDEV_CHANGEUPPER: 1334 info = ptr; 1335 /* flush all routes if dev is linked to or unlinked from 1336 * an L3 master device (e.g., VRF) 1337 */ 1338 if (info->upper_dev && netif_is_l3_master(info->upper_dev)) 1339 fib_disable_ip(dev, NETDEV_DOWN, true); 1340 break; 1341 } 1342 return NOTIFY_DONE; 1343 } 1344 1345 static struct notifier_block fib_inetaddr_notifier = { 1346 .notifier_call = fib_inetaddr_event, 1347 }; 1348 1349 static struct notifier_block fib_netdev_notifier = { 1350 .notifier_call = fib_netdev_event, 1351 }; 1352 1353 static int __net_init ip_fib_net_init(struct net *net) 1354 { 1355 int err; 1356 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ; 1357 1358 err = fib4_notifier_init(net); 1359 if (err) 1360 return err; 1361 1362 /* Avoid false sharing : Use at least a full cache line */ 1363 size = max_t(size_t, size, L1_CACHE_BYTES); 1364 1365 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL); 1366 if (!net->ipv4.fib_table_hash) { 1367 err = -ENOMEM; 1368 goto err_table_hash_alloc; 1369 } 1370 1371 err = fib4_rules_init(net); 1372 if (err < 0) 1373 goto err_rules_init; 1374 return 0; 1375 1376 err_rules_init: 1377 kfree(net->ipv4.fib_table_hash); 1378 err_table_hash_alloc: 1379 fib4_notifier_exit(net); 1380 return err; 1381 } 1382 1383 static void ip_fib_net_exit(struct net *net) 1384 { 1385 int i; 1386 1387 rtnl_lock(); 1388 #ifdef CONFIG_IP_MULTIPLE_TABLES 1389 RCU_INIT_POINTER(net->ipv4.fib_main, NULL); 1390 RCU_INIT_POINTER(net->ipv4.fib_default, NULL); 1391 #endif 1392 /* Destroy the tables in reverse order to guarantee that the 1393 * local table, ID 255, is destroyed before the main table, ID 1394 * 254. This is necessary as the local table may contain 1395 * references to data contained in the main table. 1396 */ 1397 for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) { 1398 struct hlist_head *head = &net->ipv4.fib_table_hash[i]; 1399 struct hlist_node *tmp; 1400 struct fib_table *tb; 1401 1402 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) { 1403 hlist_del(&tb->tb_hlist); 1404 fib_table_flush(net, tb); 1405 fib_free_table(tb); 1406 } 1407 } 1408 1409 #ifdef CONFIG_IP_MULTIPLE_TABLES 1410 fib4_rules_exit(net); 1411 #endif 1412 rtnl_unlock(); 1413 kfree(net->ipv4.fib_table_hash); 1414 fib4_notifier_exit(net); 1415 } 1416 1417 static int __net_init fib_net_init(struct net *net) 1418 { 1419 int error; 1420 1421 #ifdef CONFIG_IP_ROUTE_CLASSID 1422 net->ipv4.fib_num_tclassid_users = 0; 1423 #endif 1424 error = ip_fib_net_init(net); 1425 if (error < 0) 1426 goto out; 1427 error = nl_fib_lookup_init(net); 1428 if (error < 0) 1429 goto out_nlfl; 1430 error = fib_proc_init(net); 1431 if (error < 0) 1432 goto out_proc; 1433 out: 1434 return error; 1435 1436 out_proc: 1437 nl_fib_lookup_exit(net); 1438 out_nlfl: 1439 ip_fib_net_exit(net); 1440 goto out; 1441 } 1442 1443 static void __net_exit fib_net_exit(struct net *net) 1444 { 1445 fib_proc_exit(net); 1446 nl_fib_lookup_exit(net); 1447 ip_fib_net_exit(net); 1448 } 1449 1450 static struct pernet_operations fib_net_ops = { 1451 .init = fib_net_init, 1452 .exit = fib_net_exit, 1453 }; 1454 1455 void __init ip_fib_init(void) 1456 { 1457 fib_trie_init(); 1458 1459 register_pernet_subsys(&fib_net_ops); 1460 1461 register_netdevice_notifier(&fib_netdev_notifier); 1462 register_inetaddr_notifier(&fib_inetaddr_notifier); 1463 1464 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0); 1465 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0); 1466 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0); 1467 } 1468