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 #else 77 78 struct fib_table *fib_new_table(struct net *net, u32 id) 79 { 80 struct fib_table *tb, *alias = NULL; 81 unsigned int h; 82 83 if (id == 0) 84 id = RT_TABLE_MAIN; 85 tb = fib_get_table(net, id); 86 if (tb) 87 return tb; 88 89 if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules) 90 alias = fib_new_table(net, RT_TABLE_MAIN); 91 92 tb = fib_trie_table(id, alias); 93 if (!tb) 94 return NULL; 95 96 switch (id) { 97 case RT_TABLE_MAIN: 98 rcu_assign_pointer(net->ipv4.fib_main, tb); 99 break; 100 case RT_TABLE_DEFAULT: 101 rcu_assign_pointer(net->ipv4.fib_default, tb); 102 break; 103 default: 104 break; 105 } 106 107 h = id & (FIB_TABLE_HASHSZ - 1); 108 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]); 109 return tb; 110 } 111 EXPORT_SYMBOL_GPL(fib_new_table); 112 113 /* caller must hold either rtnl or rcu read lock */ 114 struct fib_table *fib_get_table(struct net *net, u32 id) 115 { 116 struct fib_table *tb; 117 struct hlist_head *head; 118 unsigned int h; 119 120 if (id == 0) 121 id = RT_TABLE_MAIN; 122 h = id & (FIB_TABLE_HASHSZ - 1); 123 124 head = &net->ipv4.fib_table_hash[h]; 125 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 126 if (tb->tb_id == id) 127 return tb; 128 } 129 return NULL; 130 } 131 #endif /* CONFIG_IP_MULTIPLE_TABLES */ 132 133 static void fib_replace_table(struct net *net, struct fib_table *old, 134 struct fib_table *new) 135 { 136 #ifdef CONFIG_IP_MULTIPLE_TABLES 137 switch (new->tb_id) { 138 case RT_TABLE_MAIN: 139 rcu_assign_pointer(net->ipv4.fib_main, new); 140 break; 141 case RT_TABLE_DEFAULT: 142 rcu_assign_pointer(net->ipv4.fib_default, new); 143 break; 144 default: 145 break; 146 } 147 148 #endif 149 /* replace the old table in the hlist */ 150 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist); 151 } 152 153 int fib_unmerge(struct net *net) 154 { 155 struct fib_table *old, *new, *main_table; 156 157 /* attempt to fetch local table if it has been allocated */ 158 old = fib_get_table(net, RT_TABLE_LOCAL); 159 if (!old) 160 return 0; 161 162 new = fib_trie_unmerge(old); 163 if (!new) 164 return -ENOMEM; 165 166 /* table is already unmerged */ 167 if (new == old) 168 return 0; 169 170 /* replace merged table with clean table */ 171 fib_replace_table(net, old, new); 172 fib_free_table(old); 173 174 /* attempt to fetch main table if it has been allocated */ 175 main_table = fib_get_table(net, RT_TABLE_MAIN); 176 if (!main_table) 177 return 0; 178 179 /* flush local entries from main table */ 180 fib_table_flush_external(main_table); 181 182 return 0; 183 } 184 185 static void fib_flush(struct net *net) 186 { 187 int flushed = 0; 188 unsigned int h; 189 190 for (h = 0; h < FIB_TABLE_HASHSZ; h++) { 191 struct hlist_head *head = &net->ipv4.fib_table_hash[h]; 192 struct hlist_node *tmp; 193 struct fib_table *tb; 194 195 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) 196 flushed += fib_table_flush(net, tb); 197 } 198 199 if (flushed) 200 rt_cache_flush(net); 201 } 202 203 /* 204 * Find address type as if only "dev" was present in the system. If 205 * on_dev is NULL then all interfaces are taken into consideration. 206 */ 207 static inline unsigned int __inet_dev_addr_type(struct net *net, 208 const struct net_device *dev, 209 __be32 addr, u32 tb_id) 210 { 211 struct flowi4 fl4 = { .daddr = addr }; 212 struct fib_result res; 213 unsigned int ret = RTN_BROADCAST; 214 struct fib_table *table; 215 216 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr)) 217 return RTN_BROADCAST; 218 if (ipv4_is_multicast(addr)) 219 return RTN_MULTICAST; 220 221 rcu_read_lock(); 222 223 table = fib_get_table(net, tb_id); 224 if (table) { 225 ret = RTN_UNICAST; 226 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) { 227 if (!dev || dev == res.fi->fib_dev) 228 ret = res.type; 229 } 230 } 231 232 rcu_read_unlock(); 233 return ret; 234 } 235 236 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id) 237 { 238 return __inet_dev_addr_type(net, NULL, addr, tb_id); 239 } 240 EXPORT_SYMBOL(inet_addr_type_table); 241 242 unsigned int inet_addr_type(struct net *net, __be32 addr) 243 { 244 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL); 245 } 246 EXPORT_SYMBOL(inet_addr_type); 247 248 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev, 249 __be32 addr) 250 { 251 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL; 252 253 return __inet_dev_addr_type(net, dev, addr, rt_table); 254 } 255 EXPORT_SYMBOL(inet_dev_addr_type); 256 257 /* inet_addr_type with dev == NULL but using the table from a dev 258 * if one is associated 259 */ 260 unsigned int inet_addr_type_dev_table(struct net *net, 261 const struct net_device *dev, 262 __be32 addr) 263 { 264 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL; 265 266 return __inet_dev_addr_type(net, NULL, addr, rt_table); 267 } 268 EXPORT_SYMBOL(inet_addr_type_dev_table); 269 270 __be32 fib_compute_spec_dst(struct sk_buff *skb) 271 { 272 struct net_device *dev = skb->dev; 273 struct in_device *in_dev; 274 struct fib_result res; 275 struct rtable *rt; 276 struct net *net; 277 int scope; 278 279 rt = skb_rtable(skb); 280 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) == 281 RTCF_LOCAL) 282 return ip_hdr(skb)->daddr; 283 284 in_dev = __in_dev_get_rcu(dev); 285 BUG_ON(!in_dev); 286 287 net = dev_net(dev); 288 289 scope = RT_SCOPE_UNIVERSE; 290 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) { 291 struct flowi4 fl4 = { 292 .flowi4_iif = LOOPBACK_IFINDEX, 293 .daddr = ip_hdr(skb)->saddr, 294 .flowi4_tos = RT_TOS(ip_hdr(skb)->tos), 295 .flowi4_scope = scope, 296 .flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0, 297 }; 298 if (!fib_lookup(net, &fl4, &res, 0)) 299 return FIB_RES_PREFSRC(net, res); 300 } else { 301 scope = RT_SCOPE_LINK; 302 } 303 304 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope); 305 } 306 307 /* Given (packet source, input interface) and optional (dst, oif, tos): 308 * - (main) check, that source is valid i.e. not broadcast or our local 309 * address. 310 * - figure out what "logical" interface this packet arrived 311 * and calculate "specific destination" address. 312 * - check, that packet arrived from expected physical interface. 313 * called with rcu_read_lock() 314 */ 315 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 316 u8 tos, int oif, struct net_device *dev, 317 int rpf, struct in_device *idev, u32 *itag) 318 { 319 int ret, no_addr; 320 struct fib_result res; 321 struct flowi4 fl4; 322 struct net *net = dev_net(dev); 323 bool dev_match; 324 325 fl4.flowi4_oif = 0; 326 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev); 327 if (!fl4.flowi4_iif) 328 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX; 329 fl4.daddr = src; 330 fl4.saddr = dst; 331 fl4.flowi4_tos = tos; 332 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 333 fl4.flowi4_tun_key.tun_id = 0; 334 fl4.flowi4_flags = 0; 335 fl4.flowi4_uid = sock_net_uid(net, NULL); 336 337 no_addr = idev->ifa_list == NULL; 338 339 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0; 340 341 trace_fib_validate_source(dev, &fl4); 342 343 if (fib_lookup(net, &fl4, &res, 0)) 344 goto last_resort; 345 if (res.type != RTN_UNICAST && 346 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev))) 347 goto e_inval; 348 if (!rpf && !fib_num_tclassid_users(net) && 349 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) 350 goto last_resort; 351 fib_combine_itag(itag, &res); 352 dev_match = false; 353 354 #ifdef CONFIG_IP_ROUTE_MULTIPATH 355 for (ret = 0; ret < res.fi->fib_nhs; ret++) { 356 struct fib_nh *nh = &res.fi->fib_nh[ret]; 357 358 if (nh->nh_dev == dev) { 359 dev_match = true; 360 break; 361 } else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) { 362 dev_match = true; 363 break; 364 } 365 } 366 #else 367 if (FIB_RES_DEV(res) == dev) 368 dev_match = true; 369 #endif 370 if (dev_match) { 371 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 372 return ret; 373 } 374 if (no_addr) 375 goto last_resort; 376 if (rpf == 1) 377 goto e_rpf; 378 fl4.flowi4_oif = dev->ifindex; 379 380 ret = 0; 381 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) { 382 if (res.type == RTN_UNICAST) 383 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 384 } 385 return ret; 386 387 last_resort: 388 if (rpf) 389 goto e_rpf; 390 *itag = 0; 391 return 0; 392 393 e_inval: 394 return -EINVAL; 395 e_rpf: 396 return -EXDEV; 397 } 398 399 /* Ignore rp_filter for packets protected by IPsec. */ 400 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 401 u8 tos, int oif, struct net_device *dev, 402 struct in_device *idev, u32 *itag) 403 { 404 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev); 405 406 if (!r && !fib_num_tclassid_users(dev_net(dev)) && 407 IN_DEV_ACCEPT_LOCAL(idev) && 408 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) { 409 *itag = 0; 410 return 0; 411 } 412 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag); 413 } 414 415 static inline __be32 sk_extract_addr(struct sockaddr *addr) 416 { 417 return ((struct sockaddr_in *) addr)->sin_addr.s_addr; 418 } 419 420 static int put_rtax(struct nlattr *mx, int len, int type, u32 value) 421 { 422 struct nlattr *nla; 423 424 nla = (struct nlattr *) ((char *) mx + len); 425 nla->nla_type = type; 426 nla->nla_len = nla_attr_size(4); 427 *(u32 *) nla_data(nla) = value; 428 429 return len + nla_total_size(4); 430 } 431 432 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt, 433 struct fib_config *cfg) 434 { 435 __be32 addr; 436 int plen; 437 438 memset(cfg, 0, sizeof(*cfg)); 439 cfg->fc_nlinfo.nl_net = net; 440 441 if (rt->rt_dst.sa_family != AF_INET) 442 return -EAFNOSUPPORT; 443 444 /* 445 * Check mask for validity: 446 * a) it must be contiguous. 447 * b) destination must have all host bits clear. 448 * c) if application forgot to set correct family (AF_INET), 449 * reject request unless it is absolutely clear i.e. 450 * both family and mask are zero. 451 */ 452 plen = 32; 453 addr = sk_extract_addr(&rt->rt_dst); 454 if (!(rt->rt_flags & RTF_HOST)) { 455 __be32 mask = sk_extract_addr(&rt->rt_genmask); 456 457 if (rt->rt_genmask.sa_family != AF_INET) { 458 if (mask || rt->rt_genmask.sa_family) 459 return -EAFNOSUPPORT; 460 } 461 462 if (bad_mask(mask, addr)) 463 return -EINVAL; 464 465 plen = inet_mask_len(mask); 466 } 467 468 cfg->fc_dst_len = plen; 469 cfg->fc_dst = addr; 470 471 if (cmd != SIOCDELRT) { 472 cfg->fc_nlflags = NLM_F_CREATE; 473 cfg->fc_protocol = RTPROT_BOOT; 474 } 475 476 if (rt->rt_metric) 477 cfg->fc_priority = rt->rt_metric - 1; 478 479 if (rt->rt_flags & RTF_REJECT) { 480 cfg->fc_scope = RT_SCOPE_HOST; 481 cfg->fc_type = RTN_UNREACHABLE; 482 return 0; 483 } 484 485 cfg->fc_scope = RT_SCOPE_NOWHERE; 486 cfg->fc_type = RTN_UNICAST; 487 488 if (rt->rt_dev) { 489 char *colon; 490 struct net_device *dev; 491 char devname[IFNAMSIZ]; 492 493 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1)) 494 return -EFAULT; 495 496 devname[IFNAMSIZ-1] = 0; 497 colon = strchr(devname, ':'); 498 if (colon) 499 *colon = 0; 500 dev = __dev_get_by_name(net, devname); 501 if (!dev) 502 return -ENODEV; 503 cfg->fc_oif = dev->ifindex; 504 cfg->fc_table = l3mdev_fib_table(dev); 505 if (colon) { 506 struct in_ifaddr *ifa; 507 struct in_device *in_dev = __in_dev_get_rtnl(dev); 508 if (!in_dev) 509 return -ENODEV; 510 *colon = ':'; 511 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) 512 if (strcmp(ifa->ifa_label, devname) == 0) 513 break; 514 if (!ifa) 515 return -ENODEV; 516 cfg->fc_prefsrc = ifa->ifa_local; 517 } 518 } 519 520 addr = sk_extract_addr(&rt->rt_gateway); 521 if (rt->rt_gateway.sa_family == AF_INET && addr) { 522 unsigned int addr_type; 523 524 cfg->fc_gw = addr; 525 addr_type = inet_addr_type_table(net, addr, cfg->fc_table); 526 if (rt->rt_flags & RTF_GATEWAY && 527 addr_type == RTN_UNICAST) 528 cfg->fc_scope = RT_SCOPE_UNIVERSE; 529 } 530 531 if (cmd == SIOCDELRT) 532 return 0; 533 534 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw) 535 return -EINVAL; 536 537 if (cfg->fc_scope == RT_SCOPE_NOWHERE) 538 cfg->fc_scope = RT_SCOPE_LINK; 539 540 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) { 541 struct nlattr *mx; 542 int len = 0; 543 544 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL); 545 if (!mx) 546 return -ENOMEM; 547 548 if (rt->rt_flags & RTF_MTU) 549 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40); 550 551 if (rt->rt_flags & RTF_WINDOW) 552 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window); 553 554 if (rt->rt_flags & RTF_IRTT) 555 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3); 556 557 cfg->fc_mx = mx; 558 cfg->fc_mx_len = len; 559 } 560 561 return 0; 562 } 563 564 /* 565 * Handle IP routing ioctl calls. 566 * These are used to manipulate the routing tables 567 */ 568 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg) 569 { 570 struct fib_config cfg; 571 struct rtentry rt; 572 int err; 573 574 switch (cmd) { 575 case SIOCADDRT: /* Add a route */ 576 case SIOCDELRT: /* Delete a route */ 577 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 578 return -EPERM; 579 580 if (copy_from_user(&rt, arg, sizeof(rt))) 581 return -EFAULT; 582 583 rtnl_lock(); 584 err = rtentry_to_fib_config(net, cmd, &rt, &cfg); 585 if (err == 0) { 586 struct fib_table *tb; 587 588 if (cmd == SIOCDELRT) { 589 tb = fib_get_table(net, cfg.fc_table); 590 if (tb) 591 err = fib_table_delete(net, tb, &cfg); 592 else 593 err = -ESRCH; 594 } else { 595 tb = fib_new_table(net, cfg.fc_table); 596 if (tb) 597 err = fib_table_insert(net, tb, &cfg); 598 else 599 err = -ENOBUFS; 600 } 601 602 /* allocated by rtentry_to_fib_config() */ 603 kfree(cfg.fc_mx); 604 } 605 rtnl_unlock(); 606 return err; 607 } 608 return -EINVAL; 609 } 610 611 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = { 612 [RTA_DST] = { .type = NLA_U32 }, 613 [RTA_SRC] = { .type = NLA_U32 }, 614 [RTA_IIF] = { .type = NLA_U32 }, 615 [RTA_OIF] = { .type = NLA_U32 }, 616 [RTA_GATEWAY] = { .type = NLA_U32 }, 617 [RTA_PRIORITY] = { .type = NLA_U32 }, 618 [RTA_PREFSRC] = { .type = NLA_U32 }, 619 [RTA_METRICS] = { .type = NLA_NESTED }, 620 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) }, 621 [RTA_FLOW] = { .type = NLA_U32 }, 622 [RTA_ENCAP_TYPE] = { .type = NLA_U16 }, 623 [RTA_ENCAP] = { .type = NLA_NESTED }, 624 [RTA_UID] = { .type = NLA_U32 }, 625 [RTA_MARK] = { .type = NLA_U32 }, 626 }; 627 628 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb, 629 struct nlmsghdr *nlh, struct fib_config *cfg) 630 { 631 struct nlattr *attr; 632 int err, remaining; 633 struct rtmsg *rtm; 634 635 err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy, 636 NULL); 637 if (err < 0) 638 goto errout; 639 640 memset(cfg, 0, sizeof(*cfg)); 641 642 rtm = nlmsg_data(nlh); 643 cfg->fc_dst_len = rtm->rtm_dst_len; 644 cfg->fc_tos = rtm->rtm_tos; 645 cfg->fc_table = rtm->rtm_table; 646 cfg->fc_protocol = rtm->rtm_protocol; 647 cfg->fc_scope = rtm->rtm_scope; 648 cfg->fc_type = rtm->rtm_type; 649 cfg->fc_flags = rtm->rtm_flags; 650 cfg->fc_nlflags = nlh->nlmsg_flags; 651 652 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid; 653 cfg->fc_nlinfo.nlh = nlh; 654 cfg->fc_nlinfo.nl_net = net; 655 656 if (cfg->fc_type > RTN_MAX) { 657 err = -EINVAL; 658 goto errout; 659 } 660 661 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) { 662 switch (nla_type(attr)) { 663 case RTA_DST: 664 cfg->fc_dst = nla_get_be32(attr); 665 break; 666 case RTA_OIF: 667 cfg->fc_oif = nla_get_u32(attr); 668 break; 669 case RTA_GATEWAY: 670 cfg->fc_gw = nla_get_be32(attr); 671 break; 672 case RTA_PRIORITY: 673 cfg->fc_priority = nla_get_u32(attr); 674 break; 675 case RTA_PREFSRC: 676 cfg->fc_prefsrc = nla_get_be32(attr); 677 break; 678 case RTA_METRICS: 679 cfg->fc_mx = nla_data(attr); 680 cfg->fc_mx_len = nla_len(attr); 681 break; 682 case RTA_MULTIPATH: 683 err = lwtunnel_valid_encap_type_attr(nla_data(attr), 684 nla_len(attr)); 685 if (err < 0) 686 goto errout; 687 cfg->fc_mp = nla_data(attr); 688 cfg->fc_mp_len = nla_len(attr); 689 break; 690 case RTA_FLOW: 691 cfg->fc_flow = nla_get_u32(attr); 692 break; 693 case RTA_TABLE: 694 cfg->fc_table = nla_get_u32(attr); 695 break; 696 case RTA_ENCAP: 697 cfg->fc_encap = attr; 698 break; 699 case RTA_ENCAP_TYPE: 700 cfg->fc_encap_type = nla_get_u16(attr); 701 err = lwtunnel_valid_encap_type(cfg->fc_encap_type); 702 if (err < 0) 703 goto errout; 704 break; 705 } 706 } 707 708 return 0; 709 errout: 710 return err; 711 } 712 713 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, 714 struct netlink_ext_ack *extack) 715 { 716 struct net *net = sock_net(skb->sk); 717 struct fib_config cfg; 718 struct fib_table *tb; 719 int err; 720 721 err = rtm_to_fib_config(net, skb, nlh, &cfg); 722 if (err < 0) 723 goto errout; 724 725 tb = fib_get_table(net, cfg.fc_table); 726 if (!tb) { 727 err = -ESRCH; 728 goto errout; 729 } 730 731 err = fib_table_delete(net, tb, &cfg); 732 errout: 733 return err; 734 } 735 736 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, 737 struct netlink_ext_ack *extack) 738 { 739 struct net *net = sock_net(skb->sk); 740 struct fib_config cfg; 741 struct fib_table *tb; 742 int err; 743 744 err = rtm_to_fib_config(net, skb, nlh, &cfg); 745 if (err < 0) 746 goto errout; 747 748 tb = fib_new_table(net, cfg.fc_table); 749 if (!tb) { 750 err = -ENOBUFS; 751 goto errout; 752 } 753 754 err = fib_table_insert(net, tb, &cfg); 755 errout: 756 return err; 757 } 758 759 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 760 { 761 struct net *net = sock_net(skb->sk); 762 unsigned int h, s_h; 763 unsigned int e = 0, s_e; 764 struct fib_table *tb; 765 struct hlist_head *head; 766 int dumped = 0; 767 768 if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) && 769 ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED) 770 return skb->len; 771 772 s_h = cb->args[0]; 773 s_e = cb->args[1]; 774 775 rcu_read_lock(); 776 777 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) { 778 e = 0; 779 head = &net->ipv4.fib_table_hash[h]; 780 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 781 if (e < s_e) 782 goto next; 783 if (dumped) 784 memset(&cb->args[2], 0, sizeof(cb->args) - 785 2 * sizeof(cb->args[0])); 786 if (fib_table_dump(tb, skb, cb) < 0) 787 goto out; 788 dumped = 1; 789 next: 790 e++; 791 } 792 } 793 out: 794 rcu_read_unlock(); 795 796 cb->args[1] = e; 797 cb->args[0] = h; 798 799 return skb->len; 800 } 801 802 /* Prepare and feed intra-kernel routing request. 803 * Really, it should be netlink message, but :-( netlink 804 * can be not configured, so that we feed it directly 805 * to fib engine. It is legal, because all events occur 806 * only when netlink is already locked. 807 */ 808 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa) 809 { 810 struct net *net = dev_net(ifa->ifa_dev->dev); 811 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev); 812 struct fib_table *tb; 813 struct fib_config cfg = { 814 .fc_protocol = RTPROT_KERNEL, 815 .fc_type = type, 816 .fc_dst = dst, 817 .fc_dst_len = dst_len, 818 .fc_prefsrc = ifa->ifa_local, 819 .fc_oif = ifa->ifa_dev->dev->ifindex, 820 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND, 821 .fc_nlinfo = { 822 .nl_net = net, 823 }, 824 }; 825 826 if (!tb_id) 827 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL; 828 829 tb = fib_new_table(net, tb_id); 830 if (!tb) 831 return; 832 833 cfg.fc_table = tb->tb_id; 834 835 if (type != RTN_LOCAL) 836 cfg.fc_scope = RT_SCOPE_LINK; 837 else 838 cfg.fc_scope = RT_SCOPE_HOST; 839 840 if (cmd == RTM_NEWROUTE) 841 fib_table_insert(net, tb, &cfg); 842 else 843 fib_table_delete(net, tb, &cfg); 844 } 845 846 void fib_add_ifaddr(struct in_ifaddr *ifa) 847 { 848 struct in_device *in_dev = ifa->ifa_dev; 849 struct net_device *dev = in_dev->dev; 850 struct in_ifaddr *prim = ifa; 851 __be32 mask = ifa->ifa_mask; 852 __be32 addr = ifa->ifa_local; 853 __be32 prefix = ifa->ifa_address & mask; 854 855 if (ifa->ifa_flags & IFA_F_SECONDARY) { 856 prim = inet_ifa_byprefix(in_dev, prefix, mask); 857 if (!prim) { 858 pr_warn("%s: bug: prim == NULL\n", __func__); 859 return; 860 } 861 } 862 863 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim); 864 865 if (!(dev->flags & IFF_UP)) 866 return; 867 868 /* Add broadcast address, if it is explicitly assigned. */ 869 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) 870 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 871 872 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) && 873 (prefix != addr || ifa->ifa_prefixlen < 32)) { 874 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 875 fib_magic(RTM_NEWROUTE, 876 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 877 prefix, ifa->ifa_prefixlen, prim); 878 879 /* Add network specific broadcasts, when it takes a sense */ 880 if (ifa->ifa_prefixlen < 31) { 881 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim); 882 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask, 883 32, prim); 884 } 885 } 886 } 887 888 /* Delete primary or secondary address. 889 * Optionally, on secondary address promotion consider the addresses 890 * from subnet iprim as deleted, even if they are in device list. 891 * In this case the secondary ifa can be in device list. 892 */ 893 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim) 894 { 895 struct in_device *in_dev = ifa->ifa_dev; 896 struct net_device *dev = in_dev->dev; 897 struct in_ifaddr *ifa1; 898 struct in_ifaddr *prim = ifa, *prim1 = NULL; 899 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask; 900 __be32 any = ifa->ifa_address & ifa->ifa_mask; 901 #define LOCAL_OK 1 902 #define BRD_OK 2 903 #define BRD0_OK 4 904 #define BRD1_OK 8 905 unsigned int ok = 0; 906 int subnet = 0; /* Primary network */ 907 int gone = 1; /* Address is missing */ 908 int same_prefsrc = 0; /* Another primary with same IP */ 909 910 if (ifa->ifa_flags & IFA_F_SECONDARY) { 911 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask); 912 if (!prim) { 913 /* if the device has been deleted, we don't perform 914 * address promotion 915 */ 916 if (!in_dev->dead) 917 pr_warn("%s: bug: prim == NULL\n", __func__); 918 return; 919 } 920 if (iprim && iprim != prim) { 921 pr_warn("%s: bug: iprim != prim\n", __func__); 922 return; 923 } 924 } else if (!ipv4_is_zeronet(any) && 925 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) { 926 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 927 fib_magic(RTM_DELROUTE, 928 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 929 any, ifa->ifa_prefixlen, prim); 930 subnet = 1; 931 } 932 933 if (in_dev->dead) 934 goto no_promotions; 935 936 /* Deletion is more complicated than add. 937 * We should take care of not to delete too much :-) 938 * 939 * Scan address list to be sure that addresses are really gone. 940 */ 941 942 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) { 943 if (ifa1 == ifa) { 944 /* promotion, keep the IP */ 945 gone = 0; 946 continue; 947 } 948 /* Ignore IFAs from our subnet */ 949 if (iprim && ifa1->ifa_mask == iprim->ifa_mask && 950 inet_ifa_match(ifa1->ifa_address, iprim)) 951 continue; 952 953 /* Ignore ifa1 if it uses different primary IP (prefsrc) */ 954 if (ifa1->ifa_flags & IFA_F_SECONDARY) { 955 /* Another address from our subnet? */ 956 if (ifa1->ifa_mask == prim->ifa_mask && 957 inet_ifa_match(ifa1->ifa_address, prim)) 958 prim1 = prim; 959 else { 960 /* We reached the secondaries, so 961 * same_prefsrc should be determined. 962 */ 963 if (!same_prefsrc) 964 continue; 965 /* Search new prim1 if ifa1 is not 966 * using the current prim1 967 */ 968 if (!prim1 || 969 ifa1->ifa_mask != prim1->ifa_mask || 970 !inet_ifa_match(ifa1->ifa_address, prim1)) 971 prim1 = inet_ifa_byprefix(in_dev, 972 ifa1->ifa_address, 973 ifa1->ifa_mask); 974 if (!prim1) 975 continue; 976 if (prim1->ifa_local != prim->ifa_local) 977 continue; 978 } 979 } else { 980 if (prim->ifa_local != ifa1->ifa_local) 981 continue; 982 prim1 = ifa1; 983 if (prim != prim1) 984 same_prefsrc = 1; 985 } 986 if (ifa->ifa_local == ifa1->ifa_local) 987 ok |= LOCAL_OK; 988 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 989 ok |= BRD_OK; 990 if (brd == ifa1->ifa_broadcast) 991 ok |= BRD1_OK; 992 if (any == ifa1->ifa_broadcast) 993 ok |= BRD0_OK; 994 /* primary has network specific broadcasts */ 995 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) { 996 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask; 997 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask; 998 999 if (!ipv4_is_zeronet(any1)) { 1000 if (ifa->ifa_broadcast == brd1 || 1001 ifa->ifa_broadcast == any1) 1002 ok |= BRD_OK; 1003 if (brd == brd1 || brd == any1) 1004 ok |= BRD1_OK; 1005 if (any == brd1 || any == any1) 1006 ok |= BRD0_OK; 1007 } 1008 } 1009 } 1010 1011 no_promotions: 1012 if (!(ok & BRD_OK)) 1013 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 1014 if (subnet && ifa->ifa_prefixlen < 31) { 1015 if (!(ok & BRD1_OK)) 1016 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim); 1017 if (!(ok & BRD0_OK)) 1018 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim); 1019 } 1020 if (!(ok & LOCAL_OK)) { 1021 unsigned int addr_type; 1022 1023 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim); 1024 1025 /* Check, that this local address finally disappeared. */ 1026 addr_type = inet_addr_type_dev_table(dev_net(dev), dev, 1027 ifa->ifa_local); 1028 if (gone && addr_type != RTN_LOCAL) { 1029 /* And the last, but not the least thing. 1030 * We must flush stray FIB entries. 1031 * 1032 * First of all, we scan fib_info list searching 1033 * for stray nexthop entries, then ignite fib_flush. 1034 */ 1035 if (fib_sync_down_addr(dev, ifa->ifa_local)) 1036 fib_flush(dev_net(dev)); 1037 } 1038 } 1039 #undef LOCAL_OK 1040 #undef BRD_OK 1041 #undef BRD0_OK 1042 #undef BRD1_OK 1043 } 1044 1045 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn) 1046 { 1047 1048 struct fib_result res; 1049 struct flowi4 fl4 = { 1050 .flowi4_mark = frn->fl_mark, 1051 .daddr = frn->fl_addr, 1052 .flowi4_tos = frn->fl_tos, 1053 .flowi4_scope = frn->fl_scope, 1054 }; 1055 struct fib_table *tb; 1056 1057 rcu_read_lock(); 1058 1059 tb = fib_get_table(net, frn->tb_id_in); 1060 1061 frn->err = -ENOENT; 1062 if (tb) { 1063 local_bh_disable(); 1064 1065 frn->tb_id = tb->tb_id; 1066 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 1067 1068 if (!frn->err) { 1069 frn->prefixlen = res.prefixlen; 1070 frn->nh_sel = res.nh_sel; 1071 frn->type = res.type; 1072 frn->scope = res.scope; 1073 } 1074 local_bh_enable(); 1075 } 1076 1077 rcu_read_unlock(); 1078 } 1079 1080 static void nl_fib_input(struct sk_buff *skb) 1081 { 1082 struct net *net; 1083 struct fib_result_nl *frn; 1084 struct nlmsghdr *nlh; 1085 u32 portid; 1086 1087 net = sock_net(skb->sk); 1088 nlh = nlmsg_hdr(skb); 1089 if (skb->len < nlmsg_total_size(sizeof(*frn)) || 1090 skb->len < nlh->nlmsg_len || 1091 nlmsg_len(nlh) < sizeof(*frn)) 1092 return; 1093 1094 skb = netlink_skb_clone(skb, GFP_KERNEL); 1095 if (!skb) 1096 return; 1097 nlh = nlmsg_hdr(skb); 1098 1099 frn = (struct fib_result_nl *) nlmsg_data(nlh); 1100 nl_fib_lookup(net, frn); 1101 1102 portid = NETLINK_CB(skb).portid; /* netlink portid */ 1103 NETLINK_CB(skb).portid = 0; /* from kernel */ 1104 NETLINK_CB(skb).dst_group = 0; /* unicast */ 1105 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT); 1106 } 1107 1108 static int __net_init nl_fib_lookup_init(struct net *net) 1109 { 1110 struct sock *sk; 1111 struct netlink_kernel_cfg cfg = { 1112 .input = nl_fib_input, 1113 }; 1114 1115 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg); 1116 if (!sk) 1117 return -EAFNOSUPPORT; 1118 net->ipv4.fibnl = sk; 1119 return 0; 1120 } 1121 1122 static void nl_fib_lookup_exit(struct net *net) 1123 { 1124 netlink_kernel_release(net->ipv4.fibnl); 1125 net->ipv4.fibnl = NULL; 1126 } 1127 1128 static void fib_disable_ip(struct net_device *dev, unsigned long event, 1129 bool force) 1130 { 1131 if (fib_sync_down_dev(dev, event, force)) 1132 fib_flush(dev_net(dev)); 1133 else 1134 rt_cache_flush(dev_net(dev)); 1135 arp_ifdown(dev); 1136 } 1137 1138 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 1139 { 1140 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; 1141 struct net_device *dev = ifa->ifa_dev->dev; 1142 struct net *net = dev_net(dev); 1143 1144 switch (event) { 1145 case NETDEV_UP: 1146 fib_add_ifaddr(ifa); 1147 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1148 fib_sync_up(dev, RTNH_F_DEAD); 1149 #endif 1150 atomic_inc(&net->ipv4.dev_addr_genid); 1151 rt_cache_flush(dev_net(dev)); 1152 break; 1153 case NETDEV_DOWN: 1154 fib_del_ifaddr(ifa, NULL); 1155 atomic_inc(&net->ipv4.dev_addr_genid); 1156 if (!ifa->ifa_dev->ifa_list) { 1157 /* Last address was deleted from this interface. 1158 * Disable IP. 1159 */ 1160 fib_disable_ip(dev, event, true); 1161 } else { 1162 rt_cache_flush(dev_net(dev)); 1163 } 1164 break; 1165 } 1166 return NOTIFY_DONE; 1167 } 1168 1169 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 1170 { 1171 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1172 struct netdev_notifier_changeupper_info *info; 1173 struct in_device *in_dev; 1174 struct net *net = dev_net(dev); 1175 unsigned int flags; 1176 1177 if (event == NETDEV_UNREGISTER) { 1178 fib_disable_ip(dev, event, true); 1179 rt_flush_dev(dev); 1180 return NOTIFY_DONE; 1181 } 1182 1183 in_dev = __in_dev_get_rtnl(dev); 1184 if (!in_dev) 1185 return NOTIFY_DONE; 1186 1187 switch (event) { 1188 case NETDEV_UP: 1189 for_ifa(in_dev) { 1190 fib_add_ifaddr(ifa); 1191 } endfor_ifa(in_dev); 1192 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1193 fib_sync_up(dev, RTNH_F_DEAD); 1194 #endif 1195 atomic_inc(&net->ipv4.dev_addr_genid); 1196 rt_cache_flush(net); 1197 break; 1198 case NETDEV_DOWN: 1199 fib_disable_ip(dev, event, false); 1200 break; 1201 case NETDEV_CHANGE: 1202 flags = dev_get_flags(dev); 1203 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 1204 fib_sync_up(dev, RTNH_F_LINKDOWN); 1205 else 1206 fib_sync_down_dev(dev, event, false); 1207 /* fall through */ 1208 case NETDEV_CHANGEMTU: 1209 rt_cache_flush(net); 1210 break; 1211 case NETDEV_CHANGEUPPER: 1212 info = ptr; 1213 /* flush all routes if dev is linked to or unlinked from 1214 * an L3 master device (e.g., VRF) 1215 */ 1216 if (info->upper_dev && netif_is_l3_master(info->upper_dev)) 1217 fib_disable_ip(dev, NETDEV_DOWN, true); 1218 break; 1219 } 1220 return NOTIFY_DONE; 1221 } 1222 1223 static struct notifier_block fib_inetaddr_notifier = { 1224 .notifier_call = fib_inetaddr_event, 1225 }; 1226 1227 static struct notifier_block fib_netdev_notifier = { 1228 .notifier_call = fib_netdev_event, 1229 }; 1230 1231 static int __net_init ip_fib_net_init(struct net *net) 1232 { 1233 int err; 1234 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ; 1235 1236 net->ipv4.fib_seq = 0; 1237 1238 /* Avoid false sharing : Use at least a full cache line */ 1239 size = max_t(size_t, size, L1_CACHE_BYTES); 1240 1241 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL); 1242 if (!net->ipv4.fib_table_hash) 1243 return -ENOMEM; 1244 1245 err = fib4_rules_init(net); 1246 if (err < 0) 1247 goto fail; 1248 return 0; 1249 1250 fail: 1251 kfree(net->ipv4.fib_table_hash); 1252 return err; 1253 } 1254 1255 static void ip_fib_net_exit(struct net *net) 1256 { 1257 unsigned int i; 1258 1259 rtnl_lock(); 1260 #ifdef CONFIG_IP_MULTIPLE_TABLES 1261 RCU_INIT_POINTER(net->ipv4.fib_main, NULL); 1262 RCU_INIT_POINTER(net->ipv4.fib_default, NULL); 1263 #endif 1264 for (i = 0; i < FIB_TABLE_HASHSZ; i++) { 1265 struct hlist_head *head = &net->ipv4.fib_table_hash[i]; 1266 struct hlist_node *tmp; 1267 struct fib_table *tb; 1268 1269 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) { 1270 hlist_del(&tb->tb_hlist); 1271 fib_table_flush(net, tb); 1272 fib_free_table(tb); 1273 } 1274 } 1275 1276 #ifdef CONFIG_IP_MULTIPLE_TABLES 1277 fib4_rules_exit(net); 1278 #endif 1279 rtnl_unlock(); 1280 kfree(net->ipv4.fib_table_hash); 1281 } 1282 1283 static int __net_init fib_net_init(struct net *net) 1284 { 1285 int error; 1286 1287 #ifdef CONFIG_IP_ROUTE_CLASSID 1288 net->ipv4.fib_num_tclassid_users = 0; 1289 #endif 1290 error = ip_fib_net_init(net); 1291 if (error < 0) 1292 goto out; 1293 error = nl_fib_lookup_init(net); 1294 if (error < 0) 1295 goto out_nlfl; 1296 error = fib_proc_init(net); 1297 if (error < 0) 1298 goto out_proc; 1299 out: 1300 return error; 1301 1302 out_proc: 1303 nl_fib_lookup_exit(net); 1304 out_nlfl: 1305 ip_fib_net_exit(net); 1306 goto out; 1307 } 1308 1309 static void __net_exit fib_net_exit(struct net *net) 1310 { 1311 fib_proc_exit(net); 1312 nl_fib_lookup_exit(net); 1313 ip_fib_net_exit(net); 1314 } 1315 1316 static struct pernet_operations fib_net_ops = { 1317 .init = fib_net_init, 1318 .exit = fib_net_exit, 1319 }; 1320 1321 void __init ip_fib_init(void) 1322 { 1323 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL); 1324 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL); 1325 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL); 1326 1327 register_pernet_subsys(&fib_net_ops); 1328 register_netdevice_notifier(&fib_netdev_notifier); 1329 register_inetaddr_notifier(&fib_inetaddr_notifier); 1330 1331 fib_trie_init(); 1332 } 1333