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 if (err < 0) 637 goto errout; 638 639 memset(cfg, 0, sizeof(*cfg)); 640 641 rtm = nlmsg_data(nlh); 642 cfg->fc_dst_len = rtm->rtm_dst_len; 643 cfg->fc_tos = rtm->rtm_tos; 644 cfg->fc_table = rtm->rtm_table; 645 cfg->fc_protocol = rtm->rtm_protocol; 646 cfg->fc_scope = rtm->rtm_scope; 647 cfg->fc_type = rtm->rtm_type; 648 cfg->fc_flags = rtm->rtm_flags; 649 cfg->fc_nlflags = nlh->nlmsg_flags; 650 651 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid; 652 cfg->fc_nlinfo.nlh = nlh; 653 cfg->fc_nlinfo.nl_net = net; 654 655 if (cfg->fc_type > RTN_MAX) { 656 err = -EINVAL; 657 goto errout; 658 } 659 660 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) { 661 switch (nla_type(attr)) { 662 case RTA_DST: 663 cfg->fc_dst = nla_get_be32(attr); 664 break; 665 case RTA_OIF: 666 cfg->fc_oif = nla_get_u32(attr); 667 break; 668 case RTA_GATEWAY: 669 cfg->fc_gw = nla_get_be32(attr); 670 break; 671 case RTA_PRIORITY: 672 cfg->fc_priority = nla_get_u32(attr); 673 break; 674 case RTA_PREFSRC: 675 cfg->fc_prefsrc = nla_get_be32(attr); 676 break; 677 case RTA_METRICS: 678 cfg->fc_mx = nla_data(attr); 679 cfg->fc_mx_len = nla_len(attr); 680 break; 681 case RTA_MULTIPATH: 682 err = lwtunnel_valid_encap_type_attr(nla_data(attr), 683 nla_len(attr)); 684 if (err < 0) 685 goto errout; 686 cfg->fc_mp = nla_data(attr); 687 cfg->fc_mp_len = nla_len(attr); 688 break; 689 case RTA_FLOW: 690 cfg->fc_flow = nla_get_u32(attr); 691 break; 692 case RTA_TABLE: 693 cfg->fc_table = nla_get_u32(attr); 694 break; 695 case RTA_ENCAP: 696 cfg->fc_encap = attr; 697 break; 698 case RTA_ENCAP_TYPE: 699 cfg->fc_encap_type = nla_get_u16(attr); 700 err = lwtunnel_valid_encap_type(cfg->fc_encap_type); 701 if (err < 0) 702 goto errout; 703 break; 704 } 705 } 706 707 return 0; 708 errout: 709 return err; 710 } 711 712 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh) 713 { 714 struct net *net = sock_net(skb->sk); 715 struct fib_config cfg; 716 struct fib_table *tb; 717 int err; 718 719 err = rtm_to_fib_config(net, skb, nlh, &cfg); 720 if (err < 0) 721 goto errout; 722 723 tb = fib_get_table(net, cfg.fc_table); 724 if (!tb) { 725 err = -ESRCH; 726 goto errout; 727 } 728 729 err = fib_table_delete(net, tb, &cfg); 730 errout: 731 return err; 732 } 733 734 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh) 735 { 736 struct net *net = sock_net(skb->sk); 737 struct fib_config cfg; 738 struct fib_table *tb; 739 int err; 740 741 err = rtm_to_fib_config(net, skb, nlh, &cfg); 742 if (err < 0) 743 goto errout; 744 745 tb = fib_new_table(net, cfg.fc_table); 746 if (!tb) { 747 err = -ENOBUFS; 748 goto errout; 749 } 750 751 err = fib_table_insert(net, tb, &cfg); 752 errout: 753 return err; 754 } 755 756 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 757 { 758 struct net *net = sock_net(skb->sk); 759 unsigned int h, s_h; 760 unsigned int e = 0, s_e; 761 struct fib_table *tb; 762 struct hlist_head *head; 763 int dumped = 0; 764 765 if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) && 766 ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED) 767 return skb->len; 768 769 s_h = cb->args[0]; 770 s_e = cb->args[1]; 771 772 rcu_read_lock(); 773 774 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) { 775 e = 0; 776 head = &net->ipv4.fib_table_hash[h]; 777 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 778 if (e < s_e) 779 goto next; 780 if (dumped) 781 memset(&cb->args[2], 0, sizeof(cb->args) - 782 2 * sizeof(cb->args[0])); 783 if (fib_table_dump(tb, skb, cb) < 0) 784 goto out; 785 dumped = 1; 786 next: 787 e++; 788 } 789 } 790 out: 791 rcu_read_unlock(); 792 793 cb->args[1] = e; 794 cb->args[0] = h; 795 796 return skb->len; 797 } 798 799 /* Prepare and feed intra-kernel routing request. 800 * Really, it should be netlink message, but :-( netlink 801 * can be not configured, so that we feed it directly 802 * to fib engine. It is legal, because all events occur 803 * only when netlink is already locked. 804 */ 805 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa) 806 { 807 struct net *net = dev_net(ifa->ifa_dev->dev); 808 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev); 809 struct fib_table *tb; 810 struct fib_config cfg = { 811 .fc_protocol = RTPROT_KERNEL, 812 .fc_type = type, 813 .fc_dst = dst, 814 .fc_dst_len = dst_len, 815 .fc_prefsrc = ifa->ifa_local, 816 .fc_oif = ifa->ifa_dev->dev->ifindex, 817 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND, 818 .fc_nlinfo = { 819 .nl_net = net, 820 }, 821 }; 822 823 if (!tb_id) 824 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL; 825 826 tb = fib_new_table(net, tb_id); 827 if (!tb) 828 return; 829 830 cfg.fc_table = tb->tb_id; 831 832 if (type != RTN_LOCAL) 833 cfg.fc_scope = RT_SCOPE_LINK; 834 else 835 cfg.fc_scope = RT_SCOPE_HOST; 836 837 if (cmd == RTM_NEWROUTE) 838 fib_table_insert(net, tb, &cfg); 839 else 840 fib_table_delete(net, tb, &cfg); 841 } 842 843 void fib_add_ifaddr(struct in_ifaddr *ifa) 844 { 845 struct in_device *in_dev = ifa->ifa_dev; 846 struct net_device *dev = in_dev->dev; 847 struct in_ifaddr *prim = ifa; 848 __be32 mask = ifa->ifa_mask; 849 __be32 addr = ifa->ifa_local; 850 __be32 prefix = ifa->ifa_address & mask; 851 852 if (ifa->ifa_flags & IFA_F_SECONDARY) { 853 prim = inet_ifa_byprefix(in_dev, prefix, mask); 854 if (!prim) { 855 pr_warn("%s: bug: prim == NULL\n", __func__); 856 return; 857 } 858 } 859 860 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim); 861 862 if (!(dev->flags & IFF_UP)) 863 return; 864 865 /* Add broadcast address, if it is explicitly assigned. */ 866 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) 867 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 868 869 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) && 870 (prefix != addr || ifa->ifa_prefixlen < 32)) { 871 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 872 fib_magic(RTM_NEWROUTE, 873 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 874 prefix, ifa->ifa_prefixlen, prim); 875 876 /* Add network specific broadcasts, when it takes a sense */ 877 if (ifa->ifa_prefixlen < 31) { 878 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim); 879 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask, 880 32, prim); 881 } 882 } 883 } 884 885 /* Delete primary or secondary address. 886 * Optionally, on secondary address promotion consider the addresses 887 * from subnet iprim as deleted, even if they are in device list. 888 * In this case the secondary ifa can be in device list. 889 */ 890 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim) 891 { 892 struct in_device *in_dev = ifa->ifa_dev; 893 struct net_device *dev = in_dev->dev; 894 struct in_ifaddr *ifa1; 895 struct in_ifaddr *prim = ifa, *prim1 = NULL; 896 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask; 897 __be32 any = ifa->ifa_address & ifa->ifa_mask; 898 #define LOCAL_OK 1 899 #define BRD_OK 2 900 #define BRD0_OK 4 901 #define BRD1_OK 8 902 unsigned int ok = 0; 903 int subnet = 0; /* Primary network */ 904 int gone = 1; /* Address is missing */ 905 int same_prefsrc = 0; /* Another primary with same IP */ 906 907 if (ifa->ifa_flags & IFA_F_SECONDARY) { 908 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask); 909 if (!prim) { 910 /* if the device has been deleted, we don't perform 911 * address promotion 912 */ 913 if (!in_dev->dead) 914 pr_warn("%s: bug: prim == NULL\n", __func__); 915 return; 916 } 917 if (iprim && iprim != prim) { 918 pr_warn("%s: bug: iprim != prim\n", __func__); 919 return; 920 } 921 } else if (!ipv4_is_zeronet(any) && 922 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) { 923 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 924 fib_magic(RTM_DELROUTE, 925 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 926 any, ifa->ifa_prefixlen, prim); 927 subnet = 1; 928 } 929 930 if (in_dev->dead) 931 goto no_promotions; 932 933 /* Deletion is more complicated than add. 934 * We should take care of not to delete too much :-) 935 * 936 * Scan address list to be sure that addresses are really gone. 937 */ 938 939 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) { 940 if (ifa1 == ifa) { 941 /* promotion, keep the IP */ 942 gone = 0; 943 continue; 944 } 945 /* Ignore IFAs from our subnet */ 946 if (iprim && ifa1->ifa_mask == iprim->ifa_mask && 947 inet_ifa_match(ifa1->ifa_address, iprim)) 948 continue; 949 950 /* Ignore ifa1 if it uses different primary IP (prefsrc) */ 951 if (ifa1->ifa_flags & IFA_F_SECONDARY) { 952 /* Another address from our subnet? */ 953 if (ifa1->ifa_mask == prim->ifa_mask && 954 inet_ifa_match(ifa1->ifa_address, prim)) 955 prim1 = prim; 956 else { 957 /* We reached the secondaries, so 958 * same_prefsrc should be determined. 959 */ 960 if (!same_prefsrc) 961 continue; 962 /* Search new prim1 if ifa1 is not 963 * using the current prim1 964 */ 965 if (!prim1 || 966 ifa1->ifa_mask != prim1->ifa_mask || 967 !inet_ifa_match(ifa1->ifa_address, prim1)) 968 prim1 = inet_ifa_byprefix(in_dev, 969 ifa1->ifa_address, 970 ifa1->ifa_mask); 971 if (!prim1) 972 continue; 973 if (prim1->ifa_local != prim->ifa_local) 974 continue; 975 } 976 } else { 977 if (prim->ifa_local != ifa1->ifa_local) 978 continue; 979 prim1 = ifa1; 980 if (prim != prim1) 981 same_prefsrc = 1; 982 } 983 if (ifa->ifa_local == ifa1->ifa_local) 984 ok |= LOCAL_OK; 985 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 986 ok |= BRD_OK; 987 if (brd == ifa1->ifa_broadcast) 988 ok |= BRD1_OK; 989 if (any == ifa1->ifa_broadcast) 990 ok |= BRD0_OK; 991 /* primary has network specific broadcasts */ 992 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) { 993 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask; 994 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask; 995 996 if (!ipv4_is_zeronet(any1)) { 997 if (ifa->ifa_broadcast == brd1 || 998 ifa->ifa_broadcast == any1) 999 ok |= BRD_OK; 1000 if (brd == brd1 || brd == any1) 1001 ok |= BRD1_OK; 1002 if (any == brd1 || any == any1) 1003 ok |= BRD0_OK; 1004 } 1005 } 1006 } 1007 1008 no_promotions: 1009 if (!(ok & BRD_OK)) 1010 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 1011 if (subnet && ifa->ifa_prefixlen < 31) { 1012 if (!(ok & BRD1_OK)) 1013 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim); 1014 if (!(ok & BRD0_OK)) 1015 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim); 1016 } 1017 if (!(ok & LOCAL_OK)) { 1018 unsigned int addr_type; 1019 1020 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim); 1021 1022 /* Check, that this local address finally disappeared. */ 1023 addr_type = inet_addr_type_dev_table(dev_net(dev), dev, 1024 ifa->ifa_local); 1025 if (gone && addr_type != RTN_LOCAL) { 1026 /* And the last, but not the least thing. 1027 * We must flush stray FIB entries. 1028 * 1029 * First of all, we scan fib_info list searching 1030 * for stray nexthop entries, then ignite fib_flush. 1031 */ 1032 if (fib_sync_down_addr(dev, ifa->ifa_local)) 1033 fib_flush(dev_net(dev)); 1034 } 1035 } 1036 #undef LOCAL_OK 1037 #undef BRD_OK 1038 #undef BRD0_OK 1039 #undef BRD1_OK 1040 } 1041 1042 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn) 1043 { 1044 1045 struct fib_result res; 1046 struct flowi4 fl4 = { 1047 .flowi4_mark = frn->fl_mark, 1048 .daddr = frn->fl_addr, 1049 .flowi4_tos = frn->fl_tos, 1050 .flowi4_scope = frn->fl_scope, 1051 }; 1052 struct fib_table *tb; 1053 1054 rcu_read_lock(); 1055 1056 tb = fib_get_table(net, frn->tb_id_in); 1057 1058 frn->err = -ENOENT; 1059 if (tb) { 1060 local_bh_disable(); 1061 1062 frn->tb_id = tb->tb_id; 1063 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 1064 1065 if (!frn->err) { 1066 frn->prefixlen = res.prefixlen; 1067 frn->nh_sel = res.nh_sel; 1068 frn->type = res.type; 1069 frn->scope = res.scope; 1070 } 1071 local_bh_enable(); 1072 } 1073 1074 rcu_read_unlock(); 1075 } 1076 1077 static void nl_fib_input(struct sk_buff *skb) 1078 { 1079 struct net *net; 1080 struct fib_result_nl *frn; 1081 struct nlmsghdr *nlh; 1082 u32 portid; 1083 1084 net = sock_net(skb->sk); 1085 nlh = nlmsg_hdr(skb); 1086 if (skb->len < nlmsg_total_size(sizeof(*frn)) || 1087 skb->len < nlh->nlmsg_len || 1088 nlmsg_len(nlh) < sizeof(*frn)) 1089 return; 1090 1091 skb = netlink_skb_clone(skb, GFP_KERNEL); 1092 if (!skb) 1093 return; 1094 nlh = nlmsg_hdr(skb); 1095 1096 frn = (struct fib_result_nl *) nlmsg_data(nlh); 1097 nl_fib_lookup(net, frn); 1098 1099 portid = NETLINK_CB(skb).portid; /* netlink portid */ 1100 NETLINK_CB(skb).portid = 0; /* from kernel */ 1101 NETLINK_CB(skb).dst_group = 0; /* unicast */ 1102 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT); 1103 } 1104 1105 static int __net_init nl_fib_lookup_init(struct net *net) 1106 { 1107 struct sock *sk; 1108 struct netlink_kernel_cfg cfg = { 1109 .input = nl_fib_input, 1110 }; 1111 1112 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg); 1113 if (!sk) 1114 return -EAFNOSUPPORT; 1115 net->ipv4.fibnl = sk; 1116 return 0; 1117 } 1118 1119 static void nl_fib_lookup_exit(struct net *net) 1120 { 1121 netlink_kernel_release(net->ipv4.fibnl); 1122 net->ipv4.fibnl = NULL; 1123 } 1124 1125 static void fib_disable_ip(struct net_device *dev, unsigned long event, 1126 bool force) 1127 { 1128 if (fib_sync_down_dev(dev, event, force)) 1129 fib_flush(dev_net(dev)); 1130 rt_cache_flush(dev_net(dev)); 1131 arp_ifdown(dev); 1132 } 1133 1134 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 1135 { 1136 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; 1137 struct net_device *dev = ifa->ifa_dev->dev; 1138 struct net *net = dev_net(dev); 1139 1140 switch (event) { 1141 case NETDEV_UP: 1142 fib_add_ifaddr(ifa); 1143 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1144 fib_sync_up(dev, RTNH_F_DEAD); 1145 #endif 1146 atomic_inc(&net->ipv4.dev_addr_genid); 1147 rt_cache_flush(dev_net(dev)); 1148 break; 1149 case NETDEV_DOWN: 1150 fib_del_ifaddr(ifa, NULL); 1151 atomic_inc(&net->ipv4.dev_addr_genid); 1152 if (!ifa->ifa_dev->ifa_list) { 1153 /* Last address was deleted from this interface. 1154 * Disable IP. 1155 */ 1156 fib_disable_ip(dev, event, true); 1157 } else { 1158 rt_cache_flush(dev_net(dev)); 1159 } 1160 break; 1161 } 1162 return NOTIFY_DONE; 1163 } 1164 1165 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 1166 { 1167 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1168 struct netdev_notifier_changeupper_info *info; 1169 struct in_device *in_dev; 1170 struct net *net = dev_net(dev); 1171 unsigned int flags; 1172 1173 if (event == NETDEV_UNREGISTER) { 1174 fib_disable_ip(dev, event, true); 1175 rt_flush_dev(dev); 1176 return NOTIFY_DONE; 1177 } 1178 1179 in_dev = __in_dev_get_rtnl(dev); 1180 if (!in_dev) 1181 return NOTIFY_DONE; 1182 1183 switch (event) { 1184 case NETDEV_UP: 1185 for_ifa(in_dev) { 1186 fib_add_ifaddr(ifa); 1187 } endfor_ifa(in_dev); 1188 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1189 fib_sync_up(dev, RTNH_F_DEAD); 1190 #endif 1191 atomic_inc(&net->ipv4.dev_addr_genid); 1192 rt_cache_flush(net); 1193 break; 1194 case NETDEV_DOWN: 1195 fib_disable_ip(dev, event, false); 1196 break; 1197 case NETDEV_CHANGE: 1198 flags = dev_get_flags(dev); 1199 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 1200 fib_sync_up(dev, RTNH_F_LINKDOWN); 1201 else 1202 fib_sync_down_dev(dev, event, false); 1203 /* fall through */ 1204 case NETDEV_CHANGEMTU: 1205 rt_cache_flush(net); 1206 break; 1207 case NETDEV_CHANGEUPPER: 1208 info = ptr; 1209 /* flush all routes if dev is linked to or unlinked from 1210 * an L3 master device (e.g., VRF) 1211 */ 1212 if (info->upper_dev && netif_is_l3_master(info->upper_dev)) 1213 fib_disable_ip(dev, NETDEV_DOWN, true); 1214 break; 1215 } 1216 return NOTIFY_DONE; 1217 } 1218 1219 static struct notifier_block fib_inetaddr_notifier = { 1220 .notifier_call = fib_inetaddr_event, 1221 }; 1222 1223 static struct notifier_block fib_netdev_notifier = { 1224 .notifier_call = fib_netdev_event, 1225 }; 1226 1227 static int __net_init ip_fib_net_init(struct net *net) 1228 { 1229 int err; 1230 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ; 1231 1232 net->ipv4.fib_seq = 0; 1233 1234 /* Avoid false sharing : Use at least a full cache line */ 1235 size = max_t(size_t, size, L1_CACHE_BYTES); 1236 1237 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL); 1238 if (!net->ipv4.fib_table_hash) 1239 return -ENOMEM; 1240 1241 err = fib4_rules_init(net); 1242 if (err < 0) 1243 goto fail; 1244 return 0; 1245 1246 fail: 1247 kfree(net->ipv4.fib_table_hash); 1248 return err; 1249 } 1250 1251 static void ip_fib_net_exit(struct net *net) 1252 { 1253 unsigned int i; 1254 1255 rtnl_lock(); 1256 #ifdef CONFIG_IP_MULTIPLE_TABLES 1257 RCU_INIT_POINTER(net->ipv4.fib_main, NULL); 1258 RCU_INIT_POINTER(net->ipv4.fib_default, NULL); 1259 #endif 1260 for (i = 0; i < FIB_TABLE_HASHSZ; i++) { 1261 struct hlist_head *head = &net->ipv4.fib_table_hash[i]; 1262 struct hlist_node *tmp; 1263 struct fib_table *tb; 1264 1265 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) { 1266 hlist_del(&tb->tb_hlist); 1267 fib_table_flush(net, tb); 1268 fib_free_table(tb); 1269 } 1270 } 1271 1272 #ifdef CONFIG_IP_MULTIPLE_TABLES 1273 fib4_rules_exit(net); 1274 #endif 1275 rtnl_unlock(); 1276 kfree(net->ipv4.fib_table_hash); 1277 } 1278 1279 static int __net_init fib_net_init(struct net *net) 1280 { 1281 int error; 1282 1283 #ifdef CONFIG_IP_ROUTE_CLASSID 1284 net->ipv4.fib_num_tclassid_users = 0; 1285 #endif 1286 error = ip_fib_net_init(net); 1287 if (error < 0) 1288 goto out; 1289 error = nl_fib_lookup_init(net); 1290 if (error < 0) 1291 goto out_nlfl; 1292 error = fib_proc_init(net); 1293 if (error < 0) 1294 goto out_proc; 1295 out: 1296 return error; 1297 1298 out_proc: 1299 nl_fib_lookup_exit(net); 1300 out_nlfl: 1301 ip_fib_net_exit(net); 1302 goto out; 1303 } 1304 1305 static void __net_exit fib_net_exit(struct net *net) 1306 { 1307 fib_proc_exit(net); 1308 nl_fib_lookup_exit(net); 1309 ip_fib_net_exit(net); 1310 } 1311 1312 static struct pernet_operations fib_net_ops = { 1313 .init = fib_net_init, 1314 .exit = fib_net_exit, 1315 }; 1316 1317 void __init ip_fib_init(void) 1318 { 1319 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL); 1320 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL); 1321 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL); 1322 1323 register_pernet_subsys(&fib_net_ops); 1324 register_netdevice_notifier(&fib_netdev_notifier); 1325 register_inetaddr_notifier(&fib_inetaddr_notifier); 1326 1327 fib_trie_init(); 1328 } 1329