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 <trace/events/fib.h> 50 51 #ifndef CONFIG_IP_MULTIPLE_TABLES 52 53 static int __net_init fib4_rules_init(struct net *net) 54 { 55 struct fib_table *local_table, *main_table; 56 57 main_table = fib_trie_table(RT_TABLE_MAIN, NULL); 58 if (!main_table) 59 return -ENOMEM; 60 61 local_table = fib_trie_table(RT_TABLE_LOCAL, main_table); 62 if (!local_table) 63 goto fail; 64 65 hlist_add_head_rcu(&local_table->tb_hlist, 66 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]); 67 hlist_add_head_rcu(&main_table->tb_hlist, 68 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]); 69 return 0; 70 71 fail: 72 fib_free_table(main_table); 73 return -ENOMEM; 74 } 75 #else 76 77 struct fib_table *fib_new_table(struct net *net, u32 id) 78 { 79 struct fib_table *tb, *alias = NULL; 80 unsigned int h; 81 82 if (id == 0) 83 id = RT_TABLE_MAIN; 84 tb = fib_get_table(net, id); 85 if (tb) 86 return tb; 87 88 if (id == RT_TABLE_LOCAL) 89 alias = fib_new_table(net, RT_TABLE_MAIN); 90 91 tb = fib_trie_table(id, alias); 92 if (!tb) 93 return NULL; 94 95 switch (id) { 96 case RT_TABLE_MAIN: 97 rcu_assign_pointer(net->ipv4.fib_main, tb); 98 break; 99 case RT_TABLE_DEFAULT: 100 rcu_assign_pointer(net->ipv4.fib_default, tb); 101 break; 102 default: 103 break; 104 } 105 106 h = id & (FIB_TABLE_HASHSZ - 1); 107 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]); 108 return tb; 109 } 110 EXPORT_SYMBOL_GPL(fib_new_table); 111 112 /* caller must hold either rtnl or rcu read lock */ 113 struct fib_table *fib_get_table(struct net *net, u32 id) 114 { 115 struct fib_table *tb; 116 struct hlist_head *head; 117 unsigned int h; 118 119 if (id == 0) 120 id = RT_TABLE_MAIN; 121 h = id & (FIB_TABLE_HASHSZ - 1); 122 123 head = &net->ipv4.fib_table_hash[h]; 124 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 125 if (tb->tb_id == id) 126 return tb; 127 } 128 return NULL; 129 } 130 #endif /* CONFIG_IP_MULTIPLE_TABLES */ 131 132 static void fib_replace_table(struct net *net, struct fib_table *old, 133 struct fib_table *new) 134 { 135 #ifdef CONFIG_IP_MULTIPLE_TABLES 136 switch (new->tb_id) { 137 case RT_TABLE_MAIN: 138 rcu_assign_pointer(net->ipv4.fib_main, new); 139 break; 140 case RT_TABLE_DEFAULT: 141 rcu_assign_pointer(net->ipv4.fib_default, new); 142 break; 143 default: 144 break; 145 } 146 147 #endif 148 /* replace the old table in the hlist */ 149 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist); 150 } 151 152 int fib_unmerge(struct net *net) 153 { 154 struct fib_table *old, *new, *main_table; 155 156 /* attempt to fetch local table if it has been allocated */ 157 old = fib_get_table(net, RT_TABLE_LOCAL); 158 if (!old) 159 return 0; 160 161 new = fib_trie_unmerge(old); 162 if (!new) 163 return -ENOMEM; 164 165 /* table is already unmerged */ 166 if (new == old) 167 return 0; 168 169 /* replace merged table with clean table */ 170 fib_replace_table(net, old, new); 171 fib_free_table(old); 172 173 /* attempt to fetch main table if it has been allocated */ 174 main_table = fib_get_table(net, RT_TABLE_MAIN); 175 if (!main_table) 176 return 0; 177 178 /* flush local entries from main table */ 179 fib_table_flush_external(main_table); 180 181 return 0; 182 } 183 184 static void fib_flush(struct net *net) 185 { 186 int flushed = 0; 187 unsigned int h; 188 189 for (h = 0; h < FIB_TABLE_HASHSZ; h++) { 190 struct hlist_head *head = &net->ipv4.fib_table_hash[h]; 191 struct hlist_node *tmp; 192 struct fib_table *tb; 193 194 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) 195 flushed += fib_table_flush(net, tb); 196 } 197 198 if (flushed) 199 rt_cache_flush(net); 200 } 201 202 /* 203 * Find address type as if only "dev" was present in the system. If 204 * on_dev is NULL then all interfaces are taken into consideration. 205 */ 206 static inline unsigned int __inet_dev_addr_type(struct net *net, 207 const struct net_device *dev, 208 __be32 addr, u32 tb_id) 209 { 210 struct flowi4 fl4 = { .daddr = addr }; 211 struct fib_result res; 212 unsigned int ret = RTN_BROADCAST; 213 struct fib_table *table; 214 215 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr)) 216 return RTN_BROADCAST; 217 if (ipv4_is_multicast(addr)) 218 return RTN_MULTICAST; 219 220 rcu_read_lock(); 221 222 table = fib_get_table(net, tb_id); 223 if (table) { 224 ret = RTN_UNICAST; 225 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) { 226 if (!dev || dev == res.fi->fib_dev) 227 ret = res.type; 228 } 229 } 230 231 rcu_read_unlock(); 232 return ret; 233 } 234 235 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id) 236 { 237 return __inet_dev_addr_type(net, NULL, addr, tb_id); 238 } 239 EXPORT_SYMBOL(inet_addr_type_table); 240 241 unsigned int inet_addr_type(struct net *net, __be32 addr) 242 { 243 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL); 244 } 245 EXPORT_SYMBOL(inet_addr_type); 246 247 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev, 248 __be32 addr) 249 { 250 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL; 251 252 return __inet_dev_addr_type(net, dev, addr, rt_table); 253 } 254 EXPORT_SYMBOL(inet_dev_addr_type); 255 256 /* inet_addr_type with dev == NULL but using the table from a dev 257 * if one is associated 258 */ 259 unsigned int inet_addr_type_dev_table(struct net *net, 260 const struct net_device *dev, 261 __be32 addr) 262 { 263 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL; 264 265 return __inet_dev_addr_type(net, NULL, addr, rt_table); 266 } 267 EXPORT_SYMBOL(inet_addr_type_dev_table); 268 269 __be32 fib_compute_spec_dst(struct sk_buff *skb) 270 { 271 struct net_device *dev = skb->dev; 272 struct in_device *in_dev; 273 struct fib_result res; 274 struct rtable *rt; 275 struct net *net; 276 int scope; 277 278 rt = skb_rtable(skb); 279 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) == 280 RTCF_LOCAL) 281 return ip_hdr(skb)->daddr; 282 283 in_dev = __in_dev_get_rcu(dev); 284 BUG_ON(!in_dev); 285 286 net = dev_net(dev); 287 288 scope = RT_SCOPE_UNIVERSE; 289 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) { 290 struct flowi4 fl4 = { 291 .flowi4_iif = LOOPBACK_IFINDEX, 292 .daddr = ip_hdr(skb)->saddr, 293 .flowi4_tos = RT_TOS(ip_hdr(skb)->tos), 294 .flowi4_scope = scope, 295 .flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0, 296 }; 297 if (!fib_lookup(net, &fl4, &res, 0)) 298 return FIB_RES_PREFSRC(net, res); 299 } else { 300 scope = RT_SCOPE_LINK; 301 } 302 303 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope); 304 } 305 306 /* Given (packet source, input interface) and optional (dst, oif, tos): 307 * - (main) check, that source is valid i.e. not broadcast or our local 308 * address. 309 * - figure out what "logical" interface this packet arrived 310 * and calculate "specific destination" address. 311 * - check, that packet arrived from expected physical interface. 312 * called with rcu_read_lock() 313 */ 314 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 315 u8 tos, int oif, struct net_device *dev, 316 int rpf, struct in_device *idev, u32 *itag) 317 { 318 int ret, no_addr; 319 struct fib_result res; 320 struct flowi4 fl4; 321 struct net *net; 322 bool dev_match; 323 324 fl4.flowi4_oif = 0; 325 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev); 326 if (!fl4.flowi4_iif) 327 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX; 328 fl4.daddr = src; 329 fl4.saddr = dst; 330 fl4.flowi4_tos = tos; 331 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 332 fl4.flowi4_tun_key.tun_id = 0; 333 fl4.flowi4_flags = 0; 334 335 no_addr = idev->ifa_list == NULL; 336 337 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0; 338 339 trace_fib_validate_source(dev, &fl4); 340 341 net = dev_net(dev); 342 if (fib_lookup(net, &fl4, &res, 0)) 343 goto last_resort; 344 if (res.type != RTN_UNICAST && 345 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev))) 346 goto e_inval; 347 if (!rpf && !fib_num_tclassid_users(dev_net(dev)) && 348 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) 349 goto last_resort; 350 fib_combine_itag(itag, &res); 351 dev_match = false; 352 353 #ifdef CONFIG_IP_ROUTE_MULTIPATH 354 for (ret = 0; ret < res.fi->fib_nhs; ret++) { 355 struct fib_nh *nh = &res.fi->fib_nh[ret]; 356 357 if (nh->nh_dev == dev) { 358 dev_match = true; 359 break; 360 } else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) { 361 dev_match = true; 362 break; 363 } 364 } 365 #else 366 if (FIB_RES_DEV(res) == dev) 367 dev_match = true; 368 #endif 369 if (dev_match) { 370 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 371 return ret; 372 } 373 if (no_addr) 374 goto last_resort; 375 if (rpf == 1) 376 goto e_rpf; 377 fl4.flowi4_oif = dev->ifindex; 378 379 ret = 0; 380 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) { 381 if (res.type == RTN_UNICAST) 382 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 383 } 384 return ret; 385 386 last_resort: 387 if (rpf) 388 goto e_rpf; 389 *itag = 0; 390 return 0; 391 392 e_inval: 393 return -EINVAL; 394 e_rpf: 395 return -EXDEV; 396 } 397 398 /* Ignore rp_filter for packets protected by IPsec. */ 399 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 400 u8 tos, int oif, struct net_device *dev, 401 struct in_device *idev, u32 *itag) 402 { 403 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev); 404 405 if (!r && !fib_num_tclassid_users(dev_net(dev)) && 406 IN_DEV_ACCEPT_LOCAL(idev) && 407 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) { 408 *itag = 0; 409 return 0; 410 } 411 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag); 412 } 413 414 static inline __be32 sk_extract_addr(struct sockaddr *addr) 415 { 416 return ((struct sockaddr_in *) addr)->sin_addr.s_addr; 417 } 418 419 static int put_rtax(struct nlattr *mx, int len, int type, u32 value) 420 { 421 struct nlattr *nla; 422 423 nla = (struct nlattr *) ((char *) mx + len); 424 nla->nla_type = type; 425 nla->nla_len = nla_attr_size(4); 426 *(u32 *) nla_data(nla) = value; 427 428 return len + nla_total_size(4); 429 } 430 431 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt, 432 struct fib_config *cfg) 433 { 434 __be32 addr; 435 int plen; 436 437 memset(cfg, 0, sizeof(*cfg)); 438 cfg->fc_nlinfo.nl_net = net; 439 440 if (rt->rt_dst.sa_family != AF_INET) 441 return -EAFNOSUPPORT; 442 443 /* 444 * Check mask for validity: 445 * a) it must be contiguous. 446 * b) destination must have all host bits clear. 447 * c) if application forgot to set correct family (AF_INET), 448 * reject request unless it is absolutely clear i.e. 449 * both family and mask are zero. 450 */ 451 plen = 32; 452 addr = sk_extract_addr(&rt->rt_dst); 453 if (!(rt->rt_flags & RTF_HOST)) { 454 __be32 mask = sk_extract_addr(&rt->rt_genmask); 455 456 if (rt->rt_genmask.sa_family != AF_INET) { 457 if (mask || rt->rt_genmask.sa_family) 458 return -EAFNOSUPPORT; 459 } 460 461 if (bad_mask(mask, addr)) 462 return -EINVAL; 463 464 plen = inet_mask_len(mask); 465 } 466 467 cfg->fc_dst_len = plen; 468 cfg->fc_dst = addr; 469 470 if (cmd != SIOCDELRT) { 471 cfg->fc_nlflags = NLM_F_CREATE; 472 cfg->fc_protocol = RTPROT_BOOT; 473 } 474 475 if (rt->rt_metric) 476 cfg->fc_priority = rt->rt_metric - 1; 477 478 if (rt->rt_flags & RTF_REJECT) { 479 cfg->fc_scope = RT_SCOPE_HOST; 480 cfg->fc_type = RTN_UNREACHABLE; 481 return 0; 482 } 483 484 cfg->fc_scope = RT_SCOPE_NOWHERE; 485 cfg->fc_type = RTN_UNICAST; 486 487 if (rt->rt_dev) { 488 char *colon; 489 struct net_device *dev; 490 char devname[IFNAMSIZ]; 491 492 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1)) 493 return -EFAULT; 494 495 devname[IFNAMSIZ-1] = 0; 496 colon = strchr(devname, ':'); 497 if (colon) 498 *colon = 0; 499 dev = __dev_get_by_name(net, devname); 500 if (!dev) 501 return -ENODEV; 502 cfg->fc_oif = dev->ifindex; 503 cfg->fc_table = l3mdev_fib_table(dev); 504 if (colon) { 505 struct in_ifaddr *ifa; 506 struct in_device *in_dev = __in_dev_get_rtnl(dev); 507 if (!in_dev) 508 return -ENODEV; 509 *colon = ':'; 510 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) 511 if (strcmp(ifa->ifa_label, devname) == 0) 512 break; 513 if (!ifa) 514 return -ENODEV; 515 cfg->fc_prefsrc = ifa->ifa_local; 516 } 517 } 518 519 addr = sk_extract_addr(&rt->rt_gateway); 520 if (rt->rt_gateway.sa_family == AF_INET && addr) { 521 unsigned int addr_type; 522 523 cfg->fc_gw = addr; 524 addr_type = inet_addr_type_table(net, addr, cfg->fc_table); 525 if (rt->rt_flags & RTF_GATEWAY && 526 addr_type == RTN_UNICAST) 527 cfg->fc_scope = RT_SCOPE_UNIVERSE; 528 } 529 530 if (cmd == SIOCDELRT) 531 return 0; 532 533 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw) 534 return -EINVAL; 535 536 if (cfg->fc_scope == RT_SCOPE_NOWHERE) 537 cfg->fc_scope = RT_SCOPE_LINK; 538 539 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) { 540 struct nlattr *mx; 541 int len = 0; 542 543 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL); 544 if (!mx) 545 return -ENOMEM; 546 547 if (rt->rt_flags & RTF_MTU) 548 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40); 549 550 if (rt->rt_flags & RTF_WINDOW) 551 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window); 552 553 if (rt->rt_flags & RTF_IRTT) 554 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3); 555 556 cfg->fc_mx = mx; 557 cfg->fc_mx_len = len; 558 } 559 560 return 0; 561 } 562 563 /* 564 * Handle IP routing ioctl calls. 565 * These are used to manipulate the routing tables 566 */ 567 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg) 568 { 569 struct fib_config cfg; 570 struct rtentry rt; 571 int err; 572 573 switch (cmd) { 574 case SIOCADDRT: /* Add a route */ 575 case SIOCDELRT: /* Delete a route */ 576 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 577 return -EPERM; 578 579 if (copy_from_user(&rt, arg, sizeof(rt))) 580 return -EFAULT; 581 582 rtnl_lock(); 583 err = rtentry_to_fib_config(net, cmd, &rt, &cfg); 584 if (err == 0) { 585 struct fib_table *tb; 586 587 if (cmd == SIOCDELRT) { 588 tb = fib_get_table(net, cfg.fc_table); 589 if (tb) 590 err = fib_table_delete(net, tb, &cfg); 591 else 592 err = -ESRCH; 593 } else { 594 tb = fib_new_table(net, cfg.fc_table); 595 if (tb) 596 err = fib_table_insert(net, tb, &cfg); 597 else 598 err = -ENOBUFS; 599 } 600 601 /* allocated by rtentry_to_fib_config() */ 602 kfree(cfg.fc_mx); 603 } 604 rtnl_unlock(); 605 return err; 606 } 607 return -EINVAL; 608 } 609 610 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = { 611 [RTA_DST] = { .type = NLA_U32 }, 612 [RTA_SRC] = { .type = NLA_U32 }, 613 [RTA_IIF] = { .type = NLA_U32 }, 614 [RTA_OIF] = { .type = NLA_U32 }, 615 [RTA_GATEWAY] = { .type = NLA_U32 }, 616 [RTA_PRIORITY] = { .type = NLA_U32 }, 617 [RTA_PREFSRC] = { .type = NLA_U32 }, 618 [RTA_METRICS] = { .type = NLA_NESTED }, 619 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) }, 620 [RTA_FLOW] = { .type = NLA_U32 }, 621 [RTA_ENCAP_TYPE] = { .type = NLA_U16 }, 622 [RTA_ENCAP] = { .type = NLA_NESTED }, 623 [RTA_UID] = { .type = NLA_U32 }, 624 }; 625 626 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb, 627 struct nlmsghdr *nlh, struct fib_config *cfg) 628 { 629 struct nlattr *attr; 630 int err, remaining; 631 struct rtmsg *rtm; 632 633 err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy); 634 if (err < 0) 635 goto errout; 636 637 memset(cfg, 0, sizeof(*cfg)); 638 639 rtm = nlmsg_data(nlh); 640 cfg->fc_dst_len = rtm->rtm_dst_len; 641 cfg->fc_tos = rtm->rtm_tos; 642 cfg->fc_table = rtm->rtm_table; 643 cfg->fc_protocol = rtm->rtm_protocol; 644 cfg->fc_scope = rtm->rtm_scope; 645 cfg->fc_type = rtm->rtm_type; 646 cfg->fc_flags = rtm->rtm_flags; 647 cfg->fc_nlflags = nlh->nlmsg_flags; 648 649 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid; 650 cfg->fc_nlinfo.nlh = nlh; 651 cfg->fc_nlinfo.nl_net = net; 652 653 if (cfg->fc_type > RTN_MAX) { 654 err = -EINVAL; 655 goto errout; 656 } 657 658 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) { 659 switch (nla_type(attr)) { 660 case RTA_DST: 661 cfg->fc_dst = nla_get_be32(attr); 662 break; 663 case RTA_OIF: 664 cfg->fc_oif = nla_get_u32(attr); 665 break; 666 case RTA_GATEWAY: 667 cfg->fc_gw = nla_get_be32(attr); 668 break; 669 case RTA_PRIORITY: 670 cfg->fc_priority = nla_get_u32(attr); 671 break; 672 case RTA_PREFSRC: 673 cfg->fc_prefsrc = nla_get_be32(attr); 674 break; 675 case RTA_METRICS: 676 cfg->fc_mx = nla_data(attr); 677 cfg->fc_mx_len = nla_len(attr); 678 break; 679 case RTA_MULTIPATH: 680 cfg->fc_mp = nla_data(attr); 681 cfg->fc_mp_len = nla_len(attr); 682 break; 683 case RTA_FLOW: 684 cfg->fc_flow = nla_get_u32(attr); 685 break; 686 case RTA_TABLE: 687 cfg->fc_table = nla_get_u32(attr); 688 break; 689 case RTA_ENCAP: 690 cfg->fc_encap = attr; 691 break; 692 case RTA_ENCAP_TYPE: 693 cfg->fc_encap_type = nla_get_u16(attr); 694 break; 695 } 696 } 697 698 return 0; 699 errout: 700 return err; 701 } 702 703 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh) 704 { 705 struct net *net = sock_net(skb->sk); 706 struct fib_config cfg; 707 struct fib_table *tb; 708 int err; 709 710 err = rtm_to_fib_config(net, skb, nlh, &cfg); 711 if (err < 0) 712 goto errout; 713 714 tb = fib_get_table(net, cfg.fc_table); 715 if (!tb) { 716 err = -ESRCH; 717 goto errout; 718 } 719 720 err = fib_table_delete(net, tb, &cfg); 721 errout: 722 return err; 723 } 724 725 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh) 726 { 727 struct net *net = sock_net(skb->sk); 728 struct fib_config cfg; 729 struct fib_table *tb; 730 int err; 731 732 err = rtm_to_fib_config(net, skb, nlh, &cfg); 733 if (err < 0) 734 goto errout; 735 736 tb = fib_new_table(net, cfg.fc_table); 737 if (!tb) { 738 err = -ENOBUFS; 739 goto errout; 740 } 741 742 err = fib_table_insert(net, tb, &cfg); 743 errout: 744 return err; 745 } 746 747 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 748 { 749 struct net *net = sock_net(skb->sk); 750 unsigned int h, s_h; 751 unsigned int e = 0, s_e; 752 struct fib_table *tb; 753 struct hlist_head *head; 754 int dumped = 0; 755 756 if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) && 757 ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED) 758 return skb->len; 759 760 s_h = cb->args[0]; 761 s_e = cb->args[1]; 762 763 rcu_read_lock(); 764 765 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) { 766 e = 0; 767 head = &net->ipv4.fib_table_hash[h]; 768 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 769 if (e < s_e) 770 goto next; 771 if (dumped) 772 memset(&cb->args[2], 0, sizeof(cb->args) - 773 2 * sizeof(cb->args[0])); 774 if (fib_table_dump(tb, skb, cb) < 0) 775 goto out; 776 dumped = 1; 777 next: 778 e++; 779 } 780 } 781 out: 782 rcu_read_unlock(); 783 784 cb->args[1] = e; 785 cb->args[0] = h; 786 787 return skb->len; 788 } 789 790 /* Prepare and feed intra-kernel routing request. 791 * Really, it should be netlink message, but :-( netlink 792 * can be not configured, so that we feed it directly 793 * to fib engine. It is legal, because all events occur 794 * only when netlink is already locked. 795 */ 796 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa) 797 { 798 struct net *net = dev_net(ifa->ifa_dev->dev); 799 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev); 800 struct fib_table *tb; 801 struct fib_config cfg = { 802 .fc_protocol = RTPROT_KERNEL, 803 .fc_type = type, 804 .fc_dst = dst, 805 .fc_dst_len = dst_len, 806 .fc_prefsrc = ifa->ifa_local, 807 .fc_oif = ifa->ifa_dev->dev->ifindex, 808 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND, 809 .fc_nlinfo = { 810 .nl_net = net, 811 }, 812 }; 813 814 if (!tb_id) 815 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL; 816 817 tb = fib_new_table(net, tb_id); 818 if (!tb) 819 return; 820 821 cfg.fc_table = tb->tb_id; 822 823 if (type != RTN_LOCAL) 824 cfg.fc_scope = RT_SCOPE_LINK; 825 else 826 cfg.fc_scope = RT_SCOPE_HOST; 827 828 if (cmd == RTM_NEWROUTE) 829 fib_table_insert(net, tb, &cfg); 830 else 831 fib_table_delete(net, tb, &cfg); 832 } 833 834 void fib_add_ifaddr(struct in_ifaddr *ifa) 835 { 836 struct in_device *in_dev = ifa->ifa_dev; 837 struct net_device *dev = in_dev->dev; 838 struct in_ifaddr *prim = ifa; 839 __be32 mask = ifa->ifa_mask; 840 __be32 addr = ifa->ifa_local; 841 __be32 prefix = ifa->ifa_address & mask; 842 843 if (ifa->ifa_flags & IFA_F_SECONDARY) { 844 prim = inet_ifa_byprefix(in_dev, prefix, mask); 845 if (!prim) { 846 pr_warn("%s: bug: prim == NULL\n", __func__); 847 return; 848 } 849 } 850 851 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim); 852 853 if (!(dev->flags & IFF_UP)) 854 return; 855 856 /* Add broadcast address, if it is explicitly assigned. */ 857 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) 858 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 859 860 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) && 861 (prefix != addr || ifa->ifa_prefixlen < 32)) { 862 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 863 fib_magic(RTM_NEWROUTE, 864 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 865 prefix, ifa->ifa_prefixlen, prim); 866 867 /* Add network specific broadcasts, when it takes a sense */ 868 if (ifa->ifa_prefixlen < 31) { 869 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim); 870 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask, 871 32, prim); 872 } 873 } 874 } 875 876 /* Delete primary or secondary address. 877 * Optionally, on secondary address promotion consider the addresses 878 * from subnet iprim as deleted, even if they are in device list. 879 * In this case the secondary ifa can be in device list. 880 */ 881 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim) 882 { 883 struct in_device *in_dev = ifa->ifa_dev; 884 struct net_device *dev = in_dev->dev; 885 struct in_ifaddr *ifa1; 886 struct in_ifaddr *prim = ifa, *prim1 = NULL; 887 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask; 888 __be32 any = ifa->ifa_address & ifa->ifa_mask; 889 #define LOCAL_OK 1 890 #define BRD_OK 2 891 #define BRD0_OK 4 892 #define BRD1_OK 8 893 unsigned int ok = 0; 894 int subnet = 0; /* Primary network */ 895 int gone = 1; /* Address is missing */ 896 int same_prefsrc = 0; /* Another primary with same IP */ 897 898 if (ifa->ifa_flags & IFA_F_SECONDARY) { 899 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask); 900 if (!prim) { 901 /* if the device has been deleted, we don't perform 902 * address promotion 903 */ 904 if (!in_dev->dead) 905 pr_warn("%s: bug: prim == NULL\n", __func__); 906 return; 907 } 908 if (iprim && iprim != prim) { 909 pr_warn("%s: bug: iprim != prim\n", __func__); 910 return; 911 } 912 } else if (!ipv4_is_zeronet(any) && 913 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) { 914 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 915 fib_magic(RTM_DELROUTE, 916 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 917 any, ifa->ifa_prefixlen, prim); 918 subnet = 1; 919 } 920 921 if (in_dev->dead) 922 goto no_promotions; 923 924 /* Deletion is more complicated than add. 925 * We should take care of not to delete too much :-) 926 * 927 * Scan address list to be sure that addresses are really gone. 928 */ 929 930 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) { 931 if (ifa1 == ifa) { 932 /* promotion, keep the IP */ 933 gone = 0; 934 continue; 935 } 936 /* Ignore IFAs from our subnet */ 937 if (iprim && ifa1->ifa_mask == iprim->ifa_mask && 938 inet_ifa_match(ifa1->ifa_address, iprim)) 939 continue; 940 941 /* Ignore ifa1 if it uses different primary IP (prefsrc) */ 942 if (ifa1->ifa_flags & IFA_F_SECONDARY) { 943 /* Another address from our subnet? */ 944 if (ifa1->ifa_mask == prim->ifa_mask && 945 inet_ifa_match(ifa1->ifa_address, prim)) 946 prim1 = prim; 947 else { 948 /* We reached the secondaries, so 949 * same_prefsrc should be determined. 950 */ 951 if (!same_prefsrc) 952 continue; 953 /* Search new prim1 if ifa1 is not 954 * using the current prim1 955 */ 956 if (!prim1 || 957 ifa1->ifa_mask != prim1->ifa_mask || 958 !inet_ifa_match(ifa1->ifa_address, prim1)) 959 prim1 = inet_ifa_byprefix(in_dev, 960 ifa1->ifa_address, 961 ifa1->ifa_mask); 962 if (!prim1) 963 continue; 964 if (prim1->ifa_local != prim->ifa_local) 965 continue; 966 } 967 } else { 968 if (prim->ifa_local != ifa1->ifa_local) 969 continue; 970 prim1 = ifa1; 971 if (prim != prim1) 972 same_prefsrc = 1; 973 } 974 if (ifa->ifa_local == ifa1->ifa_local) 975 ok |= LOCAL_OK; 976 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 977 ok |= BRD_OK; 978 if (brd == ifa1->ifa_broadcast) 979 ok |= BRD1_OK; 980 if (any == ifa1->ifa_broadcast) 981 ok |= BRD0_OK; 982 /* primary has network specific broadcasts */ 983 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) { 984 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask; 985 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask; 986 987 if (!ipv4_is_zeronet(any1)) { 988 if (ifa->ifa_broadcast == brd1 || 989 ifa->ifa_broadcast == any1) 990 ok |= BRD_OK; 991 if (brd == brd1 || brd == any1) 992 ok |= BRD1_OK; 993 if (any == brd1 || any == any1) 994 ok |= BRD0_OK; 995 } 996 } 997 } 998 999 no_promotions: 1000 if (!(ok & BRD_OK)) 1001 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 1002 if (subnet && ifa->ifa_prefixlen < 31) { 1003 if (!(ok & BRD1_OK)) 1004 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim); 1005 if (!(ok & BRD0_OK)) 1006 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim); 1007 } 1008 if (!(ok & LOCAL_OK)) { 1009 unsigned int addr_type; 1010 1011 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim); 1012 1013 /* Check, that this local address finally disappeared. */ 1014 addr_type = inet_addr_type_dev_table(dev_net(dev), dev, 1015 ifa->ifa_local); 1016 if (gone && addr_type != RTN_LOCAL) { 1017 /* And the last, but not the least thing. 1018 * We must flush stray FIB entries. 1019 * 1020 * First of all, we scan fib_info list searching 1021 * for stray nexthop entries, then ignite fib_flush. 1022 */ 1023 if (fib_sync_down_addr(dev, ifa->ifa_local)) 1024 fib_flush(dev_net(dev)); 1025 } 1026 } 1027 #undef LOCAL_OK 1028 #undef BRD_OK 1029 #undef BRD0_OK 1030 #undef BRD1_OK 1031 } 1032 1033 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn) 1034 { 1035 1036 struct fib_result res; 1037 struct flowi4 fl4 = { 1038 .flowi4_mark = frn->fl_mark, 1039 .daddr = frn->fl_addr, 1040 .flowi4_tos = frn->fl_tos, 1041 .flowi4_scope = frn->fl_scope, 1042 }; 1043 struct fib_table *tb; 1044 1045 rcu_read_lock(); 1046 1047 tb = fib_get_table(net, frn->tb_id_in); 1048 1049 frn->err = -ENOENT; 1050 if (tb) { 1051 local_bh_disable(); 1052 1053 frn->tb_id = tb->tb_id; 1054 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 1055 1056 if (!frn->err) { 1057 frn->prefixlen = res.prefixlen; 1058 frn->nh_sel = res.nh_sel; 1059 frn->type = res.type; 1060 frn->scope = res.scope; 1061 } 1062 local_bh_enable(); 1063 } 1064 1065 rcu_read_unlock(); 1066 } 1067 1068 static void nl_fib_input(struct sk_buff *skb) 1069 { 1070 struct net *net; 1071 struct fib_result_nl *frn; 1072 struct nlmsghdr *nlh; 1073 u32 portid; 1074 1075 net = sock_net(skb->sk); 1076 nlh = nlmsg_hdr(skb); 1077 if (skb->len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len || 1078 nlmsg_len(nlh) < sizeof(*frn)) 1079 return; 1080 1081 skb = netlink_skb_clone(skb, GFP_KERNEL); 1082 if (!skb) 1083 return; 1084 nlh = nlmsg_hdr(skb); 1085 1086 frn = (struct fib_result_nl *) nlmsg_data(nlh); 1087 nl_fib_lookup(net, frn); 1088 1089 portid = NETLINK_CB(skb).portid; /* netlink portid */ 1090 NETLINK_CB(skb).portid = 0; /* from kernel */ 1091 NETLINK_CB(skb).dst_group = 0; /* unicast */ 1092 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT); 1093 } 1094 1095 static int __net_init nl_fib_lookup_init(struct net *net) 1096 { 1097 struct sock *sk; 1098 struct netlink_kernel_cfg cfg = { 1099 .input = nl_fib_input, 1100 }; 1101 1102 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg); 1103 if (!sk) 1104 return -EAFNOSUPPORT; 1105 net->ipv4.fibnl = sk; 1106 return 0; 1107 } 1108 1109 static void nl_fib_lookup_exit(struct net *net) 1110 { 1111 netlink_kernel_release(net->ipv4.fibnl); 1112 net->ipv4.fibnl = NULL; 1113 } 1114 1115 static void fib_disable_ip(struct net_device *dev, unsigned long event, 1116 bool force) 1117 { 1118 if (fib_sync_down_dev(dev, event, force)) 1119 fib_flush(dev_net(dev)); 1120 rt_cache_flush(dev_net(dev)); 1121 arp_ifdown(dev); 1122 } 1123 1124 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 1125 { 1126 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; 1127 struct net_device *dev = ifa->ifa_dev->dev; 1128 struct net *net = dev_net(dev); 1129 1130 switch (event) { 1131 case NETDEV_UP: 1132 fib_add_ifaddr(ifa); 1133 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1134 fib_sync_up(dev, RTNH_F_DEAD); 1135 #endif 1136 atomic_inc(&net->ipv4.dev_addr_genid); 1137 rt_cache_flush(dev_net(dev)); 1138 break; 1139 case NETDEV_DOWN: 1140 fib_del_ifaddr(ifa, NULL); 1141 atomic_inc(&net->ipv4.dev_addr_genid); 1142 if (!ifa->ifa_dev->ifa_list) { 1143 /* Last address was deleted from this interface. 1144 * Disable IP. 1145 */ 1146 fib_disable_ip(dev, event, true); 1147 } else { 1148 rt_cache_flush(dev_net(dev)); 1149 } 1150 break; 1151 } 1152 return NOTIFY_DONE; 1153 } 1154 1155 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 1156 { 1157 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1158 struct netdev_notifier_changeupper_info *info; 1159 struct in_device *in_dev; 1160 struct net *net = dev_net(dev); 1161 unsigned int flags; 1162 1163 if (event == NETDEV_UNREGISTER) { 1164 fib_disable_ip(dev, event, true); 1165 rt_flush_dev(dev); 1166 return NOTIFY_DONE; 1167 } 1168 1169 in_dev = __in_dev_get_rtnl(dev); 1170 if (!in_dev) 1171 return NOTIFY_DONE; 1172 1173 switch (event) { 1174 case NETDEV_UP: 1175 for_ifa(in_dev) { 1176 fib_add_ifaddr(ifa); 1177 } endfor_ifa(in_dev); 1178 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1179 fib_sync_up(dev, RTNH_F_DEAD); 1180 #endif 1181 atomic_inc(&net->ipv4.dev_addr_genid); 1182 rt_cache_flush(net); 1183 break; 1184 case NETDEV_DOWN: 1185 fib_disable_ip(dev, event, false); 1186 break; 1187 case NETDEV_CHANGE: 1188 flags = dev_get_flags(dev); 1189 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 1190 fib_sync_up(dev, RTNH_F_LINKDOWN); 1191 else 1192 fib_sync_down_dev(dev, event, false); 1193 /* fall through */ 1194 case NETDEV_CHANGEMTU: 1195 rt_cache_flush(net); 1196 break; 1197 case NETDEV_CHANGEUPPER: 1198 info = ptr; 1199 /* flush all routes if dev is linked to or unlinked from 1200 * an L3 master device (e.g., VRF) 1201 */ 1202 if (info->upper_dev && netif_is_l3_master(info->upper_dev)) 1203 fib_disable_ip(dev, NETDEV_DOWN, true); 1204 break; 1205 } 1206 return NOTIFY_DONE; 1207 } 1208 1209 static struct notifier_block fib_inetaddr_notifier = { 1210 .notifier_call = fib_inetaddr_event, 1211 }; 1212 1213 static struct notifier_block fib_netdev_notifier = { 1214 .notifier_call = fib_netdev_event, 1215 }; 1216 1217 static int __net_init ip_fib_net_init(struct net *net) 1218 { 1219 int err; 1220 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ; 1221 1222 net->ipv4.fib_seq = 0; 1223 1224 /* Avoid false sharing : Use at least a full cache line */ 1225 size = max_t(size_t, size, L1_CACHE_BYTES); 1226 1227 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL); 1228 if (!net->ipv4.fib_table_hash) 1229 return -ENOMEM; 1230 1231 err = fib4_rules_init(net); 1232 if (err < 0) 1233 goto fail; 1234 return 0; 1235 1236 fail: 1237 kfree(net->ipv4.fib_table_hash); 1238 return err; 1239 } 1240 1241 static void ip_fib_net_exit(struct net *net) 1242 { 1243 unsigned int i; 1244 1245 rtnl_lock(); 1246 #ifdef CONFIG_IP_MULTIPLE_TABLES 1247 RCU_INIT_POINTER(net->ipv4.fib_main, NULL); 1248 RCU_INIT_POINTER(net->ipv4.fib_default, NULL); 1249 #endif 1250 for (i = 0; i < FIB_TABLE_HASHSZ; i++) { 1251 struct hlist_head *head = &net->ipv4.fib_table_hash[i]; 1252 struct hlist_node *tmp; 1253 struct fib_table *tb; 1254 1255 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) { 1256 hlist_del(&tb->tb_hlist); 1257 fib_table_flush(net, tb); 1258 fib_free_table(tb); 1259 } 1260 } 1261 1262 #ifdef CONFIG_IP_MULTIPLE_TABLES 1263 fib4_rules_exit(net); 1264 #endif 1265 rtnl_unlock(); 1266 kfree(net->ipv4.fib_table_hash); 1267 } 1268 1269 static int __net_init fib_net_init(struct net *net) 1270 { 1271 int error; 1272 1273 #ifdef CONFIG_IP_ROUTE_CLASSID 1274 net->ipv4.fib_num_tclassid_users = 0; 1275 #endif 1276 error = ip_fib_net_init(net); 1277 if (error < 0) 1278 goto out; 1279 error = nl_fib_lookup_init(net); 1280 if (error < 0) 1281 goto out_nlfl; 1282 error = fib_proc_init(net); 1283 if (error < 0) 1284 goto out_proc; 1285 out: 1286 return error; 1287 1288 out_proc: 1289 nl_fib_lookup_exit(net); 1290 out_nlfl: 1291 ip_fib_net_exit(net); 1292 goto out; 1293 } 1294 1295 static void __net_exit fib_net_exit(struct net *net) 1296 { 1297 fib_proc_exit(net); 1298 nl_fib_lookup_exit(net); 1299 ip_fib_net_exit(net); 1300 } 1301 1302 static struct pernet_operations fib_net_ops = { 1303 .init = fib_net_init, 1304 .exit = fib_net_exit, 1305 }; 1306 1307 void __init ip_fib_init(void) 1308 { 1309 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL); 1310 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL); 1311 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL); 1312 1313 register_pernet_subsys(&fib_net_ops); 1314 register_netdevice_notifier(&fib_netdev_notifier); 1315 register_inetaddr_notifier(&fib_inetaddr_notifier); 1316 1317 fib_trie_init(); 1318 } 1319