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 NULL); 593 else 594 err = -ESRCH; 595 } else { 596 tb = fib_new_table(net, cfg.fc_table); 597 if (tb) 598 err = fib_table_insert(net, tb, 599 &cfg, NULL); 600 else 601 err = -ENOBUFS; 602 } 603 604 /* allocated by rtentry_to_fib_config() */ 605 kfree(cfg.fc_mx); 606 } 607 rtnl_unlock(); 608 return err; 609 } 610 return -EINVAL; 611 } 612 613 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = { 614 [RTA_DST] = { .type = NLA_U32 }, 615 [RTA_SRC] = { .type = NLA_U32 }, 616 [RTA_IIF] = { .type = NLA_U32 }, 617 [RTA_OIF] = { .type = NLA_U32 }, 618 [RTA_GATEWAY] = { .type = NLA_U32 }, 619 [RTA_PRIORITY] = { .type = NLA_U32 }, 620 [RTA_PREFSRC] = { .type = NLA_U32 }, 621 [RTA_METRICS] = { .type = NLA_NESTED }, 622 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) }, 623 [RTA_FLOW] = { .type = NLA_U32 }, 624 [RTA_ENCAP_TYPE] = { .type = NLA_U16 }, 625 [RTA_ENCAP] = { .type = NLA_NESTED }, 626 [RTA_UID] = { .type = NLA_U32 }, 627 [RTA_MARK] = { .type = NLA_U32 }, 628 }; 629 EXPORT_SYMBOL_GPL(rtm_ipv4_policy); 630 631 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb, 632 struct nlmsghdr *nlh, struct fib_config *cfg, 633 struct netlink_ext_ack *extack) 634 { 635 struct nlattr *attr; 636 int err, remaining; 637 struct rtmsg *rtm; 638 639 err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy, 640 extack); 641 if (err < 0) 642 goto errout; 643 644 memset(cfg, 0, sizeof(*cfg)); 645 646 rtm = nlmsg_data(nlh); 647 cfg->fc_dst_len = rtm->rtm_dst_len; 648 cfg->fc_tos = rtm->rtm_tos; 649 cfg->fc_table = rtm->rtm_table; 650 cfg->fc_protocol = rtm->rtm_protocol; 651 cfg->fc_scope = rtm->rtm_scope; 652 cfg->fc_type = rtm->rtm_type; 653 cfg->fc_flags = rtm->rtm_flags; 654 cfg->fc_nlflags = nlh->nlmsg_flags; 655 656 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid; 657 cfg->fc_nlinfo.nlh = nlh; 658 cfg->fc_nlinfo.nl_net = net; 659 660 if (cfg->fc_type > RTN_MAX) { 661 NL_SET_ERR_MSG(extack, "Invalid route type"); 662 err = -EINVAL; 663 goto errout; 664 } 665 666 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) { 667 switch (nla_type(attr)) { 668 case RTA_DST: 669 cfg->fc_dst = nla_get_be32(attr); 670 break; 671 case RTA_OIF: 672 cfg->fc_oif = nla_get_u32(attr); 673 break; 674 case RTA_GATEWAY: 675 cfg->fc_gw = nla_get_be32(attr); 676 break; 677 case RTA_PRIORITY: 678 cfg->fc_priority = nla_get_u32(attr); 679 break; 680 case RTA_PREFSRC: 681 cfg->fc_prefsrc = nla_get_be32(attr); 682 break; 683 case RTA_METRICS: 684 cfg->fc_mx = nla_data(attr); 685 cfg->fc_mx_len = nla_len(attr); 686 break; 687 case RTA_MULTIPATH: 688 err = lwtunnel_valid_encap_type_attr(nla_data(attr), 689 nla_len(attr), 690 extack); 691 if (err < 0) 692 goto errout; 693 cfg->fc_mp = nla_data(attr); 694 cfg->fc_mp_len = nla_len(attr); 695 break; 696 case RTA_FLOW: 697 cfg->fc_flow = nla_get_u32(attr); 698 break; 699 case RTA_TABLE: 700 cfg->fc_table = nla_get_u32(attr); 701 break; 702 case RTA_ENCAP: 703 cfg->fc_encap = attr; 704 break; 705 case RTA_ENCAP_TYPE: 706 cfg->fc_encap_type = nla_get_u16(attr); 707 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, 708 extack); 709 if (err < 0) 710 goto errout; 711 break; 712 } 713 } 714 715 return 0; 716 errout: 717 return err; 718 } 719 720 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, 721 struct netlink_ext_ack *extack) 722 { 723 struct net *net = sock_net(skb->sk); 724 struct fib_config cfg; 725 struct fib_table *tb; 726 int err; 727 728 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack); 729 if (err < 0) 730 goto errout; 731 732 tb = fib_get_table(net, cfg.fc_table); 733 if (!tb) { 734 NL_SET_ERR_MSG(extack, "FIB table does not exist"); 735 err = -ESRCH; 736 goto errout; 737 } 738 739 err = fib_table_delete(net, tb, &cfg, extack); 740 errout: 741 return err; 742 } 743 744 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, 745 struct netlink_ext_ack *extack) 746 { 747 struct net *net = sock_net(skb->sk); 748 struct fib_config cfg; 749 struct fib_table *tb; 750 int err; 751 752 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack); 753 if (err < 0) 754 goto errout; 755 756 tb = fib_new_table(net, cfg.fc_table); 757 if (!tb) { 758 err = -ENOBUFS; 759 goto errout; 760 } 761 762 err = fib_table_insert(net, tb, &cfg, extack); 763 errout: 764 return err; 765 } 766 767 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 768 { 769 struct net *net = sock_net(skb->sk); 770 unsigned int h, s_h; 771 unsigned int e = 0, s_e; 772 struct fib_table *tb; 773 struct hlist_head *head; 774 int dumped = 0, err; 775 776 if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) && 777 ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED) 778 return skb->len; 779 780 s_h = cb->args[0]; 781 s_e = cb->args[1]; 782 783 rcu_read_lock(); 784 785 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) { 786 e = 0; 787 head = &net->ipv4.fib_table_hash[h]; 788 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 789 if (e < s_e) 790 goto next; 791 if (dumped) 792 memset(&cb->args[2], 0, sizeof(cb->args) - 793 2 * sizeof(cb->args[0])); 794 err = fib_table_dump(tb, skb, cb); 795 if (err < 0) { 796 if (likely(skb->len)) 797 goto out; 798 799 goto out_err; 800 } 801 dumped = 1; 802 next: 803 e++; 804 } 805 } 806 out: 807 err = skb->len; 808 out_err: 809 rcu_read_unlock(); 810 811 cb->args[1] = e; 812 cb->args[0] = h; 813 814 return err; 815 } 816 817 /* Prepare and feed intra-kernel routing request. 818 * Really, it should be netlink message, but :-( netlink 819 * can be not configured, so that we feed it directly 820 * to fib engine. It is legal, because all events occur 821 * only when netlink is already locked. 822 */ 823 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa) 824 { 825 struct net *net = dev_net(ifa->ifa_dev->dev); 826 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev); 827 struct fib_table *tb; 828 struct fib_config cfg = { 829 .fc_protocol = RTPROT_KERNEL, 830 .fc_type = type, 831 .fc_dst = dst, 832 .fc_dst_len = dst_len, 833 .fc_prefsrc = ifa->ifa_local, 834 .fc_oif = ifa->ifa_dev->dev->ifindex, 835 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND, 836 .fc_nlinfo = { 837 .nl_net = net, 838 }, 839 }; 840 841 if (!tb_id) 842 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL; 843 844 tb = fib_new_table(net, tb_id); 845 if (!tb) 846 return; 847 848 cfg.fc_table = tb->tb_id; 849 850 if (type != RTN_LOCAL) 851 cfg.fc_scope = RT_SCOPE_LINK; 852 else 853 cfg.fc_scope = RT_SCOPE_HOST; 854 855 if (cmd == RTM_NEWROUTE) 856 fib_table_insert(net, tb, &cfg, NULL); 857 else 858 fib_table_delete(net, tb, &cfg, NULL); 859 } 860 861 void fib_add_ifaddr(struct in_ifaddr *ifa) 862 { 863 struct in_device *in_dev = ifa->ifa_dev; 864 struct net_device *dev = in_dev->dev; 865 struct in_ifaddr *prim = ifa; 866 __be32 mask = ifa->ifa_mask; 867 __be32 addr = ifa->ifa_local; 868 __be32 prefix = ifa->ifa_address & mask; 869 870 if (ifa->ifa_flags & IFA_F_SECONDARY) { 871 prim = inet_ifa_byprefix(in_dev, prefix, mask); 872 if (!prim) { 873 pr_warn("%s: bug: prim == NULL\n", __func__); 874 return; 875 } 876 } 877 878 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim); 879 880 if (!(dev->flags & IFF_UP)) 881 return; 882 883 /* Add broadcast address, if it is explicitly assigned. */ 884 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) 885 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 886 887 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) && 888 (prefix != addr || ifa->ifa_prefixlen < 32)) { 889 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 890 fib_magic(RTM_NEWROUTE, 891 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 892 prefix, ifa->ifa_prefixlen, prim); 893 894 /* Add network specific broadcasts, when it takes a sense */ 895 if (ifa->ifa_prefixlen < 31) { 896 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim); 897 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask, 898 32, prim); 899 } 900 } 901 } 902 903 /* Delete primary or secondary address. 904 * Optionally, on secondary address promotion consider the addresses 905 * from subnet iprim as deleted, even if they are in device list. 906 * In this case the secondary ifa can be in device list. 907 */ 908 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim) 909 { 910 struct in_device *in_dev = ifa->ifa_dev; 911 struct net_device *dev = in_dev->dev; 912 struct in_ifaddr *ifa1; 913 struct in_ifaddr *prim = ifa, *prim1 = NULL; 914 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask; 915 __be32 any = ifa->ifa_address & ifa->ifa_mask; 916 #define LOCAL_OK 1 917 #define BRD_OK 2 918 #define BRD0_OK 4 919 #define BRD1_OK 8 920 unsigned int ok = 0; 921 int subnet = 0; /* Primary network */ 922 int gone = 1; /* Address is missing */ 923 int same_prefsrc = 0; /* Another primary with same IP */ 924 925 if (ifa->ifa_flags & IFA_F_SECONDARY) { 926 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask); 927 if (!prim) { 928 /* if the device has been deleted, we don't perform 929 * address promotion 930 */ 931 if (!in_dev->dead) 932 pr_warn("%s: bug: prim == NULL\n", __func__); 933 return; 934 } 935 if (iprim && iprim != prim) { 936 pr_warn("%s: bug: iprim != prim\n", __func__); 937 return; 938 } 939 } else if (!ipv4_is_zeronet(any) && 940 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) { 941 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 942 fib_magic(RTM_DELROUTE, 943 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 944 any, ifa->ifa_prefixlen, prim); 945 subnet = 1; 946 } 947 948 if (in_dev->dead) 949 goto no_promotions; 950 951 /* Deletion is more complicated than add. 952 * We should take care of not to delete too much :-) 953 * 954 * Scan address list to be sure that addresses are really gone. 955 */ 956 957 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) { 958 if (ifa1 == ifa) { 959 /* promotion, keep the IP */ 960 gone = 0; 961 continue; 962 } 963 /* Ignore IFAs from our subnet */ 964 if (iprim && ifa1->ifa_mask == iprim->ifa_mask && 965 inet_ifa_match(ifa1->ifa_address, iprim)) 966 continue; 967 968 /* Ignore ifa1 if it uses different primary IP (prefsrc) */ 969 if (ifa1->ifa_flags & IFA_F_SECONDARY) { 970 /* Another address from our subnet? */ 971 if (ifa1->ifa_mask == prim->ifa_mask && 972 inet_ifa_match(ifa1->ifa_address, prim)) 973 prim1 = prim; 974 else { 975 /* We reached the secondaries, so 976 * same_prefsrc should be determined. 977 */ 978 if (!same_prefsrc) 979 continue; 980 /* Search new prim1 if ifa1 is not 981 * using the current prim1 982 */ 983 if (!prim1 || 984 ifa1->ifa_mask != prim1->ifa_mask || 985 !inet_ifa_match(ifa1->ifa_address, prim1)) 986 prim1 = inet_ifa_byprefix(in_dev, 987 ifa1->ifa_address, 988 ifa1->ifa_mask); 989 if (!prim1) 990 continue; 991 if (prim1->ifa_local != prim->ifa_local) 992 continue; 993 } 994 } else { 995 if (prim->ifa_local != ifa1->ifa_local) 996 continue; 997 prim1 = ifa1; 998 if (prim != prim1) 999 same_prefsrc = 1; 1000 } 1001 if (ifa->ifa_local == ifa1->ifa_local) 1002 ok |= LOCAL_OK; 1003 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 1004 ok |= BRD_OK; 1005 if (brd == ifa1->ifa_broadcast) 1006 ok |= BRD1_OK; 1007 if (any == ifa1->ifa_broadcast) 1008 ok |= BRD0_OK; 1009 /* primary has network specific broadcasts */ 1010 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) { 1011 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask; 1012 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask; 1013 1014 if (!ipv4_is_zeronet(any1)) { 1015 if (ifa->ifa_broadcast == brd1 || 1016 ifa->ifa_broadcast == any1) 1017 ok |= BRD_OK; 1018 if (brd == brd1 || brd == any1) 1019 ok |= BRD1_OK; 1020 if (any == brd1 || any == any1) 1021 ok |= BRD0_OK; 1022 } 1023 } 1024 } 1025 1026 no_promotions: 1027 if (!(ok & BRD_OK)) 1028 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 1029 if (subnet && ifa->ifa_prefixlen < 31) { 1030 if (!(ok & BRD1_OK)) 1031 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim); 1032 if (!(ok & BRD0_OK)) 1033 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim); 1034 } 1035 if (!(ok & LOCAL_OK)) { 1036 unsigned int addr_type; 1037 1038 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim); 1039 1040 /* Check, that this local address finally disappeared. */ 1041 addr_type = inet_addr_type_dev_table(dev_net(dev), dev, 1042 ifa->ifa_local); 1043 if (gone && addr_type != RTN_LOCAL) { 1044 /* And the last, but not the least thing. 1045 * We must flush stray FIB entries. 1046 * 1047 * First of all, we scan fib_info list searching 1048 * for stray nexthop entries, then ignite fib_flush. 1049 */ 1050 if (fib_sync_down_addr(dev, ifa->ifa_local)) 1051 fib_flush(dev_net(dev)); 1052 } 1053 } 1054 #undef LOCAL_OK 1055 #undef BRD_OK 1056 #undef BRD0_OK 1057 #undef BRD1_OK 1058 } 1059 1060 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn) 1061 { 1062 1063 struct fib_result res; 1064 struct flowi4 fl4 = { 1065 .flowi4_mark = frn->fl_mark, 1066 .daddr = frn->fl_addr, 1067 .flowi4_tos = frn->fl_tos, 1068 .flowi4_scope = frn->fl_scope, 1069 }; 1070 struct fib_table *tb; 1071 1072 rcu_read_lock(); 1073 1074 tb = fib_get_table(net, frn->tb_id_in); 1075 1076 frn->err = -ENOENT; 1077 if (tb) { 1078 local_bh_disable(); 1079 1080 frn->tb_id = tb->tb_id; 1081 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 1082 1083 if (!frn->err) { 1084 frn->prefixlen = res.prefixlen; 1085 frn->nh_sel = res.nh_sel; 1086 frn->type = res.type; 1087 frn->scope = res.scope; 1088 } 1089 local_bh_enable(); 1090 } 1091 1092 rcu_read_unlock(); 1093 } 1094 1095 static void nl_fib_input(struct sk_buff *skb) 1096 { 1097 struct net *net; 1098 struct fib_result_nl *frn; 1099 struct nlmsghdr *nlh; 1100 u32 portid; 1101 1102 net = sock_net(skb->sk); 1103 nlh = nlmsg_hdr(skb); 1104 if (skb->len < nlmsg_total_size(sizeof(*frn)) || 1105 skb->len < nlh->nlmsg_len || 1106 nlmsg_len(nlh) < sizeof(*frn)) 1107 return; 1108 1109 skb = netlink_skb_clone(skb, GFP_KERNEL); 1110 if (!skb) 1111 return; 1112 nlh = nlmsg_hdr(skb); 1113 1114 frn = (struct fib_result_nl *) nlmsg_data(nlh); 1115 nl_fib_lookup(net, frn); 1116 1117 portid = NETLINK_CB(skb).portid; /* netlink portid */ 1118 NETLINK_CB(skb).portid = 0; /* from kernel */ 1119 NETLINK_CB(skb).dst_group = 0; /* unicast */ 1120 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT); 1121 } 1122 1123 static int __net_init nl_fib_lookup_init(struct net *net) 1124 { 1125 struct sock *sk; 1126 struct netlink_kernel_cfg cfg = { 1127 .input = nl_fib_input, 1128 }; 1129 1130 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg); 1131 if (!sk) 1132 return -EAFNOSUPPORT; 1133 net->ipv4.fibnl = sk; 1134 return 0; 1135 } 1136 1137 static void nl_fib_lookup_exit(struct net *net) 1138 { 1139 netlink_kernel_release(net->ipv4.fibnl); 1140 net->ipv4.fibnl = NULL; 1141 } 1142 1143 static void fib_disable_ip(struct net_device *dev, unsigned long event, 1144 bool force) 1145 { 1146 if (fib_sync_down_dev(dev, event, force)) 1147 fib_flush(dev_net(dev)); 1148 else 1149 rt_cache_flush(dev_net(dev)); 1150 arp_ifdown(dev); 1151 } 1152 1153 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 1154 { 1155 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; 1156 struct net_device *dev = ifa->ifa_dev->dev; 1157 struct net *net = dev_net(dev); 1158 1159 switch (event) { 1160 case NETDEV_UP: 1161 fib_add_ifaddr(ifa); 1162 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1163 fib_sync_up(dev, RTNH_F_DEAD); 1164 #endif 1165 atomic_inc(&net->ipv4.dev_addr_genid); 1166 rt_cache_flush(dev_net(dev)); 1167 break; 1168 case NETDEV_DOWN: 1169 fib_del_ifaddr(ifa, NULL); 1170 atomic_inc(&net->ipv4.dev_addr_genid); 1171 if (!ifa->ifa_dev->ifa_list) { 1172 /* Last address was deleted from this interface. 1173 * Disable IP. 1174 */ 1175 fib_disable_ip(dev, event, true); 1176 } else { 1177 rt_cache_flush(dev_net(dev)); 1178 } 1179 break; 1180 } 1181 return NOTIFY_DONE; 1182 } 1183 1184 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 1185 { 1186 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1187 struct netdev_notifier_changeupper_info *info; 1188 struct in_device *in_dev; 1189 struct net *net = dev_net(dev); 1190 unsigned int flags; 1191 1192 if (event == NETDEV_UNREGISTER) { 1193 fib_disable_ip(dev, event, true); 1194 rt_flush_dev(dev); 1195 return NOTIFY_DONE; 1196 } 1197 1198 in_dev = __in_dev_get_rtnl(dev); 1199 if (!in_dev) 1200 return NOTIFY_DONE; 1201 1202 switch (event) { 1203 case NETDEV_UP: 1204 for_ifa(in_dev) { 1205 fib_add_ifaddr(ifa); 1206 } endfor_ifa(in_dev); 1207 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1208 fib_sync_up(dev, RTNH_F_DEAD); 1209 #endif 1210 atomic_inc(&net->ipv4.dev_addr_genid); 1211 rt_cache_flush(net); 1212 break; 1213 case NETDEV_DOWN: 1214 fib_disable_ip(dev, event, false); 1215 break; 1216 case NETDEV_CHANGE: 1217 flags = dev_get_flags(dev); 1218 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 1219 fib_sync_up(dev, RTNH_F_LINKDOWN); 1220 else 1221 fib_sync_down_dev(dev, event, false); 1222 /* fall through */ 1223 case NETDEV_CHANGEMTU: 1224 rt_cache_flush(net); 1225 break; 1226 case NETDEV_CHANGEUPPER: 1227 info = ptr; 1228 /* flush all routes if dev is linked to or unlinked from 1229 * an L3 master device (e.g., VRF) 1230 */ 1231 if (info->upper_dev && netif_is_l3_master(info->upper_dev)) 1232 fib_disable_ip(dev, NETDEV_DOWN, true); 1233 break; 1234 } 1235 return NOTIFY_DONE; 1236 } 1237 1238 static struct notifier_block fib_inetaddr_notifier = { 1239 .notifier_call = fib_inetaddr_event, 1240 }; 1241 1242 static struct notifier_block fib_netdev_notifier = { 1243 .notifier_call = fib_netdev_event, 1244 }; 1245 1246 static int __net_init ip_fib_net_init(struct net *net) 1247 { 1248 int err; 1249 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ; 1250 1251 net->ipv4.fib_seq = 0; 1252 1253 /* Avoid false sharing : Use at least a full cache line */ 1254 size = max_t(size_t, size, L1_CACHE_BYTES); 1255 1256 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL); 1257 if (!net->ipv4.fib_table_hash) 1258 return -ENOMEM; 1259 1260 err = fib4_rules_init(net); 1261 if (err < 0) 1262 goto fail; 1263 return 0; 1264 1265 fail: 1266 kfree(net->ipv4.fib_table_hash); 1267 return err; 1268 } 1269 1270 static void ip_fib_net_exit(struct net *net) 1271 { 1272 unsigned int i; 1273 1274 rtnl_lock(); 1275 #ifdef CONFIG_IP_MULTIPLE_TABLES 1276 RCU_INIT_POINTER(net->ipv4.fib_main, NULL); 1277 RCU_INIT_POINTER(net->ipv4.fib_default, NULL); 1278 #endif 1279 for (i = 0; i < FIB_TABLE_HASHSZ; i++) { 1280 struct hlist_head *head = &net->ipv4.fib_table_hash[i]; 1281 struct hlist_node *tmp; 1282 struct fib_table *tb; 1283 1284 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) { 1285 hlist_del(&tb->tb_hlist); 1286 fib_table_flush(net, tb); 1287 fib_free_table(tb); 1288 } 1289 } 1290 1291 #ifdef CONFIG_IP_MULTIPLE_TABLES 1292 fib4_rules_exit(net); 1293 #endif 1294 rtnl_unlock(); 1295 kfree(net->ipv4.fib_table_hash); 1296 } 1297 1298 static int __net_init fib_net_init(struct net *net) 1299 { 1300 int error; 1301 1302 #ifdef CONFIG_IP_ROUTE_CLASSID 1303 net->ipv4.fib_num_tclassid_users = 0; 1304 #endif 1305 error = ip_fib_net_init(net); 1306 if (error < 0) 1307 goto out; 1308 error = nl_fib_lookup_init(net); 1309 if (error < 0) 1310 goto out_nlfl; 1311 error = fib_proc_init(net); 1312 if (error < 0) 1313 goto out_proc; 1314 out: 1315 return error; 1316 1317 out_proc: 1318 nl_fib_lookup_exit(net); 1319 out_nlfl: 1320 ip_fib_net_exit(net); 1321 goto out; 1322 } 1323 1324 static void __net_exit fib_net_exit(struct net *net) 1325 { 1326 fib_proc_exit(net); 1327 nl_fib_lookup_exit(net); 1328 ip_fib_net_exit(net); 1329 } 1330 1331 static struct pernet_operations fib_net_ops = { 1332 .init = fib_net_init, 1333 .exit = fib_net_exit, 1334 }; 1335 1336 void __init ip_fib_init(void) 1337 { 1338 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL); 1339 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL); 1340 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL); 1341 1342 register_pernet_subsys(&fib_net_ops); 1343 register_netdevice_notifier(&fib_netdev_notifier); 1344 register_inetaddr_notifier(&fib_inetaddr_notifier); 1345 1346 fib_trie_init(); 1347 } 1348