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