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 <asm/system.h> 19 #include <linux/bitops.h> 20 #include <linux/capability.h> 21 #include <linux/types.h> 22 #include <linux/kernel.h> 23 #include <linux/mm.h> 24 #include <linux/string.h> 25 #include <linux/socket.h> 26 #include <linux/sockios.h> 27 #include <linux/errno.h> 28 #include <linux/in.h> 29 #include <linux/inet.h> 30 #include <linux/inetdevice.h> 31 #include <linux/netdevice.h> 32 #include <linux/if_addr.h> 33 #include <linux/if_arp.h> 34 #include <linux/skbuff.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 48 #ifndef CONFIG_IP_MULTIPLE_TABLES 49 50 static int __net_init fib4_rules_init(struct net *net) 51 { 52 struct fib_table *local_table, *main_table; 53 54 local_table = fib_hash_table(RT_TABLE_LOCAL); 55 if (local_table == NULL) 56 return -ENOMEM; 57 58 main_table = fib_hash_table(RT_TABLE_MAIN); 59 if (main_table == NULL) 60 goto fail; 61 62 hlist_add_head_rcu(&local_table->tb_hlist, 63 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]); 64 hlist_add_head_rcu(&main_table->tb_hlist, 65 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]); 66 return 0; 67 68 fail: 69 kfree(local_table); 70 return -ENOMEM; 71 } 72 #else 73 74 struct fib_table *fib_new_table(struct net *net, u32 id) 75 { 76 struct fib_table *tb; 77 unsigned int h; 78 79 if (id == 0) 80 id = RT_TABLE_MAIN; 81 tb = fib_get_table(net, id); 82 if (tb) 83 return tb; 84 85 tb = fib_hash_table(id); 86 if (!tb) 87 return NULL; 88 h = id & (FIB_TABLE_HASHSZ - 1); 89 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]); 90 return tb; 91 } 92 93 struct fib_table *fib_get_table(struct net *net, u32 id) 94 { 95 struct fib_table *tb; 96 struct hlist_node *node; 97 struct hlist_head *head; 98 unsigned int h; 99 100 if (id == 0) 101 id = RT_TABLE_MAIN; 102 h = id & (FIB_TABLE_HASHSZ - 1); 103 104 rcu_read_lock(); 105 head = &net->ipv4.fib_table_hash[h]; 106 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) { 107 if (tb->tb_id == id) { 108 rcu_read_unlock(); 109 return tb; 110 } 111 } 112 rcu_read_unlock(); 113 return NULL; 114 } 115 #endif /* CONFIG_IP_MULTIPLE_TABLES */ 116 117 void fib_select_default(struct net *net, 118 const struct flowi *flp, struct fib_result *res) 119 { 120 struct fib_table *tb; 121 int table = RT_TABLE_MAIN; 122 #ifdef CONFIG_IP_MULTIPLE_TABLES 123 if (res->r == NULL || res->r->action != FR_ACT_TO_TBL) 124 return; 125 table = res->r->table; 126 #endif 127 tb = fib_get_table(net, table); 128 if (FIB_RES_GW(*res) && FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK) 129 fib_table_select_default(tb, flp, res); 130 } 131 132 static void fib_flush(struct net *net) 133 { 134 int flushed = 0; 135 struct fib_table *tb; 136 struct hlist_node *node; 137 struct hlist_head *head; 138 unsigned int h; 139 140 for (h = 0; h < FIB_TABLE_HASHSZ; h++) { 141 head = &net->ipv4.fib_table_hash[h]; 142 hlist_for_each_entry(tb, node, head, tb_hlist) 143 flushed += fib_table_flush(tb); 144 } 145 146 if (flushed) 147 rt_cache_flush(net, -1); 148 } 149 150 /** 151 * __ip_dev_find - find the first device with a given source address. 152 * @net: the net namespace 153 * @addr: the source address 154 * @devref: if true, take a reference on the found device 155 * 156 * If a caller uses devref=false, it should be protected by RCU, or RTNL 157 */ 158 struct net_device *__ip_dev_find(struct net *net, __be32 addr, bool devref) 159 { 160 struct flowi fl = { 161 .nl_u = { 162 .ip4_u = { 163 .daddr = addr 164 } 165 }, 166 .flags = FLOWI_FLAG_MATCH_ANY_IIF 167 }; 168 struct fib_result res = { 0 }; 169 struct net_device *dev = NULL; 170 171 rcu_read_lock(); 172 if (fib_lookup(net, &fl, &res)) { 173 rcu_read_unlock(); 174 return NULL; 175 } 176 if (res.type != RTN_LOCAL) 177 goto out; 178 dev = FIB_RES_DEV(res); 179 180 if (dev && devref) 181 dev_hold(dev); 182 out: 183 rcu_read_unlock(); 184 return dev; 185 } 186 EXPORT_SYMBOL(__ip_dev_find); 187 188 /* 189 * Find address type as if only "dev" was present in the system. If 190 * on_dev is NULL then all interfaces are taken into consideration. 191 */ 192 static inline unsigned __inet_dev_addr_type(struct net *net, 193 const struct net_device *dev, 194 __be32 addr) 195 { 196 struct flowi fl = { .nl_u = { .ip4_u = { .daddr = addr } } }; 197 struct fib_result res; 198 unsigned ret = RTN_BROADCAST; 199 struct fib_table *local_table; 200 201 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr)) 202 return RTN_BROADCAST; 203 if (ipv4_is_multicast(addr)) 204 return RTN_MULTICAST; 205 206 #ifdef CONFIG_IP_MULTIPLE_TABLES 207 res.r = NULL; 208 #endif 209 210 local_table = fib_get_table(net, RT_TABLE_LOCAL); 211 if (local_table) { 212 ret = RTN_UNICAST; 213 rcu_read_lock(); 214 if (!fib_table_lookup(local_table, &fl, &res, FIB_LOOKUP_NOREF)) { 215 if (!dev || dev == res.fi->fib_dev) 216 ret = res.type; 217 } 218 rcu_read_unlock(); 219 } 220 return ret; 221 } 222 223 unsigned int inet_addr_type(struct net *net, __be32 addr) 224 { 225 return __inet_dev_addr_type(net, NULL, addr); 226 } 227 EXPORT_SYMBOL(inet_addr_type); 228 229 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev, 230 __be32 addr) 231 { 232 return __inet_dev_addr_type(net, dev, addr); 233 } 234 EXPORT_SYMBOL(inet_dev_addr_type); 235 236 /* Given (packet source, input interface) and optional (dst, oif, tos): 237 * - (main) check, that source is valid i.e. not broadcast or our local 238 * address. 239 * - figure out what "logical" interface this packet arrived 240 * and calculate "specific destination" address. 241 * - check, that packet arrived from expected physical interface. 242 * called with rcu_read_lock() 243 */ 244 int fib_validate_source(__be32 src, __be32 dst, u8 tos, int oif, 245 struct net_device *dev, __be32 *spec_dst, 246 u32 *itag, u32 mark) 247 { 248 struct in_device *in_dev; 249 struct flowi fl = { 250 .nl_u = { 251 .ip4_u = { 252 .daddr = src, 253 .saddr = dst, 254 .tos = tos 255 } 256 }, 257 .mark = mark, 258 .iif = oif 259 }; 260 struct fib_result res; 261 int no_addr, rpf, accept_local; 262 bool dev_match; 263 int ret; 264 struct net *net; 265 266 no_addr = rpf = accept_local = 0; 267 in_dev = __in_dev_get_rcu(dev); 268 if (in_dev) { 269 no_addr = in_dev->ifa_list == NULL; 270 rpf = IN_DEV_RPFILTER(in_dev); 271 accept_local = IN_DEV_ACCEPT_LOCAL(in_dev); 272 if (mark && !IN_DEV_SRC_VMARK(in_dev)) 273 fl.mark = 0; 274 } 275 276 if (in_dev == NULL) 277 goto e_inval; 278 279 net = dev_net(dev); 280 if (fib_lookup(net, &fl, &res)) 281 goto last_resort; 282 if (res.type != RTN_UNICAST) { 283 if (res.type != RTN_LOCAL || !accept_local) 284 goto e_inval; 285 } 286 *spec_dst = FIB_RES_PREFSRC(res); 287 fib_combine_itag(itag, &res); 288 dev_match = false; 289 290 #ifdef CONFIG_IP_ROUTE_MULTIPATH 291 for (ret = 0; ret < res.fi->fib_nhs; ret++) { 292 struct fib_nh *nh = &res.fi->fib_nh[ret]; 293 294 if (nh->nh_dev == dev) { 295 dev_match = true; 296 break; 297 } 298 } 299 #else 300 if (FIB_RES_DEV(res) == dev) 301 dev_match = true; 302 #endif 303 if (dev_match) { 304 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 305 return ret; 306 } 307 if (no_addr) 308 goto last_resort; 309 if (rpf == 1) 310 goto e_rpf; 311 fl.oif = dev->ifindex; 312 313 ret = 0; 314 if (fib_lookup(net, &fl, &res) == 0) { 315 if (res.type == RTN_UNICAST) { 316 *spec_dst = FIB_RES_PREFSRC(res); 317 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 318 } 319 } 320 return ret; 321 322 last_resort: 323 if (rpf) 324 goto e_rpf; 325 *spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE); 326 *itag = 0; 327 return 0; 328 329 e_inval: 330 return -EINVAL; 331 e_rpf: 332 return -EXDEV; 333 } 334 335 static inline __be32 sk_extract_addr(struct sockaddr *addr) 336 { 337 return ((struct sockaddr_in *) addr)->sin_addr.s_addr; 338 } 339 340 static int put_rtax(struct nlattr *mx, int len, int type, u32 value) 341 { 342 struct nlattr *nla; 343 344 nla = (struct nlattr *) ((char *) mx + len); 345 nla->nla_type = type; 346 nla->nla_len = nla_attr_size(4); 347 *(u32 *) nla_data(nla) = value; 348 349 return len + nla_total_size(4); 350 } 351 352 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt, 353 struct fib_config *cfg) 354 { 355 __be32 addr; 356 int plen; 357 358 memset(cfg, 0, sizeof(*cfg)); 359 cfg->fc_nlinfo.nl_net = net; 360 361 if (rt->rt_dst.sa_family != AF_INET) 362 return -EAFNOSUPPORT; 363 364 /* 365 * Check mask for validity: 366 * a) it must be contiguous. 367 * b) destination must have all host bits clear. 368 * c) if application forgot to set correct family (AF_INET), 369 * reject request unless it is absolutely clear i.e. 370 * both family and mask are zero. 371 */ 372 plen = 32; 373 addr = sk_extract_addr(&rt->rt_dst); 374 if (!(rt->rt_flags & RTF_HOST)) { 375 __be32 mask = sk_extract_addr(&rt->rt_genmask); 376 377 if (rt->rt_genmask.sa_family != AF_INET) { 378 if (mask || rt->rt_genmask.sa_family) 379 return -EAFNOSUPPORT; 380 } 381 382 if (bad_mask(mask, addr)) 383 return -EINVAL; 384 385 plen = inet_mask_len(mask); 386 } 387 388 cfg->fc_dst_len = plen; 389 cfg->fc_dst = addr; 390 391 if (cmd != SIOCDELRT) { 392 cfg->fc_nlflags = NLM_F_CREATE; 393 cfg->fc_protocol = RTPROT_BOOT; 394 } 395 396 if (rt->rt_metric) 397 cfg->fc_priority = rt->rt_metric - 1; 398 399 if (rt->rt_flags & RTF_REJECT) { 400 cfg->fc_scope = RT_SCOPE_HOST; 401 cfg->fc_type = RTN_UNREACHABLE; 402 return 0; 403 } 404 405 cfg->fc_scope = RT_SCOPE_NOWHERE; 406 cfg->fc_type = RTN_UNICAST; 407 408 if (rt->rt_dev) { 409 char *colon; 410 struct net_device *dev; 411 char devname[IFNAMSIZ]; 412 413 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1)) 414 return -EFAULT; 415 416 devname[IFNAMSIZ-1] = 0; 417 colon = strchr(devname, ':'); 418 if (colon) 419 *colon = 0; 420 dev = __dev_get_by_name(net, devname); 421 if (!dev) 422 return -ENODEV; 423 cfg->fc_oif = dev->ifindex; 424 if (colon) { 425 struct in_ifaddr *ifa; 426 struct in_device *in_dev = __in_dev_get_rtnl(dev); 427 if (!in_dev) 428 return -ENODEV; 429 *colon = ':'; 430 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) 431 if (strcmp(ifa->ifa_label, devname) == 0) 432 break; 433 if (ifa == NULL) 434 return -ENODEV; 435 cfg->fc_prefsrc = ifa->ifa_local; 436 } 437 } 438 439 addr = sk_extract_addr(&rt->rt_gateway); 440 if (rt->rt_gateway.sa_family == AF_INET && addr) { 441 cfg->fc_gw = addr; 442 if (rt->rt_flags & RTF_GATEWAY && 443 inet_addr_type(net, addr) == RTN_UNICAST) 444 cfg->fc_scope = RT_SCOPE_UNIVERSE; 445 } 446 447 if (cmd == SIOCDELRT) 448 return 0; 449 450 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw) 451 return -EINVAL; 452 453 if (cfg->fc_scope == RT_SCOPE_NOWHERE) 454 cfg->fc_scope = RT_SCOPE_LINK; 455 456 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) { 457 struct nlattr *mx; 458 int len = 0; 459 460 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL); 461 if (mx == NULL) 462 return -ENOMEM; 463 464 if (rt->rt_flags & RTF_MTU) 465 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40); 466 467 if (rt->rt_flags & RTF_WINDOW) 468 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window); 469 470 if (rt->rt_flags & RTF_IRTT) 471 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3); 472 473 cfg->fc_mx = mx; 474 cfg->fc_mx_len = len; 475 } 476 477 return 0; 478 } 479 480 /* 481 * Handle IP routing ioctl calls. 482 * These are used to manipulate the routing tables 483 */ 484 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg) 485 { 486 struct fib_config cfg; 487 struct rtentry rt; 488 int err; 489 490 switch (cmd) { 491 case SIOCADDRT: /* Add a route */ 492 case SIOCDELRT: /* Delete a route */ 493 if (!capable(CAP_NET_ADMIN)) 494 return -EPERM; 495 496 if (copy_from_user(&rt, arg, sizeof(rt))) 497 return -EFAULT; 498 499 rtnl_lock(); 500 err = rtentry_to_fib_config(net, cmd, &rt, &cfg); 501 if (err == 0) { 502 struct fib_table *tb; 503 504 if (cmd == SIOCDELRT) { 505 tb = fib_get_table(net, cfg.fc_table); 506 if (tb) 507 err = fib_table_delete(tb, &cfg); 508 else 509 err = -ESRCH; 510 } else { 511 tb = fib_new_table(net, cfg.fc_table); 512 if (tb) 513 err = fib_table_insert(tb, &cfg); 514 else 515 err = -ENOBUFS; 516 } 517 518 /* allocated by rtentry_to_fib_config() */ 519 kfree(cfg.fc_mx); 520 } 521 rtnl_unlock(); 522 return err; 523 } 524 return -EINVAL; 525 } 526 527 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = { 528 [RTA_DST] = { .type = NLA_U32 }, 529 [RTA_SRC] = { .type = NLA_U32 }, 530 [RTA_IIF] = { .type = NLA_U32 }, 531 [RTA_OIF] = { .type = NLA_U32 }, 532 [RTA_GATEWAY] = { .type = NLA_U32 }, 533 [RTA_PRIORITY] = { .type = NLA_U32 }, 534 [RTA_PREFSRC] = { .type = NLA_U32 }, 535 [RTA_METRICS] = { .type = NLA_NESTED }, 536 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) }, 537 [RTA_FLOW] = { .type = NLA_U32 }, 538 }; 539 540 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb, 541 struct nlmsghdr *nlh, struct fib_config *cfg) 542 { 543 struct nlattr *attr; 544 int err, remaining; 545 struct rtmsg *rtm; 546 547 err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy); 548 if (err < 0) 549 goto errout; 550 551 memset(cfg, 0, sizeof(*cfg)); 552 553 rtm = nlmsg_data(nlh); 554 cfg->fc_dst_len = rtm->rtm_dst_len; 555 cfg->fc_tos = rtm->rtm_tos; 556 cfg->fc_table = rtm->rtm_table; 557 cfg->fc_protocol = rtm->rtm_protocol; 558 cfg->fc_scope = rtm->rtm_scope; 559 cfg->fc_type = rtm->rtm_type; 560 cfg->fc_flags = rtm->rtm_flags; 561 cfg->fc_nlflags = nlh->nlmsg_flags; 562 563 cfg->fc_nlinfo.pid = NETLINK_CB(skb).pid; 564 cfg->fc_nlinfo.nlh = nlh; 565 cfg->fc_nlinfo.nl_net = net; 566 567 if (cfg->fc_type > RTN_MAX) { 568 err = -EINVAL; 569 goto errout; 570 } 571 572 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) { 573 switch (nla_type(attr)) { 574 case RTA_DST: 575 cfg->fc_dst = nla_get_be32(attr); 576 break; 577 case RTA_OIF: 578 cfg->fc_oif = nla_get_u32(attr); 579 break; 580 case RTA_GATEWAY: 581 cfg->fc_gw = nla_get_be32(attr); 582 break; 583 case RTA_PRIORITY: 584 cfg->fc_priority = nla_get_u32(attr); 585 break; 586 case RTA_PREFSRC: 587 cfg->fc_prefsrc = nla_get_be32(attr); 588 break; 589 case RTA_METRICS: 590 cfg->fc_mx = nla_data(attr); 591 cfg->fc_mx_len = nla_len(attr); 592 break; 593 case RTA_MULTIPATH: 594 cfg->fc_mp = nla_data(attr); 595 cfg->fc_mp_len = nla_len(attr); 596 break; 597 case RTA_FLOW: 598 cfg->fc_flow = nla_get_u32(attr); 599 break; 600 case RTA_TABLE: 601 cfg->fc_table = nla_get_u32(attr); 602 break; 603 } 604 } 605 606 return 0; 607 errout: 608 return err; 609 } 610 611 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg) 612 { 613 struct net *net = sock_net(skb->sk); 614 struct fib_config cfg; 615 struct fib_table *tb; 616 int err; 617 618 err = rtm_to_fib_config(net, skb, nlh, &cfg); 619 if (err < 0) 620 goto errout; 621 622 tb = fib_get_table(net, cfg.fc_table); 623 if (tb == NULL) { 624 err = -ESRCH; 625 goto errout; 626 } 627 628 err = fib_table_delete(tb, &cfg); 629 errout: 630 return err; 631 } 632 633 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg) 634 { 635 struct net *net = sock_net(skb->sk); 636 struct fib_config cfg; 637 struct fib_table *tb; 638 int err; 639 640 err = rtm_to_fib_config(net, skb, nlh, &cfg); 641 if (err < 0) 642 goto errout; 643 644 tb = fib_new_table(net, cfg.fc_table); 645 if (tb == NULL) { 646 err = -ENOBUFS; 647 goto errout; 648 } 649 650 err = fib_table_insert(tb, &cfg); 651 errout: 652 return err; 653 } 654 655 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 656 { 657 struct net *net = sock_net(skb->sk); 658 unsigned int h, s_h; 659 unsigned int e = 0, s_e; 660 struct fib_table *tb; 661 struct hlist_node *node; 662 struct hlist_head *head; 663 int dumped = 0; 664 665 if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) && 666 ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED) 667 return ip_rt_dump(skb, cb); 668 669 s_h = cb->args[0]; 670 s_e = cb->args[1]; 671 672 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) { 673 e = 0; 674 head = &net->ipv4.fib_table_hash[h]; 675 hlist_for_each_entry(tb, node, head, tb_hlist) { 676 if (e < s_e) 677 goto next; 678 if (dumped) 679 memset(&cb->args[2], 0, sizeof(cb->args) - 680 2 * sizeof(cb->args[0])); 681 if (fib_table_dump(tb, skb, cb) < 0) 682 goto out; 683 dumped = 1; 684 next: 685 e++; 686 } 687 } 688 out: 689 cb->args[1] = e; 690 cb->args[0] = h; 691 692 return skb->len; 693 } 694 695 /* Prepare and feed intra-kernel routing request. 696 * Really, it should be netlink message, but :-( netlink 697 * can be not configured, so that we feed it directly 698 * to fib engine. It is legal, because all events occur 699 * only when netlink is already locked. 700 */ 701 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa) 702 { 703 struct net *net = dev_net(ifa->ifa_dev->dev); 704 struct fib_table *tb; 705 struct fib_config cfg = { 706 .fc_protocol = RTPROT_KERNEL, 707 .fc_type = type, 708 .fc_dst = dst, 709 .fc_dst_len = dst_len, 710 .fc_prefsrc = ifa->ifa_local, 711 .fc_oif = ifa->ifa_dev->dev->ifindex, 712 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND, 713 .fc_nlinfo = { 714 .nl_net = net, 715 }, 716 }; 717 718 if (type == RTN_UNICAST) 719 tb = fib_new_table(net, RT_TABLE_MAIN); 720 else 721 tb = fib_new_table(net, RT_TABLE_LOCAL); 722 723 if (tb == NULL) 724 return; 725 726 cfg.fc_table = tb->tb_id; 727 728 if (type != RTN_LOCAL) 729 cfg.fc_scope = RT_SCOPE_LINK; 730 else 731 cfg.fc_scope = RT_SCOPE_HOST; 732 733 if (cmd == RTM_NEWROUTE) 734 fib_table_insert(tb, &cfg); 735 else 736 fib_table_delete(tb, &cfg); 737 } 738 739 void fib_add_ifaddr(struct in_ifaddr *ifa) 740 { 741 struct in_device *in_dev = ifa->ifa_dev; 742 struct net_device *dev = in_dev->dev; 743 struct in_ifaddr *prim = ifa; 744 __be32 mask = ifa->ifa_mask; 745 __be32 addr = ifa->ifa_local; 746 __be32 prefix = ifa->ifa_address & mask; 747 748 if (ifa->ifa_flags & IFA_F_SECONDARY) { 749 prim = inet_ifa_byprefix(in_dev, prefix, mask); 750 if (prim == NULL) { 751 printk(KERN_WARNING "fib_add_ifaddr: bug: prim == NULL\n"); 752 return; 753 } 754 } 755 756 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim); 757 758 if (!(dev->flags & IFF_UP)) 759 return; 760 761 /* Add broadcast address, if it is explicitly assigned. */ 762 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) 763 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 764 765 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) && 766 (prefix != addr || ifa->ifa_prefixlen < 32)) { 767 fib_magic(RTM_NEWROUTE, 768 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 769 prefix, ifa->ifa_prefixlen, prim); 770 771 /* Add network specific broadcasts, when it takes a sense */ 772 if (ifa->ifa_prefixlen < 31) { 773 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim); 774 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask, 775 32, prim); 776 } 777 } 778 } 779 780 static void fib_del_ifaddr(struct in_ifaddr *ifa) 781 { 782 struct in_device *in_dev = ifa->ifa_dev; 783 struct net_device *dev = in_dev->dev; 784 struct in_ifaddr *ifa1; 785 struct in_ifaddr *prim = ifa; 786 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask; 787 __be32 any = ifa->ifa_address & ifa->ifa_mask; 788 #define LOCAL_OK 1 789 #define BRD_OK 2 790 #define BRD0_OK 4 791 #define BRD1_OK 8 792 unsigned ok = 0; 793 794 if (!(ifa->ifa_flags & IFA_F_SECONDARY)) 795 fib_magic(RTM_DELROUTE, 796 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 797 any, ifa->ifa_prefixlen, prim); 798 else { 799 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask); 800 if (prim == NULL) { 801 printk(KERN_WARNING "fib_del_ifaddr: bug: prim == NULL\n"); 802 return; 803 } 804 } 805 806 /* Deletion is more complicated than add. 807 * We should take care of not to delete too much :-) 808 * 809 * Scan address list to be sure that addresses are really gone. 810 */ 811 812 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) { 813 if (ifa->ifa_local == ifa1->ifa_local) 814 ok |= LOCAL_OK; 815 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 816 ok |= BRD_OK; 817 if (brd == ifa1->ifa_broadcast) 818 ok |= BRD1_OK; 819 if (any == ifa1->ifa_broadcast) 820 ok |= BRD0_OK; 821 } 822 823 if (!(ok & BRD_OK)) 824 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 825 if (!(ok & BRD1_OK)) 826 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim); 827 if (!(ok & BRD0_OK)) 828 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim); 829 if (!(ok & LOCAL_OK)) { 830 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim); 831 832 /* Check, that this local address finally disappeared. */ 833 if (inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) { 834 /* And the last, but not the least thing. 835 * We must flush stray FIB entries. 836 * 837 * First of all, we scan fib_info list searching 838 * for stray nexthop entries, then ignite fib_flush. 839 */ 840 if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local)) 841 fib_flush(dev_net(dev)); 842 } 843 } 844 #undef LOCAL_OK 845 #undef BRD_OK 846 #undef BRD0_OK 847 #undef BRD1_OK 848 } 849 850 static void nl_fib_lookup(struct fib_result_nl *frn, struct fib_table *tb) 851 { 852 853 struct fib_result res; 854 struct flowi fl = { 855 .mark = frn->fl_mark, 856 .nl_u = { 857 .ip4_u = { 858 .daddr = frn->fl_addr, 859 .tos = frn->fl_tos, 860 .scope = frn->fl_scope 861 } 862 } 863 }; 864 865 #ifdef CONFIG_IP_MULTIPLE_TABLES 866 res.r = NULL; 867 #endif 868 869 frn->err = -ENOENT; 870 if (tb) { 871 local_bh_disable(); 872 873 frn->tb_id = tb->tb_id; 874 rcu_read_lock(); 875 frn->err = fib_table_lookup(tb, &fl, &res, FIB_LOOKUP_NOREF); 876 877 if (!frn->err) { 878 frn->prefixlen = res.prefixlen; 879 frn->nh_sel = res.nh_sel; 880 frn->type = res.type; 881 frn->scope = res.scope; 882 } 883 rcu_read_unlock(); 884 local_bh_enable(); 885 } 886 } 887 888 static void nl_fib_input(struct sk_buff *skb) 889 { 890 struct net *net; 891 struct fib_result_nl *frn; 892 struct nlmsghdr *nlh; 893 struct fib_table *tb; 894 u32 pid; 895 896 net = sock_net(skb->sk); 897 nlh = nlmsg_hdr(skb); 898 if (skb->len < NLMSG_SPACE(0) || skb->len < nlh->nlmsg_len || 899 nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*frn))) 900 return; 901 902 skb = skb_clone(skb, GFP_KERNEL); 903 if (skb == NULL) 904 return; 905 nlh = nlmsg_hdr(skb); 906 907 frn = (struct fib_result_nl *) NLMSG_DATA(nlh); 908 tb = fib_get_table(net, frn->tb_id_in); 909 910 nl_fib_lookup(frn, tb); 911 912 pid = NETLINK_CB(skb).pid; /* pid of sending process */ 913 NETLINK_CB(skb).pid = 0; /* from kernel */ 914 NETLINK_CB(skb).dst_group = 0; /* unicast */ 915 netlink_unicast(net->ipv4.fibnl, skb, pid, MSG_DONTWAIT); 916 } 917 918 static int __net_init nl_fib_lookup_init(struct net *net) 919 { 920 struct sock *sk; 921 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, 0, 922 nl_fib_input, NULL, THIS_MODULE); 923 if (sk == NULL) 924 return -EAFNOSUPPORT; 925 net->ipv4.fibnl = sk; 926 return 0; 927 } 928 929 static void nl_fib_lookup_exit(struct net *net) 930 { 931 netlink_kernel_release(net->ipv4.fibnl); 932 net->ipv4.fibnl = NULL; 933 } 934 935 static void fib_disable_ip(struct net_device *dev, int force, int delay) 936 { 937 if (fib_sync_down_dev(dev, force)) 938 fib_flush(dev_net(dev)); 939 rt_cache_flush(dev_net(dev), delay); 940 arp_ifdown(dev); 941 } 942 943 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 944 { 945 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; 946 struct net_device *dev = ifa->ifa_dev->dev; 947 948 switch (event) { 949 case NETDEV_UP: 950 fib_add_ifaddr(ifa); 951 #ifdef CONFIG_IP_ROUTE_MULTIPATH 952 fib_sync_up(dev); 953 #endif 954 rt_cache_flush(dev_net(dev), -1); 955 break; 956 case NETDEV_DOWN: 957 fib_del_ifaddr(ifa); 958 if (ifa->ifa_dev->ifa_list == NULL) { 959 /* Last address was deleted from this interface. 960 * Disable IP. 961 */ 962 fib_disable_ip(dev, 1, 0); 963 } else { 964 rt_cache_flush(dev_net(dev), -1); 965 } 966 break; 967 } 968 return NOTIFY_DONE; 969 } 970 971 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 972 { 973 struct net_device *dev = ptr; 974 struct in_device *in_dev = __in_dev_get_rtnl(dev); 975 976 if (event == NETDEV_UNREGISTER) { 977 fib_disable_ip(dev, 2, -1); 978 return NOTIFY_DONE; 979 } 980 981 if (!in_dev) 982 return NOTIFY_DONE; 983 984 switch (event) { 985 case NETDEV_UP: 986 for_ifa(in_dev) { 987 fib_add_ifaddr(ifa); 988 } endfor_ifa(in_dev); 989 #ifdef CONFIG_IP_ROUTE_MULTIPATH 990 fib_sync_up(dev); 991 #endif 992 rt_cache_flush(dev_net(dev), -1); 993 break; 994 case NETDEV_DOWN: 995 fib_disable_ip(dev, 0, 0); 996 break; 997 case NETDEV_CHANGEMTU: 998 case NETDEV_CHANGE: 999 rt_cache_flush(dev_net(dev), 0); 1000 break; 1001 case NETDEV_UNREGISTER_BATCH: 1002 rt_cache_flush_batch(); 1003 break; 1004 } 1005 return NOTIFY_DONE; 1006 } 1007 1008 static struct notifier_block fib_inetaddr_notifier = { 1009 .notifier_call = fib_inetaddr_event, 1010 }; 1011 1012 static struct notifier_block fib_netdev_notifier = { 1013 .notifier_call = fib_netdev_event, 1014 }; 1015 1016 static int __net_init ip_fib_net_init(struct net *net) 1017 { 1018 int err; 1019 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ; 1020 1021 /* Avoid false sharing : Use at least a full cache line */ 1022 size = max_t(size_t, size, L1_CACHE_BYTES); 1023 1024 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL); 1025 if (net->ipv4.fib_table_hash == NULL) 1026 return -ENOMEM; 1027 1028 err = fib4_rules_init(net); 1029 if (err < 0) 1030 goto fail; 1031 return 0; 1032 1033 fail: 1034 kfree(net->ipv4.fib_table_hash); 1035 return err; 1036 } 1037 1038 static void ip_fib_net_exit(struct net *net) 1039 { 1040 unsigned int i; 1041 1042 #ifdef CONFIG_IP_MULTIPLE_TABLES 1043 fib4_rules_exit(net); 1044 #endif 1045 1046 for (i = 0; i < FIB_TABLE_HASHSZ; i++) { 1047 struct fib_table *tb; 1048 struct hlist_head *head; 1049 struct hlist_node *node, *tmp; 1050 1051 head = &net->ipv4.fib_table_hash[i]; 1052 hlist_for_each_entry_safe(tb, node, tmp, head, tb_hlist) { 1053 hlist_del(node); 1054 fib_table_flush(tb); 1055 fib_free_table(tb); 1056 } 1057 } 1058 kfree(net->ipv4.fib_table_hash); 1059 } 1060 1061 static int __net_init fib_net_init(struct net *net) 1062 { 1063 int error; 1064 1065 error = ip_fib_net_init(net); 1066 if (error < 0) 1067 goto out; 1068 error = nl_fib_lookup_init(net); 1069 if (error < 0) 1070 goto out_nlfl; 1071 error = fib_proc_init(net); 1072 if (error < 0) 1073 goto out_proc; 1074 out: 1075 return error; 1076 1077 out_proc: 1078 nl_fib_lookup_exit(net); 1079 out_nlfl: 1080 ip_fib_net_exit(net); 1081 goto out; 1082 } 1083 1084 static void __net_exit fib_net_exit(struct net *net) 1085 { 1086 fib_proc_exit(net); 1087 nl_fib_lookup_exit(net); 1088 ip_fib_net_exit(net); 1089 } 1090 1091 static struct pernet_operations fib_net_ops = { 1092 .init = fib_net_init, 1093 .exit = fib_net_exit, 1094 }; 1095 1096 void __init ip_fib_init(void) 1097 { 1098 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL); 1099 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL); 1100 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib); 1101 1102 register_pernet_subsys(&fib_net_ops); 1103 register_netdevice_notifier(&fib_netdev_notifier); 1104 register_inetaddr_notifier(&fib_inetaddr_notifier); 1105 1106 fib_hash_init(); 1107 } 1108