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