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