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 fl4.flowi4_flags = 0; 344 345 no_addr = idev->ifa_list == NULL; 346 347 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0; 348 349 trace_fib_validate_source(dev, &fl4); 350 351 net = dev_net(dev); 352 if (fib_lookup(net, &fl4, &res, 0)) 353 goto last_resort; 354 if (res.type != RTN_UNICAST && 355 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev))) 356 goto e_inval; 357 if (!rpf && !fib_num_tclassid_users(dev_net(dev)) && 358 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) 359 goto last_resort; 360 fib_combine_itag(itag, &res); 361 dev_match = false; 362 363 #ifdef CONFIG_IP_ROUTE_MULTIPATH 364 for (ret = 0; ret < res.fi->fib_nhs; ret++) { 365 struct fib_nh *nh = &res.fi->fib_nh[ret]; 366 367 if (nh->nh_dev == dev) { 368 dev_match = true; 369 break; 370 } else if (vrf_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) { 371 dev_match = true; 372 break; 373 } 374 } 375 #else 376 if (FIB_RES_DEV(res) == dev) 377 dev_match = true; 378 #endif 379 if (dev_match) { 380 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 381 return ret; 382 } 383 if (no_addr) 384 goto last_resort; 385 if (rpf == 1) 386 goto e_rpf; 387 fl4.flowi4_oif = dev->ifindex; 388 389 ret = 0; 390 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) { 391 if (res.type == RTN_UNICAST) 392 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 393 } 394 return ret; 395 396 last_resort: 397 if (rpf) 398 goto e_rpf; 399 *itag = 0; 400 return 0; 401 402 e_inval: 403 return -EINVAL; 404 e_rpf: 405 return -EXDEV; 406 } 407 408 /* Ignore rp_filter for packets protected by IPsec. */ 409 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 410 u8 tos, int oif, struct net_device *dev, 411 struct in_device *idev, u32 *itag) 412 { 413 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev); 414 415 if (!r && !fib_num_tclassid_users(dev_net(dev)) && 416 IN_DEV_ACCEPT_LOCAL(idev) && 417 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) { 418 *itag = 0; 419 return 0; 420 } 421 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag); 422 } 423 424 static inline __be32 sk_extract_addr(struct sockaddr *addr) 425 { 426 return ((struct sockaddr_in *) addr)->sin_addr.s_addr; 427 } 428 429 static int put_rtax(struct nlattr *mx, int len, int type, u32 value) 430 { 431 struct nlattr *nla; 432 433 nla = (struct nlattr *) ((char *) mx + len); 434 nla->nla_type = type; 435 nla->nla_len = nla_attr_size(4); 436 *(u32 *) nla_data(nla) = value; 437 438 return len + nla_total_size(4); 439 } 440 441 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt, 442 struct fib_config *cfg) 443 { 444 __be32 addr; 445 int plen; 446 447 memset(cfg, 0, sizeof(*cfg)); 448 cfg->fc_nlinfo.nl_net = net; 449 450 if (rt->rt_dst.sa_family != AF_INET) 451 return -EAFNOSUPPORT; 452 453 /* 454 * Check mask for validity: 455 * a) it must be contiguous. 456 * b) destination must have all host bits clear. 457 * c) if application forgot to set correct family (AF_INET), 458 * reject request unless it is absolutely clear i.e. 459 * both family and mask are zero. 460 */ 461 plen = 32; 462 addr = sk_extract_addr(&rt->rt_dst); 463 if (!(rt->rt_flags & RTF_HOST)) { 464 __be32 mask = sk_extract_addr(&rt->rt_genmask); 465 466 if (rt->rt_genmask.sa_family != AF_INET) { 467 if (mask || rt->rt_genmask.sa_family) 468 return -EAFNOSUPPORT; 469 } 470 471 if (bad_mask(mask, addr)) 472 return -EINVAL; 473 474 plen = inet_mask_len(mask); 475 } 476 477 cfg->fc_dst_len = plen; 478 cfg->fc_dst = addr; 479 480 if (cmd != SIOCDELRT) { 481 cfg->fc_nlflags = NLM_F_CREATE; 482 cfg->fc_protocol = RTPROT_BOOT; 483 } 484 485 if (rt->rt_metric) 486 cfg->fc_priority = rt->rt_metric - 1; 487 488 if (rt->rt_flags & RTF_REJECT) { 489 cfg->fc_scope = RT_SCOPE_HOST; 490 cfg->fc_type = RTN_UNREACHABLE; 491 return 0; 492 } 493 494 cfg->fc_scope = RT_SCOPE_NOWHERE; 495 cfg->fc_type = RTN_UNICAST; 496 497 if (rt->rt_dev) { 498 char *colon; 499 struct net_device *dev; 500 char devname[IFNAMSIZ]; 501 502 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1)) 503 return -EFAULT; 504 505 devname[IFNAMSIZ-1] = 0; 506 colon = strchr(devname, ':'); 507 if (colon) 508 *colon = 0; 509 dev = __dev_get_by_name(net, devname); 510 if (!dev) 511 return -ENODEV; 512 cfg->fc_oif = dev->ifindex; 513 if (colon) { 514 struct in_ifaddr *ifa; 515 struct in_device *in_dev = __in_dev_get_rtnl(dev); 516 if (!in_dev) 517 return -ENODEV; 518 *colon = ':'; 519 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) 520 if (strcmp(ifa->ifa_label, devname) == 0) 521 break; 522 if (!ifa) 523 return -ENODEV; 524 cfg->fc_prefsrc = ifa->ifa_local; 525 } 526 } 527 528 addr = sk_extract_addr(&rt->rt_gateway); 529 if (rt->rt_gateway.sa_family == AF_INET && addr) { 530 unsigned int addr_type; 531 532 cfg->fc_gw = addr; 533 addr_type = inet_addr_type_table(net, addr, cfg->fc_table); 534 if (rt->rt_flags & RTF_GATEWAY && 535 addr_type == RTN_UNICAST) 536 cfg->fc_scope = RT_SCOPE_UNIVERSE; 537 } 538 539 if (cmd == SIOCDELRT) 540 return 0; 541 542 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw) 543 return -EINVAL; 544 545 if (cfg->fc_scope == RT_SCOPE_NOWHERE) 546 cfg->fc_scope = RT_SCOPE_LINK; 547 548 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) { 549 struct nlattr *mx; 550 int len = 0; 551 552 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL); 553 if (!mx) 554 return -ENOMEM; 555 556 if (rt->rt_flags & RTF_MTU) 557 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40); 558 559 if (rt->rt_flags & RTF_WINDOW) 560 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window); 561 562 if (rt->rt_flags & RTF_IRTT) 563 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3); 564 565 cfg->fc_mx = mx; 566 cfg->fc_mx_len = len; 567 } 568 569 return 0; 570 } 571 572 /* 573 * Handle IP routing ioctl calls. 574 * These are used to manipulate the routing tables 575 */ 576 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg) 577 { 578 struct fib_config cfg; 579 struct rtentry rt; 580 int err; 581 582 switch (cmd) { 583 case SIOCADDRT: /* Add a route */ 584 case SIOCDELRT: /* Delete a route */ 585 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 586 return -EPERM; 587 588 if (copy_from_user(&rt, arg, sizeof(rt))) 589 return -EFAULT; 590 591 rtnl_lock(); 592 err = rtentry_to_fib_config(net, cmd, &rt, &cfg); 593 if (err == 0) { 594 struct fib_table *tb; 595 596 if (cmd == SIOCDELRT) { 597 tb = fib_get_table(net, cfg.fc_table); 598 if (tb) 599 err = fib_table_delete(tb, &cfg); 600 else 601 err = -ESRCH; 602 } else { 603 tb = fib_new_table(net, cfg.fc_table); 604 if (tb) 605 err = fib_table_insert(tb, &cfg); 606 else 607 err = -ENOBUFS; 608 } 609 610 /* allocated by rtentry_to_fib_config() */ 611 kfree(cfg.fc_mx); 612 } 613 rtnl_unlock(); 614 return err; 615 } 616 return -EINVAL; 617 } 618 619 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = { 620 [RTA_DST] = { .type = NLA_U32 }, 621 [RTA_SRC] = { .type = NLA_U32 }, 622 [RTA_IIF] = { .type = NLA_U32 }, 623 [RTA_OIF] = { .type = NLA_U32 }, 624 [RTA_GATEWAY] = { .type = NLA_U32 }, 625 [RTA_PRIORITY] = { .type = NLA_U32 }, 626 [RTA_PREFSRC] = { .type = NLA_U32 }, 627 [RTA_METRICS] = { .type = NLA_NESTED }, 628 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) }, 629 [RTA_FLOW] = { .type = NLA_U32 }, 630 [RTA_ENCAP_TYPE] = { .type = NLA_U16 }, 631 [RTA_ENCAP] = { .type = NLA_NESTED }, 632 }; 633 634 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb, 635 struct nlmsghdr *nlh, struct fib_config *cfg) 636 { 637 struct nlattr *attr; 638 int err, remaining; 639 struct rtmsg *rtm; 640 641 err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy); 642 if (err < 0) 643 goto errout; 644 645 memset(cfg, 0, sizeof(*cfg)); 646 647 rtm = nlmsg_data(nlh); 648 cfg->fc_dst_len = rtm->rtm_dst_len; 649 cfg->fc_tos = rtm->rtm_tos; 650 cfg->fc_table = rtm->rtm_table; 651 cfg->fc_protocol = rtm->rtm_protocol; 652 cfg->fc_scope = rtm->rtm_scope; 653 cfg->fc_type = rtm->rtm_type; 654 cfg->fc_flags = rtm->rtm_flags; 655 cfg->fc_nlflags = nlh->nlmsg_flags; 656 657 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid; 658 cfg->fc_nlinfo.nlh = nlh; 659 cfg->fc_nlinfo.nl_net = net; 660 661 if (cfg->fc_type > RTN_MAX) { 662 err = -EINVAL; 663 goto errout; 664 } 665 666 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) { 667 switch (nla_type(attr)) { 668 case RTA_DST: 669 cfg->fc_dst = nla_get_be32(attr); 670 break; 671 case RTA_OIF: 672 cfg->fc_oif = nla_get_u32(attr); 673 break; 674 case RTA_GATEWAY: 675 cfg->fc_gw = nla_get_be32(attr); 676 break; 677 case RTA_PRIORITY: 678 cfg->fc_priority = nla_get_u32(attr); 679 break; 680 case RTA_PREFSRC: 681 cfg->fc_prefsrc = nla_get_be32(attr); 682 break; 683 case RTA_METRICS: 684 cfg->fc_mx = nla_data(attr); 685 cfg->fc_mx_len = nla_len(attr); 686 break; 687 case RTA_MULTIPATH: 688 cfg->fc_mp = nla_data(attr); 689 cfg->fc_mp_len = nla_len(attr); 690 break; 691 case RTA_FLOW: 692 cfg->fc_flow = nla_get_u32(attr); 693 break; 694 case RTA_TABLE: 695 cfg->fc_table = nla_get_u32(attr); 696 break; 697 case RTA_ENCAP: 698 cfg->fc_encap = attr; 699 break; 700 case RTA_ENCAP_TYPE: 701 cfg->fc_encap_type = nla_get_u16(attr); 702 break; 703 } 704 } 705 706 return 0; 707 errout: 708 return err; 709 } 710 711 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh) 712 { 713 struct net *net = sock_net(skb->sk); 714 struct fib_config cfg; 715 struct fib_table *tb; 716 int err; 717 718 err = rtm_to_fib_config(net, skb, nlh, &cfg); 719 if (err < 0) 720 goto errout; 721 722 tb = fib_get_table(net, cfg.fc_table); 723 if (!tb) { 724 err = -ESRCH; 725 goto errout; 726 } 727 728 err = fib_table_delete(tb, &cfg); 729 errout: 730 return err; 731 } 732 733 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh) 734 { 735 struct net *net = sock_net(skb->sk); 736 struct fib_config cfg; 737 struct fib_table *tb; 738 int err; 739 740 err = rtm_to_fib_config(net, skb, nlh, &cfg); 741 if (err < 0) 742 goto errout; 743 744 tb = fib_new_table(net, cfg.fc_table); 745 if (!tb) { 746 err = -ENOBUFS; 747 goto errout; 748 } 749 750 err = fib_table_insert(tb, &cfg); 751 errout: 752 return err; 753 } 754 755 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 756 { 757 struct net *net = sock_net(skb->sk); 758 unsigned int h, s_h; 759 unsigned int e = 0, s_e; 760 struct fib_table *tb; 761 struct hlist_head *head; 762 int dumped = 0; 763 764 if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) && 765 ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED) 766 return skb->len; 767 768 s_h = cb->args[0]; 769 s_e = cb->args[1]; 770 771 rcu_read_lock(); 772 773 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) { 774 e = 0; 775 head = &net->ipv4.fib_table_hash[h]; 776 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 777 if (e < s_e) 778 goto next; 779 if (dumped) 780 memset(&cb->args[2], 0, sizeof(cb->args) - 781 2 * sizeof(cb->args[0])); 782 if (fib_table_dump(tb, skb, cb) < 0) 783 goto out; 784 dumped = 1; 785 next: 786 e++; 787 } 788 } 789 out: 790 rcu_read_unlock(); 791 792 cb->args[1] = e; 793 cb->args[0] = h; 794 795 return skb->len; 796 } 797 798 /* Prepare and feed intra-kernel routing request. 799 * Really, it should be netlink message, but :-( netlink 800 * can be not configured, so that we feed it directly 801 * to fib engine. It is legal, because all events occur 802 * only when netlink is already locked. 803 */ 804 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa) 805 { 806 struct net *net = dev_net(ifa->ifa_dev->dev); 807 u32 tb_id = vrf_dev_table_rtnl(ifa->ifa_dev->dev); 808 struct fib_table *tb; 809 struct fib_config cfg = { 810 .fc_protocol = RTPROT_KERNEL, 811 .fc_type = type, 812 .fc_dst = dst, 813 .fc_dst_len = dst_len, 814 .fc_prefsrc = ifa->ifa_local, 815 .fc_oif = ifa->ifa_dev->dev->ifindex, 816 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND, 817 .fc_nlinfo = { 818 .nl_net = net, 819 }, 820 }; 821 822 if (!tb_id) 823 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL; 824 825 tb = fib_new_table(net, tb_id); 826 if (!tb) 827 return; 828 829 cfg.fc_table = tb->tb_id; 830 831 if (type != RTN_LOCAL) 832 cfg.fc_scope = RT_SCOPE_LINK; 833 else 834 cfg.fc_scope = RT_SCOPE_HOST; 835 836 if (cmd == RTM_NEWROUTE) 837 fib_table_insert(tb, &cfg); 838 else 839 fib_table_delete(tb, &cfg); 840 } 841 842 void fib_add_ifaddr(struct in_ifaddr *ifa) 843 { 844 struct in_device *in_dev = ifa->ifa_dev; 845 struct net_device *dev = in_dev->dev; 846 struct in_ifaddr *prim = ifa; 847 __be32 mask = ifa->ifa_mask; 848 __be32 addr = ifa->ifa_local; 849 __be32 prefix = ifa->ifa_address & mask; 850 851 if (ifa->ifa_flags & IFA_F_SECONDARY) { 852 prim = inet_ifa_byprefix(in_dev, prefix, mask); 853 if (!prim) { 854 pr_warn("%s: bug: prim == NULL\n", __func__); 855 return; 856 } 857 } 858 859 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim); 860 861 if (!(dev->flags & IFF_UP)) 862 return; 863 864 /* Add broadcast address, if it is explicitly assigned. */ 865 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) 866 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 867 868 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) && 869 (prefix != addr || ifa->ifa_prefixlen < 32)) { 870 fib_magic(RTM_NEWROUTE, 871 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 872 prefix, ifa->ifa_prefixlen, prim); 873 874 /* Add network specific broadcasts, when it takes a sense */ 875 if (ifa->ifa_prefixlen < 31) { 876 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim); 877 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask, 878 32, prim); 879 } 880 } 881 } 882 883 /* Delete primary or secondary address. 884 * Optionally, on secondary address promotion consider the addresses 885 * from subnet iprim as deleted, even if they are in device list. 886 * In this case the secondary ifa can be in device list. 887 */ 888 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim) 889 { 890 struct in_device *in_dev = ifa->ifa_dev; 891 struct net_device *dev = in_dev->dev; 892 struct in_ifaddr *ifa1; 893 struct in_ifaddr *prim = ifa, *prim1 = NULL; 894 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask; 895 __be32 any = ifa->ifa_address & ifa->ifa_mask; 896 #define LOCAL_OK 1 897 #define BRD_OK 2 898 #define BRD0_OK 4 899 #define BRD1_OK 8 900 unsigned int ok = 0; 901 int subnet = 0; /* Primary network */ 902 int gone = 1; /* Address is missing */ 903 int same_prefsrc = 0; /* Another primary with same IP */ 904 905 if (ifa->ifa_flags & IFA_F_SECONDARY) { 906 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask); 907 if (!prim) { 908 pr_warn("%s: bug: prim == NULL\n", __func__); 909 return; 910 } 911 if (iprim && iprim != prim) { 912 pr_warn("%s: bug: iprim != prim\n", __func__); 913 return; 914 } 915 } else if (!ipv4_is_zeronet(any) && 916 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) { 917 fib_magic(RTM_DELROUTE, 918 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 919 any, ifa->ifa_prefixlen, prim); 920 subnet = 1; 921 } 922 923 /* Deletion is more complicated than add. 924 * We should take care of not to delete too much :-) 925 * 926 * Scan address list to be sure that addresses are really gone. 927 */ 928 929 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) { 930 if (ifa1 == ifa) { 931 /* promotion, keep the IP */ 932 gone = 0; 933 continue; 934 } 935 /* Ignore IFAs from our subnet */ 936 if (iprim && ifa1->ifa_mask == iprim->ifa_mask && 937 inet_ifa_match(ifa1->ifa_address, iprim)) 938 continue; 939 940 /* Ignore ifa1 if it uses different primary IP (prefsrc) */ 941 if (ifa1->ifa_flags & IFA_F_SECONDARY) { 942 /* Another address from our subnet? */ 943 if (ifa1->ifa_mask == prim->ifa_mask && 944 inet_ifa_match(ifa1->ifa_address, prim)) 945 prim1 = prim; 946 else { 947 /* We reached the secondaries, so 948 * same_prefsrc should be determined. 949 */ 950 if (!same_prefsrc) 951 continue; 952 /* Search new prim1 if ifa1 is not 953 * using the current prim1 954 */ 955 if (!prim1 || 956 ifa1->ifa_mask != prim1->ifa_mask || 957 !inet_ifa_match(ifa1->ifa_address, prim1)) 958 prim1 = inet_ifa_byprefix(in_dev, 959 ifa1->ifa_address, 960 ifa1->ifa_mask); 961 if (!prim1) 962 continue; 963 if (prim1->ifa_local != prim->ifa_local) 964 continue; 965 } 966 } else { 967 if (prim->ifa_local != ifa1->ifa_local) 968 continue; 969 prim1 = ifa1; 970 if (prim != prim1) 971 same_prefsrc = 1; 972 } 973 if (ifa->ifa_local == ifa1->ifa_local) 974 ok |= LOCAL_OK; 975 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 976 ok |= BRD_OK; 977 if (brd == ifa1->ifa_broadcast) 978 ok |= BRD1_OK; 979 if (any == ifa1->ifa_broadcast) 980 ok |= BRD0_OK; 981 /* primary has network specific broadcasts */ 982 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) { 983 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask; 984 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask; 985 986 if (!ipv4_is_zeronet(any1)) { 987 if (ifa->ifa_broadcast == brd1 || 988 ifa->ifa_broadcast == any1) 989 ok |= BRD_OK; 990 if (brd == brd1 || brd == any1) 991 ok |= BRD1_OK; 992 if (any == brd1 || any == any1) 993 ok |= BRD0_OK; 994 } 995 } 996 } 997 998 if (!(ok & BRD_OK)) 999 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 1000 if (subnet && ifa->ifa_prefixlen < 31) { 1001 if (!(ok & BRD1_OK)) 1002 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim); 1003 if (!(ok & BRD0_OK)) 1004 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim); 1005 } 1006 if (!(ok & LOCAL_OK)) { 1007 unsigned int addr_type; 1008 1009 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim); 1010 1011 /* Check, that this local address finally disappeared. */ 1012 addr_type = inet_addr_type_dev_table(dev_net(dev), dev, 1013 ifa->ifa_local); 1014 if (gone && addr_type != RTN_LOCAL) { 1015 /* And the last, but not the least thing. 1016 * We must flush stray FIB entries. 1017 * 1018 * First of all, we scan fib_info list searching 1019 * for stray nexthop entries, then ignite fib_flush. 1020 */ 1021 if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local)) 1022 fib_flush(dev_net(dev)); 1023 } 1024 } 1025 #undef LOCAL_OK 1026 #undef BRD_OK 1027 #undef BRD0_OK 1028 #undef BRD1_OK 1029 } 1030 1031 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn) 1032 { 1033 1034 struct fib_result res; 1035 struct flowi4 fl4 = { 1036 .flowi4_mark = frn->fl_mark, 1037 .daddr = frn->fl_addr, 1038 .flowi4_tos = frn->fl_tos, 1039 .flowi4_scope = frn->fl_scope, 1040 }; 1041 struct fib_table *tb; 1042 1043 rcu_read_lock(); 1044 1045 tb = fib_get_table(net, frn->tb_id_in); 1046 1047 frn->err = -ENOENT; 1048 if (tb) { 1049 local_bh_disable(); 1050 1051 frn->tb_id = tb->tb_id; 1052 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 1053 1054 if (!frn->err) { 1055 frn->prefixlen = res.prefixlen; 1056 frn->nh_sel = res.nh_sel; 1057 frn->type = res.type; 1058 frn->scope = res.scope; 1059 } 1060 local_bh_enable(); 1061 } 1062 1063 rcu_read_unlock(); 1064 } 1065 1066 static void nl_fib_input(struct sk_buff *skb) 1067 { 1068 struct net *net; 1069 struct fib_result_nl *frn; 1070 struct nlmsghdr *nlh; 1071 u32 portid; 1072 1073 net = sock_net(skb->sk); 1074 nlh = nlmsg_hdr(skb); 1075 if (skb->len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len || 1076 nlmsg_len(nlh) < sizeof(*frn)) 1077 return; 1078 1079 skb = netlink_skb_clone(skb, GFP_KERNEL); 1080 if (!skb) 1081 return; 1082 nlh = nlmsg_hdr(skb); 1083 1084 frn = (struct fib_result_nl *) nlmsg_data(nlh); 1085 nl_fib_lookup(net, frn); 1086 1087 portid = NETLINK_CB(skb).portid; /* netlink portid */ 1088 NETLINK_CB(skb).portid = 0; /* from kernel */ 1089 NETLINK_CB(skb).dst_group = 0; /* unicast */ 1090 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT); 1091 } 1092 1093 static int __net_init nl_fib_lookup_init(struct net *net) 1094 { 1095 struct sock *sk; 1096 struct netlink_kernel_cfg cfg = { 1097 .input = nl_fib_input, 1098 }; 1099 1100 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg); 1101 if (!sk) 1102 return -EAFNOSUPPORT; 1103 net->ipv4.fibnl = sk; 1104 return 0; 1105 } 1106 1107 static void nl_fib_lookup_exit(struct net *net) 1108 { 1109 netlink_kernel_release(net->ipv4.fibnl); 1110 net->ipv4.fibnl = NULL; 1111 } 1112 1113 static void fib_disable_ip(struct net_device *dev, unsigned long event) 1114 { 1115 if (fib_sync_down_dev(dev, event)) 1116 fib_flush(dev_net(dev)); 1117 rt_cache_flush(dev_net(dev)); 1118 arp_ifdown(dev); 1119 } 1120 1121 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 1122 { 1123 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; 1124 struct net_device *dev = ifa->ifa_dev->dev; 1125 struct net *net = dev_net(dev); 1126 1127 switch (event) { 1128 case NETDEV_UP: 1129 fib_add_ifaddr(ifa); 1130 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1131 fib_sync_up(dev, RTNH_F_DEAD); 1132 #endif 1133 atomic_inc(&net->ipv4.dev_addr_genid); 1134 rt_cache_flush(dev_net(dev)); 1135 break; 1136 case NETDEV_DOWN: 1137 fib_del_ifaddr(ifa, NULL); 1138 atomic_inc(&net->ipv4.dev_addr_genid); 1139 if (!ifa->ifa_dev->ifa_list) { 1140 /* Last address was deleted from this interface. 1141 * Disable IP. 1142 */ 1143 fib_disable_ip(dev, event); 1144 } else { 1145 rt_cache_flush(dev_net(dev)); 1146 } 1147 break; 1148 } 1149 return NOTIFY_DONE; 1150 } 1151 1152 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 1153 { 1154 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1155 struct in_device *in_dev; 1156 struct net *net = dev_net(dev); 1157 unsigned int flags; 1158 1159 if (event == NETDEV_UNREGISTER) { 1160 fib_disable_ip(dev, event); 1161 rt_flush_dev(dev); 1162 return NOTIFY_DONE; 1163 } 1164 1165 in_dev = __in_dev_get_rtnl(dev); 1166 if (!in_dev) 1167 return NOTIFY_DONE; 1168 1169 switch (event) { 1170 case NETDEV_UP: 1171 for_ifa(in_dev) { 1172 fib_add_ifaddr(ifa); 1173 } endfor_ifa(in_dev); 1174 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1175 fib_sync_up(dev, RTNH_F_DEAD); 1176 #endif 1177 atomic_inc(&net->ipv4.dev_addr_genid); 1178 rt_cache_flush(net); 1179 break; 1180 case NETDEV_DOWN: 1181 fib_disable_ip(dev, event); 1182 break; 1183 case NETDEV_CHANGE: 1184 flags = dev_get_flags(dev); 1185 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 1186 fib_sync_up(dev, RTNH_F_LINKDOWN); 1187 else 1188 fib_sync_down_dev(dev, event); 1189 /* fall through */ 1190 case NETDEV_CHANGEMTU: 1191 rt_cache_flush(net); 1192 break; 1193 } 1194 return NOTIFY_DONE; 1195 } 1196 1197 static struct notifier_block fib_inetaddr_notifier = { 1198 .notifier_call = fib_inetaddr_event, 1199 }; 1200 1201 static struct notifier_block fib_netdev_notifier = { 1202 .notifier_call = fib_netdev_event, 1203 }; 1204 1205 static int __net_init ip_fib_net_init(struct net *net) 1206 { 1207 int err; 1208 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ; 1209 1210 /* Avoid false sharing : Use at least a full cache line */ 1211 size = max_t(size_t, size, L1_CACHE_BYTES); 1212 1213 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL); 1214 if (!net->ipv4.fib_table_hash) 1215 return -ENOMEM; 1216 1217 err = fib4_rules_init(net); 1218 if (err < 0) 1219 goto fail; 1220 return 0; 1221 1222 fail: 1223 kfree(net->ipv4.fib_table_hash); 1224 return err; 1225 } 1226 1227 static void ip_fib_net_exit(struct net *net) 1228 { 1229 unsigned int i; 1230 1231 rtnl_lock(); 1232 #ifdef CONFIG_IP_MULTIPLE_TABLES 1233 RCU_INIT_POINTER(net->ipv4.fib_local, NULL); 1234 RCU_INIT_POINTER(net->ipv4.fib_main, NULL); 1235 RCU_INIT_POINTER(net->ipv4.fib_default, NULL); 1236 #endif 1237 for (i = 0; i < FIB_TABLE_HASHSZ; i++) { 1238 struct hlist_head *head = &net->ipv4.fib_table_hash[i]; 1239 struct hlist_node *tmp; 1240 struct fib_table *tb; 1241 1242 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) { 1243 hlist_del(&tb->tb_hlist); 1244 fib_table_flush(tb); 1245 fib_free_table(tb); 1246 } 1247 } 1248 1249 #ifdef CONFIG_IP_MULTIPLE_TABLES 1250 fib4_rules_exit(net); 1251 #endif 1252 rtnl_unlock(); 1253 kfree(net->ipv4.fib_table_hash); 1254 } 1255 1256 static int __net_init fib_net_init(struct net *net) 1257 { 1258 int error; 1259 1260 #ifdef CONFIG_IP_ROUTE_CLASSID 1261 net->ipv4.fib_num_tclassid_users = 0; 1262 #endif 1263 error = ip_fib_net_init(net); 1264 if (error < 0) 1265 goto out; 1266 error = nl_fib_lookup_init(net); 1267 if (error < 0) 1268 goto out_nlfl; 1269 error = fib_proc_init(net); 1270 if (error < 0) 1271 goto out_proc; 1272 out: 1273 return error; 1274 1275 out_proc: 1276 nl_fib_lookup_exit(net); 1277 out_nlfl: 1278 ip_fib_net_exit(net); 1279 goto out; 1280 } 1281 1282 static void __net_exit fib_net_exit(struct net *net) 1283 { 1284 fib_proc_exit(net); 1285 nl_fib_lookup_exit(net); 1286 ip_fib_net_exit(net); 1287 } 1288 1289 static struct pernet_operations fib_net_ops = { 1290 .init = fib_net_init, 1291 .exit = fib_net_exit, 1292 }; 1293 1294 void __init ip_fib_init(void) 1295 { 1296 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL); 1297 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL); 1298 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL); 1299 1300 register_pernet_subsys(&fib_net_ops); 1301 register_netdevice_notifier(&fib_netdev_notifier); 1302 register_inetaddr_notifier(&fib_inetaddr_notifier); 1303 1304 fib_trie_init(); 1305 } 1306