1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * IPv4 Forwarding Information Base: FIB frontend. 7 * 8 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> 9 * 10 * This program is free software; you can redistribute it and/or 11 * modify it under the terms of the GNU General Public License 12 * as published by the Free Software Foundation; either version 13 * 2 of the License, or (at your option) any later version. 14 */ 15 16 #include <linux/module.h> 17 #include <asm/uaccess.h> 18 #include <linux/bitops.h> 19 #include <linux/capability.h> 20 #include <linux/types.h> 21 #include <linux/kernel.h> 22 #include <linux/mm.h> 23 #include <linux/string.h> 24 #include <linux/socket.h> 25 #include <linux/sockios.h> 26 #include <linux/errno.h> 27 #include <linux/in.h> 28 #include <linux/inet.h> 29 #include <linux/inetdevice.h> 30 #include <linux/netdevice.h> 31 #include <linux/if_addr.h> 32 #include <linux/if_arp.h> 33 #include <linux/skbuff.h> 34 #include <linux/cache.h> 35 #include <linux/init.h> 36 #include <linux/list.h> 37 #include <linux/slab.h> 38 39 #include <net/ip.h> 40 #include <net/protocol.h> 41 #include <net/route.h> 42 #include <net/tcp.h> 43 #include <net/sock.h> 44 #include <net/arp.h> 45 #include <net/ip_fib.h> 46 #include <net/rtnetlink.h> 47 #include <net/xfrm.h> 48 #include <net/l3mdev.h> 49 #include <trace/events/fib.h> 50 51 #ifndef CONFIG_IP_MULTIPLE_TABLES 52 53 static int __net_init fib4_rules_init(struct net *net) 54 { 55 struct fib_table *local_table, *main_table; 56 57 main_table = fib_trie_table(RT_TABLE_MAIN, NULL); 58 if (!main_table) 59 return -ENOMEM; 60 61 local_table = fib_trie_table(RT_TABLE_LOCAL, main_table); 62 if (!local_table) 63 goto fail; 64 65 hlist_add_head_rcu(&local_table->tb_hlist, 66 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]); 67 hlist_add_head_rcu(&main_table->tb_hlist, 68 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]); 69 return 0; 70 71 fail: 72 fib_free_table(main_table); 73 return -ENOMEM; 74 } 75 #else 76 77 struct fib_table *fib_new_table(struct net *net, u32 id) 78 { 79 struct fib_table *tb, *alias = NULL; 80 unsigned int h; 81 82 if (id == 0) 83 id = RT_TABLE_MAIN; 84 tb = fib_get_table(net, id); 85 if (tb) 86 return tb; 87 88 if (id == RT_TABLE_LOCAL) 89 alias = fib_new_table(net, RT_TABLE_MAIN); 90 91 tb = fib_trie_table(id, alias); 92 if (!tb) 93 return NULL; 94 95 switch (id) { 96 case RT_TABLE_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 EXPORT_SYMBOL_GPL(fib_new_table); 114 115 /* caller must hold either rtnl or rcu read lock */ 116 struct fib_table *fib_get_table(struct net *net, u32 id) 117 { 118 struct fib_table *tb; 119 struct hlist_head *head; 120 unsigned int h; 121 122 if (id == 0) 123 id = RT_TABLE_MAIN; 124 h = id & (FIB_TABLE_HASHSZ - 1); 125 126 head = &net->ipv4.fib_table_hash[h]; 127 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 128 if (tb->tb_id == id) 129 return tb; 130 } 131 return NULL; 132 } 133 #endif /* CONFIG_IP_MULTIPLE_TABLES */ 134 135 static void fib_replace_table(struct net *net, struct fib_table *old, 136 struct fib_table *new) 137 { 138 #ifdef CONFIG_IP_MULTIPLE_TABLES 139 switch (new->tb_id) { 140 case RT_TABLE_LOCAL: 141 rcu_assign_pointer(net->ipv4.fib_local, new); 142 break; 143 case RT_TABLE_MAIN: 144 rcu_assign_pointer(net->ipv4.fib_main, new); 145 break; 146 case RT_TABLE_DEFAULT: 147 rcu_assign_pointer(net->ipv4.fib_default, new); 148 break; 149 default: 150 break; 151 } 152 153 #endif 154 /* replace the old table in the hlist */ 155 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist); 156 } 157 158 int fib_unmerge(struct net *net) 159 { 160 struct fib_table *old, *new; 161 162 /* attempt to fetch local table if it has been allocated */ 163 old = fib_get_table(net, RT_TABLE_LOCAL); 164 if (!old) 165 return 0; 166 167 new = fib_trie_unmerge(old); 168 if (!new) 169 return -ENOMEM; 170 171 /* replace merged table with clean table */ 172 if (new != old) { 173 fib_replace_table(net, old, new); 174 fib_free_table(old); 175 } 176 177 return 0; 178 } 179 180 static void fib_flush(struct net *net) 181 { 182 int flushed = 0; 183 unsigned int h; 184 185 for (h = 0; h < FIB_TABLE_HASHSZ; h++) { 186 struct hlist_head *head = &net->ipv4.fib_table_hash[h]; 187 struct hlist_node *tmp; 188 struct fib_table *tb; 189 190 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) 191 flushed += fib_table_flush(tb); 192 } 193 194 if (flushed) 195 rt_cache_flush(net); 196 } 197 198 void fib_flush_external(struct net *net) 199 { 200 struct fib_table *tb; 201 struct hlist_head *head; 202 unsigned int h; 203 204 for (h = 0; h < FIB_TABLE_HASHSZ; h++) { 205 head = &net->ipv4.fib_table_hash[h]; 206 hlist_for_each_entry(tb, head, tb_hlist) 207 fib_table_flush_external(tb); 208 } 209 } 210 211 /* 212 * Find address type as if only "dev" was present in the system. If 213 * on_dev is NULL then all interfaces are taken into consideration. 214 */ 215 static inline unsigned int __inet_dev_addr_type(struct net *net, 216 const struct net_device *dev, 217 __be32 addr, u32 tb_id) 218 { 219 struct flowi4 fl4 = { .daddr = addr }; 220 struct fib_result res; 221 unsigned int ret = RTN_BROADCAST; 222 struct fib_table *table; 223 224 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr)) 225 return RTN_BROADCAST; 226 if (ipv4_is_multicast(addr)) 227 return RTN_MULTICAST; 228 229 rcu_read_lock(); 230 231 table = fib_get_table(net, tb_id); 232 if (table) { 233 ret = RTN_UNICAST; 234 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) { 235 if (!dev || dev == res.fi->fib_dev) 236 ret = res.type; 237 } 238 } 239 240 rcu_read_unlock(); 241 return ret; 242 } 243 244 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id) 245 { 246 return __inet_dev_addr_type(net, NULL, addr, tb_id); 247 } 248 EXPORT_SYMBOL(inet_addr_type_table); 249 250 unsigned int inet_addr_type(struct net *net, __be32 addr) 251 { 252 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL); 253 } 254 EXPORT_SYMBOL(inet_addr_type); 255 256 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev, 257 __be32 addr) 258 { 259 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL; 260 261 return __inet_dev_addr_type(net, dev, addr, rt_table); 262 } 263 EXPORT_SYMBOL(inet_dev_addr_type); 264 265 /* inet_addr_type with dev == NULL but using the table from a dev 266 * if one is associated 267 */ 268 unsigned int inet_addr_type_dev_table(struct net *net, 269 const struct net_device *dev, 270 __be32 addr) 271 { 272 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL; 273 274 return __inet_dev_addr_type(net, NULL, addr, rt_table); 275 } 276 EXPORT_SYMBOL(inet_addr_type_dev_table); 277 278 __be32 fib_compute_spec_dst(struct sk_buff *skb) 279 { 280 struct net_device *dev = skb->dev; 281 struct in_device *in_dev; 282 struct fib_result res; 283 struct rtable *rt; 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 struct flowi4 fl4 = { 300 .flowi4_iif = LOOPBACK_IFINDEX, 301 .daddr = ip_hdr(skb)->saddr, 302 .flowi4_tos = RT_TOS(ip_hdr(skb)->tos), 303 .flowi4_scope = scope, 304 .flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0, 305 }; 306 if (!fib_lookup(net, &fl4, &res, 0)) 307 return FIB_RES_PREFSRC(net, res); 308 } else { 309 scope = RT_SCOPE_LINK; 310 } 311 312 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope); 313 } 314 315 /* Given (packet source, input interface) and optional (dst, oif, tos): 316 * - (main) check, that source is valid i.e. not broadcast or our local 317 * address. 318 * - figure out what "logical" interface this packet arrived 319 * and calculate "specific destination" address. 320 * - check, that packet arrived from expected physical interface. 321 * called with rcu_read_lock() 322 */ 323 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 324 u8 tos, int oif, struct net_device *dev, 325 int rpf, struct in_device *idev, u32 *itag) 326 { 327 int ret, no_addr; 328 struct fib_result res; 329 struct flowi4 fl4; 330 struct net *net; 331 bool dev_match; 332 333 fl4.flowi4_oif = 0; 334 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev); 335 if (!fl4.flowi4_iif) 336 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX; 337 fl4.daddr = src; 338 fl4.saddr = dst; 339 fl4.flowi4_tos = tos; 340 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 341 fl4.flowi4_tun_key.tun_id = 0; 342 fl4.flowi4_flags = 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 (l3mdev_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 = l3mdev_fib_table(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 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 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 /* if the device has been deleted, we don't perform 909 * address promotion 910 */ 911 if (!in_dev->dead) 912 pr_warn("%s: bug: prim == NULL\n", __func__); 913 return; 914 } 915 if (iprim && iprim != prim) { 916 pr_warn("%s: bug: iprim != prim\n", __func__); 917 return; 918 } 919 } else if (!ipv4_is_zeronet(any) && 920 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) { 921 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 922 fib_magic(RTM_DELROUTE, 923 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 924 any, ifa->ifa_prefixlen, prim); 925 subnet = 1; 926 } 927 928 if (in_dev->dead) 929 goto no_promotions; 930 931 /* Deletion is more complicated than add. 932 * We should take care of not to delete too much :-) 933 * 934 * Scan address list to be sure that addresses are really gone. 935 */ 936 937 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) { 938 if (ifa1 == ifa) { 939 /* promotion, keep the IP */ 940 gone = 0; 941 continue; 942 } 943 /* Ignore IFAs from our subnet */ 944 if (iprim && ifa1->ifa_mask == iprim->ifa_mask && 945 inet_ifa_match(ifa1->ifa_address, iprim)) 946 continue; 947 948 /* Ignore ifa1 if it uses different primary IP (prefsrc) */ 949 if (ifa1->ifa_flags & IFA_F_SECONDARY) { 950 /* Another address from our subnet? */ 951 if (ifa1->ifa_mask == prim->ifa_mask && 952 inet_ifa_match(ifa1->ifa_address, prim)) 953 prim1 = prim; 954 else { 955 /* We reached the secondaries, so 956 * same_prefsrc should be determined. 957 */ 958 if (!same_prefsrc) 959 continue; 960 /* Search new prim1 if ifa1 is not 961 * using the current prim1 962 */ 963 if (!prim1 || 964 ifa1->ifa_mask != prim1->ifa_mask || 965 !inet_ifa_match(ifa1->ifa_address, prim1)) 966 prim1 = inet_ifa_byprefix(in_dev, 967 ifa1->ifa_address, 968 ifa1->ifa_mask); 969 if (!prim1) 970 continue; 971 if (prim1->ifa_local != prim->ifa_local) 972 continue; 973 } 974 } else { 975 if (prim->ifa_local != ifa1->ifa_local) 976 continue; 977 prim1 = ifa1; 978 if (prim != prim1) 979 same_prefsrc = 1; 980 } 981 if (ifa->ifa_local == ifa1->ifa_local) 982 ok |= LOCAL_OK; 983 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 984 ok |= BRD_OK; 985 if (brd == ifa1->ifa_broadcast) 986 ok |= BRD1_OK; 987 if (any == ifa1->ifa_broadcast) 988 ok |= BRD0_OK; 989 /* primary has network specific broadcasts */ 990 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) { 991 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask; 992 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask; 993 994 if (!ipv4_is_zeronet(any1)) { 995 if (ifa->ifa_broadcast == brd1 || 996 ifa->ifa_broadcast == any1) 997 ok |= BRD_OK; 998 if (brd == brd1 || brd == any1) 999 ok |= BRD1_OK; 1000 if (any == brd1 || any == any1) 1001 ok |= BRD0_OK; 1002 } 1003 } 1004 } 1005 1006 no_promotions: 1007 if (!(ok & BRD_OK)) 1008 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 1009 if (subnet && ifa->ifa_prefixlen < 31) { 1010 if (!(ok & BRD1_OK)) 1011 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim); 1012 if (!(ok & BRD0_OK)) 1013 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim); 1014 } 1015 if (!(ok & LOCAL_OK)) { 1016 unsigned int addr_type; 1017 1018 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim); 1019 1020 /* Check, that this local address finally disappeared. */ 1021 addr_type = inet_addr_type_dev_table(dev_net(dev), dev, 1022 ifa->ifa_local); 1023 if (gone && addr_type != RTN_LOCAL) { 1024 /* And the last, but not the least thing. 1025 * We must flush stray FIB entries. 1026 * 1027 * First of all, we scan fib_info list searching 1028 * for stray nexthop entries, then ignite fib_flush. 1029 */ 1030 if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local)) 1031 fib_flush(dev_net(dev)); 1032 } 1033 } 1034 #undef LOCAL_OK 1035 #undef BRD_OK 1036 #undef BRD0_OK 1037 #undef BRD1_OK 1038 } 1039 1040 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn) 1041 { 1042 1043 struct fib_result res; 1044 struct flowi4 fl4 = { 1045 .flowi4_mark = frn->fl_mark, 1046 .daddr = frn->fl_addr, 1047 .flowi4_tos = frn->fl_tos, 1048 .flowi4_scope = frn->fl_scope, 1049 }; 1050 struct fib_table *tb; 1051 1052 rcu_read_lock(); 1053 1054 tb = fib_get_table(net, frn->tb_id_in); 1055 1056 frn->err = -ENOENT; 1057 if (tb) { 1058 local_bh_disable(); 1059 1060 frn->tb_id = tb->tb_id; 1061 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 1062 1063 if (!frn->err) { 1064 frn->prefixlen = res.prefixlen; 1065 frn->nh_sel = res.nh_sel; 1066 frn->type = res.type; 1067 frn->scope = res.scope; 1068 } 1069 local_bh_enable(); 1070 } 1071 1072 rcu_read_unlock(); 1073 } 1074 1075 static void nl_fib_input(struct sk_buff *skb) 1076 { 1077 struct net *net; 1078 struct fib_result_nl *frn; 1079 struct nlmsghdr *nlh; 1080 u32 portid; 1081 1082 net = sock_net(skb->sk); 1083 nlh = nlmsg_hdr(skb); 1084 if (skb->len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len || 1085 nlmsg_len(nlh) < sizeof(*frn)) 1086 return; 1087 1088 skb = netlink_skb_clone(skb, GFP_KERNEL); 1089 if (!skb) 1090 return; 1091 nlh = nlmsg_hdr(skb); 1092 1093 frn = (struct fib_result_nl *) nlmsg_data(nlh); 1094 nl_fib_lookup(net, frn); 1095 1096 portid = NETLINK_CB(skb).portid; /* netlink portid */ 1097 NETLINK_CB(skb).portid = 0; /* from kernel */ 1098 NETLINK_CB(skb).dst_group = 0; /* unicast */ 1099 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT); 1100 } 1101 1102 static int __net_init nl_fib_lookup_init(struct net *net) 1103 { 1104 struct sock *sk; 1105 struct netlink_kernel_cfg cfg = { 1106 .input = nl_fib_input, 1107 }; 1108 1109 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg); 1110 if (!sk) 1111 return -EAFNOSUPPORT; 1112 net->ipv4.fibnl = sk; 1113 return 0; 1114 } 1115 1116 static void nl_fib_lookup_exit(struct net *net) 1117 { 1118 netlink_kernel_release(net->ipv4.fibnl); 1119 net->ipv4.fibnl = NULL; 1120 } 1121 1122 static void fib_disable_ip(struct net_device *dev, unsigned long event, 1123 bool force) 1124 { 1125 if (fib_sync_down_dev(dev, event, force)) 1126 fib_flush(dev_net(dev)); 1127 rt_cache_flush(dev_net(dev)); 1128 arp_ifdown(dev); 1129 } 1130 1131 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 1132 { 1133 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; 1134 struct net_device *dev = ifa->ifa_dev->dev; 1135 struct net *net = dev_net(dev); 1136 1137 switch (event) { 1138 case NETDEV_UP: 1139 fib_add_ifaddr(ifa); 1140 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1141 fib_sync_up(dev, RTNH_F_DEAD); 1142 #endif 1143 atomic_inc(&net->ipv4.dev_addr_genid); 1144 rt_cache_flush(dev_net(dev)); 1145 break; 1146 case NETDEV_DOWN: 1147 fib_del_ifaddr(ifa, NULL); 1148 atomic_inc(&net->ipv4.dev_addr_genid); 1149 if (!ifa->ifa_dev->ifa_list) { 1150 /* Last address was deleted from this interface. 1151 * Disable IP. 1152 */ 1153 fib_disable_ip(dev, event, true); 1154 } else { 1155 rt_cache_flush(dev_net(dev)); 1156 } 1157 break; 1158 } 1159 return NOTIFY_DONE; 1160 } 1161 1162 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 1163 { 1164 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1165 struct netdev_notifier_changeupper_info *info; 1166 struct in_device *in_dev; 1167 struct net *net = dev_net(dev); 1168 unsigned int flags; 1169 1170 if (event == NETDEV_UNREGISTER) { 1171 fib_disable_ip(dev, event, true); 1172 rt_flush_dev(dev); 1173 return NOTIFY_DONE; 1174 } 1175 1176 in_dev = __in_dev_get_rtnl(dev); 1177 if (!in_dev) 1178 return NOTIFY_DONE; 1179 1180 switch (event) { 1181 case NETDEV_UP: 1182 for_ifa(in_dev) { 1183 fib_add_ifaddr(ifa); 1184 } endfor_ifa(in_dev); 1185 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1186 fib_sync_up(dev, RTNH_F_DEAD); 1187 #endif 1188 atomic_inc(&net->ipv4.dev_addr_genid); 1189 rt_cache_flush(net); 1190 break; 1191 case NETDEV_DOWN: 1192 fib_disable_ip(dev, event, false); 1193 break; 1194 case NETDEV_CHANGE: 1195 flags = dev_get_flags(dev); 1196 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 1197 fib_sync_up(dev, RTNH_F_LINKDOWN); 1198 else 1199 fib_sync_down_dev(dev, event, false); 1200 /* fall through */ 1201 case NETDEV_CHANGEMTU: 1202 rt_cache_flush(net); 1203 break; 1204 case NETDEV_CHANGEUPPER: 1205 info = ptr; 1206 /* flush all routes if dev is linked to or unlinked from 1207 * an L3 master device (e.g., VRF) 1208 */ 1209 if (info->upper_dev && netif_is_l3_master(info->upper_dev)) 1210 fib_disable_ip(dev, NETDEV_DOWN, true); 1211 break; 1212 } 1213 return NOTIFY_DONE; 1214 } 1215 1216 static struct notifier_block fib_inetaddr_notifier = { 1217 .notifier_call = fib_inetaddr_event, 1218 }; 1219 1220 static struct notifier_block fib_netdev_notifier = { 1221 .notifier_call = fib_netdev_event, 1222 }; 1223 1224 static int __net_init ip_fib_net_init(struct net *net) 1225 { 1226 int err; 1227 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ; 1228 1229 /* Avoid false sharing : Use at least a full cache line */ 1230 size = max_t(size_t, size, L1_CACHE_BYTES); 1231 1232 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL); 1233 if (!net->ipv4.fib_table_hash) 1234 return -ENOMEM; 1235 1236 err = fib4_rules_init(net); 1237 if (err < 0) 1238 goto fail; 1239 return 0; 1240 1241 fail: 1242 kfree(net->ipv4.fib_table_hash); 1243 return err; 1244 } 1245 1246 static void ip_fib_net_exit(struct net *net) 1247 { 1248 unsigned int i; 1249 1250 rtnl_lock(); 1251 #ifdef CONFIG_IP_MULTIPLE_TABLES 1252 RCU_INIT_POINTER(net->ipv4.fib_local, NULL); 1253 RCU_INIT_POINTER(net->ipv4.fib_main, NULL); 1254 RCU_INIT_POINTER(net->ipv4.fib_default, NULL); 1255 #endif 1256 for (i = 0; i < FIB_TABLE_HASHSZ; i++) { 1257 struct hlist_head *head = &net->ipv4.fib_table_hash[i]; 1258 struct hlist_node *tmp; 1259 struct fib_table *tb; 1260 1261 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) { 1262 hlist_del(&tb->tb_hlist); 1263 fib_table_flush(tb); 1264 fib_free_table(tb); 1265 } 1266 } 1267 1268 #ifdef CONFIG_IP_MULTIPLE_TABLES 1269 fib4_rules_exit(net); 1270 #endif 1271 rtnl_unlock(); 1272 kfree(net->ipv4.fib_table_hash); 1273 } 1274 1275 static int __net_init fib_net_init(struct net *net) 1276 { 1277 int error; 1278 1279 #ifdef CONFIG_IP_ROUTE_CLASSID 1280 net->ipv4.fib_num_tclassid_users = 0; 1281 #endif 1282 error = ip_fib_net_init(net); 1283 if (error < 0) 1284 goto out; 1285 error = nl_fib_lookup_init(net); 1286 if (error < 0) 1287 goto out_nlfl; 1288 error = fib_proc_init(net); 1289 if (error < 0) 1290 goto out_proc; 1291 out: 1292 return error; 1293 1294 out_proc: 1295 nl_fib_lookup_exit(net); 1296 out_nlfl: 1297 ip_fib_net_exit(net); 1298 goto out; 1299 } 1300 1301 static void __net_exit fib_net_exit(struct net *net) 1302 { 1303 fib_proc_exit(net); 1304 nl_fib_lookup_exit(net); 1305 ip_fib_net_exit(net); 1306 } 1307 1308 static struct pernet_operations fib_net_ops = { 1309 .init = fib_net_init, 1310 .exit = fib_net_exit, 1311 }; 1312 1313 void __init ip_fib_init(void) 1314 { 1315 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL); 1316 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL); 1317 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL); 1318 1319 register_pernet_subsys(&fib_net_ops); 1320 register_netdevice_notifier(&fib_netdev_notifier); 1321 register_inetaddr_notifier(&fib_inetaddr_notifier); 1322 1323 fib_trie_init(); 1324 } 1325