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_MAIN: 97 rcu_assign_pointer(net->ipv4.fib_main, tb); 98 break; 99 case RT_TABLE_DEFAULT: 100 rcu_assign_pointer(net->ipv4.fib_default, tb); 101 break; 102 default: 103 break; 104 } 105 106 h = id & (FIB_TABLE_HASHSZ - 1); 107 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]); 108 return tb; 109 } 110 EXPORT_SYMBOL_GPL(fib_new_table); 111 112 /* caller must hold either rtnl or rcu read lock */ 113 struct fib_table *fib_get_table(struct net *net, u32 id) 114 { 115 struct fib_table *tb; 116 struct hlist_head *head; 117 unsigned int h; 118 119 if (id == 0) 120 id = RT_TABLE_MAIN; 121 h = id & (FIB_TABLE_HASHSZ - 1); 122 123 head = &net->ipv4.fib_table_hash[h]; 124 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 125 if (tb->tb_id == id) 126 return tb; 127 } 128 return NULL; 129 } 130 #endif /* CONFIG_IP_MULTIPLE_TABLES */ 131 132 static void fib_replace_table(struct net *net, struct fib_table *old, 133 struct fib_table *new) 134 { 135 #ifdef CONFIG_IP_MULTIPLE_TABLES 136 switch (new->tb_id) { 137 case RT_TABLE_MAIN: 138 rcu_assign_pointer(net->ipv4.fib_main, new); 139 break; 140 case RT_TABLE_DEFAULT: 141 rcu_assign_pointer(net->ipv4.fib_default, new); 142 break; 143 default: 144 break; 145 } 146 147 #endif 148 /* replace the old table in the hlist */ 149 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist); 150 } 151 152 int fib_unmerge(struct net *net) 153 { 154 struct fib_table *old, *new; 155 156 /* attempt to fetch local table if it has been allocated */ 157 old = fib_get_table(net, RT_TABLE_LOCAL); 158 if (!old) 159 return 0; 160 161 new = fib_trie_unmerge(old); 162 if (!new) 163 return -ENOMEM; 164 165 /* replace merged table with clean table */ 166 if (new != old) { 167 fib_replace_table(net, old, new); 168 fib_free_table(old); 169 } 170 171 return 0; 172 } 173 174 static void fib_flush(struct net *net) 175 { 176 int flushed = 0; 177 unsigned int h; 178 179 for (h = 0; h < FIB_TABLE_HASHSZ; h++) { 180 struct hlist_head *head = &net->ipv4.fib_table_hash[h]; 181 struct hlist_node *tmp; 182 struct fib_table *tb; 183 184 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) 185 flushed += fib_table_flush(tb); 186 } 187 188 if (flushed) 189 rt_cache_flush(net); 190 } 191 192 void fib_flush_external(struct net *net) 193 { 194 struct fib_table *tb; 195 struct hlist_head *head; 196 unsigned int h; 197 198 for (h = 0; h < FIB_TABLE_HASHSZ; h++) { 199 head = &net->ipv4.fib_table_hash[h]; 200 hlist_for_each_entry(tb, head, tb_hlist) 201 fib_table_flush_external(tb); 202 } 203 } 204 205 /* 206 * Find address type as if only "dev" was present in the system. If 207 * on_dev is NULL then all interfaces are taken into consideration. 208 */ 209 static inline unsigned int __inet_dev_addr_type(struct net *net, 210 const struct net_device *dev, 211 __be32 addr, u32 tb_id) 212 { 213 struct flowi4 fl4 = { .daddr = addr }; 214 struct fib_result res; 215 unsigned int ret = RTN_BROADCAST; 216 struct fib_table *table; 217 218 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr)) 219 return RTN_BROADCAST; 220 if (ipv4_is_multicast(addr)) 221 return RTN_MULTICAST; 222 223 rcu_read_lock(); 224 225 table = fib_get_table(net, tb_id); 226 if (table) { 227 ret = RTN_UNICAST; 228 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) { 229 if (!dev || dev == res.fi->fib_dev) 230 ret = res.type; 231 } 232 } 233 234 rcu_read_unlock(); 235 return ret; 236 } 237 238 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id) 239 { 240 return __inet_dev_addr_type(net, NULL, addr, tb_id); 241 } 242 EXPORT_SYMBOL(inet_addr_type_table); 243 244 unsigned int inet_addr_type(struct net *net, __be32 addr) 245 { 246 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL); 247 } 248 EXPORT_SYMBOL(inet_addr_type); 249 250 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev, 251 __be32 addr) 252 { 253 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL; 254 255 return __inet_dev_addr_type(net, dev, addr, rt_table); 256 } 257 EXPORT_SYMBOL(inet_dev_addr_type); 258 259 /* inet_addr_type with dev == NULL but using the table from a dev 260 * if one is associated 261 */ 262 unsigned int inet_addr_type_dev_table(struct net *net, 263 const struct net_device *dev, 264 __be32 addr) 265 { 266 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL; 267 268 return __inet_dev_addr_type(net, NULL, addr, rt_table); 269 } 270 EXPORT_SYMBOL(inet_addr_type_dev_table); 271 272 __be32 fib_compute_spec_dst(struct sk_buff *skb) 273 { 274 struct net_device *dev = skb->dev; 275 struct in_device *in_dev; 276 struct fib_result res; 277 struct rtable *rt; 278 struct net *net; 279 int scope; 280 281 rt = skb_rtable(skb); 282 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) == 283 RTCF_LOCAL) 284 return ip_hdr(skb)->daddr; 285 286 in_dev = __in_dev_get_rcu(dev); 287 BUG_ON(!in_dev); 288 289 net = dev_net(dev); 290 291 scope = RT_SCOPE_UNIVERSE; 292 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) { 293 struct flowi4 fl4 = { 294 .flowi4_iif = LOOPBACK_IFINDEX, 295 .daddr = ip_hdr(skb)->saddr, 296 .flowi4_tos = RT_TOS(ip_hdr(skb)->tos), 297 .flowi4_scope = scope, 298 .flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0, 299 }; 300 if (!fib_lookup(net, &fl4, &res, 0)) 301 return FIB_RES_PREFSRC(net, res); 302 } else { 303 scope = RT_SCOPE_LINK; 304 } 305 306 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope); 307 } 308 309 /* Given (packet source, input interface) and optional (dst, oif, tos): 310 * - (main) check, that source is valid i.e. not broadcast or our local 311 * address. 312 * - figure out what "logical" interface this packet arrived 313 * and calculate "specific destination" address. 314 * - check, that packet arrived from expected physical interface. 315 * called with rcu_read_lock() 316 */ 317 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 318 u8 tos, int oif, struct net_device *dev, 319 int rpf, struct in_device *idev, u32 *itag) 320 { 321 int ret, no_addr; 322 struct fib_result res; 323 struct flowi4 fl4; 324 struct net *net; 325 bool dev_match; 326 327 fl4.flowi4_oif = 0; 328 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev); 329 if (!fl4.flowi4_iif) 330 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX; 331 fl4.daddr = src; 332 fl4.saddr = dst; 333 fl4.flowi4_tos = tos; 334 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 335 fl4.flowi4_tun_key.tun_id = 0; 336 fl4.flowi4_flags = 0; 337 338 no_addr = idev->ifa_list == NULL; 339 340 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0; 341 342 trace_fib_validate_source(dev, &fl4); 343 344 net = dev_net(dev); 345 if (fib_lookup(net, &fl4, &res, 0)) 346 goto last_resort; 347 if (res.type != RTN_UNICAST && 348 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev))) 349 goto e_inval; 350 if (!rpf && !fib_num_tclassid_users(dev_net(dev)) && 351 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) 352 goto last_resort; 353 fib_combine_itag(itag, &res); 354 dev_match = false; 355 356 #ifdef CONFIG_IP_ROUTE_MULTIPATH 357 for (ret = 0; ret < res.fi->fib_nhs; ret++) { 358 struct fib_nh *nh = &res.fi->fib_nh[ret]; 359 360 if (nh->nh_dev == dev) { 361 dev_match = true; 362 break; 363 } else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) { 364 dev_match = true; 365 break; 366 } 367 } 368 #else 369 if (FIB_RES_DEV(res) == dev) 370 dev_match = true; 371 #endif 372 if (dev_match) { 373 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 374 return ret; 375 } 376 if (no_addr) 377 goto last_resort; 378 if (rpf == 1) 379 goto e_rpf; 380 fl4.flowi4_oif = dev->ifindex; 381 382 ret = 0; 383 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) { 384 if (res.type == RTN_UNICAST) 385 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 386 } 387 return ret; 388 389 last_resort: 390 if (rpf) 391 goto e_rpf; 392 *itag = 0; 393 return 0; 394 395 e_inval: 396 return -EINVAL; 397 e_rpf: 398 return -EXDEV; 399 } 400 401 /* Ignore rp_filter for packets protected by IPsec. */ 402 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 403 u8 tos, int oif, struct net_device *dev, 404 struct in_device *idev, u32 *itag) 405 { 406 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev); 407 408 if (!r && !fib_num_tclassid_users(dev_net(dev)) && 409 IN_DEV_ACCEPT_LOCAL(idev) && 410 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) { 411 *itag = 0; 412 return 0; 413 } 414 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag); 415 } 416 417 static inline __be32 sk_extract_addr(struct sockaddr *addr) 418 { 419 return ((struct sockaddr_in *) addr)->sin_addr.s_addr; 420 } 421 422 static int put_rtax(struct nlattr *mx, int len, int type, u32 value) 423 { 424 struct nlattr *nla; 425 426 nla = (struct nlattr *) ((char *) mx + len); 427 nla->nla_type = type; 428 nla->nla_len = nla_attr_size(4); 429 *(u32 *) nla_data(nla) = value; 430 431 return len + nla_total_size(4); 432 } 433 434 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt, 435 struct fib_config *cfg) 436 { 437 __be32 addr; 438 int plen; 439 440 memset(cfg, 0, sizeof(*cfg)); 441 cfg->fc_nlinfo.nl_net = net; 442 443 if (rt->rt_dst.sa_family != AF_INET) 444 return -EAFNOSUPPORT; 445 446 /* 447 * Check mask for validity: 448 * a) it must be contiguous. 449 * b) destination must have all host bits clear. 450 * c) if application forgot to set correct family (AF_INET), 451 * reject request unless it is absolutely clear i.e. 452 * both family and mask are zero. 453 */ 454 plen = 32; 455 addr = sk_extract_addr(&rt->rt_dst); 456 if (!(rt->rt_flags & RTF_HOST)) { 457 __be32 mask = sk_extract_addr(&rt->rt_genmask); 458 459 if (rt->rt_genmask.sa_family != AF_INET) { 460 if (mask || rt->rt_genmask.sa_family) 461 return -EAFNOSUPPORT; 462 } 463 464 if (bad_mask(mask, addr)) 465 return -EINVAL; 466 467 plen = inet_mask_len(mask); 468 } 469 470 cfg->fc_dst_len = plen; 471 cfg->fc_dst = addr; 472 473 if (cmd != SIOCDELRT) { 474 cfg->fc_nlflags = NLM_F_CREATE; 475 cfg->fc_protocol = RTPROT_BOOT; 476 } 477 478 if (rt->rt_metric) 479 cfg->fc_priority = rt->rt_metric - 1; 480 481 if (rt->rt_flags & RTF_REJECT) { 482 cfg->fc_scope = RT_SCOPE_HOST; 483 cfg->fc_type = RTN_UNREACHABLE; 484 return 0; 485 } 486 487 cfg->fc_scope = RT_SCOPE_NOWHERE; 488 cfg->fc_type = RTN_UNICAST; 489 490 if (rt->rt_dev) { 491 char *colon; 492 struct net_device *dev; 493 char devname[IFNAMSIZ]; 494 495 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1)) 496 return -EFAULT; 497 498 devname[IFNAMSIZ-1] = 0; 499 colon = strchr(devname, ':'); 500 if (colon) 501 *colon = 0; 502 dev = __dev_get_by_name(net, devname); 503 if (!dev) 504 return -ENODEV; 505 cfg->fc_oif = dev->ifindex; 506 if (colon) { 507 struct in_ifaddr *ifa; 508 struct in_device *in_dev = __in_dev_get_rtnl(dev); 509 if (!in_dev) 510 return -ENODEV; 511 *colon = ':'; 512 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) 513 if (strcmp(ifa->ifa_label, devname) == 0) 514 break; 515 if (!ifa) 516 return -ENODEV; 517 cfg->fc_prefsrc = ifa->ifa_local; 518 } 519 } 520 521 addr = sk_extract_addr(&rt->rt_gateway); 522 if (rt->rt_gateway.sa_family == AF_INET && addr) { 523 unsigned int addr_type; 524 525 cfg->fc_gw = addr; 526 addr_type = inet_addr_type_table(net, addr, cfg->fc_table); 527 if (rt->rt_flags & RTF_GATEWAY && 528 addr_type == RTN_UNICAST) 529 cfg->fc_scope = RT_SCOPE_UNIVERSE; 530 } 531 532 if (cmd == SIOCDELRT) 533 return 0; 534 535 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw) 536 return -EINVAL; 537 538 if (cfg->fc_scope == RT_SCOPE_NOWHERE) 539 cfg->fc_scope = RT_SCOPE_LINK; 540 541 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) { 542 struct nlattr *mx; 543 int len = 0; 544 545 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL); 546 if (!mx) 547 return -ENOMEM; 548 549 if (rt->rt_flags & RTF_MTU) 550 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40); 551 552 if (rt->rt_flags & RTF_WINDOW) 553 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window); 554 555 if (rt->rt_flags & RTF_IRTT) 556 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3); 557 558 cfg->fc_mx = mx; 559 cfg->fc_mx_len = len; 560 } 561 562 return 0; 563 } 564 565 /* 566 * Handle IP routing ioctl calls. 567 * These are used to manipulate the routing tables 568 */ 569 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg) 570 { 571 struct fib_config cfg; 572 struct rtentry rt; 573 int err; 574 575 switch (cmd) { 576 case SIOCADDRT: /* Add a route */ 577 case SIOCDELRT: /* Delete a route */ 578 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 579 return -EPERM; 580 581 if (copy_from_user(&rt, arg, sizeof(rt))) 582 return -EFAULT; 583 584 rtnl_lock(); 585 err = rtentry_to_fib_config(net, cmd, &rt, &cfg); 586 if (err == 0) { 587 struct fib_table *tb; 588 589 if (cmd == SIOCDELRT) { 590 tb = fib_get_table(net, cfg.fc_table); 591 if (tb) 592 err = fib_table_delete(tb, &cfg); 593 else 594 err = -ESRCH; 595 } else { 596 tb = fib_new_table(net, cfg.fc_table); 597 if (tb) 598 err = fib_table_insert(tb, &cfg); 599 else 600 err = -ENOBUFS; 601 } 602 603 /* allocated by rtentry_to_fib_config() */ 604 kfree(cfg.fc_mx); 605 } 606 rtnl_unlock(); 607 return err; 608 } 609 return -EINVAL; 610 } 611 612 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = { 613 [RTA_DST] = { .type = NLA_U32 }, 614 [RTA_SRC] = { .type = NLA_U32 }, 615 [RTA_IIF] = { .type = NLA_U32 }, 616 [RTA_OIF] = { .type = NLA_U32 }, 617 [RTA_GATEWAY] = { .type = NLA_U32 }, 618 [RTA_PRIORITY] = { .type = NLA_U32 }, 619 [RTA_PREFSRC] = { .type = NLA_U32 }, 620 [RTA_METRICS] = { .type = NLA_NESTED }, 621 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) }, 622 [RTA_FLOW] = { .type = NLA_U32 }, 623 [RTA_ENCAP_TYPE] = { .type = NLA_U16 }, 624 [RTA_ENCAP] = { .type = NLA_NESTED }, 625 }; 626 627 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb, 628 struct nlmsghdr *nlh, struct fib_config *cfg) 629 { 630 struct nlattr *attr; 631 int err, remaining; 632 struct rtmsg *rtm; 633 634 err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy); 635 if (err < 0) 636 goto errout; 637 638 memset(cfg, 0, sizeof(*cfg)); 639 640 rtm = nlmsg_data(nlh); 641 cfg->fc_dst_len = rtm->rtm_dst_len; 642 cfg->fc_tos = rtm->rtm_tos; 643 cfg->fc_table = rtm->rtm_table; 644 cfg->fc_protocol = rtm->rtm_protocol; 645 cfg->fc_scope = rtm->rtm_scope; 646 cfg->fc_type = rtm->rtm_type; 647 cfg->fc_flags = rtm->rtm_flags; 648 cfg->fc_nlflags = nlh->nlmsg_flags; 649 650 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid; 651 cfg->fc_nlinfo.nlh = nlh; 652 cfg->fc_nlinfo.nl_net = net; 653 654 if (cfg->fc_type > RTN_MAX) { 655 err = -EINVAL; 656 goto errout; 657 } 658 659 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) { 660 switch (nla_type(attr)) { 661 case RTA_DST: 662 cfg->fc_dst = nla_get_be32(attr); 663 break; 664 case RTA_OIF: 665 cfg->fc_oif = nla_get_u32(attr); 666 break; 667 case RTA_GATEWAY: 668 cfg->fc_gw = nla_get_be32(attr); 669 break; 670 case RTA_PRIORITY: 671 cfg->fc_priority = nla_get_u32(attr); 672 break; 673 case RTA_PREFSRC: 674 cfg->fc_prefsrc = nla_get_be32(attr); 675 break; 676 case RTA_METRICS: 677 cfg->fc_mx = nla_data(attr); 678 cfg->fc_mx_len = nla_len(attr); 679 break; 680 case RTA_MULTIPATH: 681 cfg->fc_mp = nla_data(attr); 682 cfg->fc_mp_len = nla_len(attr); 683 break; 684 case RTA_FLOW: 685 cfg->fc_flow = nla_get_u32(attr); 686 break; 687 case RTA_TABLE: 688 cfg->fc_table = nla_get_u32(attr); 689 break; 690 case RTA_ENCAP: 691 cfg->fc_encap = attr; 692 break; 693 case RTA_ENCAP_TYPE: 694 cfg->fc_encap_type = nla_get_u16(attr); 695 break; 696 } 697 } 698 699 return 0; 700 errout: 701 return err; 702 } 703 704 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh) 705 { 706 struct net *net = sock_net(skb->sk); 707 struct fib_config cfg; 708 struct fib_table *tb; 709 int err; 710 711 err = rtm_to_fib_config(net, skb, nlh, &cfg); 712 if (err < 0) 713 goto errout; 714 715 tb = fib_get_table(net, cfg.fc_table); 716 if (!tb) { 717 err = -ESRCH; 718 goto errout; 719 } 720 721 err = fib_table_delete(tb, &cfg); 722 errout: 723 return err; 724 } 725 726 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh) 727 { 728 struct net *net = sock_net(skb->sk); 729 struct fib_config cfg; 730 struct fib_table *tb; 731 int err; 732 733 err = rtm_to_fib_config(net, skb, nlh, &cfg); 734 if (err < 0) 735 goto errout; 736 737 tb = fib_new_table(net, cfg.fc_table); 738 if (!tb) { 739 err = -ENOBUFS; 740 goto errout; 741 } 742 743 err = fib_table_insert(tb, &cfg); 744 errout: 745 return err; 746 } 747 748 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 749 { 750 struct net *net = sock_net(skb->sk); 751 unsigned int h, s_h; 752 unsigned int e = 0, s_e; 753 struct fib_table *tb; 754 struct hlist_head *head; 755 int dumped = 0; 756 757 if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) && 758 ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED) 759 return skb->len; 760 761 s_h = cb->args[0]; 762 s_e = cb->args[1]; 763 764 rcu_read_lock(); 765 766 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) { 767 e = 0; 768 head = &net->ipv4.fib_table_hash[h]; 769 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 770 if (e < s_e) 771 goto next; 772 if (dumped) 773 memset(&cb->args[2], 0, sizeof(cb->args) - 774 2 * sizeof(cb->args[0])); 775 if (fib_table_dump(tb, skb, cb) < 0) 776 goto out; 777 dumped = 1; 778 next: 779 e++; 780 } 781 } 782 out: 783 rcu_read_unlock(); 784 785 cb->args[1] = e; 786 cb->args[0] = h; 787 788 return skb->len; 789 } 790 791 /* Prepare and feed intra-kernel routing request. 792 * Really, it should be netlink message, but :-( netlink 793 * can be not configured, so that we feed it directly 794 * to fib engine. It is legal, because all events occur 795 * only when netlink is already locked. 796 */ 797 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa) 798 { 799 struct net *net = dev_net(ifa->ifa_dev->dev); 800 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev); 801 struct fib_table *tb; 802 struct fib_config cfg = { 803 .fc_protocol = RTPROT_KERNEL, 804 .fc_type = type, 805 .fc_dst = dst, 806 .fc_dst_len = dst_len, 807 .fc_prefsrc = ifa->ifa_local, 808 .fc_oif = ifa->ifa_dev->dev->ifindex, 809 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND, 810 .fc_nlinfo = { 811 .nl_net = net, 812 }, 813 }; 814 815 if (!tb_id) 816 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL; 817 818 tb = fib_new_table(net, tb_id); 819 if (!tb) 820 return; 821 822 cfg.fc_table = tb->tb_id; 823 824 if (type != RTN_LOCAL) 825 cfg.fc_scope = RT_SCOPE_LINK; 826 else 827 cfg.fc_scope = RT_SCOPE_HOST; 828 829 if (cmd == RTM_NEWROUTE) 830 fib_table_insert(tb, &cfg); 831 else 832 fib_table_delete(tb, &cfg); 833 } 834 835 void fib_add_ifaddr(struct in_ifaddr *ifa) 836 { 837 struct in_device *in_dev = ifa->ifa_dev; 838 struct net_device *dev = in_dev->dev; 839 struct in_ifaddr *prim = ifa; 840 __be32 mask = ifa->ifa_mask; 841 __be32 addr = ifa->ifa_local; 842 __be32 prefix = ifa->ifa_address & mask; 843 844 if (ifa->ifa_flags & IFA_F_SECONDARY) { 845 prim = inet_ifa_byprefix(in_dev, prefix, mask); 846 if (!prim) { 847 pr_warn("%s: bug: prim == NULL\n", __func__); 848 return; 849 } 850 } 851 852 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim); 853 854 if (!(dev->flags & IFF_UP)) 855 return; 856 857 /* Add broadcast address, if it is explicitly assigned. */ 858 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) 859 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 860 861 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) && 862 (prefix != addr || ifa->ifa_prefixlen < 32)) { 863 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 864 fib_magic(RTM_NEWROUTE, 865 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 866 prefix, ifa->ifa_prefixlen, prim); 867 868 /* Add network specific broadcasts, when it takes a sense */ 869 if (ifa->ifa_prefixlen < 31) { 870 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim); 871 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask, 872 32, prim); 873 } 874 } 875 } 876 877 /* Delete primary or secondary address. 878 * Optionally, on secondary address promotion consider the addresses 879 * from subnet iprim as deleted, even if they are in device list. 880 * In this case the secondary ifa can be in device list. 881 */ 882 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim) 883 { 884 struct in_device *in_dev = ifa->ifa_dev; 885 struct net_device *dev = in_dev->dev; 886 struct in_ifaddr *ifa1; 887 struct in_ifaddr *prim = ifa, *prim1 = NULL; 888 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask; 889 __be32 any = ifa->ifa_address & ifa->ifa_mask; 890 #define LOCAL_OK 1 891 #define BRD_OK 2 892 #define BRD0_OK 4 893 #define BRD1_OK 8 894 unsigned int ok = 0; 895 int subnet = 0; /* Primary network */ 896 int gone = 1; /* Address is missing */ 897 int same_prefsrc = 0; /* Another primary with same IP */ 898 899 if (ifa->ifa_flags & IFA_F_SECONDARY) { 900 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask); 901 if (!prim) { 902 /* if the device has been deleted, we don't perform 903 * address promotion 904 */ 905 if (!in_dev->dead) 906 pr_warn("%s: bug: prim == NULL\n", __func__); 907 return; 908 } 909 if (iprim && iprim != prim) { 910 pr_warn("%s: bug: iprim != prim\n", __func__); 911 return; 912 } 913 } else if (!ipv4_is_zeronet(any) && 914 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) { 915 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 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 if (in_dev->dead) 923 goto no_promotions; 924 925 /* Deletion is more complicated than add. 926 * We should take care of not to delete too much :-) 927 * 928 * Scan address list to be sure that addresses are really gone. 929 */ 930 931 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) { 932 if (ifa1 == ifa) { 933 /* promotion, keep the IP */ 934 gone = 0; 935 continue; 936 } 937 /* Ignore IFAs from our subnet */ 938 if (iprim && ifa1->ifa_mask == iprim->ifa_mask && 939 inet_ifa_match(ifa1->ifa_address, iprim)) 940 continue; 941 942 /* Ignore ifa1 if it uses different primary IP (prefsrc) */ 943 if (ifa1->ifa_flags & IFA_F_SECONDARY) { 944 /* Another address from our subnet? */ 945 if (ifa1->ifa_mask == prim->ifa_mask && 946 inet_ifa_match(ifa1->ifa_address, prim)) 947 prim1 = prim; 948 else { 949 /* We reached the secondaries, so 950 * same_prefsrc should be determined. 951 */ 952 if (!same_prefsrc) 953 continue; 954 /* Search new prim1 if ifa1 is not 955 * using the current prim1 956 */ 957 if (!prim1 || 958 ifa1->ifa_mask != prim1->ifa_mask || 959 !inet_ifa_match(ifa1->ifa_address, prim1)) 960 prim1 = inet_ifa_byprefix(in_dev, 961 ifa1->ifa_address, 962 ifa1->ifa_mask); 963 if (!prim1) 964 continue; 965 if (prim1->ifa_local != prim->ifa_local) 966 continue; 967 } 968 } else { 969 if (prim->ifa_local != ifa1->ifa_local) 970 continue; 971 prim1 = ifa1; 972 if (prim != prim1) 973 same_prefsrc = 1; 974 } 975 if (ifa->ifa_local == ifa1->ifa_local) 976 ok |= LOCAL_OK; 977 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 978 ok |= BRD_OK; 979 if (brd == ifa1->ifa_broadcast) 980 ok |= BRD1_OK; 981 if (any == ifa1->ifa_broadcast) 982 ok |= BRD0_OK; 983 /* primary has network specific broadcasts */ 984 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) { 985 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask; 986 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask; 987 988 if (!ipv4_is_zeronet(any1)) { 989 if (ifa->ifa_broadcast == brd1 || 990 ifa->ifa_broadcast == any1) 991 ok |= BRD_OK; 992 if (brd == brd1 || brd == any1) 993 ok |= BRD1_OK; 994 if (any == brd1 || any == any1) 995 ok |= BRD0_OK; 996 } 997 } 998 } 999 1000 no_promotions: 1001 if (!(ok & BRD_OK)) 1002 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 1003 if (subnet && ifa->ifa_prefixlen < 31) { 1004 if (!(ok & BRD1_OK)) 1005 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim); 1006 if (!(ok & BRD0_OK)) 1007 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim); 1008 } 1009 if (!(ok & LOCAL_OK)) { 1010 unsigned int addr_type; 1011 1012 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim); 1013 1014 /* Check, that this local address finally disappeared. */ 1015 addr_type = inet_addr_type_dev_table(dev_net(dev), dev, 1016 ifa->ifa_local); 1017 if (gone && addr_type != RTN_LOCAL) { 1018 /* And the last, but not the least thing. 1019 * We must flush stray FIB entries. 1020 * 1021 * First of all, we scan fib_info list searching 1022 * for stray nexthop entries, then ignite fib_flush. 1023 */ 1024 if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local)) 1025 fib_flush(dev_net(dev)); 1026 } 1027 } 1028 #undef LOCAL_OK 1029 #undef BRD_OK 1030 #undef BRD0_OK 1031 #undef BRD1_OK 1032 } 1033 1034 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn) 1035 { 1036 1037 struct fib_result res; 1038 struct flowi4 fl4 = { 1039 .flowi4_mark = frn->fl_mark, 1040 .daddr = frn->fl_addr, 1041 .flowi4_tos = frn->fl_tos, 1042 .flowi4_scope = frn->fl_scope, 1043 }; 1044 struct fib_table *tb; 1045 1046 rcu_read_lock(); 1047 1048 tb = fib_get_table(net, frn->tb_id_in); 1049 1050 frn->err = -ENOENT; 1051 if (tb) { 1052 local_bh_disable(); 1053 1054 frn->tb_id = tb->tb_id; 1055 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 1056 1057 if (!frn->err) { 1058 frn->prefixlen = res.prefixlen; 1059 frn->nh_sel = res.nh_sel; 1060 frn->type = res.type; 1061 frn->scope = res.scope; 1062 } 1063 local_bh_enable(); 1064 } 1065 1066 rcu_read_unlock(); 1067 } 1068 1069 static void nl_fib_input(struct sk_buff *skb) 1070 { 1071 struct net *net; 1072 struct fib_result_nl *frn; 1073 struct nlmsghdr *nlh; 1074 u32 portid; 1075 1076 net = sock_net(skb->sk); 1077 nlh = nlmsg_hdr(skb); 1078 if (skb->len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len || 1079 nlmsg_len(nlh) < sizeof(*frn)) 1080 return; 1081 1082 skb = netlink_skb_clone(skb, GFP_KERNEL); 1083 if (!skb) 1084 return; 1085 nlh = nlmsg_hdr(skb); 1086 1087 frn = (struct fib_result_nl *) nlmsg_data(nlh); 1088 nl_fib_lookup(net, frn); 1089 1090 portid = NETLINK_CB(skb).portid; /* netlink portid */ 1091 NETLINK_CB(skb).portid = 0; /* from kernel */ 1092 NETLINK_CB(skb).dst_group = 0; /* unicast */ 1093 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT); 1094 } 1095 1096 static int __net_init nl_fib_lookup_init(struct net *net) 1097 { 1098 struct sock *sk; 1099 struct netlink_kernel_cfg cfg = { 1100 .input = nl_fib_input, 1101 }; 1102 1103 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg); 1104 if (!sk) 1105 return -EAFNOSUPPORT; 1106 net->ipv4.fibnl = sk; 1107 return 0; 1108 } 1109 1110 static void nl_fib_lookup_exit(struct net *net) 1111 { 1112 netlink_kernel_release(net->ipv4.fibnl); 1113 net->ipv4.fibnl = NULL; 1114 } 1115 1116 static void fib_disable_ip(struct net_device *dev, unsigned long event, 1117 bool force) 1118 { 1119 if (fib_sync_down_dev(dev, event, force)) 1120 fib_flush(dev_net(dev)); 1121 rt_cache_flush(dev_net(dev)); 1122 arp_ifdown(dev); 1123 } 1124 1125 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 1126 { 1127 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; 1128 struct net_device *dev = ifa->ifa_dev->dev; 1129 struct net *net = dev_net(dev); 1130 1131 switch (event) { 1132 case NETDEV_UP: 1133 fib_add_ifaddr(ifa); 1134 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1135 fib_sync_up(dev, RTNH_F_DEAD); 1136 #endif 1137 atomic_inc(&net->ipv4.dev_addr_genid); 1138 rt_cache_flush(dev_net(dev)); 1139 break; 1140 case NETDEV_DOWN: 1141 fib_del_ifaddr(ifa, NULL); 1142 atomic_inc(&net->ipv4.dev_addr_genid); 1143 if (!ifa->ifa_dev->ifa_list) { 1144 /* Last address was deleted from this interface. 1145 * Disable IP. 1146 */ 1147 fib_disable_ip(dev, event, true); 1148 } else { 1149 rt_cache_flush(dev_net(dev)); 1150 } 1151 break; 1152 } 1153 return NOTIFY_DONE; 1154 } 1155 1156 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 1157 { 1158 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1159 struct netdev_notifier_changeupper_info *info; 1160 struct in_device *in_dev; 1161 struct net *net = dev_net(dev); 1162 unsigned int flags; 1163 1164 if (event == NETDEV_UNREGISTER) { 1165 fib_disable_ip(dev, event, true); 1166 rt_flush_dev(dev); 1167 return NOTIFY_DONE; 1168 } 1169 1170 in_dev = __in_dev_get_rtnl(dev); 1171 if (!in_dev) 1172 return NOTIFY_DONE; 1173 1174 switch (event) { 1175 case NETDEV_UP: 1176 for_ifa(in_dev) { 1177 fib_add_ifaddr(ifa); 1178 } endfor_ifa(in_dev); 1179 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1180 fib_sync_up(dev, RTNH_F_DEAD); 1181 #endif 1182 atomic_inc(&net->ipv4.dev_addr_genid); 1183 rt_cache_flush(net); 1184 break; 1185 case NETDEV_DOWN: 1186 fib_disable_ip(dev, event, false); 1187 break; 1188 case NETDEV_CHANGE: 1189 flags = dev_get_flags(dev); 1190 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 1191 fib_sync_up(dev, RTNH_F_LINKDOWN); 1192 else 1193 fib_sync_down_dev(dev, event, false); 1194 /* fall through */ 1195 case NETDEV_CHANGEMTU: 1196 rt_cache_flush(net); 1197 break; 1198 case NETDEV_CHANGEUPPER: 1199 info = ptr; 1200 /* flush all routes if dev is linked to or unlinked from 1201 * an L3 master device (e.g., VRF) 1202 */ 1203 if (info->upper_dev && netif_is_l3_master(info->upper_dev)) 1204 fib_disable_ip(dev, NETDEV_DOWN, true); 1205 break; 1206 } 1207 return NOTIFY_DONE; 1208 } 1209 1210 static struct notifier_block fib_inetaddr_notifier = { 1211 .notifier_call = fib_inetaddr_event, 1212 }; 1213 1214 static struct notifier_block fib_netdev_notifier = { 1215 .notifier_call = fib_netdev_event, 1216 }; 1217 1218 static int __net_init ip_fib_net_init(struct net *net) 1219 { 1220 int err; 1221 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ; 1222 1223 /* Avoid false sharing : Use at least a full cache line */ 1224 size = max_t(size_t, size, L1_CACHE_BYTES); 1225 1226 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL); 1227 if (!net->ipv4.fib_table_hash) 1228 return -ENOMEM; 1229 1230 err = fib4_rules_init(net); 1231 if (err < 0) 1232 goto fail; 1233 return 0; 1234 1235 fail: 1236 kfree(net->ipv4.fib_table_hash); 1237 return err; 1238 } 1239 1240 static void ip_fib_net_exit(struct net *net) 1241 { 1242 unsigned int i; 1243 1244 rtnl_lock(); 1245 #ifdef CONFIG_IP_MULTIPLE_TABLES 1246 RCU_INIT_POINTER(net->ipv4.fib_main, NULL); 1247 RCU_INIT_POINTER(net->ipv4.fib_default, NULL); 1248 #endif 1249 for (i = 0; i < FIB_TABLE_HASHSZ; i++) { 1250 struct hlist_head *head = &net->ipv4.fib_table_hash[i]; 1251 struct hlist_node *tmp; 1252 struct fib_table *tb; 1253 1254 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) { 1255 hlist_del(&tb->tb_hlist); 1256 fib_table_flush(tb); 1257 fib_free_table(tb); 1258 } 1259 } 1260 1261 #ifdef CONFIG_IP_MULTIPLE_TABLES 1262 fib4_rules_exit(net); 1263 #endif 1264 rtnl_unlock(); 1265 kfree(net->ipv4.fib_table_hash); 1266 } 1267 1268 static int __net_init fib_net_init(struct net *net) 1269 { 1270 int error; 1271 1272 #ifdef CONFIG_IP_ROUTE_CLASSID 1273 net->ipv4.fib_num_tclassid_users = 0; 1274 #endif 1275 error = ip_fib_net_init(net); 1276 if (error < 0) 1277 goto out; 1278 error = nl_fib_lookup_init(net); 1279 if (error < 0) 1280 goto out_nlfl; 1281 error = fib_proc_init(net); 1282 if (error < 0) 1283 goto out_proc; 1284 out: 1285 return error; 1286 1287 out_proc: 1288 nl_fib_lookup_exit(net); 1289 out_nlfl: 1290 ip_fib_net_exit(net); 1291 goto out; 1292 } 1293 1294 static void __net_exit fib_net_exit(struct net *net) 1295 { 1296 fib_proc_exit(net); 1297 nl_fib_lookup_exit(net); 1298 ip_fib_net_exit(net); 1299 } 1300 1301 static struct pernet_operations fib_net_ops = { 1302 .init = fib_net_init, 1303 .exit = fib_net_exit, 1304 }; 1305 1306 void __init ip_fib_init(void) 1307 { 1308 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL); 1309 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL); 1310 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL); 1311 1312 register_pernet_subsys(&fib_net_ops); 1313 register_netdevice_notifier(&fib_netdev_notifier); 1314 register_inetaddr_notifier(&fib_inetaddr_notifier); 1315 1316 fib_trie_init(); 1317 } 1318