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(net, tb); 186 } 187 188 if (flushed) 189 rt_cache_flush(net); 190 } 191 192 /* 193 * Find address type as if only "dev" was present in the system. If 194 * on_dev is NULL then all interfaces are taken into consideration. 195 */ 196 static inline unsigned int __inet_dev_addr_type(struct net *net, 197 const struct net_device *dev, 198 __be32 addr, u32 tb_id) 199 { 200 struct flowi4 fl4 = { .daddr = addr }; 201 struct fib_result res; 202 unsigned int ret = RTN_BROADCAST; 203 struct fib_table *table; 204 205 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr)) 206 return RTN_BROADCAST; 207 if (ipv4_is_multicast(addr)) 208 return RTN_MULTICAST; 209 210 rcu_read_lock(); 211 212 table = fib_get_table(net, tb_id); 213 if (table) { 214 ret = RTN_UNICAST; 215 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) { 216 if (!dev || dev == res.fi->fib_dev) 217 ret = res.type; 218 } 219 } 220 221 rcu_read_unlock(); 222 return ret; 223 } 224 225 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id) 226 { 227 return __inet_dev_addr_type(net, NULL, addr, tb_id); 228 } 229 EXPORT_SYMBOL(inet_addr_type_table); 230 231 unsigned int inet_addr_type(struct net *net, __be32 addr) 232 { 233 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL); 234 } 235 EXPORT_SYMBOL(inet_addr_type); 236 237 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev, 238 __be32 addr) 239 { 240 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL; 241 242 return __inet_dev_addr_type(net, dev, addr, rt_table); 243 } 244 EXPORT_SYMBOL(inet_dev_addr_type); 245 246 /* inet_addr_type with dev == NULL but using the table from a dev 247 * if one is associated 248 */ 249 unsigned int inet_addr_type_dev_table(struct net *net, 250 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, NULL, addr, rt_table); 256 } 257 EXPORT_SYMBOL(inet_addr_type_dev_table); 258 259 __be32 fib_compute_spec_dst(struct sk_buff *skb) 260 { 261 struct net_device *dev = skb->dev; 262 struct in_device *in_dev; 263 struct fib_result res; 264 struct rtable *rt; 265 struct net *net; 266 int scope; 267 268 rt = skb_rtable(skb); 269 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) == 270 RTCF_LOCAL) 271 return ip_hdr(skb)->daddr; 272 273 in_dev = __in_dev_get_rcu(dev); 274 BUG_ON(!in_dev); 275 276 net = dev_net(dev); 277 278 scope = RT_SCOPE_UNIVERSE; 279 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) { 280 struct flowi4 fl4 = { 281 .flowi4_iif = LOOPBACK_IFINDEX, 282 .daddr = ip_hdr(skb)->saddr, 283 .flowi4_tos = RT_TOS(ip_hdr(skb)->tos), 284 .flowi4_scope = scope, 285 .flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0, 286 }; 287 if (!fib_lookup(net, &fl4, &res, 0)) 288 return FIB_RES_PREFSRC(net, res); 289 } else { 290 scope = RT_SCOPE_LINK; 291 } 292 293 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope); 294 } 295 296 /* Given (packet source, input interface) and optional (dst, oif, tos): 297 * - (main) check, that source is valid i.e. not broadcast or our local 298 * address. 299 * - figure out what "logical" interface this packet arrived 300 * and calculate "specific destination" address. 301 * - check, that packet arrived from expected physical interface. 302 * called with rcu_read_lock() 303 */ 304 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 305 u8 tos, int oif, struct net_device *dev, 306 int rpf, struct in_device *idev, u32 *itag) 307 { 308 int ret, no_addr; 309 struct fib_result res; 310 struct flowi4 fl4; 311 struct net *net; 312 bool dev_match; 313 314 fl4.flowi4_oif = 0; 315 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev); 316 if (!fl4.flowi4_iif) 317 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX; 318 fl4.daddr = src; 319 fl4.saddr = dst; 320 fl4.flowi4_tos = tos; 321 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 322 fl4.flowi4_tun_key.tun_id = 0; 323 fl4.flowi4_flags = 0; 324 325 no_addr = idev->ifa_list == NULL; 326 327 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0; 328 329 trace_fib_validate_source(dev, &fl4); 330 331 net = dev_net(dev); 332 if (fib_lookup(net, &fl4, &res, 0)) 333 goto last_resort; 334 if (res.type != RTN_UNICAST && 335 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev))) 336 goto e_inval; 337 if (!rpf && !fib_num_tclassid_users(dev_net(dev)) && 338 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) 339 goto last_resort; 340 fib_combine_itag(itag, &res); 341 dev_match = false; 342 343 #ifdef CONFIG_IP_ROUTE_MULTIPATH 344 for (ret = 0; ret < res.fi->fib_nhs; ret++) { 345 struct fib_nh *nh = &res.fi->fib_nh[ret]; 346 347 if (nh->nh_dev == dev) { 348 dev_match = true; 349 break; 350 } else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) { 351 dev_match = true; 352 break; 353 } 354 } 355 #else 356 if (FIB_RES_DEV(res) == dev) 357 dev_match = true; 358 #endif 359 if (dev_match) { 360 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 361 return ret; 362 } 363 if (no_addr) 364 goto last_resort; 365 if (rpf == 1) 366 goto e_rpf; 367 fl4.flowi4_oif = dev->ifindex; 368 369 ret = 0; 370 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) { 371 if (res.type == RTN_UNICAST) 372 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 373 } 374 return ret; 375 376 last_resort: 377 if (rpf) 378 goto e_rpf; 379 *itag = 0; 380 return 0; 381 382 e_inval: 383 return -EINVAL; 384 e_rpf: 385 return -EXDEV; 386 } 387 388 /* Ignore rp_filter for packets protected by IPsec. */ 389 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 390 u8 tos, int oif, struct net_device *dev, 391 struct in_device *idev, u32 *itag) 392 { 393 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev); 394 395 if (!r && !fib_num_tclassid_users(dev_net(dev)) && 396 IN_DEV_ACCEPT_LOCAL(idev) && 397 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) { 398 *itag = 0; 399 return 0; 400 } 401 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag); 402 } 403 404 static inline __be32 sk_extract_addr(struct sockaddr *addr) 405 { 406 return ((struct sockaddr_in *) addr)->sin_addr.s_addr; 407 } 408 409 static int put_rtax(struct nlattr *mx, int len, int type, u32 value) 410 { 411 struct nlattr *nla; 412 413 nla = (struct nlattr *) ((char *) mx + len); 414 nla->nla_type = type; 415 nla->nla_len = nla_attr_size(4); 416 *(u32 *) nla_data(nla) = value; 417 418 return len + nla_total_size(4); 419 } 420 421 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt, 422 struct fib_config *cfg) 423 { 424 __be32 addr; 425 int plen; 426 427 memset(cfg, 0, sizeof(*cfg)); 428 cfg->fc_nlinfo.nl_net = net; 429 430 if (rt->rt_dst.sa_family != AF_INET) 431 return -EAFNOSUPPORT; 432 433 /* 434 * Check mask for validity: 435 * a) it must be contiguous. 436 * b) destination must have all host bits clear. 437 * c) if application forgot to set correct family (AF_INET), 438 * reject request unless it is absolutely clear i.e. 439 * both family and mask are zero. 440 */ 441 plen = 32; 442 addr = sk_extract_addr(&rt->rt_dst); 443 if (!(rt->rt_flags & RTF_HOST)) { 444 __be32 mask = sk_extract_addr(&rt->rt_genmask); 445 446 if (rt->rt_genmask.sa_family != AF_INET) { 447 if (mask || rt->rt_genmask.sa_family) 448 return -EAFNOSUPPORT; 449 } 450 451 if (bad_mask(mask, addr)) 452 return -EINVAL; 453 454 plen = inet_mask_len(mask); 455 } 456 457 cfg->fc_dst_len = plen; 458 cfg->fc_dst = addr; 459 460 if (cmd != SIOCDELRT) { 461 cfg->fc_nlflags = NLM_F_CREATE; 462 cfg->fc_protocol = RTPROT_BOOT; 463 } 464 465 if (rt->rt_metric) 466 cfg->fc_priority = rt->rt_metric - 1; 467 468 if (rt->rt_flags & RTF_REJECT) { 469 cfg->fc_scope = RT_SCOPE_HOST; 470 cfg->fc_type = RTN_UNREACHABLE; 471 return 0; 472 } 473 474 cfg->fc_scope = RT_SCOPE_NOWHERE; 475 cfg->fc_type = RTN_UNICAST; 476 477 if (rt->rt_dev) { 478 char *colon; 479 struct net_device *dev; 480 char devname[IFNAMSIZ]; 481 482 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1)) 483 return -EFAULT; 484 485 devname[IFNAMSIZ-1] = 0; 486 colon = strchr(devname, ':'); 487 if (colon) 488 *colon = 0; 489 dev = __dev_get_by_name(net, devname); 490 if (!dev) 491 return -ENODEV; 492 cfg->fc_oif = dev->ifindex; 493 cfg->fc_table = l3mdev_fib_table(dev); 494 if (colon) { 495 struct in_ifaddr *ifa; 496 struct in_device *in_dev = __in_dev_get_rtnl(dev); 497 if (!in_dev) 498 return -ENODEV; 499 *colon = ':'; 500 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) 501 if (strcmp(ifa->ifa_label, devname) == 0) 502 break; 503 if (!ifa) 504 return -ENODEV; 505 cfg->fc_prefsrc = ifa->ifa_local; 506 } 507 } 508 509 addr = sk_extract_addr(&rt->rt_gateway); 510 if (rt->rt_gateway.sa_family == AF_INET && addr) { 511 unsigned int addr_type; 512 513 cfg->fc_gw = addr; 514 addr_type = inet_addr_type_table(net, addr, cfg->fc_table); 515 if (rt->rt_flags & RTF_GATEWAY && 516 addr_type == RTN_UNICAST) 517 cfg->fc_scope = RT_SCOPE_UNIVERSE; 518 } 519 520 if (cmd == SIOCDELRT) 521 return 0; 522 523 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw) 524 return -EINVAL; 525 526 if (cfg->fc_scope == RT_SCOPE_NOWHERE) 527 cfg->fc_scope = RT_SCOPE_LINK; 528 529 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) { 530 struct nlattr *mx; 531 int len = 0; 532 533 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL); 534 if (!mx) 535 return -ENOMEM; 536 537 if (rt->rt_flags & RTF_MTU) 538 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40); 539 540 if (rt->rt_flags & RTF_WINDOW) 541 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window); 542 543 if (rt->rt_flags & RTF_IRTT) 544 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3); 545 546 cfg->fc_mx = mx; 547 cfg->fc_mx_len = len; 548 } 549 550 return 0; 551 } 552 553 /* 554 * Handle IP routing ioctl calls. 555 * These are used to manipulate the routing tables 556 */ 557 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg) 558 { 559 struct fib_config cfg; 560 struct rtentry rt; 561 int err; 562 563 switch (cmd) { 564 case SIOCADDRT: /* Add a route */ 565 case SIOCDELRT: /* Delete a route */ 566 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 567 return -EPERM; 568 569 if (copy_from_user(&rt, arg, sizeof(rt))) 570 return -EFAULT; 571 572 rtnl_lock(); 573 err = rtentry_to_fib_config(net, cmd, &rt, &cfg); 574 if (err == 0) { 575 struct fib_table *tb; 576 577 if (cmd == SIOCDELRT) { 578 tb = fib_get_table(net, cfg.fc_table); 579 if (tb) 580 err = fib_table_delete(net, tb, &cfg); 581 else 582 err = -ESRCH; 583 } else { 584 tb = fib_new_table(net, cfg.fc_table); 585 if (tb) 586 err = fib_table_insert(net, tb, &cfg); 587 else 588 err = -ENOBUFS; 589 } 590 591 /* allocated by rtentry_to_fib_config() */ 592 kfree(cfg.fc_mx); 593 } 594 rtnl_unlock(); 595 return err; 596 } 597 return -EINVAL; 598 } 599 600 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = { 601 [RTA_DST] = { .type = NLA_U32 }, 602 [RTA_SRC] = { .type = NLA_U32 }, 603 [RTA_IIF] = { .type = NLA_U32 }, 604 [RTA_OIF] = { .type = NLA_U32 }, 605 [RTA_GATEWAY] = { .type = NLA_U32 }, 606 [RTA_PRIORITY] = { .type = NLA_U32 }, 607 [RTA_PREFSRC] = { .type = NLA_U32 }, 608 [RTA_METRICS] = { .type = NLA_NESTED }, 609 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) }, 610 [RTA_FLOW] = { .type = NLA_U32 }, 611 [RTA_ENCAP_TYPE] = { .type = NLA_U16 }, 612 [RTA_ENCAP] = { .type = NLA_NESTED }, 613 }; 614 615 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb, 616 struct nlmsghdr *nlh, struct fib_config *cfg) 617 { 618 struct nlattr *attr; 619 int err, remaining; 620 struct rtmsg *rtm; 621 622 err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy); 623 if (err < 0) 624 goto errout; 625 626 memset(cfg, 0, sizeof(*cfg)); 627 628 rtm = nlmsg_data(nlh); 629 cfg->fc_dst_len = rtm->rtm_dst_len; 630 cfg->fc_tos = rtm->rtm_tos; 631 cfg->fc_table = rtm->rtm_table; 632 cfg->fc_protocol = rtm->rtm_protocol; 633 cfg->fc_scope = rtm->rtm_scope; 634 cfg->fc_type = rtm->rtm_type; 635 cfg->fc_flags = rtm->rtm_flags; 636 cfg->fc_nlflags = nlh->nlmsg_flags; 637 638 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid; 639 cfg->fc_nlinfo.nlh = nlh; 640 cfg->fc_nlinfo.nl_net = net; 641 642 if (cfg->fc_type > RTN_MAX) { 643 err = -EINVAL; 644 goto errout; 645 } 646 647 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) { 648 switch (nla_type(attr)) { 649 case RTA_DST: 650 cfg->fc_dst = nla_get_be32(attr); 651 break; 652 case RTA_OIF: 653 cfg->fc_oif = nla_get_u32(attr); 654 break; 655 case RTA_GATEWAY: 656 cfg->fc_gw = nla_get_be32(attr); 657 break; 658 case RTA_PRIORITY: 659 cfg->fc_priority = nla_get_u32(attr); 660 break; 661 case RTA_PREFSRC: 662 cfg->fc_prefsrc = nla_get_be32(attr); 663 break; 664 case RTA_METRICS: 665 cfg->fc_mx = nla_data(attr); 666 cfg->fc_mx_len = nla_len(attr); 667 break; 668 case RTA_MULTIPATH: 669 cfg->fc_mp = nla_data(attr); 670 cfg->fc_mp_len = nla_len(attr); 671 break; 672 case RTA_FLOW: 673 cfg->fc_flow = nla_get_u32(attr); 674 break; 675 case RTA_TABLE: 676 cfg->fc_table = nla_get_u32(attr); 677 break; 678 case RTA_ENCAP: 679 cfg->fc_encap = attr; 680 break; 681 case RTA_ENCAP_TYPE: 682 cfg->fc_encap_type = nla_get_u16(attr); 683 break; 684 } 685 } 686 687 return 0; 688 errout: 689 return err; 690 } 691 692 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh) 693 { 694 struct net *net = sock_net(skb->sk); 695 struct fib_config cfg; 696 struct fib_table *tb; 697 int err; 698 699 err = rtm_to_fib_config(net, skb, nlh, &cfg); 700 if (err < 0) 701 goto errout; 702 703 tb = fib_get_table(net, cfg.fc_table); 704 if (!tb) { 705 err = -ESRCH; 706 goto errout; 707 } 708 709 err = fib_table_delete(net, tb, &cfg); 710 errout: 711 return err; 712 } 713 714 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh) 715 { 716 struct net *net = sock_net(skb->sk); 717 struct fib_config cfg; 718 struct fib_table *tb; 719 int err; 720 721 err = rtm_to_fib_config(net, skb, nlh, &cfg); 722 if (err < 0) 723 goto errout; 724 725 tb = fib_new_table(net, cfg.fc_table); 726 if (!tb) { 727 err = -ENOBUFS; 728 goto errout; 729 } 730 731 err = fib_table_insert(net, tb, &cfg); 732 errout: 733 return err; 734 } 735 736 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 737 { 738 struct net *net = sock_net(skb->sk); 739 unsigned int h, s_h; 740 unsigned int e = 0, s_e; 741 struct fib_table *tb; 742 struct hlist_head *head; 743 int dumped = 0; 744 745 if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) && 746 ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED) 747 return skb->len; 748 749 s_h = cb->args[0]; 750 s_e = cb->args[1]; 751 752 rcu_read_lock(); 753 754 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) { 755 e = 0; 756 head = &net->ipv4.fib_table_hash[h]; 757 hlist_for_each_entry_rcu(tb, head, tb_hlist) { 758 if (e < s_e) 759 goto next; 760 if (dumped) 761 memset(&cb->args[2], 0, sizeof(cb->args) - 762 2 * sizeof(cb->args[0])); 763 if (fib_table_dump(tb, skb, cb) < 0) 764 goto out; 765 dumped = 1; 766 next: 767 e++; 768 } 769 } 770 out: 771 rcu_read_unlock(); 772 773 cb->args[1] = e; 774 cb->args[0] = h; 775 776 return skb->len; 777 } 778 779 /* Prepare and feed intra-kernel routing request. 780 * Really, it should be netlink message, but :-( netlink 781 * can be not configured, so that we feed it directly 782 * to fib engine. It is legal, because all events occur 783 * only when netlink is already locked. 784 */ 785 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa) 786 { 787 struct net *net = dev_net(ifa->ifa_dev->dev); 788 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev); 789 struct fib_table *tb; 790 struct fib_config cfg = { 791 .fc_protocol = RTPROT_KERNEL, 792 .fc_type = type, 793 .fc_dst = dst, 794 .fc_dst_len = dst_len, 795 .fc_prefsrc = ifa->ifa_local, 796 .fc_oif = ifa->ifa_dev->dev->ifindex, 797 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND, 798 .fc_nlinfo = { 799 .nl_net = net, 800 }, 801 }; 802 803 if (!tb_id) 804 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL; 805 806 tb = fib_new_table(net, tb_id); 807 if (!tb) 808 return; 809 810 cfg.fc_table = tb->tb_id; 811 812 if (type != RTN_LOCAL) 813 cfg.fc_scope = RT_SCOPE_LINK; 814 else 815 cfg.fc_scope = RT_SCOPE_HOST; 816 817 if (cmd == RTM_NEWROUTE) 818 fib_table_insert(net, tb, &cfg); 819 else 820 fib_table_delete(net, tb, &cfg); 821 } 822 823 void fib_add_ifaddr(struct in_ifaddr *ifa) 824 { 825 struct in_device *in_dev = ifa->ifa_dev; 826 struct net_device *dev = in_dev->dev; 827 struct in_ifaddr *prim = ifa; 828 __be32 mask = ifa->ifa_mask; 829 __be32 addr = ifa->ifa_local; 830 __be32 prefix = ifa->ifa_address & mask; 831 832 if (ifa->ifa_flags & IFA_F_SECONDARY) { 833 prim = inet_ifa_byprefix(in_dev, prefix, mask); 834 if (!prim) { 835 pr_warn("%s: bug: prim == NULL\n", __func__); 836 return; 837 } 838 } 839 840 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim); 841 842 if (!(dev->flags & IFF_UP)) 843 return; 844 845 /* Add broadcast address, if it is explicitly assigned. */ 846 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) 847 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 848 849 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) && 850 (prefix != addr || ifa->ifa_prefixlen < 32)) { 851 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 852 fib_magic(RTM_NEWROUTE, 853 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 854 prefix, ifa->ifa_prefixlen, prim); 855 856 /* Add network specific broadcasts, when it takes a sense */ 857 if (ifa->ifa_prefixlen < 31) { 858 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim); 859 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask, 860 32, prim); 861 } 862 } 863 } 864 865 /* Delete primary or secondary address. 866 * Optionally, on secondary address promotion consider the addresses 867 * from subnet iprim as deleted, even if they are in device list. 868 * In this case the secondary ifa can be in device list. 869 */ 870 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim) 871 { 872 struct in_device *in_dev = ifa->ifa_dev; 873 struct net_device *dev = in_dev->dev; 874 struct in_ifaddr *ifa1; 875 struct in_ifaddr *prim = ifa, *prim1 = NULL; 876 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask; 877 __be32 any = ifa->ifa_address & ifa->ifa_mask; 878 #define LOCAL_OK 1 879 #define BRD_OK 2 880 #define BRD0_OK 4 881 #define BRD1_OK 8 882 unsigned int ok = 0; 883 int subnet = 0; /* Primary network */ 884 int gone = 1; /* Address is missing */ 885 int same_prefsrc = 0; /* Another primary with same IP */ 886 887 if (ifa->ifa_flags & IFA_F_SECONDARY) { 888 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask); 889 if (!prim) { 890 /* if the device has been deleted, we don't perform 891 * address promotion 892 */ 893 if (!in_dev->dead) 894 pr_warn("%s: bug: prim == NULL\n", __func__); 895 return; 896 } 897 if (iprim && iprim != prim) { 898 pr_warn("%s: bug: iprim != prim\n", __func__); 899 return; 900 } 901 } else if (!ipv4_is_zeronet(any) && 902 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) { 903 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE)) 904 fib_magic(RTM_DELROUTE, 905 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 906 any, ifa->ifa_prefixlen, prim); 907 subnet = 1; 908 } 909 910 if (in_dev->dead) 911 goto no_promotions; 912 913 /* Deletion is more complicated than add. 914 * We should take care of not to delete too much :-) 915 * 916 * Scan address list to be sure that addresses are really gone. 917 */ 918 919 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) { 920 if (ifa1 == ifa) { 921 /* promotion, keep the IP */ 922 gone = 0; 923 continue; 924 } 925 /* Ignore IFAs from our subnet */ 926 if (iprim && ifa1->ifa_mask == iprim->ifa_mask && 927 inet_ifa_match(ifa1->ifa_address, iprim)) 928 continue; 929 930 /* Ignore ifa1 if it uses different primary IP (prefsrc) */ 931 if (ifa1->ifa_flags & IFA_F_SECONDARY) { 932 /* Another address from our subnet? */ 933 if (ifa1->ifa_mask == prim->ifa_mask && 934 inet_ifa_match(ifa1->ifa_address, prim)) 935 prim1 = prim; 936 else { 937 /* We reached the secondaries, so 938 * same_prefsrc should be determined. 939 */ 940 if (!same_prefsrc) 941 continue; 942 /* Search new prim1 if ifa1 is not 943 * using the current prim1 944 */ 945 if (!prim1 || 946 ifa1->ifa_mask != prim1->ifa_mask || 947 !inet_ifa_match(ifa1->ifa_address, prim1)) 948 prim1 = inet_ifa_byprefix(in_dev, 949 ifa1->ifa_address, 950 ifa1->ifa_mask); 951 if (!prim1) 952 continue; 953 if (prim1->ifa_local != prim->ifa_local) 954 continue; 955 } 956 } else { 957 if (prim->ifa_local != ifa1->ifa_local) 958 continue; 959 prim1 = ifa1; 960 if (prim != prim1) 961 same_prefsrc = 1; 962 } 963 if (ifa->ifa_local == ifa1->ifa_local) 964 ok |= LOCAL_OK; 965 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 966 ok |= BRD_OK; 967 if (brd == ifa1->ifa_broadcast) 968 ok |= BRD1_OK; 969 if (any == ifa1->ifa_broadcast) 970 ok |= BRD0_OK; 971 /* primary has network specific broadcasts */ 972 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) { 973 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask; 974 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask; 975 976 if (!ipv4_is_zeronet(any1)) { 977 if (ifa->ifa_broadcast == brd1 || 978 ifa->ifa_broadcast == any1) 979 ok |= BRD_OK; 980 if (brd == brd1 || brd == any1) 981 ok |= BRD1_OK; 982 if (any == brd1 || any == any1) 983 ok |= BRD0_OK; 984 } 985 } 986 } 987 988 no_promotions: 989 if (!(ok & BRD_OK)) 990 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 991 if (subnet && ifa->ifa_prefixlen < 31) { 992 if (!(ok & BRD1_OK)) 993 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim); 994 if (!(ok & BRD0_OK)) 995 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim); 996 } 997 if (!(ok & LOCAL_OK)) { 998 unsigned int addr_type; 999 1000 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim); 1001 1002 /* Check, that this local address finally disappeared. */ 1003 addr_type = inet_addr_type_dev_table(dev_net(dev), dev, 1004 ifa->ifa_local); 1005 if (gone && addr_type != RTN_LOCAL) { 1006 /* And the last, but not the least thing. 1007 * We must flush stray FIB entries. 1008 * 1009 * First of all, we scan fib_info list searching 1010 * for stray nexthop entries, then ignite fib_flush. 1011 */ 1012 if (fib_sync_down_addr(dev, ifa->ifa_local)) 1013 fib_flush(dev_net(dev)); 1014 } 1015 } 1016 #undef LOCAL_OK 1017 #undef BRD_OK 1018 #undef BRD0_OK 1019 #undef BRD1_OK 1020 } 1021 1022 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn) 1023 { 1024 1025 struct fib_result res; 1026 struct flowi4 fl4 = { 1027 .flowi4_mark = frn->fl_mark, 1028 .daddr = frn->fl_addr, 1029 .flowi4_tos = frn->fl_tos, 1030 .flowi4_scope = frn->fl_scope, 1031 }; 1032 struct fib_table *tb; 1033 1034 rcu_read_lock(); 1035 1036 tb = fib_get_table(net, frn->tb_id_in); 1037 1038 frn->err = -ENOENT; 1039 if (tb) { 1040 local_bh_disable(); 1041 1042 frn->tb_id = tb->tb_id; 1043 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 1044 1045 if (!frn->err) { 1046 frn->prefixlen = res.prefixlen; 1047 frn->nh_sel = res.nh_sel; 1048 frn->type = res.type; 1049 frn->scope = res.scope; 1050 } 1051 local_bh_enable(); 1052 } 1053 1054 rcu_read_unlock(); 1055 } 1056 1057 static void nl_fib_input(struct sk_buff *skb) 1058 { 1059 struct net *net; 1060 struct fib_result_nl *frn; 1061 struct nlmsghdr *nlh; 1062 u32 portid; 1063 1064 net = sock_net(skb->sk); 1065 nlh = nlmsg_hdr(skb); 1066 if (skb->len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len || 1067 nlmsg_len(nlh) < sizeof(*frn)) 1068 return; 1069 1070 skb = netlink_skb_clone(skb, GFP_KERNEL); 1071 if (!skb) 1072 return; 1073 nlh = nlmsg_hdr(skb); 1074 1075 frn = (struct fib_result_nl *) nlmsg_data(nlh); 1076 nl_fib_lookup(net, frn); 1077 1078 portid = NETLINK_CB(skb).portid; /* netlink portid */ 1079 NETLINK_CB(skb).portid = 0; /* from kernel */ 1080 NETLINK_CB(skb).dst_group = 0; /* unicast */ 1081 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT); 1082 } 1083 1084 static int __net_init nl_fib_lookup_init(struct net *net) 1085 { 1086 struct sock *sk; 1087 struct netlink_kernel_cfg cfg = { 1088 .input = nl_fib_input, 1089 }; 1090 1091 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg); 1092 if (!sk) 1093 return -EAFNOSUPPORT; 1094 net->ipv4.fibnl = sk; 1095 return 0; 1096 } 1097 1098 static void nl_fib_lookup_exit(struct net *net) 1099 { 1100 netlink_kernel_release(net->ipv4.fibnl); 1101 net->ipv4.fibnl = NULL; 1102 } 1103 1104 static void fib_disable_ip(struct net_device *dev, unsigned long event, 1105 bool force) 1106 { 1107 if (fib_sync_down_dev(dev, event, force)) 1108 fib_flush(dev_net(dev)); 1109 rt_cache_flush(dev_net(dev)); 1110 arp_ifdown(dev); 1111 } 1112 1113 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 1114 { 1115 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; 1116 struct net_device *dev = ifa->ifa_dev->dev; 1117 struct net *net = dev_net(dev); 1118 1119 switch (event) { 1120 case NETDEV_UP: 1121 fib_add_ifaddr(ifa); 1122 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1123 fib_sync_up(dev, RTNH_F_DEAD); 1124 #endif 1125 atomic_inc(&net->ipv4.dev_addr_genid); 1126 rt_cache_flush(dev_net(dev)); 1127 break; 1128 case NETDEV_DOWN: 1129 fib_del_ifaddr(ifa, NULL); 1130 atomic_inc(&net->ipv4.dev_addr_genid); 1131 if (!ifa->ifa_dev->ifa_list) { 1132 /* Last address was deleted from this interface. 1133 * Disable IP. 1134 */ 1135 fib_disable_ip(dev, event, true); 1136 } else { 1137 rt_cache_flush(dev_net(dev)); 1138 } 1139 break; 1140 } 1141 return NOTIFY_DONE; 1142 } 1143 1144 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 1145 { 1146 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1147 struct netdev_notifier_changeupper_info *info; 1148 struct in_device *in_dev; 1149 struct net *net = dev_net(dev); 1150 unsigned int flags; 1151 1152 if (event == NETDEV_UNREGISTER) { 1153 fib_disable_ip(dev, event, true); 1154 rt_flush_dev(dev); 1155 return NOTIFY_DONE; 1156 } 1157 1158 in_dev = __in_dev_get_rtnl(dev); 1159 if (!in_dev) 1160 return NOTIFY_DONE; 1161 1162 switch (event) { 1163 case NETDEV_UP: 1164 for_ifa(in_dev) { 1165 fib_add_ifaddr(ifa); 1166 } endfor_ifa(in_dev); 1167 #ifdef CONFIG_IP_ROUTE_MULTIPATH 1168 fib_sync_up(dev, RTNH_F_DEAD); 1169 #endif 1170 atomic_inc(&net->ipv4.dev_addr_genid); 1171 rt_cache_flush(net); 1172 break; 1173 case NETDEV_DOWN: 1174 fib_disable_ip(dev, event, false); 1175 break; 1176 case NETDEV_CHANGE: 1177 flags = dev_get_flags(dev); 1178 if (flags & (IFF_RUNNING | IFF_LOWER_UP)) 1179 fib_sync_up(dev, RTNH_F_LINKDOWN); 1180 else 1181 fib_sync_down_dev(dev, event, false); 1182 /* fall through */ 1183 case NETDEV_CHANGEMTU: 1184 rt_cache_flush(net); 1185 break; 1186 case NETDEV_CHANGEUPPER: 1187 info = ptr; 1188 /* flush all routes if dev is linked to or unlinked from 1189 * an L3 master device (e.g., VRF) 1190 */ 1191 if (info->upper_dev && netif_is_l3_master(info->upper_dev)) 1192 fib_disable_ip(dev, NETDEV_DOWN, true); 1193 break; 1194 } 1195 return NOTIFY_DONE; 1196 } 1197 1198 static struct notifier_block fib_inetaddr_notifier = { 1199 .notifier_call = fib_inetaddr_event, 1200 }; 1201 1202 static struct notifier_block fib_netdev_notifier = { 1203 .notifier_call = fib_netdev_event, 1204 }; 1205 1206 static int __net_init ip_fib_net_init(struct net *net) 1207 { 1208 int err; 1209 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ; 1210 1211 /* Avoid false sharing : Use at least a full cache line */ 1212 size = max_t(size_t, size, L1_CACHE_BYTES); 1213 1214 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL); 1215 if (!net->ipv4.fib_table_hash) 1216 return -ENOMEM; 1217 1218 err = fib4_rules_init(net); 1219 if (err < 0) 1220 goto fail; 1221 return 0; 1222 1223 fail: 1224 kfree(net->ipv4.fib_table_hash); 1225 return err; 1226 } 1227 1228 static void ip_fib_net_exit(struct net *net) 1229 { 1230 unsigned int i; 1231 1232 rtnl_lock(); 1233 #ifdef CONFIG_IP_MULTIPLE_TABLES 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(net, 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