fib_frontend.c revision 52e804c6dfaa5df1e4b0e290357b82ad4e4cda2c
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 49#ifndef CONFIG_IP_MULTIPLE_TABLES 50 51static int __net_init fib4_rules_init(struct net *net) 52{ 53 struct fib_table *local_table, *main_table; 54 55 local_table = fib_trie_table(RT_TABLE_LOCAL); 56 if (local_table == NULL) 57 return -ENOMEM; 58 59 main_table = fib_trie_table(RT_TABLE_MAIN); 60 if (main_table == NULL) 61 goto fail; 62 63 hlist_add_head_rcu(&local_table->tb_hlist, 64 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]); 65 hlist_add_head_rcu(&main_table->tb_hlist, 66 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]); 67 return 0; 68 69fail: 70 kfree(local_table); 71 return -ENOMEM; 72} 73#else 74 75struct fib_table *fib_new_table(struct net *net, u32 id) 76{ 77 struct fib_table *tb; 78 unsigned int h; 79 80 if (id == 0) 81 id = RT_TABLE_MAIN; 82 tb = fib_get_table(net, id); 83 if (tb) 84 return tb; 85 86 tb = fib_trie_table(id); 87 if (!tb) 88 return NULL; 89 90 switch (id) { 91 case RT_TABLE_LOCAL: 92 net->ipv4.fib_local = tb; 93 break; 94 95 case RT_TABLE_MAIN: 96 net->ipv4.fib_main = tb; 97 break; 98 99 case RT_TABLE_DEFAULT: 100 net->ipv4.fib_default = tb; 101 break; 102 103 default: 104 break; 105 } 106 107 h = id & (FIB_TABLE_HASHSZ - 1); 108 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]); 109 return tb; 110} 111 112struct fib_table *fib_get_table(struct net *net, u32 id) 113{ 114 struct fib_table *tb; 115 struct hlist_node *node; 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 rcu_read_lock(); 124 head = &net->ipv4.fib_table_hash[h]; 125 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) { 126 if (tb->tb_id == id) { 127 rcu_read_unlock(); 128 return tb; 129 } 130 } 131 rcu_read_unlock(); 132 return NULL; 133} 134#endif /* CONFIG_IP_MULTIPLE_TABLES */ 135 136static void fib_flush(struct net *net) 137{ 138 int flushed = 0; 139 struct fib_table *tb; 140 struct hlist_node *node; 141 struct hlist_head *head; 142 unsigned int h; 143 144 for (h = 0; h < FIB_TABLE_HASHSZ; h++) { 145 head = &net->ipv4.fib_table_hash[h]; 146 hlist_for_each_entry(tb, node, head, tb_hlist) 147 flushed += fib_table_flush(tb); 148 } 149 150 if (flushed) 151 rt_cache_flush(net); 152} 153 154/* 155 * Find address type as if only "dev" was present in the system. If 156 * on_dev is NULL then all interfaces are taken into consideration. 157 */ 158static inline unsigned int __inet_dev_addr_type(struct net *net, 159 const struct net_device *dev, 160 __be32 addr) 161{ 162 struct flowi4 fl4 = { .daddr = addr }; 163 struct fib_result res; 164 unsigned int ret = RTN_BROADCAST; 165 struct fib_table *local_table; 166 167 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr)) 168 return RTN_BROADCAST; 169 if (ipv4_is_multicast(addr)) 170 return RTN_MULTICAST; 171 172 local_table = fib_get_table(net, RT_TABLE_LOCAL); 173 if (local_table) { 174 ret = RTN_UNICAST; 175 rcu_read_lock(); 176 if (!fib_table_lookup(local_table, &fl4, &res, FIB_LOOKUP_NOREF)) { 177 if (!dev || dev == res.fi->fib_dev) 178 ret = res.type; 179 } 180 rcu_read_unlock(); 181 } 182 return ret; 183} 184 185unsigned int inet_addr_type(struct net *net, __be32 addr) 186{ 187 return __inet_dev_addr_type(net, NULL, addr); 188} 189EXPORT_SYMBOL(inet_addr_type); 190 191unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev, 192 __be32 addr) 193{ 194 return __inet_dev_addr_type(net, dev, addr); 195} 196EXPORT_SYMBOL(inet_dev_addr_type); 197 198__be32 fib_compute_spec_dst(struct sk_buff *skb) 199{ 200 struct net_device *dev = skb->dev; 201 struct in_device *in_dev; 202 struct fib_result res; 203 struct rtable *rt; 204 struct flowi4 fl4; 205 struct net *net; 206 int scope; 207 208 rt = skb_rtable(skb); 209 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) == 210 RTCF_LOCAL) 211 return ip_hdr(skb)->daddr; 212 213 in_dev = __in_dev_get_rcu(dev); 214 BUG_ON(!in_dev); 215 216 net = dev_net(dev); 217 218 scope = RT_SCOPE_UNIVERSE; 219 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) { 220 fl4.flowi4_oif = 0; 221 fl4.flowi4_iif = LOOPBACK_IFINDEX; 222 fl4.daddr = ip_hdr(skb)->saddr; 223 fl4.saddr = 0; 224 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos); 225 fl4.flowi4_scope = scope; 226 fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0; 227 if (!fib_lookup(net, &fl4, &res)) 228 return FIB_RES_PREFSRC(net, res); 229 } else { 230 scope = RT_SCOPE_LINK; 231 } 232 233 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope); 234} 235 236/* Given (packet source, input interface) and optional (dst, oif, tos): 237 * - (main) check, that source is valid i.e. not broadcast or our local 238 * address. 239 * - figure out what "logical" interface this packet arrived 240 * and calculate "specific destination" address. 241 * - check, that packet arrived from expected physical interface. 242 * called with rcu_read_lock() 243 */ 244static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 245 u8 tos, int oif, struct net_device *dev, 246 int rpf, struct in_device *idev, u32 *itag) 247{ 248 int ret, no_addr, accept_local; 249 struct fib_result res; 250 struct flowi4 fl4; 251 struct net *net; 252 bool dev_match; 253 254 fl4.flowi4_oif = 0; 255 fl4.flowi4_iif = oif; 256 fl4.daddr = src; 257 fl4.saddr = dst; 258 fl4.flowi4_tos = tos; 259 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 260 261 no_addr = idev->ifa_list == NULL; 262 263 accept_local = IN_DEV_ACCEPT_LOCAL(idev); 264 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0; 265 266 net = dev_net(dev); 267 if (fib_lookup(net, &fl4, &res)) 268 goto last_resort; 269 if (res.type != RTN_UNICAST) { 270 if (res.type != RTN_LOCAL || !accept_local) 271 goto e_inval; 272 } 273 fib_combine_itag(itag, &res); 274 dev_match = false; 275 276#ifdef CONFIG_IP_ROUTE_MULTIPATH 277 for (ret = 0; ret < res.fi->fib_nhs; ret++) { 278 struct fib_nh *nh = &res.fi->fib_nh[ret]; 279 280 if (nh->nh_dev == dev) { 281 dev_match = true; 282 break; 283 } 284 } 285#else 286 if (FIB_RES_DEV(res) == dev) 287 dev_match = true; 288#endif 289 if (dev_match) { 290 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 291 return ret; 292 } 293 if (no_addr) 294 goto last_resort; 295 if (rpf == 1) 296 goto e_rpf; 297 fl4.flowi4_oif = dev->ifindex; 298 299 ret = 0; 300 if (fib_lookup(net, &fl4, &res) == 0) { 301 if (res.type == RTN_UNICAST) 302 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 303 } 304 return ret; 305 306last_resort: 307 if (rpf) 308 goto e_rpf; 309 *itag = 0; 310 return 0; 311 312e_inval: 313 return -EINVAL; 314e_rpf: 315 return -EXDEV; 316} 317 318/* Ignore rp_filter for packets protected by IPsec. */ 319int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, 320 u8 tos, int oif, struct net_device *dev, 321 struct in_device *idev, u32 *itag) 322{ 323 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev); 324 325 if (!r && !fib_num_tclassid_users(dev_net(dev)) && 326 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) { 327 *itag = 0; 328 return 0; 329 } 330 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag); 331} 332 333static inline __be32 sk_extract_addr(struct sockaddr *addr) 334{ 335 return ((struct sockaddr_in *) addr)->sin_addr.s_addr; 336} 337 338static int put_rtax(struct nlattr *mx, int len, int type, u32 value) 339{ 340 struct nlattr *nla; 341 342 nla = (struct nlattr *) ((char *) mx + len); 343 nla->nla_type = type; 344 nla->nla_len = nla_attr_size(4); 345 *(u32 *) nla_data(nla) = value; 346 347 return len + nla_total_size(4); 348} 349 350static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt, 351 struct fib_config *cfg) 352{ 353 __be32 addr; 354 int plen; 355 356 memset(cfg, 0, sizeof(*cfg)); 357 cfg->fc_nlinfo.nl_net = net; 358 359 if (rt->rt_dst.sa_family != AF_INET) 360 return -EAFNOSUPPORT; 361 362 /* 363 * Check mask for validity: 364 * a) it must be contiguous. 365 * b) destination must have all host bits clear. 366 * c) if application forgot to set correct family (AF_INET), 367 * reject request unless it is absolutely clear i.e. 368 * both family and mask are zero. 369 */ 370 plen = 32; 371 addr = sk_extract_addr(&rt->rt_dst); 372 if (!(rt->rt_flags & RTF_HOST)) { 373 __be32 mask = sk_extract_addr(&rt->rt_genmask); 374 375 if (rt->rt_genmask.sa_family != AF_INET) { 376 if (mask || rt->rt_genmask.sa_family) 377 return -EAFNOSUPPORT; 378 } 379 380 if (bad_mask(mask, addr)) 381 return -EINVAL; 382 383 plen = inet_mask_len(mask); 384 } 385 386 cfg->fc_dst_len = plen; 387 cfg->fc_dst = addr; 388 389 if (cmd != SIOCDELRT) { 390 cfg->fc_nlflags = NLM_F_CREATE; 391 cfg->fc_protocol = RTPROT_BOOT; 392 } 393 394 if (rt->rt_metric) 395 cfg->fc_priority = rt->rt_metric - 1; 396 397 if (rt->rt_flags & RTF_REJECT) { 398 cfg->fc_scope = RT_SCOPE_HOST; 399 cfg->fc_type = RTN_UNREACHABLE; 400 return 0; 401 } 402 403 cfg->fc_scope = RT_SCOPE_NOWHERE; 404 cfg->fc_type = RTN_UNICAST; 405 406 if (rt->rt_dev) { 407 char *colon; 408 struct net_device *dev; 409 char devname[IFNAMSIZ]; 410 411 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1)) 412 return -EFAULT; 413 414 devname[IFNAMSIZ-1] = 0; 415 colon = strchr(devname, ':'); 416 if (colon) 417 *colon = 0; 418 dev = __dev_get_by_name(net, devname); 419 if (!dev) 420 return -ENODEV; 421 cfg->fc_oif = dev->ifindex; 422 if (colon) { 423 struct in_ifaddr *ifa; 424 struct in_device *in_dev = __in_dev_get_rtnl(dev); 425 if (!in_dev) 426 return -ENODEV; 427 *colon = ':'; 428 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) 429 if (strcmp(ifa->ifa_label, devname) == 0) 430 break; 431 if (ifa == NULL) 432 return -ENODEV; 433 cfg->fc_prefsrc = ifa->ifa_local; 434 } 435 } 436 437 addr = sk_extract_addr(&rt->rt_gateway); 438 if (rt->rt_gateway.sa_family == AF_INET && addr) { 439 cfg->fc_gw = addr; 440 if (rt->rt_flags & RTF_GATEWAY && 441 inet_addr_type(net, addr) == RTN_UNICAST) 442 cfg->fc_scope = RT_SCOPE_UNIVERSE; 443 } 444 445 if (cmd == SIOCDELRT) 446 return 0; 447 448 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw) 449 return -EINVAL; 450 451 if (cfg->fc_scope == RT_SCOPE_NOWHERE) 452 cfg->fc_scope = RT_SCOPE_LINK; 453 454 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) { 455 struct nlattr *mx; 456 int len = 0; 457 458 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL); 459 if (mx == NULL) 460 return -ENOMEM; 461 462 if (rt->rt_flags & RTF_MTU) 463 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40); 464 465 if (rt->rt_flags & RTF_WINDOW) 466 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window); 467 468 if (rt->rt_flags & RTF_IRTT) 469 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3); 470 471 cfg->fc_mx = mx; 472 cfg->fc_mx_len = len; 473 } 474 475 return 0; 476} 477 478/* 479 * Handle IP routing ioctl calls. 480 * These are used to manipulate the routing tables 481 */ 482int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg) 483{ 484 struct fib_config cfg; 485 struct rtentry rt; 486 int err; 487 488 switch (cmd) { 489 case SIOCADDRT: /* Add a route */ 490 case SIOCDELRT: /* Delete a route */ 491 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 492 return -EPERM; 493 494 if (copy_from_user(&rt, arg, sizeof(rt))) 495 return -EFAULT; 496 497 rtnl_lock(); 498 err = rtentry_to_fib_config(net, cmd, &rt, &cfg); 499 if (err == 0) { 500 struct fib_table *tb; 501 502 if (cmd == SIOCDELRT) { 503 tb = fib_get_table(net, cfg.fc_table); 504 if (tb) 505 err = fib_table_delete(tb, &cfg); 506 else 507 err = -ESRCH; 508 } else { 509 tb = fib_new_table(net, cfg.fc_table); 510 if (tb) 511 err = fib_table_insert(tb, &cfg); 512 else 513 err = -ENOBUFS; 514 } 515 516 /* allocated by rtentry_to_fib_config() */ 517 kfree(cfg.fc_mx); 518 } 519 rtnl_unlock(); 520 return err; 521 } 522 return -EINVAL; 523} 524 525const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = { 526 [RTA_DST] = { .type = NLA_U32 }, 527 [RTA_SRC] = { .type = NLA_U32 }, 528 [RTA_IIF] = { .type = NLA_U32 }, 529 [RTA_OIF] = { .type = NLA_U32 }, 530 [RTA_GATEWAY] = { .type = NLA_U32 }, 531 [RTA_PRIORITY] = { .type = NLA_U32 }, 532 [RTA_PREFSRC] = { .type = NLA_U32 }, 533 [RTA_METRICS] = { .type = NLA_NESTED }, 534 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) }, 535 [RTA_FLOW] = { .type = NLA_U32 }, 536}; 537 538static int rtm_to_fib_config(struct net *net, struct sk_buff *skb, 539 struct nlmsghdr *nlh, struct fib_config *cfg) 540{ 541 struct nlattr *attr; 542 int err, remaining; 543 struct rtmsg *rtm; 544 545 err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy); 546 if (err < 0) 547 goto errout; 548 549 memset(cfg, 0, sizeof(*cfg)); 550 551 rtm = nlmsg_data(nlh); 552 cfg->fc_dst_len = rtm->rtm_dst_len; 553 cfg->fc_tos = rtm->rtm_tos; 554 cfg->fc_table = rtm->rtm_table; 555 cfg->fc_protocol = rtm->rtm_protocol; 556 cfg->fc_scope = rtm->rtm_scope; 557 cfg->fc_type = rtm->rtm_type; 558 cfg->fc_flags = rtm->rtm_flags; 559 cfg->fc_nlflags = nlh->nlmsg_flags; 560 561 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid; 562 cfg->fc_nlinfo.nlh = nlh; 563 cfg->fc_nlinfo.nl_net = net; 564 565 if (cfg->fc_type > RTN_MAX) { 566 err = -EINVAL; 567 goto errout; 568 } 569 570 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) { 571 switch (nla_type(attr)) { 572 case RTA_DST: 573 cfg->fc_dst = nla_get_be32(attr); 574 break; 575 case RTA_OIF: 576 cfg->fc_oif = nla_get_u32(attr); 577 break; 578 case RTA_GATEWAY: 579 cfg->fc_gw = nla_get_be32(attr); 580 break; 581 case RTA_PRIORITY: 582 cfg->fc_priority = nla_get_u32(attr); 583 break; 584 case RTA_PREFSRC: 585 cfg->fc_prefsrc = nla_get_be32(attr); 586 break; 587 case RTA_METRICS: 588 cfg->fc_mx = nla_data(attr); 589 cfg->fc_mx_len = nla_len(attr); 590 break; 591 case RTA_MULTIPATH: 592 cfg->fc_mp = nla_data(attr); 593 cfg->fc_mp_len = nla_len(attr); 594 break; 595 case RTA_FLOW: 596 cfg->fc_flow = nla_get_u32(attr); 597 break; 598 case RTA_TABLE: 599 cfg->fc_table = nla_get_u32(attr); 600 break; 601 } 602 } 603 604 return 0; 605errout: 606 return err; 607} 608 609static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg) 610{ 611 struct net *net = sock_net(skb->sk); 612 struct fib_config cfg; 613 struct fib_table *tb; 614 int err; 615 616 if (!capable(CAP_NET_ADMIN)) 617 return -EPERM; 618 619 err = rtm_to_fib_config(net, skb, nlh, &cfg); 620 if (err < 0) 621 goto errout; 622 623 tb = fib_get_table(net, cfg.fc_table); 624 if (tb == NULL) { 625 err = -ESRCH; 626 goto errout; 627 } 628 629 err = fib_table_delete(tb, &cfg); 630errout: 631 return err; 632} 633 634static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg) 635{ 636 struct net *net = sock_net(skb->sk); 637 struct fib_config cfg; 638 struct fib_table *tb; 639 int err; 640 641 if (!capable(CAP_NET_ADMIN)) 642 return -EPERM; 643 644 err = rtm_to_fib_config(net, skb, nlh, &cfg); 645 if (err < 0) 646 goto errout; 647 648 tb = fib_new_table(net, cfg.fc_table); 649 if (tb == NULL) { 650 err = -ENOBUFS; 651 goto errout; 652 } 653 654 err = fib_table_insert(tb, &cfg); 655errout: 656 return err; 657} 658 659static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 660{ 661 struct net *net = sock_net(skb->sk); 662 unsigned int h, s_h; 663 unsigned int e = 0, s_e; 664 struct fib_table *tb; 665 struct hlist_node *node; 666 struct hlist_head *head; 667 int dumped = 0; 668 669 if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) && 670 ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED) 671 return ip_rt_dump(skb, cb); 672 673 s_h = cb->args[0]; 674 s_e = cb->args[1]; 675 676 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) { 677 e = 0; 678 head = &net->ipv4.fib_table_hash[h]; 679 hlist_for_each_entry(tb, node, head, tb_hlist) { 680 if (e < s_e) 681 goto next; 682 if (dumped) 683 memset(&cb->args[2], 0, sizeof(cb->args) - 684 2 * sizeof(cb->args[0])); 685 if (fib_table_dump(tb, skb, cb) < 0) 686 goto out; 687 dumped = 1; 688next: 689 e++; 690 } 691 } 692out: 693 cb->args[1] = e; 694 cb->args[0] = h; 695 696 return skb->len; 697} 698 699/* Prepare and feed intra-kernel routing request. 700 * Really, it should be netlink message, but :-( netlink 701 * can be not configured, so that we feed it directly 702 * to fib engine. It is legal, because all events occur 703 * only when netlink is already locked. 704 */ 705static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa) 706{ 707 struct net *net = dev_net(ifa->ifa_dev->dev); 708 struct fib_table *tb; 709 struct fib_config cfg = { 710 .fc_protocol = RTPROT_KERNEL, 711 .fc_type = type, 712 .fc_dst = dst, 713 .fc_dst_len = dst_len, 714 .fc_prefsrc = ifa->ifa_local, 715 .fc_oif = ifa->ifa_dev->dev->ifindex, 716 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND, 717 .fc_nlinfo = { 718 .nl_net = net, 719 }, 720 }; 721 722 if (type == RTN_UNICAST) 723 tb = fib_new_table(net, RT_TABLE_MAIN); 724 else 725 tb = fib_new_table(net, RT_TABLE_LOCAL); 726 727 if (tb == NULL) 728 return; 729 730 cfg.fc_table = tb->tb_id; 731 732 if (type != RTN_LOCAL) 733 cfg.fc_scope = RT_SCOPE_LINK; 734 else 735 cfg.fc_scope = RT_SCOPE_HOST; 736 737 if (cmd == RTM_NEWROUTE) 738 fib_table_insert(tb, &cfg); 739 else 740 fib_table_delete(tb, &cfg); 741} 742 743void fib_add_ifaddr(struct in_ifaddr *ifa) 744{ 745 struct in_device *in_dev = ifa->ifa_dev; 746 struct net_device *dev = in_dev->dev; 747 struct in_ifaddr *prim = ifa; 748 __be32 mask = ifa->ifa_mask; 749 __be32 addr = ifa->ifa_local; 750 __be32 prefix = ifa->ifa_address & mask; 751 752 if (ifa->ifa_flags & IFA_F_SECONDARY) { 753 prim = inet_ifa_byprefix(in_dev, prefix, mask); 754 if (prim == NULL) { 755 pr_warn("%s: bug: prim == NULL\n", __func__); 756 return; 757 } 758 } 759 760 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim); 761 762 if (!(dev->flags & IFF_UP)) 763 return; 764 765 /* Add broadcast address, if it is explicitly assigned. */ 766 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) 767 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 768 769 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) && 770 (prefix != addr || ifa->ifa_prefixlen < 32)) { 771 fib_magic(RTM_NEWROUTE, 772 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 773 prefix, ifa->ifa_prefixlen, prim); 774 775 /* Add network specific broadcasts, when it takes a sense */ 776 if (ifa->ifa_prefixlen < 31) { 777 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim); 778 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask, 779 32, prim); 780 } 781 } 782} 783 784/* Delete primary or secondary address. 785 * Optionally, on secondary address promotion consider the addresses 786 * from subnet iprim as deleted, even if they are in device list. 787 * In this case the secondary ifa can be in device list. 788 */ 789void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim) 790{ 791 struct in_device *in_dev = ifa->ifa_dev; 792 struct net_device *dev = in_dev->dev; 793 struct in_ifaddr *ifa1; 794 struct in_ifaddr *prim = ifa, *prim1 = NULL; 795 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask; 796 __be32 any = ifa->ifa_address & ifa->ifa_mask; 797#define LOCAL_OK 1 798#define BRD_OK 2 799#define BRD0_OK 4 800#define BRD1_OK 8 801 unsigned int ok = 0; 802 int subnet = 0; /* Primary network */ 803 int gone = 1; /* Address is missing */ 804 int same_prefsrc = 0; /* Another primary with same IP */ 805 806 if (ifa->ifa_flags & IFA_F_SECONDARY) { 807 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask); 808 if (prim == NULL) { 809 pr_warn("%s: bug: prim == NULL\n", __func__); 810 return; 811 } 812 if (iprim && iprim != prim) { 813 pr_warn("%s: bug: iprim != prim\n", __func__); 814 return; 815 } 816 } else if (!ipv4_is_zeronet(any) && 817 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) { 818 fib_magic(RTM_DELROUTE, 819 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 820 any, ifa->ifa_prefixlen, prim); 821 subnet = 1; 822 } 823 824 /* Deletion is more complicated than add. 825 * We should take care of not to delete too much :-) 826 * 827 * Scan address list to be sure that addresses are really gone. 828 */ 829 830 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) { 831 if (ifa1 == ifa) { 832 /* promotion, keep the IP */ 833 gone = 0; 834 continue; 835 } 836 /* Ignore IFAs from our subnet */ 837 if (iprim && ifa1->ifa_mask == iprim->ifa_mask && 838 inet_ifa_match(ifa1->ifa_address, iprim)) 839 continue; 840 841 /* Ignore ifa1 if it uses different primary IP (prefsrc) */ 842 if (ifa1->ifa_flags & IFA_F_SECONDARY) { 843 /* Another address from our subnet? */ 844 if (ifa1->ifa_mask == prim->ifa_mask && 845 inet_ifa_match(ifa1->ifa_address, prim)) 846 prim1 = prim; 847 else { 848 /* We reached the secondaries, so 849 * same_prefsrc should be determined. 850 */ 851 if (!same_prefsrc) 852 continue; 853 /* Search new prim1 if ifa1 is not 854 * using the current prim1 855 */ 856 if (!prim1 || 857 ifa1->ifa_mask != prim1->ifa_mask || 858 !inet_ifa_match(ifa1->ifa_address, prim1)) 859 prim1 = inet_ifa_byprefix(in_dev, 860 ifa1->ifa_address, 861 ifa1->ifa_mask); 862 if (!prim1) 863 continue; 864 if (prim1->ifa_local != prim->ifa_local) 865 continue; 866 } 867 } else { 868 if (prim->ifa_local != ifa1->ifa_local) 869 continue; 870 prim1 = ifa1; 871 if (prim != prim1) 872 same_prefsrc = 1; 873 } 874 if (ifa->ifa_local == ifa1->ifa_local) 875 ok |= LOCAL_OK; 876 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 877 ok |= BRD_OK; 878 if (brd == ifa1->ifa_broadcast) 879 ok |= BRD1_OK; 880 if (any == ifa1->ifa_broadcast) 881 ok |= BRD0_OK; 882 /* primary has network specific broadcasts */ 883 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) { 884 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask; 885 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask; 886 887 if (!ipv4_is_zeronet(any1)) { 888 if (ifa->ifa_broadcast == brd1 || 889 ifa->ifa_broadcast == any1) 890 ok |= BRD_OK; 891 if (brd == brd1 || brd == any1) 892 ok |= BRD1_OK; 893 if (any == brd1 || any == any1) 894 ok |= BRD0_OK; 895 } 896 } 897 } 898 899 if (!(ok & BRD_OK)) 900 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 901 if (subnet && ifa->ifa_prefixlen < 31) { 902 if (!(ok & BRD1_OK)) 903 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim); 904 if (!(ok & BRD0_OK)) 905 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim); 906 } 907 if (!(ok & LOCAL_OK)) { 908 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim); 909 910 /* Check, that this local address finally disappeared. */ 911 if (gone && 912 inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) { 913 /* And the last, but not the least thing. 914 * We must flush stray FIB entries. 915 * 916 * First of all, we scan fib_info list searching 917 * for stray nexthop entries, then ignite fib_flush. 918 */ 919 if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local)) 920 fib_flush(dev_net(dev)); 921 } 922 } 923#undef LOCAL_OK 924#undef BRD_OK 925#undef BRD0_OK 926#undef BRD1_OK 927} 928 929static void nl_fib_lookup(struct fib_result_nl *frn, struct fib_table *tb) 930{ 931 932 struct fib_result res; 933 struct flowi4 fl4 = { 934 .flowi4_mark = frn->fl_mark, 935 .daddr = frn->fl_addr, 936 .flowi4_tos = frn->fl_tos, 937 .flowi4_scope = frn->fl_scope, 938 }; 939 940 frn->err = -ENOENT; 941 if (tb) { 942 local_bh_disable(); 943 944 frn->tb_id = tb->tb_id; 945 rcu_read_lock(); 946 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 947 948 if (!frn->err) { 949 frn->prefixlen = res.prefixlen; 950 frn->nh_sel = res.nh_sel; 951 frn->type = res.type; 952 frn->scope = res.scope; 953 } 954 rcu_read_unlock(); 955 local_bh_enable(); 956 } 957} 958 959static void nl_fib_input(struct sk_buff *skb) 960{ 961 struct net *net; 962 struct fib_result_nl *frn; 963 struct nlmsghdr *nlh; 964 struct fib_table *tb; 965 u32 portid; 966 967 net = sock_net(skb->sk); 968 nlh = nlmsg_hdr(skb); 969 if (skb->len < NLMSG_SPACE(0) || skb->len < nlh->nlmsg_len || 970 nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*frn))) 971 return; 972 973 skb = skb_clone(skb, GFP_KERNEL); 974 if (skb == NULL) 975 return; 976 nlh = nlmsg_hdr(skb); 977 978 frn = (struct fib_result_nl *) NLMSG_DATA(nlh); 979 tb = fib_get_table(net, frn->tb_id_in); 980 981 nl_fib_lookup(frn, tb); 982 983 portid = NETLINK_CB(skb).portid; /* pid of sending process */ 984 NETLINK_CB(skb).portid = 0; /* from kernel */ 985 NETLINK_CB(skb).dst_group = 0; /* unicast */ 986 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT); 987} 988 989static int __net_init nl_fib_lookup_init(struct net *net) 990{ 991 struct sock *sk; 992 struct netlink_kernel_cfg cfg = { 993 .input = nl_fib_input, 994 }; 995 996 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg); 997 if (sk == NULL) 998 return -EAFNOSUPPORT; 999 net->ipv4.fibnl = sk; 1000 return 0; 1001} 1002 1003static void nl_fib_lookup_exit(struct net *net) 1004{ 1005 netlink_kernel_release(net->ipv4.fibnl); 1006 net->ipv4.fibnl = NULL; 1007} 1008 1009static void fib_disable_ip(struct net_device *dev, int force) 1010{ 1011 if (fib_sync_down_dev(dev, force)) 1012 fib_flush(dev_net(dev)); 1013 rt_cache_flush(dev_net(dev)); 1014 arp_ifdown(dev); 1015} 1016 1017static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 1018{ 1019 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; 1020 struct net_device *dev = ifa->ifa_dev->dev; 1021 struct net *net = dev_net(dev); 1022 1023 switch (event) { 1024 case NETDEV_UP: 1025 fib_add_ifaddr(ifa); 1026#ifdef CONFIG_IP_ROUTE_MULTIPATH 1027 fib_sync_up(dev); 1028#endif 1029 atomic_inc(&net->ipv4.dev_addr_genid); 1030 rt_cache_flush(dev_net(dev)); 1031 break; 1032 case NETDEV_DOWN: 1033 fib_del_ifaddr(ifa, NULL); 1034 atomic_inc(&net->ipv4.dev_addr_genid); 1035 if (ifa->ifa_dev->ifa_list == NULL) { 1036 /* Last address was deleted from this interface. 1037 * Disable IP. 1038 */ 1039 fib_disable_ip(dev, 1); 1040 } else { 1041 rt_cache_flush(dev_net(dev)); 1042 } 1043 break; 1044 } 1045 return NOTIFY_DONE; 1046} 1047 1048static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 1049{ 1050 struct net_device *dev = ptr; 1051 struct in_device *in_dev; 1052 struct net *net = dev_net(dev); 1053 1054 if (event == NETDEV_UNREGISTER) { 1055 fib_disable_ip(dev, 2); 1056 rt_flush_dev(dev); 1057 return NOTIFY_DONE; 1058 } 1059 1060 in_dev = __in_dev_get_rtnl(dev); 1061 1062 switch (event) { 1063 case NETDEV_UP: 1064 for_ifa(in_dev) { 1065 fib_add_ifaddr(ifa); 1066 } endfor_ifa(in_dev); 1067#ifdef CONFIG_IP_ROUTE_MULTIPATH 1068 fib_sync_up(dev); 1069#endif 1070 atomic_inc(&net->ipv4.dev_addr_genid); 1071 rt_cache_flush(net); 1072 break; 1073 case NETDEV_DOWN: 1074 fib_disable_ip(dev, 0); 1075 break; 1076 case NETDEV_CHANGEMTU: 1077 case NETDEV_CHANGE: 1078 rt_cache_flush(net); 1079 break; 1080 } 1081 return NOTIFY_DONE; 1082} 1083 1084static struct notifier_block fib_inetaddr_notifier = { 1085 .notifier_call = fib_inetaddr_event, 1086}; 1087 1088static struct notifier_block fib_netdev_notifier = { 1089 .notifier_call = fib_netdev_event, 1090}; 1091 1092static int __net_init ip_fib_net_init(struct net *net) 1093{ 1094 int err; 1095 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ; 1096 1097 /* Avoid false sharing : Use at least a full cache line */ 1098 size = max_t(size_t, size, L1_CACHE_BYTES); 1099 1100 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL); 1101 if (net->ipv4.fib_table_hash == NULL) 1102 return -ENOMEM; 1103 1104 err = fib4_rules_init(net); 1105 if (err < 0) 1106 goto fail; 1107 return 0; 1108 1109fail: 1110 kfree(net->ipv4.fib_table_hash); 1111 return err; 1112} 1113 1114static void ip_fib_net_exit(struct net *net) 1115{ 1116 unsigned int i; 1117 1118#ifdef CONFIG_IP_MULTIPLE_TABLES 1119 fib4_rules_exit(net); 1120#endif 1121 1122 rtnl_lock(); 1123 for (i = 0; i < FIB_TABLE_HASHSZ; i++) { 1124 struct fib_table *tb; 1125 struct hlist_head *head; 1126 struct hlist_node *node, *tmp; 1127 1128 head = &net->ipv4.fib_table_hash[i]; 1129 hlist_for_each_entry_safe(tb, node, tmp, head, tb_hlist) { 1130 hlist_del(node); 1131 fib_table_flush(tb); 1132 fib_free_table(tb); 1133 } 1134 } 1135 rtnl_unlock(); 1136 kfree(net->ipv4.fib_table_hash); 1137} 1138 1139static int __net_init fib_net_init(struct net *net) 1140{ 1141 int error; 1142 1143#ifdef CONFIG_IP_ROUTE_CLASSID 1144 net->ipv4.fib_num_tclassid_users = 0; 1145#endif 1146 error = ip_fib_net_init(net); 1147 if (error < 0) 1148 goto out; 1149 error = nl_fib_lookup_init(net); 1150 if (error < 0) 1151 goto out_nlfl; 1152 error = fib_proc_init(net); 1153 if (error < 0) 1154 goto out_proc; 1155out: 1156 return error; 1157 1158out_proc: 1159 nl_fib_lookup_exit(net); 1160out_nlfl: 1161 ip_fib_net_exit(net); 1162 goto out; 1163} 1164 1165static void __net_exit fib_net_exit(struct net *net) 1166{ 1167 fib_proc_exit(net); 1168 nl_fib_lookup_exit(net); 1169 ip_fib_net_exit(net); 1170} 1171 1172static struct pernet_operations fib_net_ops = { 1173 .init = fib_net_init, 1174 .exit = fib_net_exit, 1175}; 1176 1177void __init ip_fib_init(void) 1178{ 1179 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL); 1180 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL); 1181 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL); 1182 1183 register_pernet_subsys(&fib_net_ops); 1184 register_netdevice_notifier(&fib_netdev_notifier); 1185 register_inetaddr_notifier(&fib_inetaddr_notifier); 1186 1187 fib_trie_init(); 1188} 1189