fib_frontend.c revision a207a4b2e8067cbc7f33924e7f2c0fa4ef43b459
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/init.h> 35#include <linux/list.h> 36#include <linux/slab.h> 37 38#include <net/ip.h> 39#include <net/protocol.h> 40#include <net/route.h> 41#include <net/tcp.h> 42#include <net/sock.h> 43#include <net/arp.h> 44#include <net/ip_fib.h> 45#include <net/rtnetlink.h> 46#include <net/xfrm.h> 47 48#ifndef CONFIG_IP_MULTIPLE_TABLES 49 50static int __net_init fib4_rules_init(struct net *net) 51{ 52 struct fib_table *local_table, *main_table; 53 54 local_table = fib_trie_table(RT_TABLE_LOCAL); 55 if (local_table == NULL) 56 return -ENOMEM; 57 58 main_table = fib_trie_table(RT_TABLE_MAIN); 59 if (main_table == NULL) 60 goto fail; 61 62 hlist_add_head_rcu(&local_table->tb_hlist, 63 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]); 64 hlist_add_head_rcu(&main_table->tb_hlist, 65 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]); 66 return 0; 67 68fail: 69 kfree(local_table); 70 return -ENOMEM; 71} 72#else 73 74struct fib_table *fib_new_table(struct net *net, u32 id) 75{ 76 struct fib_table *tb; 77 unsigned int h; 78 79 if (id == 0) 80 id = RT_TABLE_MAIN; 81 tb = fib_get_table(net, id); 82 if (tb) 83 return tb; 84 85 tb = fib_trie_table(id); 86 if (!tb) 87 return NULL; 88 h = id & (FIB_TABLE_HASHSZ - 1); 89 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]); 90 return tb; 91} 92 93struct fib_table *fib_get_table(struct net *net, u32 id) 94{ 95 struct fib_table *tb; 96 struct hlist_node *node; 97 struct hlist_head *head; 98 unsigned int h; 99 100 if (id == 0) 101 id = RT_TABLE_MAIN; 102 h = id & (FIB_TABLE_HASHSZ - 1); 103 104 rcu_read_lock(); 105 head = &net->ipv4.fib_table_hash[h]; 106 hlist_for_each_entry_rcu(tb, node, head, tb_hlist) { 107 if (tb->tb_id == id) { 108 rcu_read_unlock(); 109 return tb; 110 } 111 } 112 rcu_read_unlock(); 113 return NULL; 114} 115#endif /* CONFIG_IP_MULTIPLE_TABLES */ 116 117static void fib_flush(struct net *net) 118{ 119 int flushed = 0; 120 struct fib_table *tb; 121 struct hlist_node *node; 122 struct hlist_head *head; 123 unsigned int h; 124 125 for (h = 0; h < FIB_TABLE_HASHSZ; h++) { 126 head = &net->ipv4.fib_table_hash[h]; 127 hlist_for_each_entry(tb, node, head, tb_hlist) 128 flushed += fib_table_flush(tb); 129 } 130 131 if (flushed) 132 rt_cache_flush(net, -1); 133} 134 135/* 136 * Find address type as if only "dev" was present in the system. If 137 * on_dev is NULL then all interfaces are taken into consideration. 138 */ 139static inline unsigned int __inet_dev_addr_type(struct net *net, 140 const struct net_device *dev, 141 __be32 addr) 142{ 143 struct flowi4 fl4 = { .daddr = addr }; 144 struct fib_result res; 145 unsigned int ret = RTN_BROADCAST; 146 struct fib_table *local_table; 147 148 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr)) 149 return RTN_BROADCAST; 150 if (ipv4_is_multicast(addr)) 151 return RTN_MULTICAST; 152 153#ifdef CONFIG_IP_MULTIPLE_TABLES 154 res.r = NULL; 155#endif 156 157 local_table = fib_get_table(net, RT_TABLE_LOCAL); 158 if (local_table) { 159 ret = RTN_UNICAST; 160 rcu_read_lock(); 161 if (!fib_table_lookup(local_table, &fl4, &res, FIB_LOOKUP_NOREF)) { 162 if (!dev || dev == res.fi->fib_dev) 163 ret = res.type; 164 } 165 rcu_read_unlock(); 166 } 167 return ret; 168} 169 170unsigned int inet_addr_type(struct net *net, __be32 addr) 171{ 172 return __inet_dev_addr_type(net, NULL, addr); 173} 174EXPORT_SYMBOL(inet_addr_type); 175 176unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev, 177 __be32 addr) 178{ 179 return __inet_dev_addr_type(net, dev, addr); 180} 181EXPORT_SYMBOL(inet_dev_addr_type); 182 183__be32 fib_compute_spec_dst(struct sk_buff *skb) 184{ 185 struct net_device *dev = skb->dev; 186 struct in_device *in_dev; 187 struct fib_result res; 188 struct rtable *rt; 189 struct flowi4 fl4; 190 struct net *net; 191 int scope; 192 193 rt = skb_rtable(skb); 194 if (!(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) 195 return ip_hdr(skb)->daddr; 196 197 in_dev = __in_dev_get_rcu(dev); 198 BUG_ON(!in_dev); 199 200 net = dev_net(dev); 201 202 scope = RT_SCOPE_UNIVERSE; 203 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) { 204 fl4.flowi4_oif = 0; 205 fl4.flowi4_iif = net->loopback_dev->ifindex; 206 fl4.daddr = ip_hdr(skb)->saddr; 207 fl4.saddr = 0; 208 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos); 209 fl4.flowi4_scope = scope; 210 fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0; 211 if (!fib_lookup(net, &fl4, &res)) 212 return FIB_RES_PREFSRC(net, res); 213 } else { 214 scope = RT_SCOPE_LINK; 215 } 216 217 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope); 218} 219 220/* Given (packet source, input interface) and optional (dst, oif, tos): 221 * - (main) check, that source is valid i.e. not broadcast or our local 222 * address. 223 * - figure out what "logical" interface this packet arrived 224 * and calculate "specific destination" address. 225 * - check, that packet arrived from expected physical interface. 226 * called with rcu_read_lock() 227 */ 228int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, u8 tos, 229 int oif, struct net_device *dev, u32 *itag) 230{ 231 struct in_device *in_dev; 232 struct flowi4 fl4; 233 struct fib_result res; 234 int no_addr, rpf, accept_local; 235 bool dev_match; 236 int ret; 237 struct net *net; 238 239 fl4.flowi4_oif = 0; 240 fl4.flowi4_iif = oif; 241 fl4.daddr = src; 242 fl4.saddr = dst; 243 fl4.flowi4_tos = tos; 244 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 245 246 no_addr = rpf = accept_local = 0; 247 in_dev = __in_dev_get_rcu(dev); 248 if (in_dev) { 249 no_addr = in_dev->ifa_list == NULL; 250 251 /* Ignore rp_filter for packets protected by IPsec. */ 252 rpf = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(in_dev); 253 254 accept_local = IN_DEV_ACCEPT_LOCAL(in_dev); 255 fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0; 256 } 257 258 if (in_dev == NULL) 259 goto e_inval; 260 261 net = dev_net(dev); 262 if (fib_lookup(net, &fl4, &res)) 263 goto last_resort; 264 if (res.type != RTN_UNICAST) { 265 if (res.type != RTN_LOCAL || !accept_local) 266 goto e_inval; 267 } 268 fib_combine_itag(itag, &res); 269 dev_match = false; 270 271#ifdef CONFIG_IP_ROUTE_MULTIPATH 272 for (ret = 0; ret < res.fi->fib_nhs; ret++) { 273 struct fib_nh *nh = &res.fi->fib_nh[ret]; 274 275 if (nh->nh_dev == dev) { 276 dev_match = true; 277 break; 278 } 279 } 280#else 281 if (FIB_RES_DEV(res) == dev) 282 dev_match = true; 283#endif 284 if (dev_match) { 285 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 286 return ret; 287 } 288 if (no_addr) 289 goto last_resort; 290 if (rpf == 1) 291 goto e_rpf; 292 fl4.flowi4_oif = dev->ifindex; 293 294 ret = 0; 295 if (fib_lookup(net, &fl4, &res) == 0) { 296 if (res.type == RTN_UNICAST) 297 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST; 298 } 299 return ret; 300 301last_resort: 302 if (rpf) 303 goto e_rpf; 304 *itag = 0; 305 return 0; 306 307e_inval: 308 return -EINVAL; 309e_rpf: 310 return -EXDEV; 311} 312 313static inline __be32 sk_extract_addr(struct sockaddr *addr) 314{ 315 return ((struct sockaddr_in *) addr)->sin_addr.s_addr; 316} 317 318static int put_rtax(struct nlattr *mx, int len, int type, u32 value) 319{ 320 struct nlattr *nla; 321 322 nla = (struct nlattr *) ((char *) mx + len); 323 nla->nla_type = type; 324 nla->nla_len = nla_attr_size(4); 325 *(u32 *) nla_data(nla) = value; 326 327 return len + nla_total_size(4); 328} 329 330static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt, 331 struct fib_config *cfg) 332{ 333 __be32 addr; 334 int plen; 335 336 memset(cfg, 0, sizeof(*cfg)); 337 cfg->fc_nlinfo.nl_net = net; 338 339 if (rt->rt_dst.sa_family != AF_INET) 340 return -EAFNOSUPPORT; 341 342 /* 343 * Check mask for validity: 344 * a) it must be contiguous. 345 * b) destination must have all host bits clear. 346 * c) if application forgot to set correct family (AF_INET), 347 * reject request unless it is absolutely clear i.e. 348 * both family and mask are zero. 349 */ 350 plen = 32; 351 addr = sk_extract_addr(&rt->rt_dst); 352 if (!(rt->rt_flags & RTF_HOST)) { 353 __be32 mask = sk_extract_addr(&rt->rt_genmask); 354 355 if (rt->rt_genmask.sa_family != AF_INET) { 356 if (mask || rt->rt_genmask.sa_family) 357 return -EAFNOSUPPORT; 358 } 359 360 if (bad_mask(mask, addr)) 361 return -EINVAL; 362 363 plen = inet_mask_len(mask); 364 } 365 366 cfg->fc_dst_len = plen; 367 cfg->fc_dst = addr; 368 369 if (cmd != SIOCDELRT) { 370 cfg->fc_nlflags = NLM_F_CREATE; 371 cfg->fc_protocol = RTPROT_BOOT; 372 } 373 374 if (rt->rt_metric) 375 cfg->fc_priority = rt->rt_metric - 1; 376 377 if (rt->rt_flags & RTF_REJECT) { 378 cfg->fc_scope = RT_SCOPE_HOST; 379 cfg->fc_type = RTN_UNREACHABLE; 380 return 0; 381 } 382 383 cfg->fc_scope = RT_SCOPE_NOWHERE; 384 cfg->fc_type = RTN_UNICAST; 385 386 if (rt->rt_dev) { 387 char *colon; 388 struct net_device *dev; 389 char devname[IFNAMSIZ]; 390 391 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1)) 392 return -EFAULT; 393 394 devname[IFNAMSIZ-1] = 0; 395 colon = strchr(devname, ':'); 396 if (colon) 397 *colon = 0; 398 dev = __dev_get_by_name(net, devname); 399 if (!dev) 400 return -ENODEV; 401 cfg->fc_oif = dev->ifindex; 402 if (colon) { 403 struct in_ifaddr *ifa; 404 struct in_device *in_dev = __in_dev_get_rtnl(dev); 405 if (!in_dev) 406 return -ENODEV; 407 *colon = ':'; 408 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) 409 if (strcmp(ifa->ifa_label, devname) == 0) 410 break; 411 if (ifa == NULL) 412 return -ENODEV; 413 cfg->fc_prefsrc = ifa->ifa_local; 414 } 415 } 416 417 addr = sk_extract_addr(&rt->rt_gateway); 418 if (rt->rt_gateway.sa_family == AF_INET && addr) { 419 cfg->fc_gw = addr; 420 if (rt->rt_flags & RTF_GATEWAY && 421 inet_addr_type(net, addr) == RTN_UNICAST) 422 cfg->fc_scope = RT_SCOPE_UNIVERSE; 423 } 424 425 if (cmd == SIOCDELRT) 426 return 0; 427 428 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw) 429 return -EINVAL; 430 431 if (cfg->fc_scope == RT_SCOPE_NOWHERE) 432 cfg->fc_scope = RT_SCOPE_LINK; 433 434 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) { 435 struct nlattr *mx; 436 int len = 0; 437 438 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL); 439 if (mx == NULL) 440 return -ENOMEM; 441 442 if (rt->rt_flags & RTF_MTU) 443 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40); 444 445 if (rt->rt_flags & RTF_WINDOW) 446 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window); 447 448 if (rt->rt_flags & RTF_IRTT) 449 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3); 450 451 cfg->fc_mx = mx; 452 cfg->fc_mx_len = len; 453 } 454 455 return 0; 456} 457 458/* 459 * Handle IP routing ioctl calls. 460 * These are used to manipulate the routing tables 461 */ 462int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg) 463{ 464 struct fib_config cfg; 465 struct rtentry rt; 466 int err; 467 468 switch (cmd) { 469 case SIOCADDRT: /* Add a route */ 470 case SIOCDELRT: /* Delete a route */ 471 if (!capable(CAP_NET_ADMIN)) 472 return -EPERM; 473 474 if (copy_from_user(&rt, arg, sizeof(rt))) 475 return -EFAULT; 476 477 rtnl_lock(); 478 err = rtentry_to_fib_config(net, cmd, &rt, &cfg); 479 if (err == 0) { 480 struct fib_table *tb; 481 482 if (cmd == SIOCDELRT) { 483 tb = fib_get_table(net, cfg.fc_table); 484 if (tb) 485 err = fib_table_delete(tb, &cfg); 486 else 487 err = -ESRCH; 488 } else { 489 tb = fib_new_table(net, cfg.fc_table); 490 if (tb) 491 err = fib_table_insert(tb, &cfg); 492 else 493 err = -ENOBUFS; 494 } 495 496 /* allocated by rtentry_to_fib_config() */ 497 kfree(cfg.fc_mx); 498 } 499 rtnl_unlock(); 500 return err; 501 } 502 return -EINVAL; 503} 504 505const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = { 506 [RTA_DST] = { .type = NLA_U32 }, 507 [RTA_SRC] = { .type = NLA_U32 }, 508 [RTA_IIF] = { .type = NLA_U32 }, 509 [RTA_OIF] = { .type = NLA_U32 }, 510 [RTA_GATEWAY] = { .type = NLA_U32 }, 511 [RTA_PRIORITY] = { .type = NLA_U32 }, 512 [RTA_PREFSRC] = { .type = NLA_U32 }, 513 [RTA_METRICS] = { .type = NLA_NESTED }, 514 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) }, 515 [RTA_FLOW] = { .type = NLA_U32 }, 516}; 517 518static int rtm_to_fib_config(struct net *net, struct sk_buff *skb, 519 struct nlmsghdr *nlh, struct fib_config *cfg) 520{ 521 struct nlattr *attr; 522 int err, remaining; 523 struct rtmsg *rtm; 524 525 err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy); 526 if (err < 0) 527 goto errout; 528 529 memset(cfg, 0, sizeof(*cfg)); 530 531 rtm = nlmsg_data(nlh); 532 cfg->fc_dst_len = rtm->rtm_dst_len; 533 cfg->fc_tos = rtm->rtm_tos; 534 cfg->fc_table = rtm->rtm_table; 535 cfg->fc_protocol = rtm->rtm_protocol; 536 cfg->fc_scope = rtm->rtm_scope; 537 cfg->fc_type = rtm->rtm_type; 538 cfg->fc_flags = rtm->rtm_flags; 539 cfg->fc_nlflags = nlh->nlmsg_flags; 540 541 cfg->fc_nlinfo.pid = NETLINK_CB(skb).pid; 542 cfg->fc_nlinfo.nlh = nlh; 543 cfg->fc_nlinfo.nl_net = net; 544 545 if (cfg->fc_type > RTN_MAX) { 546 err = -EINVAL; 547 goto errout; 548 } 549 550 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) { 551 switch (nla_type(attr)) { 552 case RTA_DST: 553 cfg->fc_dst = nla_get_be32(attr); 554 break; 555 case RTA_OIF: 556 cfg->fc_oif = nla_get_u32(attr); 557 break; 558 case RTA_GATEWAY: 559 cfg->fc_gw = nla_get_be32(attr); 560 break; 561 case RTA_PRIORITY: 562 cfg->fc_priority = nla_get_u32(attr); 563 break; 564 case RTA_PREFSRC: 565 cfg->fc_prefsrc = nla_get_be32(attr); 566 break; 567 case RTA_METRICS: 568 cfg->fc_mx = nla_data(attr); 569 cfg->fc_mx_len = nla_len(attr); 570 break; 571 case RTA_MULTIPATH: 572 cfg->fc_mp = nla_data(attr); 573 cfg->fc_mp_len = nla_len(attr); 574 break; 575 case RTA_FLOW: 576 cfg->fc_flow = nla_get_u32(attr); 577 break; 578 case RTA_TABLE: 579 cfg->fc_table = nla_get_u32(attr); 580 break; 581 } 582 } 583 584 return 0; 585errout: 586 return err; 587} 588 589static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg) 590{ 591 struct net *net = sock_net(skb->sk); 592 struct fib_config cfg; 593 struct fib_table *tb; 594 int err; 595 596 err = rtm_to_fib_config(net, skb, nlh, &cfg); 597 if (err < 0) 598 goto errout; 599 600 tb = fib_get_table(net, cfg.fc_table); 601 if (tb == NULL) { 602 err = -ESRCH; 603 goto errout; 604 } 605 606 err = fib_table_delete(tb, &cfg); 607errout: 608 return err; 609} 610 611static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg) 612{ 613 struct net *net = sock_net(skb->sk); 614 struct fib_config cfg; 615 struct fib_table *tb; 616 int err; 617 618 err = rtm_to_fib_config(net, skb, nlh, &cfg); 619 if (err < 0) 620 goto errout; 621 622 tb = fib_new_table(net, cfg.fc_table); 623 if (tb == NULL) { 624 err = -ENOBUFS; 625 goto errout; 626 } 627 628 err = fib_table_insert(tb, &cfg); 629errout: 630 return err; 631} 632 633static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) 634{ 635 struct net *net = sock_net(skb->sk); 636 unsigned int h, s_h; 637 unsigned int e = 0, s_e; 638 struct fib_table *tb; 639 struct hlist_node *node; 640 struct hlist_head *head; 641 int dumped = 0; 642 643 if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) && 644 ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED) 645 return ip_rt_dump(skb, cb); 646 647 s_h = cb->args[0]; 648 s_e = cb->args[1]; 649 650 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) { 651 e = 0; 652 head = &net->ipv4.fib_table_hash[h]; 653 hlist_for_each_entry(tb, node, head, tb_hlist) { 654 if (e < s_e) 655 goto next; 656 if (dumped) 657 memset(&cb->args[2], 0, sizeof(cb->args) - 658 2 * sizeof(cb->args[0])); 659 if (fib_table_dump(tb, skb, cb) < 0) 660 goto out; 661 dumped = 1; 662next: 663 e++; 664 } 665 } 666out: 667 cb->args[1] = e; 668 cb->args[0] = h; 669 670 return skb->len; 671} 672 673/* Prepare and feed intra-kernel routing request. 674 * Really, it should be netlink message, but :-( netlink 675 * can be not configured, so that we feed it directly 676 * to fib engine. It is legal, because all events occur 677 * only when netlink is already locked. 678 */ 679static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa) 680{ 681 struct net *net = dev_net(ifa->ifa_dev->dev); 682 struct fib_table *tb; 683 struct fib_config cfg = { 684 .fc_protocol = RTPROT_KERNEL, 685 .fc_type = type, 686 .fc_dst = dst, 687 .fc_dst_len = dst_len, 688 .fc_prefsrc = ifa->ifa_local, 689 .fc_oif = ifa->ifa_dev->dev->ifindex, 690 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND, 691 .fc_nlinfo = { 692 .nl_net = net, 693 }, 694 }; 695 696 if (type == RTN_UNICAST) 697 tb = fib_new_table(net, RT_TABLE_MAIN); 698 else 699 tb = fib_new_table(net, RT_TABLE_LOCAL); 700 701 if (tb == NULL) 702 return; 703 704 cfg.fc_table = tb->tb_id; 705 706 if (type != RTN_LOCAL) 707 cfg.fc_scope = RT_SCOPE_LINK; 708 else 709 cfg.fc_scope = RT_SCOPE_HOST; 710 711 if (cmd == RTM_NEWROUTE) 712 fib_table_insert(tb, &cfg); 713 else 714 fib_table_delete(tb, &cfg); 715} 716 717void fib_add_ifaddr(struct in_ifaddr *ifa) 718{ 719 struct in_device *in_dev = ifa->ifa_dev; 720 struct net_device *dev = in_dev->dev; 721 struct in_ifaddr *prim = ifa; 722 __be32 mask = ifa->ifa_mask; 723 __be32 addr = ifa->ifa_local; 724 __be32 prefix = ifa->ifa_address & mask; 725 726 if (ifa->ifa_flags & IFA_F_SECONDARY) { 727 prim = inet_ifa_byprefix(in_dev, prefix, mask); 728 if (prim == NULL) { 729 pr_warn("%s: bug: prim == NULL\n", __func__); 730 return; 731 } 732 } 733 734 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim); 735 736 if (!(dev->flags & IFF_UP)) 737 return; 738 739 /* Add broadcast address, if it is explicitly assigned. */ 740 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) 741 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 742 743 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) && 744 (prefix != addr || ifa->ifa_prefixlen < 32)) { 745 fib_magic(RTM_NEWROUTE, 746 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 747 prefix, ifa->ifa_prefixlen, prim); 748 749 /* Add network specific broadcasts, when it takes a sense */ 750 if (ifa->ifa_prefixlen < 31) { 751 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim); 752 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask, 753 32, prim); 754 } 755 } 756} 757 758/* Delete primary or secondary address. 759 * Optionally, on secondary address promotion consider the addresses 760 * from subnet iprim as deleted, even if they are in device list. 761 * In this case the secondary ifa can be in device list. 762 */ 763void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim) 764{ 765 struct in_device *in_dev = ifa->ifa_dev; 766 struct net_device *dev = in_dev->dev; 767 struct in_ifaddr *ifa1; 768 struct in_ifaddr *prim = ifa, *prim1 = NULL; 769 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask; 770 __be32 any = ifa->ifa_address & ifa->ifa_mask; 771#define LOCAL_OK 1 772#define BRD_OK 2 773#define BRD0_OK 4 774#define BRD1_OK 8 775 unsigned int ok = 0; 776 int subnet = 0; /* Primary network */ 777 int gone = 1; /* Address is missing */ 778 int same_prefsrc = 0; /* Another primary with same IP */ 779 780 if (ifa->ifa_flags & IFA_F_SECONDARY) { 781 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask); 782 if (prim == NULL) { 783 pr_warn("%s: bug: prim == NULL\n", __func__); 784 return; 785 } 786 if (iprim && iprim != prim) { 787 pr_warn("%s: bug: iprim != prim\n", __func__); 788 return; 789 } 790 } else if (!ipv4_is_zeronet(any) && 791 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) { 792 fib_magic(RTM_DELROUTE, 793 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST, 794 any, ifa->ifa_prefixlen, prim); 795 subnet = 1; 796 } 797 798 /* Deletion is more complicated than add. 799 * We should take care of not to delete too much :-) 800 * 801 * Scan address list to be sure that addresses are really gone. 802 */ 803 804 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) { 805 if (ifa1 == ifa) { 806 /* promotion, keep the IP */ 807 gone = 0; 808 continue; 809 } 810 /* Ignore IFAs from our subnet */ 811 if (iprim && ifa1->ifa_mask == iprim->ifa_mask && 812 inet_ifa_match(ifa1->ifa_address, iprim)) 813 continue; 814 815 /* Ignore ifa1 if it uses different primary IP (prefsrc) */ 816 if (ifa1->ifa_flags & IFA_F_SECONDARY) { 817 /* Another address from our subnet? */ 818 if (ifa1->ifa_mask == prim->ifa_mask && 819 inet_ifa_match(ifa1->ifa_address, prim)) 820 prim1 = prim; 821 else { 822 /* We reached the secondaries, so 823 * same_prefsrc should be determined. 824 */ 825 if (!same_prefsrc) 826 continue; 827 /* Search new prim1 if ifa1 is not 828 * using the current prim1 829 */ 830 if (!prim1 || 831 ifa1->ifa_mask != prim1->ifa_mask || 832 !inet_ifa_match(ifa1->ifa_address, prim1)) 833 prim1 = inet_ifa_byprefix(in_dev, 834 ifa1->ifa_address, 835 ifa1->ifa_mask); 836 if (!prim1) 837 continue; 838 if (prim1->ifa_local != prim->ifa_local) 839 continue; 840 } 841 } else { 842 if (prim->ifa_local != ifa1->ifa_local) 843 continue; 844 prim1 = ifa1; 845 if (prim != prim1) 846 same_prefsrc = 1; 847 } 848 if (ifa->ifa_local == ifa1->ifa_local) 849 ok |= LOCAL_OK; 850 if (ifa->ifa_broadcast == ifa1->ifa_broadcast) 851 ok |= BRD_OK; 852 if (brd == ifa1->ifa_broadcast) 853 ok |= BRD1_OK; 854 if (any == ifa1->ifa_broadcast) 855 ok |= BRD0_OK; 856 /* primary has network specific broadcasts */ 857 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) { 858 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask; 859 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask; 860 861 if (!ipv4_is_zeronet(any1)) { 862 if (ifa->ifa_broadcast == brd1 || 863 ifa->ifa_broadcast == any1) 864 ok |= BRD_OK; 865 if (brd == brd1 || brd == any1) 866 ok |= BRD1_OK; 867 if (any == brd1 || any == any1) 868 ok |= BRD0_OK; 869 } 870 } 871 } 872 873 if (!(ok & BRD_OK)) 874 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim); 875 if (subnet && ifa->ifa_prefixlen < 31) { 876 if (!(ok & BRD1_OK)) 877 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim); 878 if (!(ok & BRD0_OK)) 879 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim); 880 } 881 if (!(ok & LOCAL_OK)) { 882 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim); 883 884 /* Check, that this local address finally disappeared. */ 885 if (gone && 886 inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) { 887 /* And the last, but not the least thing. 888 * We must flush stray FIB entries. 889 * 890 * First of all, we scan fib_info list searching 891 * for stray nexthop entries, then ignite fib_flush. 892 */ 893 if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local)) 894 fib_flush(dev_net(dev)); 895 } 896 } 897#undef LOCAL_OK 898#undef BRD_OK 899#undef BRD0_OK 900#undef BRD1_OK 901} 902 903static void nl_fib_lookup(struct fib_result_nl *frn, struct fib_table *tb) 904{ 905 906 struct fib_result res; 907 struct flowi4 fl4 = { 908 .flowi4_mark = frn->fl_mark, 909 .daddr = frn->fl_addr, 910 .flowi4_tos = frn->fl_tos, 911 .flowi4_scope = frn->fl_scope, 912 }; 913 914#ifdef CONFIG_IP_MULTIPLE_TABLES 915 res.r = NULL; 916#endif 917 918 frn->err = -ENOENT; 919 if (tb) { 920 local_bh_disable(); 921 922 frn->tb_id = tb->tb_id; 923 rcu_read_lock(); 924 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 925 926 if (!frn->err) { 927 frn->prefixlen = res.prefixlen; 928 frn->nh_sel = res.nh_sel; 929 frn->type = res.type; 930 frn->scope = res.scope; 931 } 932 rcu_read_unlock(); 933 local_bh_enable(); 934 } 935} 936 937static void nl_fib_input(struct sk_buff *skb) 938{ 939 struct net *net; 940 struct fib_result_nl *frn; 941 struct nlmsghdr *nlh; 942 struct fib_table *tb; 943 u32 pid; 944 945 net = sock_net(skb->sk); 946 nlh = nlmsg_hdr(skb); 947 if (skb->len < NLMSG_SPACE(0) || skb->len < nlh->nlmsg_len || 948 nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*frn))) 949 return; 950 951 skb = skb_clone(skb, GFP_KERNEL); 952 if (skb == NULL) 953 return; 954 nlh = nlmsg_hdr(skb); 955 956 frn = (struct fib_result_nl *) NLMSG_DATA(nlh); 957 tb = fib_get_table(net, frn->tb_id_in); 958 959 nl_fib_lookup(frn, tb); 960 961 pid = NETLINK_CB(skb).pid; /* pid of sending process */ 962 NETLINK_CB(skb).pid = 0; /* from kernel */ 963 NETLINK_CB(skb).dst_group = 0; /* unicast */ 964 netlink_unicast(net->ipv4.fibnl, skb, pid, MSG_DONTWAIT); 965} 966 967static int __net_init nl_fib_lookup_init(struct net *net) 968{ 969 struct sock *sk; 970 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, 0, 971 nl_fib_input, NULL, THIS_MODULE); 972 if (sk == NULL) 973 return -EAFNOSUPPORT; 974 net->ipv4.fibnl = sk; 975 return 0; 976} 977 978static void nl_fib_lookup_exit(struct net *net) 979{ 980 netlink_kernel_release(net->ipv4.fibnl); 981 net->ipv4.fibnl = NULL; 982} 983 984static void fib_disable_ip(struct net_device *dev, int force, int delay) 985{ 986 if (fib_sync_down_dev(dev, force)) 987 fib_flush(dev_net(dev)); 988 rt_cache_flush(dev_net(dev), delay); 989 arp_ifdown(dev); 990} 991 992static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr) 993{ 994 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; 995 struct net_device *dev = ifa->ifa_dev->dev; 996 struct net *net = dev_net(dev); 997 998 switch (event) { 999 case NETDEV_UP: 1000 fib_add_ifaddr(ifa); 1001#ifdef CONFIG_IP_ROUTE_MULTIPATH 1002 fib_sync_up(dev); 1003#endif 1004 atomic_inc(&net->ipv4.dev_addr_genid); 1005 rt_cache_flush(dev_net(dev), -1); 1006 break; 1007 case NETDEV_DOWN: 1008 fib_del_ifaddr(ifa, NULL); 1009 atomic_inc(&net->ipv4.dev_addr_genid); 1010 if (ifa->ifa_dev->ifa_list == NULL) { 1011 /* Last address was deleted from this interface. 1012 * Disable IP. 1013 */ 1014 fib_disable_ip(dev, 1, 0); 1015 } else { 1016 rt_cache_flush(dev_net(dev), -1); 1017 } 1018 break; 1019 } 1020 return NOTIFY_DONE; 1021} 1022 1023static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr) 1024{ 1025 struct net_device *dev = ptr; 1026 struct in_device *in_dev = __in_dev_get_rtnl(dev); 1027 struct net *net = dev_net(dev); 1028 1029 if (event == NETDEV_UNREGISTER) { 1030 fib_disable_ip(dev, 2, -1); 1031 return NOTIFY_DONE; 1032 } 1033 1034 if (!in_dev) 1035 return NOTIFY_DONE; 1036 1037 switch (event) { 1038 case NETDEV_UP: 1039 for_ifa(in_dev) { 1040 fib_add_ifaddr(ifa); 1041 } endfor_ifa(in_dev); 1042#ifdef CONFIG_IP_ROUTE_MULTIPATH 1043 fib_sync_up(dev); 1044#endif 1045 atomic_inc(&net->ipv4.dev_addr_genid); 1046 rt_cache_flush(dev_net(dev), -1); 1047 break; 1048 case NETDEV_DOWN: 1049 fib_disable_ip(dev, 0, 0); 1050 break; 1051 case NETDEV_CHANGEMTU: 1052 case NETDEV_CHANGE: 1053 rt_cache_flush(dev_net(dev), 0); 1054 break; 1055 case NETDEV_UNREGISTER_BATCH: 1056 /* The batch unregister is only called on the first 1057 * device in the list of devices being unregistered. 1058 * Therefore we should not pass dev_net(dev) in here. 1059 */ 1060 rt_cache_flush_batch(NULL); 1061 break; 1062 } 1063 return NOTIFY_DONE; 1064} 1065 1066static struct notifier_block fib_inetaddr_notifier = { 1067 .notifier_call = fib_inetaddr_event, 1068}; 1069 1070static struct notifier_block fib_netdev_notifier = { 1071 .notifier_call = fib_netdev_event, 1072}; 1073 1074static int __net_init ip_fib_net_init(struct net *net) 1075{ 1076 int err; 1077 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ; 1078 1079 /* Avoid false sharing : Use at least a full cache line */ 1080 size = max_t(size_t, size, L1_CACHE_BYTES); 1081 1082 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL); 1083 if (net->ipv4.fib_table_hash == NULL) 1084 return -ENOMEM; 1085 1086 err = fib4_rules_init(net); 1087 if (err < 0) 1088 goto fail; 1089 return 0; 1090 1091fail: 1092 kfree(net->ipv4.fib_table_hash); 1093 return err; 1094} 1095 1096static void ip_fib_net_exit(struct net *net) 1097{ 1098 unsigned int i; 1099 1100#ifdef CONFIG_IP_MULTIPLE_TABLES 1101 fib4_rules_exit(net); 1102#endif 1103 1104 rtnl_lock(); 1105 for (i = 0; i < FIB_TABLE_HASHSZ; i++) { 1106 struct fib_table *tb; 1107 struct hlist_head *head; 1108 struct hlist_node *node, *tmp; 1109 1110 head = &net->ipv4.fib_table_hash[i]; 1111 hlist_for_each_entry_safe(tb, node, tmp, head, tb_hlist) { 1112 hlist_del(node); 1113 fib_table_flush(tb); 1114 fib_free_table(tb); 1115 } 1116 } 1117 rtnl_unlock(); 1118 kfree(net->ipv4.fib_table_hash); 1119} 1120 1121static int __net_init fib_net_init(struct net *net) 1122{ 1123 int error; 1124 1125 error = ip_fib_net_init(net); 1126 if (error < 0) 1127 goto out; 1128 error = nl_fib_lookup_init(net); 1129 if (error < 0) 1130 goto out_nlfl; 1131 error = fib_proc_init(net); 1132 if (error < 0) 1133 goto out_proc; 1134out: 1135 return error; 1136 1137out_proc: 1138 nl_fib_lookup_exit(net); 1139out_nlfl: 1140 ip_fib_net_exit(net); 1141 goto out; 1142} 1143 1144static void __net_exit fib_net_exit(struct net *net) 1145{ 1146 fib_proc_exit(net); 1147 nl_fib_lookup_exit(net); 1148 ip_fib_net_exit(net); 1149} 1150 1151static struct pernet_operations fib_net_ops = { 1152 .init = fib_net_init, 1153 .exit = fib_net_exit, 1154}; 1155 1156void __init ip_fib_init(void) 1157{ 1158 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL); 1159 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL); 1160 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL); 1161 1162 register_pernet_subsys(&fib_net_ops); 1163 register_netdevice_notifier(&fib_netdev_notifier); 1164 register_inetaddr_notifier(&fib_inetaddr_notifier); 1165 1166 fib_trie_init(); 1167} 1168