net/ipv4/fib_frontend.c

   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/vrf.h>
  49
  50 #ifndef CONFIG_IP_MULTIPLE_TABLES
  51
  52 static int __net_init fib4_rules_init(struct net *net)
  53 {
  54         struct fib_table *local_table, *main_table;
  55
  56         main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
  57         if (!main_table)
  58                 return -ENOMEM;
  59
  60         local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
  61         if (!local_table)
  62                 goto fail;
  63
  64         hlist_add_head_rcu(&local_table->tb_hlist,
  65                                 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
  66         hlist_add_head_rcu(&main_table->tb_hlist,
  67                                 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
  68         return 0;
  69
  70 fail:
  71         fib_free_table(main_table);
  72         return -ENOMEM;
  73 }
  74 #else
  75
  76 struct fib_table *fib_new_table(struct net *net, u32 id)
  77 {
  78         struct fib_table *tb, *alias = NULL;
  79         unsigned int h;
  80
  81         if (id == 0)
  82                 id = RT_TABLE_MAIN;
  83         tb = fib_get_table(net, id);
  84         if (tb)
  85                 return tb;
  86
  87         if (id == RT_TABLE_LOCAL)
  88                 alias = fib_new_table(net, RT_TABLE_MAIN);
  89
  90         tb = fib_trie_table(id, alias);
  91         if (!tb)
  92                 return NULL;
  93
  94         switch (id) {
  95         case RT_TABLE_LOCAL:
  96                 rcu_assign_pointer(net->ipv4.fib_local, tb);
  97                 break;
  98         case RT_TABLE_MAIN:
  99                 rcu_assign_pointer(net->ipv4.fib_main, tb);
 100                 break;
 101         case RT_TABLE_DEFAULT:
 102                 rcu_assign_pointer(net->ipv4.fib_default, tb);
 103                 break;
 104         default:
 105                 break;
 106         }
 107
 108         h = id & (FIB_TABLE_HASHSZ - 1);
 109         hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
 110         return tb;
 111 }
 112
 113 /* caller must hold either rtnl or rcu read lock */
 114 struct fib_table *fib_get_table(struct net *net, u32 id)
 115 {
 116         struct fib_table *tb;
 117         struct hlist_head *head;
 118         unsigned int h;
 119
 120         if (id == 0)
 121                 id = RT_TABLE_MAIN;
 122         h = id & (FIB_TABLE_HASHSZ - 1);
 123
 124         head = &net->ipv4.fib_table_hash[h];
 125         hlist_for_each_entry_rcu(tb, head, tb_hlist) {
 126                 if (tb->tb_id == id)
 127                         return tb;
 128         }
 129         return NULL;
 130 }
 131 #endif /* CONFIG_IP_MULTIPLE_TABLES */
 132
 133 static void fib_replace_table(struct net *net, struct fib_table *old,
 134                               struct fib_table *new)
 135 {
 136 #ifdef CONFIG_IP_MULTIPLE_TABLES
 137         switch (new->tb_id) {
 138         case RT_TABLE_LOCAL:
 139                 rcu_assign_pointer(net->ipv4.fib_local, new);
 140                 break;
 141         case RT_TABLE_MAIN:
 142                 rcu_assign_pointer(net->ipv4.fib_main, new);
 143                 break;
 144         case RT_TABLE_DEFAULT:
 145                 rcu_assign_pointer(net->ipv4.fib_default, new);
 146                 break;
 147         default:
 148                 break;
 149         }
 150
 151 #endif
 152         /* replace the old table in the hlist */
 153         hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
 154 }
 155
 156 int fib_unmerge(struct net *net)
 157 {
 158         struct fib_table *old, *new;
 159
 160         /* attempt to fetch local table if it has been allocated */
 161         old = fib_get_table(net, RT_TABLE_LOCAL);
 162         if (!old)
 163                 return 0;
 164
 165         new = fib_trie_unmerge(old);
 166         if (!new)
 167                 return -ENOMEM;
 168
 169         /* replace merged table with clean table */
 170         if (new != old) {
 171                 fib_replace_table(net, old, new);
 172                 fib_free_table(old);
 173         }
 174
 175         return 0;
 176 }
 177
 178 static void fib_flush(struct net *net)
 179 {
 180         int flushed = 0;
 181         unsigned int h;
 182
 183         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
 184                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
 185                 struct hlist_node *tmp;
 186                 struct fib_table *tb;
 187
 188                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
 189                         flushed += fib_table_flush(tb);
 190         }
 191
 192         if (flushed)
 193                 rt_cache_flush(net);
 194 }
 195
 196 void fib_flush_external(struct net *net)
 197 {
 198         struct fib_table *tb;
 199         struct hlist_head *head;
 200         unsigned int h;
 201
 202         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
 203                 head = &net->ipv4.fib_table_hash[h];
 204                 hlist_for_each_entry(tb, head, tb_hlist)
 205                         fib_table_flush_external(tb);
 206         }
 207 }
 208
 209 /*
 210  * Find address type as if only "dev" was present in the system. If
 211  * on_dev is NULL then all interfaces are taken into consideration.
 212  */
 213 static inline unsigned int __inet_dev_addr_type(struct net *net,
 214                                                 const struct net_device *dev,
 215                                                 __be32 addr, int tb_id)
 216 {
 217         struct flowi4           fl4 = { .daddr = addr };
 218         struct fib_result       res;
 219         unsigned int ret = RTN_BROADCAST;
 220         struct fib_table *table;
 221
 222         if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
 223                 return RTN_BROADCAST;
 224         if (ipv4_is_multicast(addr))
 225                 return RTN_MULTICAST;
 226
 227         rcu_read_lock();
 228
 229         table = fib_get_table(net, tb_id);
 230         if (table) {
 231                 ret = RTN_UNICAST;
 232                 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
 233                         if (!dev || dev == res.fi->fib_dev)
 234                                 ret = res.type;
 235                 }
 236         }
 237
 238         rcu_read_unlock();
 239         return ret;
 240 }
 241
 242 unsigned int inet_addr_type_table(struct net *net, __be32 addr, int tb_id)
 243 {
 244         return __inet_dev_addr_type(net, NULL, addr, tb_id);
 245 }
 246 EXPORT_SYMBOL(inet_addr_type_table);
 247
 248 unsigned int inet_addr_type(struct net *net, __be32 addr)
 249 {
 250         return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
 251 }
 252 EXPORT_SYMBOL(inet_addr_type);
 253
 254 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
 255                                 __be32 addr)
 256 {
 257         int rt_table = vrf_dev_table(dev) ? : RT_TABLE_LOCAL;
 258
 259         return __inet_dev_addr_type(net, dev, addr, rt_table);
 260 }
 261 EXPORT_SYMBOL(inet_dev_addr_type);
 262
 263 __be32 fib_compute_spec_dst(struct sk_buff *skb)
 264 {
 265         struct net_device *dev = skb->dev;
 266         struct in_device *in_dev;
 267         struct fib_result res;
 268         struct rtable *rt;
 269         struct flowi4 fl4;
 270         struct net *net;
 271         int scope;
 272
 273         rt = skb_rtable(skb);
 274         if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
 275             RTCF_LOCAL)
 276                 return ip_hdr(skb)->daddr;
 277
 278         in_dev = __in_dev_get_rcu(dev);
 279         BUG_ON(!in_dev);
 280
 281         net = dev_net(dev);
 282
 283         scope = RT_SCOPE_UNIVERSE;
 284         if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
 285                 fl4.flowi4_oif = 0;
 286                 fl4.flowi4_iif = LOOPBACK_IFINDEX;
 287                 fl4.daddr = ip_hdr(skb)->saddr;
 288                 fl4.saddr = 0;
 289                 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
 290                 fl4.flowi4_scope = scope;
 291                 fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0;
 292                 fl4.flowi4_tun_key.tun_id = 0;
 293                 if (!fib_lookup(net, &fl4, &res, 0))
 294                         return FIB_RES_PREFSRC(net, res);
 295         } else {
 296                 scope = RT_SCOPE_LINK;
 297         }
 298
 299         return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
 300 }
 301
 302 /* Given (packet source, input interface) and optional (dst, oif, tos):
 303  * - (main) check, that source is valid i.e. not broadcast or our local
 304  *   address.
 305  * - figure out what "logical" interface this packet arrived
 306  *   and calculate "specific destination" address.
 307  * - check, that packet arrived from expected physical interface.
 308  * called with rcu_read_lock()
 309  */
 310 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 311                                  u8 tos, int oif, struct net_device *dev,
 312                                  int rpf, struct in_device *idev, u32 *itag)
 313 {
 314         int ret, no_addr;
 315         struct fib_result res;
 316         struct flowi4 fl4;
 317         struct net *net;
 318         bool dev_match;
 319
 320         fl4.flowi4_oif = 0;
 321         fl4.flowi4_iif = vrf_master_ifindex_rcu(dev);
 322         if (!fl4.flowi4_iif)
 323                 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
 324         fl4.daddr = src;
 325         fl4.saddr = dst;
 326         fl4.flowi4_tos = tos;
 327         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 328         fl4.flowi4_tun_key.tun_id = 0;
 329
 330         no_addr = idev->ifa_list == NULL;
 331
 332         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
 333
 334         net = dev_net(dev);
 335         if (fib_lookup(net, &fl4, &res, 0))
 336                 goto last_resort;
 337         if (res.type != RTN_UNICAST &&
 338             (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
 339                 goto e_inval;
 340         if (!rpf && !fib_num_tclassid_users(dev_net(dev)) &&
 341             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev)))
 342                 goto last_resort;
 343         fib_combine_itag(itag, &res);
 344         dev_match = false;
 345
 346 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 347         for (ret = 0; ret < res.fi->fib_nhs; ret++) {
 348                 struct fib_nh *nh = &res.fi->fib_nh[ret];
 349
 350                 if (nh->nh_dev == dev) {
 351                         dev_match = true;
 352                         break;
 353                 } else if (vrf_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) {
 354                         dev_match = true;
 355                         break;
 356                 }
 357         }
 358 #else
 359         if (FIB_RES_DEV(res) == dev)
 360                 dev_match = true;
 361 #endif
 362         if (dev_match) {
 363                 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 364                 return ret;
 365         }
 366         if (no_addr)
 367                 goto last_resort;
 368         if (rpf == 1)
 369                 goto e_rpf;
 370         fl4.flowi4_oif = dev->ifindex;
 371
 372         ret = 0;
 373         if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
 374                 if (res.type == RTN_UNICAST)
 375                         ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 376         }
 377         return ret;
 378
 379 last_resort:
 380         if (rpf)
 381                 goto e_rpf;
 382         *itag = 0;
 383         return 0;
 384
 385 e_inval:
 386         return -EINVAL;
 387 e_rpf:
 388         return -EXDEV;
 389 }
 390
 391 /* Ignore rp_filter for packets protected by IPsec. */
 392 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 393                         u8 tos, int oif, struct net_device *dev,
 394                         struct in_device *idev, u32 *itag)
 395 {
 396         int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
 397
 398         if (!r && !fib_num_tclassid_users(dev_net(dev)) &&
 399             IN_DEV_ACCEPT_LOCAL(idev) &&
 400             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
 401                 *itag = 0;
 402                 return 0;
 403         }
 404         return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
 405 }
 406
 407 static inline __be32 sk_extract_addr(struct sockaddr *addr)
 408 {
 409         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 410 }
 411
 412 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
 413 {
 414         struct nlattr *nla;
 415
 416         nla = (struct nlattr *) ((char *) mx + len);
 417         nla->nla_type = type;
 418         nla->nla_len = nla_attr_size(4);
 419         *(u32 *) nla_data(nla) = value;
 420
 421         return len + nla_total_size(4);
 422 }
 423
 424 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
 425                                  struct fib_config *cfg)
 426 {
 427         __be32 addr;
 428         int plen;
 429
 430         memset(cfg, 0, sizeof(*cfg));
 431         cfg->fc_nlinfo.nl_net = net;
 432
 433         if (rt->rt_dst.sa_family != AF_INET)
 434                 return -EAFNOSUPPORT;
 435
 436         /*
 437          * Check mask for validity:
 438          * a) it must be contiguous.
 439          * b) destination must have all host bits clear.
 440          * c) if application forgot to set correct family (AF_INET),
 441          *    reject request unless it is absolutely clear i.e.
 442          *    both family and mask are zero.
 443          */
 444         plen = 32;
 445         addr = sk_extract_addr(&rt->rt_dst);
 446         if (!(rt->rt_flags & RTF_HOST)) {
 447                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
 448
 449                 if (rt->rt_genmask.sa_family != AF_INET) {
 450                         if (mask || rt->rt_genmask.sa_family)
 451                                 return -EAFNOSUPPORT;
 452                 }
 453
 454                 if (bad_mask(mask, addr))
 455                         return -EINVAL;
 456
 457                 plen = inet_mask_len(mask);
 458         }
 459
 460         cfg->fc_dst_len = plen;
 461         cfg->fc_dst = addr;
 462
 463         if (cmd != SIOCDELRT) {
 464                 cfg->fc_nlflags = NLM_F_CREATE;
 465                 cfg->fc_protocol = RTPROT_BOOT;
 466         }
 467
 468         if (rt->rt_metric)
 469                 cfg->fc_priority = rt->rt_metric - 1;
 470
 471         if (rt->rt_flags & RTF_REJECT) {
 472                 cfg->fc_scope = RT_SCOPE_HOST;
 473                 cfg->fc_type = RTN_UNREACHABLE;
 474                 return 0;
 475         }
 476
 477         cfg->fc_scope = RT_SCOPE_NOWHERE;
 478         cfg->fc_type = RTN_UNICAST;
 479
 480         if (rt->rt_dev) {
 481                 char *colon;
 482                 struct net_device *dev;
 483                 char devname[IFNAMSIZ];
 484
 485                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
 486                         return -EFAULT;
 487
 488                 devname[IFNAMSIZ-1] = 0;
 489                 colon = strchr(devname, ':');
 490                 if (colon)
 491                         *colon = 0;
 492                 dev = __dev_get_by_name(net, devname);
 493                 if (!dev)
 494                         return -ENODEV;
 495                 cfg->fc_oif = dev->ifindex;
 496                 if (colon) {
 497                         struct in_ifaddr *ifa;
 498                         struct in_device *in_dev = __in_dev_get_rtnl(dev);
 499                         if (!in_dev)
 500                                 return -ENODEV;
 501                         *colon = ':';
 502                         for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
 503                                 if (strcmp(ifa->ifa_label, devname) == 0)
 504                                         break;
 505                         if (!ifa)
 506                                 return -ENODEV;
 507                         cfg->fc_prefsrc = ifa->ifa_local;
 508                 }
 509         }
 510
 511         addr = sk_extract_addr(&rt->rt_gateway);
 512         if (rt->rt_gateway.sa_family == AF_INET && addr) {
 513                 cfg->fc_gw = addr;
 514                 if (rt->rt_flags & RTF_GATEWAY &&
 515                     inet_addr_type(net, addr) == RTN_UNICAST)
 516                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
 517         }
 518
 519         if (cmd == SIOCDELRT)
 520                 return 0;
 521
 522         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
 523                 return -EINVAL;
 524
 525         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 526                 cfg->fc_scope = RT_SCOPE_LINK;
 527
 528         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 529                 struct nlattr *mx;
 530                 int len = 0;
 531
 532                 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
 533                 if (!mx)
 534                         return -ENOMEM;
 535
 536                 if (rt->rt_flags & RTF_MTU)
 537                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 538
 539                 if (rt->rt_flags & RTF_WINDOW)
 540                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 541
 542                 if (rt->rt_flags & RTF_IRTT)
 543                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 544
 545                 cfg->fc_mx = mx;
 546                 cfg->fc_mx_len = len;
 547         }
 548
 549         return 0;
 550 }
 551
 552 /*
 553  * Handle IP routing ioctl calls.
 554  * These are used to manipulate the routing tables
 555  */
 556 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
 557 {
 558         struct fib_config cfg;
 559         struct rtentry rt;
 560         int err;
 561
 562         switch (cmd) {
 563         case SIOCADDRT:         /* Add a route */
 564         case SIOCDELRT:         /* Delete a route */
 565                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 566                         return -EPERM;
 567
 568                 if (copy_from_user(&rt, arg, sizeof(rt)))
 569                         return -EFAULT;
 570
 571                 rtnl_lock();
 572                 err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
 573                 if (err == 0) {
 574                         struct fib_table *tb;
 575
 576                         if (cmd == SIOCDELRT) {
 577                                 tb = fib_get_table(net, cfg.fc_table);
 578                                 if (tb)
 579                                         err = fib_table_delete(tb, &cfg);
 580                                 else
 581                                         err = -ESRCH;
 582                         } else {
 583                                 tb = fib_new_table(net, cfg.fc_table);
 584                                 if (tb)
 585                                         err = fib_table_insert(tb, &cfg);
 586                                 else
 587                                         err = -ENOBUFS;
 588                         }
 589
 590                         /* allocated by rtentry_to_fib_config() */
 591                         kfree(cfg.fc_mx);
 592                 }
 593                 rtnl_unlock();
 594                 return err;
 595         }
 596         return -EINVAL;
 597 }
 598
 599 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 600         [RTA_DST]               = { .type = NLA_U32 },
 601         [RTA_SRC]               = { .type = NLA_U32 },
 602         [RTA_IIF]               = { .type = NLA_U32 },
 603         [RTA_OIF]               = { .type = NLA_U32 },
 604         [RTA_GATEWAY]           = { .type = NLA_U32 },
 605         [RTA_PRIORITY]          = { .type = NLA_U32 },
 606         [RTA_PREFSRC]           = { .type = NLA_U32 },
 607         [RTA_METRICS]           = { .type = NLA_NESTED },
 608         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 609         [RTA_FLOW]              = { .type = NLA_U32 },
 610         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
 611         [RTA_ENCAP]             = { .type = NLA_NESTED },
 612 };
 613
 614 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 615                              struct nlmsghdr *nlh, struct fib_config *cfg)
 616 {
 617         struct nlattr *attr;
 618         int err, remaining;
 619         struct rtmsg *rtm;
 620
 621         err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
 622         if (err < 0)
 623                 goto errout;
 624
 625         memset(cfg, 0, sizeof(*cfg));
 626
 627         rtm = nlmsg_data(nlh);
 628         cfg->fc_dst_len = rtm->rtm_dst_len;
 629         cfg->fc_tos = rtm->rtm_tos;
 630         cfg->fc_table = rtm->rtm_table;
 631         cfg->fc_protocol = rtm->rtm_protocol;
 632         cfg->fc_scope = rtm->rtm_scope;
 633         cfg->fc_type = rtm->rtm_type;
 634         cfg->fc_flags = rtm->rtm_flags;
 635         cfg->fc_nlflags = nlh->nlmsg_flags;
 636
 637         cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
 638         cfg->fc_nlinfo.nlh = nlh;
 639         cfg->fc_nlinfo.nl_net = net;
 640
 641         if (cfg->fc_type > RTN_MAX) {
 642                 err = -EINVAL;
 643                 goto errout;
 644         }
 645
 646         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 647                 switch (nla_type(attr)) {
 648                 case RTA_DST:
 649                         cfg->fc_dst = nla_get_be32(attr);
 650                         break;
 651                 case RTA_OIF:
 652                         cfg->fc_oif = nla_get_u32(attr);
 653                         break;
 654                 case RTA_GATEWAY:
 655                         cfg->fc_gw = nla_get_be32(attr);
 656                         break;
 657                 case RTA_PRIORITY:
 658                         cfg->fc_priority = nla_get_u32(attr);
 659                         break;
 660                 case RTA_PREFSRC:
 661                         cfg->fc_prefsrc = nla_get_be32(attr);
 662                         break;
 663                 case RTA_METRICS:
 664                         cfg->fc_mx = nla_data(attr);
 665                         cfg->fc_mx_len = nla_len(attr);
 666                         break;
 667                 case RTA_MULTIPATH:
 668                         cfg->fc_mp = nla_data(attr);
 669                         cfg->fc_mp_len = nla_len(attr);
 670                         break;
 671                 case RTA_FLOW:
 672                         cfg->fc_flow = nla_get_u32(attr);
 673                         break;
 674                 case RTA_TABLE:
 675                         cfg->fc_table = nla_get_u32(attr);
 676                         break;
 677                 case RTA_ENCAP:
 678                         cfg->fc_encap = attr;
 679                         break;
 680                 case RTA_ENCAP_TYPE:
 681                         cfg->fc_encap_type = nla_get_u16(attr);
 682                         break;
 683                 }
 684         }
 685
 686         return 0;
 687 errout:
 688         return err;
 689 }
 690
 691 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
 692 {
 693         struct net *net = sock_net(skb->sk);
 694         struct fib_config cfg;
 695         struct fib_table *tb;
 696         int err;
 697
 698         err = rtm_to_fib_config(net, skb, nlh, &cfg);
 699         if (err < 0)
 700                 goto errout;
 701
 702         tb = fib_get_table(net, cfg.fc_table);
 703         if (!tb) {
 704                 err = -ESRCH;
 705                 goto errout;
 706         }
 707
 708         err = fib_table_delete(tb, &cfg);
 709 errout:
 710         return err;
 711 }
 712
 713 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
 714 {
 715         struct net *net = sock_net(skb->sk);
 716         struct fib_config cfg;
 717         struct fib_table *tb;
 718         int err;
 719
 720         err = rtm_to_fib_config(net, skb, nlh, &cfg);
 721         if (err < 0)
 722                 goto errout;
 723
 724         tb = fib_new_table(net, cfg.fc_table);
 725         if (!tb) {
 726                 err = -ENOBUFS;
 727                 goto errout;
 728         }
 729
 730         err = fib_table_insert(tb, &cfg);
 731 errout:
 732         return err;
 733 }
 734
 735 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 736 {
 737         struct net *net = sock_net(skb->sk);
 738         unsigned int h, s_h;
 739         unsigned int e = 0, s_e;
 740         struct fib_table *tb;
 741         struct hlist_head *head;
 742         int dumped = 0;
 743
 744         if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
 745             ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
 746                 return skb->len;
 747
 748         s_h = cb->args[0];
 749         s_e = cb->args[1];
 750
 751         rcu_read_lock();
 752
 753         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
 754                 e = 0;
 755                 head = &net->ipv4.fib_table_hash[h];
 756                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
 757                         if (e < s_e)
 758                                 goto next;
 759                         if (dumped)
 760                                 memset(&cb->args[2], 0, sizeof(cb->args) -
 761                                                  2 * sizeof(cb->args[0]));
 762                         if (fib_table_dump(tb, skb, cb) < 0)
 763                                 goto out;
 764                         dumped = 1;
 765 next:
 766                         e++;
 767                 }
 768         }
 769 out:
 770         rcu_read_unlock();
 771
 772         cb->args[1] = e;
 773         cb->args[0] = h;
 774
 775         return skb->len;
 776 }
 777
 778 /* Prepare and feed intra-kernel routing request.
 779  * Really, it should be netlink message, but :-( netlink
 780  * can be not configured, so that we feed it directly
 781  * to fib engine. It is legal, because all events occur
 782  * only when netlink is already locked.
 783  */
 784 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
 785 {
 786         struct net *net = dev_net(ifa->ifa_dev->dev);
 787         struct fib_table *tb;
 788         struct fib_config cfg = {
 789                 .fc_protocol = RTPROT_KERNEL,
 790                 .fc_type = type,
 791                 .fc_dst = dst,
 792                 .fc_dst_len = dst_len,
 793                 .fc_prefsrc = ifa->ifa_local,
 794                 .fc_oif = ifa->ifa_dev->dev->ifindex,
 795                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
 796                 .fc_nlinfo = {
 797                         .nl_net = net,
 798                 },
 799         };
 800
 801         if (type == RTN_UNICAST)
 802                 tb = fib_new_table(net, RT_TABLE_MAIN);
 803         else
 804                 tb = fib_new_table(net, RT_TABLE_LOCAL);
 805
 806         if (!tb)
 807                 return;
 808
 809         cfg.fc_table = tb->tb_id;
 810
 811         if (type != RTN_LOCAL)
 812                 cfg.fc_scope = RT_SCOPE_LINK;
 813         else
 814                 cfg.fc_scope = RT_SCOPE_HOST;
 815
 816         if (cmd == RTM_NEWROUTE)
 817                 fib_table_insert(tb, &cfg);
 818         else
 819                 fib_table_delete(tb, &cfg);
 820 }
 821
 822 void fib_add_ifaddr(struct in_ifaddr *ifa)
 823 {
 824         struct in_device *in_dev = ifa->ifa_dev;
 825         struct net_device *dev = in_dev->dev;
 826         struct in_ifaddr *prim = ifa;
 827         __be32 mask = ifa->ifa_mask;
 828         __be32 addr = ifa->ifa_local;
 829         __be32 prefix = ifa->ifa_address & mask;
 830
 831         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 832                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
 833                 if (!prim) {
 834                         pr_warn("%s: bug: prim == NULL\n", __func__);
 835                         return;
 836                 }
 837         }
 838
 839         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
 840
 841         if (!(dev->flags & IFF_UP))
 842                 return;
 843
 844         /* Add broadcast address, if it is explicitly assigned. */
 845         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
 846                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
 847
 848         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
 849             (prefix != addr || ifa->ifa_prefixlen < 32)) {
 850                 fib_magic(RTM_NEWROUTE,
 851                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 852                           prefix, ifa->ifa_prefixlen, prim);
 853
 854                 /* Add network specific broadcasts, when it takes a sense */
 855                 if (ifa->ifa_prefixlen < 31) {
 856                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
 857                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
 858                                   32, prim);
 859                 }
 860         }
 861 }
 862
 863 /* Delete primary or secondary address.
 864  * Optionally, on secondary address promotion consider the addresses
 865  * from subnet iprim as deleted, even if they are in device list.
 866  * In this case the secondary ifa can be in device list.
 867  */
 868 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
 869 {
 870         struct in_device *in_dev = ifa->ifa_dev;
 871         struct net_device *dev = in_dev->dev;
 872         struct in_ifaddr *ifa1;
 873         struct in_ifaddr *prim = ifa, *prim1 = NULL;
 874         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
 875         __be32 any = ifa->ifa_address & ifa->ifa_mask;
 876 #define LOCAL_OK        1
 877 #define BRD_OK          2
 878 #define BRD0_OK         4
 879 #define BRD1_OK         8
 880         unsigned int ok = 0;
 881         int subnet = 0;         /* Primary network */
 882         int gone = 1;           /* Address is missing */
 883         int same_prefsrc = 0;   /* Another primary with same IP */
 884
 885         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 886                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
 887                 if (!prim) {
 888                         pr_warn("%s: bug: prim == NULL\n", __func__);
 889                         return;
 890                 }
 891                 if (iprim && iprim != prim) {
 892                         pr_warn("%s: bug: iprim != prim\n", __func__);
 893                         return;
 894                 }
 895         } else if (!ipv4_is_zeronet(any) &&
 896                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
 897                 fib_magic(RTM_DELROUTE,
 898                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 899                           any, ifa->ifa_prefixlen, prim);
 900                 subnet = 1;
 901         }
 902
 903         /* Deletion is more complicated than add.
 904          * We should take care of not to delete too much :-)
 905          *
 906          * Scan address list to be sure that addresses are really gone.
 907          */
 908
 909         for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
 910                 if (ifa1 == ifa) {
 911                         /* promotion, keep the IP */
 912                         gone = 0;
 913                         continue;
 914                 }
 915                 /* Ignore IFAs from our subnet */
 916                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
 917                     inet_ifa_match(ifa1->ifa_address, iprim))
 918                         continue;
 919
 920                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
 921                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
 922                         /* Another address from our subnet? */
 923                         if (ifa1->ifa_mask == prim->ifa_mask &&
 924                             inet_ifa_match(ifa1->ifa_address, prim))
 925                                 prim1 = prim;
 926                         else {
 927                                 /* We reached the secondaries, so
 928                                  * same_prefsrc should be determined.
 929                                  */
 930                                 if (!same_prefsrc)
 931                                         continue;
 932                                 /* Search new prim1 if ifa1 is not
 933                                  * using the current prim1
 934                                  */
 935                                 if (!prim1 ||
 936                                     ifa1->ifa_mask != prim1->ifa_mask ||
 937                                     !inet_ifa_match(ifa1->ifa_address, prim1))
 938                                         prim1 = inet_ifa_byprefix(in_dev,
 939                                                         ifa1->ifa_address,
 940                                                         ifa1->ifa_mask);
 941                                 if (!prim1)
 942                                         continue;
 943                                 if (prim1->ifa_local != prim->ifa_local)
 944                                         continue;
 945                         }
 946                 } else {
 947                         if (prim->ifa_local != ifa1->ifa_local)
 948                                 continue;
 949                         prim1 = ifa1;
 950                         if (prim != prim1)
 951                                 same_prefsrc = 1;
 952                 }
 953                 if (ifa->ifa_local == ifa1->ifa_local)
 954                         ok |= LOCAL_OK;
 955                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
 956                         ok |= BRD_OK;
 957                 if (brd == ifa1->ifa_broadcast)
 958                         ok |= BRD1_OK;
 959                 if (any == ifa1->ifa_broadcast)
 960                         ok |= BRD0_OK;
 961                 /* primary has network specific broadcasts */
 962                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
 963                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
 964                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
 965
 966                         if (!ipv4_is_zeronet(any1)) {
 967                                 if (ifa->ifa_broadcast == brd1 ||
 968                                     ifa->ifa_broadcast == any1)
 969                                         ok |= BRD_OK;
 970                                 if (brd == brd1 || brd == any1)
 971                                         ok |= BRD1_OK;
 972                                 if (any == brd1 || any == any1)
 973                                         ok |= BRD0_OK;
 974                         }
 975                 }
 976         }
 977
 978         if (!(ok & BRD_OK))
 979                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
 980         if (subnet && ifa->ifa_prefixlen < 31) {
 981                 if (!(ok & BRD1_OK))
 982                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
 983                 if (!(ok & BRD0_OK))
 984                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
 985         }
 986         if (!(ok & LOCAL_OK)) {
 987                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
 988
 989                 /* Check, that this local address finally disappeared. */
 990                 if (gone &&
 991                     inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) {
 992                         /* And the last, but not the least thing.
 993                          * We must flush stray FIB entries.
 994                          *
 995                          * First of all, we scan fib_info list searching
 996                          * for stray nexthop entries, then ignite fib_flush.
 997                          */
 998                         if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
 999                                 fib_flush(dev_net(dev));
1000                 }
1001         }
1002 #undef LOCAL_OK
1003 #undef BRD_OK
1004 #undef BRD0_OK
1005 #undef BRD1_OK
1006 }
1007
1008 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1009 {
1010
1011         struct fib_result       res;
1012         struct flowi4           fl4 = {
1013                 .flowi4_mark = frn->fl_mark,
1014                 .daddr = frn->fl_addr,
1015                 .flowi4_tos = frn->fl_tos,
1016                 .flowi4_scope = frn->fl_scope,
1017         };
1018         struct fib_table *tb;
1019
1020         rcu_read_lock();
1021
1022         tb = fib_get_table(net, frn->tb_id_in);
1023
1024         frn->err = -ENOENT;
1025         if (tb) {
1026                 local_bh_disable();
1027
1028                 frn->tb_id = tb->tb_id;
1029                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1030
1031                 if (!frn->err) {
1032                         frn->prefixlen = res.prefixlen;
1033                         frn->nh_sel = res.nh_sel;
1034                         frn->type = res.type;
1035                         frn->scope = res.scope;
1036                 }
1037                 local_bh_enable();
1038         }
1039
1040         rcu_read_unlock();
1041 }
1042
1043 static void nl_fib_input(struct sk_buff *skb)
1044 {
1045         struct net *net;
1046         struct fib_result_nl *frn;
1047         struct nlmsghdr *nlh;
1048         u32 portid;
1049
1050         net = sock_net(skb->sk);
1051         nlh = nlmsg_hdr(skb);
1052         if (skb->len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len ||
1053             nlmsg_len(nlh) < sizeof(*frn))
1054                 return;
1055
1056         skb = netlink_skb_clone(skb, GFP_KERNEL);
1057         if (!skb)
1058                 return;
1059         nlh = nlmsg_hdr(skb);
1060
1061         frn = (struct fib_result_nl *) nlmsg_data(nlh);
1062         nl_fib_lookup(net, frn);
1063
1064         portid = NETLINK_CB(skb).portid;      /* netlink portid */
1065         NETLINK_CB(skb).portid = 0;        /* from kernel */
1066         NETLINK_CB(skb).dst_group = 0;  /* unicast */
1067         netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1068 }
1069
1070 static int __net_init nl_fib_lookup_init(struct net *net)
1071 {
1072         struct sock *sk;
1073         struct netlink_kernel_cfg cfg = {
1074                 .input  = nl_fib_input,
1075         };
1076
1077         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1078         if (!sk)
1079                 return -EAFNOSUPPORT;
1080         net->ipv4.fibnl = sk;
1081         return 0;
1082 }
1083
1084 static void nl_fib_lookup_exit(struct net *net)
1085 {
1086         netlink_kernel_release(net->ipv4.fibnl);
1087         net->ipv4.fibnl = NULL;
1088 }
1089
1090 static void fib_disable_ip(struct net_device *dev, unsigned long event)
1091 {
1092         if (fib_sync_down_dev(dev, event))
1093                 fib_flush(dev_net(dev));
1094         rt_cache_flush(dev_net(dev));
1095         arp_ifdown(dev);
1096 }
1097
1098 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1099 {
1100         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1101         struct net_device *dev = ifa->ifa_dev->dev;
1102         struct net *net = dev_net(dev);
1103
1104         switch (event) {
1105         case NETDEV_UP:
1106                 fib_add_ifaddr(ifa);
1107 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1108                 fib_sync_up(dev, RTNH_F_DEAD);
1109 #endif
1110                 atomic_inc(&net->ipv4.dev_addr_genid);
1111                 rt_cache_flush(dev_net(dev));
1112                 break;
1113         case NETDEV_DOWN:
1114                 fib_del_ifaddr(ifa, NULL);
1115                 atomic_inc(&net->ipv4.dev_addr_genid);
1116                 if (!ifa->ifa_dev->ifa_list) {
1117                         /* Last address was deleted from this interface.
1118                          * Disable IP.
1119                          */
1120                         fib_disable_ip(dev, event);
1121                 } else {
1122                         rt_cache_flush(dev_net(dev));
1123                 }
1124                 break;
1125         }
1126         return NOTIFY_DONE;
1127 }
1128
1129 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1130 {
1131         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1132         struct in_device *in_dev;
1133         struct net *net = dev_net(dev);
1134         unsigned int flags;
1135
1136         if (event == NETDEV_UNREGISTER) {
1137                 fib_disable_ip(dev, event);
1138                 rt_flush_dev(dev);
1139                 return NOTIFY_DONE;
1140         }
1141
1142         in_dev = __in_dev_get_rtnl(dev);
1143         if (!in_dev)
1144                 return NOTIFY_DONE;
1145
1146         switch (event) {
1147         case NETDEV_UP:
1148                 for_ifa(in_dev) {
1149                         fib_add_ifaddr(ifa);
1150                 } endfor_ifa(in_dev);
1151 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1152                 fib_sync_up(dev, RTNH_F_DEAD);
1153 #endif
1154                 atomic_inc(&net->ipv4.dev_addr_genid);
1155                 rt_cache_flush(net);
1156                 break;
1157         case NETDEV_DOWN:
1158                 fib_disable_ip(dev, event);
1159                 break;
1160         case NETDEV_CHANGE:
1161                 flags = dev_get_flags(dev);
1162                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1163                         fib_sync_up(dev, RTNH_F_LINKDOWN);
1164                 else
1165                         fib_sync_down_dev(dev, event);
1166                 /* fall through */
1167         case NETDEV_CHANGEMTU:
1168                 rt_cache_flush(net);
1169                 break;
1170         }
1171         return NOTIFY_DONE;
1172 }
1173
1174 static struct notifier_block fib_inetaddr_notifier = {
1175         .notifier_call = fib_inetaddr_event,
1176 };
1177
1178 static struct notifier_block fib_netdev_notifier = {
1179         .notifier_call = fib_netdev_event,
1180 };
1181
1182 static int __net_init ip_fib_net_init(struct net *net)
1183 {
1184         int err;
1185         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1186
1187         /* Avoid false sharing : Use at least a full cache line */
1188         size = max_t(size_t, size, L1_CACHE_BYTES);
1189
1190         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1191         if (!net->ipv4.fib_table_hash)
1192                 return -ENOMEM;
1193
1194         err = fib4_rules_init(net);
1195         if (err < 0)
1196                 goto fail;
1197         return 0;
1198
1199 fail:
1200         kfree(net->ipv4.fib_table_hash);
1201         return err;
1202 }
1203
1204 static void ip_fib_net_exit(struct net *net)
1205 {
1206         unsigned int i;
1207
1208         rtnl_lock();
1209 #ifdef CONFIG_IP_MULTIPLE_TABLES
1210         RCU_INIT_POINTER(net->ipv4.fib_local, NULL);
1211         RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1212         RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1213 #endif
1214         for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1215                 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1216                 struct hlist_node *tmp;
1217                 struct fib_table *tb;
1218
1219                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1220                         hlist_del(&tb->tb_hlist);
1221                         fib_table_flush(tb);
1222                         fib_free_table(tb);
1223                 }
1224         }
1225
1226 #ifdef CONFIG_IP_MULTIPLE_TABLES
1227         fib4_rules_exit(net);
1228 #endif
1229         rtnl_unlock();
1230         kfree(net->ipv4.fib_table_hash);
1231 }
1232
1233 static int __net_init fib_net_init(struct net *net)
1234 {
1235         int error;
1236
1237 #ifdef CONFIG_IP_ROUTE_CLASSID
1238         net->ipv4.fib_num_tclassid_users = 0;
1239 #endif
1240         error = ip_fib_net_init(net);
1241         if (error < 0)
1242                 goto out;
1243         error = nl_fib_lookup_init(net);
1244         if (error < 0)
1245                 goto out_nlfl;
1246         error = fib_proc_init(net);
1247         if (error < 0)
1248                 goto out_proc;
1249 out:
1250         return error;
1251
1252 out_proc:
1253         nl_fib_lookup_exit(net);
1254 out_nlfl:
1255         ip_fib_net_exit(net);
1256         goto out;
1257 }
1258
1259 static void __net_exit fib_net_exit(struct net *net)
1260 {
1261         fib_proc_exit(net);
1262         nl_fib_lookup_exit(net);
1263         ip_fib_net_exit(net);
1264 }
1265
1266 static struct pernet_operations fib_net_ops = {
1267         .init = fib_net_init,
1268         .exit = fib_net_exit,
1269 };
1270
1271 void __init ip_fib_init(void)
1272 {
1273         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1274         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1275         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1276
1277         register_pernet_subsys(&fib_net_ops);
1278         register_netdevice_notifier(&fib_netdev_notifier);
1279         register_inetaddr_notifier(&fib_inetaddr_notifier);
1280
1281         fib_trie_init();
1282 }