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 <linux/uaccess.h>
  18 #include <linux/bitops.h>
  19 #include <linux/capability.h>
  20 #include <linux/types.h>
  21 #include <linux/kernel.h>
  22 #include <linux/mm.h>
  23 #include <linux/string.h>
  24 #include <linux/socket.h>
  25 #include <linux/sockios.h>
  26 #include <linux/errno.h>
  27 #include <linux/in.h>
  28 #include <linux/inet.h>
  29 #include <linux/inetdevice.h>
  30 #include <linux/netdevice.h>
  31 #include <linux/if_addr.h>
  32 #include <linux/if_arp.h>
  33 #include <linux/skbuff.h>
  34 #include <linux/cache.h>
  35 #include <linux/init.h>
  36 #include <linux/list.h>
  37 #include <linux/slab.h>
  38
  39 #include <net/ip.h>
  40 #include <net/protocol.h>
  41 #include <net/route.h>
  42 #include <net/tcp.h>
  43 #include <net/sock.h>
  44 #include <net/arp.h>
  45 #include <net/ip_fib.h>
  46 #include <net/rtnetlink.h>
  47 #include <net/xfrm.h>
  48 #include <net/l3mdev.h>
  49 #include <net/lwtunnel.h>
  50 #include <trace/events/fib.h>
  51
  52 #ifndef CONFIG_IP_MULTIPLE_TABLES
  53
  54 static int __net_init fib4_rules_init(struct net *net)
  55 {
  56         struct fib_table *local_table, *main_table;
  57
  58         main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
  59         if (!main_table)
  60                 return -ENOMEM;
  61
  62         local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
  63         if (!local_table)
  64                 goto fail;
  65
  66         hlist_add_head_rcu(&local_table->tb_hlist,
  67                                 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
  68         hlist_add_head_rcu(&main_table->tb_hlist,
  69                                 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
  70         return 0;
  71
  72 fail:
  73         fib_free_table(main_table);
  74         return -ENOMEM;
  75 }
  76 #else
  77
  78 struct fib_table *fib_new_table(struct net *net, u32 id)
  79 {
  80         struct fib_table *tb, *alias = NULL;
  81         unsigned int h;
  82
  83         if (id == 0)
  84                 id = RT_TABLE_MAIN;
  85         tb = fib_get_table(net, id);
  86         if (tb)
  87                 return tb;
  88
  89         if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
  90                 alias = fib_new_table(net, RT_TABLE_MAIN);
  91
  92         tb = fib_trie_table(id, alias);
  93         if (!tb)
  94                 return NULL;
  95
  96         switch (id) {
  97         case RT_TABLE_MAIN:
  98                 rcu_assign_pointer(net->ipv4.fib_main, tb);
  99                 break;
 100         case RT_TABLE_DEFAULT:
 101                 rcu_assign_pointer(net->ipv4.fib_default, tb);
 102                 break;
 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 EXPORT_SYMBOL_GPL(fib_new_table);
 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_MAIN:
 139                 rcu_assign_pointer(net->ipv4.fib_main, new);
 140                 break;
 141         case RT_TABLE_DEFAULT:
 142                 rcu_assign_pointer(net->ipv4.fib_default, new);
 143                 break;
 144         default:
 145                 break;
 146         }
 147
 148 #endif
 149         /* replace the old table in the hlist */
 150         hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
 151 }
 152
 153 int fib_unmerge(struct net *net)
 154 {
 155         struct fib_table *old, *new, *main_table;
 156
 157         /* attempt to fetch local table if it has been allocated */
 158         old = fib_get_table(net, RT_TABLE_LOCAL);
 159         if (!old)
 160                 return 0;
 161
 162         new = fib_trie_unmerge(old);
 163         if (!new)
 164                 return -ENOMEM;
 165
 166         /* table is already unmerged */
 167         if (new == old)
 168                 return 0;
 169
 170         /* replace merged table with clean table */
 171         fib_replace_table(net, old, new);
 172         fib_free_table(old);
 173
 174         /* attempt to fetch main table if it has been allocated */
 175         main_table = fib_get_table(net, RT_TABLE_MAIN);
 176         if (!main_table)
 177                 return 0;
 178
 179         /* flush local entries from main table */
 180         fib_table_flush_external(main_table);
 181
 182         return 0;
 183 }
 184
 185 static void fib_flush(struct net *net)
 186 {
 187         int flushed = 0;
 188         unsigned int h;
 189
 190         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
 191                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
 192                 struct hlist_node *tmp;
 193                 struct fib_table *tb;
 194
 195                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
 196                         flushed += fib_table_flush(net, tb);
 197         }
 198
 199         if (flushed)
 200                 rt_cache_flush(net);
 201 }
 202
 203 /*
 204  * Find address type as if only "dev" was present in the system. If
 205  * on_dev is NULL then all interfaces are taken into consideration.
 206  */
 207 static inline unsigned int __inet_dev_addr_type(struct net *net,
 208                                                 const struct net_device *dev,
 209                                                 __be32 addr, u32 tb_id)
 210 {
 211         struct flowi4           fl4 = { .daddr = addr };
 212         struct fib_result       res;
 213         unsigned int ret = RTN_BROADCAST;
 214         struct fib_table *table;
 215
 216         if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
 217                 return RTN_BROADCAST;
 218         if (ipv4_is_multicast(addr))
 219                 return RTN_MULTICAST;
 220
 221         rcu_read_lock();
 222
 223         table = fib_get_table(net, tb_id);
 224         if (table) {
 225                 ret = RTN_UNICAST;
 226                 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
 227                         if (!dev || dev == res.fi->fib_dev)
 228                                 ret = res.type;
 229                 }
 230         }
 231
 232         rcu_read_unlock();
 233         return ret;
 234 }
 235
 236 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
 237 {
 238         return __inet_dev_addr_type(net, NULL, addr, tb_id);
 239 }
 240 EXPORT_SYMBOL(inet_addr_type_table);
 241
 242 unsigned int inet_addr_type(struct net *net, __be32 addr)
 243 {
 244         return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
 245 }
 246 EXPORT_SYMBOL(inet_addr_type);
 247
 248 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
 249                                 __be32 addr)
 250 {
 251         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
 252
 253         return __inet_dev_addr_type(net, dev, addr, rt_table);
 254 }
 255 EXPORT_SYMBOL(inet_dev_addr_type);
 256
 257 /* inet_addr_type with dev == NULL but using the table from a dev
 258  * if one is associated
 259  */
 260 unsigned int inet_addr_type_dev_table(struct net *net,
 261                                       const struct net_device *dev,
 262                                       __be32 addr)
 263 {
 264         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
 265
 266         return __inet_dev_addr_type(net, NULL, addr, rt_table);
 267 }
 268 EXPORT_SYMBOL(inet_addr_type_dev_table);
 269
 270 __be32 fib_compute_spec_dst(struct sk_buff *skb)
 271 {
 272         struct net_device *dev = skb->dev;
 273         struct in_device *in_dev;
 274         struct fib_result res;
 275         struct rtable *rt;
 276         struct net *net;
 277         int scope;
 278
 279         rt = skb_rtable(skb);
 280         if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
 281             RTCF_LOCAL)
 282                 return ip_hdr(skb)->daddr;
 283
 284         in_dev = __in_dev_get_rcu(dev);
 285         BUG_ON(!in_dev);
 286
 287         net = dev_net(dev);
 288
 289         scope = RT_SCOPE_UNIVERSE;
 290         if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
 291                 struct flowi4 fl4 = {
 292                         .flowi4_iif = LOOPBACK_IFINDEX,
 293                         .daddr = ip_hdr(skb)->saddr,
 294                         .flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
 295                         .flowi4_scope = scope,
 296                         .flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0,
 297                 };
 298                 if (!fib_lookup(net, &fl4, &res, 0))
 299                         return FIB_RES_PREFSRC(net, res);
 300         } else {
 301                 scope = RT_SCOPE_LINK;
 302         }
 303
 304         return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
 305 }
 306
 307 /* Given (packet source, input interface) and optional (dst, oif, tos):
 308  * - (main) check, that source is valid i.e. not broadcast or our local
 309  *   address.
 310  * - figure out what "logical" interface this packet arrived
 311  *   and calculate "specific destination" address.
 312  * - check, that packet arrived from expected physical interface.
 313  * called with rcu_read_lock()
 314  */
 315 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 316                                  u8 tos, int oif, struct net_device *dev,
 317                                  int rpf, struct in_device *idev, u32 *itag)
 318 {
 319         int ret, no_addr;
 320         struct fib_result res;
 321         struct flowi4 fl4;
 322         struct net *net = dev_net(dev);
 323         bool dev_match;
 324
 325         fl4.flowi4_oif = 0;
 326         fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
 327         if (!fl4.flowi4_iif)
 328                 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
 329         fl4.daddr = src;
 330         fl4.saddr = dst;
 331         fl4.flowi4_tos = tos;
 332         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 333         fl4.flowi4_tun_key.tun_id = 0;
 334         fl4.flowi4_flags = 0;
 335         fl4.flowi4_uid = sock_net_uid(net, NULL);
 336
 337         no_addr = idev->ifa_list == NULL;
 338
 339         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
 340
 341         trace_fib_validate_source(dev, &fl4);
 342
 343         if (fib_lookup(net, &fl4, &res, 0))
 344                 goto last_resort;
 345         if (res.type != RTN_UNICAST &&
 346             (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
 347                 goto e_inval;
 348         if (!rpf && !fib_num_tclassid_users(net) &&
 349             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev)))
 350                 goto last_resort;
 351         fib_combine_itag(itag, &res);
 352         dev_match = false;
 353
 354 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 355         for (ret = 0; ret < res.fi->fib_nhs; ret++) {
 356                 struct fib_nh *nh = &res.fi->fib_nh[ret];
 357
 358                 if (nh->nh_dev == dev) {
 359                         dev_match = true;
 360                         break;
 361                 } else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) {
 362                         dev_match = true;
 363                         break;
 364                 }
 365         }
 366 #else
 367         if (FIB_RES_DEV(res) == dev)
 368                 dev_match = true;
 369 #endif
 370         if (dev_match) {
 371                 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 372                 return ret;
 373         }
 374         if (no_addr)
 375                 goto last_resort;
 376         if (rpf == 1)
 377                 goto e_rpf;
 378         fl4.flowi4_oif = dev->ifindex;
 379
 380         ret = 0;
 381         if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
 382                 if (res.type == RTN_UNICAST)
 383                         ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 384         }
 385         return ret;
 386
 387 last_resort:
 388         if (rpf)
 389                 goto e_rpf;
 390         *itag = 0;
 391         return 0;
 392
 393 e_inval:
 394         return -EINVAL;
 395 e_rpf:
 396         return -EXDEV;
 397 }
 398
 399 /* Ignore rp_filter for packets protected by IPsec. */
 400 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 401                         u8 tos, int oif, struct net_device *dev,
 402                         struct in_device *idev, u32 *itag)
 403 {
 404         int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
 405
 406         if (!r && !fib_num_tclassid_users(dev_net(dev)) &&
 407             IN_DEV_ACCEPT_LOCAL(idev) &&
 408             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
 409                 *itag = 0;
 410                 return 0;
 411         }
 412         return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
 413 }
 414
 415 static inline __be32 sk_extract_addr(struct sockaddr *addr)
 416 {
 417         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 418 }
 419
 420 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
 421 {
 422         struct nlattr *nla;
 423
 424         nla = (struct nlattr *) ((char *) mx + len);
 425         nla->nla_type = type;
 426         nla->nla_len = nla_attr_size(4);
 427         *(u32 *) nla_data(nla) = value;
 428
 429         return len + nla_total_size(4);
 430 }
 431
 432 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
 433                                  struct fib_config *cfg)
 434 {
 435         __be32 addr;
 436         int plen;
 437
 438         memset(cfg, 0, sizeof(*cfg));
 439         cfg->fc_nlinfo.nl_net = net;
 440
 441         if (rt->rt_dst.sa_family != AF_INET)
 442                 return -EAFNOSUPPORT;
 443
 444         /*
 445          * Check mask for validity:
 446          * a) it must be contiguous.
 447          * b) destination must have all host bits clear.
 448          * c) if application forgot to set correct family (AF_INET),
 449          *    reject request unless it is absolutely clear i.e.
 450          *    both family and mask are zero.
 451          */
 452         plen = 32;
 453         addr = sk_extract_addr(&rt->rt_dst);
 454         if (!(rt->rt_flags & RTF_HOST)) {
 455                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
 456
 457                 if (rt->rt_genmask.sa_family != AF_INET) {
 458                         if (mask || rt->rt_genmask.sa_family)
 459                                 return -EAFNOSUPPORT;
 460                 }
 461
 462                 if (bad_mask(mask, addr))
 463                         return -EINVAL;
 464
 465                 plen = inet_mask_len(mask);
 466         }
 467
 468         cfg->fc_dst_len = plen;
 469         cfg->fc_dst = addr;
 470
 471         if (cmd != SIOCDELRT) {
 472                 cfg->fc_nlflags = NLM_F_CREATE;
 473                 cfg->fc_protocol = RTPROT_BOOT;
 474         }
 475
 476         if (rt->rt_metric)
 477                 cfg->fc_priority = rt->rt_metric - 1;
 478
 479         if (rt->rt_flags & RTF_REJECT) {
 480                 cfg->fc_scope = RT_SCOPE_HOST;
 481                 cfg->fc_type = RTN_UNREACHABLE;
 482                 return 0;
 483         }
 484
 485         cfg->fc_scope = RT_SCOPE_NOWHERE;
 486         cfg->fc_type = RTN_UNICAST;
 487
 488         if (rt->rt_dev) {
 489                 char *colon;
 490                 struct net_device *dev;
 491                 char devname[IFNAMSIZ];
 492
 493                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
 494                         return -EFAULT;
 495
 496                 devname[IFNAMSIZ-1] = 0;
 497                 colon = strchr(devname, ':');
 498                 if (colon)
 499                         *colon = 0;
 500                 dev = __dev_get_by_name(net, devname);
 501                 if (!dev)
 502                         return -ENODEV;
 503                 cfg->fc_oif = dev->ifindex;
 504                 cfg->fc_table = l3mdev_fib_table(dev);
 505                 if (colon) {
 506                         struct in_ifaddr *ifa;
 507                         struct in_device *in_dev = __in_dev_get_rtnl(dev);
 508                         if (!in_dev)
 509                                 return -ENODEV;
 510                         *colon = ':';
 511                         for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
 512                                 if (strcmp(ifa->ifa_label, devname) == 0)
 513                                         break;
 514                         if (!ifa)
 515                                 return -ENODEV;
 516                         cfg->fc_prefsrc = ifa->ifa_local;
 517                 }
 518         }
 519
 520         addr = sk_extract_addr(&rt->rt_gateway);
 521         if (rt->rt_gateway.sa_family == AF_INET && addr) {
 522                 unsigned int addr_type;
 523
 524                 cfg->fc_gw = addr;
 525                 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
 526                 if (rt->rt_flags & RTF_GATEWAY &&
 527                     addr_type == RTN_UNICAST)
 528                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
 529         }
 530
 531         if (cmd == SIOCDELRT)
 532                 return 0;
 533
 534         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
 535                 return -EINVAL;
 536
 537         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 538                 cfg->fc_scope = RT_SCOPE_LINK;
 539
 540         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 541                 struct nlattr *mx;
 542                 int len = 0;
 543
 544                 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
 545                 if (!mx)
 546                         return -ENOMEM;
 547
 548                 if (rt->rt_flags & RTF_MTU)
 549                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 550
 551                 if (rt->rt_flags & RTF_WINDOW)
 552                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 553
 554                 if (rt->rt_flags & RTF_IRTT)
 555                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 556
 557                 cfg->fc_mx = mx;
 558                 cfg->fc_mx_len = len;
 559         }
 560
 561         return 0;
 562 }
 563
 564 /*
 565  * Handle IP routing ioctl calls.
 566  * These are used to manipulate the routing tables
 567  */
 568 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
 569 {
 570         struct fib_config cfg;
 571         struct rtentry rt;
 572         int err;
 573
 574         switch (cmd) {
 575         case SIOCADDRT:         /* Add a route */
 576         case SIOCDELRT:         /* Delete a route */
 577                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 578                         return -EPERM;
 579
 580                 if (copy_from_user(&rt, arg, sizeof(rt)))
 581                         return -EFAULT;
 582
 583                 rtnl_lock();
 584                 err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
 585                 if (err == 0) {
 586                         struct fib_table *tb;
 587
 588                         if (cmd == SIOCDELRT) {
 589                                 tb = fib_get_table(net, cfg.fc_table);
 590                                 if (tb)
 591                                         err = fib_table_delete(net, tb, &cfg,
 592                                                                NULL);
 593                                 else
 594                                         err = -ESRCH;
 595                         } else {
 596                                 tb = fib_new_table(net, cfg.fc_table);
 597                                 if (tb)
 598                                         err = fib_table_insert(net, tb,
 599                                                                &cfg, NULL);
 600                                 else
 601                                         err = -ENOBUFS;
 602                         }
 603
 604                         /* allocated by rtentry_to_fib_config() */
 605                         kfree(cfg.fc_mx);
 606                 }
 607                 rtnl_unlock();
 608                 return err;
 609         }
 610         return -EINVAL;
 611 }
 612
 613 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 614         [RTA_DST]               = { .type = NLA_U32 },
 615         [RTA_SRC]               = { .type = NLA_U32 },
 616         [RTA_IIF]               = { .type = NLA_U32 },
 617         [RTA_OIF]               = { .type = NLA_U32 },
 618         [RTA_GATEWAY]           = { .type = NLA_U32 },
 619         [RTA_PRIORITY]          = { .type = NLA_U32 },
 620         [RTA_PREFSRC]           = { .type = NLA_U32 },
 621         [RTA_METRICS]           = { .type = NLA_NESTED },
 622         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 623         [RTA_FLOW]              = { .type = NLA_U32 },
 624         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
 625         [RTA_ENCAP]             = { .type = NLA_NESTED },
 626         [RTA_UID]               = { .type = NLA_U32 },
 627         [RTA_MARK]              = { .type = NLA_U32 },
 628 };
 629
 630 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 631                              struct nlmsghdr *nlh, struct fib_config *cfg,
 632                              struct netlink_ext_ack *extack)
 633 {
 634         struct nlattr *attr;
 635         int err, remaining;
 636         struct rtmsg *rtm;
 637
 638         err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy,
 639                              extack);
 640         if (err < 0)
 641                 goto errout;
 642
 643         memset(cfg, 0, sizeof(*cfg));
 644
 645         rtm = nlmsg_data(nlh);
 646         cfg->fc_dst_len = rtm->rtm_dst_len;
 647         cfg->fc_tos = rtm->rtm_tos;
 648         cfg->fc_table = rtm->rtm_table;
 649         cfg->fc_protocol = rtm->rtm_protocol;
 650         cfg->fc_scope = rtm->rtm_scope;
 651         cfg->fc_type = rtm->rtm_type;
 652         cfg->fc_flags = rtm->rtm_flags;
 653         cfg->fc_nlflags = nlh->nlmsg_flags;
 654
 655         cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
 656         cfg->fc_nlinfo.nlh = nlh;
 657         cfg->fc_nlinfo.nl_net = net;
 658
 659         if (cfg->fc_type > RTN_MAX) {
 660                 NL_SET_ERR_MSG(extack, "Invalid route type");
 661                 err = -EINVAL;
 662                 goto errout;
 663         }
 664
 665         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 666                 switch (nla_type(attr)) {
 667                 case RTA_DST:
 668                         cfg->fc_dst = nla_get_be32(attr);
 669                         break;
 670                 case RTA_OIF:
 671                         cfg->fc_oif = nla_get_u32(attr);
 672                         break;
 673                 case RTA_GATEWAY:
 674                         cfg->fc_gw = nla_get_be32(attr);
 675                         break;
 676                 case RTA_PRIORITY:
 677                         cfg->fc_priority = nla_get_u32(attr);
 678                         break;
 679                 case RTA_PREFSRC:
 680                         cfg->fc_prefsrc = nla_get_be32(attr);
 681                         break;
 682                 case RTA_METRICS:
 683                         cfg->fc_mx = nla_data(attr);
 684                         cfg->fc_mx_len = nla_len(attr);
 685                         break;
 686                 case RTA_MULTIPATH:
 687                         err = lwtunnel_valid_encap_type_attr(nla_data(attr),
 688                                                              nla_len(attr),
 689                                                              extack);
 690                         if (err < 0)
 691                                 goto errout;
 692                         cfg->fc_mp = nla_data(attr);
 693                         cfg->fc_mp_len = nla_len(attr);
 694                         break;
 695                 case RTA_FLOW:
 696                         cfg->fc_flow = nla_get_u32(attr);
 697                         break;
 698                 case RTA_TABLE:
 699                         cfg->fc_table = nla_get_u32(attr);
 700                         break;
 701                 case RTA_ENCAP:
 702                         cfg->fc_encap = attr;
 703                         break;
 704                 case RTA_ENCAP_TYPE:
 705                         cfg->fc_encap_type = nla_get_u16(attr);
 706                         err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
 707                                                         extack);
 708                         if (err < 0)
 709                                 goto errout;
 710                         break;
 711                 }
 712         }
 713
 714         return 0;
 715 errout:
 716         return err;
 717 }
 718
 719 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 720                              struct netlink_ext_ack *extack)
 721 {
 722         struct net *net = sock_net(skb->sk);
 723         struct fib_config cfg;
 724         struct fib_table *tb;
 725         int err;
 726
 727         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 728         if (err < 0)
 729                 goto errout;
 730
 731         tb = fib_get_table(net, cfg.fc_table);
 732         if (!tb) {
 733                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
 734                 err = -ESRCH;
 735                 goto errout;
 736         }
 737
 738         err = fib_table_delete(net, tb, &cfg, extack);
 739 errout:
 740         return err;
 741 }
 742
 743 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 744                              struct netlink_ext_ack *extack)
 745 {
 746         struct net *net = sock_net(skb->sk);
 747         struct fib_config cfg;
 748         struct fib_table *tb;
 749         int err;
 750
 751         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 752         if (err < 0)
 753                 goto errout;
 754
 755         tb = fib_new_table(net, cfg.fc_table);
 756         if (!tb) {
 757                 err = -ENOBUFS;
 758                 goto errout;
 759         }
 760
 761         err = fib_table_insert(net, tb, &cfg, extack);
 762 errout:
 763         return err;
 764 }
 765
 766 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 767 {
 768         struct net *net = sock_net(skb->sk);
 769         unsigned int h, s_h;
 770         unsigned int e = 0, s_e;
 771         struct fib_table *tb;
 772         struct hlist_head *head;
 773         int dumped = 0, err;
 774
 775         if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
 776             ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
 777                 return skb->len;
 778
 779         s_h = cb->args[0];
 780         s_e = cb->args[1];
 781
 782         rcu_read_lock();
 783
 784         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
 785                 e = 0;
 786                 head = &net->ipv4.fib_table_hash[h];
 787                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
 788                         if (e < s_e)
 789                                 goto next;
 790                         if (dumped)
 791                                 memset(&cb->args[2], 0, sizeof(cb->args) -
 792                                                  2 * sizeof(cb->args[0]));
 793                         err = fib_table_dump(tb, skb, cb);
 794                         if (err < 0) {
 795                                 if (likely(skb->len))
 796                                         goto out;
 797
 798                                 goto out_err;
 799                         }
 800                         dumped = 1;
 801 next:
 802                         e++;
 803                 }
 804         }
 805 out:
 806         err = skb->len;
 807 out_err:
 808         rcu_read_unlock();
 809
 810         cb->args[1] = e;
 811         cb->args[0] = h;
 812
 813         return err;
 814 }
 815
 816 /* Prepare and feed intra-kernel routing request.
 817  * Really, it should be netlink message, but :-( netlink
 818  * can be not configured, so that we feed it directly
 819  * to fib engine. It is legal, because all events occur
 820  * only when netlink is already locked.
 821  */
 822 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
 823 {
 824         struct net *net = dev_net(ifa->ifa_dev->dev);
 825         u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
 826         struct fib_table *tb;
 827         struct fib_config cfg = {
 828                 .fc_protocol = RTPROT_KERNEL,
 829                 .fc_type = type,
 830                 .fc_dst = dst,
 831                 .fc_dst_len = dst_len,
 832                 .fc_prefsrc = ifa->ifa_local,
 833                 .fc_oif = ifa->ifa_dev->dev->ifindex,
 834                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
 835                 .fc_nlinfo = {
 836                         .nl_net = net,
 837                 },
 838         };
 839
 840         if (!tb_id)
 841                 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
 842
 843         tb = fib_new_table(net, tb_id);
 844         if (!tb)
 845                 return;
 846
 847         cfg.fc_table = tb->tb_id;
 848
 849         if (type != RTN_LOCAL)
 850                 cfg.fc_scope = RT_SCOPE_LINK;
 851         else
 852                 cfg.fc_scope = RT_SCOPE_HOST;
 853
 854         if (cmd == RTM_NEWROUTE)
 855                 fib_table_insert(net, tb, &cfg, NULL);
 856         else
 857                 fib_table_delete(net, tb, &cfg, NULL);
 858 }
 859
 860 void fib_add_ifaddr(struct in_ifaddr *ifa)
 861 {
 862         struct in_device *in_dev = ifa->ifa_dev;
 863         struct net_device *dev = in_dev->dev;
 864         struct in_ifaddr *prim = ifa;
 865         __be32 mask = ifa->ifa_mask;
 866         __be32 addr = ifa->ifa_local;
 867         __be32 prefix = ifa->ifa_address & mask;
 868
 869         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 870                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
 871                 if (!prim) {
 872                         pr_warn("%s: bug: prim == NULL\n", __func__);
 873                         return;
 874                 }
 875         }
 876
 877         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
 878
 879         if (!(dev->flags & IFF_UP))
 880                 return;
 881
 882         /* Add broadcast address, if it is explicitly assigned. */
 883         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
 884                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
 885
 886         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
 887             (prefix != addr || ifa->ifa_prefixlen < 32)) {
 888                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
 889                         fib_magic(RTM_NEWROUTE,
 890                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 891                                   prefix, ifa->ifa_prefixlen, prim);
 892
 893                 /* Add network specific broadcasts, when it takes a sense */
 894                 if (ifa->ifa_prefixlen < 31) {
 895                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
 896                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
 897                                   32, prim);
 898                 }
 899         }
 900 }
 901
 902 /* Delete primary or secondary address.
 903  * Optionally, on secondary address promotion consider the addresses
 904  * from subnet iprim as deleted, even if they are in device list.
 905  * In this case the secondary ifa can be in device list.
 906  */
 907 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
 908 {
 909         struct in_device *in_dev = ifa->ifa_dev;
 910         struct net_device *dev = in_dev->dev;
 911         struct in_ifaddr *ifa1;
 912         struct in_ifaddr *prim = ifa, *prim1 = NULL;
 913         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
 914         __be32 any = ifa->ifa_address & ifa->ifa_mask;
 915 #define LOCAL_OK        1
 916 #define BRD_OK          2
 917 #define BRD0_OK         4
 918 #define BRD1_OK         8
 919         unsigned int ok = 0;
 920         int subnet = 0;         /* Primary network */
 921         int gone = 1;           /* Address is missing */
 922         int same_prefsrc = 0;   /* Another primary with same IP */
 923
 924         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 925                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
 926                 if (!prim) {
 927                         /* if the device has been deleted, we don't perform
 928                          * address promotion
 929                          */
 930                         if (!in_dev->dead)
 931                                 pr_warn("%s: bug: prim == NULL\n", __func__);
 932                         return;
 933                 }
 934                 if (iprim && iprim != prim) {
 935                         pr_warn("%s: bug: iprim != prim\n", __func__);
 936                         return;
 937                 }
 938         } else if (!ipv4_is_zeronet(any) &&
 939                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
 940                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
 941                         fib_magic(RTM_DELROUTE,
 942                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 943                                   any, ifa->ifa_prefixlen, prim);
 944                 subnet = 1;
 945         }
 946
 947         if (in_dev->dead)
 948                 goto no_promotions;
 949
 950         /* Deletion is more complicated than add.
 951          * We should take care of not to delete too much :-)
 952          *
 953          * Scan address list to be sure that addresses are really gone.
 954          */
 955
 956         for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
 957                 if (ifa1 == ifa) {
 958                         /* promotion, keep the IP */
 959                         gone = 0;
 960                         continue;
 961                 }
 962                 /* Ignore IFAs from our subnet */
 963                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
 964                     inet_ifa_match(ifa1->ifa_address, iprim))
 965                         continue;
 966
 967                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
 968                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
 969                         /* Another address from our subnet? */
 970                         if (ifa1->ifa_mask == prim->ifa_mask &&
 971                             inet_ifa_match(ifa1->ifa_address, prim))
 972                                 prim1 = prim;
 973                         else {
 974                                 /* We reached the secondaries, so
 975                                  * same_prefsrc should be determined.
 976                                  */
 977                                 if (!same_prefsrc)
 978                                         continue;
 979                                 /* Search new prim1 if ifa1 is not
 980                                  * using the current prim1
 981                                  */
 982                                 if (!prim1 ||
 983                                     ifa1->ifa_mask != prim1->ifa_mask ||
 984                                     !inet_ifa_match(ifa1->ifa_address, prim1))
 985                                         prim1 = inet_ifa_byprefix(in_dev,
 986                                                         ifa1->ifa_address,
 987                                                         ifa1->ifa_mask);
 988                                 if (!prim1)
 989                                         continue;
 990                                 if (prim1->ifa_local != prim->ifa_local)
 991                                         continue;
 992                         }
 993                 } else {
 994                         if (prim->ifa_local != ifa1->ifa_local)
 995                                 continue;
 996                         prim1 = ifa1;
 997                         if (prim != prim1)
 998                                 same_prefsrc = 1;
 999                 }
1000                 if (ifa->ifa_local == ifa1->ifa_local)
1001                         ok |= LOCAL_OK;
1002                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1003                         ok |= BRD_OK;
1004                 if (brd == ifa1->ifa_broadcast)
1005                         ok |= BRD1_OK;
1006                 if (any == ifa1->ifa_broadcast)
1007                         ok |= BRD0_OK;
1008                 /* primary has network specific broadcasts */
1009                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1010                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1011                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1012
1013                         if (!ipv4_is_zeronet(any1)) {
1014                                 if (ifa->ifa_broadcast == brd1 ||
1015                                     ifa->ifa_broadcast == any1)
1016                                         ok |= BRD_OK;
1017                                 if (brd == brd1 || brd == any1)
1018                                         ok |= BRD1_OK;
1019                                 if (any == brd1 || any == any1)
1020                                         ok |= BRD0_OK;
1021                         }
1022                 }
1023         }
1024
1025 no_promotions:
1026         if (!(ok & BRD_OK))
1027                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
1028         if (subnet && ifa->ifa_prefixlen < 31) {
1029                 if (!(ok & BRD1_OK))
1030                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
1031                 if (!(ok & BRD0_OK))
1032                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
1033         }
1034         if (!(ok & LOCAL_OK)) {
1035                 unsigned int addr_type;
1036
1037                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
1038
1039                 /* Check, that this local address finally disappeared. */
1040                 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1041                                                      ifa->ifa_local);
1042                 if (gone && addr_type != RTN_LOCAL) {
1043                         /* And the last, but not the least thing.
1044                          * We must flush stray FIB entries.
1045                          *
1046                          * First of all, we scan fib_info list searching
1047                          * for stray nexthop entries, then ignite fib_flush.
1048                          */
1049                         if (fib_sync_down_addr(dev, ifa->ifa_local))
1050                                 fib_flush(dev_net(dev));
1051                 }
1052         }
1053 #undef LOCAL_OK
1054 #undef BRD_OK
1055 #undef BRD0_OK
1056 #undef BRD1_OK
1057 }
1058
1059 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1060 {
1061
1062         struct fib_result       res;
1063         struct flowi4           fl4 = {
1064                 .flowi4_mark = frn->fl_mark,
1065                 .daddr = frn->fl_addr,
1066                 .flowi4_tos = frn->fl_tos,
1067                 .flowi4_scope = frn->fl_scope,
1068         };
1069         struct fib_table *tb;
1070
1071         rcu_read_lock();
1072
1073         tb = fib_get_table(net, frn->tb_id_in);
1074
1075         frn->err = -ENOENT;
1076         if (tb) {
1077                 local_bh_disable();
1078
1079                 frn->tb_id = tb->tb_id;
1080                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1081
1082                 if (!frn->err) {
1083                         frn->prefixlen = res.prefixlen;
1084                         frn->nh_sel = res.nh_sel;
1085                         frn->type = res.type;
1086                         frn->scope = res.scope;
1087                 }
1088                 local_bh_enable();
1089         }
1090
1091         rcu_read_unlock();
1092 }
1093
1094 static void nl_fib_input(struct sk_buff *skb)
1095 {
1096         struct net *net;
1097         struct fib_result_nl *frn;
1098         struct nlmsghdr *nlh;
1099         u32 portid;
1100
1101         net = sock_net(skb->sk);
1102         nlh = nlmsg_hdr(skb);
1103         if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1104             skb->len < nlh->nlmsg_len ||
1105             nlmsg_len(nlh) < sizeof(*frn))
1106                 return;
1107
1108         skb = netlink_skb_clone(skb, GFP_KERNEL);
1109         if (!skb)
1110                 return;
1111         nlh = nlmsg_hdr(skb);
1112
1113         frn = (struct fib_result_nl *) nlmsg_data(nlh);
1114         nl_fib_lookup(net, frn);
1115
1116         portid = NETLINK_CB(skb).portid;      /* netlink portid */
1117         NETLINK_CB(skb).portid = 0;        /* from kernel */
1118         NETLINK_CB(skb).dst_group = 0;  /* unicast */
1119         netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1120 }
1121
1122 static int __net_init nl_fib_lookup_init(struct net *net)
1123 {
1124         struct sock *sk;
1125         struct netlink_kernel_cfg cfg = {
1126                 .input  = nl_fib_input,
1127         };
1128
1129         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1130         if (!sk)
1131                 return -EAFNOSUPPORT;
1132         net->ipv4.fibnl = sk;
1133         return 0;
1134 }
1135
1136 static void nl_fib_lookup_exit(struct net *net)
1137 {
1138         netlink_kernel_release(net->ipv4.fibnl);
1139         net->ipv4.fibnl = NULL;
1140 }
1141
1142 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1143                            bool force)
1144 {
1145         if (fib_sync_down_dev(dev, event, force))
1146                 fib_flush(dev_net(dev));
1147         else
1148                 rt_cache_flush(dev_net(dev));
1149         arp_ifdown(dev);
1150 }
1151
1152 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1153 {
1154         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1155         struct net_device *dev = ifa->ifa_dev->dev;
1156         struct net *net = dev_net(dev);
1157
1158         switch (event) {
1159         case NETDEV_UP:
1160                 fib_add_ifaddr(ifa);
1161 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1162                 fib_sync_up(dev, RTNH_F_DEAD);
1163 #endif
1164                 atomic_inc(&net->ipv4.dev_addr_genid);
1165                 rt_cache_flush(dev_net(dev));
1166                 break;
1167         case NETDEV_DOWN:
1168                 fib_del_ifaddr(ifa, NULL);
1169                 atomic_inc(&net->ipv4.dev_addr_genid);
1170                 if (!ifa->ifa_dev->ifa_list) {
1171                         /* Last address was deleted from this interface.
1172                          * Disable IP.
1173                          */
1174                         fib_disable_ip(dev, event, true);
1175                 } else {
1176                         rt_cache_flush(dev_net(dev));
1177                 }
1178                 break;
1179         }
1180         return NOTIFY_DONE;
1181 }
1182
1183 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1184 {
1185         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1186         struct netdev_notifier_changeupper_info *info;
1187         struct in_device *in_dev;
1188         struct net *net = dev_net(dev);
1189         unsigned int flags;
1190
1191         if (event == NETDEV_UNREGISTER) {
1192                 fib_disable_ip(dev, event, true);
1193                 rt_flush_dev(dev);
1194                 return NOTIFY_DONE;
1195         }
1196
1197         in_dev = __in_dev_get_rtnl(dev);
1198         if (!in_dev)
1199                 return NOTIFY_DONE;
1200
1201         switch (event) {
1202         case NETDEV_UP:
1203                 for_ifa(in_dev) {
1204                         fib_add_ifaddr(ifa);
1205                 } endfor_ifa(in_dev);
1206 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1207                 fib_sync_up(dev, RTNH_F_DEAD);
1208 #endif
1209                 atomic_inc(&net->ipv4.dev_addr_genid);
1210                 rt_cache_flush(net);
1211                 break;
1212         case NETDEV_DOWN:
1213                 fib_disable_ip(dev, event, false);
1214                 break;
1215         case NETDEV_CHANGE:
1216                 flags = dev_get_flags(dev);
1217                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1218                         fib_sync_up(dev, RTNH_F_LINKDOWN);
1219                 else
1220                         fib_sync_down_dev(dev, event, false);
1221                 /* fall through */
1222         case NETDEV_CHANGEMTU:
1223                 rt_cache_flush(net);
1224                 break;
1225         case NETDEV_CHANGEUPPER:
1226                 info = ptr;
1227                 /* flush all routes if dev is linked to or unlinked from
1228                  * an L3 master device (e.g., VRF)
1229                  */
1230                 if (info->upper_dev && netif_is_l3_master(info->upper_dev))
1231                         fib_disable_ip(dev, NETDEV_DOWN, true);
1232                 break;
1233         }
1234         return NOTIFY_DONE;
1235 }
1236
1237 static struct notifier_block fib_inetaddr_notifier = {
1238         .notifier_call = fib_inetaddr_event,
1239 };
1240
1241 static struct notifier_block fib_netdev_notifier = {
1242         .notifier_call = fib_netdev_event,
1243 };
1244
1245 static int __net_init ip_fib_net_init(struct net *net)
1246 {
1247         int err;
1248         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1249
1250         err = fib4_notifier_init(net);
1251         if (err)
1252                 return err;
1253
1254         /* Avoid false sharing : Use at least a full cache line */
1255         size = max_t(size_t, size, L1_CACHE_BYTES);
1256
1257         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1258         if (!net->ipv4.fib_table_hash) {
1259                 err = -ENOMEM;
1260                 goto err_table_hash_alloc;
1261         }
1262
1263         err = fib4_rules_init(net);
1264         if (err < 0)
1265                 goto err_rules_init;
1266         return 0;
1267
1268 err_rules_init:
1269         kfree(net->ipv4.fib_table_hash);
1270 err_table_hash_alloc:
1271         fib4_notifier_exit(net);
1272         return err;
1273 }
1274
1275 static void ip_fib_net_exit(struct net *net)
1276 {
1277         unsigned int i;
1278
1279         rtnl_lock();
1280 #ifdef CONFIG_IP_MULTIPLE_TABLES
1281         RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1282         RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1283 #endif
1284         for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1285                 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1286                 struct hlist_node *tmp;
1287                 struct fib_table *tb;
1288
1289                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1290                         hlist_del(&tb->tb_hlist);
1291                         fib_table_flush(net, tb);
1292                         fib_free_table(tb);
1293                 }
1294         }
1295
1296 #ifdef CONFIG_IP_MULTIPLE_TABLES
1297         fib4_rules_exit(net);
1298 #endif
1299         rtnl_unlock();
1300         kfree(net->ipv4.fib_table_hash);
1301         fib4_notifier_exit(net);
1302 }
1303
1304 static int __net_init fib_net_init(struct net *net)
1305 {
1306         int error;
1307
1308 #ifdef CONFIG_IP_ROUTE_CLASSID
1309         net->ipv4.fib_num_tclassid_users = 0;
1310 #endif
1311         error = ip_fib_net_init(net);
1312         if (error < 0)
1313                 goto out;
1314         error = nl_fib_lookup_init(net);
1315         if (error < 0)
1316                 goto out_nlfl;
1317         error = fib_proc_init(net);
1318         if (error < 0)
1319                 goto out_proc;
1320 out:
1321         return error;
1322
1323 out_proc:
1324         nl_fib_lookup_exit(net);
1325 out_nlfl:
1326         ip_fib_net_exit(net);
1327         goto out;
1328 }
1329
1330 static void __net_exit fib_net_exit(struct net *net)
1331 {
1332         fib_proc_exit(net);
1333         nl_fib_lookup_exit(net);
1334         ip_fib_net_exit(net);
1335 }
1336
1337 static struct pernet_operations fib_net_ops = {
1338         .init = fib_net_init,
1339         .exit = fib_net_exit,
1340 };
1341
1342 void __init ip_fib_init(void)
1343 {
1344         fib_trie_init();
1345
1346         register_pernet_subsys(&fib_net_ops);
1347
1348         register_netdevice_notifier(&fib_netdev_notifier);
1349         register_inetaddr_notifier(&fib_inetaddr_notifier);
1350
1351         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1352         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1353         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1354 }