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
  77 static bool fib4_has_custom_rules(struct net *net)
  78 {
  79         return false;
  80 }
  81 #else
  82
  83 struct fib_table *fib_new_table(struct net *net, u32 id)
  84 {
  85         struct fib_table *tb, *alias = NULL;
  86         unsigned int h;
  87
  88         if (id == 0)
  89                 id = RT_TABLE_MAIN;
  90         tb = fib_get_table(net, id);
  91         if (tb)
  92                 return tb;
  93
  94         if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
  95                 alias = fib_new_table(net, RT_TABLE_MAIN);
  96
  97         tb = fib_trie_table(id, alias);
  98         if (!tb)
  99                 return NULL;
 100
 101         switch (id) {
 102         case RT_TABLE_MAIN:
 103                 rcu_assign_pointer(net->ipv4.fib_main, tb);
 104                 break;
 105         case RT_TABLE_DEFAULT:
 106                 rcu_assign_pointer(net->ipv4.fib_default, tb);
 107                 break;
 108         default:
 109                 break;
 110         }
 111
 112         h = id & (FIB_TABLE_HASHSZ - 1);
 113         hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
 114         return tb;
 115 }
 116 EXPORT_SYMBOL_GPL(fib_new_table);
 117
 118 /* caller must hold either rtnl or rcu read lock */
 119 struct fib_table *fib_get_table(struct net *net, u32 id)
 120 {
 121         struct fib_table *tb;
 122         struct hlist_head *head;
 123         unsigned int h;
 124
 125         if (id == 0)
 126                 id = RT_TABLE_MAIN;
 127         h = id & (FIB_TABLE_HASHSZ - 1);
 128
 129         head = &net->ipv4.fib_table_hash[h];
 130         hlist_for_each_entry_rcu(tb, head, tb_hlist) {
 131                 if (tb->tb_id == id)
 132                         return tb;
 133         }
 134         return NULL;
 135 }
 136
 137 static bool fib4_has_custom_rules(struct net *net)
 138 {
 139         return net->ipv4.fib_has_custom_rules;
 140 }
 141 #endif /* CONFIG_IP_MULTIPLE_TABLES */
 142
 143 static void fib_replace_table(struct net *net, struct fib_table *old,
 144                               struct fib_table *new)
 145 {
 146 #ifdef CONFIG_IP_MULTIPLE_TABLES
 147         switch (new->tb_id) {
 148         case RT_TABLE_MAIN:
 149                 rcu_assign_pointer(net->ipv4.fib_main, new);
 150                 break;
 151         case RT_TABLE_DEFAULT:
 152                 rcu_assign_pointer(net->ipv4.fib_default, new);
 153                 break;
 154         default:
 155                 break;
 156         }
 157
 158 #endif
 159         /* replace the old table in the hlist */
 160         hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
 161 }
 162
 163 int fib_unmerge(struct net *net)
 164 {
 165         struct fib_table *old, *new, *main_table;
 166
 167         /* attempt to fetch local table if it has been allocated */
 168         old = fib_get_table(net, RT_TABLE_LOCAL);
 169         if (!old)
 170                 return 0;
 171
 172         new = fib_trie_unmerge(old);
 173         if (!new)
 174                 return -ENOMEM;
 175
 176         /* table is already unmerged */
 177         if (new == old)
 178                 return 0;
 179
 180         /* replace merged table with clean table */
 181         fib_replace_table(net, old, new);
 182         fib_free_table(old);
 183
 184         /* attempt to fetch main table if it has been allocated */
 185         main_table = fib_get_table(net, RT_TABLE_MAIN);
 186         if (!main_table)
 187                 return 0;
 188
 189         /* flush local entries from main table */
 190         fib_table_flush_external(main_table);
 191
 192         return 0;
 193 }
 194
 195 static void fib_flush(struct net *net)
 196 {
 197         int flushed = 0;
 198         unsigned int h;
 199
 200         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
 201                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
 202                 struct hlist_node *tmp;
 203                 struct fib_table *tb;
 204
 205                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
 206                         flushed += fib_table_flush(net, tb);
 207         }
 208
 209         if (flushed)
 210                 rt_cache_flush(net);
 211 }
 212
 213 /*
 214  * Find address type as if only "dev" was present in the system. If
 215  * on_dev is NULL then all interfaces are taken into consideration.
 216  */
 217 static inline unsigned int __inet_dev_addr_type(struct net *net,
 218                                                 const struct net_device *dev,
 219                                                 __be32 addr, u32 tb_id)
 220 {
 221         struct flowi4           fl4 = { .daddr = addr };
 222         struct fib_result       res;
 223         unsigned int ret = RTN_BROADCAST;
 224         struct fib_table *table;
 225
 226         if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
 227                 return RTN_BROADCAST;
 228         if (ipv4_is_multicast(addr))
 229                 return RTN_MULTICAST;
 230
 231         rcu_read_lock();
 232
 233         table = fib_get_table(net, tb_id);
 234         if (table) {
 235                 ret = RTN_UNICAST;
 236                 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
 237                         if (!dev || dev == res.fi->fib_dev)
 238                                 ret = res.type;
 239                 }
 240         }
 241
 242         rcu_read_unlock();
 243         return ret;
 244 }
 245
 246 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
 247 {
 248         return __inet_dev_addr_type(net, NULL, addr, tb_id);
 249 }
 250 EXPORT_SYMBOL(inet_addr_type_table);
 251
 252 unsigned int inet_addr_type(struct net *net, __be32 addr)
 253 {
 254         return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
 255 }
 256 EXPORT_SYMBOL(inet_addr_type);
 257
 258 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
 259                                 __be32 addr)
 260 {
 261         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
 262
 263         return __inet_dev_addr_type(net, dev, addr, rt_table);
 264 }
 265 EXPORT_SYMBOL(inet_dev_addr_type);
 266
 267 /* inet_addr_type with dev == NULL but using the table from a dev
 268  * if one is associated
 269  */
 270 unsigned int inet_addr_type_dev_table(struct net *net,
 271                                       const struct net_device *dev,
 272                                       __be32 addr)
 273 {
 274         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
 275
 276         return __inet_dev_addr_type(net, NULL, addr, rt_table);
 277 }
 278 EXPORT_SYMBOL(inet_addr_type_dev_table);
 279
 280 __be32 fib_compute_spec_dst(struct sk_buff *skb)
 281 {
 282         struct net_device *dev = skb->dev;
 283         struct in_device *in_dev;
 284         struct fib_result res;
 285         struct rtable *rt;
 286         struct net *net;
 287         int scope;
 288
 289         rt = skb_rtable(skb);
 290         if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
 291             RTCF_LOCAL)
 292                 return ip_hdr(skb)->daddr;
 293
 294         in_dev = __in_dev_get_rcu(dev);
 295         BUG_ON(!in_dev);
 296
 297         net = dev_net(dev);
 298
 299         scope = RT_SCOPE_UNIVERSE;
 300         if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
 301                 struct flowi4 fl4 = {
 302                         .flowi4_iif = LOOPBACK_IFINDEX,
 303                         .daddr = ip_hdr(skb)->saddr,
 304                         .flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
 305                         .flowi4_scope = scope,
 306                         .flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0,
 307                 };
 308                 if (!fib_lookup(net, &fl4, &res, 0))
 309                         return FIB_RES_PREFSRC(net, res);
 310         } else {
 311                 scope = RT_SCOPE_LINK;
 312         }
 313
 314         return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
 315 }
 316
 317 /* Given (packet source, input interface) and optional (dst, oif, tos):
 318  * - (main) check, that source is valid i.e. not broadcast or our local
 319  *   address.
 320  * - figure out what "logical" interface this packet arrived
 321  *   and calculate "specific destination" address.
 322  * - check, that packet arrived from expected physical interface.
 323  * called with rcu_read_lock()
 324  */
 325 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 326                                  u8 tos, int oif, struct net_device *dev,
 327                                  int rpf, struct in_device *idev, u32 *itag)
 328 {
 329         int ret, no_addr;
 330         struct fib_result res;
 331         struct flowi4 fl4;
 332         struct net *net = dev_net(dev);
 333         bool dev_match;
 334
 335         fl4.flowi4_oif = 0;
 336         fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
 337         if (!fl4.flowi4_iif)
 338                 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
 339         fl4.daddr = src;
 340         fl4.saddr = dst;
 341         fl4.flowi4_tos = tos;
 342         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 343         fl4.flowi4_tun_key.tun_id = 0;
 344         fl4.flowi4_flags = 0;
 345         fl4.flowi4_uid = sock_net_uid(net, NULL);
 346
 347         no_addr = idev->ifa_list == NULL;
 348
 349         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
 350
 351         trace_fib_validate_source(dev, &fl4);
 352
 353         if (fib_lookup(net, &fl4, &res, 0))
 354                 goto last_resort;
 355         if (res.type != RTN_UNICAST &&
 356             (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
 357                 goto e_inval;
 358         fib_combine_itag(itag, &res);
 359         dev_match = false;
 360
 361 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 362         for (ret = 0; ret < res.fi->fib_nhs; ret++) {
 363                 struct fib_nh *nh = &res.fi->fib_nh[ret];
 364
 365                 if (nh->nh_dev == dev) {
 366                         dev_match = true;
 367                         break;
 368                 } else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) {
 369                         dev_match = true;
 370                         break;
 371                 }
 372         }
 373 #else
 374         if (FIB_RES_DEV(res) == dev)
 375                 dev_match = true;
 376 #endif
 377         if (dev_match) {
 378                 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 379                 return ret;
 380         }
 381         if (no_addr)
 382                 goto last_resort;
 383         if (rpf == 1)
 384                 goto e_rpf;
 385         fl4.flowi4_oif = dev->ifindex;
 386
 387         ret = 0;
 388         if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
 389                 if (res.type == RTN_UNICAST)
 390                         ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 391         }
 392         return ret;
 393
 394 last_resort:
 395         if (rpf)
 396                 goto e_rpf;
 397         *itag = 0;
 398         return 0;
 399
 400 e_inval:
 401         return -EINVAL;
 402 e_rpf:
 403         return -EXDEV;
 404 }
 405
 406 /* Ignore rp_filter for packets protected by IPsec. */
 407 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 408                         u8 tos, int oif, struct net_device *dev,
 409                         struct in_device *idev, u32 *itag)
 410 {
 411         int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
 412         struct net *net = dev_net(dev);
 413
 414         if (!r && !fib_num_tclassid_users(net) &&
 415             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
 416                 if (IN_DEV_ACCEPT_LOCAL(idev))
 417                         goto ok;
 418                 /* with custom local routes in place, checking local addresses
 419                  * only will be too optimistic, with custom rules, checking
 420                  * local addresses only can be too strict, e.g. due to vrf
 421                  */
 422                 if (net->ipv4.fib_has_custom_local_routes ||
 423                     fib4_has_custom_rules(net))
 424                         goto full_check;
 425                 if (inet_lookup_ifaddr_rcu(net, src))
 426                         return -EINVAL;
 427
 428 ok:
 429                 *itag = 0;
 430                 return 0;
 431         }
 432
 433 full_check:
 434         return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
 435 }
 436
 437 static inline __be32 sk_extract_addr(struct sockaddr *addr)
 438 {
 439         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 440 }
 441
 442 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
 443 {
 444         struct nlattr *nla;
 445
 446         nla = (struct nlattr *) ((char *) mx + len);
 447         nla->nla_type = type;
 448         nla->nla_len = nla_attr_size(4);
 449         *(u32 *) nla_data(nla) = value;
 450
 451         return len + nla_total_size(4);
 452 }
 453
 454 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
 455                                  struct fib_config *cfg)
 456 {
 457         __be32 addr;
 458         int plen;
 459
 460         memset(cfg, 0, sizeof(*cfg));
 461         cfg->fc_nlinfo.nl_net = net;
 462
 463         if (rt->rt_dst.sa_family != AF_INET)
 464                 return -EAFNOSUPPORT;
 465
 466         /*
 467          * Check mask for validity:
 468          * a) it must be contiguous.
 469          * b) destination must have all host bits clear.
 470          * c) if application forgot to set correct family (AF_INET),
 471          *    reject request unless it is absolutely clear i.e.
 472          *    both family and mask are zero.
 473          */
 474         plen = 32;
 475         addr = sk_extract_addr(&rt->rt_dst);
 476         if (!(rt->rt_flags & RTF_HOST)) {
 477                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
 478
 479                 if (rt->rt_genmask.sa_family != AF_INET) {
 480                         if (mask || rt->rt_genmask.sa_family)
 481                                 return -EAFNOSUPPORT;
 482                 }
 483
 484                 if (bad_mask(mask, addr))
 485                         return -EINVAL;
 486
 487                 plen = inet_mask_len(mask);
 488         }
 489
 490         cfg->fc_dst_len = plen;
 491         cfg->fc_dst = addr;
 492
 493         if (cmd != SIOCDELRT) {
 494                 cfg->fc_nlflags = NLM_F_CREATE;
 495                 cfg->fc_protocol = RTPROT_BOOT;
 496         }
 497
 498         if (rt->rt_metric)
 499                 cfg->fc_priority = rt->rt_metric - 1;
 500
 501         if (rt->rt_flags & RTF_REJECT) {
 502                 cfg->fc_scope = RT_SCOPE_HOST;
 503                 cfg->fc_type = RTN_UNREACHABLE;
 504                 return 0;
 505         }
 506
 507         cfg->fc_scope = RT_SCOPE_NOWHERE;
 508         cfg->fc_type = RTN_UNICAST;
 509
 510         if (rt->rt_dev) {
 511                 char *colon;
 512                 struct net_device *dev;
 513                 char devname[IFNAMSIZ];
 514
 515                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
 516                         return -EFAULT;
 517
 518                 devname[IFNAMSIZ-1] = 0;
 519                 colon = strchr(devname, ':');
 520                 if (colon)
 521                         *colon = 0;
 522                 dev = __dev_get_by_name(net, devname);
 523                 if (!dev)
 524                         return -ENODEV;
 525                 cfg->fc_oif = dev->ifindex;
 526                 cfg->fc_table = l3mdev_fib_table(dev);
 527                 if (colon) {
 528                         struct in_ifaddr *ifa;
 529                         struct in_device *in_dev = __in_dev_get_rtnl(dev);
 530                         if (!in_dev)
 531                                 return -ENODEV;
 532                         *colon = ':';
 533                         for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
 534                                 if (strcmp(ifa->ifa_label, devname) == 0)
 535                                         break;
 536                         if (!ifa)
 537                                 return -ENODEV;
 538                         cfg->fc_prefsrc = ifa->ifa_local;
 539                 }
 540         }
 541
 542         addr = sk_extract_addr(&rt->rt_gateway);
 543         if (rt->rt_gateway.sa_family == AF_INET && addr) {
 544                 unsigned int addr_type;
 545
 546                 cfg->fc_gw = addr;
 547                 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
 548                 if (rt->rt_flags & RTF_GATEWAY &&
 549                     addr_type == RTN_UNICAST)
 550                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
 551         }
 552
 553         if (cmd == SIOCDELRT)
 554                 return 0;
 555
 556         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
 557                 return -EINVAL;
 558
 559         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 560                 cfg->fc_scope = RT_SCOPE_LINK;
 561
 562         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 563                 struct nlattr *mx;
 564                 int len = 0;
 565
 566                 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
 567                 if (!mx)
 568                         return -ENOMEM;
 569
 570                 if (rt->rt_flags & RTF_MTU)
 571                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 572
 573                 if (rt->rt_flags & RTF_WINDOW)
 574                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 575
 576                 if (rt->rt_flags & RTF_IRTT)
 577                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 578
 579                 cfg->fc_mx = mx;
 580                 cfg->fc_mx_len = len;
 581         }
 582
 583         return 0;
 584 }
 585
 586 /*
 587  * Handle IP routing ioctl calls.
 588  * These are used to manipulate the routing tables
 589  */
 590 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
 591 {
 592         struct fib_config cfg;
 593         struct rtentry rt;
 594         int err;
 595
 596         switch (cmd) {
 597         case SIOCADDRT:         /* Add a route */
 598         case SIOCDELRT:         /* Delete a route */
 599                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 600                         return -EPERM;
 601
 602                 if (copy_from_user(&rt, arg, sizeof(rt)))
 603                         return -EFAULT;
 604
 605                 rtnl_lock();
 606                 err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
 607                 if (err == 0) {
 608                         struct fib_table *tb;
 609
 610                         if (cmd == SIOCDELRT) {
 611                                 tb = fib_get_table(net, cfg.fc_table);
 612                                 if (tb)
 613                                         err = fib_table_delete(net, tb, &cfg,
 614                                                                NULL);
 615                                 else
 616                                         err = -ESRCH;
 617                         } else {
 618                                 tb = fib_new_table(net, cfg.fc_table);
 619                                 if (tb)
 620                                         err = fib_table_insert(net, tb,
 621                                                                &cfg, NULL);
 622                                 else
 623                                         err = -ENOBUFS;
 624                         }
 625
 626                         /* allocated by rtentry_to_fib_config() */
 627                         kfree(cfg.fc_mx);
 628                 }
 629                 rtnl_unlock();
 630                 return err;
 631         }
 632         return -EINVAL;
 633 }
 634
 635 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 636         [RTA_DST]               = { .type = NLA_U32 },
 637         [RTA_SRC]               = { .type = NLA_U32 },
 638         [RTA_IIF]               = { .type = NLA_U32 },
 639         [RTA_OIF]               = { .type = NLA_U32 },
 640         [RTA_GATEWAY]           = { .type = NLA_U32 },
 641         [RTA_PRIORITY]          = { .type = NLA_U32 },
 642         [RTA_PREFSRC]           = { .type = NLA_U32 },
 643         [RTA_METRICS]           = { .type = NLA_NESTED },
 644         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 645         [RTA_FLOW]              = { .type = NLA_U32 },
 646         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
 647         [RTA_ENCAP]             = { .type = NLA_NESTED },
 648         [RTA_UID]               = { .type = NLA_U32 },
 649         [RTA_MARK]              = { .type = NLA_U32 },
 650 };
 651
 652 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 653                              struct nlmsghdr *nlh, struct fib_config *cfg,
 654                              struct netlink_ext_ack *extack)
 655 {
 656         struct nlattr *attr;
 657         int err, remaining;
 658         struct rtmsg *rtm;
 659
 660         err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy,
 661                              extack);
 662         if (err < 0)
 663                 goto errout;
 664
 665         memset(cfg, 0, sizeof(*cfg));
 666
 667         rtm = nlmsg_data(nlh);
 668         cfg->fc_dst_len = rtm->rtm_dst_len;
 669         cfg->fc_tos = rtm->rtm_tos;
 670         cfg->fc_table = rtm->rtm_table;
 671         cfg->fc_protocol = rtm->rtm_protocol;
 672         cfg->fc_scope = rtm->rtm_scope;
 673         cfg->fc_type = rtm->rtm_type;
 674         cfg->fc_flags = rtm->rtm_flags;
 675         cfg->fc_nlflags = nlh->nlmsg_flags;
 676
 677         cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
 678         cfg->fc_nlinfo.nlh = nlh;
 679         cfg->fc_nlinfo.nl_net = net;
 680
 681         if (cfg->fc_type > RTN_MAX) {
 682                 NL_SET_ERR_MSG(extack, "Invalid route type");
 683                 err = -EINVAL;
 684                 goto errout;
 685         }
 686
 687         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 688                 switch (nla_type(attr)) {
 689                 case RTA_DST:
 690                         cfg->fc_dst = nla_get_be32(attr);
 691                         break;
 692                 case RTA_OIF:
 693                         cfg->fc_oif = nla_get_u32(attr);
 694                         break;
 695                 case RTA_GATEWAY:
 696                         cfg->fc_gw = nla_get_be32(attr);
 697                         break;
 698                 case RTA_PRIORITY:
 699                         cfg->fc_priority = nla_get_u32(attr);
 700                         break;
 701                 case RTA_PREFSRC:
 702                         cfg->fc_prefsrc = nla_get_be32(attr);
 703                         break;
 704                 case RTA_METRICS:
 705                         cfg->fc_mx = nla_data(attr);
 706                         cfg->fc_mx_len = nla_len(attr);
 707                         break;
 708                 case RTA_MULTIPATH:
 709                         err = lwtunnel_valid_encap_type_attr(nla_data(attr),
 710                                                              nla_len(attr),
 711                                                              extack);
 712                         if (err < 0)
 713                                 goto errout;
 714                         cfg->fc_mp = nla_data(attr);
 715                         cfg->fc_mp_len = nla_len(attr);
 716                         break;
 717                 case RTA_FLOW:
 718                         cfg->fc_flow = nla_get_u32(attr);
 719                         break;
 720                 case RTA_TABLE:
 721                         cfg->fc_table = nla_get_u32(attr);
 722                         break;
 723                 case RTA_ENCAP:
 724                         cfg->fc_encap = attr;
 725                         break;
 726                 case RTA_ENCAP_TYPE:
 727                         cfg->fc_encap_type = nla_get_u16(attr);
 728                         err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
 729                                                         extack);
 730                         if (err < 0)
 731                                 goto errout;
 732                         break;
 733                 }
 734         }
 735
 736         return 0;
 737 errout:
 738         return err;
 739 }
 740
 741 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 742                              struct netlink_ext_ack *extack)
 743 {
 744         struct net *net = sock_net(skb->sk);
 745         struct fib_config cfg;
 746         struct fib_table *tb;
 747         int err;
 748
 749         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 750         if (err < 0)
 751                 goto errout;
 752
 753         tb = fib_get_table(net, cfg.fc_table);
 754         if (!tb) {
 755                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
 756                 err = -ESRCH;
 757                 goto errout;
 758         }
 759
 760         err = fib_table_delete(net, tb, &cfg, extack);
 761 errout:
 762         return err;
 763 }
 764
 765 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 766                              struct netlink_ext_ack *extack)
 767 {
 768         struct net *net = sock_net(skb->sk);
 769         struct fib_config cfg;
 770         struct fib_table *tb;
 771         int err;
 772
 773         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 774         if (err < 0)
 775                 goto errout;
 776
 777         tb = fib_new_table(net, cfg.fc_table);
 778         if (!tb) {
 779                 err = -ENOBUFS;
 780                 goto errout;
 781         }
 782
 783         err = fib_table_insert(net, tb, &cfg, extack);
 784         if (!err && cfg.fc_type == RTN_LOCAL)
 785                 net->ipv4.fib_has_custom_local_routes = true;
 786 errout:
 787         return err;
 788 }
 789
 790 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 791 {
 792         struct net *net = sock_net(skb->sk);
 793         unsigned int h, s_h;
 794         unsigned int e = 0, s_e;
 795         struct fib_table *tb;
 796         struct hlist_head *head;
 797         int dumped = 0, err;
 798
 799         if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
 800             ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
 801                 return skb->len;
 802
 803         s_h = cb->args[0];
 804         s_e = cb->args[1];
 805
 806         rcu_read_lock();
 807
 808         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
 809                 e = 0;
 810                 head = &net->ipv4.fib_table_hash[h];
 811                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
 812                         if (e < s_e)
 813                                 goto next;
 814                         if (dumped)
 815                                 memset(&cb->args[2], 0, sizeof(cb->args) -
 816                                                  2 * sizeof(cb->args[0]));
 817                         err = fib_table_dump(tb, skb, cb);
 818                         if (err < 0) {
 819                                 if (likely(skb->len))
 820                                         goto out;
 821
 822                                 goto out_err;
 823                         }
 824                         dumped = 1;
 825 next:
 826                         e++;
 827                 }
 828         }
 829 out:
 830         err = skb->len;
 831 out_err:
 832         rcu_read_unlock();
 833
 834         cb->args[1] = e;
 835         cb->args[0] = h;
 836
 837         return err;
 838 }
 839
 840 /* Prepare and feed intra-kernel routing request.
 841  * Really, it should be netlink message, but :-( netlink
 842  * can be not configured, so that we feed it directly
 843  * to fib engine. It is legal, because all events occur
 844  * only when netlink is already locked.
 845  */
 846 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
 847 {
 848         struct net *net = dev_net(ifa->ifa_dev->dev);
 849         u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
 850         struct fib_table *tb;
 851         struct fib_config cfg = {
 852                 .fc_protocol = RTPROT_KERNEL,
 853                 .fc_type = type,
 854                 .fc_dst = dst,
 855                 .fc_dst_len = dst_len,
 856                 .fc_prefsrc = ifa->ifa_local,
 857                 .fc_oif = ifa->ifa_dev->dev->ifindex,
 858                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
 859                 .fc_nlinfo = {
 860                         .nl_net = net,
 861                 },
 862         };
 863
 864         if (!tb_id)
 865                 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
 866
 867         tb = fib_new_table(net, tb_id);
 868         if (!tb)
 869                 return;
 870
 871         cfg.fc_table = tb->tb_id;
 872
 873         if (type != RTN_LOCAL)
 874                 cfg.fc_scope = RT_SCOPE_LINK;
 875         else
 876                 cfg.fc_scope = RT_SCOPE_HOST;
 877
 878         if (cmd == RTM_NEWROUTE)
 879                 fib_table_insert(net, tb, &cfg, NULL);
 880         else
 881                 fib_table_delete(net, tb, &cfg, NULL);
 882 }
 883
 884 void fib_add_ifaddr(struct in_ifaddr *ifa)
 885 {
 886         struct in_device *in_dev = ifa->ifa_dev;
 887         struct net_device *dev = in_dev->dev;
 888         struct in_ifaddr *prim = ifa;
 889         __be32 mask = ifa->ifa_mask;
 890         __be32 addr = ifa->ifa_local;
 891         __be32 prefix = ifa->ifa_address & mask;
 892
 893         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 894                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
 895                 if (!prim) {
 896                         pr_warn("%s: bug: prim == NULL\n", __func__);
 897                         return;
 898                 }
 899         }
 900
 901         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
 902
 903         if (!(dev->flags & IFF_UP))
 904                 return;
 905
 906         /* Add broadcast address, if it is explicitly assigned. */
 907         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
 908                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
 909
 910         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
 911             (prefix != addr || ifa->ifa_prefixlen < 32)) {
 912                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
 913                         fib_magic(RTM_NEWROUTE,
 914                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 915                                   prefix, ifa->ifa_prefixlen, prim);
 916
 917                 /* Add network specific broadcasts, when it takes a sense */
 918                 if (ifa->ifa_prefixlen < 31) {
 919                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
 920                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
 921                                   32, prim);
 922                 }
 923         }
 924 }
 925
 926 /* Delete primary or secondary address.
 927  * Optionally, on secondary address promotion consider the addresses
 928  * from subnet iprim as deleted, even if they are in device list.
 929  * In this case the secondary ifa can be in device list.
 930  */
 931 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
 932 {
 933         struct in_device *in_dev = ifa->ifa_dev;
 934         struct net_device *dev = in_dev->dev;
 935         struct in_ifaddr *ifa1;
 936         struct in_ifaddr *prim = ifa, *prim1 = NULL;
 937         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
 938         __be32 any = ifa->ifa_address & ifa->ifa_mask;
 939 #define LOCAL_OK        1
 940 #define BRD_OK          2
 941 #define BRD0_OK         4
 942 #define BRD1_OK         8
 943         unsigned int ok = 0;
 944         int subnet = 0;         /* Primary network */
 945         int gone = 1;           /* Address is missing */
 946         int same_prefsrc = 0;   /* Another primary with same IP */
 947
 948         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 949                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
 950                 if (!prim) {
 951                         /* if the device has been deleted, we don't perform
 952                          * address promotion
 953                          */
 954                         if (!in_dev->dead)
 955                                 pr_warn("%s: bug: prim == NULL\n", __func__);
 956                         return;
 957                 }
 958                 if (iprim && iprim != prim) {
 959                         pr_warn("%s: bug: iprim != prim\n", __func__);
 960                         return;
 961                 }
 962         } else if (!ipv4_is_zeronet(any) &&
 963                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
 964                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
 965                         fib_magic(RTM_DELROUTE,
 966                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 967                                   any, ifa->ifa_prefixlen, prim);
 968                 subnet = 1;
 969         }
 970
 971         if (in_dev->dead)
 972                 goto no_promotions;
 973
 974         /* Deletion is more complicated than add.
 975          * We should take care of not to delete too much :-)
 976          *
 977          * Scan address list to be sure that addresses are really gone.
 978          */
 979
 980         for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
 981                 if (ifa1 == ifa) {
 982                         /* promotion, keep the IP */
 983                         gone = 0;
 984                         continue;
 985                 }
 986                 /* Ignore IFAs from our subnet */
 987                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
 988                     inet_ifa_match(ifa1->ifa_address, iprim))
 989                         continue;
 990
 991                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
 992                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
 993                         /* Another address from our subnet? */
 994                         if (ifa1->ifa_mask == prim->ifa_mask &&
 995                             inet_ifa_match(ifa1->ifa_address, prim))
 996                                 prim1 = prim;
 997                         else {
 998                                 /* We reached the secondaries, so
 999                                  * same_prefsrc should be determined.
1000                                  */
1001                                 if (!same_prefsrc)
1002                                         continue;
1003                                 /* Search new prim1 if ifa1 is not
1004                                  * using the current prim1
1005                                  */
1006                                 if (!prim1 ||
1007                                     ifa1->ifa_mask != prim1->ifa_mask ||
1008                                     !inet_ifa_match(ifa1->ifa_address, prim1))
1009                                         prim1 = inet_ifa_byprefix(in_dev,
1010                                                         ifa1->ifa_address,
1011                                                         ifa1->ifa_mask);
1012                                 if (!prim1)
1013                                         continue;
1014                                 if (prim1->ifa_local != prim->ifa_local)
1015                                         continue;
1016                         }
1017                 } else {
1018                         if (prim->ifa_local != ifa1->ifa_local)
1019                                 continue;
1020                         prim1 = ifa1;
1021                         if (prim != prim1)
1022                                 same_prefsrc = 1;
1023                 }
1024                 if (ifa->ifa_local == ifa1->ifa_local)
1025                         ok |= LOCAL_OK;
1026                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1027                         ok |= BRD_OK;
1028                 if (brd == ifa1->ifa_broadcast)
1029                         ok |= BRD1_OK;
1030                 if (any == ifa1->ifa_broadcast)
1031                         ok |= BRD0_OK;
1032                 /* primary has network specific broadcasts */
1033                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1034                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1035                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1036
1037                         if (!ipv4_is_zeronet(any1)) {
1038                                 if (ifa->ifa_broadcast == brd1 ||
1039                                     ifa->ifa_broadcast == any1)
1040                                         ok |= BRD_OK;
1041                                 if (brd == brd1 || brd == any1)
1042                                         ok |= BRD1_OK;
1043                                 if (any == brd1 || any == any1)
1044                                         ok |= BRD0_OK;
1045                         }
1046                 }
1047         }
1048
1049 no_promotions:
1050         if (!(ok & BRD_OK))
1051                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
1052         if (subnet && ifa->ifa_prefixlen < 31) {
1053                 if (!(ok & BRD1_OK))
1054                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
1055                 if (!(ok & BRD0_OK))
1056                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
1057         }
1058         if (!(ok & LOCAL_OK)) {
1059                 unsigned int addr_type;
1060
1061                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
1062
1063                 /* Check, that this local address finally disappeared. */
1064                 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1065                                                      ifa->ifa_local);
1066                 if (gone && addr_type != RTN_LOCAL) {
1067                         /* And the last, but not the least thing.
1068                          * We must flush stray FIB entries.
1069                          *
1070                          * First of all, we scan fib_info list searching
1071                          * for stray nexthop entries, then ignite fib_flush.
1072                          */
1073                         if (fib_sync_down_addr(dev, ifa->ifa_local))
1074                                 fib_flush(dev_net(dev));
1075                 }
1076         }
1077 #undef LOCAL_OK
1078 #undef BRD_OK
1079 #undef BRD0_OK
1080 #undef BRD1_OK
1081 }
1082
1083 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1084 {
1085
1086         struct fib_result       res;
1087         struct flowi4           fl4 = {
1088                 .flowi4_mark = frn->fl_mark,
1089                 .daddr = frn->fl_addr,
1090                 .flowi4_tos = frn->fl_tos,
1091                 .flowi4_scope = frn->fl_scope,
1092         };
1093         struct fib_table *tb;
1094
1095         rcu_read_lock();
1096
1097         tb = fib_get_table(net, frn->tb_id_in);
1098
1099         frn->err = -ENOENT;
1100         if (tb) {
1101                 local_bh_disable();
1102
1103                 frn->tb_id = tb->tb_id;
1104                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1105
1106                 if (!frn->err) {
1107                         frn->prefixlen = res.prefixlen;
1108                         frn->nh_sel = res.nh_sel;
1109                         frn->type = res.type;
1110                         frn->scope = res.scope;
1111                 }
1112                 local_bh_enable();
1113         }
1114
1115         rcu_read_unlock();
1116 }
1117
1118 static void nl_fib_input(struct sk_buff *skb)
1119 {
1120         struct net *net;
1121         struct fib_result_nl *frn;
1122         struct nlmsghdr *nlh;
1123         u32 portid;
1124
1125         net = sock_net(skb->sk);
1126         nlh = nlmsg_hdr(skb);
1127         if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1128             skb->len < nlh->nlmsg_len ||
1129             nlmsg_len(nlh) < sizeof(*frn))
1130                 return;
1131
1132         skb = netlink_skb_clone(skb, GFP_KERNEL);
1133         if (!skb)
1134                 return;
1135         nlh = nlmsg_hdr(skb);
1136
1137         frn = (struct fib_result_nl *) nlmsg_data(nlh);
1138         nl_fib_lookup(net, frn);
1139
1140         portid = NETLINK_CB(skb).portid;      /* netlink portid */
1141         NETLINK_CB(skb).portid = 0;        /* from kernel */
1142         NETLINK_CB(skb).dst_group = 0;  /* unicast */
1143         netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1144 }
1145
1146 static int __net_init nl_fib_lookup_init(struct net *net)
1147 {
1148         struct sock *sk;
1149         struct netlink_kernel_cfg cfg = {
1150                 .input  = nl_fib_input,
1151         };
1152
1153         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1154         if (!sk)
1155                 return -EAFNOSUPPORT;
1156         net->ipv4.fibnl = sk;
1157         return 0;
1158 }
1159
1160 static void nl_fib_lookup_exit(struct net *net)
1161 {
1162         netlink_kernel_release(net->ipv4.fibnl);
1163         net->ipv4.fibnl = NULL;
1164 }
1165
1166 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1167                            bool force)
1168 {
1169         if (fib_sync_down_dev(dev, event, force))
1170                 fib_flush(dev_net(dev));
1171         else
1172                 rt_cache_flush(dev_net(dev));
1173         arp_ifdown(dev);
1174 }
1175
1176 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1177 {
1178         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1179         struct net_device *dev = ifa->ifa_dev->dev;
1180         struct net *net = dev_net(dev);
1181
1182         switch (event) {
1183         case NETDEV_UP:
1184                 fib_add_ifaddr(ifa);
1185 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1186                 fib_sync_up(dev, RTNH_F_DEAD);
1187 #endif
1188                 atomic_inc(&net->ipv4.dev_addr_genid);
1189                 rt_cache_flush(dev_net(dev));
1190                 break;
1191         case NETDEV_DOWN:
1192                 fib_del_ifaddr(ifa, NULL);
1193                 atomic_inc(&net->ipv4.dev_addr_genid);
1194                 if (!ifa->ifa_dev->ifa_list) {
1195                         /* Last address was deleted from this interface.
1196                          * Disable IP.
1197                          */
1198                         fib_disable_ip(dev, event, true);
1199                 } else {
1200                         rt_cache_flush(dev_net(dev));
1201                 }
1202                 break;
1203         }
1204         return NOTIFY_DONE;
1205 }
1206
1207 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1208 {
1209         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1210         struct netdev_notifier_changeupper_info *info;
1211         struct in_device *in_dev;
1212         struct net *net = dev_net(dev);
1213         unsigned int flags;
1214
1215         if (event == NETDEV_UNREGISTER) {
1216                 fib_disable_ip(dev, event, true);
1217                 rt_flush_dev(dev);
1218                 return NOTIFY_DONE;
1219         }
1220
1221         in_dev = __in_dev_get_rtnl(dev);
1222         if (!in_dev)
1223                 return NOTIFY_DONE;
1224
1225         switch (event) {
1226         case NETDEV_UP:
1227                 for_ifa(in_dev) {
1228                         fib_add_ifaddr(ifa);
1229                 } endfor_ifa(in_dev);
1230 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1231                 fib_sync_up(dev, RTNH_F_DEAD);
1232 #endif
1233                 atomic_inc(&net->ipv4.dev_addr_genid);
1234                 rt_cache_flush(net);
1235                 break;
1236         case NETDEV_DOWN:
1237                 fib_disable_ip(dev, event, false);
1238                 break;
1239         case NETDEV_CHANGE:
1240                 flags = dev_get_flags(dev);
1241                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1242                         fib_sync_up(dev, RTNH_F_LINKDOWN);
1243                 else
1244                         fib_sync_down_dev(dev, event, false);
1245                 /* fall through */
1246         case NETDEV_CHANGEMTU:
1247                 rt_cache_flush(net);
1248                 break;
1249         case NETDEV_CHANGEUPPER:
1250                 info = ptr;
1251                 /* flush all routes if dev is linked to or unlinked from
1252                  * an L3 master device (e.g., VRF)
1253                  */
1254                 if (info->upper_dev && netif_is_l3_master(info->upper_dev))
1255                         fib_disable_ip(dev, NETDEV_DOWN, true);
1256                 break;
1257         }
1258         return NOTIFY_DONE;
1259 }
1260
1261 static struct notifier_block fib_inetaddr_notifier = {
1262         .notifier_call = fib_inetaddr_event,
1263 };
1264
1265 static struct notifier_block fib_netdev_notifier = {
1266         .notifier_call = fib_netdev_event,
1267 };
1268
1269 static int __net_init ip_fib_net_init(struct net *net)
1270 {
1271         int err;
1272         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1273
1274         err = fib4_notifier_init(net);
1275         if (err)
1276                 return err;
1277
1278         /* Avoid false sharing : Use at least a full cache line */
1279         size = max_t(size_t, size, L1_CACHE_BYTES);
1280
1281         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1282         if (!net->ipv4.fib_table_hash) {
1283                 err = -ENOMEM;
1284                 goto err_table_hash_alloc;
1285         }
1286
1287         err = fib4_rules_init(net);
1288         if (err < 0)
1289                 goto err_rules_init;
1290         return 0;
1291
1292 err_rules_init:
1293         kfree(net->ipv4.fib_table_hash);
1294 err_table_hash_alloc:
1295         fib4_notifier_exit(net);
1296         return err;
1297 }
1298
1299 static void ip_fib_net_exit(struct net *net)
1300 {
1301         int i;
1302
1303         rtnl_lock();
1304 #ifdef CONFIG_IP_MULTIPLE_TABLES
1305         RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1306         RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1307 #endif
1308         /* Destroy the tables in reverse order to guarantee that the
1309          * local table, ID 255, is destroyed before the main table, ID
1310          * 254. This is necessary as the local table may contain
1311          * references to data contained in the main table.
1312          */
1313         for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1314                 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1315                 struct hlist_node *tmp;
1316                 struct fib_table *tb;
1317
1318                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1319                         hlist_del(&tb->tb_hlist);
1320                         fib_table_flush(net, tb);
1321                         fib_free_table(tb);
1322                 }
1323         }
1324
1325 #ifdef CONFIG_IP_MULTIPLE_TABLES
1326         fib4_rules_exit(net);
1327 #endif
1328         rtnl_unlock();
1329         kfree(net->ipv4.fib_table_hash);
1330         fib4_notifier_exit(net);
1331 }
1332
1333 static int __net_init fib_net_init(struct net *net)
1334 {
1335         int error;
1336
1337 #ifdef CONFIG_IP_ROUTE_CLASSID
1338         net->ipv4.fib_num_tclassid_users = 0;
1339 #endif
1340         error = ip_fib_net_init(net);
1341         if (error < 0)
1342                 goto out;
1343         error = nl_fib_lookup_init(net);
1344         if (error < 0)
1345                 goto out_nlfl;
1346         error = fib_proc_init(net);
1347         if (error < 0)
1348                 goto out_proc;
1349 out:
1350         return error;
1351
1352 out_proc:
1353         nl_fib_lookup_exit(net);
1354 out_nlfl:
1355         ip_fib_net_exit(net);
1356         goto out;
1357 }
1358
1359 static void __net_exit fib_net_exit(struct net *net)
1360 {
1361         fib_proc_exit(net);
1362         nl_fib_lookup_exit(net);
1363         ip_fib_net_exit(net);
1364 }
1365
1366 static struct pernet_operations fib_net_ops = {
1367         .init = fib_net_init,
1368         .exit = fib_net_exit,
1369 };
1370
1371 void __init ip_fib_init(void)
1372 {
1373         fib_trie_init();
1374
1375         register_pernet_subsys(&fib_net_ops);
1376
1377         register_netdevice_notifier(&fib_netdev_notifier);
1378         register_inetaddr_notifier(&fib_inetaddr_notifier);
1379
1380         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1381         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1382         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1383 }