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         struct net *net = dev_net(dev);
 330         struct flow_keys flkeys;
 331         int ret, no_addr;
 332         struct fib_result res;
 333         struct flowi4 fl4;
 334         bool dev_match;
 335
 336         fl4.flowi4_oif = 0;
 337         fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
 338         if (!fl4.flowi4_iif)
 339                 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
 340         fl4.daddr = src;
 341         fl4.saddr = dst;
 342         fl4.flowi4_tos = tos;
 343         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 344         fl4.flowi4_tun_key.tun_id = 0;
 345         fl4.flowi4_flags = 0;
 346         fl4.flowi4_uid = sock_net_uid(net, NULL);
 347
 348         no_addr = idev->ifa_list == NULL;
 349
 350         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
 351         if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
 352                 fl4.flowi4_proto = 0;
 353                 fl4.fl4_sport = 0;
 354                 fl4.fl4_dport = 0;
 355         }
 356
 357         trace_fib_validate_source(dev, &fl4);
 358
 359         if (fib_lookup(net, &fl4, &res, 0))
 360                 goto last_resort;
 361         if (res.type != RTN_UNICAST &&
 362             (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
 363                 goto e_inval;
 364         fib_combine_itag(itag, &res);
 365         dev_match = false;
 366
 367 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 368         for (ret = 0; ret < res.fi->fib_nhs; ret++) {
 369                 struct fib_nh *nh = &res.fi->fib_nh[ret];
 370
 371                 if (nh->nh_dev == dev) {
 372                         dev_match = true;
 373                         break;
 374                 } else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) {
 375                         dev_match = true;
 376                         break;
 377                 }
 378         }
 379 #else
 380         if (FIB_RES_DEV(res) == dev)
 381                 dev_match = true;
 382 #endif
 383         if (dev_match) {
 384                 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 385                 return ret;
 386         }
 387         if (no_addr)
 388                 goto last_resort;
 389         if (rpf == 1)
 390                 goto e_rpf;
 391         fl4.flowi4_oif = dev->ifindex;
 392
 393         ret = 0;
 394         if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
 395                 if (res.type == RTN_UNICAST)
 396                         ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 397         }
 398         return ret;
 399
 400 last_resort:
 401         if (rpf)
 402                 goto e_rpf;
 403         *itag = 0;
 404         return 0;
 405
 406 e_inval:
 407         return -EINVAL;
 408 e_rpf:
 409         return -EXDEV;
 410 }
 411
 412 /* Ignore rp_filter for packets protected by IPsec. */
 413 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 414                         u8 tos, int oif, struct net_device *dev,
 415                         struct in_device *idev, u32 *itag)
 416 {
 417         int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
 418         struct net *net = dev_net(dev);
 419
 420         if (!r && !fib_num_tclassid_users(net) &&
 421             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
 422                 if (IN_DEV_ACCEPT_LOCAL(idev))
 423                         goto ok;
 424                 /* with custom local routes in place, checking local addresses
 425                  * only will be too optimistic, with custom rules, checking
 426                  * local addresses only can be too strict, e.g. due to vrf
 427                  */
 428                 if (net->ipv4.fib_has_custom_local_routes ||
 429                     fib4_has_custom_rules(net))
 430                         goto full_check;
 431                 if (inet_lookup_ifaddr_rcu(net, src))
 432                         return -EINVAL;
 433
 434 ok:
 435                 *itag = 0;
 436                 return 0;
 437         }
 438
 439 full_check:
 440         return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
 441 }
 442
 443 static inline __be32 sk_extract_addr(struct sockaddr *addr)
 444 {
 445         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 446 }
 447
 448 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
 449 {
 450         struct nlattr *nla;
 451
 452         nla = (struct nlattr *) ((char *) mx + len);
 453         nla->nla_type = type;
 454         nla->nla_len = nla_attr_size(4);
 455         *(u32 *) nla_data(nla) = value;
 456
 457         return len + nla_total_size(4);
 458 }
 459
 460 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
 461                                  struct fib_config *cfg)
 462 {
 463         __be32 addr;
 464         int plen;
 465
 466         memset(cfg, 0, sizeof(*cfg));
 467         cfg->fc_nlinfo.nl_net = net;
 468
 469         if (rt->rt_dst.sa_family != AF_INET)
 470                 return -EAFNOSUPPORT;
 471
 472         /*
 473          * Check mask for validity:
 474          * a) it must be contiguous.
 475          * b) destination must have all host bits clear.
 476          * c) if application forgot to set correct family (AF_INET),
 477          *    reject request unless it is absolutely clear i.e.
 478          *    both family and mask are zero.
 479          */
 480         plen = 32;
 481         addr = sk_extract_addr(&rt->rt_dst);
 482         if (!(rt->rt_flags & RTF_HOST)) {
 483                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
 484
 485                 if (rt->rt_genmask.sa_family != AF_INET) {
 486                         if (mask || rt->rt_genmask.sa_family)
 487                                 return -EAFNOSUPPORT;
 488                 }
 489
 490                 if (bad_mask(mask, addr))
 491                         return -EINVAL;
 492
 493                 plen = inet_mask_len(mask);
 494         }
 495
 496         cfg->fc_dst_len = plen;
 497         cfg->fc_dst = addr;
 498
 499         if (cmd != SIOCDELRT) {
 500                 cfg->fc_nlflags = NLM_F_CREATE;
 501                 cfg->fc_protocol = RTPROT_BOOT;
 502         }
 503
 504         if (rt->rt_metric)
 505                 cfg->fc_priority = rt->rt_metric - 1;
 506
 507         if (rt->rt_flags & RTF_REJECT) {
 508                 cfg->fc_scope = RT_SCOPE_HOST;
 509                 cfg->fc_type = RTN_UNREACHABLE;
 510                 return 0;
 511         }
 512
 513         cfg->fc_scope = RT_SCOPE_NOWHERE;
 514         cfg->fc_type = RTN_UNICAST;
 515
 516         if (rt->rt_dev) {
 517                 char *colon;
 518                 struct net_device *dev;
 519                 char devname[IFNAMSIZ];
 520
 521                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
 522                         return -EFAULT;
 523
 524                 devname[IFNAMSIZ-1] = 0;
 525                 colon = strchr(devname, ':');
 526                 if (colon)
 527                         *colon = 0;
 528                 dev = __dev_get_by_name(net, devname);
 529                 if (!dev)
 530                         return -ENODEV;
 531                 cfg->fc_oif = dev->ifindex;
 532                 cfg->fc_table = l3mdev_fib_table(dev);
 533                 if (colon) {
 534                         struct in_ifaddr *ifa;
 535                         struct in_device *in_dev = __in_dev_get_rtnl(dev);
 536                         if (!in_dev)
 537                                 return -ENODEV;
 538                         *colon = ':';
 539                         for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
 540                                 if (strcmp(ifa->ifa_label, devname) == 0)
 541                                         break;
 542                         if (!ifa)
 543                                 return -ENODEV;
 544                         cfg->fc_prefsrc = ifa->ifa_local;
 545                 }
 546         }
 547
 548         addr = sk_extract_addr(&rt->rt_gateway);
 549         if (rt->rt_gateway.sa_family == AF_INET && addr) {
 550                 unsigned int addr_type;
 551
 552                 cfg->fc_gw = addr;
 553                 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
 554                 if (rt->rt_flags & RTF_GATEWAY &&
 555                     addr_type == RTN_UNICAST)
 556                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
 557         }
 558
 559         if (cmd == SIOCDELRT)
 560                 return 0;
 561
 562         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
 563                 return -EINVAL;
 564
 565         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 566                 cfg->fc_scope = RT_SCOPE_LINK;
 567
 568         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 569                 struct nlattr *mx;
 570                 int len = 0;
 571
 572                 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
 573                 if (!mx)
 574                         return -ENOMEM;
 575
 576                 if (rt->rt_flags & RTF_MTU)
 577                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 578
 579                 if (rt->rt_flags & RTF_WINDOW)
 580                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 581
 582                 if (rt->rt_flags & RTF_IRTT)
 583                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 584
 585                 cfg->fc_mx = mx;
 586                 cfg->fc_mx_len = len;
 587         }
 588
 589         return 0;
 590 }
 591
 592 /*
 593  * Handle IP routing ioctl calls.
 594  * These are used to manipulate the routing tables
 595  */
 596 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
 597 {
 598         struct fib_config cfg;
 599         int err;
 600
 601         switch (cmd) {
 602         case SIOCADDRT:         /* Add a route */
 603         case SIOCDELRT:         /* Delete a route */
 604                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 605                         return -EPERM;
 606
 607                 rtnl_lock();
 608                 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
 609                 if (err == 0) {
 610                         struct fib_table *tb;
 611
 612                         if (cmd == SIOCDELRT) {
 613                                 tb = fib_get_table(net, cfg.fc_table);
 614                                 if (tb)
 615                                         err = fib_table_delete(net, tb, &cfg,
 616                                                                NULL);
 617                                 else
 618                                         err = -ESRCH;
 619                         } else {
 620                                 tb = fib_new_table(net, cfg.fc_table);
 621                                 if (tb)
 622                                         err = fib_table_insert(net, tb,
 623                                                                &cfg, NULL);
 624                                 else
 625                                         err = -ENOBUFS;
 626                         }
 627
 628                         /* allocated by rtentry_to_fib_config() */
 629                         kfree(cfg.fc_mx);
 630                 }
 631                 rtnl_unlock();
 632                 return err;
 633         }
 634         return -EINVAL;
 635 }
 636
 637 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 638         [RTA_DST]               = { .type = NLA_U32 },
 639         [RTA_SRC]               = { .type = NLA_U32 },
 640         [RTA_IIF]               = { .type = NLA_U32 },
 641         [RTA_OIF]               = { .type = NLA_U32 },
 642         [RTA_GATEWAY]           = { .type = NLA_U32 },
 643         [RTA_PRIORITY]          = { .type = NLA_U32 },
 644         [RTA_PREFSRC]           = { .type = NLA_U32 },
 645         [RTA_METRICS]           = { .type = NLA_NESTED },
 646         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 647         [RTA_FLOW]              = { .type = NLA_U32 },
 648         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
 649         [RTA_ENCAP]             = { .type = NLA_NESTED },
 650         [RTA_UID]               = { .type = NLA_U32 },
 651         [RTA_MARK]              = { .type = NLA_U32 },
 652         [RTA_TABLE]             = { .type = NLA_U32 },
 653 };
 654
 655 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 656                              struct nlmsghdr *nlh, struct fib_config *cfg,
 657                              struct netlink_ext_ack *extack)
 658 {
 659         struct nlattr *attr;
 660         int err, remaining;
 661         struct rtmsg *rtm;
 662
 663         err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy,
 664                              extack);
 665         if (err < 0)
 666                 goto errout;
 667
 668         memset(cfg, 0, sizeof(*cfg));
 669
 670         rtm = nlmsg_data(nlh);
 671         cfg->fc_dst_len = rtm->rtm_dst_len;
 672         cfg->fc_tos = rtm->rtm_tos;
 673         cfg->fc_table = rtm->rtm_table;
 674         cfg->fc_protocol = rtm->rtm_protocol;
 675         cfg->fc_scope = rtm->rtm_scope;
 676         cfg->fc_type = rtm->rtm_type;
 677         cfg->fc_flags = rtm->rtm_flags;
 678         cfg->fc_nlflags = nlh->nlmsg_flags;
 679
 680         cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
 681         cfg->fc_nlinfo.nlh = nlh;
 682         cfg->fc_nlinfo.nl_net = net;
 683
 684         if (cfg->fc_type > RTN_MAX) {
 685                 NL_SET_ERR_MSG(extack, "Invalid route type");
 686                 err = -EINVAL;
 687                 goto errout;
 688         }
 689
 690         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 691                 switch (nla_type(attr)) {
 692                 case RTA_DST:
 693                         cfg->fc_dst = nla_get_be32(attr);
 694                         break;
 695                 case RTA_OIF:
 696                         cfg->fc_oif = nla_get_u32(attr);
 697                         break;
 698                 case RTA_GATEWAY:
 699                         cfg->fc_gw = nla_get_be32(attr);
 700                         break;
 701                 case RTA_PRIORITY:
 702                         cfg->fc_priority = nla_get_u32(attr);
 703                         break;
 704                 case RTA_PREFSRC:
 705                         cfg->fc_prefsrc = nla_get_be32(attr);
 706                         break;
 707                 case RTA_METRICS:
 708                         cfg->fc_mx = nla_data(attr);
 709                         cfg->fc_mx_len = nla_len(attr);
 710                         break;
 711                 case RTA_MULTIPATH:
 712                         err = lwtunnel_valid_encap_type_attr(nla_data(attr),
 713                                                              nla_len(attr),
 714                                                              extack);
 715                         if (err < 0)
 716                                 goto errout;
 717                         cfg->fc_mp = nla_data(attr);
 718                         cfg->fc_mp_len = nla_len(attr);
 719                         break;
 720                 case RTA_FLOW:
 721                         cfg->fc_flow = nla_get_u32(attr);
 722                         break;
 723                 case RTA_TABLE:
 724                         cfg->fc_table = nla_get_u32(attr);
 725                         break;
 726                 case RTA_ENCAP:
 727                         cfg->fc_encap = attr;
 728                         break;
 729                 case RTA_ENCAP_TYPE:
 730                         cfg->fc_encap_type = nla_get_u16(attr);
 731                         err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
 732                                                         extack);
 733                         if (err < 0)
 734                                 goto errout;
 735                         break;
 736                 }
 737         }
 738
 739         return 0;
 740 errout:
 741         return err;
 742 }
 743
 744 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 745                              struct netlink_ext_ack *extack)
 746 {
 747         struct net *net = sock_net(skb->sk);
 748         struct fib_config cfg;
 749         struct fib_table *tb;
 750         int err;
 751
 752         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 753         if (err < 0)
 754                 goto errout;
 755
 756         tb = fib_get_table(net, cfg.fc_table);
 757         if (!tb) {
 758                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
 759                 err = -ESRCH;
 760                 goto errout;
 761         }
 762
 763         err = fib_table_delete(net, tb, &cfg, extack);
 764 errout:
 765         return err;
 766 }
 767
 768 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 769                              struct netlink_ext_ack *extack)
 770 {
 771         struct net *net = sock_net(skb->sk);
 772         struct fib_config cfg;
 773         struct fib_table *tb;
 774         int err;
 775
 776         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 777         if (err < 0)
 778                 goto errout;
 779
 780         tb = fib_new_table(net, cfg.fc_table);
 781         if (!tb) {
 782                 err = -ENOBUFS;
 783                 goto errout;
 784         }
 785
 786         err = fib_table_insert(net, tb, &cfg, extack);
 787         if (!err && cfg.fc_type == RTN_LOCAL)
 788                 net->ipv4.fib_has_custom_local_routes = true;
 789 errout:
 790         return err;
 791 }
 792
 793 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 794 {
 795         struct net *net = sock_net(skb->sk);
 796         unsigned int h, s_h;
 797         unsigned int e = 0, s_e;
 798         struct fib_table *tb;
 799         struct hlist_head *head;
 800         int dumped = 0, err;
 801
 802         if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
 803             ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
 804                 return skb->len;
 805
 806         s_h = cb->args[0];
 807         s_e = cb->args[1];
 808
 809         rcu_read_lock();
 810
 811         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
 812                 e = 0;
 813                 head = &net->ipv4.fib_table_hash[h];
 814                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
 815                         if (e < s_e)
 816                                 goto next;
 817                         if (dumped)
 818                                 memset(&cb->args[2], 0, sizeof(cb->args) -
 819                                                  2 * sizeof(cb->args[0]));
 820                         err = fib_table_dump(tb, skb, cb);
 821                         if (err < 0) {
 822                                 if (likely(skb->len))
 823                                         goto out;
 824
 825                                 goto out_err;
 826                         }
 827                         dumped = 1;
 828 next:
 829                         e++;
 830                 }
 831         }
 832 out:
 833         err = skb->len;
 834 out_err:
 835         rcu_read_unlock();
 836
 837         cb->args[1] = e;
 838         cb->args[0] = h;
 839
 840         return err;
 841 }
 842
 843 /* Prepare and feed intra-kernel routing request.
 844  * Really, it should be netlink message, but :-( netlink
 845  * can be not configured, so that we feed it directly
 846  * to fib engine. It is legal, because all events occur
 847  * only when netlink is already locked.
 848  */
 849 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
 850 {
 851         struct net *net = dev_net(ifa->ifa_dev->dev);
 852         u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
 853         struct fib_table *tb;
 854         struct fib_config cfg = {
 855                 .fc_protocol = RTPROT_KERNEL,
 856                 .fc_type = type,
 857                 .fc_dst = dst,
 858                 .fc_dst_len = dst_len,
 859                 .fc_prefsrc = ifa->ifa_local,
 860                 .fc_oif = ifa->ifa_dev->dev->ifindex,
 861                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
 862                 .fc_nlinfo = {
 863                         .nl_net = net,
 864                 },
 865         };
 866
 867         if (!tb_id)
 868                 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
 869
 870         tb = fib_new_table(net, tb_id);
 871         if (!tb)
 872                 return;
 873
 874         cfg.fc_table = tb->tb_id;
 875
 876         if (type != RTN_LOCAL)
 877                 cfg.fc_scope = RT_SCOPE_LINK;
 878         else
 879                 cfg.fc_scope = RT_SCOPE_HOST;
 880
 881         if (cmd == RTM_NEWROUTE)
 882                 fib_table_insert(net, tb, &cfg, NULL);
 883         else
 884                 fib_table_delete(net, tb, &cfg, NULL);
 885 }
 886
 887 void fib_add_ifaddr(struct in_ifaddr *ifa)
 888 {
 889         struct in_device *in_dev = ifa->ifa_dev;
 890         struct net_device *dev = in_dev->dev;
 891         struct in_ifaddr *prim = ifa;
 892         __be32 mask = ifa->ifa_mask;
 893         __be32 addr = ifa->ifa_local;
 894         __be32 prefix = ifa->ifa_address & mask;
 895
 896         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 897                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
 898                 if (!prim) {
 899                         pr_warn("%s: bug: prim == NULL\n", __func__);
 900                         return;
 901                 }
 902         }
 903
 904         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
 905
 906         if (!(dev->flags & IFF_UP))
 907                 return;
 908
 909         /* Add broadcast address, if it is explicitly assigned. */
 910         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
 911                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
 912
 913         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
 914             (prefix != addr || ifa->ifa_prefixlen < 32)) {
 915                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
 916                         fib_magic(RTM_NEWROUTE,
 917                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 918                                   prefix, ifa->ifa_prefixlen, prim);
 919
 920                 /* Add network specific broadcasts, when it takes a sense */
 921                 if (ifa->ifa_prefixlen < 31) {
 922                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
 923                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
 924                                   32, prim);
 925                 }
 926         }
 927 }
 928
 929 /* Delete primary or secondary address.
 930  * Optionally, on secondary address promotion consider the addresses
 931  * from subnet iprim as deleted, even if they are in device list.
 932  * In this case the secondary ifa can be in device list.
 933  */
 934 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
 935 {
 936         struct in_device *in_dev = ifa->ifa_dev;
 937         struct net_device *dev = in_dev->dev;
 938         struct in_ifaddr *ifa1;
 939         struct in_ifaddr *prim = ifa, *prim1 = NULL;
 940         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
 941         __be32 any = ifa->ifa_address & ifa->ifa_mask;
 942 #define LOCAL_OK        1
 943 #define BRD_OK          2
 944 #define BRD0_OK         4
 945 #define BRD1_OK         8
 946         unsigned int ok = 0;
 947         int subnet = 0;         /* Primary network */
 948         int gone = 1;           /* Address is missing */
 949         int same_prefsrc = 0;   /* Another primary with same IP */
 950
 951         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 952                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
 953                 if (!prim) {
 954                         /* if the device has been deleted, we don't perform
 955                          * address promotion
 956                          */
 957                         if (!in_dev->dead)
 958                                 pr_warn("%s: bug: prim == NULL\n", __func__);
 959                         return;
 960                 }
 961                 if (iprim && iprim != prim) {
 962                         pr_warn("%s: bug: iprim != prim\n", __func__);
 963                         return;
 964                 }
 965         } else if (!ipv4_is_zeronet(any) &&
 966                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
 967                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
 968                         fib_magic(RTM_DELROUTE,
 969                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 970                                   any, ifa->ifa_prefixlen, prim);
 971                 subnet = 1;
 972         }
 973
 974         if (in_dev->dead)
 975                 goto no_promotions;
 976
 977         /* Deletion is more complicated than add.
 978          * We should take care of not to delete too much :-)
 979          *
 980          * Scan address list to be sure that addresses are really gone.
 981          */
 982
 983         for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
 984                 if (ifa1 == ifa) {
 985                         /* promotion, keep the IP */
 986                         gone = 0;
 987                         continue;
 988                 }
 989                 /* Ignore IFAs from our subnet */
 990                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
 991                     inet_ifa_match(ifa1->ifa_address, iprim))
 992                         continue;
 993
 994                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
 995                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
 996                         /* Another address from our subnet? */
 997                         if (ifa1->ifa_mask == prim->ifa_mask &&
 998                             inet_ifa_match(ifa1->ifa_address, prim))
 999                                 prim1 = prim;
1000                         else {
1001                                 /* We reached the secondaries, so
1002                                  * same_prefsrc should be determined.
1003                                  */
1004                                 if (!same_prefsrc)
1005                                         continue;
1006                                 /* Search new prim1 if ifa1 is not
1007                                  * using the current prim1
1008                                  */
1009                                 if (!prim1 ||
1010                                     ifa1->ifa_mask != prim1->ifa_mask ||
1011                                     !inet_ifa_match(ifa1->ifa_address, prim1))
1012                                         prim1 = inet_ifa_byprefix(in_dev,
1013                                                         ifa1->ifa_address,
1014                                                         ifa1->ifa_mask);
1015                                 if (!prim1)
1016                                         continue;
1017                                 if (prim1->ifa_local != prim->ifa_local)
1018                                         continue;
1019                         }
1020                 } else {
1021                         if (prim->ifa_local != ifa1->ifa_local)
1022                                 continue;
1023                         prim1 = ifa1;
1024                         if (prim != prim1)
1025                                 same_prefsrc = 1;
1026                 }
1027                 if (ifa->ifa_local == ifa1->ifa_local)
1028                         ok |= LOCAL_OK;
1029                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1030                         ok |= BRD_OK;
1031                 if (brd == ifa1->ifa_broadcast)
1032                         ok |= BRD1_OK;
1033                 if (any == ifa1->ifa_broadcast)
1034                         ok |= BRD0_OK;
1035                 /* primary has network specific broadcasts */
1036                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1037                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1038                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1039
1040                         if (!ipv4_is_zeronet(any1)) {
1041                                 if (ifa->ifa_broadcast == brd1 ||
1042                                     ifa->ifa_broadcast == any1)
1043                                         ok |= BRD_OK;
1044                                 if (brd == brd1 || brd == any1)
1045                                         ok |= BRD1_OK;
1046                                 if (any == brd1 || any == any1)
1047                                         ok |= BRD0_OK;
1048                         }
1049                 }
1050         }
1051
1052 no_promotions:
1053         if (!(ok & BRD_OK))
1054                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
1055         if (subnet && ifa->ifa_prefixlen < 31) {
1056                 if (!(ok & BRD1_OK))
1057                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
1058                 if (!(ok & BRD0_OK))
1059                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
1060         }
1061         if (!(ok & LOCAL_OK)) {
1062                 unsigned int addr_type;
1063
1064                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
1065
1066                 /* Check, that this local address finally disappeared. */
1067                 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1068                                                      ifa->ifa_local);
1069                 if (gone && addr_type != RTN_LOCAL) {
1070                         /* And the last, but not the least thing.
1071                          * We must flush stray FIB entries.
1072                          *
1073                          * First of all, we scan fib_info list searching
1074                          * for stray nexthop entries, then ignite fib_flush.
1075                          */
1076                         if (fib_sync_down_addr(dev, ifa->ifa_local))
1077                                 fib_flush(dev_net(dev));
1078                 }
1079         }
1080 #undef LOCAL_OK
1081 #undef BRD_OK
1082 #undef BRD0_OK
1083 #undef BRD1_OK
1084 }
1085
1086 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1087 {
1088
1089         struct fib_result       res;
1090         struct flowi4           fl4 = {
1091                 .flowi4_mark = frn->fl_mark,
1092                 .daddr = frn->fl_addr,
1093                 .flowi4_tos = frn->fl_tos,
1094                 .flowi4_scope = frn->fl_scope,
1095         };
1096         struct fib_table *tb;
1097
1098         rcu_read_lock();
1099
1100         tb = fib_get_table(net, frn->tb_id_in);
1101
1102         frn->err = -ENOENT;
1103         if (tb) {
1104                 local_bh_disable();
1105
1106                 frn->tb_id = tb->tb_id;
1107                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1108
1109                 if (!frn->err) {
1110                         frn->prefixlen = res.prefixlen;
1111                         frn->nh_sel = res.nh_sel;
1112                         frn->type = res.type;
1113                         frn->scope = res.scope;
1114                 }
1115                 local_bh_enable();
1116         }
1117
1118         rcu_read_unlock();
1119 }
1120
1121 static void nl_fib_input(struct sk_buff *skb)
1122 {
1123         struct net *net;
1124         struct fib_result_nl *frn;
1125         struct nlmsghdr *nlh;
1126         u32 portid;
1127
1128         net = sock_net(skb->sk);
1129         nlh = nlmsg_hdr(skb);
1130         if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1131             skb->len < nlh->nlmsg_len ||
1132             nlmsg_len(nlh) < sizeof(*frn))
1133                 return;
1134
1135         skb = netlink_skb_clone(skb, GFP_KERNEL);
1136         if (!skb)
1137                 return;
1138         nlh = nlmsg_hdr(skb);
1139
1140         frn = (struct fib_result_nl *) nlmsg_data(nlh);
1141         nl_fib_lookup(net, frn);
1142
1143         portid = NETLINK_CB(skb).portid;      /* netlink portid */
1144         NETLINK_CB(skb).portid = 0;        /* from kernel */
1145         NETLINK_CB(skb).dst_group = 0;  /* unicast */
1146         netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1147 }
1148
1149 static int __net_init nl_fib_lookup_init(struct net *net)
1150 {
1151         struct sock *sk;
1152         struct netlink_kernel_cfg cfg = {
1153                 .input  = nl_fib_input,
1154         };
1155
1156         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1157         if (!sk)
1158                 return -EAFNOSUPPORT;
1159         net->ipv4.fibnl = sk;
1160         return 0;
1161 }
1162
1163 static void nl_fib_lookup_exit(struct net *net)
1164 {
1165         netlink_kernel_release(net->ipv4.fibnl);
1166         net->ipv4.fibnl = NULL;
1167 }
1168
1169 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1170                            bool force)
1171 {
1172         if (fib_sync_down_dev(dev, event, force))
1173                 fib_flush(dev_net(dev));
1174         else
1175                 rt_cache_flush(dev_net(dev));
1176         arp_ifdown(dev);
1177 }
1178
1179 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1180 {
1181         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1182         struct net_device *dev = ifa->ifa_dev->dev;
1183         struct net *net = dev_net(dev);
1184
1185         switch (event) {
1186         case NETDEV_UP:
1187                 fib_add_ifaddr(ifa);
1188 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1189                 fib_sync_up(dev, RTNH_F_DEAD);
1190 #endif
1191                 atomic_inc(&net->ipv4.dev_addr_genid);
1192                 rt_cache_flush(dev_net(dev));
1193                 break;
1194         case NETDEV_DOWN:
1195                 fib_del_ifaddr(ifa, NULL);
1196                 atomic_inc(&net->ipv4.dev_addr_genid);
1197                 if (!ifa->ifa_dev->ifa_list) {
1198                         /* Last address was deleted from this interface.
1199                          * Disable IP.
1200                          */
1201                         fib_disable_ip(dev, event, true);
1202                 } else {
1203                         rt_cache_flush(dev_net(dev));
1204                 }
1205                 break;
1206         }
1207         return NOTIFY_DONE;
1208 }
1209
1210 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1211 {
1212         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1213         struct netdev_notifier_changeupper_info *info;
1214         struct in_device *in_dev;
1215         struct net *net = dev_net(dev);
1216         unsigned int flags;
1217
1218         if (event == NETDEV_UNREGISTER) {
1219                 fib_disable_ip(dev, event, true);
1220                 rt_flush_dev(dev);
1221                 return NOTIFY_DONE;
1222         }
1223
1224         in_dev = __in_dev_get_rtnl(dev);
1225         if (!in_dev)
1226                 return NOTIFY_DONE;
1227
1228         switch (event) {
1229         case NETDEV_UP:
1230                 for_ifa(in_dev) {
1231                         fib_add_ifaddr(ifa);
1232                 } endfor_ifa(in_dev);
1233 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1234                 fib_sync_up(dev, RTNH_F_DEAD);
1235 #endif
1236                 atomic_inc(&net->ipv4.dev_addr_genid);
1237                 rt_cache_flush(net);
1238                 break;
1239         case NETDEV_DOWN:
1240                 fib_disable_ip(dev, event, false);
1241                 break;
1242         case NETDEV_CHANGE:
1243                 flags = dev_get_flags(dev);
1244                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1245                         fib_sync_up(dev, RTNH_F_LINKDOWN);
1246                 else
1247                         fib_sync_down_dev(dev, event, false);
1248                 /* fall through */
1249         case NETDEV_CHANGEMTU:
1250                 rt_cache_flush(net);
1251                 break;
1252         case NETDEV_CHANGEUPPER:
1253                 info = ptr;
1254                 /* flush all routes if dev is linked to or unlinked from
1255                  * an L3 master device (e.g., VRF)
1256                  */
1257                 if (info->upper_dev && netif_is_l3_master(info->upper_dev))
1258                         fib_disable_ip(dev, NETDEV_DOWN, true);
1259                 break;
1260         }
1261         return NOTIFY_DONE;
1262 }
1263
1264 static struct notifier_block fib_inetaddr_notifier = {
1265         .notifier_call = fib_inetaddr_event,
1266 };
1267
1268 static struct notifier_block fib_netdev_notifier = {
1269         .notifier_call = fib_netdev_event,
1270 };
1271
1272 static int __net_init ip_fib_net_init(struct net *net)
1273 {
1274         int err;
1275         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1276
1277         err = fib4_notifier_init(net);
1278         if (err)
1279                 return err;
1280
1281         /* Avoid false sharing : Use at least a full cache line */
1282         size = max_t(size_t, size, L1_CACHE_BYTES);
1283
1284         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1285         if (!net->ipv4.fib_table_hash) {
1286                 err = -ENOMEM;
1287                 goto err_table_hash_alloc;
1288         }
1289
1290         err = fib4_rules_init(net);
1291         if (err < 0)
1292                 goto err_rules_init;
1293         return 0;
1294
1295 err_rules_init:
1296         kfree(net->ipv4.fib_table_hash);
1297 err_table_hash_alloc:
1298         fib4_notifier_exit(net);
1299         return err;
1300 }
1301
1302 static void ip_fib_net_exit(struct net *net)
1303 {
1304         int i;
1305
1306         rtnl_lock();
1307 #ifdef CONFIG_IP_MULTIPLE_TABLES
1308         RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1309         RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1310 #endif
1311         /* Destroy the tables in reverse order to guarantee that the
1312          * local table, ID 255, is destroyed before the main table, ID
1313          * 254. This is necessary as the local table may contain
1314          * references to data contained in the main table.
1315          */
1316         for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1317                 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1318                 struct hlist_node *tmp;
1319                 struct fib_table *tb;
1320
1321                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1322                         hlist_del(&tb->tb_hlist);
1323                         fib_table_flush(net, tb);
1324                         fib_free_table(tb);
1325                 }
1326         }
1327
1328 #ifdef CONFIG_IP_MULTIPLE_TABLES
1329         fib4_rules_exit(net);
1330 #endif
1331         rtnl_unlock();
1332         kfree(net->ipv4.fib_table_hash);
1333         fib4_notifier_exit(net);
1334 }
1335
1336 static int __net_init fib_net_init(struct net *net)
1337 {
1338         int error;
1339
1340 #ifdef CONFIG_IP_ROUTE_CLASSID
1341         net->ipv4.fib_num_tclassid_users = 0;
1342 #endif
1343         error = ip_fib_net_init(net);
1344         if (error < 0)
1345                 goto out;
1346         error = nl_fib_lookup_init(net);
1347         if (error < 0)
1348                 goto out_nlfl;
1349         error = fib_proc_init(net);
1350         if (error < 0)
1351                 goto out_proc;
1352 out:
1353         return error;
1354
1355 out_proc:
1356         nl_fib_lookup_exit(net);
1357 out_nlfl:
1358         ip_fib_net_exit(net);
1359         goto out;
1360 }
1361
1362 static void __net_exit fib_net_exit(struct net *net)
1363 {
1364         fib_proc_exit(net);
1365         nl_fib_lookup_exit(net);
1366         ip_fib_net_exit(net);
1367 }
1368
1369 static struct pernet_operations fib_net_ops = {
1370         .init = fib_net_init,
1371         .exit = fib_net_exit,
1372 };
1373
1374 void __init ip_fib_init(void)
1375 {
1376         fib_trie_init();
1377
1378         register_pernet_subsys(&fib_net_ops);
1379
1380         register_netdevice_notifier(&fib_netdev_notifier);
1381         register_inetaddr_notifier(&fib_inetaddr_notifier);
1382
1383         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1384         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1385         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1386 }