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, false);
 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
 296         net = dev_net(dev);
 297
 298         scope = RT_SCOPE_UNIVERSE;
 299         if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
 300                 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
 301                 struct flowi4 fl4 = {
 302                         .flowi4_iif = LOOPBACK_IFINDEX,
 303                         .flowi4_oif = l3mdev_master_ifindex_rcu(dev),
 304                         .daddr = ip_hdr(skb)->saddr,
 305                         .flowi4_tos = ip_hdr(skb)->tos & IPTOS_RT_MASK,
 306                         .flowi4_scope = scope,
 307                         .flowi4_mark = vmark ? skb->mark : 0,
 308                 };
 309                 if (!fib_lookup(net, &fl4, &res, 0))
 310                         return FIB_RES_PREFSRC(net, res);
 311         } else {
 312                 scope = RT_SCOPE_LINK;
 313         }
 314
 315         return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
 316 }
 317
 318 /* Given (packet source, input interface) and optional (dst, oif, tos):
 319  * - (main) check, that source is valid i.e. not broadcast or our local
 320  *   address.
 321  * - figure out what "logical" interface this packet arrived
 322  *   and calculate "specific destination" address.
 323  * - check, that packet arrived from expected physical interface.
 324  * called with rcu_read_lock()
 325  */
 326 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 327                                  u8 tos, int oif, struct net_device *dev,
 328                                  int rpf, struct in_device *idev, u32 *itag)
 329 {
 330         struct net *net = dev_net(dev);
 331         struct flow_keys flkeys;
 332         int ret, no_addr;
 333         struct fib_result res;
 334         struct flowi4 fl4;
 335         bool dev_match;
 336
 337         fl4.flowi4_oif = 0;
 338         fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
 339         if (!fl4.flowi4_iif)
 340                 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
 341         fl4.daddr = src;
 342         fl4.saddr = dst;
 343         fl4.flowi4_tos = tos;
 344         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 345         fl4.flowi4_tun_key.tun_id = 0;
 346         fl4.flowi4_flags = 0;
 347         fl4.flowi4_uid = sock_net_uid(net, NULL);
 348
 349         no_addr = idev->ifa_list == NULL;
 350
 351         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
 352         if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
 353                 fl4.flowi4_proto = 0;
 354                 fl4.fl4_sport = 0;
 355                 fl4.fl4_dport = 0;
 356         } else {
 357                 swap(fl4.fl4_sport, fl4.fl4_dport);
 358         }
 359
 360         if (fib_lookup(net, &fl4, &res, 0))
 361                 goto last_resort;
 362         if (res.type != RTN_UNICAST &&
 363             (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
 364                 goto e_inval;
 365         fib_combine_itag(itag, &res);
 366         dev_match = false;
 367
 368 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 369         for (ret = 0; ret < res.fi->fib_nhs; ret++) {
 370                 struct fib_nh *nh = &res.fi->fib_nh[ret];
 371
 372                 if (nh->nh_dev == dev) {
 373                         dev_match = true;
 374                         break;
 375                 } else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) {
 376                         dev_match = true;
 377                         break;
 378                 }
 379         }
 380 #else
 381         if (FIB_RES_DEV(res) == dev)
 382                 dev_match = true;
 383 #endif
 384         if (dev_match) {
 385                 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 386                 return ret;
 387         }
 388         if (no_addr)
 389                 goto last_resort;
 390         if (rpf == 1)
 391                 goto e_rpf;
 392         fl4.flowi4_oif = dev->ifindex;
 393
 394         ret = 0;
 395         if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
 396                 if (res.type == RTN_UNICAST)
 397                         ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 398         }
 399         return ret;
 400
 401 last_resort:
 402         if (rpf)
 403                 goto e_rpf;
 404         *itag = 0;
 405         return 0;
 406
 407 e_inval:
 408         return -EINVAL;
 409 e_rpf:
 410         return -EXDEV;
 411 }
 412
 413 /* Ignore rp_filter for packets protected by IPsec. */
 414 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 415                         u8 tos, int oif, struct net_device *dev,
 416                         struct in_device *idev, u32 *itag)
 417 {
 418         int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
 419         struct net *net = dev_net(dev);
 420
 421         if (!r && !fib_num_tclassid_users(net) &&
 422             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
 423                 if (IN_DEV_ACCEPT_LOCAL(idev))
 424                         goto ok;
 425                 /* with custom local routes in place, checking local addresses
 426                  * only will be too optimistic, with custom rules, checking
 427                  * local addresses only can be too strict, e.g. due to vrf
 428                  */
 429                 if (net->ipv4.fib_has_custom_local_routes ||
 430                     fib4_has_custom_rules(net))
 431                         goto full_check;
 432                 if (inet_lookup_ifaddr_rcu(net, src))
 433                         return -EINVAL;
 434
 435 ok:
 436                 *itag = 0;
 437                 return 0;
 438         }
 439
 440 full_check:
 441         return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
 442 }
 443
 444 static inline __be32 sk_extract_addr(struct sockaddr *addr)
 445 {
 446         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 447 }
 448
 449 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
 450 {
 451         struct nlattr *nla;
 452
 453         nla = (struct nlattr *) ((char *) mx + len);
 454         nla->nla_type = type;
 455         nla->nla_len = nla_attr_size(4);
 456         *(u32 *) nla_data(nla) = value;
 457
 458         return len + nla_total_size(4);
 459 }
 460
 461 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
 462                                  struct fib_config *cfg)
 463 {
 464         __be32 addr;
 465         int plen;
 466
 467         memset(cfg, 0, sizeof(*cfg));
 468         cfg->fc_nlinfo.nl_net = net;
 469
 470         if (rt->rt_dst.sa_family != AF_INET)
 471                 return -EAFNOSUPPORT;
 472
 473         /*
 474          * Check mask for validity:
 475          * a) it must be contiguous.
 476          * b) destination must have all host bits clear.
 477          * c) if application forgot to set correct family (AF_INET),
 478          *    reject request unless it is absolutely clear i.e.
 479          *    both family and mask are zero.
 480          */
 481         plen = 32;
 482         addr = sk_extract_addr(&rt->rt_dst);
 483         if (!(rt->rt_flags & RTF_HOST)) {
 484                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
 485
 486                 if (rt->rt_genmask.sa_family != AF_INET) {
 487                         if (mask || rt->rt_genmask.sa_family)
 488                                 return -EAFNOSUPPORT;
 489                 }
 490
 491                 if (bad_mask(mask, addr))
 492                         return -EINVAL;
 493
 494                 plen = inet_mask_len(mask);
 495         }
 496
 497         cfg->fc_dst_len = plen;
 498         cfg->fc_dst = addr;
 499
 500         if (cmd != SIOCDELRT) {
 501                 cfg->fc_nlflags = NLM_F_CREATE;
 502                 cfg->fc_protocol = RTPROT_BOOT;
 503         }
 504
 505         if (rt->rt_metric)
 506                 cfg->fc_priority = rt->rt_metric - 1;
 507
 508         if (rt->rt_flags & RTF_REJECT) {
 509                 cfg->fc_scope = RT_SCOPE_HOST;
 510                 cfg->fc_type = RTN_UNREACHABLE;
 511                 return 0;
 512         }
 513
 514         cfg->fc_scope = RT_SCOPE_NOWHERE;
 515         cfg->fc_type = RTN_UNICAST;
 516
 517         if (rt->rt_dev) {
 518                 char *colon;
 519                 struct net_device *dev;
 520                 char devname[IFNAMSIZ];
 521
 522                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
 523                         return -EFAULT;
 524
 525                 devname[IFNAMSIZ-1] = 0;
 526                 colon = strchr(devname, ':');
 527                 if (colon)
 528                         *colon = 0;
 529                 dev = __dev_get_by_name(net, devname);
 530                 if (!dev)
 531                         return -ENODEV;
 532                 cfg->fc_oif = dev->ifindex;
 533                 cfg->fc_table = l3mdev_fib_table(dev);
 534                 if (colon) {
 535                         struct in_ifaddr *ifa;
 536                         struct in_device *in_dev = __in_dev_get_rtnl(dev);
 537                         if (!in_dev)
 538                                 return -ENODEV;
 539                         *colon = ':';
 540                         for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
 541                                 if (strcmp(ifa->ifa_label, devname) == 0)
 542                                         break;
 543                         if (!ifa)
 544                                 return -ENODEV;
 545                         cfg->fc_prefsrc = ifa->ifa_local;
 546                 }
 547         }
 548
 549         addr = sk_extract_addr(&rt->rt_gateway);
 550         if (rt->rt_gateway.sa_family == AF_INET && addr) {
 551                 unsigned int addr_type;
 552
 553                 cfg->fc_gw = addr;
 554                 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
 555                 if (rt->rt_flags & RTF_GATEWAY &&
 556                     addr_type == RTN_UNICAST)
 557                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
 558         }
 559
 560         if (cmd == SIOCDELRT)
 561                 return 0;
 562
 563         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
 564                 return -EINVAL;
 565
 566         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 567                 cfg->fc_scope = RT_SCOPE_LINK;
 568
 569         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 570                 struct nlattr *mx;
 571                 int len = 0;
 572
 573                 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
 574                 if (!mx)
 575                         return -ENOMEM;
 576
 577                 if (rt->rt_flags & RTF_MTU)
 578                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 579
 580                 if (rt->rt_flags & RTF_WINDOW)
 581                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 582
 583                 if (rt->rt_flags & RTF_IRTT)
 584                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 585
 586                 cfg->fc_mx = mx;
 587                 cfg->fc_mx_len = len;
 588         }
 589
 590         return 0;
 591 }
 592
 593 /*
 594  * Handle IP routing ioctl calls.
 595  * These are used to manipulate the routing tables
 596  */
 597 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
 598 {
 599         struct fib_config cfg;
 600         int err;
 601
 602         switch (cmd) {
 603         case SIOCADDRT:         /* Add a route */
 604         case SIOCDELRT:         /* Delete a route */
 605                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 606                         return -EPERM;
 607
 608                 rtnl_lock();
 609                 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
 610                 if (err == 0) {
 611                         struct fib_table *tb;
 612
 613                         if (cmd == SIOCDELRT) {
 614                                 tb = fib_get_table(net, cfg.fc_table);
 615                                 if (tb)
 616                                         err = fib_table_delete(net, tb, &cfg,
 617                                                                NULL);
 618                                 else
 619                                         err = -ESRCH;
 620                         } else {
 621                                 tb = fib_new_table(net, cfg.fc_table);
 622                                 if (tb)
 623                                         err = fib_table_insert(net, tb,
 624                                                                &cfg, NULL);
 625                                 else
 626                                         err = -ENOBUFS;
 627                         }
 628
 629                         /* allocated by rtentry_to_fib_config() */
 630                         kfree(cfg.fc_mx);
 631                 }
 632                 rtnl_unlock();
 633                 return err;
 634         }
 635         return -EINVAL;
 636 }
 637
 638 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 639         [RTA_DST]               = { .type = NLA_U32 },
 640         [RTA_SRC]               = { .type = NLA_U32 },
 641         [RTA_IIF]               = { .type = NLA_U32 },
 642         [RTA_OIF]               = { .type = NLA_U32 },
 643         [RTA_GATEWAY]           = { .type = NLA_U32 },
 644         [RTA_PRIORITY]          = { .type = NLA_U32 },
 645         [RTA_PREFSRC]           = { .type = NLA_U32 },
 646         [RTA_METRICS]           = { .type = NLA_NESTED },
 647         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 648         [RTA_FLOW]              = { .type = NLA_U32 },
 649         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
 650         [RTA_ENCAP]             = { .type = NLA_NESTED },
 651         [RTA_UID]               = { .type = NLA_U32 },
 652         [RTA_MARK]              = { .type = NLA_U32 },
 653         [RTA_TABLE]             = { .type = NLA_U32 },
 654         [RTA_IP_PROTO]          = { .type = NLA_U8 },
 655         [RTA_SPORT]             = { .type = NLA_U16 },
 656         [RTA_DPORT]             = { .type = NLA_U16 },
 657 };
 658
 659 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 660                              struct nlmsghdr *nlh, struct fib_config *cfg,
 661                              struct netlink_ext_ack *extack)
 662 {
 663         struct nlattr *attr;
 664         int err, remaining;
 665         struct rtmsg *rtm;
 666
 667         err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy,
 668                              extack);
 669         if (err < 0)
 670                 goto errout;
 671
 672         memset(cfg, 0, sizeof(*cfg));
 673
 674         rtm = nlmsg_data(nlh);
 675         cfg->fc_dst_len = rtm->rtm_dst_len;
 676         cfg->fc_tos = rtm->rtm_tos;
 677         cfg->fc_table = rtm->rtm_table;
 678         cfg->fc_protocol = rtm->rtm_protocol;
 679         cfg->fc_scope = rtm->rtm_scope;
 680         cfg->fc_type = rtm->rtm_type;
 681         cfg->fc_flags = rtm->rtm_flags;
 682         cfg->fc_nlflags = nlh->nlmsg_flags;
 683
 684         cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
 685         cfg->fc_nlinfo.nlh = nlh;
 686         cfg->fc_nlinfo.nl_net = net;
 687
 688         if (cfg->fc_type > RTN_MAX) {
 689                 NL_SET_ERR_MSG(extack, "Invalid route type");
 690                 err = -EINVAL;
 691                 goto errout;
 692         }
 693
 694         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 695                 switch (nla_type(attr)) {
 696                 case RTA_DST:
 697                         cfg->fc_dst = nla_get_be32(attr);
 698                         break;
 699                 case RTA_OIF:
 700                         cfg->fc_oif = nla_get_u32(attr);
 701                         break;
 702                 case RTA_GATEWAY:
 703                         cfg->fc_gw = nla_get_be32(attr);
 704                         break;
 705                 case RTA_VIA:
 706                         NL_SET_ERR_MSG(extack, "IPv4 does not support RTA_VIA attribute");
 707                         err = -EINVAL;
 708                         goto errout;
 709                 case RTA_PRIORITY:
 710                         cfg->fc_priority = nla_get_u32(attr);
 711                         break;
 712                 case RTA_PREFSRC:
 713                         cfg->fc_prefsrc = nla_get_be32(attr);
 714                         break;
 715                 case RTA_METRICS:
 716                         cfg->fc_mx = nla_data(attr);
 717                         cfg->fc_mx_len = nla_len(attr);
 718                         break;
 719                 case RTA_MULTIPATH:
 720                         err = lwtunnel_valid_encap_type_attr(nla_data(attr),
 721                                                              nla_len(attr),
 722                                                              extack);
 723                         if (err < 0)
 724                                 goto errout;
 725                         cfg->fc_mp = nla_data(attr);
 726                         cfg->fc_mp_len = nla_len(attr);
 727                         break;
 728                 case RTA_FLOW:
 729                         cfg->fc_flow = nla_get_u32(attr);
 730                         break;
 731                 case RTA_TABLE:
 732                         cfg->fc_table = nla_get_u32(attr);
 733                         break;
 734                 case RTA_ENCAP:
 735                         cfg->fc_encap = attr;
 736                         break;
 737                 case RTA_ENCAP_TYPE:
 738                         cfg->fc_encap_type = nla_get_u16(attr);
 739                         err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
 740                                                         extack);
 741                         if (err < 0)
 742                                 goto errout;
 743                         break;
 744                 }
 745         }
 746
 747         return 0;
 748 errout:
 749         return err;
 750 }
 751
 752 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 753                              struct netlink_ext_ack *extack)
 754 {
 755         struct net *net = sock_net(skb->sk);
 756         struct fib_config cfg;
 757         struct fib_table *tb;
 758         int err;
 759
 760         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 761         if (err < 0)
 762                 goto errout;
 763
 764         tb = fib_get_table(net, cfg.fc_table);
 765         if (!tb) {
 766                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
 767                 err = -ESRCH;
 768                 goto errout;
 769         }
 770
 771         err = fib_table_delete(net, tb, &cfg, extack);
 772 errout:
 773         return err;
 774 }
 775
 776 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 777                              struct netlink_ext_ack *extack)
 778 {
 779         struct net *net = sock_net(skb->sk);
 780         struct fib_config cfg;
 781         struct fib_table *tb;
 782         int err;
 783
 784         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 785         if (err < 0)
 786                 goto errout;
 787
 788         tb = fib_new_table(net, cfg.fc_table);
 789         if (!tb) {
 790                 err = -ENOBUFS;
 791                 goto errout;
 792         }
 793
 794         err = fib_table_insert(net, tb, &cfg, extack);
 795         if (!err && cfg.fc_type == RTN_LOCAL)
 796                 net->ipv4.fib_has_custom_local_routes = true;
 797 errout:
 798         return err;
 799 }
 800
 801 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 802 {
 803         struct net *net = sock_net(skb->sk);
 804         unsigned int h, s_h;
 805         unsigned int e = 0, s_e;
 806         struct fib_table *tb;
 807         struct hlist_head *head;
 808         int dumped = 0, err;
 809
 810         if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
 811             ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
 812                 return skb->len;
 813
 814         s_h = cb->args[0];
 815         s_e = cb->args[1];
 816
 817         rcu_read_lock();
 818
 819         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
 820                 e = 0;
 821                 head = &net->ipv4.fib_table_hash[h];
 822                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
 823                         if (e < s_e)
 824                                 goto next;
 825                         if (dumped)
 826                                 memset(&cb->args[2], 0, sizeof(cb->args) -
 827                                                  2 * sizeof(cb->args[0]));
 828                         err = fib_table_dump(tb, skb, cb);
 829                         if (err < 0) {
 830                                 if (likely(skb->len))
 831                                         goto out;
 832
 833                                 goto out_err;
 834                         }
 835                         dumped = 1;
 836 next:
 837                         e++;
 838                 }
 839         }
 840 out:
 841         err = skb->len;
 842 out_err:
 843         rcu_read_unlock();
 844
 845         cb->args[1] = e;
 846         cb->args[0] = h;
 847
 848         return err;
 849 }
 850
 851 /* Prepare and feed intra-kernel routing request.
 852  * Really, it should be netlink message, but :-( netlink
 853  * can be not configured, so that we feed it directly
 854  * to fib engine. It is legal, because all events occur
 855  * only when netlink is already locked.
 856  */
 857 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
 858                       struct in_ifaddr *ifa, u32 rt_priority)
 859 {
 860         struct net *net = dev_net(ifa->ifa_dev->dev);
 861         u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
 862         struct fib_table *tb;
 863         struct fib_config cfg = {
 864                 .fc_protocol = RTPROT_KERNEL,
 865                 .fc_type = type,
 866                 .fc_dst = dst,
 867                 .fc_dst_len = dst_len,
 868                 .fc_priority = rt_priority,
 869                 .fc_prefsrc = ifa->ifa_local,
 870                 .fc_oif = ifa->ifa_dev->dev->ifindex,
 871                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
 872                 .fc_nlinfo = {
 873                         .nl_net = net,
 874                 },
 875         };
 876
 877         if (!tb_id)
 878                 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
 879
 880         tb = fib_new_table(net, tb_id);
 881         if (!tb)
 882                 return;
 883
 884         cfg.fc_table = tb->tb_id;
 885
 886         if (type != RTN_LOCAL)
 887                 cfg.fc_scope = RT_SCOPE_LINK;
 888         else
 889                 cfg.fc_scope = RT_SCOPE_HOST;
 890
 891         if (cmd == RTM_NEWROUTE)
 892                 fib_table_insert(net, tb, &cfg, NULL);
 893         else
 894                 fib_table_delete(net, tb, &cfg, NULL);
 895 }
 896
 897 void fib_add_ifaddr(struct in_ifaddr *ifa)
 898 {
 899         struct in_device *in_dev = ifa->ifa_dev;
 900         struct net_device *dev = in_dev->dev;
 901         struct in_ifaddr *prim = ifa;
 902         __be32 mask = ifa->ifa_mask;
 903         __be32 addr = ifa->ifa_local;
 904         __be32 prefix = ifa->ifa_address & mask;
 905
 906         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 907                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
 908                 if (!prim) {
 909                         pr_warn("%s: bug: prim == NULL\n", __func__);
 910                         return;
 911                 }
 912         }
 913
 914         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
 915
 916         if (!(dev->flags & IFF_UP))
 917                 return;
 918
 919         /* Add broadcast address, if it is explicitly assigned. */
 920         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
 921                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
 922                           prim, 0);
 923
 924         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
 925             (prefix != addr || ifa->ifa_prefixlen < 32)) {
 926                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
 927                         fib_magic(RTM_NEWROUTE,
 928                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 929                                   prefix, ifa->ifa_prefixlen, prim,
 930                                   ifa->ifa_rt_priority);
 931
 932                 /* Add network specific broadcasts, when it takes a sense */
 933                 if (ifa->ifa_prefixlen < 31) {
 934                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
 935                                   prim, 0);
 936                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
 937                                   32, prim, 0);
 938                 }
 939         }
 940 }
 941
 942 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
 943 {
 944         __be32 prefix = ifa->ifa_address & ifa->ifa_mask;
 945         struct in_device *in_dev = ifa->ifa_dev;
 946         struct net_device *dev = in_dev->dev;
 947
 948         if (!(dev->flags & IFF_UP) ||
 949             ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
 950             ipv4_is_zeronet(prefix) ||
 951             (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32))
 952                 return;
 953
 954         /* add the new */
 955         fib_magic(RTM_NEWROUTE,
 956                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 957                   prefix, ifa->ifa_prefixlen, ifa, new_metric);
 958
 959         /* delete the old */
 960         fib_magic(RTM_DELROUTE,
 961                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 962                   prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
 963 }
 964
 965 /* Delete primary or secondary address.
 966  * Optionally, on secondary address promotion consider the addresses
 967  * from subnet iprim as deleted, even if they are in device list.
 968  * In this case the secondary ifa can be in device list.
 969  */
 970 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
 971 {
 972         struct in_device *in_dev = ifa->ifa_dev;
 973         struct net_device *dev = in_dev->dev;
 974         struct in_ifaddr *ifa1;
 975         struct in_ifaddr *prim = ifa, *prim1 = NULL;
 976         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
 977         __be32 any = ifa->ifa_address & ifa->ifa_mask;
 978 #define LOCAL_OK        1
 979 #define BRD_OK          2
 980 #define BRD0_OK         4
 981 #define BRD1_OK         8
 982         unsigned int ok = 0;
 983         int subnet = 0;         /* Primary network */
 984         int gone = 1;           /* Address is missing */
 985         int same_prefsrc = 0;   /* Another primary with same IP */
 986
 987         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 988                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
 989                 if (!prim) {
 990                         /* if the device has been deleted, we don't perform
 991                          * address promotion
 992                          */
 993                         if (!in_dev->dead)
 994                                 pr_warn("%s: bug: prim == NULL\n", __func__);
 995                         return;
 996                 }
 997                 if (iprim && iprim != prim) {
 998                         pr_warn("%s: bug: iprim != prim\n", __func__);
 999                         return;
1000                 }
1001         } else if (!ipv4_is_zeronet(any) &&
1002                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1003                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1004                         fib_magic(RTM_DELROUTE,
1005                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1006                                   any, ifa->ifa_prefixlen, prim, 0);
1007                 subnet = 1;
1008         }
1009
1010         if (in_dev->dead)
1011                 goto no_promotions;
1012
1013         /* Deletion is more complicated than add.
1014          * We should take care of not to delete too much :-)
1015          *
1016          * Scan address list to be sure that addresses are really gone.
1017          */
1018
1019         for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
1020                 if (ifa1 == ifa) {
1021                         /* promotion, keep the IP */
1022                         gone = 0;
1023                         continue;
1024                 }
1025                 /* Ignore IFAs from our subnet */
1026                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1027                     inet_ifa_match(ifa1->ifa_address, iprim))
1028                         continue;
1029
1030                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
1031                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1032                         /* Another address from our subnet? */
1033                         if (ifa1->ifa_mask == prim->ifa_mask &&
1034                             inet_ifa_match(ifa1->ifa_address, prim))
1035                                 prim1 = prim;
1036                         else {
1037                                 /* We reached the secondaries, so
1038                                  * same_prefsrc should be determined.
1039                                  */
1040                                 if (!same_prefsrc)
1041                                         continue;
1042                                 /* Search new prim1 if ifa1 is not
1043                                  * using the current prim1
1044                                  */
1045                                 if (!prim1 ||
1046                                     ifa1->ifa_mask != prim1->ifa_mask ||
1047                                     !inet_ifa_match(ifa1->ifa_address, prim1))
1048                                         prim1 = inet_ifa_byprefix(in_dev,
1049                                                         ifa1->ifa_address,
1050                                                         ifa1->ifa_mask);
1051                                 if (!prim1)
1052                                         continue;
1053                                 if (prim1->ifa_local != prim->ifa_local)
1054                                         continue;
1055                         }
1056                 } else {
1057                         if (prim->ifa_local != ifa1->ifa_local)
1058                                 continue;
1059                         prim1 = ifa1;
1060                         if (prim != prim1)
1061                                 same_prefsrc = 1;
1062                 }
1063                 if (ifa->ifa_local == ifa1->ifa_local)
1064                         ok |= LOCAL_OK;
1065                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1066                         ok |= BRD_OK;
1067                 if (brd == ifa1->ifa_broadcast)
1068                         ok |= BRD1_OK;
1069                 if (any == ifa1->ifa_broadcast)
1070                         ok |= BRD0_OK;
1071                 /* primary has network specific broadcasts */
1072                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1073                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1074                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1075
1076                         if (!ipv4_is_zeronet(any1)) {
1077                                 if (ifa->ifa_broadcast == brd1 ||
1078                                     ifa->ifa_broadcast == any1)
1079                                         ok |= BRD_OK;
1080                                 if (brd == brd1 || brd == any1)
1081                                         ok |= BRD1_OK;
1082                                 if (any == brd1 || any == any1)
1083                                         ok |= BRD0_OK;
1084                         }
1085                 }
1086         }
1087
1088 no_promotions:
1089         if (!(ok & BRD_OK))
1090                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1091                           prim, 0);
1092         if (subnet && ifa->ifa_prefixlen < 31) {
1093                 if (!(ok & BRD1_OK))
1094                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1095                                   prim, 0);
1096                 if (!(ok & BRD0_OK))
1097                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1098                                   prim, 0);
1099         }
1100         if (!(ok & LOCAL_OK)) {
1101                 unsigned int addr_type;
1102
1103                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1104
1105                 /* Check, that this local address finally disappeared. */
1106                 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1107                                                      ifa->ifa_local);
1108                 if (gone && addr_type != RTN_LOCAL) {
1109                         /* And the last, but not the least thing.
1110                          * We must flush stray FIB entries.
1111                          *
1112                          * First of all, we scan fib_info list searching
1113                          * for stray nexthop entries, then ignite fib_flush.
1114                          */
1115                         if (fib_sync_down_addr(dev, ifa->ifa_local))
1116                                 fib_flush(dev_net(dev));
1117                 }
1118         }
1119 #undef LOCAL_OK
1120 #undef BRD_OK
1121 #undef BRD0_OK
1122 #undef BRD1_OK
1123 }
1124
1125 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1126 {
1127
1128         struct fib_result       res;
1129         struct flowi4           fl4 = {
1130                 .flowi4_mark = frn->fl_mark,
1131                 .daddr = frn->fl_addr,
1132                 .flowi4_tos = frn->fl_tos,
1133                 .flowi4_scope = frn->fl_scope,
1134         };
1135         struct fib_table *tb;
1136
1137         rcu_read_lock();
1138
1139         tb = fib_get_table(net, frn->tb_id_in);
1140
1141         frn->err = -ENOENT;
1142         if (tb) {
1143                 local_bh_disable();
1144
1145                 frn->tb_id = tb->tb_id;
1146                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1147
1148                 if (!frn->err) {
1149                         frn->prefixlen = res.prefixlen;
1150                         frn->nh_sel = res.nh_sel;
1151                         frn->type = res.type;
1152                         frn->scope = res.scope;
1153                 }
1154                 local_bh_enable();
1155         }
1156
1157         rcu_read_unlock();
1158 }
1159
1160 static void nl_fib_input(struct sk_buff *skb)
1161 {
1162         struct net *net;
1163         struct fib_result_nl *frn;
1164         struct nlmsghdr *nlh;
1165         u32 portid;
1166
1167         net = sock_net(skb->sk);
1168         nlh = nlmsg_hdr(skb);
1169         if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1170             skb->len < nlh->nlmsg_len ||
1171             nlmsg_len(nlh) < sizeof(*frn))
1172                 return;
1173
1174         skb = netlink_skb_clone(skb, GFP_KERNEL);
1175         if (!skb)
1176                 return;
1177         nlh = nlmsg_hdr(skb);
1178
1179         frn = (struct fib_result_nl *) nlmsg_data(nlh);
1180         nl_fib_lookup(net, frn);
1181
1182         portid = NETLINK_CB(skb).portid;      /* netlink portid */
1183         NETLINK_CB(skb).portid = 0;        /* from kernel */
1184         NETLINK_CB(skb).dst_group = 0;  /* unicast */
1185         netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1186 }
1187
1188 static int __net_init nl_fib_lookup_init(struct net *net)
1189 {
1190         struct sock *sk;
1191         struct netlink_kernel_cfg cfg = {
1192                 .input  = nl_fib_input,
1193         };
1194
1195         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1196         if (!sk)
1197                 return -EAFNOSUPPORT;
1198         net->ipv4.fibnl = sk;
1199         return 0;
1200 }
1201
1202 static void nl_fib_lookup_exit(struct net *net)
1203 {
1204         netlink_kernel_release(net->ipv4.fibnl);
1205         net->ipv4.fibnl = NULL;
1206 }
1207
1208 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1209                            bool force)
1210 {
1211         if (fib_sync_down_dev(dev, event, force))
1212                 fib_flush(dev_net(dev));
1213         else
1214                 rt_cache_flush(dev_net(dev));
1215         arp_ifdown(dev);
1216 }
1217
1218 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1219 {
1220         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1221         struct net_device *dev = ifa->ifa_dev->dev;
1222         struct net *net = dev_net(dev);
1223
1224         switch (event) {
1225         case NETDEV_UP:
1226                 fib_add_ifaddr(ifa);
1227 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1228                 fib_sync_up(dev, RTNH_F_DEAD);
1229 #endif
1230                 atomic_inc(&net->ipv4.dev_addr_genid);
1231                 rt_cache_flush(dev_net(dev));
1232                 break;
1233         case NETDEV_DOWN:
1234                 fib_del_ifaddr(ifa, NULL);
1235                 atomic_inc(&net->ipv4.dev_addr_genid);
1236                 if (!ifa->ifa_dev->ifa_list) {
1237                         /* Last address was deleted from this interface.
1238                          * Disable IP.
1239                          */
1240                         fib_disable_ip(dev, event, true);
1241                 } else {
1242                         rt_cache_flush(dev_net(dev));
1243                 }
1244                 break;
1245         }
1246         return NOTIFY_DONE;
1247 }
1248
1249 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1250 {
1251         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1252         struct netdev_notifier_changeupper_info *upper_info = ptr;
1253         struct netdev_notifier_info_ext *info_ext = ptr;
1254         struct in_device *in_dev;
1255         struct net *net = dev_net(dev);
1256         unsigned int flags;
1257
1258         if (event == NETDEV_UNREGISTER) {
1259                 fib_disable_ip(dev, event, true);
1260                 rt_flush_dev(dev);
1261                 return NOTIFY_DONE;
1262         }
1263
1264         in_dev = __in_dev_get_rtnl(dev);
1265         if (!in_dev)
1266                 return NOTIFY_DONE;
1267
1268         switch (event) {
1269         case NETDEV_UP:
1270                 for_ifa(in_dev) {
1271                         fib_add_ifaddr(ifa);
1272                 } endfor_ifa(in_dev);
1273 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1274                 fib_sync_up(dev, RTNH_F_DEAD);
1275 #endif
1276                 atomic_inc(&net->ipv4.dev_addr_genid);
1277                 rt_cache_flush(net);
1278                 break;
1279         case NETDEV_DOWN:
1280                 fib_disable_ip(dev, event, false);
1281                 break;
1282         case NETDEV_CHANGE:
1283                 flags = dev_get_flags(dev);
1284                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1285                         fib_sync_up(dev, RTNH_F_LINKDOWN);
1286                 else
1287                         fib_sync_down_dev(dev, event, false);
1288                 rt_cache_flush(net);
1289                 break;
1290         case NETDEV_CHANGEMTU:
1291                 fib_sync_mtu(dev, info_ext->ext.mtu);
1292                 rt_cache_flush(net);
1293                 break;
1294         case NETDEV_CHANGEUPPER:
1295                 upper_info = ptr;
1296                 /* flush all routes if dev is linked to or unlinked from
1297                  * an L3 master device (e.g., VRF)
1298                  */
1299                 if (upper_info->upper_dev &&
1300                     netif_is_l3_master(upper_info->upper_dev))
1301                         fib_disable_ip(dev, NETDEV_DOWN, true);
1302                 break;
1303         }
1304         return NOTIFY_DONE;
1305 }
1306
1307 static struct notifier_block fib_inetaddr_notifier = {
1308         .notifier_call = fib_inetaddr_event,
1309 };
1310
1311 static struct notifier_block fib_netdev_notifier = {
1312         .notifier_call = fib_netdev_event,
1313 };
1314
1315 static int __net_init ip_fib_net_init(struct net *net)
1316 {
1317         int err;
1318         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1319
1320         err = fib4_notifier_init(net);
1321         if (err)
1322                 return err;
1323
1324         /* Avoid false sharing : Use at least a full cache line */
1325         size = max_t(size_t, size, L1_CACHE_BYTES);
1326
1327         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1328         if (!net->ipv4.fib_table_hash) {
1329                 err = -ENOMEM;
1330                 goto err_table_hash_alloc;
1331         }
1332
1333         err = fib4_rules_init(net);
1334         if (err < 0)
1335                 goto err_rules_init;
1336         return 0;
1337
1338 err_rules_init:
1339         kfree(net->ipv4.fib_table_hash);
1340 err_table_hash_alloc:
1341         fib4_notifier_exit(net);
1342         return err;
1343 }
1344
1345 static void ip_fib_net_exit(struct net *net)
1346 {
1347         int i;
1348
1349         rtnl_lock();
1350 #ifdef CONFIG_IP_MULTIPLE_TABLES
1351         RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1352         RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1353 #endif
1354         /* Destroy the tables in reverse order to guarantee that the
1355          * local table, ID 255, is destroyed before the main table, ID
1356          * 254. This is necessary as the local table may contain
1357          * references to data contained in the main table.
1358          */
1359         for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1360                 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1361                 struct hlist_node *tmp;
1362                 struct fib_table *tb;
1363
1364                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1365                         hlist_del(&tb->tb_hlist);
1366                         fib_table_flush(net, tb, true);
1367                         fib_free_table(tb);
1368                 }
1369         }
1370
1371 #ifdef CONFIG_IP_MULTIPLE_TABLES
1372         fib4_rules_exit(net);
1373 #endif
1374         rtnl_unlock();
1375         kfree(net->ipv4.fib_table_hash);
1376         fib4_notifier_exit(net);
1377 }
1378
1379 static int __net_init fib_net_init(struct net *net)
1380 {
1381         int error;
1382
1383 #ifdef CONFIG_IP_ROUTE_CLASSID
1384         net->ipv4.fib_num_tclassid_users = 0;
1385 #endif
1386         error = ip_fib_net_init(net);
1387         if (error < 0)
1388                 goto out;
1389         error = nl_fib_lookup_init(net);
1390         if (error < 0)
1391                 goto out_nlfl;
1392         error = fib_proc_init(net);
1393         if (error < 0)
1394                 goto out_proc;
1395 out:
1396         return error;
1397
1398 out_proc:
1399         nl_fib_lookup_exit(net);
1400 out_nlfl:
1401         ip_fib_net_exit(net);
1402         goto out;
1403 }
1404
1405 static void __net_exit fib_net_exit(struct net *net)
1406 {
1407         fib_proc_exit(net);
1408         nl_fib_lookup_exit(net);
1409         ip_fib_net_exit(net);
1410 }
1411
1412 static struct pernet_operations fib_net_ops = {
1413         .init = fib_net_init,
1414         .exit = fib_net_exit,
1415 };
1416
1417 void __init ip_fib_init(void)
1418 {
1419         fib_trie_init();
1420
1421         register_pernet_subsys(&fib_net_ops);
1422
1423         register_netdevice_notifier(&fib_netdev_notifier);
1424         register_inetaddr_notifier(&fib_inetaddr_notifier);
1425
1426         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1427         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1428         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1429 }