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