net/ipv4/fib_frontend.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              IPv4 Forwarding Information Base: FIB frontend.
   8  *
   9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/uaccess.h>
  14 #include <linux/bitops.h>
  15 #include <linux/capability.h>
  16 #include <linux/types.h>
  17 #include <linux/kernel.h>
  18 #include <linux/mm.h>
  19 #include <linux/string.h>
  20 #include <linux/socket.h>
  21 #include <linux/sockios.h>
  22 #include <linux/errno.h>
  23 #include <linux/in.h>
  24 #include <linux/inet.h>
  25 #include <linux/inetdevice.h>
  26 #include <linux/netdevice.h>
  27 #include <linux/if_addr.h>
  28 #include <linux/if_arp.h>
  29 #include <linux/skbuff.h>
  30 #include <linux/cache.h>
  31 #include <linux/init.h>
  32 #include <linux/list.h>
  33 #include <linux/slab.h>
  34
  35 #include <net/ip.h>
  36 #include <net/protocol.h>
  37 #include <net/route.h>
  38 #include <net/tcp.h>
  39 #include <net/sock.h>
  40 #include <net/arp.h>
  41 #include <net/ip_fib.h>
  42 #include <net/nexthop.h>
  43 #include <net/rtnetlink.h>
  44 #include <net/xfrm.h>
  45 #include <net/l3mdev.h>
  46 #include <net/lwtunnel.h>
  47 #include <trace/events/fib.h>
  48
  49 #ifndef CONFIG_IP_MULTIPLE_TABLES
  50
  51 static int __net_init fib4_rules_init(struct net *net)
  52 {
  53         struct fib_table *local_table, *main_table;
  54
  55         main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
  56         if (!main_table)
  57                 return -ENOMEM;
  58
  59         local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
  60         if (!local_table)
  61                 goto fail;
  62
  63         hlist_add_head_rcu(&local_table->tb_hlist,
  64                                 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
  65         hlist_add_head_rcu(&main_table->tb_hlist,
  66                                 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
  67         return 0;
  68
  69 fail:
  70         fib_free_table(main_table);
  71         return -ENOMEM;
  72 }
  73
  74 static bool fib4_has_custom_rules(struct net *net)
  75 {
  76         return false;
  77 }
  78 #else
  79
  80 struct fib_table *fib_new_table(struct net *net, u32 id)
  81 {
  82         struct fib_table *tb, *alias = NULL;
  83         unsigned int h;
  84
  85         if (id == 0)
  86                 id = RT_TABLE_MAIN;
  87         tb = fib_get_table(net, id);
  88         if (tb)
  89                 return tb;
  90
  91         if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
  92                 alias = fib_new_table(net, RT_TABLE_MAIN);
  93
  94         tb = fib_trie_table(id, alias);
  95         if (!tb)
  96                 return NULL;
  97
  98         switch (id) {
  99         case RT_TABLE_MAIN:
 100                 rcu_assign_pointer(net->ipv4.fib_main, tb);
 101                 break;
 102         case RT_TABLE_DEFAULT:
 103                 rcu_assign_pointer(net->ipv4.fib_default, tb);
 104                 break;
 105         default:
 106                 break;
 107         }
 108
 109         h = id & (FIB_TABLE_HASHSZ - 1);
 110         hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
 111         return tb;
 112 }
 113 EXPORT_SYMBOL_GPL(fib_new_table);
 114
 115 /* caller must hold either rtnl or rcu read lock */
 116 struct fib_table *fib_get_table(struct net *net, u32 id)
 117 {
 118         struct fib_table *tb;
 119         struct hlist_head *head;
 120         unsigned int h;
 121
 122         if (id == 0)
 123                 id = RT_TABLE_MAIN;
 124         h = id & (FIB_TABLE_HASHSZ - 1);
 125
 126         head = &net->ipv4.fib_table_hash[h];
 127         hlist_for_each_entry_rcu(tb, head, tb_hlist) {
 128                 if (tb->tb_id == id)
 129                         return tb;
 130         }
 131         return NULL;
 132 }
 133
 134 static bool fib4_has_custom_rules(struct net *net)
 135 {
 136         return net->ipv4.fib_has_custom_rules;
 137 }
 138 #endif /* CONFIG_IP_MULTIPLE_TABLES */
 139
 140 static void fib_replace_table(struct net *net, struct fib_table *old,
 141                               struct fib_table *new)
 142 {
 143 #ifdef CONFIG_IP_MULTIPLE_TABLES
 144         switch (new->tb_id) {
 145         case RT_TABLE_MAIN:
 146                 rcu_assign_pointer(net->ipv4.fib_main, new);
 147                 break;
 148         case RT_TABLE_DEFAULT:
 149                 rcu_assign_pointer(net->ipv4.fib_default, new);
 150                 break;
 151         default:
 152                 break;
 153         }
 154
 155 #endif
 156         /* replace the old table in the hlist */
 157         hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
 158 }
 159
 160 int fib_unmerge(struct net *net)
 161 {
 162         struct fib_table *old, *new, *main_table;
 163
 164         /* attempt to fetch local table if it has been allocated */
 165         old = fib_get_table(net, RT_TABLE_LOCAL);
 166         if (!old)
 167                 return 0;
 168
 169         new = fib_trie_unmerge(old);
 170         if (!new)
 171                 return -ENOMEM;
 172
 173         /* table is already unmerged */
 174         if (new == old)
 175                 return 0;
 176
 177         /* replace merged table with clean table */
 178         fib_replace_table(net, old, new);
 179         fib_free_table(old);
 180
 181         /* attempt to fetch main table if it has been allocated */
 182         main_table = fib_get_table(net, RT_TABLE_MAIN);
 183         if (!main_table)
 184                 return 0;
 185
 186         /* flush local entries from main table */
 187         fib_table_flush_external(main_table);
 188
 189         return 0;
 190 }
 191
 192 void fib_flush(struct net *net)
 193 {
 194         int flushed = 0;
 195         unsigned int h;
 196
 197         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
 198                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
 199                 struct hlist_node *tmp;
 200                 struct fib_table *tb;
 201
 202                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
 203                         flushed += fib_table_flush(net, tb, false);
 204         }
 205
 206         if (flushed)
 207                 rt_cache_flush(net);
 208 }
 209
 210 /*
 211  * Find address type as if only "dev" was present in the system. If
 212  * on_dev is NULL then all interfaces are taken into consideration.
 213  */
 214 static inline unsigned int __inet_dev_addr_type(struct net *net,
 215                                                 const struct net_device *dev,
 216                                                 __be32 addr, u32 tb_id)
 217 {
 218         struct flowi4           fl4 = { .daddr = addr };
 219         struct fib_result       res;
 220         unsigned int ret = RTN_BROADCAST;
 221         struct fib_table *table;
 222
 223         if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
 224                 return RTN_BROADCAST;
 225         if (ipv4_is_multicast(addr))
 226                 return RTN_MULTICAST;
 227
 228         rcu_read_lock();
 229
 230         table = fib_get_table(net, tb_id);
 231         if (table) {
 232                 ret = RTN_UNICAST;
 233                 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
 234                         struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0);
 235
 236                         if (!dev || dev == nhc->nhc_dev)
 237                                 ret = res.type;
 238                 }
 239         }
 240
 241         rcu_read_unlock();
 242         return ret;
 243 }
 244
 245 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
 246 {
 247         return __inet_dev_addr_type(net, NULL, addr, tb_id);
 248 }
 249 EXPORT_SYMBOL(inet_addr_type_table);
 250
 251 unsigned int inet_addr_type(struct net *net, __be32 addr)
 252 {
 253         return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
 254 }
 255 EXPORT_SYMBOL(inet_addr_type);
 256
 257 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
 258                                 __be32 addr)
 259 {
 260         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
 261
 262         return __inet_dev_addr_type(net, dev, addr, rt_table);
 263 }
 264 EXPORT_SYMBOL(inet_dev_addr_type);
 265
 266 /* inet_addr_type with dev == NULL but using the table from a dev
 267  * if one is associated
 268  */
 269 unsigned int inet_addr_type_dev_table(struct net *net,
 270                                       const struct net_device *dev,
 271                                       __be32 addr)
 272 {
 273         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
 274
 275         return __inet_dev_addr_type(net, NULL, addr, rt_table);
 276 }
 277 EXPORT_SYMBOL(inet_addr_type_dev_table);
 278
 279 __be32 fib_compute_spec_dst(struct sk_buff *skb)
 280 {
 281         struct net_device *dev = skb->dev;
 282         struct in_device *in_dev;
 283         struct fib_result res;
 284         struct rtable *rt;
 285         struct net *net;
 286         int scope;
 287
 288         rt = skb_rtable(skb);
 289         if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
 290             RTCF_LOCAL)
 291                 return ip_hdr(skb)->daddr;
 292
 293         in_dev = __in_dev_get_rcu(dev);
 294
 295         net = dev_net(dev);
 296
 297         scope = RT_SCOPE_UNIVERSE;
 298         if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
 299                 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
 300                 struct flowi4 fl4 = {
 301                         .flowi4_iif = LOOPBACK_IFINDEX,
 302                         .flowi4_oif = l3mdev_master_ifindex_rcu(dev),
 303                         .daddr = ip_hdr(skb)->saddr,
 304                         .flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
 305                         .flowi4_scope = scope,
 306                         .flowi4_mark = vmark ? skb->mark : 0,
 307                 };
 308                 if (!fib_lookup(net, &fl4, &res, 0))
 309                         return fib_result_prefsrc(net, &res);
 310         } else {
 311                 scope = RT_SCOPE_LINK;
 312         }
 313
 314         return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
 315 }
 316
 317 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
 318 {
 319         bool dev_match = false;
 320 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 321         int ret;
 322
 323         for (ret = 0; ret < fib_info_num_path(fi); ret++) {
 324                 const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
 325
 326                 if (nhc->nhc_dev == dev) {
 327                         dev_match = true;
 328                         break;
 329                 } else if (l3mdev_master_ifindex_rcu(nhc->nhc_dev) == dev->ifindex) {
 330                         dev_match = true;
 331                         break;
 332                 }
 333         }
 334 #else
 335         if (fib_info_nhc(fi, 0)->nhc_dev == dev)
 336                 dev_match = true;
 337 #endif
 338
 339         return dev_match;
 340 }
 341 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
 342
 343 /* Given (packet source, input interface) and optional (dst, oif, tos):
 344  * - (main) check, that source is valid i.e. not broadcast or our local
 345  *   address.
 346  * - figure out what "logical" interface this packet arrived
 347  *   and calculate "specific destination" address.
 348  * - check, that packet arrived from expected physical interface.
 349  * called with rcu_read_lock()
 350  */
 351 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 352                                  u8 tos, int oif, struct net_device *dev,
 353                                  int rpf, struct in_device *idev, u32 *itag)
 354 {
 355         struct net *net = dev_net(dev);
 356         struct flow_keys flkeys;
 357         int ret, no_addr;
 358         struct fib_result res;
 359         struct flowi4 fl4;
 360         bool dev_match;
 361
 362         fl4.flowi4_oif = 0;
 363         fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
 364         if (!fl4.flowi4_iif)
 365                 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
 366         fl4.daddr = src;
 367         fl4.saddr = dst;
 368         fl4.flowi4_tos = tos;
 369         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 370         fl4.flowi4_tun_key.tun_id = 0;
 371         fl4.flowi4_flags = 0;
 372         fl4.flowi4_uid = sock_net_uid(net, NULL);
 373
 374         no_addr = idev->ifa_list == NULL;
 375
 376         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
 377         if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
 378                 fl4.flowi4_proto = 0;
 379                 fl4.fl4_sport = 0;
 380                 fl4.fl4_dport = 0;
 381         }
 382
 383         if (fib_lookup(net, &fl4, &res, 0))
 384                 goto last_resort;
 385         if (res.type != RTN_UNICAST &&
 386             (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
 387                 goto e_inval;
 388         fib_combine_itag(itag, &res);
 389
 390         dev_match = fib_info_nh_uses_dev(res.fi, dev);
 391         if (dev_match) {
 392                 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
 393                 return ret;
 394         }
 395         if (no_addr)
 396                 goto last_resort;
 397         if (rpf == 1)
 398                 goto e_rpf;
 399         fl4.flowi4_oif = dev->ifindex;
 400
 401         ret = 0;
 402         if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
 403                 if (res.type == RTN_UNICAST)
 404                         ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
 405         }
 406         return ret;
 407
 408 last_resort:
 409         if (rpf)
 410                 goto e_rpf;
 411         *itag = 0;
 412         return 0;
 413
 414 e_inval:
 415         return -EINVAL;
 416 e_rpf:
 417         return -EXDEV;
 418 }
 419
 420 /* Ignore rp_filter for packets protected by IPsec. */
 421 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 422                         u8 tos, int oif, struct net_device *dev,
 423                         struct in_device *idev, u32 *itag)
 424 {
 425         int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
 426         struct net *net = dev_net(dev);
 427
 428         if (!r && !fib_num_tclassid_users(net) &&
 429             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
 430                 if (IN_DEV_ACCEPT_LOCAL(idev))
 431                         goto ok;
 432                 /* with custom local routes in place, checking local addresses
 433                  * only will be too optimistic, with custom rules, checking
 434                  * local addresses only can be too strict, e.g. due to vrf
 435                  */
 436                 if (net->ipv4.fib_has_custom_local_routes ||
 437                     fib4_has_custom_rules(net))
 438                         goto full_check;
 439                 if (inet_lookup_ifaddr_rcu(net, src))
 440                         return -EINVAL;
 441
 442 ok:
 443                 *itag = 0;
 444                 return 0;
 445         }
 446
 447 full_check:
 448         return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
 449 }
 450
 451 static inline __be32 sk_extract_addr(struct sockaddr *addr)
 452 {
 453         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 454 }
 455
 456 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
 457 {
 458         struct nlattr *nla;
 459
 460         nla = (struct nlattr *) ((char *) mx + len);
 461         nla->nla_type = type;
 462         nla->nla_len = nla_attr_size(4);
 463         *(u32 *) nla_data(nla) = value;
 464
 465         return len + nla_total_size(4);
 466 }
 467
 468 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
 469                                  struct fib_config *cfg)
 470 {
 471         __be32 addr;
 472         int plen;
 473
 474         memset(cfg, 0, sizeof(*cfg));
 475         cfg->fc_nlinfo.nl_net = net;
 476
 477         if (rt->rt_dst.sa_family != AF_INET)
 478                 return -EAFNOSUPPORT;
 479
 480         /*
 481          * Check mask for validity:
 482          * a) it must be contiguous.
 483          * b) destination must have all host bits clear.
 484          * c) if application forgot to set correct family (AF_INET),
 485          *    reject request unless it is absolutely clear i.e.
 486          *    both family and mask are zero.
 487          */
 488         plen = 32;
 489         addr = sk_extract_addr(&rt->rt_dst);
 490         if (!(rt->rt_flags & RTF_HOST)) {
 491                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
 492
 493                 if (rt->rt_genmask.sa_family != AF_INET) {
 494                         if (mask || rt->rt_genmask.sa_family)
 495                                 return -EAFNOSUPPORT;
 496                 }
 497
 498                 if (bad_mask(mask, addr))
 499                         return -EINVAL;
 500
 501                 plen = inet_mask_len(mask);
 502         }
 503
 504         cfg->fc_dst_len = plen;
 505         cfg->fc_dst = addr;
 506
 507         if (cmd != SIOCDELRT) {
 508                 cfg->fc_nlflags = NLM_F_CREATE;
 509                 cfg->fc_protocol = RTPROT_BOOT;
 510         }
 511
 512         if (rt->rt_metric)
 513                 cfg->fc_priority = rt->rt_metric - 1;
 514
 515         if (rt->rt_flags & RTF_REJECT) {
 516                 cfg->fc_scope = RT_SCOPE_HOST;
 517                 cfg->fc_type = RTN_UNREACHABLE;
 518                 return 0;
 519         }
 520
 521         cfg->fc_scope = RT_SCOPE_NOWHERE;
 522         cfg->fc_type = RTN_UNICAST;
 523
 524         if (rt->rt_dev) {
 525                 char *colon;
 526                 struct net_device *dev;
 527                 char devname[IFNAMSIZ];
 528
 529                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
 530                         return -EFAULT;
 531
 532                 devname[IFNAMSIZ-1] = 0;
 533                 colon = strchr(devname, ':');
 534                 if (colon)
 535                         *colon = 0;
 536                 dev = __dev_get_by_name(net, devname);
 537                 if (!dev)
 538                         return -ENODEV;
 539                 cfg->fc_oif = dev->ifindex;
 540                 cfg->fc_table = l3mdev_fib_table(dev);
 541                 if (colon) {
 542                         const struct in_ifaddr *ifa;
 543                         struct in_device *in_dev;
 544
 545                         in_dev = __in_dev_get_rtnl(dev);
 546                         if (!in_dev)
 547                                 return -ENODEV;
 548
 549                         *colon = ':';
 550
 551                         rcu_read_lock();
 552                         in_dev_for_each_ifa_rcu(ifa, in_dev) {
 553                                 if (strcmp(ifa->ifa_label, devname) == 0)
 554                                         break;
 555                         }
 556                         rcu_read_unlock();
 557
 558                         if (!ifa)
 559                                 return -ENODEV;
 560                         cfg->fc_prefsrc = ifa->ifa_local;
 561                 }
 562         }
 563
 564         addr = sk_extract_addr(&rt->rt_gateway);
 565         if (rt->rt_gateway.sa_family == AF_INET && addr) {
 566                 unsigned int addr_type;
 567
 568                 cfg->fc_gw4 = addr;
 569                 cfg->fc_gw_family = AF_INET;
 570                 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
 571                 if (rt->rt_flags & RTF_GATEWAY &&
 572                     addr_type == RTN_UNICAST)
 573                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
 574         }
 575
 576         if (cmd == SIOCDELRT)
 577                 return 0;
 578
 579         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
 580                 return -EINVAL;
 581
 582         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 583                 cfg->fc_scope = RT_SCOPE_LINK;
 584
 585         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 586                 struct nlattr *mx;
 587                 int len = 0;
 588
 589                 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
 590                 if (!mx)
 591                         return -ENOMEM;
 592
 593                 if (rt->rt_flags & RTF_MTU)
 594                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 595
 596                 if (rt->rt_flags & RTF_WINDOW)
 597                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 598
 599                 if (rt->rt_flags & RTF_IRTT)
 600                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 601
 602                 cfg->fc_mx = mx;
 603                 cfg->fc_mx_len = len;
 604         }
 605
 606         return 0;
 607 }
 608
 609 /*
 610  * Handle IP routing ioctl calls.
 611  * These are used to manipulate the routing tables
 612  */
 613 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
 614 {
 615         struct fib_config cfg;
 616         int err;
 617
 618         switch (cmd) {
 619         case SIOCADDRT:         /* Add a route */
 620         case SIOCDELRT:         /* Delete a route */
 621                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 622                         return -EPERM;
 623
 624                 rtnl_lock();
 625                 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
 626                 if (err == 0) {
 627                         struct fib_table *tb;
 628
 629                         if (cmd == SIOCDELRT) {
 630                                 tb = fib_get_table(net, cfg.fc_table);
 631                                 if (tb)
 632                                         err = fib_table_delete(net, tb, &cfg,
 633                                                                NULL);
 634                                 else
 635                                         err = -ESRCH;
 636                         } else {
 637                                 tb = fib_new_table(net, cfg.fc_table);
 638                                 if (tb)
 639                                         err = fib_table_insert(net, tb,
 640                                                                &cfg, NULL);
 641                                 else
 642                                         err = -ENOBUFS;
 643                         }
 644
 645                         /* allocated by rtentry_to_fib_config() */
 646                         kfree(cfg.fc_mx);
 647                 }
 648                 rtnl_unlock();
 649                 return err;
 650         }
 651         return -EINVAL;
 652 }
 653
 654 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 655         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
 656         [RTA_DST]               = { .type = NLA_U32 },
 657         [RTA_SRC]               = { .type = NLA_U32 },
 658         [RTA_IIF]               = { .type = NLA_U32 },
 659         [RTA_OIF]               = { .type = NLA_U32 },
 660         [RTA_GATEWAY]           = { .type = NLA_U32 },
 661         [RTA_PRIORITY]          = { .type = NLA_U32 },
 662         [RTA_PREFSRC]           = { .type = NLA_U32 },
 663         [RTA_METRICS]           = { .type = NLA_NESTED },
 664         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 665         [RTA_FLOW]              = { .type = NLA_U32 },
 666         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
 667         [RTA_ENCAP]             = { .type = NLA_NESTED },
 668         [RTA_UID]               = { .type = NLA_U32 },
 669         [RTA_MARK]              = { .type = NLA_U32 },
 670         [RTA_TABLE]             = { .type = NLA_U32 },
 671         [RTA_IP_PROTO]          = { .type = NLA_U8 },
 672         [RTA_SPORT]             = { .type = NLA_U16 },
 673         [RTA_DPORT]             = { .type = NLA_U16 },
 674         [RTA_NH_ID]             = { .type = NLA_U32 },
 675 };
 676
 677 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
 678                     struct netlink_ext_ack *extack)
 679 {
 680         struct rtvia *via;
 681         int alen;
 682
 683         if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
 684                 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
 685                 return -EINVAL;
 686         }
 687
 688         via = nla_data(nla);
 689         alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
 690
 691         switch (via->rtvia_family) {
 692         case AF_INET:
 693                 if (alen != sizeof(__be32)) {
 694                         NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
 695                         return -EINVAL;
 696                 }
 697                 cfg->fc_gw_family = AF_INET;
 698                 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
 699                 break;
 700         case AF_INET6:
 701 #ifdef CONFIG_IPV6
 702                 if (alen != sizeof(struct in6_addr)) {
 703                         NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
 704                         return -EINVAL;
 705                 }
 706                 cfg->fc_gw_family = AF_INET6;
 707                 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
 708 #else
 709                 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
 710                 return -EINVAL;
 711 #endif
 712                 break;
 713         default:
 714                 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
 715                 return -EINVAL;
 716         }
 717
 718         return 0;
 719 }
 720
 721 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 722                              struct nlmsghdr *nlh, struct fib_config *cfg,
 723                              struct netlink_ext_ack *extack)
 724 {
 725         bool has_gw = false, has_via = false;
 726         struct nlattr *attr;
 727         int err, remaining;
 728         struct rtmsg *rtm;
 729
 730         err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
 731                                         rtm_ipv4_policy, extack);
 732         if (err < 0)
 733                 goto errout;
 734
 735         memset(cfg, 0, sizeof(*cfg));
 736
 737         rtm = nlmsg_data(nlh);
 738         cfg->fc_dst_len = rtm->rtm_dst_len;
 739         cfg->fc_tos = rtm->rtm_tos;
 740         cfg->fc_table = rtm->rtm_table;
 741         cfg->fc_protocol = rtm->rtm_protocol;
 742         cfg->fc_scope = rtm->rtm_scope;
 743         cfg->fc_type = rtm->rtm_type;
 744         cfg->fc_flags = rtm->rtm_flags;
 745         cfg->fc_nlflags = nlh->nlmsg_flags;
 746
 747         cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
 748         cfg->fc_nlinfo.nlh = nlh;
 749         cfg->fc_nlinfo.nl_net = net;
 750
 751         if (cfg->fc_type > RTN_MAX) {
 752                 NL_SET_ERR_MSG(extack, "Invalid route type");
 753                 err = -EINVAL;
 754                 goto errout;
 755         }
 756
 757         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 758                 switch (nla_type(attr)) {
 759                 case RTA_DST:
 760                         cfg->fc_dst = nla_get_be32(attr);
 761                         break;
 762                 case RTA_OIF:
 763                         cfg->fc_oif = nla_get_u32(attr);
 764                         break;
 765                 case RTA_GATEWAY:
 766                         has_gw = true;
 767                         cfg->fc_gw4 = nla_get_be32(attr);
 768                         if (cfg->fc_gw4)
 769                                 cfg->fc_gw_family = AF_INET;
 770                         break;
 771                 case RTA_VIA:
 772                         has_via = true;
 773                         err = fib_gw_from_via(cfg, attr, extack);
 774                         if (err)
 775                                 goto errout;
 776                         break;
 777                 case RTA_PRIORITY:
 778                         cfg->fc_priority = nla_get_u32(attr);
 779                         break;
 780                 case RTA_PREFSRC:
 781                         cfg->fc_prefsrc = nla_get_be32(attr);
 782                         break;
 783                 case RTA_METRICS:
 784                         cfg->fc_mx = nla_data(attr);
 785                         cfg->fc_mx_len = nla_len(attr);
 786                         break;
 787                 case RTA_MULTIPATH:
 788                         err = lwtunnel_valid_encap_type_attr(nla_data(attr),
 789                                                              nla_len(attr),
 790                                                              extack);
 791                         if (err < 0)
 792                                 goto errout;
 793                         cfg->fc_mp = nla_data(attr);
 794                         cfg->fc_mp_len = nla_len(attr);
 795                         break;
 796                 case RTA_FLOW:
 797                         cfg->fc_flow = nla_get_u32(attr);
 798                         break;
 799                 case RTA_TABLE:
 800                         cfg->fc_table = nla_get_u32(attr);
 801                         break;
 802                 case RTA_ENCAP:
 803                         cfg->fc_encap = attr;
 804                         break;
 805                 case RTA_ENCAP_TYPE:
 806                         cfg->fc_encap_type = nla_get_u16(attr);
 807                         err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
 808                                                         extack);
 809                         if (err < 0)
 810                                 goto errout;
 811                         break;
 812                 case RTA_NH_ID:
 813                         cfg->fc_nh_id = nla_get_u32(attr);
 814                         break;
 815                 }
 816         }
 817
 818         if (cfg->fc_nh_id) {
 819                 if (cfg->fc_oif || cfg->fc_gw_family ||
 820                     cfg->fc_encap || cfg->fc_mp) {
 821                         NL_SET_ERR_MSG(extack,
 822                                        "Nexthop specification and nexthop id are mutually exclusive");
 823                         return -EINVAL;
 824                 }
 825         }
 826
 827         if (has_gw && has_via) {
 828                 NL_SET_ERR_MSG(extack,
 829                                "Nexthop configuration can not contain both GATEWAY and VIA");
 830                 goto errout;
 831         }
 832
 833         return 0;
 834 errout:
 835         return err;
 836 }
 837
 838 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 839                              struct netlink_ext_ack *extack)
 840 {
 841         struct net *net = sock_net(skb->sk);
 842         struct fib_config cfg;
 843         struct fib_table *tb;
 844         int err;
 845
 846         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 847         if (err < 0)
 848                 goto errout;
 849
 850         if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
 851                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
 852                 err = -EINVAL;
 853                 goto errout;
 854         }
 855
 856         tb = fib_get_table(net, cfg.fc_table);
 857         if (!tb) {
 858                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
 859                 err = -ESRCH;
 860                 goto errout;
 861         }
 862
 863         err = fib_table_delete(net, tb, &cfg, extack);
 864 errout:
 865         return err;
 866 }
 867
 868 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 869                              struct netlink_ext_ack *extack)
 870 {
 871         struct net *net = sock_net(skb->sk);
 872         struct fib_config cfg;
 873         struct fib_table *tb;
 874         int err;
 875
 876         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 877         if (err < 0)
 878                 goto errout;
 879
 880         tb = fib_new_table(net, cfg.fc_table);
 881         if (!tb) {
 882                 err = -ENOBUFS;
 883                 goto errout;
 884         }
 885
 886         err = fib_table_insert(net, tb, &cfg, extack);
 887         if (!err && cfg.fc_type == RTN_LOCAL)
 888                 net->ipv4.fib_has_custom_local_routes = true;
 889 errout:
 890         return err;
 891 }
 892
 893 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
 894                           struct fib_dump_filter *filter,
 895                           struct netlink_callback *cb)
 896 {
 897         struct netlink_ext_ack *extack = cb->extack;
 898         struct nlattr *tb[RTA_MAX + 1];
 899         struct rtmsg *rtm;
 900         int err, i;
 901
 902         ASSERT_RTNL();
 903
 904         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
 905                 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
 906                 return -EINVAL;
 907         }
 908
 909         rtm = nlmsg_data(nlh);
 910         if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
 911             rtm->rtm_scope) {
 912                 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
 913                 return -EINVAL;
 914         }
 915
 916         if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
 917                 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
 918                 return -EINVAL;
 919         }
 920         if (rtm->rtm_flags & RTM_F_CLONED)
 921                 filter->dump_routes = false;
 922         else
 923                 filter->dump_exceptions = false;
 924
 925         filter->dump_all_families = (rtm->rtm_family == AF_UNSPEC);
 926         filter->flags    = rtm->rtm_flags;
 927         filter->protocol = rtm->rtm_protocol;
 928         filter->rt_type  = rtm->rtm_type;
 929         filter->table_id = rtm->rtm_table;
 930
 931         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
 932                                             rtm_ipv4_policy, extack);
 933         if (err < 0)
 934                 return err;
 935
 936         for (i = 0; i <= RTA_MAX; ++i) {
 937                 int ifindex;
 938
 939                 if (!tb[i])
 940                         continue;
 941
 942                 switch (i) {
 943                 case RTA_TABLE:
 944                         filter->table_id = nla_get_u32(tb[i]);
 945                         break;
 946                 case RTA_OIF:
 947                         ifindex = nla_get_u32(tb[i]);
 948                         filter->dev = __dev_get_by_index(net, ifindex);
 949                         if (!filter->dev)
 950                                 return -ENODEV;
 951                         break;
 952                 default:
 953                         NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
 954                         return -EINVAL;
 955                 }
 956         }
 957
 958         if (filter->flags || filter->protocol || filter->rt_type ||
 959             filter->table_id || filter->dev) {
 960                 filter->filter_set = 1;
 961                 cb->answer_flags = NLM_F_DUMP_FILTERED;
 962         }
 963
 964         return 0;
 965 }
 966 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
 967
 968 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 969 {
 970         struct fib_dump_filter filter = { .dump_routes = true,
 971                                           .dump_exceptions = true };
 972         const struct nlmsghdr *nlh = cb->nlh;
 973         struct net *net = sock_net(skb->sk);
 974         unsigned int h, s_h;
 975         unsigned int e = 0, s_e;
 976         struct fib_table *tb;
 977         struct hlist_head *head;
 978         int dumped = 0, err;
 979
 980         if (cb->strict_check) {
 981                 err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
 982                 if (err < 0)
 983                         return err;
 984         } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
 985                 struct rtmsg *rtm = nlmsg_data(nlh);
 986
 987                 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
 988         }
 989
 990         /* ipv4 does not use prefix flag */
 991         if (filter.flags & RTM_F_PREFIX)
 992                 return skb->len;
 993
 994         if (filter.table_id) {
 995                 tb = fib_get_table(net, filter.table_id);
 996                 if (!tb) {
 997                         if (filter.dump_all_families)
 998                                 return skb->len;
 999
1000                         NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1001                         return -ENOENT;
1002                 }
1003
1004                 err = fib_table_dump(tb, skb, cb, &filter);
1005                 return skb->len ? : err;
1006         }
1007
1008         s_h = cb->args[0];
1009         s_e = cb->args[1];
1010
1011         rcu_read_lock();
1012
1013         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1014                 e = 0;
1015                 head = &net->ipv4.fib_table_hash[h];
1016                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1017                         if (e < s_e)
1018                                 goto next;
1019                         if (dumped)
1020                                 memset(&cb->args[2], 0, sizeof(cb->args) -
1021                                                  2 * sizeof(cb->args[0]));
1022                         err = fib_table_dump(tb, skb, cb, &filter);
1023                         if (err < 0) {
1024                                 if (likely(skb->len))
1025                                         goto out;
1026
1027                                 goto out_err;
1028                         }
1029                         dumped = 1;
1030 next:
1031                         e++;
1032                 }
1033         }
1034 out:
1035         err = skb->len;
1036 out_err:
1037         rcu_read_unlock();
1038
1039         cb->args[1] = e;
1040         cb->args[0] = h;
1041
1042         return err;
1043 }
1044
1045 /* Prepare and feed intra-kernel routing request.
1046  * Really, it should be netlink message, but :-( netlink
1047  * can be not configured, so that we feed it directly
1048  * to fib engine. It is legal, because all events occur
1049  * only when netlink is already locked.
1050  */
1051 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1052                       struct in_ifaddr *ifa, u32 rt_priority)
1053 {
1054         struct net *net = dev_net(ifa->ifa_dev->dev);
1055         u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1056         struct fib_table *tb;
1057         struct fib_config cfg = {
1058                 .fc_protocol = RTPROT_KERNEL,
1059                 .fc_type = type,
1060                 .fc_dst = dst,
1061                 .fc_dst_len = dst_len,
1062                 .fc_priority = rt_priority,
1063                 .fc_prefsrc = ifa->ifa_local,
1064                 .fc_oif = ifa->ifa_dev->dev->ifindex,
1065                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1066                 .fc_nlinfo = {
1067                         .nl_net = net,
1068                 },
1069         };
1070
1071         if (!tb_id)
1072                 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1073
1074         tb = fib_new_table(net, tb_id);
1075         if (!tb)
1076                 return;
1077
1078         cfg.fc_table = tb->tb_id;
1079
1080         if (type != RTN_LOCAL)
1081                 cfg.fc_scope = RT_SCOPE_LINK;
1082         else
1083                 cfg.fc_scope = RT_SCOPE_HOST;
1084
1085         if (cmd == RTM_NEWROUTE)
1086                 fib_table_insert(net, tb, &cfg, NULL);
1087         else
1088                 fib_table_delete(net, tb, &cfg, NULL);
1089 }
1090
1091 void fib_add_ifaddr(struct in_ifaddr *ifa)
1092 {
1093         struct in_device *in_dev = ifa->ifa_dev;
1094         struct net_device *dev = in_dev->dev;
1095         struct in_ifaddr *prim = ifa;
1096         __be32 mask = ifa->ifa_mask;
1097         __be32 addr = ifa->ifa_local;
1098         __be32 prefix = ifa->ifa_address & mask;
1099
1100         if (ifa->ifa_flags & IFA_F_SECONDARY) {
1101                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
1102                 if (!prim) {
1103                         pr_warn("%s: bug: prim == NULL\n", __func__);
1104                         return;
1105                 }
1106         }
1107
1108         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1109
1110         if (!(dev->flags & IFF_UP))
1111                 return;
1112
1113         /* Add broadcast address, if it is explicitly assigned. */
1114         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
1115                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1116                           prim, 0);
1117
1118         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1119             (prefix != addr || ifa->ifa_prefixlen < 32)) {
1120                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1121                         fib_magic(RTM_NEWROUTE,
1122                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1123                                   prefix, ifa->ifa_prefixlen, prim,
1124                                   ifa->ifa_rt_priority);
1125
1126                 /* Add network specific broadcasts, when it takes a sense */
1127                 if (ifa->ifa_prefixlen < 31) {
1128                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1129                                   prim, 0);
1130                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1131                                   32, prim, 0);
1132                 }
1133         }
1134 }
1135
1136 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1137 {
1138         __be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1139         struct in_device *in_dev = ifa->ifa_dev;
1140         struct net_device *dev = in_dev->dev;
1141
1142         if (!(dev->flags & IFF_UP) ||
1143             ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1144             ipv4_is_zeronet(prefix) ||
1145             prefix == ifa->ifa_local || ifa->ifa_prefixlen == 32)
1146                 return;
1147
1148         /* add the new */
1149         fib_magic(RTM_NEWROUTE,
1150                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1151                   prefix, ifa->ifa_prefixlen, ifa, new_metric);
1152
1153         /* delete the old */
1154         fib_magic(RTM_DELROUTE,
1155                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1156                   prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1157 }
1158
1159 /* Delete primary or secondary address.
1160  * Optionally, on secondary address promotion consider the addresses
1161  * from subnet iprim as deleted, even if they are in device list.
1162  * In this case the secondary ifa can be in device list.
1163  */
1164 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1165 {
1166         struct in_device *in_dev = ifa->ifa_dev;
1167         struct net_device *dev = in_dev->dev;
1168         struct in_ifaddr *ifa1;
1169         struct in_ifaddr *prim = ifa, *prim1 = NULL;
1170         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1171         __be32 any = ifa->ifa_address & ifa->ifa_mask;
1172 #define LOCAL_OK        1
1173 #define BRD_OK          2
1174 #define BRD0_OK         4
1175 #define BRD1_OK         8
1176         unsigned int ok = 0;
1177         int subnet = 0;         /* Primary network */
1178         int gone = 1;           /* Address is missing */
1179         int same_prefsrc = 0;   /* Another primary with same IP */
1180
1181         if (ifa->ifa_flags & IFA_F_SECONDARY) {
1182                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1183                 if (!prim) {
1184                         /* if the device has been deleted, we don't perform
1185                          * address promotion
1186                          */
1187                         if (!in_dev->dead)
1188                                 pr_warn("%s: bug: prim == NULL\n", __func__);
1189                         return;
1190                 }
1191                 if (iprim && iprim != prim) {
1192                         pr_warn("%s: bug: iprim != prim\n", __func__);
1193                         return;
1194                 }
1195         } else if (!ipv4_is_zeronet(any) &&
1196                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1197                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1198                         fib_magic(RTM_DELROUTE,
1199                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1200                                   any, ifa->ifa_prefixlen, prim, 0);
1201                 subnet = 1;
1202         }
1203
1204         if (in_dev->dead)
1205                 goto no_promotions;
1206
1207         /* Deletion is more complicated than add.
1208          * We should take care of not to delete too much :-)
1209          *
1210          * Scan address list to be sure that addresses are really gone.
1211          */
1212         rcu_read_lock();
1213         in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1214                 if (ifa1 == ifa) {
1215                         /* promotion, keep the IP */
1216                         gone = 0;
1217                         continue;
1218                 }
1219                 /* Ignore IFAs from our subnet */
1220                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1221                     inet_ifa_match(ifa1->ifa_address, iprim))
1222                         continue;
1223
1224                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
1225                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1226                         /* Another address from our subnet? */
1227                         if (ifa1->ifa_mask == prim->ifa_mask &&
1228                             inet_ifa_match(ifa1->ifa_address, prim))
1229                                 prim1 = prim;
1230                         else {
1231                                 /* We reached the secondaries, so
1232                                  * same_prefsrc should be determined.
1233                                  */
1234                                 if (!same_prefsrc)
1235                                         continue;
1236                                 /* Search new prim1 if ifa1 is not
1237                                  * using the current prim1
1238                                  */
1239                                 if (!prim1 ||
1240                                     ifa1->ifa_mask != prim1->ifa_mask ||
1241                                     !inet_ifa_match(ifa1->ifa_address, prim1))
1242                                         prim1 = inet_ifa_byprefix(in_dev,
1243                                                         ifa1->ifa_address,
1244                                                         ifa1->ifa_mask);
1245                                 if (!prim1)
1246                                         continue;
1247                                 if (prim1->ifa_local != prim->ifa_local)
1248                                         continue;
1249                         }
1250                 } else {
1251                         if (prim->ifa_local != ifa1->ifa_local)
1252                                 continue;
1253                         prim1 = ifa1;
1254                         if (prim != prim1)
1255                                 same_prefsrc = 1;
1256                 }
1257                 if (ifa->ifa_local == ifa1->ifa_local)
1258                         ok |= LOCAL_OK;
1259                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1260                         ok |= BRD_OK;
1261                 if (brd == ifa1->ifa_broadcast)
1262                         ok |= BRD1_OK;
1263                 if (any == ifa1->ifa_broadcast)
1264                         ok |= BRD0_OK;
1265                 /* primary has network specific broadcasts */
1266                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1267                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1268                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1269
1270                         if (!ipv4_is_zeronet(any1)) {
1271                                 if (ifa->ifa_broadcast == brd1 ||
1272                                     ifa->ifa_broadcast == any1)
1273                                         ok |= BRD_OK;
1274                                 if (brd == brd1 || brd == any1)
1275                                         ok |= BRD1_OK;
1276                                 if (any == brd1 || any == any1)
1277                                         ok |= BRD0_OK;
1278                         }
1279                 }
1280         }
1281         rcu_read_unlock();
1282
1283 no_promotions:
1284         if (!(ok & BRD_OK))
1285                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1286                           prim, 0);
1287         if (subnet && ifa->ifa_prefixlen < 31) {
1288                 if (!(ok & BRD1_OK))
1289                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1290                                   prim, 0);
1291                 if (!(ok & BRD0_OK))
1292                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1293                                   prim, 0);
1294         }
1295         if (!(ok & LOCAL_OK)) {
1296                 unsigned int addr_type;
1297
1298                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1299
1300                 /* Check, that this local address finally disappeared. */
1301                 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1302                                                      ifa->ifa_local);
1303                 if (gone && addr_type != RTN_LOCAL) {
1304                         /* And the last, but not the least thing.
1305                          * We must flush stray FIB entries.
1306                          *
1307                          * First of all, we scan fib_info list searching
1308                          * for stray nexthop entries, then ignite fib_flush.
1309                          */
1310                         if (fib_sync_down_addr(dev, ifa->ifa_local))
1311                                 fib_flush(dev_net(dev));
1312                 }
1313         }
1314 #undef LOCAL_OK
1315 #undef BRD_OK
1316 #undef BRD0_OK
1317 #undef BRD1_OK
1318 }
1319
1320 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1321 {
1322
1323         struct fib_result       res;
1324         struct flowi4           fl4 = {
1325                 .flowi4_mark = frn->fl_mark,
1326                 .daddr = frn->fl_addr,
1327                 .flowi4_tos = frn->fl_tos,
1328                 .flowi4_scope = frn->fl_scope,
1329         };
1330         struct fib_table *tb;
1331
1332         rcu_read_lock();
1333
1334         tb = fib_get_table(net, frn->tb_id_in);
1335
1336         frn->err = -ENOENT;
1337         if (tb) {
1338                 local_bh_disable();
1339
1340                 frn->tb_id = tb->tb_id;
1341                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1342
1343                 if (!frn->err) {
1344                         frn->prefixlen = res.prefixlen;
1345                         frn->nh_sel = res.nh_sel;
1346                         frn->type = res.type;
1347                         frn->scope = res.scope;
1348                 }
1349                 local_bh_enable();
1350         }
1351
1352         rcu_read_unlock();
1353 }
1354
1355 static void nl_fib_input(struct sk_buff *skb)
1356 {
1357         struct net *net;
1358         struct fib_result_nl *frn;
1359         struct nlmsghdr *nlh;
1360         u32 portid;
1361
1362         net = sock_net(skb->sk);
1363         nlh = nlmsg_hdr(skb);
1364         if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1365             skb->len < nlh->nlmsg_len ||
1366             nlmsg_len(nlh) < sizeof(*frn))
1367                 return;
1368
1369         skb = netlink_skb_clone(skb, GFP_KERNEL);
1370         if (!skb)
1371                 return;
1372         nlh = nlmsg_hdr(skb);
1373
1374         frn = (struct fib_result_nl *) nlmsg_data(nlh);
1375         nl_fib_lookup(net, frn);
1376
1377         portid = NETLINK_CB(skb).portid;      /* netlink portid */
1378         NETLINK_CB(skb).portid = 0;        /* from kernel */
1379         NETLINK_CB(skb).dst_group = 0;  /* unicast */
1380         netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1381 }
1382
1383 static int __net_init nl_fib_lookup_init(struct net *net)
1384 {
1385         struct sock *sk;
1386         struct netlink_kernel_cfg cfg = {
1387                 .input  = nl_fib_input,
1388         };
1389
1390         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1391         if (!sk)
1392                 return -EAFNOSUPPORT;
1393         net->ipv4.fibnl = sk;
1394         return 0;
1395 }
1396
1397 static void nl_fib_lookup_exit(struct net *net)
1398 {
1399         netlink_kernel_release(net->ipv4.fibnl);
1400         net->ipv4.fibnl = NULL;
1401 }
1402
1403 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1404                            bool force)
1405 {
1406         if (fib_sync_down_dev(dev, event, force))
1407                 fib_flush(dev_net(dev));
1408         else
1409                 rt_cache_flush(dev_net(dev));
1410         arp_ifdown(dev);
1411 }
1412
1413 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1414 {
1415         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1416         struct net_device *dev = ifa->ifa_dev->dev;
1417         struct net *net = dev_net(dev);
1418
1419         switch (event) {
1420         case NETDEV_UP:
1421                 fib_add_ifaddr(ifa);
1422 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1423                 fib_sync_up(dev, RTNH_F_DEAD);
1424 #endif
1425                 atomic_inc(&net->ipv4.dev_addr_genid);
1426                 rt_cache_flush(dev_net(dev));
1427                 break;
1428         case NETDEV_DOWN:
1429                 fib_del_ifaddr(ifa, NULL);
1430                 atomic_inc(&net->ipv4.dev_addr_genid);
1431                 if (!ifa->ifa_dev->ifa_list) {
1432                         /* Last address was deleted from this interface.
1433                          * Disable IP.
1434                          */
1435                         fib_disable_ip(dev, event, true);
1436                 } else {
1437                         rt_cache_flush(dev_net(dev));
1438                 }
1439                 break;
1440         }
1441         return NOTIFY_DONE;
1442 }
1443
1444 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1445 {
1446         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1447         struct netdev_notifier_changeupper_info *upper_info = ptr;
1448         struct netdev_notifier_info_ext *info_ext = ptr;
1449         struct in_device *in_dev;
1450         struct net *net = dev_net(dev);
1451         struct in_ifaddr *ifa;
1452         unsigned int flags;
1453
1454         if (event == NETDEV_UNREGISTER) {
1455                 fib_disable_ip(dev, event, true);
1456                 rt_flush_dev(dev);
1457                 return NOTIFY_DONE;
1458         }
1459
1460         in_dev = __in_dev_get_rtnl(dev);
1461         if (!in_dev)
1462                 return NOTIFY_DONE;
1463
1464         switch (event) {
1465         case NETDEV_UP:
1466                 in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1467                         fib_add_ifaddr(ifa);
1468                 }
1469 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1470                 fib_sync_up(dev, RTNH_F_DEAD);
1471 #endif
1472                 atomic_inc(&net->ipv4.dev_addr_genid);
1473                 rt_cache_flush(net);
1474                 break;
1475         case NETDEV_DOWN:
1476                 fib_disable_ip(dev, event, false);
1477                 break;
1478         case NETDEV_CHANGE:
1479                 flags = dev_get_flags(dev);
1480                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1481                         fib_sync_up(dev, RTNH_F_LINKDOWN);
1482                 else
1483                         fib_sync_down_dev(dev, event, false);
1484                 rt_cache_flush(net);
1485                 break;
1486         case NETDEV_CHANGEMTU:
1487                 fib_sync_mtu(dev, info_ext->ext.mtu);
1488                 rt_cache_flush(net);
1489                 break;
1490         case NETDEV_CHANGEUPPER:
1491                 upper_info = ptr;
1492                 /* flush all routes if dev is linked to or unlinked from
1493                  * an L3 master device (e.g., VRF)
1494                  */
1495                 if (upper_info->upper_dev &&
1496                     netif_is_l3_master(upper_info->upper_dev))
1497                         fib_disable_ip(dev, NETDEV_DOWN, true);
1498                 break;
1499         }
1500         return NOTIFY_DONE;
1501 }
1502
1503 static struct notifier_block fib_inetaddr_notifier = {
1504         .notifier_call = fib_inetaddr_event,
1505 };
1506
1507 static struct notifier_block fib_netdev_notifier = {
1508         .notifier_call = fib_netdev_event,
1509 };
1510
1511 static int __net_init ip_fib_net_init(struct net *net)
1512 {
1513         int err;
1514         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1515
1516         err = fib4_notifier_init(net);
1517         if (err)
1518                 return err;
1519
1520         /* Avoid false sharing : Use at least a full cache line */
1521         size = max_t(size_t, size, L1_CACHE_BYTES);
1522
1523         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1524         if (!net->ipv4.fib_table_hash) {
1525                 err = -ENOMEM;
1526                 goto err_table_hash_alloc;
1527         }
1528
1529         err = fib4_rules_init(net);
1530         if (err < 0)
1531                 goto err_rules_init;
1532         return 0;
1533
1534 err_rules_init:
1535         kfree(net->ipv4.fib_table_hash);
1536 err_table_hash_alloc:
1537         fib4_notifier_exit(net);
1538         return err;
1539 }
1540
1541 static void ip_fib_net_exit(struct net *net)
1542 {
1543         int i;
1544
1545         rtnl_lock();
1546 #ifdef CONFIG_IP_MULTIPLE_TABLES
1547         RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1548         RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1549 #endif
1550         /* Destroy the tables in reverse order to guarantee that the
1551          * local table, ID 255, is destroyed before the main table, ID
1552          * 254. This is necessary as the local table may contain
1553          * references to data contained in the main table.
1554          */
1555         for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1556                 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1557                 struct hlist_node *tmp;
1558                 struct fib_table *tb;
1559
1560                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1561                         hlist_del(&tb->tb_hlist);
1562                         fib_table_flush(net, tb, true);
1563                         fib_free_table(tb);
1564                 }
1565         }
1566
1567 #ifdef CONFIG_IP_MULTIPLE_TABLES
1568         fib4_rules_exit(net);
1569 #endif
1570         rtnl_unlock();
1571         kfree(net->ipv4.fib_table_hash);
1572         fib4_notifier_exit(net);
1573 }
1574
1575 static int __net_init fib_net_init(struct net *net)
1576 {
1577         int error;
1578
1579 #ifdef CONFIG_IP_ROUTE_CLASSID
1580         net->ipv4.fib_num_tclassid_users = 0;
1581 #endif
1582         error = ip_fib_net_init(net);
1583         if (error < 0)
1584                 goto out;
1585         error = nl_fib_lookup_init(net);
1586         if (error < 0)
1587                 goto out_nlfl;
1588         error = fib_proc_init(net);
1589         if (error < 0)
1590                 goto out_proc;
1591 out:
1592         return error;
1593
1594 out_proc:
1595         nl_fib_lookup_exit(net);
1596 out_nlfl:
1597         ip_fib_net_exit(net);
1598         goto out;
1599 }
1600
1601 static void __net_exit fib_net_exit(struct net *net)
1602 {
1603         fib_proc_exit(net);
1604         nl_fib_lookup_exit(net);
1605         ip_fib_net_exit(net);
1606 }
1607
1608 static struct pernet_operations fib_net_ops = {
1609         .init = fib_net_init,
1610         .exit = fib_net_exit,
1611 };
1612
1613 void __init ip_fib_init(void)
1614 {
1615         fib_trie_init();
1616
1617         register_pernet_subsys(&fib_net_ops);
1618
1619         register_netdevice_notifier(&fib_netdev_notifier);
1620         register_inetaddr_notifier(&fib_inetaddr_notifier);
1621
1622         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1623         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1624         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1625 }