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                                  lockdep_rtnl_is_held()) {
 129                 if (tb->tb_id == id)
 130                         return tb;
 131         }
 132         return NULL;
 133 }
 134
 135 static bool fib4_has_custom_rules(struct net *net)
 136 {
 137         return net->ipv4.fib_has_custom_rules;
 138 }
 139 #endif /* CONFIG_IP_MULTIPLE_TABLES */
 140
 141 static void fib_replace_table(struct net *net, struct fib_table *old,
 142                               struct fib_table *new)
 143 {
 144 #ifdef CONFIG_IP_MULTIPLE_TABLES
 145         switch (new->tb_id) {
 146         case RT_TABLE_MAIN:
 147                 rcu_assign_pointer(net->ipv4.fib_main, new);
 148                 break;
 149         case RT_TABLE_DEFAULT:
 150                 rcu_assign_pointer(net->ipv4.fib_default, new);
 151                 break;
 152         default:
 153                 break;
 154         }
 155
 156 #endif
 157         /* replace the old table in the hlist */
 158         hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
 159 }
 160
 161 int fib_unmerge(struct net *net)
 162 {
 163         struct fib_table *old, *new, *main_table;
 164
 165         /* attempt to fetch local table if it has been allocated */
 166         old = fib_get_table(net, RT_TABLE_LOCAL);
 167         if (!old)
 168                 return 0;
 169
 170         new = fib_trie_unmerge(old);
 171         if (!new)
 172                 return -ENOMEM;
 173
 174         /* table is already unmerged */
 175         if (new == old)
 176                 return 0;
 177
 178         /* replace merged table with clean table */
 179         fib_replace_table(net, old, new);
 180         fib_free_table(old);
 181
 182         /* attempt to fetch main table if it has been allocated */
 183         main_table = fib_get_table(net, RT_TABLE_MAIN);
 184         if (!main_table)
 185                 return 0;
 186
 187         /* flush local entries from main table */
 188         fib_table_flush_external(main_table);
 189
 190         return 0;
 191 }
 192
 193 void fib_flush(struct net *net)
 194 {
 195         int flushed = 0;
 196         unsigned int h;
 197
 198         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
 199                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
 200                 struct hlist_node *tmp;
 201                 struct fib_table *tb;
 202
 203                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
 204                         flushed += fib_table_flush(net, tb, false);
 205         }
 206
 207         if (flushed)
 208                 rt_cache_flush(net);
 209 }
 210
 211 /*
 212  * Find address type as if only "dev" was present in the system. If
 213  * on_dev is NULL then all interfaces are taken into consideration.
 214  */
 215 static inline unsigned int __inet_dev_addr_type(struct net *net,
 216                                                 const struct net_device *dev,
 217                                                 __be32 addr, u32 tb_id)
 218 {
 219         struct flowi4           fl4 = { .daddr = addr };
 220         struct fib_result       res;
 221         unsigned int ret = RTN_BROADCAST;
 222         struct fib_table *table;
 223
 224         if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
 225                 return RTN_BROADCAST;
 226         if (ipv4_is_multicast(addr))
 227                 return RTN_MULTICAST;
 228
 229         rcu_read_lock();
 230
 231         table = fib_get_table(net, tb_id);
 232         if (table) {
 233                 ret = RTN_UNICAST;
 234                 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
 235                         struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0);
 236
 237                         if (!dev || dev == nhc->nhc_dev)
 238                                 ret = res.type;
 239                 }
 240         }
 241
 242         rcu_read_unlock();
 243         return ret;
 244 }
 245
 246 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
 247 {
 248         return __inet_dev_addr_type(net, NULL, addr, tb_id);
 249 }
 250 EXPORT_SYMBOL(inet_addr_type_table);
 251
 252 unsigned int inet_addr_type(struct net *net, __be32 addr)
 253 {
 254         return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
 255 }
 256 EXPORT_SYMBOL(inet_addr_type);
 257
 258 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
 259                                 __be32 addr)
 260 {
 261         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
 262
 263         return __inet_dev_addr_type(net, dev, addr, rt_table);
 264 }
 265 EXPORT_SYMBOL(inet_dev_addr_type);
 266
 267 /* inet_addr_type with dev == NULL but using the table from a dev
 268  * if one is associated
 269  */
 270 unsigned int inet_addr_type_dev_table(struct net *net,
 271                                       const struct net_device *dev,
 272                                       __be32 addr)
 273 {
 274         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
 275
 276         return __inet_dev_addr_type(net, NULL, addr, rt_table);
 277 }
 278 EXPORT_SYMBOL(inet_addr_type_dev_table);
 279
 280 __be32 fib_compute_spec_dst(struct sk_buff *skb)
 281 {
 282         struct net_device *dev = skb->dev;
 283         struct in_device *in_dev;
 284         struct fib_result res;
 285         struct rtable *rt;
 286         struct net *net;
 287         int scope;
 288
 289         rt = skb_rtable(skb);
 290         if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
 291             RTCF_LOCAL)
 292                 return ip_hdr(skb)->daddr;
 293
 294         in_dev = __in_dev_get_rcu(dev);
 295
 296         net = dev_net(dev);
 297
 298         scope = RT_SCOPE_UNIVERSE;
 299         if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
 300                 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
 301                 struct flowi4 fl4 = {
 302                         .flowi4_iif = LOOPBACK_IFINDEX,
 303                         .flowi4_oif = l3mdev_master_ifindex_rcu(dev),
 304                         .daddr = ip_hdr(skb)->saddr,
 305                         .flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
 306                         .flowi4_scope = scope,
 307                         .flowi4_mark = vmark ? skb->mark : 0,
 308                 };
 309                 if (!fib_lookup(net, &fl4, &res, 0))
 310                         return fib_result_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 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
 319 {
 320         bool dev_match = false;
 321 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 322         int ret;
 323
 324         for (ret = 0; ret < fib_info_num_path(fi); ret++) {
 325                 const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
 326
 327                 if (nhc->nhc_dev == dev) {
 328                         dev_match = true;
 329                         break;
 330                 } else if (l3mdev_master_ifindex_rcu(nhc->nhc_dev) == dev->ifindex) {
 331                         dev_match = true;
 332                         break;
 333                 }
 334         }
 335 #else
 336         if (fib_info_nhc(fi, 0)->nhc_dev == dev)
 337                 dev_match = true;
 338 #endif
 339
 340         return dev_match;
 341 }
 342 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
 343
 344 /* Given (packet source, input interface) and optional (dst, oif, tos):
 345  * - (main) check, that source is valid i.e. not broadcast or our local
 346  *   address.
 347  * - figure out what "logical" interface this packet arrived
 348  *   and calculate "specific destination" address.
 349  * - check, that packet arrived from expected physical interface.
 350  * called with rcu_read_lock()
 351  */
 352 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 353                                  u8 tos, int oif, struct net_device *dev,
 354                                  int rpf, struct in_device *idev, u32 *itag)
 355 {
 356         struct net *net = dev_net(dev);
 357         struct flow_keys flkeys;
 358         int ret, no_addr;
 359         struct fib_result res;
 360         struct flowi4 fl4;
 361         bool dev_match;
 362
 363         fl4.flowi4_oif = 0;
 364         fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
 365         if (!fl4.flowi4_iif)
 366                 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
 367         fl4.daddr = src;
 368         fl4.saddr = dst;
 369         fl4.flowi4_tos = tos;
 370         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 371         fl4.flowi4_tun_key.tun_id = 0;
 372         fl4.flowi4_flags = 0;
 373         fl4.flowi4_uid = sock_net_uid(net, NULL);
 374
 375         no_addr = idev->ifa_list == NULL;
 376
 377         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
 378         if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
 379                 fl4.flowi4_proto = 0;
 380                 fl4.fl4_sport = 0;
 381                 fl4.fl4_dport = 0;
 382         }
 383
 384         if (fib_lookup(net, &fl4, &res, 0))
 385                 goto last_resort;
 386         if (res.type != RTN_UNICAST &&
 387             (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
 388                 goto e_inval;
 389         fib_combine_itag(itag, &res);
 390
 391         dev_match = fib_info_nh_uses_dev(res.fi, dev);
 392         /* This is not common, loopback packets retain skb_dst so normally they
 393          * would not even hit this slow path.
 394          */
 395         dev_match = dev_match || (res.type == RTN_LOCAL &&
 396                                   dev == net->loopback_dev);
 397         if (dev_match) {
 398                 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
 399                 return ret;
 400         }
 401         if (no_addr)
 402                 goto last_resort;
 403         if (rpf == 1)
 404                 goto e_rpf;
 405         fl4.flowi4_oif = dev->ifindex;
 406
 407         ret = 0;
 408         if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
 409                 if (res.type == RTN_UNICAST)
 410                         ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
 411         }
 412         return ret;
 413
 414 last_resort:
 415         if (rpf)
 416                 goto e_rpf;
 417         *itag = 0;
 418         return 0;
 419
 420 e_inval:
 421         return -EINVAL;
 422 e_rpf:
 423         return -EXDEV;
 424 }
 425
 426 /* Ignore rp_filter for packets protected by IPsec. */
 427 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 428                         u8 tos, int oif, struct net_device *dev,
 429                         struct in_device *idev, u32 *itag)
 430 {
 431         int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
 432         struct net *net = dev_net(dev);
 433
 434         if (!r && !fib_num_tclassid_users(net) &&
 435             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
 436                 if (IN_DEV_ACCEPT_LOCAL(idev))
 437                         goto ok;
 438                 /* with custom local routes in place, checking local addresses
 439                  * only will be too optimistic, with custom rules, checking
 440                  * local addresses only can be too strict, e.g. due to vrf
 441                  */
 442                 if (net->ipv4.fib_has_custom_local_routes ||
 443                     fib4_has_custom_rules(net))
 444                         goto full_check;
 445                 if (inet_lookup_ifaddr_rcu(net, src))
 446                         return -EINVAL;
 447
 448 ok:
 449                 *itag = 0;
 450                 return 0;
 451         }
 452
 453 full_check:
 454         return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
 455 }
 456
 457 static inline __be32 sk_extract_addr(struct sockaddr *addr)
 458 {
 459         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 460 }
 461
 462 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
 463 {
 464         struct nlattr *nla;
 465
 466         nla = (struct nlattr *) ((char *) mx + len);
 467         nla->nla_type = type;
 468         nla->nla_len = nla_attr_size(4);
 469         *(u32 *) nla_data(nla) = value;
 470
 471         return len + nla_total_size(4);
 472 }
 473
 474 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
 475                                  struct fib_config *cfg)
 476 {
 477         __be32 addr;
 478         int plen;
 479
 480         memset(cfg, 0, sizeof(*cfg));
 481         cfg->fc_nlinfo.nl_net = net;
 482
 483         if (rt->rt_dst.sa_family != AF_INET)
 484                 return -EAFNOSUPPORT;
 485
 486         /*
 487          * Check mask for validity:
 488          * a) it must be contiguous.
 489          * b) destination must have all host bits clear.
 490          * c) if application forgot to set correct family (AF_INET),
 491          *    reject request unless it is absolutely clear i.e.
 492          *    both family and mask are zero.
 493          */
 494         plen = 32;
 495         addr = sk_extract_addr(&rt->rt_dst);
 496         if (!(rt->rt_flags & RTF_HOST)) {
 497                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
 498
 499                 if (rt->rt_genmask.sa_family != AF_INET) {
 500                         if (mask || rt->rt_genmask.sa_family)
 501                                 return -EAFNOSUPPORT;
 502                 }
 503
 504                 if (bad_mask(mask, addr))
 505                         return -EINVAL;
 506
 507                 plen = inet_mask_len(mask);
 508         }
 509
 510         cfg->fc_dst_len = plen;
 511         cfg->fc_dst = addr;
 512
 513         if (cmd != SIOCDELRT) {
 514                 cfg->fc_nlflags = NLM_F_CREATE;
 515                 cfg->fc_protocol = RTPROT_BOOT;
 516         }
 517
 518         if (rt->rt_metric)
 519                 cfg->fc_priority = rt->rt_metric - 1;
 520
 521         if (rt->rt_flags & RTF_REJECT) {
 522                 cfg->fc_scope = RT_SCOPE_HOST;
 523                 cfg->fc_type = RTN_UNREACHABLE;
 524                 return 0;
 525         }
 526
 527         cfg->fc_scope = RT_SCOPE_NOWHERE;
 528         cfg->fc_type = RTN_UNICAST;
 529
 530         if (rt->rt_dev) {
 531                 char *colon;
 532                 struct net_device *dev;
 533                 char devname[IFNAMSIZ];
 534
 535                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
 536                         return -EFAULT;
 537
 538                 devname[IFNAMSIZ-1] = 0;
 539                 colon = strchr(devname, ':');
 540                 if (colon)
 541                         *colon = 0;
 542                 dev = __dev_get_by_name(net, devname);
 543                 if (!dev)
 544                         return -ENODEV;
 545                 cfg->fc_oif = dev->ifindex;
 546                 cfg->fc_table = l3mdev_fib_table(dev);
 547                 if (colon) {
 548                         const struct in_ifaddr *ifa;
 549                         struct in_device *in_dev;
 550
 551                         in_dev = __in_dev_get_rtnl(dev);
 552                         if (!in_dev)
 553                                 return -ENODEV;
 554
 555                         *colon = ':';
 556
 557                         rcu_read_lock();
 558                         in_dev_for_each_ifa_rcu(ifa, in_dev) {
 559                                 if (strcmp(ifa->ifa_label, devname) == 0)
 560                                         break;
 561                         }
 562                         rcu_read_unlock();
 563
 564                         if (!ifa)
 565                                 return -ENODEV;
 566                         cfg->fc_prefsrc = ifa->ifa_local;
 567                 }
 568         }
 569
 570         addr = sk_extract_addr(&rt->rt_gateway);
 571         if (rt->rt_gateway.sa_family == AF_INET && addr) {
 572                 unsigned int addr_type;
 573
 574                 cfg->fc_gw4 = addr;
 575                 cfg->fc_gw_family = AF_INET;
 576                 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
 577                 if (rt->rt_flags & RTF_GATEWAY &&
 578                     addr_type == RTN_UNICAST)
 579                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
 580         }
 581
 582         if (cmd == SIOCDELRT)
 583                 return 0;
 584
 585         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
 586                 return -EINVAL;
 587
 588         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 589                 cfg->fc_scope = RT_SCOPE_LINK;
 590
 591         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 592                 struct nlattr *mx;
 593                 int len = 0;
 594
 595                 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
 596                 if (!mx)
 597                         return -ENOMEM;
 598
 599                 if (rt->rt_flags & RTF_MTU)
 600                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 601
 602                 if (rt->rt_flags & RTF_WINDOW)
 603                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 604
 605                 if (rt->rt_flags & RTF_IRTT)
 606                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 607
 608                 cfg->fc_mx = mx;
 609                 cfg->fc_mx_len = len;
 610         }
 611
 612         return 0;
 613 }
 614
 615 /*
 616  * Handle IP routing ioctl calls.
 617  * These are used to manipulate the routing tables
 618  */
 619 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
 620 {
 621         struct fib_config cfg;
 622         int err;
 623
 624         switch (cmd) {
 625         case SIOCADDRT:         /* Add a route */
 626         case SIOCDELRT:         /* Delete a route */
 627                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 628                         return -EPERM;
 629
 630                 rtnl_lock();
 631                 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
 632                 if (err == 0) {
 633                         struct fib_table *tb;
 634
 635                         if (cmd == SIOCDELRT) {
 636                                 tb = fib_get_table(net, cfg.fc_table);
 637                                 if (tb)
 638                                         err = fib_table_delete(net, tb, &cfg,
 639                                                                NULL);
 640                                 else
 641                                         err = -ESRCH;
 642                         } else {
 643                                 tb = fib_new_table(net, cfg.fc_table);
 644                                 if (tb)
 645                                         err = fib_table_insert(net, tb,
 646                                                                &cfg, NULL);
 647                                 else
 648                                         err = -ENOBUFS;
 649                         }
 650
 651                         /* allocated by rtentry_to_fib_config() */
 652                         kfree(cfg.fc_mx);
 653                 }
 654                 rtnl_unlock();
 655                 return err;
 656         }
 657         return -EINVAL;
 658 }
 659
 660 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 661         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
 662         [RTA_DST]               = { .type = NLA_U32 },
 663         [RTA_SRC]               = { .type = NLA_U32 },
 664         [RTA_IIF]               = { .type = NLA_U32 },
 665         [RTA_OIF]               = { .type = NLA_U32 },
 666         [RTA_GATEWAY]           = { .type = NLA_U32 },
 667         [RTA_PRIORITY]          = { .type = NLA_U32 },
 668         [RTA_PREFSRC]           = { .type = NLA_U32 },
 669         [RTA_METRICS]           = { .type = NLA_NESTED },
 670         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 671         [RTA_FLOW]              = { .type = NLA_U32 },
 672         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
 673         [RTA_ENCAP]             = { .type = NLA_NESTED },
 674         [RTA_UID]               = { .type = NLA_U32 },
 675         [RTA_MARK]              = { .type = NLA_U32 },
 676         [RTA_TABLE]             = { .type = NLA_U32 },
 677         [RTA_IP_PROTO]          = { .type = NLA_U8 },
 678         [RTA_SPORT]             = { .type = NLA_U16 },
 679         [RTA_DPORT]             = { .type = NLA_U16 },
 680         [RTA_NH_ID]             = { .type = NLA_U32 },
 681 };
 682
 683 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
 684                     struct netlink_ext_ack *extack)
 685 {
 686         struct rtvia *via;
 687         int alen;
 688
 689         if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
 690                 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
 691                 return -EINVAL;
 692         }
 693
 694         via = nla_data(nla);
 695         alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
 696
 697         switch (via->rtvia_family) {
 698         case AF_INET:
 699                 if (alen != sizeof(__be32)) {
 700                         NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
 701                         return -EINVAL;
 702                 }
 703                 cfg->fc_gw_family = AF_INET;
 704                 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
 705                 break;
 706         case AF_INET6:
 707 #ifdef CONFIG_IPV6
 708                 if (alen != sizeof(struct in6_addr)) {
 709                         NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
 710                         return -EINVAL;
 711                 }
 712                 cfg->fc_gw_family = AF_INET6;
 713                 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
 714 #else
 715                 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
 716                 return -EINVAL;
 717 #endif
 718                 break;
 719         default:
 720                 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
 721                 return -EINVAL;
 722         }
 723
 724         return 0;
 725 }
 726
 727 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 728                              struct nlmsghdr *nlh, struct fib_config *cfg,
 729                              struct netlink_ext_ack *extack)
 730 {
 731         bool has_gw = false, has_via = false;
 732         struct nlattr *attr;
 733         int err, remaining;
 734         struct rtmsg *rtm;
 735
 736         err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
 737                                         rtm_ipv4_policy, extack);
 738         if (err < 0)
 739                 goto errout;
 740
 741         memset(cfg, 0, sizeof(*cfg));
 742
 743         rtm = nlmsg_data(nlh);
 744         cfg->fc_dst_len = rtm->rtm_dst_len;
 745         cfg->fc_tos = rtm->rtm_tos;
 746         cfg->fc_table = rtm->rtm_table;
 747         cfg->fc_protocol = rtm->rtm_protocol;
 748         cfg->fc_scope = rtm->rtm_scope;
 749         cfg->fc_type = rtm->rtm_type;
 750         cfg->fc_flags = rtm->rtm_flags;
 751         cfg->fc_nlflags = nlh->nlmsg_flags;
 752
 753         cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
 754         cfg->fc_nlinfo.nlh = nlh;
 755         cfg->fc_nlinfo.nl_net = net;
 756
 757         if (cfg->fc_type > RTN_MAX) {
 758                 NL_SET_ERR_MSG(extack, "Invalid route type");
 759                 err = -EINVAL;
 760                 goto errout;
 761         }
 762
 763         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 764                 switch (nla_type(attr)) {
 765                 case RTA_DST:
 766                         cfg->fc_dst = nla_get_be32(attr);
 767                         break;
 768                 case RTA_OIF:
 769                         cfg->fc_oif = nla_get_u32(attr);
 770                         break;
 771                 case RTA_GATEWAY:
 772                         has_gw = true;
 773                         cfg->fc_gw4 = nla_get_be32(attr);
 774                         if (cfg->fc_gw4)
 775                                 cfg->fc_gw_family = AF_INET;
 776                         break;
 777                 case RTA_VIA:
 778                         has_via = true;
 779                         err = fib_gw_from_via(cfg, attr, extack);
 780                         if (err)
 781                                 goto errout;
 782                         break;
 783                 case RTA_PRIORITY:
 784                         cfg->fc_priority = nla_get_u32(attr);
 785                         break;
 786                 case RTA_PREFSRC:
 787                         cfg->fc_prefsrc = nla_get_be32(attr);
 788                         break;
 789                 case RTA_METRICS:
 790                         cfg->fc_mx = nla_data(attr);
 791                         cfg->fc_mx_len = nla_len(attr);
 792                         break;
 793                 case RTA_MULTIPATH:
 794                         err = lwtunnel_valid_encap_type_attr(nla_data(attr),
 795                                                              nla_len(attr),
 796                                                              extack);
 797                         if (err < 0)
 798                                 goto errout;
 799                         cfg->fc_mp = nla_data(attr);
 800                         cfg->fc_mp_len = nla_len(attr);
 801                         break;
 802                 case RTA_FLOW:
 803                         cfg->fc_flow = nla_get_u32(attr);
 804                         break;
 805                 case RTA_TABLE:
 806                         cfg->fc_table = nla_get_u32(attr);
 807                         break;
 808                 case RTA_ENCAP:
 809                         cfg->fc_encap = attr;
 810                         break;
 811                 case RTA_ENCAP_TYPE:
 812                         cfg->fc_encap_type = nla_get_u16(attr);
 813                         err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
 814                                                         extack);
 815                         if (err < 0)
 816                                 goto errout;
 817                         break;
 818                 case RTA_NH_ID:
 819                         cfg->fc_nh_id = nla_get_u32(attr);
 820                         break;
 821                 }
 822         }
 823
 824         if (cfg->fc_nh_id) {
 825                 if (cfg->fc_oif || cfg->fc_gw_family ||
 826                     cfg->fc_encap || cfg->fc_mp) {
 827                         NL_SET_ERR_MSG(extack,
 828                                        "Nexthop specification and nexthop id are mutually exclusive");
 829                         return -EINVAL;
 830                 }
 831         }
 832
 833         if (has_gw && has_via) {
 834                 NL_SET_ERR_MSG(extack,
 835                                "Nexthop configuration can not contain both GATEWAY and VIA");
 836                 goto errout;
 837         }
 838
 839         return 0;
 840 errout:
 841         return err;
 842 }
 843
 844 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 845                              struct netlink_ext_ack *extack)
 846 {
 847         struct net *net = sock_net(skb->sk);
 848         struct fib_config cfg;
 849         struct fib_table *tb;
 850         int err;
 851
 852         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 853         if (err < 0)
 854                 goto errout;
 855
 856         if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
 857                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
 858                 err = -EINVAL;
 859                 goto errout;
 860         }
 861
 862         tb = fib_get_table(net, cfg.fc_table);
 863         if (!tb) {
 864                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
 865                 err = -ESRCH;
 866                 goto errout;
 867         }
 868
 869         err = fib_table_delete(net, tb, &cfg, extack);
 870 errout:
 871         return err;
 872 }
 873
 874 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 875                              struct netlink_ext_ack *extack)
 876 {
 877         struct net *net = sock_net(skb->sk);
 878         struct fib_config cfg;
 879         struct fib_table *tb;
 880         int err;
 881
 882         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 883         if (err < 0)
 884                 goto errout;
 885
 886         tb = fib_new_table(net, cfg.fc_table);
 887         if (!tb) {
 888                 err = -ENOBUFS;
 889                 goto errout;
 890         }
 891
 892         err = fib_table_insert(net, tb, &cfg, extack);
 893         if (!err && cfg.fc_type == RTN_LOCAL)
 894                 net->ipv4.fib_has_custom_local_routes = true;
 895 errout:
 896         return err;
 897 }
 898
 899 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
 900                           struct fib_dump_filter *filter,
 901                           struct netlink_callback *cb)
 902 {
 903         struct netlink_ext_ack *extack = cb->extack;
 904         struct nlattr *tb[RTA_MAX + 1];
 905         struct rtmsg *rtm;
 906         int err, i;
 907
 908         ASSERT_RTNL();
 909
 910         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
 911                 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
 912                 return -EINVAL;
 913         }
 914
 915         rtm = nlmsg_data(nlh);
 916         if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
 917             rtm->rtm_scope) {
 918                 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
 919                 return -EINVAL;
 920         }
 921
 922         if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
 923                 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
 924                 return -EINVAL;
 925         }
 926         if (rtm->rtm_flags & RTM_F_CLONED)
 927                 filter->dump_routes = false;
 928         else
 929                 filter->dump_exceptions = false;
 930
 931         filter->dump_all_families = (rtm->rtm_family == AF_UNSPEC);
 932         filter->flags    = rtm->rtm_flags;
 933         filter->protocol = rtm->rtm_protocol;
 934         filter->rt_type  = rtm->rtm_type;
 935         filter->table_id = rtm->rtm_table;
 936
 937         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
 938                                             rtm_ipv4_policy, extack);
 939         if (err < 0)
 940                 return err;
 941
 942         for (i = 0; i <= RTA_MAX; ++i) {
 943                 int ifindex;
 944
 945                 if (!tb[i])
 946                         continue;
 947
 948                 switch (i) {
 949                 case RTA_TABLE:
 950                         filter->table_id = nla_get_u32(tb[i]);
 951                         break;
 952                 case RTA_OIF:
 953                         ifindex = nla_get_u32(tb[i]);
 954                         filter->dev = __dev_get_by_index(net, ifindex);
 955                         if (!filter->dev)
 956                                 return -ENODEV;
 957                         break;
 958                 default:
 959                         NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
 960                         return -EINVAL;
 961                 }
 962         }
 963
 964         if (filter->flags || filter->protocol || filter->rt_type ||
 965             filter->table_id || filter->dev) {
 966                 filter->filter_set = 1;
 967                 cb->answer_flags = NLM_F_DUMP_FILTERED;
 968         }
 969
 970         return 0;
 971 }
 972 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
 973
 974 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 975 {
 976         struct fib_dump_filter filter = { .dump_routes = true,
 977                                           .dump_exceptions = true };
 978         const struct nlmsghdr *nlh = cb->nlh;
 979         struct net *net = sock_net(skb->sk);
 980         unsigned int h, s_h;
 981         unsigned int e = 0, s_e;
 982         struct fib_table *tb;
 983         struct hlist_head *head;
 984         int dumped = 0, err;
 985
 986         if (cb->strict_check) {
 987                 err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
 988                 if (err < 0)
 989                         return err;
 990         } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
 991                 struct rtmsg *rtm = nlmsg_data(nlh);
 992
 993                 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
 994         }
 995
 996         /* ipv4 does not use prefix flag */
 997         if (filter.flags & RTM_F_PREFIX)
 998                 return skb->len;
 999
1000         if (filter.table_id) {
1001                 tb = fib_get_table(net, filter.table_id);
1002                 if (!tb) {
1003                         if (filter.dump_all_families)
1004                                 return skb->len;
1005
1006                         NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1007                         return -ENOENT;
1008                 }
1009
1010                 err = fib_table_dump(tb, skb, cb, &filter);
1011                 return skb->len ? : err;
1012         }
1013
1014         s_h = cb->args[0];
1015         s_e = cb->args[1];
1016
1017         rcu_read_lock();
1018
1019         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1020                 e = 0;
1021                 head = &net->ipv4.fib_table_hash[h];
1022                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1023                         if (e < s_e)
1024                                 goto next;
1025                         if (dumped)
1026                                 memset(&cb->args[2], 0, sizeof(cb->args) -
1027                                                  2 * sizeof(cb->args[0]));
1028                         err = fib_table_dump(tb, skb, cb, &filter);
1029                         if (err < 0) {
1030                                 if (likely(skb->len))
1031                                         goto out;
1032
1033                                 goto out_err;
1034                         }
1035                         dumped = 1;
1036 next:
1037                         e++;
1038                 }
1039         }
1040 out:
1041         err = skb->len;
1042 out_err:
1043         rcu_read_unlock();
1044
1045         cb->args[1] = e;
1046         cb->args[0] = h;
1047
1048         return err;
1049 }
1050
1051 /* Prepare and feed intra-kernel routing request.
1052  * Really, it should be netlink message, but :-( netlink
1053  * can be not configured, so that we feed it directly
1054  * to fib engine. It is legal, because all events occur
1055  * only when netlink is already locked.
1056  */
1057 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1058                       struct in_ifaddr *ifa, u32 rt_priority)
1059 {
1060         struct net *net = dev_net(ifa->ifa_dev->dev);
1061         u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1062         struct fib_table *tb;
1063         struct fib_config cfg = {
1064                 .fc_protocol = RTPROT_KERNEL,
1065                 .fc_type = type,
1066                 .fc_dst = dst,
1067                 .fc_dst_len = dst_len,
1068                 .fc_priority = rt_priority,
1069                 .fc_prefsrc = ifa->ifa_local,
1070                 .fc_oif = ifa->ifa_dev->dev->ifindex,
1071                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1072                 .fc_nlinfo = {
1073                         .nl_net = net,
1074                 },
1075         };
1076
1077         if (!tb_id)
1078                 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1079
1080         tb = fib_new_table(net, tb_id);
1081         if (!tb)
1082                 return;
1083
1084         cfg.fc_table = tb->tb_id;
1085
1086         if (type != RTN_LOCAL)
1087                 cfg.fc_scope = RT_SCOPE_LINK;
1088         else
1089                 cfg.fc_scope = RT_SCOPE_HOST;
1090
1091         if (cmd == RTM_NEWROUTE)
1092                 fib_table_insert(net, tb, &cfg, NULL);
1093         else
1094                 fib_table_delete(net, tb, &cfg, NULL);
1095 }
1096
1097 void fib_add_ifaddr(struct in_ifaddr *ifa)
1098 {
1099         struct in_device *in_dev = ifa->ifa_dev;
1100         struct net_device *dev = in_dev->dev;
1101         struct in_ifaddr *prim = ifa;
1102         __be32 mask = ifa->ifa_mask;
1103         __be32 addr = ifa->ifa_local;
1104         __be32 prefix = ifa->ifa_address & mask;
1105
1106         if (ifa->ifa_flags & IFA_F_SECONDARY) {
1107                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
1108                 if (!prim) {
1109                         pr_warn("%s: bug: prim == NULL\n", __func__);
1110                         return;
1111                 }
1112         }
1113
1114         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1115
1116         if (!(dev->flags & IFF_UP))
1117                 return;
1118
1119         /* Add broadcast address, if it is explicitly assigned. */
1120         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
1121                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1122                           prim, 0);
1123
1124         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1125             (prefix != addr || ifa->ifa_prefixlen < 32)) {
1126                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1127                         fib_magic(RTM_NEWROUTE,
1128                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1129                                   prefix, ifa->ifa_prefixlen, prim,
1130                                   ifa->ifa_rt_priority);
1131
1132                 /* Add network specific broadcasts, when it takes a sense */
1133                 if (ifa->ifa_prefixlen < 31) {
1134                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1135                                   prim, 0);
1136                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1137                                   32, prim, 0);
1138                 }
1139         }
1140 }
1141
1142 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1143 {
1144         __be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1145         struct in_device *in_dev = ifa->ifa_dev;
1146         struct net_device *dev = in_dev->dev;
1147
1148         if (!(dev->flags & IFF_UP) ||
1149             ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1150             ipv4_is_zeronet(prefix) ||
1151             prefix == ifa->ifa_local || ifa->ifa_prefixlen == 32)
1152                 return;
1153
1154         /* add the new */
1155         fib_magic(RTM_NEWROUTE,
1156                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1157                   prefix, ifa->ifa_prefixlen, ifa, new_metric);
1158
1159         /* delete the old */
1160         fib_magic(RTM_DELROUTE,
1161                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1162                   prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1163 }
1164
1165 /* Delete primary or secondary address.
1166  * Optionally, on secondary address promotion consider the addresses
1167  * from subnet iprim as deleted, even if they are in device list.
1168  * In this case the secondary ifa can be in device list.
1169  */
1170 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1171 {
1172         struct in_device *in_dev = ifa->ifa_dev;
1173         struct net_device *dev = in_dev->dev;
1174         struct in_ifaddr *ifa1;
1175         struct in_ifaddr *prim = ifa, *prim1 = NULL;
1176         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1177         __be32 any = ifa->ifa_address & ifa->ifa_mask;
1178 #define LOCAL_OK        1
1179 #define BRD_OK          2
1180 #define BRD0_OK         4
1181 #define BRD1_OK         8
1182         unsigned int ok = 0;
1183         int subnet = 0;         /* Primary network */
1184         int gone = 1;           /* Address is missing */
1185         int same_prefsrc = 0;   /* Another primary with same IP */
1186
1187         if (ifa->ifa_flags & IFA_F_SECONDARY) {
1188                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1189                 if (!prim) {
1190                         /* if the device has been deleted, we don't perform
1191                          * address promotion
1192                          */
1193                         if (!in_dev->dead)
1194                                 pr_warn("%s: bug: prim == NULL\n", __func__);
1195                         return;
1196                 }
1197                 if (iprim && iprim != prim) {
1198                         pr_warn("%s: bug: iprim != prim\n", __func__);
1199                         return;
1200                 }
1201         } else if (!ipv4_is_zeronet(any) &&
1202                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1203                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1204                         fib_magic(RTM_DELROUTE,
1205                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1206                                   any, ifa->ifa_prefixlen, prim, 0);
1207                 subnet = 1;
1208         }
1209
1210         if (in_dev->dead)
1211                 goto no_promotions;
1212
1213         /* Deletion is more complicated than add.
1214          * We should take care of not to delete too much :-)
1215          *
1216          * Scan address list to be sure that addresses are really gone.
1217          */
1218         rcu_read_lock();
1219         in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1220                 if (ifa1 == ifa) {
1221                         /* promotion, keep the IP */
1222                         gone = 0;
1223                         continue;
1224                 }
1225                 /* Ignore IFAs from our subnet */
1226                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1227                     inet_ifa_match(ifa1->ifa_address, iprim))
1228                         continue;
1229
1230                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
1231                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1232                         /* Another address from our subnet? */
1233                         if (ifa1->ifa_mask == prim->ifa_mask &&
1234                             inet_ifa_match(ifa1->ifa_address, prim))
1235                                 prim1 = prim;
1236                         else {
1237                                 /* We reached the secondaries, so
1238                                  * same_prefsrc should be determined.
1239                                  */
1240                                 if (!same_prefsrc)
1241                                         continue;
1242                                 /* Search new prim1 if ifa1 is not
1243                                  * using the current prim1
1244                                  */
1245                                 if (!prim1 ||
1246                                     ifa1->ifa_mask != prim1->ifa_mask ||
1247                                     !inet_ifa_match(ifa1->ifa_address, prim1))
1248                                         prim1 = inet_ifa_byprefix(in_dev,
1249                                                         ifa1->ifa_address,
1250                                                         ifa1->ifa_mask);
1251                                 if (!prim1)
1252                                         continue;
1253                                 if (prim1->ifa_local != prim->ifa_local)
1254                                         continue;
1255                         }
1256                 } else {
1257                         if (prim->ifa_local != ifa1->ifa_local)
1258                                 continue;
1259                         prim1 = ifa1;
1260                         if (prim != prim1)
1261                                 same_prefsrc = 1;
1262                 }
1263                 if (ifa->ifa_local == ifa1->ifa_local)
1264                         ok |= LOCAL_OK;
1265                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1266                         ok |= BRD_OK;
1267                 if (brd == ifa1->ifa_broadcast)
1268                         ok |= BRD1_OK;
1269                 if (any == ifa1->ifa_broadcast)
1270                         ok |= BRD0_OK;
1271                 /* primary has network specific broadcasts */
1272                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1273                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1274                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1275
1276                         if (!ipv4_is_zeronet(any1)) {
1277                                 if (ifa->ifa_broadcast == brd1 ||
1278                                     ifa->ifa_broadcast == any1)
1279                                         ok |= BRD_OK;
1280                                 if (brd == brd1 || brd == any1)
1281                                         ok |= BRD1_OK;
1282                                 if (any == brd1 || any == any1)
1283                                         ok |= BRD0_OK;
1284                         }
1285                 }
1286         }
1287         rcu_read_unlock();
1288
1289 no_promotions:
1290         if (!(ok & BRD_OK))
1291                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1292                           prim, 0);
1293         if (subnet && ifa->ifa_prefixlen < 31) {
1294                 if (!(ok & BRD1_OK))
1295                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1296                                   prim, 0);
1297                 if (!(ok & BRD0_OK))
1298                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1299                                   prim, 0);
1300         }
1301         if (!(ok & LOCAL_OK)) {
1302                 unsigned int addr_type;
1303
1304                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1305
1306                 /* Check, that this local address finally disappeared. */
1307                 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1308                                                      ifa->ifa_local);
1309                 if (gone && addr_type != RTN_LOCAL) {
1310                         /* And the last, but not the least thing.
1311                          * We must flush stray FIB entries.
1312                          *
1313                          * First of all, we scan fib_info list searching
1314                          * for stray nexthop entries, then ignite fib_flush.
1315                          */
1316                         if (fib_sync_down_addr(dev, ifa->ifa_local))
1317                                 fib_flush(dev_net(dev));
1318                 }
1319         }
1320 #undef LOCAL_OK
1321 #undef BRD_OK
1322 #undef BRD0_OK
1323 #undef BRD1_OK
1324 }
1325
1326 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1327 {
1328
1329         struct fib_result       res;
1330         struct flowi4           fl4 = {
1331                 .flowi4_mark = frn->fl_mark,
1332                 .daddr = frn->fl_addr,
1333                 .flowi4_tos = frn->fl_tos,
1334                 .flowi4_scope = frn->fl_scope,
1335         };
1336         struct fib_table *tb;
1337
1338         rcu_read_lock();
1339
1340         tb = fib_get_table(net, frn->tb_id_in);
1341
1342         frn->err = -ENOENT;
1343         if (tb) {
1344                 local_bh_disable();
1345
1346                 frn->tb_id = tb->tb_id;
1347                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1348
1349                 if (!frn->err) {
1350                         frn->prefixlen = res.prefixlen;
1351                         frn->nh_sel = res.nh_sel;
1352                         frn->type = res.type;
1353                         frn->scope = res.scope;
1354                 }
1355                 local_bh_enable();
1356         }
1357
1358         rcu_read_unlock();
1359 }
1360
1361 static void nl_fib_input(struct sk_buff *skb)
1362 {
1363         struct net *net;
1364         struct fib_result_nl *frn;
1365         struct nlmsghdr *nlh;
1366         u32 portid;
1367
1368         net = sock_net(skb->sk);
1369         nlh = nlmsg_hdr(skb);
1370         if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1371             skb->len < nlh->nlmsg_len ||
1372             nlmsg_len(nlh) < sizeof(*frn))
1373                 return;
1374
1375         skb = netlink_skb_clone(skb, GFP_KERNEL);
1376         if (!skb)
1377                 return;
1378         nlh = nlmsg_hdr(skb);
1379
1380         frn = (struct fib_result_nl *) nlmsg_data(nlh);
1381         nl_fib_lookup(net, frn);
1382
1383         portid = NETLINK_CB(skb).portid;      /* netlink portid */
1384         NETLINK_CB(skb).portid = 0;        /* from kernel */
1385         NETLINK_CB(skb).dst_group = 0;  /* unicast */
1386         netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1387 }
1388
1389 static int __net_init nl_fib_lookup_init(struct net *net)
1390 {
1391         struct sock *sk;
1392         struct netlink_kernel_cfg cfg = {
1393                 .input  = nl_fib_input,
1394         };
1395
1396         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1397         if (!sk)
1398                 return -EAFNOSUPPORT;
1399         net->ipv4.fibnl = sk;
1400         return 0;
1401 }
1402
1403 static void nl_fib_lookup_exit(struct net *net)
1404 {
1405         netlink_kernel_release(net->ipv4.fibnl);
1406         net->ipv4.fibnl = NULL;
1407 }
1408
1409 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1410                            bool force)
1411 {
1412         if (fib_sync_down_dev(dev, event, force))
1413                 fib_flush(dev_net(dev));
1414         else
1415                 rt_cache_flush(dev_net(dev));
1416         arp_ifdown(dev);
1417 }
1418
1419 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1420 {
1421         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1422         struct net_device *dev = ifa->ifa_dev->dev;
1423         struct net *net = dev_net(dev);
1424
1425         switch (event) {
1426         case NETDEV_UP:
1427                 fib_add_ifaddr(ifa);
1428 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1429                 fib_sync_up(dev, RTNH_F_DEAD);
1430 #endif
1431                 atomic_inc(&net->ipv4.dev_addr_genid);
1432                 rt_cache_flush(dev_net(dev));
1433                 break;
1434         case NETDEV_DOWN:
1435                 fib_del_ifaddr(ifa, NULL);
1436                 atomic_inc(&net->ipv4.dev_addr_genid);
1437                 if (!ifa->ifa_dev->ifa_list) {
1438                         /* Last address was deleted from this interface.
1439                          * Disable IP.
1440                          */
1441                         fib_disable_ip(dev, event, true);
1442                 } else {
1443                         rt_cache_flush(dev_net(dev));
1444                 }
1445                 break;
1446         }
1447         return NOTIFY_DONE;
1448 }
1449
1450 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1451 {
1452         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1453         struct netdev_notifier_changeupper_info *upper_info = ptr;
1454         struct netdev_notifier_info_ext *info_ext = ptr;
1455         struct in_device *in_dev;
1456         struct net *net = dev_net(dev);
1457         struct in_ifaddr *ifa;
1458         unsigned int flags;
1459
1460         if (event == NETDEV_UNREGISTER) {
1461                 fib_disable_ip(dev, event, true);
1462                 rt_flush_dev(dev);
1463                 return NOTIFY_DONE;
1464         }
1465
1466         in_dev = __in_dev_get_rtnl(dev);
1467         if (!in_dev)
1468                 return NOTIFY_DONE;
1469
1470         switch (event) {
1471         case NETDEV_UP:
1472                 in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1473                         fib_add_ifaddr(ifa);
1474                 }
1475 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1476                 fib_sync_up(dev, RTNH_F_DEAD);
1477 #endif
1478                 atomic_inc(&net->ipv4.dev_addr_genid);
1479                 rt_cache_flush(net);
1480                 break;
1481         case NETDEV_DOWN:
1482                 fib_disable_ip(dev, event, false);
1483                 break;
1484         case NETDEV_CHANGE:
1485                 flags = dev_get_flags(dev);
1486                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1487                         fib_sync_up(dev, RTNH_F_LINKDOWN);
1488                 else
1489                         fib_sync_down_dev(dev, event, false);
1490                 rt_cache_flush(net);
1491                 break;
1492         case NETDEV_CHANGEMTU:
1493                 fib_sync_mtu(dev, info_ext->ext.mtu);
1494                 rt_cache_flush(net);
1495                 break;
1496         case NETDEV_CHANGEUPPER:
1497                 upper_info = ptr;
1498                 /* flush all routes if dev is linked to or unlinked from
1499                  * an L3 master device (e.g., VRF)
1500                  */
1501                 if (upper_info->upper_dev &&
1502                     netif_is_l3_master(upper_info->upper_dev))
1503                         fib_disable_ip(dev, NETDEV_DOWN, true);
1504                 break;
1505         }
1506         return NOTIFY_DONE;
1507 }
1508
1509 static struct notifier_block fib_inetaddr_notifier = {
1510         .notifier_call = fib_inetaddr_event,
1511 };
1512
1513 static struct notifier_block fib_netdev_notifier = {
1514         .notifier_call = fib_netdev_event,
1515 };
1516
1517 static int __net_init ip_fib_net_init(struct net *net)
1518 {
1519         int err;
1520         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1521
1522         err = fib4_notifier_init(net);
1523         if (err)
1524                 return err;
1525
1526         /* Avoid false sharing : Use at least a full cache line */
1527         size = max_t(size_t, size, L1_CACHE_BYTES);
1528
1529         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1530         if (!net->ipv4.fib_table_hash) {
1531                 err = -ENOMEM;
1532                 goto err_table_hash_alloc;
1533         }
1534
1535         err = fib4_rules_init(net);
1536         if (err < 0)
1537                 goto err_rules_init;
1538         return 0;
1539
1540 err_rules_init:
1541         kfree(net->ipv4.fib_table_hash);
1542 err_table_hash_alloc:
1543         fib4_notifier_exit(net);
1544         return err;
1545 }
1546
1547 static void ip_fib_net_exit(struct net *net)
1548 {
1549         int i;
1550
1551         rtnl_lock();
1552 #ifdef CONFIG_IP_MULTIPLE_TABLES
1553         RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1554         RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1555 #endif
1556         /* Destroy the tables in reverse order to guarantee that the
1557          * local table, ID 255, is destroyed before the main table, ID
1558          * 254. This is necessary as the local table may contain
1559          * references to data contained in the main table.
1560          */
1561         for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1562                 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1563                 struct hlist_node *tmp;
1564                 struct fib_table *tb;
1565
1566                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1567                         hlist_del(&tb->tb_hlist);
1568                         fib_table_flush(net, tb, true);
1569                         fib_free_table(tb);
1570                 }
1571         }
1572
1573 #ifdef CONFIG_IP_MULTIPLE_TABLES
1574         fib4_rules_exit(net);
1575 #endif
1576         rtnl_unlock();
1577         kfree(net->ipv4.fib_table_hash);
1578         fib4_notifier_exit(net);
1579 }
1580
1581 static int __net_init fib_net_init(struct net *net)
1582 {
1583         int error;
1584
1585 #ifdef CONFIG_IP_ROUTE_CLASSID
1586         net->ipv4.fib_num_tclassid_users = 0;
1587 #endif
1588         error = ip_fib_net_init(net);
1589         if (error < 0)
1590                 goto out;
1591         error = nl_fib_lookup_init(net);
1592         if (error < 0)
1593                 goto out_nlfl;
1594         error = fib_proc_init(net);
1595         if (error < 0)
1596                 goto out_proc;
1597 out:
1598         return error;
1599
1600 out_proc:
1601         nl_fib_lookup_exit(net);
1602 out_nlfl:
1603         ip_fib_net_exit(net);
1604         goto out;
1605 }
1606
1607 static void __net_exit fib_net_exit(struct net *net)
1608 {
1609         fib_proc_exit(net);
1610         nl_fib_lookup_exit(net);
1611         ip_fib_net_exit(net);
1612 }
1613
1614 static struct pernet_operations fib_net_ops = {
1615         .init = fib_net_init,
1616         .exit = fib_net_exit,
1617 };
1618
1619 void __init ip_fib_init(void)
1620 {
1621         fib_trie_init();
1622
1623         register_pernet_subsys(&fib_net_ops);
1624
1625         register_netdevice_notifier(&fib_netdev_notifier);
1626         register_inetaddr_notifier(&fib_inetaddr_notifier);
1627
1628         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1629         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1630         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1631 }