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 #else
  74
  75 struct fib_table *fib_new_table(struct net *net, u32 id)
  76 {
  77         struct fib_table *tb, *alias = NULL;
  78         unsigned int h;
  79
  80         if (id == 0)
  81                 id = RT_TABLE_MAIN;
  82         tb = fib_get_table(net, id);
  83         if (tb)
  84                 return tb;
  85
  86         if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
  87                 alias = fib_new_table(net, RT_TABLE_MAIN);
  88
  89         tb = fib_trie_table(id, alias);
  90         if (!tb)
  91                 return NULL;
  92
  93         switch (id) {
  94         case RT_TABLE_MAIN:
  95                 rcu_assign_pointer(net->ipv4.fib_main, tb);
  96                 break;
  97         case RT_TABLE_DEFAULT:
  98                 rcu_assign_pointer(net->ipv4.fib_default, tb);
  99                 break;
 100         default:
 101                 break;
 102         }
 103
 104         h = id & (FIB_TABLE_HASHSZ - 1);
 105         hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
 106         return tb;
 107 }
 108 EXPORT_SYMBOL_GPL(fib_new_table);
 109
 110 /* caller must hold either rtnl or rcu read lock */
 111 struct fib_table *fib_get_table(struct net *net, u32 id)
 112 {
 113         struct fib_table *tb;
 114         struct hlist_head *head;
 115         unsigned int h;
 116
 117         if (id == 0)
 118                 id = RT_TABLE_MAIN;
 119         h = id & (FIB_TABLE_HASHSZ - 1);
 120
 121         head = &net->ipv4.fib_table_hash[h];
 122         hlist_for_each_entry_rcu(tb, head, tb_hlist,
 123                                  lockdep_rtnl_is_held()) {
 124                 if (tb->tb_id == id)
 125                         return tb;
 126         }
 127         return NULL;
 128 }
 129 #endif /* CONFIG_IP_MULTIPLE_TABLES */
 130
 131 static void fib_replace_table(struct net *net, struct fib_table *old,
 132                               struct fib_table *new)
 133 {
 134 #ifdef CONFIG_IP_MULTIPLE_TABLES
 135         switch (new->tb_id) {
 136         case RT_TABLE_MAIN:
 137                 rcu_assign_pointer(net->ipv4.fib_main, new);
 138                 break;
 139         case RT_TABLE_DEFAULT:
 140                 rcu_assign_pointer(net->ipv4.fib_default, new);
 141                 break;
 142         default:
 143                 break;
 144         }
 145
 146 #endif
 147         /* replace the old table in the hlist */
 148         hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
 149 }
 150
 151 int fib_unmerge(struct net *net)
 152 {
 153         struct fib_table *old, *new, *main_table;
 154
 155         /* attempt to fetch local table if it has been allocated */
 156         old = fib_get_table(net, RT_TABLE_LOCAL);
 157         if (!old)
 158                 return 0;
 159
 160         new = fib_trie_unmerge(old);
 161         if (!new)
 162                 return -ENOMEM;
 163
 164         /* table is already unmerged */
 165         if (new == old)
 166                 return 0;
 167
 168         /* replace merged table with clean table */
 169         fib_replace_table(net, old, new);
 170         fib_free_table(old);
 171
 172         /* attempt to fetch main table if it has been allocated */
 173         main_table = fib_get_table(net, RT_TABLE_MAIN);
 174         if (!main_table)
 175                 return 0;
 176
 177         /* flush local entries from main table */
 178         fib_table_flush_external(main_table);
 179
 180         return 0;
 181 }
 182
 183 void fib_flush(struct net *net)
 184 {
 185         int flushed = 0;
 186         unsigned int h;
 187
 188         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
 189                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
 190                 struct hlist_node *tmp;
 191                 struct fib_table *tb;
 192
 193                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
 194                         flushed += fib_table_flush(net, tb, false);
 195         }
 196
 197         if (flushed)
 198                 rt_cache_flush(net);
 199 }
 200
 201 /*
 202  * Find address type as if only "dev" was present in the system. If
 203  * on_dev is NULL then all interfaces are taken into consideration.
 204  */
 205 static inline unsigned int __inet_dev_addr_type(struct net *net,
 206                                                 const struct net_device *dev,
 207                                                 __be32 addr, u32 tb_id)
 208 {
 209         struct flowi4           fl4 = { .daddr = addr };
 210         struct fib_result       res;
 211         unsigned int ret = RTN_BROADCAST;
 212         struct fib_table *table;
 213
 214         if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
 215                 return RTN_BROADCAST;
 216         if (ipv4_is_multicast(addr))
 217                 return RTN_MULTICAST;
 218
 219         rcu_read_lock();
 220
 221         table = fib_get_table(net, tb_id);
 222         if (table) {
 223                 ret = RTN_UNICAST;
 224                 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
 225                         struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0);
 226
 227                         if (!dev || dev == nhc->nhc_dev)
 228                                 ret = res.type;
 229                 }
 230         }
 231
 232         rcu_read_unlock();
 233         return ret;
 234 }
 235
 236 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
 237 {
 238         return __inet_dev_addr_type(net, NULL, addr, tb_id);
 239 }
 240 EXPORT_SYMBOL(inet_addr_type_table);
 241
 242 unsigned int inet_addr_type(struct net *net, __be32 addr)
 243 {
 244         return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
 245 }
 246 EXPORT_SYMBOL(inet_addr_type);
 247
 248 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
 249                                 __be32 addr)
 250 {
 251         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
 252
 253         return __inet_dev_addr_type(net, dev, addr, rt_table);
 254 }
 255 EXPORT_SYMBOL(inet_dev_addr_type);
 256
 257 /* inet_addr_type with dev == NULL but using the table from a dev
 258  * if one is associated
 259  */
 260 unsigned int inet_addr_type_dev_table(struct net *net,
 261                                       const struct net_device *dev,
 262                                       __be32 addr)
 263 {
 264         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
 265
 266         return __inet_dev_addr_type(net, NULL, addr, rt_table);
 267 }
 268 EXPORT_SYMBOL(inet_addr_type_dev_table);
 269
 270 __be32 fib_compute_spec_dst(struct sk_buff *skb)
 271 {
 272         struct net_device *dev = skb->dev;
 273         struct in_device *in_dev;
 274         struct fib_result res;
 275         struct rtable *rt;
 276         struct net *net;
 277         int scope;
 278
 279         rt = skb_rtable(skb);
 280         if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
 281             RTCF_LOCAL)
 282                 return ip_hdr(skb)->daddr;
 283
 284         in_dev = __in_dev_get_rcu(dev);
 285
 286         net = dev_net(dev);
 287
 288         scope = RT_SCOPE_UNIVERSE;
 289         if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
 290                 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
 291                 struct flowi4 fl4 = {
 292                         .flowi4_iif = LOOPBACK_IFINDEX,
 293                         .flowi4_oif = l3mdev_master_ifindex_rcu(dev),
 294                         .daddr = ip_hdr(skb)->saddr,
 295                         .flowi4_tos = ip_hdr(skb)->tos & IPTOS_RT_MASK,
 296                         .flowi4_scope = scope,
 297                         .flowi4_mark = vmark ? skb->mark : 0,
 298                 };
 299                 if (!fib_lookup(net, &fl4, &res, 0))
 300                         return fib_result_prefsrc(net, &res);
 301         } else {
 302                 scope = RT_SCOPE_LINK;
 303         }
 304
 305         return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
 306 }
 307
 308 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
 309 {
 310         bool dev_match = false;
 311 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 312         if (unlikely(fi->nh)) {
 313                 dev_match = nexthop_uses_dev(fi->nh, dev);
 314         } else {
 315                 int ret;
 316
 317                 for (ret = 0; ret < fib_info_num_path(fi); ret++) {
 318                         const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
 319
 320                         if (nhc_l3mdev_matches_dev(nhc, dev)) {
 321                                 dev_match = true;
 322                                 break;
 323                         }
 324                 }
 325         }
 326 #else
 327         if (fib_info_nhc(fi, 0)->nhc_dev == dev)
 328                 dev_match = true;
 329 #endif
 330
 331         return dev_match;
 332 }
 333 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
 334
 335 /* Given (packet source, input interface) and optional (dst, oif, tos):
 336  * - (main) check, that source is valid i.e. not broadcast or our local
 337  *   address.
 338  * - figure out what "logical" interface this packet arrived
 339  *   and calculate "specific destination" address.
 340  * - check, that packet arrived from expected physical interface.
 341  * called with rcu_read_lock()
 342  */
 343 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 344                                  u8 tos, int oif, struct net_device *dev,
 345                                  int rpf, struct in_device *idev, u32 *itag)
 346 {
 347         struct net *net = dev_net(dev);
 348         struct flow_keys flkeys;
 349         int ret, no_addr;
 350         struct fib_result res;
 351         struct flowi4 fl4;
 352         bool dev_match;
 353
 354         fl4.flowi4_oif = 0;
 355         fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
 356         if (!fl4.flowi4_iif)
 357                 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
 358         fl4.daddr = src;
 359         fl4.saddr = dst;
 360         fl4.flowi4_tos = tos;
 361         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 362         fl4.flowi4_tun_key.tun_id = 0;
 363         fl4.flowi4_flags = 0;
 364         fl4.flowi4_uid = sock_net_uid(net, NULL);
 365         fl4.flowi4_multipath_hash = 0;
 366
 367         no_addr = idev->ifa_list == NULL;
 368
 369         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
 370         if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
 371                 fl4.flowi4_proto = 0;
 372                 fl4.fl4_sport = 0;
 373                 fl4.fl4_dport = 0;
 374         } else {
 375                 swap(fl4.fl4_sport, fl4.fl4_dport);
 376         }
 377
 378         if (fib_lookup(net, &fl4, &res, 0))
 379                 goto last_resort;
 380         if (res.type != RTN_UNICAST &&
 381             (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
 382                 goto e_inval;
 383         fib_combine_itag(itag, &res);
 384
 385         dev_match = fib_info_nh_uses_dev(res.fi, dev);
 386         /* This is not common, loopback packets retain skb_dst so normally they
 387          * would not even hit this slow path.
 388          */
 389         dev_match = dev_match || (res.type == RTN_LOCAL &&
 390                                   dev == net->loopback_dev);
 391         if (dev_match) {
 392                 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_LINK;
 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_LINK;
 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 (!cfg->fc_table)
 577                 cfg->fc_table = RT_TABLE_MAIN;
 578
 579         if (cmd == SIOCDELRT)
 580                 return 0;
 581
 582         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
 583                 return -EINVAL;
 584
 585         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 586                 cfg->fc_scope = RT_SCOPE_LINK;
 587
 588         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 589                 struct nlattr *mx;
 590                 int len = 0;
 591
 592                 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
 593                 if (!mx)
 594                         return -ENOMEM;
 595
 596                 if (rt->rt_flags & RTF_MTU)
 597                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 598
 599                 if (rt->rt_flags & RTF_WINDOW)
 600                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 601
 602                 if (rt->rt_flags & RTF_IRTT)
 603                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 604
 605                 cfg->fc_mx = mx;
 606                 cfg->fc_mx_len = len;
 607         }
 608
 609         return 0;
 610 }
 611
 612 /*
 613  * Handle IP routing ioctl calls.
 614  * These are used to manipulate the routing tables
 615  */
 616 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
 617 {
 618         struct fib_config cfg;
 619         int err;
 620
 621         switch (cmd) {
 622         case SIOCADDRT:         /* Add a route */
 623         case SIOCDELRT:         /* Delete a route */
 624                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 625                         return -EPERM;
 626
 627                 rtnl_lock();
 628                 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
 629                 if (err == 0) {
 630                         struct fib_table *tb;
 631
 632                         if (cmd == SIOCDELRT) {
 633                                 tb = fib_get_table(net, cfg.fc_table);
 634                                 if (tb)
 635                                         err = fib_table_delete(net, tb, &cfg,
 636                                                                NULL);
 637                                 else
 638                                         err = -ESRCH;
 639                         } else {
 640                                 tb = fib_new_table(net, cfg.fc_table);
 641                                 if (tb)
 642                                         err = fib_table_insert(net, tb,
 643                                                                &cfg, NULL);
 644                                 else
 645                                         err = -ENOBUFS;
 646                         }
 647
 648                         /* allocated by rtentry_to_fib_config() */
 649                         kfree(cfg.fc_mx);
 650                 }
 651                 rtnl_unlock();
 652                 return err;
 653         }
 654         return -EINVAL;
 655 }
 656
 657 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 658         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
 659         [RTA_DST]               = { .type = NLA_U32 },
 660         [RTA_SRC]               = { .type = NLA_U32 },
 661         [RTA_IIF]               = { .type = NLA_U32 },
 662         [RTA_OIF]               = { .type = NLA_U32 },
 663         [RTA_GATEWAY]           = { .type = NLA_U32 },
 664         [RTA_PRIORITY]          = { .type = NLA_U32 },
 665         [RTA_PREFSRC]           = { .type = NLA_U32 },
 666         [RTA_METRICS]           = { .type = NLA_NESTED },
 667         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 668         [RTA_FLOW]              = { .type = NLA_U32 },
 669         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
 670         [RTA_ENCAP]             = { .type = NLA_NESTED },
 671         [RTA_UID]               = { .type = NLA_U32 },
 672         [RTA_MARK]              = { .type = NLA_U32 },
 673         [RTA_TABLE]             = { .type = NLA_U32 },
 674         [RTA_IP_PROTO]          = { .type = NLA_U8 },
 675         [RTA_SPORT]             = { .type = NLA_U16 },
 676         [RTA_DPORT]             = { .type = NLA_U16 },
 677         [RTA_NH_ID]             = { .type = NLA_U32 },
 678 };
 679
 680 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
 681                     struct netlink_ext_ack *extack)
 682 {
 683         struct rtvia *via;
 684         int alen;
 685
 686         if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
 687                 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
 688                 return -EINVAL;
 689         }
 690
 691         via = nla_data(nla);
 692         alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
 693
 694         switch (via->rtvia_family) {
 695         case AF_INET:
 696                 if (alen != sizeof(__be32)) {
 697                         NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
 698                         return -EINVAL;
 699                 }
 700                 cfg->fc_gw_family = AF_INET;
 701                 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
 702                 break;
 703         case AF_INET6:
 704 #if IS_ENABLED(CONFIG_IPV6)
 705                 if (alen != sizeof(struct in6_addr)) {
 706                         NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
 707                         return -EINVAL;
 708                 }
 709                 cfg->fc_gw_family = AF_INET6;
 710                 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
 711 #else
 712                 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
 713                 return -EINVAL;
 714 #endif
 715                 break;
 716         default:
 717                 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
 718                 return -EINVAL;
 719         }
 720
 721         return 0;
 722 }
 723
 724 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 725                              struct nlmsghdr *nlh, struct fib_config *cfg,
 726                              struct netlink_ext_ack *extack)
 727 {
 728         bool has_gw = false, has_via = false;
 729         struct nlattr *attr;
 730         int err, remaining;
 731         struct rtmsg *rtm;
 732
 733         err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
 734                                         rtm_ipv4_policy, extack);
 735         if (err < 0)
 736                 goto errout;
 737
 738         memset(cfg, 0, sizeof(*cfg));
 739
 740         rtm = nlmsg_data(nlh);
 741         cfg->fc_dst_len = rtm->rtm_dst_len;
 742         cfg->fc_tos = rtm->rtm_tos;
 743         cfg->fc_table = rtm->rtm_table;
 744         cfg->fc_protocol = rtm->rtm_protocol;
 745         cfg->fc_scope = rtm->rtm_scope;
 746         cfg->fc_type = rtm->rtm_type;
 747         cfg->fc_flags = rtm->rtm_flags;
 748         cfg->fc_nlflags = nlh->nlmsg_flags;
 749
 750         cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
 751         cfg->fc_nlinfo.nlh = nlh;
 752         cfg->fc_nlinfo.nl_net = net;
 753
 754         if (cfg->fc_type > RTN_MAX) {
 755                 NL_SET_ERR_MSG(extack, "Invalid route type");
 756                 err = -EINVAL;
 757                 goto errout;
 758         }
 759
 760         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 761                 switch (nla_type(attr)) {
 762                 case RTA_DST:
 763                         cfg->fc_dst = nla_get_be32(attr);
 764                         break;
 765                 case RTA_OIF:
 766                         cfg->fc_oif = nla_get_u32(attr);
 767                         break;
 768                 case RTA_GATEWAY:
 769                         has_gw = true;
 770                         cfg->fc_gw4 = nla_get_be32(attr);
 771                         if (cfg->fc_gw4)
 772                                 cfg->fc_gw_family = AF_INET;
 773                         break;
 774                 case RTA_VIA:
 775                         has_via = true;
 776                         err = fib_gw_from_via(cfg, attr, extack);
 777                         if (err)
 778                                 goto errout;
 779                         break;
 780                 case RTA_PRIORITY:
 781                         cfg->fc_priority = nla_get_u32(attr);
 782                         break;
 783                 case RTA_PREFSRC:
 784                         cfg->fc_prefsrc = nla_get_be32(attr);
 785                         break;
 786                 case RTA_METRICS:
 787                         cfg->fc_mx = nla_data(attr);
 788                         cfg->fc_mx_len = nla_len(attr);
 789                         break;
 790                 case RTA_MULTIPATH:
 791                         err = lwtunnel_valid_encap_type_attr(nla_data(attr),
 792                                                              nla_len(attr),
 793                                                              extack);
 794                         if (err < 0)
 795                                 goto errout;
 796                         cfg->fc_mp = nla_data(attr);
 797                         cfg->fc_mp_len = nla_len(attr);
 798                         break;
 799                 case RTA_FLOW:
 800                         cfg->fc_flow = nla_get_u32(attr);
 801                         break;
 802                 case RTA_TABLE:
 803                         cfg->fc_table = nla_get_u32(attr);
 804                         break;
 805                 case RTA_ENCAP:
 806                         cfg->fc_encap = attr;
 807                         break;
 808                 case RTA_ENCAP_TYPE:
 809                         cfg->fc_encap_type = nla_get_u16(attr);
 810                         err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
 811                                                         extack);
 812                         if (err < 0)
 813                                 goto errout;
 814                         break;
 815                 case RTA_NH_ID:
 816                         cfg->fc_nh_id = nla_get_u32(attr);
 817                         break;
 818                 }
 819         }
 820
 821         if (cfg->fc_nh_id) {
 822                 if (cfg->fc_oif || cfg->fc_gw_family ||
 823                     cfg->fc_encap || cfg->fc_mp) {
 824                         NL_SET_ERR_MSG(extack,
 825                                        "Nexthop specification and nexthop id are mutually exclusive");
 826                         return -EINVAL;
 827                 }
 828         }
 829
 830         if (has_gw && has_via) {
 831                 NL_SET_ERR_MSG(extack,
 832                                "Nexthop configuration can not contain both GATEWAY and VIA");
 833                 return -EINVAL;
 834         }
 835
 836         if (!cfg->fc_table)
 837                 cfg->fc_table = RT_TABLE_MAIN;
 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->flags    = rtm->rtm_flags;
 932         filter->protocol = rtm->rtm_protocol;
 933         filter->rt_type  = rtm->rtm_type;
 934         filter->table_id = rtm->rtm_table;
 935
 936         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
 937                                             rtm_ipv4_policy, extack);
 938         if (err < 0)
 939                 return err;
 940
 941         for (i = 0; i <= RTA_MAX; ++i) {
 942                 int ifindex;
 943
 944                 if (!tb[i])
 945                         continue;
 946
 947                 switch (i) {
 948                 case RTA_TABLE:
 949                         filter->table_id = nla_get_u32(tb[i]);
 950                         break;
 951                 case RTA_OIF:
 952                         ifindex = nla_get_u32(tb[i]);
 953                         filter->dev = __dev_get_by_index(net, ifindex);
 954                         if (!filter->dev)
 955                                 return -ENODEV;
 956                         break;
 957                 default:
 958                         NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
 959                         return -EINVAL;
 960                 }
 961         }
 962
 963         if (filter->flags || filter->protocol || filter->rt_type ||
 964             filter->table_id || filter->dev) {
 965                 filter->filter_set = 1;
 966                 cb->answer_flags = NLM_F_DUMP_FILTERED;
 967         }
 968
 969         return 0;
 970 }
 971 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
 972
 973 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 974 {
 975         struct fib_dump_filter filter = { .dump_routes = true,
 976                                           .dump_exceptions = true };
 977         const struct nlmsghdr *nlh = cb->nlh;
 978         struct net *net = sock_net(skb->sk);
 979         unsigned int h, s_h;
 980         unsigned int e = 0, s_e;
 981         struct fib_table *tb;
 982         struct hlist_head *head;
 983         int dumped = 0, err;
 984
 985         if (cb->strict_check) {
 986                 err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
 987                 if (err < 0)
 988                         return err;
 989         } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
 990                 struct rtmsg *rtm = nlmsg_data(nlh);
 991
 992                 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
 993         }
 994
 995         /* ipv4 does not use prefix flag */
 996         if (filter.flags & RTM_F_PREFIX)
 997                 return skb->len;
 998
 999         if (filter.table_id) {
1000                 tb = fib_get_table(net, filter.table_id);
1001                 if (!tb) {
1002                         if (rtnl_msg_family(cb->nlh) != PF_INET)
1003                                 return skb->len;
1004
1005                         NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1006                         return -ENOENT;
1007                 }
1008
1009                 rcu_read_lock();
1010                 err = fib_table_dump(tb, skb, cb, &filter);
1011                 rcu_read_unlock();
1012                 return skb->len ? : err;
1013         }
1014
1015         s_h = cb->args[0];
1016         s_e = cb->args[1];
1017
1018         rcu_read_lock();
1019
1020         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1021                 e = 0;
1022                 head = &net->ipv4.fib_table_hash[h];
1023                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1024                         if (e < s_e)
1025                                 goto next;
1026                         if (dumped)
1027                                 memset(&cb->args[2], 0, sizeof(cb->args) -
1028                                                  2 * sizeof(cb->args[0]));
1029                         err = fib_table_dump(tb, skb, cb, &filter);
1030                         if (err < 0) {
1031                                 if (likely(skb->len))
1032                                         goto out;
1033
1034                                 goto out_err;
1035                         }
1036                         dumped = 1;
1037 next:
1038                         e++;
1039                 }
1040         }
1041 out:
1042         err = skb->len;
1043 out_err:
1044         rcu_read_unlock();
1045
1046         cb->args[1] = e;
1047         cb->args[0] = h;
1048
1049         return err;
1050 }
1051
1052 /* Prepare and feed intra-kernel routing request.
1053  * Really, it should be netlink message, but :-( netlink
1054  * can be not configured, so that we feed it directly
1055  * to fib engine. It is legal, because all events occur
1056  * only when netlink is already locked.
1057  */
1058 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1059                       struct in_ifaddr *ifa, u32 rt_priority)
1060 {
1061         struct net *net = dev_net(ifa->ifa_dev->dev);
1062         u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1063         struct fib_table *tb;
1064         struct fib_config cfg = {
1065                 .fc_protocol = RTPROT_KERNEL,
1066                 .fc_type = type,
1067                 .fc_dst = dst,
1068                 .fc_dst_len = dst_len,
1069                 .fc_priority = rt_priority,
1070                 .fc_prefsrc = ifa->ifa_local,
1071                 .fc_oif = ifa->ifa_dev->dev->ifindex,
1072                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1073                 .fc_nlinfo = {
1074                         .nl_net = net,
1075                 },
1076         };
1077
1078         if (!tb_id)
1079                 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1080
1081         tb = fib_new_table(net, tb_id);
1082         if (!tb)
1083                 return;
1084
1085         cfg.fc_table = tb->tb_id;
1086
1087         if (type != RTN_LOCAL)
1088                 cfg.fc_scope = RT_SCOPE_LINK;
1089         else
1090                 cfg.fc_scope = RT_SCOPE_HOST;
1091
1092         if (cmd == RTM_NEWROUTE)
1093                 fib_table_insert(net, tb, &cfg, NULL);
1094         else
1095                 fib_table_delete(net, tb, &cfg, NULL);
1096 }
1097
1098 void fib_add_ifaddr(struct in_ifaddr *ifa)
1099 {
1100         struct in_device *in_dev = ifa->ifa_dev;
1101         struct net_device *dev = in_dev->dev;
1102         struct in_ifaddr *prim = ifa;
1103         __be32 mask = ifa->ifa_mask;
1104         __be32 addr = ifa->ifa_local;
1105         __be32 prefix = ifa->ifa_address & mask;
1106
1107         if (ifa->ifa_flags & IFA_F_SECONDARY) {
1108                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
1109                 if (!prim) {
1110                         pr_warn("%s: bug: prim == NULL\n", __func__);
1111                         return;
1112                 }
1113         }
1114
1115         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1116
1117         if (!(dev->flags & IFF_UP))
1118                 return;
1119
1120         /* Add broadcast address, if it is explicitly assigned. */
1121         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) {
1122                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1123                           prim, 0);
1124                 arp_invalidate(dev, ifa->ifa_broadcast, false);
1125         }
1126
1127         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1128             (prefix != addr || ifa->ifa_prefixlen < 32)) {
1129                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1130                         fib_magic(RTM_NEWROUTE,
1131                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1132                                   prefix, ifa->ifa_prefixlen, prim,
1133                                   ifa->ifa_rt_priority);
1134
1135                 /* Add the network broadcast address, when it makes sense */
1136                 if (ifa->ifa_prefixlen < 31) {
1137                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1138                                   32, prim, 0);
1139                         arp_invalidate(dev, prefix | ~mask, false);
1140                 }
1141         }
1142 }
1143
1144 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1145 {
1146         __be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1147         struct in_device *in_dev = ifa->ifa_dev;
1148         struct net_device *dev = in_dev->dev;
1149
1150         if (!(dev->flags & IFF_UP) ||
1151             ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1152             ipv4_is_zeronet(prefix) ||
1153             (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32))
1154                 return;
1155
1156         /* add the new */
1157         fib_magic(RTM_NEWROUTE,
1158                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1159                   prefix, ifa->ifa_prefixlen, ifa, new_metric);
1160
1161         /* delete the old */
1162         fib_magic(RTM_DELROUTE,
1163                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1164                   prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1165 }
1166
1167 /* Delete primary or secondary address.
1168  * Optionally, on secondary address promotion consider the addresses
1169  * from subnet iprim as deleted, even if they are in device list.
1170  * In this case the secondary ifa can be in device list.
1171  */
1172 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1173 {
1174         struct in_device *in_dev = ifa->ifa_dev;
1175         struct net_device *dev = in_dev->dev;
1176         struct in_ifaddr *ifa1;
1177         struct in_ifaddr *prim = ifa, *prim1 = NULL;
1178         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1179         __be32 any = ifa->ifa_address & ifa->ifa_mask;
1180 #define LOCAL_OK        1
1181 #define BRD_OK          2
1182 #define BRD0_OK         4
1183 #define BRD1_OK         8
1184         unsigned int ok = 0;
1185         int subnet = 0;         /* Primary network */
1186         int gone = 1;           /* Address is missing */
1187         int same_prefsrc = 0;   /* Another primary with same IP */
1188
1189         if (ifa->ifa_flags & IFA_F_SECONDARY) {
1190                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1191                 if (!prim) {
1192                         /* if the device has been deleted, we don't perform
1193                          * address promotion
1194                          */
1195                         if (!in_dev->dead)
1196                                 pr_warn("%s: bug: prim == NULL\n", __func__);
1197                         return;
1198                 }
1199                 if (iprim && iprim != prim) {
1200                         pr_warn("%s: bug: iprim != prim\n", __func__);
1201                         return;
1202                 }
1203         } else if (!ipv4_is_zeronet(any) &&
1204                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1205                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1206                         fib_magic(RTM_DELROUTE,
1207                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1208                                   any, ifa->ifa_prefixlen, prim, 0);
1209                 subnet = 1;
1210         }
1211
1212         if (in_dev->dead)
1213                 goto no_promotions;
1214
1215         /* Deletion is more complicated than add.
1216          * We should take care of not to delete too much :-)
1217          *
1218          * Scan address list to be sure that addresses are really gone.
1219          */
1220         rcu_read_lock();
1221         in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1222                 if (ifa1 == ifa) {
1223                         /* promotion, keep the IP */
1224                         gone = 0;
1225                         continue;
1226                 }
1227                 /* Ignore IFAs from our subnet */
1228                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1229                     inet_ifa_match(ifa1->ifa_address, iprim))
1230                         continue;
1231
1232                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
1233                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1234                         /* Another address from our subnet? */
1235                         if (ifa1->ifa_mask == prim->ifa_mask &&
1236                             inet_ifa_match(ifa1->ifa_address, prim))
1237                                 prim1 = prim;
1238                         else {
1239                                 /* We reached the secondaries, so
1240                                  * same_prefsrc should be determined.
1241                                  */
1242                                 if (!same_prefsrc)
1243                                         continue;
1244                                 /* Search new prim1 if ifa1 is not
1245                                  * using the current prim1
1246                                  */
1247                                 if (!prim1 ||
1248                                     ifa1->ifa_mask != prim1->ifa_mask ||
1249                                     !inet_ifa_match(ifa1->ifa_address, prim1))
1250                                         prim1 = inet_ifa_byprefix(in_dev,
1251                                                         ifa1->ifa_address,
1252                                                         ifa1->ifa_mask);
1253                                 if (!prim1)
1254                                         continue;
1255                                 if (prim1->ifa_local != prim->ifa_local)
1256                                         continue;
1257                         }
1258                 } else {
1259                         if (prim->ifa_local != ifa1->ifa_local)
1260                                 continue;
1261                         prim1 = ifa1;
1262                         if (prim != prim1)
1263                                 same_prefsrc = 1;
1264                 }
1265                 if (ifa->ifa_local == ifa1->ifa_local)
1266                         ok |= LOCAL_OK;
1267                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1268                         ok |= BRD_OK;
1269                 if (brd == ifa1->ifa_broadcast)
1270                         ok |= BRD1_OK;
1271                 if (any == ifa1->ifa_broadcast)
1272                         ok |= BRD0_OK;
1273                 /* primary has network specific broadcasts */
1274                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1275                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1276                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1277
1278                         if (!ipv4_is_zeronet(any1)) {
1279                                 if (ifa->ifa_broadcast == brd1 ||
1280                                     ifa->ifa_broadcast == any1)
1281                                         ok |= BRD_OK;
1282                                 if (brd == brd1 || brd == any1)
1283                                         ok |= BRD1_OK;
1284                                 if (any == brd1 || any == any1)
1285                                         ok |= BRD0_OK;
1286                         }
1287                 }
1288         }
1289         rcu_read_unlock();
1290
1291 no_promotions:
1292         if (!(ok & BRD_OK))
1293                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1294                           prim, 0);
1295         if (subnet && ifa->ifa_prefixlen < 31) {
1296                 if (!(ok & BRD1_OK))
1297                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1298                                   prim, 0);
1299                 if (!(ok & BRD0_OK))
1300                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1301                                   prim, 0);
1302         }
1303         if (!(ok & LOCAL_OK)) {
1304                 unsigned int addr_type;
1305
1306                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1307
1308                 /* Check, that this local address finally disappeared. */
1309                 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1310                                                      ifa->ifa_local);
1311                 if (gone && addr_type != RTN_LOCAL) {
1312                         /* And the last, but not the least thing.
1313                          * We must flush stray FIB entries.
1314                          *
1315                          * First of all, we scan fib_info list searching
1316                          * for stray nexthop entries, then ignite fib_flush.
1317                          */
1318                         if (fib_sync_down_addr(dev, ifa->ifa_local))
1319                                 fib_flush(dev_net(dev));
1320                 }
1321         }
1322 #undef LOCAL_OK
1323 #undef BRD_OK
1324 #undef BRD0_OK
1325 #undef BRD1_OK
1326 }
1327
1328 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1329 {
1330
1331         struct fib_result       res;
1332         struct flowi4           fl4 = {
1333                 .flowi4_mark = frn->fl_mark,
1334                 .daddr = frn->fl_addr,
1335                 .flowi4_tos = frn->fl_tos,
1336                 .flowi4_scope = frn->fl_scope,
1337         };
1338         struct fib_table *tb;
1339
1340         rcu_read_lock();
1341
1342         tb = fib_get_table(net, frn->tb_id_in);
1343
1344         frn->err = -ENOENT;
1345         if (tb) {
1346                 local_bh_disable();
1347
1348                 frn->tb_id = tb->tb_id;
1349                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1350
1351                 if (!frn->err) {
1352                         frn->prefixlen = res.prefixlen;
1353                         frn->nh_sel = res.nh_sel;
1354                         frn->type = res.type;
1355                         frn->scope = res.scope;
1356                 }
1357                 local_bh_enable();
1358         }
1359
1360         rcu_read_unlock();
1361 }
1362
1363 static void nl_fib_input(struct sk_buff *skb)
1364 {
1365         struct net *net;
1366         struct fib_result_nl *frn;
1367         struct nlmsghdr *nlh;
1368         u32 portid;
1369
1370         net = sock_net(skb->sk);
1371         nlh = nlmsg_hdr(skb);
1372         if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1373             skb->len < nlh->nlmsg_len ||
1374             nlmsg_len(nlh) < sizeof(*frn))
1375                 return;
1376
1377         skb = netlink_skb_clone(skb, GFP_KERNEL);
1378         if (!skb)
1379                 return;
1380         nlh = nlmsg_hdr(skb);
1381
1382         frn = (struct fib_result_nl *) nlmsg_data(nlh);
1383         nl_fib_lookup(net, frn);
1384
1385         portid = NETLINK_CB(skb).portid;      /* netlink portid */
1386         NETLINK_CB(skb).portid = 0;        /* from kernel */
1387         NETLINK_CB(skb).dst_group = 0;  /* unicast */
1388         nlmsg_unicast(net->ipv4.fibnl, skb, portid);
1389 }
1390
1391 static int __net_init nl_fib_lookup_init(struct net *net)
1392 {
1393         struct sock *sk;
1394         struct netlink_kernel_cfg cfg = {
1395                 .input  = nl_fib_input,
1396         };
1397
1398         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1399         if (!sk)
1400                 return -EAFNOSUPPORT;
1401         net->ipv4.fibnl = sk;
1402         return 0;
1403 }
1404
1405 static void nl_fib_lookup_exit(struct net *net)
1406 {
1407         netlink_kernel_release(net->ipv4.fibnl);
1408         net->ipv4.fibnl = NULL;
1409 }
1410
1411 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1412                            bool force)
1413 {
1414         if (fib_sync_down_dev(dev, event, force))
1415                 fib_flush(dev_net(dev));
1416         else
1417                 rt_cache_flush(dev_net(dev));
1418         arp_ifdown(dev);
1419 }
1420
1421 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1422 {
1423         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1424         struct net_device *dev = ifa->ifa_dev->dev;
1425         struct net *net = dev_net(dev);
1426
1427         switch (event) {
1428         case NETDEV_UP:
1429                 fib_add_ifaddr(ifa);
1430 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1431                 fib_sync_up(dev, RTNH_F_DEAD);
1432 #endif
1433                 atomic_inc(&net->ipv4.dev_addr_genid);
1434                 rt_cache_flush(dev_net(dev));
1435                 break;
1436         case NETDEV_DOWN:
1437                 fib_del_ifaddr(ifa, NULL);
1438                 atomic_inc(&net->ipv4.dev_addr_genid);
1439                 if (!ifa->ifa_dev->ifa_list) {
1440                         /* Last address was deleted from this interface.
1441                          * Disable IP.
1442                          */
1443                         fib_disable_ip(dev, event, true);
1444                 } else {
1445                         rt_cache_flush(dev_net(dev));
1446                 }
1447                 break;
1448         }
1449         return NOTIFY_DONE;
1450 }
1451
1452 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1453 {
1454         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1455         struct netdev_notifier_changeupper_info *upper_info = ptr;
1456         struct netdev_notifier_info_ext *info_ext = ptr;
1457         struct in_device *in_dev;
1458         struct net *net = dev_net(dev);
1459         struct in_ifaddr *ifa;
1460         unsigned int flags;
1461
1462         if (event == NETDEV_UNREGISTER) {
1463                 fib_disable_ip(dev, event, true);
1464                 rt_flush_dev(dev);
1465                 return NOTIFY_DONE;
1466         }
1467
1468         in_dev = __in_dev_get_rtnl(dev);
1469         if (!in_dev)
1470                 return NOTIFY_DONE;
1471
1472         switch (event) {
1473         case NETDEV_UP:
1474                 in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1475                         fib_add_ifaddr(ifa);
1476                 }
1477 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1478                 fib_sync_up(dev, RTNH_F_DEAD);
1479 #endif
1480                 atomic_inc(&net->ipv4.dev_addr_genid);
1481                 rt_cache_flush(net);
1482                 break;
1483         case NETDEV_DOWN:
1484                 fib_disable_ip(dev, event, false);
1485                 break;
1486         case NETDEV_CHANGE:
1487                 flags = dev_get_flags(dev);
1488                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1489                         fib_sync_up(dev, RTNH_F_LINKDOWN);
1490                 else
1491                         fib_sync_down_dev(dev, event, false);
1492                 rt_cache_flush(net);
1493                 break;
1494         case NETDEV_CHANGEMTU:
1495                 fib_sync_mtu(dev, info_ext->ext.mtu);
1496                 rt_cache_flush(net);
1497                 break;
1498         case NETDEV_CHANGEUPPER:
1499                 upper_info = ptr;
1500                 /* flush all routes if dev is linked to or unlinked from
1501                  * an L3 master device (e.g., VRF)
1502                  */
1503                 if (upper_info->upper_dev &&
1504                     netif_is_l3_master(upper_info->upper_dev))
1505                         fib_disable_ip(dev, NETDEV_DOWN, true);
1506                 break;
1507         }
1508         return NOTIFY_DONE;
1509 }
1510
1511 static struct notifier_block fib_inetaddr_notifier = {
1512         .notifier_call = fib_inetaddr_event,
1513 };
1514
1515 static struct notifier_block fib_netdev_notifier = {
1516         .notifier_call = fib_netdev_event,
1517 };
1518
1519 static int __net_init ip_fib_net_init(struct net *net)
1520 {
1521         int err;
1522         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1523
1524         err = fib4_notifier_init(net);
1525         if (err)
1526                 return err;
1527
1528 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1529         /* Default to 3-tuple */
1530         net->ipv4.sysctl_fib_multipath_hash_fields =
1531                 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK;
1532 #endif
1533
1534         /* Avoid false sharing : Use at least a full cache line */
1535         size = max_t(size_t, size, L1_CACHE_BYTES);
1536
1537         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1538         if (!net->ipv4.fib_table_hash) {
1539                 err = -ENOMEM;
1540                 goto err_table_hash_alloc;
1541         }
1542
1543         err = fib4_rules_init(net);
1544         if (err < 0)
1545                 goto err_rules_init;
1546         return 0;
1547
1548 err_rules_init:
1549         kfree(net->ipv4.fib_table_hash);
1550 err_table_hash_alloc:
1551         fib4_notifier_exit(net);
1552         return err;
1553 }
1554
1555 static void ip_fib_net_exit(struct net *net)
1556 {
1557         int i;
1558
1559         rtnl_lock();
1560 #ifdef CONFIG_IP_MULTIPLE_TABLES
1561         RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1562         RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1563 #endif
1564         /* Destroy the tables in reverse order to guarantee that the
1565          * local table, ID 255, is destroyed before the main table, ID
1566          * 254. This is necessary as the local table may contain
1567          * references to data contained in the main table.
1568          */
1569         for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1570                 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1571                 struct hlist_node *tmp;
1572                 struct fib_table *tb;
1573
1574                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1575                         hlist_del(&tb->tb_hlist);
1576                         fib_table_flush(net, tb, true);
1577                         fib_free_table(tb);
1578                 }
1579         }
1580
1581 #ifdef CONFIG_IP_MULTIPLE_TABLES
1582         fib4_rules_exit(net);
1583 #endif
1584         rtnl_unlock();
1585         kfree(net->ipv4.fib_table_hash);
1586         fib4_notifier_exit(net);
1587 }
1588
1589 static int __net_init fib_net_init(struct net *net)
1590 {
1591         int error;
1592
1593 #ifdef CONFIG_IP_ROUTE_CLASSID
1594         atomic_set(&net->ipv4.fib_num_tclassid_users, 0);
1595 #endif
1596         error = ip_fib_net_init(net);
1597         if (error < 0)
1598                 goto out;
1599         error = nl_fib_lookup_init(net);
1600         if (error < 0)
1601                 goto out_nlfl;
1602         error = fib_proc_init(net);
1603         if (error < 0)
1604                 goto out_proc;
1605 out:
1606         return error;
1607
1608 out_proc:
1609         nl_fib_lookup_exit(net);
1610 out_nlfl:
1611         ip_fib_net_exit(net);
1612         goto out;
1613 }
1614
1615 static void __net_exit fib_net_exit(struct net *net)
1616 {
1617         fib_proc_exit(net);
1618         nl_fib_lookup_exit(net);
1619         ip_fib_net_exit(net);
1620 }
1621
1622 static struct pernet_operations fib_net_ops = {
1623         .init = fib_net_init,
1624         .exit = fib_net_exit,
1625 };
1626
1627 void __init ip_fib_init(void)
1628 {
1629         fib_trie_init();
1630
1631         register_pernet_subsys(&fib_net_ops);
1632
1633         register_netdevice_notifier(&fib_netdev_notifier);
1634         register_inetaddr_notifier(&fib_inetaddr_notifier);
1635
1636         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1637         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1638         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1639 }