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