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