net/ipv4/fib_frontend.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              IPv4 Forwarding Information Base: FIB frontend.
   8  *
   9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
  10  */
  11
  12 #include <linux/module.h>
  13 #include <linux/uaccess.h>
  14 #include <linux/bitops.h>
  15 #include <linux/capability.h>
  16 #include <linux/types.h>
  17 #include <linux/kernel.h>
  18 #include <linux/mm.h>
  19 #include <linux/string.h>
  20 #include <linux/socket.h>
  21 #include <linux/sockios.h>
  22 #include <linux/errno.h>
  23 #include <linux/in.h>
  24 #include <linux/inet.h>
  25 #include <linux/inetdevice.h>
  26 #include <linux/netdevice.h>
  27 #include <linux/if_addr.h>
  28 #include <linux/if_arp.h>
  29 #include <linux/skbuff.h>
  30 #include <linux/cache.h>
  31 #include <linux/init.h>
  32 #include <linux/list.h>
  33 #include <linux/slab.h>
  34
  35 #include <net/ip.h>
  36 #include <net/protocol.h>
  37 #include <net/route.h>
  38 #include <net/tcp.h>
  39 #include <net/sock.h>
  40 #include <net/arp.h>
  41 #include <net/ip_fib.h>
  42 #include <net/nexthop.h>
  43 #include <net/rtnetlink.h>
  44 #include <net/xfrm.h>
  45 #include <net/l3mdev.h>
  46 #include <net/lwtunnel.h>
  47 #include <trace/events/fib.h>
  48
  49 #ifndef CONFIG_IP_MULTIPLE_TABLES
  50
  51 static int __net_init fib4_rules_init(struct net *net)
  52 {
  53         struct fib_table *local_table, *main_table;
  54
  55         main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
  56         if (!main_table)
  57                 return -ENOMEM;
  58
  59         local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
  60         if (!local_table)
  61                 goto fail;
  62
  63         hlist_add_head_rcu(&local_table->tb_hlist,
  64                                 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
  65         hlist_add_head_rcu(&main_table->tb_hlist,
  66                                 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
  67         return 0;
  68
  69 fail:
  70         fib_free_table(main_table);
  71         return -ENOMEM;
  72 }
  73
  74 static bool fib4_has_custom_rules(struct net *net)
  75 {
  76         return false;
  77 }
  78 #else
  79
  80 struct fib_table *fib_new_table(struct net *net, u32 id)
  81 {
  82         struct fib_table *tb, *alias = NULL;
  83         unsigned int h;
  84
  85         if (id == 0)
  86                 id = RT_TABLE_MAIN;
  87         tb = fib_get_table(net, id);
  88         if (tb)
  89                 return tb;
  90
  91         if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
  92                 alias = fib_new_table(net, RT_TABLE_MAIN);
  93
  94         tb = fib_trie_table(id, alias);
  95         if (!tb)
  96                 return NULL;
  97
  98         switch (id) {
  99         case RT_TABLE_MAIN:
 100                 rcu_assign_pointer(net->ipv4.fib_main, tb);
 101                 break;
 102         case RT_TABLE_DEFAULT:
 103                 rcu_assign_pointer(net->ipv4.fib_default, tb);
 104                 break;
 105         default:
 106                 break;
 107         }
 108
 109         h = id & (FIB_TABLE_HASHSZ - 1);
 110         hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
 111         return tb;
 112 }
 113 EXPORT_SYMBOL_GPL(fib_new_table);
 114
 115 /* caller must hold either rtnl or rcu read lock */
 116 struct fib_table *fib_get_table(struct net *net, u32 id)
 117 {
 118         struct fib_table *tb;
 119         struct hlist_head *head;
 120         unsigned int h;
 121
 122         if (id == 0)
 123                 id = RT_TABLE_MAIN;
 124         h = id & (FIB_TABLE_HASHSZ - 1);
 125
 126         head = &net->ipv4.fib_table_hash[h];
 127         hlist_for_each_entry_rcu(tb, head, tb_hlist,
 128                                  lockdep_rtnl_is_held()) {
 129                 if (tb->tb_id == id)
 130                         return tb;
 131         }
 132         return NULL;
 133 }
 134
 135 static bool fib4_has_custom_rules(struct net *net)
 136 {
 137         return net->ipv4.fib_has_custom_rules;
 138 }
 139 #endif /* CONFIG_IP_MULTIPLE_TABLES */
 140
 141 static void fib_replace_table(struct net *net, struct fib_table *old,
 142                               struct fib_table *new)
 143 {
 144 #ifdef CONFIG_IP_MULTIPLE_TABLES
 145         switch (new->tb_id) {
 146         case RT_TABLE_MAIN:
 147                 rcu_assign_pointer(net->ipv4.fib_main, new);
 148                 break;
 149         case RT_TABLE_DEFAULT:
 150                 rcu_assign_pointer(net->ipv4.fib_default, new);
 151                 break;
 152         default:
 153                 break;
 154         }
 155
 156 #endif
 157         /* replace the old table in the hlist */
 158         hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
 159 }
 160
 161 int fib_unmerge(struct net *net)
 162 {
 163         struct fib_table *old, *new, *main_table;
 164
 165         /* attempt to fetch local table if it has been allocated */
 166         old = fib_get_table(net, RT_TABLE_LOCAL);
 167         if (!old)
 168                 return 0;
 169
 170         new = fib_trie_unmerge(old);
 171         if (!new)
 172                 return -ENOMEM;
 173
 174         /* table is already unmerged */
 175         if (new == old)
 176                 return 0;
 177
 178         /* replace merged table with clean table */
 179         fib_replace_table(net, old, new);
 180         fib_free_table(old);
 181
 182         /* attempt to fetch main table if it has been allocated */
 183         main_table = fib_get_table(net, RT_TABLE_MAIN);
 184         if (!main_table)
 185                 return 0;
 186
 187         /* flush local entries from main table */
 188         fib_table_flush_external(main_table);
 189
 190         return 0;
 191 }
 192
 193 void fib_flush(struct net *net)
 194 {
 195         int flushed = 0;
 196         unsigned int h;
 197
 198         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
 199                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
 200                 struct hlist_node *tmp;
 201                 struct fib_table *tb;
 202
 203                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
 204                         flushed += fib_table_flush(net, tb, false);
 205         }
 206
 207         if (flushed)
 208                 rt_cache_flush(net);
 209 }
 210
 211 /*
 212  * Find address type as if only "dev" was present in the system. If
 213  * on_dev is NULL then all interfaces are taken into consideration.
 214  */
 215 static inline unsigned int __inet_dev_addr_type(struct net *net,
 216                                                 const struct net_device *dev,
 217                                                 __be32 addr, u32 tb_id)
 218 {
 219         struct flowi4           fl4 = { .daddr = addr };
 220         struct fib_result       res;
 221         unsigned int ret = RTN_BROADCAST;
 222         struct fib_table *table;
 223
 224         if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
 225                 return RTN_BROADCAST;
 226         if (ipv4_is_multicast(addr))
 227                 return RTN_MULTICAST;
 228
 229         rcu_read_lock();
 230
 231         table = fib_get_table(net, tb_id);
 232         if (table) {
 233                 ret = RTN_UNICAST;
 234                 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
 235                         struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0);
 236
 237                         if (!dev || dev == nhc->nhc_dev)
 238                                 ret = res.type;
 239                 }
 240         }
 241
 242         rcu_read_unlock();
 243         return ret;
 244 }
 245
 246 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
 247 {
 248         return __inet_dev_addr_type(net, NULL, addr, tb_id);
 249 }
 250 EXPORT_SYMBOL(inet_addr_type_table);
 251
 252 unsigned int inet_addr_type(struct net *net, __be32 addr)
 253 {
 254         return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
 255 }
 256 EXPORT_SYMBOL(inet_addr_type);
 257
 258 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
 259                                 __be32 addr)
 260 {
 261         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
 262
 263         return __inet_dev_addr_type(net, dev, addr, rt_table);
 264 }
 265 EXPORT_SYMBOL(inet_dev_addr_type);
 266
 267 /* inet_addr_type with dev == NULL but using the table from a dev
 268  * if one is associated
 269  */
 270 unsigned int inet_addr_type_dev_table(struct net *net,
 271                                       const struct net_device *dev,
 272                                       __be32 addr)
 273 {
 274         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
 275
 276         return __inet_dev_addr_type(net, NULL, addr, rt_table);
 277 }
 278 EXPORT_SYMBOL(inet_addr_type_dev_table);
 279
 280 __be32 fib_compute_spec_dst(struct sk_buff *skb)
 281 {
 282         struct net_device *dev = skb->dev;
 283         struct in_device *in_dev;
 284         struct fib_result res;
 285         struct rtable *rt;
 286         struct net *net;
 287         int scope;
 288
 289         rt = skb_rtable(skb);
 290         if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
 291             RTCF_LOCAL)
 292                 return ip_hdr(skb)->daddr;
 293
 294         in_dev = __in_dev_get_rcu(dev);
 295
 296         net = dev_net(dev);
 297
 298         scope = RT_SCOPE_UNIVERSE;
 299         if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
 300                 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
 301                 struct flowi4 fl4 = {
 302                         .flowi4_iif = LOOPBACK_IFINDEX,
 303                         .flowi4_oif = l3mdev_master_ifindex_rcu(dev),
 304                         .daddr = ip_hdr(skb)->saddr,
 305                         .flowi4_tos = ip_hdr(skb)->tos & IPTOS_RT_MASK,
 306                         .flowi4_scope = scope,
 307                         .flowi4_mark = vmark ? skb->mark : 0,
 308                 };
 309                 if (!fib_lookup(net, &fl4, &res, 0))
 310                         return fib_result_prefsrc(net, &res);
 311         } else {
 312                 scope = RT_SCOPE_LINK;
 313         }
 314
 315         return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
 316 }
 317
 318 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
 319 {
 320         bool dev_match = false;
 321 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 322         if (unlikely(fi->nh)) {
 323                 dev_match = nexthop_uses_dev(fi->nh, dev);
 324         } else {
 325                 int ret;
 326
 327                 for (ret = 0; ret < fib_info_num_path(fi); ret++) {
 328                         const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
 329
 330                         if (nhc_l3mdev_matches_dev(nhc, dev)) {
 331                                 dev_match = true;
 332                                 break;
 333                         }
 334                 }
 335         }
 336 #else
 337         if (fib_info_nhc(fi, 0)->nhc_dev == dev)
 338                 dev_match = true;
 339 #endif
 340
 341         return dev_match;
 342 }
 343 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
 344
 345 /* Given (packet source, input interface) and optional (dst, oif, tos):
 346  * - (main) check, that source is valid i.e. not broadcast or our local
 347  *   address.
 348  * - figure out what "logical" interface this packet arrived
 349  *   and calculate "specific destination" address.
 350  * - check, that packet arrived from expected physical interface.
 351  * called with rcu_read_lock()
 352  */
 353 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 354                                  u8 tos, int oif, struct net_device *dev,
 355                                  int rpf, struct in_device *idev, u32 *itag)
 356 {
 357         struct net *net = dev_net(dev);
 358         struct flow_keys flkeys;
 359         int ret, no_addr;
 360         struct fib_result res;
 361         struct flowi4 fl4;
 362         bool dev_match;
 363
 364         fl4.flowi4_oif = 0;
 365         fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
 366         if (!fl4.flowi4_iif)
 367                 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
 368         fl4.daddr = src;
 369         fl4.saddr = dst;
 370         fl4.flowi4_tos = tos;
 371         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 372         fl4.flowi4_tun_key.tun_id = 0;
 373         fl4.flowi4_flags = 0;
 374         fl4.flowi4_uid = sock_net_uid(net, NULL);
 375         fl4.flowi4_multipath_hash = 0;
 376
 377         no_addr = idev->ifa_list == NULL;
 378
 379         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
 380         if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
 381                 fl4.flowi4_proto = 0;
 382                 fl4.fl4_sport = 0;
 383                 fl4.fl4_dport = 0;
 384         } else {
 385                 swap(fl4.fl4_sport, fl4.fl4_dport);
 386         }
 387
 388         if (fib_lookup(net, &fl4, &res, 0))
 389                 goto last_resort;
 390         if (res.type != RTN_UNICAST &&
 391             (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
 392                 goto e_inval;
 393         fib_combine_itag(itag, &res);
 394
 395         dev_match = fib_info_nh_uses_dev(res.fi, dev);
 396         /* This is not common, loopback packets retain skb_dst so normally they
 397          * would not even hit this slow path.
 398          */
 399         dev_match = dev_match || (res.type == RTN_LOCAL &&
 400                                   dev == net->loopback_dev);
 401         if (dev_match) {
 402                 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_LINK;
 403                 return ret;
 404         }
 405         if (no_addr)
 406                 goto last_resort;
 407         if (rpf == 1)
 408                 goto e_rpf;
 409         fl4.flowi4_oif = dev->ifindex;
 410
 411         ret = 0;
 412         if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
 413                 if (res.type == RTN_UNICAST)
 414                         ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_LINK;
 415         }
 416         return ret;
 417
 418 last_resort:
 419         if (rpf)
 420                 goto e_rpf;
 421         *itag = 0;
 422         return 0;
 423
 424 e_inval:
 425         return -EINVAL;
 426 e_rpf:
 427         return -EXDEV;
 428 }
 429
 430 /* Ignore rp_filter for packets protected by IPsec. */
 431 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 432                         u8 tos, int oif, struct net_device *dev,
 433                         struct in_device *idev, u32 *itag)
 434 {
 435         int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
 436         struct net *net = dev_net(dev);
 437
 438         if (!r && !fib_num_tclassid_users(net) &&
 439             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
 440                 if (IN_DEV_ACCEPT_LOCAL(idev))
 441                         goto ok;
 442                 /* with custom local routes in place, checking local addresses
 443                  * only will be too optimistic, with custom rules, checking
 444                  * local addresses only can be too strict, e.g. due to vrf
 445                  */
 446                 if (net->ipv4.fib_has_custom_local_routes ||
 447                     fib4_has_custom_rules(net))
 448                         goto full_check;
 449                 if (inet_lookup_ifaddr_rcu(net, src))
 450                         return -EINVAL;
 451
 452 ok:
 453                 *itag = 0;
 454                 return 0;
 455         }
 456
 457 full_check:
 458         return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
 459 }
 460
 461 static inline __be32 sk_extract_addr(struct sockaddr *addr)
 462 {
 463         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 464 }
 465
 466 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
 467 {
 468         struct nlattr *nla;
 469
 470         nla = (struct nlattr *) ((char *) mx + len);
 471         nla->nla_type = type;
 472         nla->nla_len = nla_attr_size(4);
 473         *(u32 *) nla_data(nla) = value;
 474
 475         return len + nla_total_size(4);
 476 }
 477
 478 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
 479                                  struct fib_config *cfg)
 480 {
 481         __be32 addr;
 482         int plen;
 483
 484         memset(cfg, 0, sizeof(*cfg));
 485         cfg->fc_nlinfo.nl_net = net;
 486
 487         if (rt->rt_dst.sa_family != AF_INET)
 488                 return -EAFNOSUPPORT;
 489
 490         /*
 491          * Check mask for validity:
 492          * a) it must be contiguous.
 493          * b) destination must have all host bits clear.
 494          * c) if application forgot to set correct family (AF_INET),
 495          *    reject request unless it is absolutely clear i.e.
 496          *    both family and mask are zero.
 497          */
 498         plen = 32;
 499         addr = sk_extract_addr(&rt->rt_dst);
 500         if (!(rt->rt_flags & RTF_HOST)) {
 501                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
 502
 503                 if (rt->rt_genmask.sa_family != AF_INET) {
 504                         if (mask || rt->rt_genmask.sa_family)
 505                                 return -EAFNOSUPPORT;
 506                 }
 507
 508                 if (bad_mask(mask, addr))
 509                         return -EINVAL;
 510
 511                 plen = inet_mask_len(mask);
 512         }
 513
 514         cfg->fc_dst_len = plen;
 515         cfg->fc_dst = addr;
 516
 517         if (cmd != SIOCDELRT) {
 518                 cfg->fc_nlflags = NLM_F_CREATE;
 519                 cfg->fc_protocol = RTPROT_BOOT;
 520         }
 521
 522         if (rt->rt_metric)
 523                 cfg->fc_priority = rt->rt_metric - 1;
 524
 525         if (rt->rt_flags & RTF_REJECT) {
 526                 cfg->fc_scope = RT_SCOPE_HOST;
 527                 cfg->fc_type = RTN_UNREACHABLE;
 528                 return 0;
 529         }
 530
 531         cfg->fc_scope = RT_SCOPE_NOWHERE;
 532         cfg->fc_type = RTN_UNICAST;
 533
 534         if (rt->rt_dev) {
 535                 char *colon;
 536                 struct net_device *dev;
 537                 char devname[IFNAMSIZ];
 538
 539                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
 540                         return -EFAULT;
 541
 542                 devname[IFNAMSIZ-1] = 0;
 543                 colon = strchr(devname, ':');
 544                 if (colon)
 545                         *colon = 0;
 546                 dev = __dev_get_by_name(net, devname);
 547                 if (!dev)
 548                         return -ENODEV;
 549                 cfg->fc_oif = dev->ifindex;
 550                 cfg->fc_table = l3mdev_fib_table(dev);
 551                 if (colon) {
 552                         const struct in_ifaddr *ifa;
 553                         struct in_device *in_dev;
 554
 555                         in_dev = __in_dev_get_rtnl(dev);
 556                         if (!in_dev)
 557                                 return -ENODEV;
 558
 559                         *colon = ':';
 560
 561                         rcu_read_lock();
 562                         in_dev_for_each_ifa_rcu(ifa, in_dev) {
 563                                 if (strcmp(ifa->ifa_label, devname) == 0)
 564                                         break;
 565                         }
 566                         rcu_read_unlock();
 567
 568                         if (!ifa)
 569                                 return -ENODEV;
 570                         cfg->fc_prefsrc = ifa->ifa_local;
 571                 }
 572         }
 573
 574         addr = sk_extract_addr(&rt->rt_gateway);
 575         if (rt->rt_gateway.sa_family == AF_INET && addr) {
 576                 unsigned int addr_type;
 577
 578                 cfg->fc_gw4 = addr;
 579                 cfg->fc_gw_family = AF_INET;
 580                 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
 581                 if (rt->rt_flags & RTF_GATEWAY &&
 582                     addr_type == RTN_UNICAST)
 583                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
 584         }
 585
 586         if (!cfg->fc_table)
 587                 cfg->fc_table = RT_TABLE_MAIN;
 588
 589         if (cmd == SIOCDELRT)
 590                 return 0;
 591
 592         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
 593                 return -EINVAL;
 594
 595         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 596                 cfg->fc_scope = RT_SCOPE_LINK;
 597
 598         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 599                 struct nlattr *mx;
 600                 int len = 0;
 601
 602                 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
 603                 if (!mx)
 604                         return -ENOMEM;
 605
 606                 if (rt->rt_flags & RTF_MTU)
 607                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 608
 609                 if (rt->rt_flags & RTF_WINDOW)
 610                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 611
 612                 if (rt->rt_flags & RTF_IRTT)
 613                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 614
 615                 cfg->fc_mx = mx;
 616                 cfg->fc_mx_len = len;
 617         }
 618
 619         return 0;
 620 }
 621
 622 /*
 623  * Handle IP routing ioctl calls.
 624  * These are used to manipulate the routing tables
 625  */
 626 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
 627 {
 628         struct fib_config cfg;
 629         int err;
 630
 631         switch (cmd) {
 632         case SIOCADDRT:         /* Add a route */
 633         case SIOCDELRT:         /* Delete a route */
 634                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 635                         return -EPERM;
 636
 637                 rtnl_lock();
 638                 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
 639                 if (err == 0) {
 640                         struct fib_table *tb;
 641
 642                         if (cmd == SIOCDELRT) {
 643                                 tb = fib_get_table(net, cfg.fc_table);
 644                                 if (tb)
 645                                         err = fib_table_delete(net, tb, &cfg,
 646                                                                NULL);
 647                                 else
 648                                         err = -ESRCH;
 649                         } else {
 650                                 tb = fib_new_table(net, cfg.fc_table);
 651                                 if (tb)
 652                                         err = fib_table_insert(net, tb,
 653                                                                &cfg, NULL);
 654                                 else
 655                                         err = -ENOBUFS;
 656                         }
 657
 658                         /* allocated by rtentry_to_fib_config() */
 659                         kfree(cfg.fc_mx);
 660                 }
 661                 rtnl_unlock();
 662                 return err;
 663         }
 664         return -EINVAL;
 665 }
 666
 667 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 668         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
 669         [RTA_DST]               = { .type = NLA_U32 },
 670         [RTA_SRC]               = { .type = NLA_U32 },
 671         [RTA_IIF]               = { .type = NLA_U32 },
 672         [RTA_OIF]               = { .type = NLA_U32 },
 673         [RTA_GATEWAY]           = { .type = NLA_U32 },
 674         [RTA_PRIORITY]          = { .type = NLA_U32 },
 675         [RTA_PREFSRC]           = { .type = NLA_U32 },
 676         [RTA_METRICS]           = { .type = NLA_NESTED },
 677         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 678         [RTA_FLOW]              = { .type = NLA_U32 },
 679         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
 680         [RTA_ENCAP]             = { .type = NLA_NESTED },
 681         [RTA_UID]               = { .type = NLA_U32 },
 682         [RTA_MARK]              = { .type = NLA_U32 },
 683         [RTA_TABLE]             = { .type = NLA_U32 },
 684         [RTA_IP_PROTO]          = { .type = NLA_U8 },
 685         [RTA_SPORT]             = { .type = NLA_U16 },
 686         [RTA_DPORT]             = { .type = NLA_U16 },
 687         [RTA_NH_ID]             = { .type = NLA_U32 },
 688 };
 689
 690 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
 691                     struct netlink_ext_ack *extack)
 692 {
 693         struct rtvia *via;
 694         int alen;
 695
 696         if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
 697                 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
 698                 return -EINVAL;
 699         }
 700
 701         via = nla_data(nla);
 702         alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
 703
 704         switch (via->rtvia_family) {
 705         case AF_INET:
 706                 if (alen != sizeof(__be32)) {
 707                         NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
 708                         return -EINVAL;
 709                 }
 710                 cfg->fc_gw_family = AF_INET;
 711                 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
 712                 break;
 713         case AF_INET6:
 714 #if IS_ENABLED(CONFIG_IPV6)
 715                 if (alen != sizeof(struct in6_addr)) {
 716                         NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
 717                         return -EINVAL;
 718                 }
 719                 cfg->fc_gw_family = AF_INET6;
 720                 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
 721 #else
 722                 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
 723                 return -EINVAL;
 724 #endif
 725                 break;
 726         default:
 727                 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
 728                 return -EINVAL;
 729         }
 730
 731         return 0;
 732 }
 733
 734 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 735                              struct nlmsghdr *nlh, struct fib_config *cfg,
 736                              struct netlink_ext_ack *extack)
 737 {
 738         bool has_gw = false, has_via = false;
 739         struct nlattr *attr;
 740         int err, remaining;
 741         struct rtmsg *rtm;
 742
 743         err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
 744                                         rtm_ipv4_policy, extack);
 745         if (err < 0)
 746                 goto errout;
 747
 748         memset(cfg, 0, sizeof(*cfg));
 749
 750         rtm = nlmsg_data(nlh);
 751         cfg->fc_dst_len = rtm->rtm_dst_len;
 752         cfg->fc_tos = rtm->rtm_tos;
 753         cfg->fc_table = rtm->rtm_table;
 754         cfg->fc_protocol = rtm->rtm_protocol;
 755         cfg->fc_scope = rtm->rtm_scope;
 756         cfg->fc_type = rtm->rtm_type;
 757         cfg->fc_flags = rtm->rtm_flags;
 758         cfg->fc_nlflags = nlh->nlmsg_flags;
 759
 760         cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
 761         cfg->fc_nlinfo.nlh = nlh;
 762         cfg->fc_nlinfo.nl_net = net;
 763
 764         if (cfg->fc_type > RTN_MAX) {
 765                 NL_SET_ERR_MSG(extack, "Invalid route type");
 766                 err = -EINVAL;
 767                 goto errout;
 768         }
 769
 770         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 771                 switch (nla_type(attr)) {
 772                 case RTA_DST:
 773                         cfg->fc_dst = nla_get_be32(attr);
 774                         break;
 775                 case RTA_OIF:
 776                         cfg->fc_oif = nla_get_u32(attr);
 777                         break;
 778                 case RTA_GATEWAY:
 779                         has_gw = true;
 780                         cfg->fc_gw4 = nla_get_be32(attr);
 781                         if (cfg->fc_gw4)
 782                                 cfg->fc_gw_family = AF_INET;
 783                         break;
 784                 case RTA_VIA:
 785                         has_via = true;
 786                         err = fib_gw_from_via(cfg, attr, extack);
 787                         if (err)
 788                                 goto errout;
 789                         break;
 790                 case RTA_PRIORITY:
 791                         cfg->fc_priority = nla_get_u32(attr);
 792                         break;
 793                 case RTA_PREFSRC:
 794                         cfg->fc_prefsrc = nla_get_be32(attr);
 795                         break;
 796                 case RTA_METRICS:
 797                         cfg->fc_mx = nla_data(attr);
 798                         cfg->fc_mx_len = nla_len(attr);
 799                         break;
 800                 case RTA_MULTIPATH:
 801                         err = lwtunnel_valid_encap_type_attr(nla_data(attr),
 802                                                              nla_len(attr),
 803                                                              extack);
 804                         if (err < 0)
 805                                 goto errout;
 806                         cfg->fc_mp = nla_data(attr);
 807                         cfg->fc_mp_len = nla_len(attr);
 808                         break;
 809                 case RTA_FLOW:
 810                         cfg->fc_flow = nla_get_u32(attr);
 811                         break;
 812                 case RTA_TABLE:
 813                         cfg->fc_table = nla_get_u32(attr);
 814                         break;
 815                 case RTA_ENCAP:
 816                         cfg->fc_encap = attr;
 817                         break;
 818                 case RTA_ENCAP_TYPE:
 819                         cfg->fc_encap_type = nla_get_u16(attr);
 820                         err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
 821                                                         extack);
 822                         if (err < 0)
 823                                 goto errout;
 824                         break;
 825                 case RTA_NH_ID:
 826                         cfg->fc_nh_id = nla_get_u32(attr);
 827                         break;
 828                 }
 829         }
 830
 831         if (cfg->fc_nh_id) {
 832                 if (cfg->fc_oif || cfg->fc_gw_family ||
 833                     cfg->fc_encap || cfg->fc_mp) {
 834                         NL_SET_ERR_MSG(extack,
 835                                        "Nexthop specification and nexthop id are mutually exclusive");
 836                         return -EINVAL;
 837                 }
 838         }
 839
 840         if (has_gw && has_via) {
 841                 NL_SET_ERR_MSG(extack,
 842                                "Nexthop configuration can not contain both GATEWAY and VIA");
 843                 return -EINVAL;
 844         }
 845
 846         if (!cfg->fc_table)
 847                 cfg->fc_table = RT_TABLE_MAIN;
 848
 849         return 0;
 850 errout:
 851         return err;
 852 }
 853
 854 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 855                              struct netlink_ext_ack *extack)
 856 {
 857         struct net *net = sock_net(skb->sk);
 858         struct fib_config cfg;
 859         struct fib_table *tb;
 860         int err;
 861
 862         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 863         if (err < 0)
 864                 goto errout;
 865
 866         if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
 867                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
 868                 err = -EINVAL;
 869                 goto errout;
 870         }
 871
 872         tb = fib_get_table(net, cfg.fc_table);
 873         if (!tb) {
 874                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
 875                 err = -ESRCH;
 876                 goto errout;
 877         }
 878
 879         err = fib_table_delete(net, tb, &cfg, extack);
 880 errout:
 881         return err;
 882 }
 883
 884 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 885                              struct netlink_ext_ack *extack)
 886 {
 887         struct net *net = sock_net(skb->sk);
 888         struct fib_config cfg;
 889         struct fib_table *tb;
 890         int err;
 891
 892         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 893         if (err < 0)
 894                 goto errout;
 895
 896         tb = fib_new_table(net, cfg.fc_table);
 897         if (!tb) {
 898                 err = -ENOBUFS;
 899                 goto errout;
 900         }
 901
 902         err = fib_table_insert(net, tb, &cfg, extack);
 903         if (!err && cfg.fc_type == RTN_LOCAL)
 904                 net->ipv4.fib_has_custom_local_routes = true;
 905 errout:
 906         return err;
 907 }
 908
 909 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
 910                           struct fib_dump_filter *filter,
 911                           struct netlink_callback *cb)
 912 {
 913         struct netlink_ext_ack *extack = cb->extack;
 914         struct nlattr *tb[RTA_MAX + 1];
 915         struct rtmsg *rtm;
 916         int err, i;
 917
 918         ASSERT_RTNL();
 919
 920         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
 921                 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
 922                 return -EINVAL;
 923         }
 924
 925         rtm = nlmsg_data(nlh);
 926         if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
 927             rtm->rtm_scope) {
 928                 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
 929                 return -EINVAL;
 930         }
 931
 932         if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
 933                 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
 934                 return -EINVAL;
 935         }
 936         if (rtm->rtm_flags & RTM_F_CLONED)
 937                 filter->dump_routes = false;
 938         else
 939                 filter->dump_exceptions = false;
 940
 941         filter->flags    = rtm->rtm_flags;
 942         filter->protocol = rtm->rtm_protocol;
 943         filter->rt_type  = rtm->rtm_type;
 944         filter->table_id = rtm->rtm_table;
 945
 946         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
 947                                             rtm_ipv4_policy, extack);
 948         if (err < 0)
 949                 return err;
 950
 951         for (i = 0; i <= RTA_MAX; ++i) {
 952                 int ifindex;
 953
 954                 if (!tb[i])
 955                         continue;
 956
 957                 switch (i) {
 958                 case RTA_TABLE:
 959                         filter->table_id = nla_get_u32(tb[i]);
 960                         break;
 961                 case RTA_OIF:
 962                         ifindex = nla_get_u32(tb[i]);
 963                         filter->dev = __dev_get_by_index(net, ifindex);
 964                         if (!filter->dev)
 965                                 return -ENODEV;
 966                         break;
 967                 default:
 968                         NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
 969                         return -EINVAL;
 970                 }
 971         }
 972
 973         if (filter->flags || filter->protocol || filter->rt_type ||
 974             filter->table_id || filter->dev) {
 975                 filter->filter_set = 1;
 976                 cb->answer_flags = NLM_F_DUMP_FILTERED;
 977         }
 978
 979         return 0;
 980 }
 981 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
 982
 983 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 984 {
 985         struct fib_dump_filter filter = { .dump_routes = true,
 986                                           .dump_exceptions = true };
 987         const struct nlmsghdr *nlh = cb->nlh;
 988         struct net *net = sock_net(skb->sk);
 989         unsigned int h, s_h;
 990         unsigned int e = 0, s_e;
 991         struct fib_table *tb;
 992         struct hlist_head *head;
 993         int dumped = 0, err;
 994
 995         if (cb->strict_check) {
 996                 err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
 997                 if (err < 0)
 998                         return err;
 999         } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
1000                 struct rtmsg *rtm = nlmsg_data(nlh);
1001
1002                 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
1003         }
1004
1005         /* ipv4 does not use prefix flag */
1006         if (filter.flags & RTM_F_PREFIX)
1007                 return skb->len;
1008
1009         if (filter.table_id) {
1010                 tb = fib_get_table(net, filter.table_id);
1011                 if (!tb) {
1012                         if (rtnl_msg_family(cb->nlh) != PF_INET)
1013                                 return skb->len;
1014
1015                         NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1016                         return -ENOENT;
1017                 }
1018
1019                 rcu_read_lock();
1020                 err = fib_table_dump(tb, skb, cb, &filter);
1021                 rcu_read_unlock();
1022                 return skb->len ? : err;
1023         }
1024
1025         s_h = cb->args[0];
1026         s_e = cb->args[1];
1027
1028         rcu_read_lock();
1029
1030         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1031                 e = 0;
1032                 head = &net->ipv4.fib_table_hash[h];
1033                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1034                         if (e < s_e)
1035                                 goto next;
1036                         if (dumped)
1037                                 memset(&cb->args[2], 0, sizeof(cb->args) -
1038                                                  2 * sizeof(cb->args[0]));
1039                         err = fib_table_dump(tb, skb, cb, &filter);
1040                         if (err < 0) {
1041                                 if (likely(skb->len))
1042                                         goto out;
1043
1044                                 goto out_err;
1045                         }
1046                         dumped = 1;
1047 next:
1048                         e++;
1049                 }
1050         }
1051 out:
1052         err = skb->len;
1053 out_err:
1054         rcu_read_unlock();
1055
1056         cb->args[1] = e;
1057         cb->args[0] = h;
1058
1059         return err;
1060 }
1061
1062 /* Prepare and feed intra-kernel routing request.
1063  * Really, it should be netlink message, but :-( netlink
1064  * can be not configured, so that we feed it directly
1065  * to fib engine. It is legal, because all events occur
1066  * only when netlink is already locked.
1067  */
1068 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1069                       struct in_ifaddr *ifa, u32 rt_priority)
1070 {
1071         struct net *net = dev_net(ifa->ifa_dev->dev);
1072         u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1073         struct fib_table *tb;
1074         struct fib_config cfg = {
1075                 .fc_protocol = RTPROT_KERNEL,
1076                 .fc_type = type,
1077                 .fc_dst = dst,
1078                 .fc_dst_len = dst_len,
1079                 .fc_priority = rt_priority,
1080                 .fc_prefsrc = ifa->ifa_local,
1081                 .fc_oif = ifa->ifa_dev->dev->ifindex,
1082                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1083                 .fc_nlinfo = {
1084                         .nl_net = net,
1085                 },
1086         };
1087
1088         if (!tb_id)
1089                 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1090
1091         tb = fib_new_table(net, tb_id);
1092         if (!tb)
1093                 return;
1094
1095         cfg.fc_table = tb->tb_id;
1096
1097         if (type != RTN_LOCAL)
1098                 cfg.fc_scope = RT_SCOPE_LINK;
1099         else
1100                 cfg.fc_scope = RT_SCOPE_HOST;
1101
1102         if (cmd == RTM_NEWROUTE)
1103                 fib_table_insert(net, tb, &cfg, NULL);
1104         else
1105                 fib_table_delete(net, tb, &cfg, NULL);
1106 }
1107
1108 void fib_add_ifaddr(struct in_ifaddr *ifa)
1109 {
1110         struct in_device *in_dev = ifa->ifa_dev;
1111         struct net_device *dev = in_dev->dev;
1112         struct in_ifaddr *prim = ifa;
1113         __be32 mask = ifa->ifa_mask;
1114         __be32 addr = ifa->ifa_local;
1115         __be32 prefix = ifa->ifa_address & mask;
1116
1117         if (ifa->ifa_flags & IFA_F_SECONDARY) {
1118                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
1119                 if (!prim) {
1120                         pr_warn("%s: bug: prim == NULL\n", __func__);
1121                         return;
1122                 }
1123         }
1124
1125         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1126
1127         if (!(dev->flags & IFF_UP))
1128                 return;
1129
1130         /* Add broadcast address, if it is explicitly assigned. */
1131         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) {
1132                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1133                           prim, 0);
1134                 arp_invalidate(dev, ifa->ifa_broadcast, false);
1135         }
1136
1137         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1138             (prefix != addr || ifa->ifa_prefixlen < 32)) {
1139                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1140                         fib_magic(RTM_NEWROUTE,
1141                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1142                                   prefix, ifa->ifa_prefixlen, prim,
1143                                   ifa->ifa_rt_priority);
1144
1145                 /* Add network specific broadcasts, when it takes a sense */
1146                 if (ifa->ifa_prefixlen < 31) {
1147                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1148                                   prim, 0);
1149                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1150                                   32, prim, 0);
1151                         arp_invalidate(dev, prefix | ~mask, false);
1152                 }
1153         }
1154 }
1155
1156 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1157 {
1158         __be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1159         struct in_device *in_dev = ifa->ifa_dev;
1160         struct net_device *dev = in_dev->dev;
1161
1162         if (!(dev->flags & IFF_UP) ||
1163             ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1164             ipv4_is_zeronet(prefix) ||
1165             (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32))
1166                 return;
1167
1168         /* add the new */
1169         fib_magic(RTM_NEWROUTE,
1170                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1171                   prefix, ifa->ifa_prefixlen, ifa, new_metric);
1172
1173         /* delete the old */
1174         fib_magic(RTM_DELROUTE,
1175                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1176                   prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1177 }
1178
1179 /* Delete primary or secondary address.
1180  * Optionally, on secondary address promotion consider the addresses
1181  * from subnet iprim as deleted, even if they are in device list.
1182  * In this case the secondary ifa can be in device list.
1183  */
1184 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1185 {
1186         struct in_device *in_dev = ifa->ifa_dev;
1187         struct net_device *dev = in_dev->dev;
1188         struct in_ifaddr *ifa1;
1189         struct in_ifaddr *prim = ifa, *prim1 = NULL;
1190         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1191         __be32 any = ifa->ifa_address & ifa->ifa_mask;
1192 #define LOCAL_OK        1
1193 #define BRD_OK          2
1194 #define BRD0_OK         4
1195 #define BRD1_OK         8
1196         unsigned int ok = 0;
1197         int subnet = 0;         /* Primary network */
1198         int gone = 1;           /* Address is missing */
1199         int same_prefsrc = 0;   /* Another primary with same IP */
1200
1201         if (ifa->ifa_flags & IFA_F_SECONDARY) {
1202                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1203                 if (!prim) {
1204                         /* if the device has been deleted, we don't perform
1205                          * address promotion
1206                          */
1207                         if (!in_dev->dead)
1208                                 pr_warn("%s: bug: prim == NULL\n", __func__);
1209                         return;
1210                 }
1211                 if (iprim && iprim != prim) {
1212                         pr_warn("%s: bug: iprim != prim\n", __func__);
1213                         return;
1214                 }
1215         } else if (!ipv4_is_zeronet(any) &&
1216                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1217                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1218                         fib_magic(RTM_DELROUTE,
1219                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1220                                   any, ifa->ifa_prefixlen, prim, 0);
1221                 subnet = 1;
1222         }
1223
1224         if (in_dev->dead)
1225                 goto no_promotions;
1226
1227         /* Deletion is more complicated than add.
1228          * We should take care of not to delete too much :-)
1229          *
1230          * Scan address list to be sure that addresses are really gone.
1231          */
1232         rcu_read_lock();
1233         in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1234                 if (ifa1 == ifa) {
1235                         /* promotion, keep the IP */
1236                         gone = 0;
1237                         continue;
1238                 }
1239                 /* Ignore IFAs from our subnet */
1240                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1241                     inet_ifa_match(ifa1->ifa_address, iprim))
1242                         continue;
1243
1244                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
1245                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1246                         /* Another address from our subnet? */
1247                         if (ifa1->ifa_mask == prim->ifa_mask &&
1248                             inet_ifa_match(ifa1->ifa_address, prim))
1249                                 prim1 = prim;
1250                         else {
1251                                 /* We reached the secondaries, so
1252                                  * same_prefsrc should be determined.
1253                                  */
1254                                 if (!same_prefsrc)
1255                                         continue;
1256                                 /* Search new prim1 if ifa1 is not
1257                                  * using the current prim1
1258                                  */
1259                                 if (!prim1 ||
1260                                     ifa1->ifa_mask != prim1->ifa_mask ||
1261                                     !inet_ifa_match(ifa1->ifa_address, prim1))
1262                                         prim1 = inet_ifa_byprefix(in_dev,
1263                                                         ifa1->ifa_address,
1264                                                         ifa1->ifa_mask);
1265                                 if (!prim1)
1266                                         continue;
1267                                 if (prim1->ifa_local != prim->ifa_local)
1268                                         continue;
1269                         }
1270                 } else {
1271                         if (prim->ifa_local != ifa1->ifa_local)
1272                                 continue;
1273                         prim1 = ifa1;
1274                         if (prim != prim1)
1275                                 same_prefsrc = 1;
1276                 }
1277                 if (ifa->ifa_local == ifa1->ifa_local)
1278                         ok |= LOCAL_OK;
1279                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1280                         ok |= BRD_OK;
1281                 if (brd == ifa1->ifa_broadcast)
1282                         ok |= BRD1_OK;
1283                 if (any == ifa1->ifa_broadcast)
1284                         ok |= BRD0_OK;
1285                 /* primary has network specific broadcasts */
1286                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1287                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1288                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1289
1290                         if (!ipv4_is_zeronet(any1)) {
1291                                 if (ifa->ifa_broadcast == brd1 ||
1292                                     ifa->ifa_broadcast == any1)
1293                                         ok |= BRD_OK;
1294                                 if (brd == brd1 || brd == any1)
1295                                         ok |= BRD1_OK;
1296                                 if (any == brd1 || any == any1)
1297                                         ok |= BRD0_OK;
1298                         }
1299                 }
1300         }
1301         rcu_read_unlock();
1302
1303 no_promotions:
1304         if (!(ok & BRD_OK))
1305                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1306                           prim, 0);
1307         if (subnet && ifa->ifa_prefixlen < 31) {
1308                 if (!(ok & BRD1_OK))
1309                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1310                                   prim, 0);
1311                 if (!(ok & BRD0_OK))
1312                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1313                                   prim, 0);
1314         }
1315         if (!(ok & LOCAL_OK)) {
1316                 unsigned int addr_type;
1317
1318                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1319
1320                 /* Check, that this local address finally disappeared. */
1321                 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1322                                                      ifa->ifa_local);
1323                 if (gone && addr_type != RTN_LOCAL) {
1324                         /* And the last, but not the least thing.
1325                          * We must flush stray FIB entries.
1326                          *
1327                          * First of all, we scan fib_info list searching
1328                          * for stray nexthop entries, then ignite fib_flush.
1329                          */
1330                         if (fib_sync_down_addr(dev, ifa->ifa_local))
1331                                 fib_flush(dev_net(dev));
1332                 }
1333         }
1334 #undef LOCAL_OK
1335 #undef BRD_OK
1336 #undef BRD0_OK
1337 #undef BRD1_OK
1338 }
1339
1340 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1341 {
1342
1343         struct fib_result       res;
1344         struct flowi4           fl4 = {
1345                 .flowi4_mark = frn->fl_mark,
1346                 .daddr = frn->fl_addr,
1347                 .flowi4_tos = frn->fl_tos,
1348                 .flowi4_scope = frn->fl_scope,
1349         };
1350         struct fib_table *tb;
1351
1352         rcu_read_lock();
1353
1354         tb = fib_get_table(net, frn->tb_id_in);
1355
1356         frn->err = -ENOENT;
1357         if (tb) {
1358                 local_bh_disable();
1359
1360                 frn->tb_id = tb->tb_id;
1361                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1362
1363                 if (!frn->err) {
1364                         frn->prefixlen = res.prefixlen;
1365                         frn->nh_sel = res.nh_sel;
1366                         frn->type = res.type;
1367                         frn->scope = res.scope;
1368                 }
1369                 local_bh_enable();
1370         }
1371
1372         rcu_read_unlock();
1373 }
1374
1375 static void nl_fib_input(struct sk_buff *skb)
1376 {
1377         struct net *net;
1378         struct fib_result_nl *frn;
1379         struct nlmsghdr *nlh;
1380         u32 portid;
1381
1382         net = sock_net(skb->sk);
1383         nlh = nlmsg_hdr(skb);
1384         if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1385             skb->len < nlh->nlmsg_len ||
1386             nlmsg_len(nlh) < sizeof(*frn))
1387                 return;
1388
1389         skb = netlink_skb_clone(skb, GFP_KERNEL);
1390         if (!skb)
1391                 return;
1392         nlh = nlmsg_hdr(skb);
1393
1394         frn = (struct fib_result_nl *) nlmsg_data(nlh);
1395         nl_fib_lookup(net, frn);
1396
1397         portid = NETLINK_CB(skb).portid;      /* netlink portid */
1398         NETLINK_CB(skb).portid = 0;        /* from kernel */
1399         NETLINK_CB(skb).dst_group = 0;  /* unicast */
1400         netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1401 }
1402
1403 static int __net_init nl_fib_lookup_init(struct net *net)
1404 {
1405         struct sock *sk;
1406         struct netlink_kernel_cfg cfg = {
1407                 .input  = nl_fib_input,
1408         };
1409
1410         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1411         if (!sk)
1412                 return -EAFNOSUPPORT;
1413         net->ipv4.fibnl = sk;
1414         return 0;
1415 }
1416
1417 static void nl_fib_lookup_exit(struct net *net)
1418 {
1419         netlink_kernel_release(net->ipv4.fibnl);
1420         net->ipv4.fibnl = NULL;
1421 }
1422
1423 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1424                            bool force)
1425 {
1426         if (fib_sync_down_dev(dev, event, force))
1427                 fib_flush(dev_net(dev));
1428         else
1429                 rt_cache_flush(dev_net(dev));
1430         arp_ifdown(dev);
1431 }
1432
1433 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1434 {
1435         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1436         struct net_device *dev = ifa->ifa_dev->dev;
1437         struct net *net = dev_net(dev);
1438
1439         switch (event) {
1440         case NETDEV_UP:
1441                 fib_add_ifaddr(ifa);
1442 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1443                 fib_sync_up(dev, RTNH_F_DEAD);
1444 #endif
1445                 atomic_inc(&net->ipv4.dev_addr_genid);
1446                 rt_cache_flush(dev_net(dev));
1447                 break;
1448         case NETDEV_DOWN:
1449                 fib_del_ifaddr(ifa, NULL);
1450                 atomic_inc(&net->ipv4.dev_addr_genid);
1451                 if (!ifa->ifa_dev->ifa_list) {
1452                         /* Last address was deleted from this interface.
1453                          * Disable IP.
1454                          */
1455                         fib_disable_ip(dev, event, true);
1456                 } else {
1457                         rt_cache_flush(dev_net(dev));
1458                 }
1459                 break;
1460         }
1461         return NOTIFY_DONE;
1462 }
1463
1464 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1465 {
1466         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1467         struct netdev_notifier_changeupper_info *upper_info = ptr;
1468         struct netdev_notifier_info_ext *info_ext = ptr;
1469         struct in_device *in_dev;
1470         struct net *net = dev_net(dev);
1471         struct in_ifaddr *ifa;
1472         unsigned int flags;
1473
1474         if (event == NETDEV_UNREGISTER) {
1475                 fib_disable_ip(dev, event, true);
1476                 rt_flush_dev(dev);
1477                 return NOTIFY_DONE;
1478         }
1479
1480         in_dev = __in_dev_get_rtnl(dev);
1481         if (!in_dev)
1482                 return NOTIFY_DONE;
1483
1484         switch (event) {
1485         case NETDEV_UP:
1486                 in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1487                         fib_add_ifaddr(ifa);
1488                 }
1489 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1490                 fib_sync_up(dev, RTNH_F_DEAD);
1491 #endif
1492                 atomic_inc(&net->ipv4.dev_addr_genid);
1493                 rt_cache_flush(net);
1494                 break;
1495         case NETDEV_DOWN:
1496                 fib_disable_ip(dev, event, false);
1497                 break;
1498         case NETDEV_CHANGE:
1499                 flags = dev_get_flags(dev);
1500                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1501                         fib_sync_up(dev, RTNH_F_LINKDOWN);
1502                 else
1503                         fib_sync_down_dev(dev, event, false);
1504                 rt_cache_flush(net);
1505                 break;
1506         case NETDEV_CHANGEMTU:
1507                 fib_sync_mtu(dev, info_ext->ext.mtu);
1508                 rt_cache_flush(net);
1509                 break;
1510         case NETDEV_CHANGEUPPER:
1511                 upper_info = ptr;
1512                 /* flush all routes if dev is linked to or unlinked from
1513                  * an L3 master device (e.g., VRF)
1514                  */
1515                 if (upper_info->upper_dev &&
1516                     netif_is_l3_master(upper_info->upper_dev))
1517                         fib_disable_ip(dev, NETDEV_DOWN, true);
1518                 break;
1519         }
1520         return NOTIFY_DONE;
1521 }
1522
1523 static struct notifier_block fib_inetaddr_notifier = {
1524         .notifier_call = fib_inetaddr_event,
1525 };
1526
1527 static struct notifier_block fib_netdev_notifier = {
1528         .notifier_call = fib_netdev_event,
1529 };
1530
1531 static int __net_init ip_fib_net_init(struct net *net)
1532 {
1533         int err;
1534         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1535
1536         err = fib4_notifier_init(net);
1537         if (err)
1538                 return err;
1539
1540         /* Avoid false sharing : Use at least a full cache line */
1541         size = max_t(size_t, size, L1_CACHE_BYTES);
1542
1543         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1544         if (!net->ipv4.fib_table_hash) {
1545                 err = -ENOMEM;
1546                 goto err_table_hash_alloc;
1547         }
1548
1549         err = fib4_rules_init(net);
1550         if (err < 0)
1551                 goto err_rules_init;
1552         return 0;
1553
1554 err_rules_init:
1555         kfree(net->ipv4.fib_table_hash);
1556 err_table_hash_alloc:
1557         fib4_notifier_exit(net);
1558         return err;
1559 }
1560
1561 static void ip_fib_net_exit(struct net *net)
1562 {
1563         int i;
1564
1565         rtnl_lock();
1566 #ifdef CONFIG_IP_MULTIPLE_TABLES
1567         RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1568         RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1569 #endif
1570         /* Destroy the tables in reverse order to guarantee that the
1571          * local table, ID 255, is destroyed before the main table, ID
1572          * 254. This is necessary as the local table may contain
1573          * references to data contained in the main table.
1574          */
1575         for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1576                 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1577                 struct hlist_node *tmp;
1578                 struct fib_table *tb;
1579
1580                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1581                         hlist_del(&tb->tb_hlist);
1582                         fib_table_flush(net, tb, true);
1583                         fib_free_table(tb);
1584                 }
1585         }
1586
1587 #ifdef CONFIG_IP_MULTIPLE_TABLES
1588         fib4_rules_exit(net);
1589 #endif
1590         rtnl_unlock();
1591         kfree(net->ipv4.fib_table_hash);
1592         fib4_notifier_exit(net);
1593 }
1594
1595 static int __net_init fib_net_init(struct net *net)
1596 {
1597         int error;
1598
1599 #ifdef CONFIG_IP_ROUTE_CLASSID
1600         atomic_set(&net->ipv4.fib_num_tclassid_users, 0);
1601 #endif
1602         error = ip_fib_net_init(net);
1603         if (error < 0)
1604                 goto out;
1605         error = nl_fib_lookup_init(net);
1606         if (error < 0)
1607                 goto out_nlfl;
1608         error = fib_proc_init(net);
1609         if (error < 0)
1610                 goto out_proc;
1611 out:
1612         return error;
1613
1614 out_proc:
1615         nl_fib_lookup_exit(net);
1616 out_nlfl:
1617         ip_fib_net_exit(net);
1618         goto out;
1619 }
1620
1621 static void __net_exit fib_net_exit(struct net *net)
1622 {
1623         fib_proc_exit(net);
1624         nl_fib_lookup_exit(net);
1625         ip_fib_net_exit(net);
1626 }
1627
1628 static struct pernet_operations fib_net_ops = {
1629         .init = fib_net_init,
1630         .exit = fib_net_exit,
1631 };
1632
1633 void __init ip_fib_init(void)
1634 {
1635         fib_trie_init();
1636
1637         register_pernet_subsys(&fib_net_ops);
1638
1639         register_netdevice_notifier(&fib_netdev_notifier);
1640         register_inetaddr_notifier(&fib_inetaddr_notifier);
1641
1642         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1643         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1644         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1645 }