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, <[email protected]>
  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/inet_dscp.h>
  36 #include <net/ip.h>
  37 #include <net/protocol.h>
  38 #include <net/route.h>
  39 #include <net/tcp.h>
  40 #include <net/sock.h>
  41 #include <net/arp.h>
  42 #include <net/ip_fib.h>
  43 #include <net/nexthop.h>
  44 #include <net/rtnetlink.h>
  45 #include <net/xfrm.h>
  46 #include <net/l3mdev.h>
  47 #include <net/lwtunnel.h>
  48 #include <trace/events/fib.h>
  49
  50 #ifndef CONFIG_IP_MULTIPLE_TABLES
  51
  52 static int __net_init fib4_rules_init(struct net *net)
  53 {
  54         struct fib_table *local_table, *main_table;
  55
  56         main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
  57         if (!main_table)
  58                 return -ENOMEM;
  59
  60         local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
  61         if (!local_table)
  62                 goto fail;
  63
  64         hlist_add_head_rcu(&local_table->tb_hlist,
  65                                 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
  66         hlist_add_head_rcu(&main_table->tb_hlist,
  67                                 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
  68         return 0;
  69
  70 fail:
  71         fib_free_table(main_table);
  72         return -ENOMEM;
  73 }
  74 #else
  75
  76 struct fib_table *fib_new_table(struct net *net, u32 id)
  77 {
  78         struct fib_table *tb, *alias = NULL;
  79         unsigned int h;
  80
  81         if (id == 0)
  82                 id = RT_TABLE_MAIN;
  83         tb = fib_get_table(net, id);
  84         if (tb)
  85                 return tb;
  86
  87         if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
  88                 alias = fib_new_table(net, RT_TABLE_MAIN);
  89
  90         tb = fib_trie_table(id, alias);
  91         if (!tb)
  92                 return NULL;
  93
  94         switch (id) {
  95         case RT_TABLE_MAIN:
  96                 rcu_assign_pointer(net->ipv4.fib_main, tb);
  97                 break;
  98         case RT_TABLE_DEFAULT:
  99                 rcu_assign_pointer(net->ipv4.fib_default, tb);
 100                 break;
 101         default:
 102                 break;
 103         }
 104
 105         h = id & (FIB_TABLE_HASHSZ - 1);
 106         hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
 107         return tb;
 108 }
 109 EXPORT_SYMBOL_GPL(fib_new_table);
 110
 111 /* caller must hold either rtnl or rcu read lock */
 112 struct fib_table *fib_get_table(struct net *net, u32 id)
 113 {
 114         struct fib_table *tb;
 115         struct hlist_head *head;
 116         unsigned int h;
 117
 118         if (id == 0)
 119                 id = RT_TABLE_MAIN;
 120         h = id & (FIB_TABLE_HASHSZ - 1);
 121
 122         head = &net->ipv4.fib_table_hash[h];
 123         hlist_for_each_entry_rcu(tb, head, tb_hlist,
 124                                  lockdep_rtnl_is_held()) {
 125                 if (tb->tb_id == id)
 126                         return tb;
 127         }
 128         return NULL;
 129 }
 130 #endif /* CONFIG_IP_MULTIPLE_TABLES */
 131
 132 static void fib_replace_table(struct net *net, struct fib_table *old,
 133                               struct fib_table *new)
 134 {
 135 #ifdef CONFIG_IP_MULTIPLE_TABLES
 136         switch (new->tb_id) {
 137         case RT_TABLE_MAIN:
 138                 rcu_assign_pointer(net->ipv4.fib_main, new);
 139                 break;
 140         case RT_TABLE_DEFAULT:
 141                 rcu_assign_pointer(net->ipv4.fib_default, new);
 142                 break;
 143         default:
 144                 break;
 145         }
 146
 147 #endif
 148         /* replace the old table in the hlist */
 149         hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
 150 }
 151
 152 int fib_unmerge(struct net *net)
 153 {
 154         struct fib_table *old, *new, *main_table;
 155
 156         /* attempt to fetch local table if it has been allocated */
 157         old = fib_get_table(net, RT_TABLE_LOCAL);
 158         if (!old)
 159                 return 0;
 160
 161         new = fib_trie_unmerge(old);
 162         if (!new)
 163                 return -ENOMEM;
 164
 165         /* table is already unmerged */
 166         if (new == old)
 167                 return 0;
 168
 169         /* replace merged table with clean table */
 170         fib_replace_table(net, old, new);
 171         fib_free_table(old);
 172
 173         /* attempt to fetch main table if it has been allocated */
 174         main_table = fib_get_table(net, RT_TABLE_MAIN);
 175         if (!main_table)
 176                 return 0;
 177
 178         /* flush local entries from main table */
 179         fib_table_flush_external(main_table);
 180
 181         return 0;
 182 }
 183
 184 void fib_flush(struct net *net)
 185 {
 186         int flushed = 0;
 187         unsigned int h;
 188
 189         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
 190                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
 191                 struct hlist_node *tmp;
 192                 struct fib_table *tb;
 193
 194                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
 195                         flushed += fib_table_flush(net, tb, false);
 196         }
 197
 198         if (flushed)
 199                 rt_cache_flush(net);
 200 }
 201
 202 /*
 203  * Find address type as if only "dev" was present in the system. If
 204  * on_dev is NULL then all interfaces are taken into consideration.
 205  */
 206 static inline unsigned int __inet_dev_addr_type(struct net *net,
 207                                                 const struct net_device *dev,
 208                                                 __be32 addr, u32 tb_id)
 209 {
 210         struct flowi4           fl4 = { .daddr = addr };
 211         struct fib_result       res;
 212         unsigned int ret = RTN_BROADCAST;
 213         struct fib_table *table;
 214
 215         if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
 216                 return RTN_BROADCAST;
 217         if (ipv4_is_multicast(addr))
 218                 return RTN_MULTICAST;
 219
 220         rcu_read_lock();
 221
 222         table = fib_get_table(net, tb_id);
 223         if (table) {
 224                 ret = RTN_UNICAST;
 225                 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
 226                         struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0);
 227
 228                         if (!dev || dev == nhc->nhc_dev)
 229                                 ret = res.type;
 230                 }
 231         }
 232
 233         rcu_read_unlock();
 234         return ret;
 235 }
 236
 237 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
 238 {
 239         return __inet_dev_addr_type(net, NULL, addr, tb_id);
 240 }
 241 EXPORT_SYMBOL(inet_addr_type_table);
 242
 243 unsigned int inet_addr_type(struct net *net, __be32 addr)
 244 {
 245         return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
 246 }
 247 EXPORT_SYMBOL(inet_addr_type);
 248
 249 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
 250                                 __be32 addr)
 251 {
 252         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
 253
 254         return __inet_dev_addr_type(net, dev, addr, rt_table);
 255 }
 256 EXPORT_SYMBOL(inet_dev_addr_type);
 257
 258 /* inet_addr_type with dev == NULL but using the table from a dev
 259  * if one is associated
 260  */
 261 unsigned int inet_addr_type_dev_table(struct net *net,
 262                                       const struct net_device *dev,
 263                                       __be32 addr)
 264 {
 265         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
 266
 267         return __inet_dev_addr_type(net, NULL, addr, rt_table);
 268 }
 269 EXPORT_SYMBOL(inet_addr_type_dev_table);
 270
 271 __be32 fib_compute_spec_dst(struct sk_buff *skb)
 272 {
 273         struct net_device *dev = skb->dev;
 274         struct in_device *in_dev;
 275         struct fib_result res;
 276         struct rtable *rt;
 277         struct net *net;
 278         int scope;
 279
 280         rt = skb_rtable(skb);
 281         if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
 282             RTCF_LOCAL)
 283                 return ip_hdr(skb)->daddr;
 284
 285         in_dev = __in_dev_get_rcu(dev);
 286
 287         net = dev_net(dev);
 288
 289         scope = RT_SCOPE_UNIVERSE;
 290         if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
 291                 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
 292                 struct flowi4 fl4 = {
 293                         .flowi4_iif = LOOPBACK_IFINDEX,
 294                         .flowi4_l3mdev = l3mdev_master_ifindex_rcu(dev),
 295                         .daddr = ip_hdr(skb)->saddr,
 296                         .flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(ip_hdr(skb))),
 297                         .flowi4_scope = scope,
 298                         .flowi4_mark = vmark ? skb->mark : 0,
 299                 };
 300                 if (!fib_lookup(net, &fl4, &res, 0))
 301                         return fib_result_prefsrc(net, &res);
 302         } else {
 303                 scope = RT_SCOPE_LINK;
 304         }
 305
 306         return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
 307 }
 308
 309 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
 310 {
 311         bool dev_match = false;
 312 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 313         if (unlikely(fi->nh)) {
 314                 dev_match = nexthop_uses_dev(fi->nh, dev);
 315         } else {
 316                 int ret;
 317
 318                 for (ret = 0; ret < fib_info_num_path(fi); ret++) {
 319                         const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
 320
 321                         if (nhc_l3mdev_matches_dev(nhc, dev)) {
 322                                 dev_match = true;
 323                                 break;
 324                         }
 325                 }
 326         }
 327 #else
 328         if (fib_info_nhc(fi, 0)->nhc_dev == dev)
 329                 dev_match = true;
 330 #endif
 331
 332         return dev_match;
 333 }
 334 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
 335
 336 /* Given (packet source, input interface) and optional (dst, oif, tos):
 337  * - (main) check, that source is valid i.e. not broadcast or our local
 338  *   address.
 339  * - figure out what "logical" interface this packet arrived
 340  *   and calculate "specific destination" address.
 341  * - check, that packet arrived from expected physical interface.
 342  * called with rcu_read_lock()
 343  */
 344 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 345                                  dscp_t dscp, int oif, struct net_device *dev,
 346                                  int rpf, struct in_device *idev, u32 *itag)
 347 {
 348         struct net *net = dev_net(dev);
 349         enum skb_drop_reason reason;
 350         struct flow_keys flkeys;
 351         int ret, no_addr;
 352         struct fib_result res;
 353         struct flowi4 fl4;
 354         bool dev_match;
 355
 356         fl4.flowi4_oif = 0;
 357         fl4.flowi4_l3mdev = l3mdev_master_ifindex_rcu(dev);
 358         fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
 359         fl4.daddr = src;
 360         fl4.saddr = dst;
 361         fl4.flowi4_tos = inet_dscp_to_dsfield(dscp);
 362         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 363         fl4.flowi4_tun_key.tun_id = 0;
 364         fl4.flowi4_flags = 0;
 365         fl4.flowi4_uid = sock_net_uid(net, NULL);
 366         fl4.flowi4_multipath_hash = 0;
 367
 368         no_addr = idev->ifa_list == NULL;
 369
 370         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
 371         if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
 372                 fl4.flowi4_proto = 0;
 373                 fl4.fl4_sport = 0;
 374                 fl4.fl4_dport = 0;
 375         } else {
 376                 swap(fl4.fl4_sport, fl4.fl4_dport);
 377         }
 378
 379         if (fib_lookup(net, &fl4, &res, 0))
 380                 goto last_resort;
 381         if (res.type != RTN_UNICAST) {
 382                 if (res.type != RTN_LOCAL) {
 383                         reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
 384                         goto e_inval;
 385                 } else if (!IN_DEV_ACCEPT_LOCAL(idev)) {
 386                         reason = SKB_DROP_REASON_IP_LOCAL_SOURCE;
 387                         goto e_inval;
 388                 }
 389         }
 390         fib_combine_itag(itag, &res);
 391
 392         dev_match = fib_info_nh_uses_dev(res.fi, dev);
 393         /* This is not common, loopback packets retain skb_dst so normally they
 394          * would not even hit this slow path.
 395          */
 396         dev_match = dev_match || (res.type == RTN_LOCAL &&
 397                                   dev == net->loopback_dev);
 398         if (dev_match) {
 399                 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
 400                 return ret;
 401         }
 402         if (no_addr)
 403                 goto last_resort;
 404         if (rpf == 1)
 405                 goto e_rpf;
 406         fl4.flowi4_oif = dev->ifindex;
 407
 408         ret = 0;
 409         if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
 410                 if (res.type == RTN_UNICAST)
 411                         ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
 412         }
 413         return ret;
 414
 415 last_resort:
 416         if (rpf)
 417                 goto e_rpf;
 418         *itag = 0;
 419         return 0;
 420
 421 e_inval:
 422         return -reason;
 423 e_rpf:
 424         return -SKB_DROP_REASON_IP_RPFILTER;
 425 }
 426
 427 /* Ignore rp_filter for packets protected by IPsec. */
 428 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 429                         dscp_t dscp, int oif, struct net_device *dev,
 430                         struct in_device *idev, u32 *itag)
 431 {
 432         int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
 433         struct net *net = dev_net(dev);
 434
 435         if (!r && !fib_num_tclassid_users(net) &&
 436             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
 437                 if (IN_DEV_ACCEPT_LOCAL(idev))
 438                         goto ok;
 439                 /* with custom local routes in place, checking local addresses
 440                  * only will be too optimistic, with custom rules, checking
 441                  * local addresses only can be too strict, e.g. due to vrf
 442                  */
 443                 if (net->ipv4.fib_has_custom_local_routes ||
 444                     fib4_has_custom_rules(net))
 445                         goto full_check;
 446                 /* Within the same container, it is regarded as a martian source,
 447                  * and the same host but different containers are not.
 448                  */
 449                 if (inet_lookup_ifaddr_rcu(net, src))
 450                         return -SKB_DROP_REASON_IP_LOCAL_SOURCE;
 451
 452 ok:
 453                 *itag = 0;
 454                 return 0;
 455         }
 456
 457 full_check:
 458         return __fib_validate_source(skb, src, dst, dscp, oif, dev, r, idev,
 459                                      itag);
 460 }
 461
 462 static inline __be32 sk_extract_addr(struct sockaddr *addr)
 463 {
 464         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 465 }
 466
 467 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
 468 {
 469         struct nlattr *nla;
 470
 471         nla = (struct nlattr *) ((char *) mx + len);
 472         nla->nla_type = type;
 473         nla->nla_len = nla_attr_size(4);
 474         *(u32 *) nla_data(nla) = value;
 475
 476         return len + nla_total_size(4);
 477 }
 478
 479 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
 480                                  struct fib_config *cfg)
 481 {
 482         __be32 addr;
 483         int plen;
 484
 485         memset(cfg, 0, sizeof(*cfg));
 486         cfg->fc_nlinfo.nl_net = net;
 487
 488         if (rt->rt_dst.sa_family != AF_INET)
 489                 return -EAFNOSUPPORT;
 490
 491         /*
 492          * Check mask for validity:
 493          * a) it must be contiguous.
 494          * b) destination must have all host bits clear.
 495          * c) if application forgot to set correct family (AF_INET),
 496          *    reject request unless it is absolutely clear i.e.
 497          *    both family and mask are zero.
 498          */
 499         plen = 32;
 500         addr = sk_extract_addr(&rt->rt_dst);
 501         if (!(rt->rt_flags & RTF_HOST)) {
 502                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
 503
 504                 if (rt->rt_genmask.sa_family != AF_INET) {
 505                         if (mask || rt->rt_genmask.sa_family)
 506                                 return -EAFNOSUPPORT;
 507                 }
 508
 509                 if (bad_mask(mask, addr))
 510                         return -EINVAL;
 511
 512                 plen = inet_mask_len(mask);
 513         }
 514
 515         cfg->fc_dst_len = plen;
 516         cfg->fc_dst = addr;
 517
 518         if (cmd != SIOCDELRT) {
 519                 cfg->fc_nlflags = NLM_F_CREATE;
 520                 cfg->fc_protocol = RTPROT_BOOT;
 521         }
 522
 523         if (rt->rt_metric)
 524                 cfg->fc_priority = rt->rt_metric - 1;
 525
 526         if (rt->rt_flags & RTF_REJECT) {
 527                 cfg->fc_scope = RT_SCOPE_HOST;
 528                 cfg->fc_type = RTN_UNREACHABLE;
 529                 return 0;
 530         }
 531
 532         cfg->fc_scope = RT_SCOPE_NOWHERE;
 533         cfg->fc_type = RTN_UNICAST;
 534
 535         if (rt->rt_dev) {
 536                 char *colon;
 537                 struct net_device *dev;
 538                 char devname[IFNAMSIZ];
 539
 540                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
 541                         return -EFAULT;
 542
 543                 devname[IFNAMSIZ-1] = 0;
 544                 colon = strchr(devname, ':');
 545                 if (colon)
 546                         *colon = 0;
 547                 dev = __dev_get_by_name(net, devname);
 548                 if (!dev)
 549                         return -ENODEV;
 550                 cfg->fc_oif = dev->ifindex;
 551                 cfg->fc_table = l3mdev_fib_table(dev);
 552                 if (colon) {
 553                         const struct in_ifaddr *ifa;
 554                         struct in_device *in_dev;
 555
 556                         in_dev = __in_dev_get_rtnl(dev);
 557                         if (!in_dev)
 558                                 return -ENODEV;
 559
 560                         *colon = ':';
 561
 562                         rcu_read_lock();
 563                         in_dev_for_each_ifa_rcu(ifa, in_dev) {
 564                                 if (strcmp(ifa->ifa_label, devname) == 0)
 565                                         break;
 566                         }
 567                         rcu_read_unlock();
 568
 569                         if (!ifa)
 570                                 return -ENODEV;
 571                         cfg->fc_prefsrc = ifa->ifa_local;
 572                 }
 573         }
 574
 575         addr = sk_extract_addr(&rt->rt_gateway);
 576         if (rt->rt_gateway.sa_family == AF_INET && addr) {
 577                 unsigned int addr_type;
 578
 579                 cfg->fc_gw4 = addr;
 580                 cfg->fc_gw_family = AF_INET;
 581                 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
 582                 if (rt->rt_flags & RTF_GATEWAY &&
 583                     addr_type == RTN_UNICAST)
 584                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
 585         }
 586
 587         if (!cfg->fc_table)
 588                 cfg->fc_table = RT_TABLE_MAIN;
 589
 590         if (cmd == SIOCDELRT)
 591                 return 0;
 592
 593         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
 594                 return -EINVAL;
 595
 596         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 597                 cfg->fc_scope = RT_SCOPE_LINK;
 598
 599         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 600                 struct nlattr *mx;
 601                 int len = 0;
 602
 603                 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
 604                 if (!mx)
 605                         return -ENOMEM;
 606
 607                 if (rt->rt_flags & RTF_MTU)
 608                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 609
 610                 if (rt->rt_flags & RTF_WINDOW)
 611                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 612
 613                 if (rt->rt_flags & RTF_IRTT)
 614                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 615
 616                 cfg->fc_mx = mx;
 617                 cfg->fc_mx_len = len;
 618         }
 619
 620         return 0;
 621 }
 622
 623 /*
 624  * Handle IP routing ioctl calls.
 625  * These are used to manipulate the routing tables
 626  */
 627 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
 628 {
 629         struct fib_config cfg;
 630         int err;
 631
 632         switch (cmd) {
 633         case SIOCADDRT:         /* Add a route */
 634         case SIOCDELRT:         /* Delete a route */
 635                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 636                         return -EPERM;
 637
 638                 rtnl_lock();
 639                 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
 640                 if (err == 0) {
 641                         struct fib_table *tb;
 642
 643                         if (cmd == SIOCDELRT) {
 644                                 tb = fib_get_table(net, cfg.fc_table);
 645                                 if (tb)
 646                                         err = fib_table_delete(net, tb, &cfg,
 647                                                                NULL);
 648                                 else
 649                                         err = -ESRCH;
 650                         } else {
 651                                 tb = fib_new_table(net, cfg.fc_table);
 652                                 if (tb)
 653                                         err = fib_table_insert(net, tb,
 654                                                                &cfg, NULL);
 655                                 else
 656                                         err = -ENOBUFS;
 657                         }
 658
 659                         /* allocated by rtentry_to_fib_config() */
 660                         kfree(cfg.fc_mx);
 661                 }
 662                 rtnl_unlock();
 663                 return err;
 664         }
 665         return -EINVAL;
 666 }
 667
 668 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 669         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
 670         [RTA_DST]               = { .type = NLA_U32 },
 671         [RTA_SRC]               = { .type = NLA_U32 },
 672         [RTA_IIF]               = { .type = NLA_U32 },
 673         [RTA_OIF]               = { .type = NLA_U32 },
 674         [RTA_GATEWAY]           = { .type = NLA_U32 },
 675         [RTA_PRIORITY]          = { .type = NLA_U32 },
 676         [RTA_PREFSRC]           = { .type = NLA_U32 },
 677         [RTA_METRICS]           = { .type = NLA_NESTED },
 678         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 679         [RTA_FLOW]              = { .type = NLA_U32 },
 680         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
 681         [RTA_ENCAP]             = { .type = NLA_NESTED },
 682         [RTA_UID]               = { .type = NLA_U32 },
 683         [RTA_MARK]              = { .type = NLA_U32 },
 684         [RTA_TABLE]             = { .type = NLA_U32 },
 685         [RTA_IP_PROTO]          = { .type = NLA_U8 },
 686         [RTA_SPORT]             = { .type = NLA_U16 },
 687         [RTA_DPORT]             = { .type = NLA_U16 },
 688         [RTA_NH_ID]             = { .type = NLA_U32 },
 689 };
 690
 691 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
 692                     struct netlink_ext_ack *extack)
 693 {
 694         struct rtvia *via;
 695         int alen;
 696
 697         if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
 698                 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
 699                 return -EINVAL;
 700         }
 701
 702         via = nla_data(nla);
 703         alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
 704
 705         switch (via->rtvia_family) {
 706         case AF_INET:
 707                 if (alen != sizeof(__be32)) {
 708                         NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
 709                         return -EINVAL;
 710                 }
 711                 cfg->fc_gw_family = AF_INET;
 712                 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
 713                 break;
 714         case AF_INET6:
 715 #if IS_ENABLED(CONFIG_IPV6)
 716                 if (alen != sizeof(struct in6_addr)) {
 717                         NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
 718                         return -EINVAL;
 719                 }
 720                 cfg->fc_gw_family = AF_INET6;
 721                 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
 722 #else
 723                 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
 724                 return -EINVAL;
 725 #endif
 726                 break;
 727         default:
 728                 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
 729                 return -EINVAL;
 730         }
 731
 732         return 0;
 733 }
 734
 735 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 736                              struct nlmsghdr *nlh, struct fib_config *cfg,
 737                              struct netlink_ext_ack *extack)
 738 {
 739         bool has_gw = false, has_via = false;
 740         struct nlattr *attr;
 741         int err, remaining;
 742         struct rtmsg *rtm;
 743
 744         err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
 745                                         rtm_ipv4_policy, extack);
 746         if (err < 0)
 747                 goto errout;
 748
 749         memset(cfg, 0, sizeof(*cfg));
 750
 751         rtm = nlmsg_data(nlh);
 752
 753         if (!inet_validate_dscp(rtm->rtm_tos)) {
 754                 NL_SET_ERR_MSG(extack,
 755                                "Invalid dsfield (tos): ECN bits must be 0");
 756                 err = -EINVAL;
 757                 goto errout;
 758         }
 759         cfg->fc_dscp = inet_dsfield_to_dscp(rtm->rtm_tos);
 760
 761         cfg->fc_dst_len = rtm->rtm_dst_len;
 762         cfg->fc_table = rtm->rtm_table;
 763         cfg->fc_protocol = rtm->rtm_protocol;
 764         cfg->fc_scope = rtm->rtm_scope;
 765         cfg->fc_type = rtm->rtm_type;
 766         cfg->fc_flags = rtm->rtm_flags;
 767         cfg->fc_nlflags = nlh->nlmsg_flags;
 768
 769         cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
 770         cfg->fc_nlinfo.nlh = nlh;
 771         cfg->fc_nlinfo.nl_net = net;
 772
 773         if (cfg->fc_type > RTN_MAX) {
 774                 NL_SET_ERR_MSG(extack, "Invalid route type");
 775                 err = -EINVAL;
 776                 goto errout;
 777         }
 778
 779         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 780                 switch (nla_type(attr)) {
 781                 case RTA_DST:
 782                         cfg->fc_dst = nla_get_be32(attr);
 783                         break;
 784                 case RTA_OIF:
 785                         cfg->fc_oif = nla_get_u32(attr);
 786                         break;
 787                 case RTA_GATEWAY:
 788                         has_gw = true;
 789                         cfg->fc_gw4 = nla_get_be32(attr);
 790                         if (cfg->fc_gw4)
 791                                 cfg->fc_gw_family = AF_INET;
 792                         break;
 793                 case RTA_VIA:
 794                         has_via = true;
 795                         err = fib_gw_from_via(cfg, attr, extack);
 796                         if (err)
 797                                 goto errout;
 798                         break;
 799                 case RTA_PRIORITY:
 800                         cfg->fc_priority = nla_get_u32(attr);
 801                         break;
 802                 case RTA_PREFSRC:
 803                         cfg->fc_prefsrc = nla_get_be32(attr);
 804                         break;
 805                 case RTA_METRICS:
 806                         cfg->fc_mx = nla_data(attr);
 807                         cfg->fc_mx_len = nla_len(attr);
 808                         break;
 809                 case RTA_MULTIPATH:
 810                         err = lwtunnel_valid_encap_type_attr(nla_data(attr),
 811                                                              nla_len(attr),
 812                                                              extack);
 813                         if (err < 0)
 814                                 goto errout;
 815                         cfg->fc_mp = nla_data(attr);
 816                         cfg->fc_mp_len = nla_len(attr);
 817                         break;
 818                 case RTA_FLOW:
 819                         cfg->fc_flow = nla_get_u32(attr);
 820                         break;
 821                 case RTA_TABLE:
 822                         cfg->fc_table = nla_get_u32(attr);
 823                         break;
 824                 case RTA_ENCAP:
 825                         cfg->fc_encap = attr;
 826                         break;
 827                 case RTA_ENCAP_TYPE:
 828                         cfg->fc_encap_type = nla_get_u16(attr);
 829                         err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
 830                                                         extack);
 831                         if (err < 0)
 832                                 goto errout;
 833                         break;
 834                 case RTA_NH_ID:
 835                         cfg->fc_nh_id = nla_get_u32(attr);
 836                         break;
 837                 }
 838         }
 839
 840         if (cfg->fc_nh_id) {
 841                 if (cfg->fc_oif || cfg->fc_gw_family ||
 842                     cfg->fc_encap || cfg->fc_mp) {
 843                         NL_SET_ERR_MSG(extack,
 844                                        "Nexthop specification and nexthop id are mutually exclusive");
 845                         return -EINVAL;
 846                 }
 847         }
 848
 849         if (has_gw && has_via) {
 850                 NL_SET_ERR_MSG(extack,
 851                                "Nexthop configuration can not contain both GATEWAY and VIA");
 852                 return -EINVAL;
 853         }
 854
 855         if (!cfg->fc_table)
 856                 cfg->fc_table = RT_TABLE_MAIN;
 857
 858         return 0;
 859 errout:
 860         return err;
 861 }
 862
 863 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 864                              struct netlink_ext_ack *extack)
 865 {
 866         struct net *net = sock_net(skb->sk);
 867         struct fib_config cfg;
 868         struct fib_table *tb;
 869         int err;
 870
 871         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 872         if (err < 0)
 873                 goto errout;
 874
 875         if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
 876                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
 877                 err = -EINVAL;
 878                 goto errout;
 879         }
 880
 881         tb = fib_get_table(net, cfg.fc_table);
 882         if (!tb) {
 883                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
 884                 err = -ESRCH;
 885                 goto errout;
 886         }
 887
 888         err = fib_table_delete(net, tb, &cfg, extack);
 889 errout:
 890         return err;
 891 }
 892
 893 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 894                              struct netlink_ext_ack *extack)
 895 {
 896         struct net *net = sock_net(skb->sk);
 897         struct fib_config cfg;
 898         struct fib_table *tb;
 899         int err;
 900
 901         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 902         if (err < 0)
 903                 goto errout;
 904
 905         tb = fib_new_table(net, cfg.fc_table);
 906         if (!tb) {
 907                 err = -ENOBUFS;
 908                 goto errout;
 909         }
 910
 911         err = fib_table_insert(net, tb, &cfg, extack);
 912         if (!err && cfg.fc_type == RTN_LOCAL)
 913                 net->ipv4.fib_has_custom_local_routes = true;
 914 errout:
 915         return err;
 916 }
 917
 918 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
 919                           struct fib_dump_filter *filter,
 920                           struct netlink_callback *cb)
 921 {
 922         struct netlink_ext_ack *extack = cb->extack;
 923         struct nlattr *tb[RTA_MAX + 1];
 924         struct rtmsg *rtm;
 925         int err, i;
 926
 927         if (filter->rtnl_held)
 928                 ASSERT_RTNL();
 929
 930         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
 931                 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
 932                 return -EINVAL;
 933         }
 934
 935         rtm = nlmsg_data(nlh);
 936         if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
 937             rtm->rtm_scope) {
 938                 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
 939                 return -EINVAL;
 940         }
 941
 942         if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
 943                 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
 944                 return -EINVAL;
 945         }
 946         if (rtm->rtm_flags & RTM_F_CLONED)
 947                 filter->dump_routes = false;
 948         else
 949                 filter->dump_exceptions = false;
 950
 951         filter->flags    = rtm->rtm_flags;
 952         filter->protocol = rtm->rtm_protocol;
 953         filter->rt_type  = rtm->rtm_type;
 954         filter->table_id = rtm->rtm_table;
 955
 956         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
 957                                             rtm_ipv4_policy, extack);
 958         if (err < 0)
 959                 return err;
 960
 961         for (i = 0; i <= RTA_MAX; ++i) {
 962                 int ifindex;
 963
 964                 if (!tb[i])
 965                         continue;
 966
 967                 switch (i) {
 968                 case RTA_TABLE:
 969                         filter->table_id = nla_get_u32(tb[i]);
 970                         break;
 971                 case RTA_OIF:
 972                         ifindex = nla_get_u32(tb[i]);
 973                         if (filter->rtnl_held)
 974                                 filter->dev = __dev_get_by_index(net, ifindex);
 975                         else
 976                                 filter->dev = dev_get_by_index_rcu(net, ifindex);
 977                         if (!filter->dev)
 978                                 return -ENODEV;
 979                         break;
 980                 default:
 981                         NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
 982                         return -EINVAL;
 983                 }
 984         }
 985
 986         if (filter->flags || filter->protocol || filter->rt_type ||
 987             filter->table_id || filter->dev) {
 988                 filter->filter_set = 1;
 989                 cb->answer_flags = NLM_F_DUMP_FILTERED;
 990         }
 991
 992         return 0;
 993 }
 994 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
 995
 996 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 997 {
 998         struct fib_dump_filter filter = {
 999                 .dump_routes = true,
1000                 .dump_exceptions = true,
1001                 .rtnl_held = false,
1002         };
1003         const struct nlmsghdr *nlh = cb->nlh;
1004         struct net *net = sock_net(skb->sk);
1005         unsigned int h, s_h;
1006         unsigned int e = 0, s_e;
1007         struct fib_table *tb;
1008         struct hlist_head *head;
1009         int dumped = 0, err = 0;
1010
1011         rcu_read_lock();
1012         if (cb->strict_check) {
1013                 err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
1014                 if (err < 0)
1015                         goto unlock;
1016         } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
1017                 struct rtmsg *rtm = nlmsg_data(nlh);
1018
1019                 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
1020         }
1021
1022         /* ipv4 does not use prefix flag */
1023         if (filter.flags & RTM_F_PREFIX)
1024                 goto unlock;
1025
1026         if (filter.table_id) {
1027                 tb = fib_get_table(net, filter.table_id);
1028                 if (!tb) {
1029                         if (rtnl_msg_family(cb->nlh) != PF_INET)
1030                                 goto unlock;
1031
1032                         NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1033                         err = -ENOENT;
1034                         goto unlock;
1035                 }
1036                 err = fib_table_dump(tb, skb, cb, &filter);
1037                 goto unlock;
1038         }
1039
1040         s_h = cb->args[0];
1041         s_e = cb->args[1];
1042
1043         err = 0;
1044         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1045                 e = 0;
1046                 head = &net->ipv4.fib_table_hash[h];
1047                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1048                         if (e < s_e)
1049                                 goto next;
1050                         if (dumped)
1051                                 memset(&cb->args[2], 0, sizeof(cb->args) -
1052                                                  2 * sizeof(cb->args[0]));
1053                         err = fib_table_dump(tb, skb, cb, &filter);
1054                         if (err < 0)
1055                                 goto out;
1056                         dumped = 1;
1057 next:
1058                         e++;
1059                 }
1060         }
1061 out:
1062
1063         cb->args[1] = e;
1064         cb->args[0] = h;
1065
1066 unlock:
1067         rcu_read_unlock();
1068         return err;
1069 }
1070
1071 /* Prepare and feed intra-kernel routing request.
1072  * Really, it should be netlink message, but :-( netlink
1073  * can be not configured, so that we feed it directly
1074  * to fib engine. It is legal, because all events occur
1075  * only when netlink is already locked.
1076  */
1077 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1078                       struct in_ifaddr *ifa, u32 rt_priority)
1079 {
1080         struct net *net = dev_net(ifa->ifa_dev->dev);
1081         u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1082         struct fib_table *tb;
1083         struct fib_config cfg = {
1084                 .fc_protocol = RTPROT_KERNEL,
1085                 .fc_type = type,
1086                 .fc_dst = dst,
1087                 .fc_dst_len = dst_len,
1088                 .fc_priority = rt_priority,
1089                 .fc_prefsrc = ifa->ifa_local,
1090                 .fc_oif = ifa->ifa_dev->dev->ifindex,
1091                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1092                 .fc_nlinfo = {
1093                         .nl_net = net,
1094                 },
1095         };
1096
1097         if (!tb_id)
1098                 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1099
1100         tb = fib_new_table(net, tb_id);
1101         if (!tb)
1102                 return;
1103
1104         cfg.fc_table = tb->tb_id;
1105
1106         if (type != RTN_LOCAL)
1107                 cfg.fc_scope = RT_SCOPE_LINK;
1108         else
1109                 cfg.fc_scope = RT_SCOPE_HOST;
1110
1111         if (cmd == RTM_NEWROUTE)
1112                 fib_table_insert(net, tb, &cfg, NULL);
1113         else
1114                 fib_table_delete(net, tb, &cfg, NULL);
1115 }
1116
1117 void fib_add_ifaddr(struct in_ifaddr *ifa)
1118 {
1119         struct in_device *in_dev = ifa->ifa_dev;
1120         struct net_device *dev = in_dev->dev;
1121         struct in_ifaddr *prim = ifa;
1122         __be32 mask = ifa->ifa_mask;
1123         __be32 addr = ifa->ifa_local;
1124         __be32 prefix = ifa->ifa_address & mask;
1125
1126         if (ifa->ifa_flags & IFA_F_SECONDARY) {
1127                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
1128                 if (!prim) {
1129                         pr_warn("%s: bug: prim == NULL\n", __func__);
1130                         return;
1131                 }
1132         }
1133
1134         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1135
1136         if (!(dev->flags & IFF_UP))
1137                 return;
1138
1139         /* Add broadcast address, if it is explicitly assigned. */
1140         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) {
1141                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1142                           prim, 0);
1143                 arp_invalidate(dev, ifa->ifa_broadcast, false);
1144         }
1145
1146         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1147             (prefix != addr || ifa->ifa_prefixlen < 32)) {
1148                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1149                         fib_magic(RTM_NEWROUTE,
1150                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1151                                   prefix, ifa->ifa_prefixlen, prim,
1152                                   ifa->ifa_rt_priority);
1153
1154                 /* Add the network broadcast address, when it makes sense */
1155                 if (ifa->ifa_prefixlen < 31) {
1156                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1157                                   32, prim, 0);
1158                         arp_invalidate(dev, prefix | ~mask, false);
1159                 }
1160         }
1161 }
1162
1163 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1164 {
1165         __be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1166         struct in_device *in_dev = ifa->ifa_dev;
1167         struct net_device *dev = in_dev->dev;
1168
1169         if (!(dev->flags & IFF_UP) ||
1170             ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1171             ipv4_is_zeronet(prefix) ||
1172             (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32))
1173                 return;
1174
1175         /* add the new */
1176         fib_magic(RTM_NEWROUTE,
1177                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1178                   prefix, ifa->ifa_prefixlen, ifa, new_metric);
1179
1180         /* delete the old */
1181         fib_magic(RTM_DELROUTE,
1182                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1183                   prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1184 }
1185
1186 /* Delete primary or secondary address.
1187  * Optionally, on secondary address promotion consider the addresses
1188  * from subnet iprim as deleted, even if they are in device list.
1189  * In this case the secondary ifa can be in device list.
1190  */
1191 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1192 {
1193         struct in_device *in_dev = ifa->ifa_dev;
1194         struct net_device *dev = in_dev->dev;
1195         struct in_ifaddr *ifa1;
1196         struct in_ifaddr *prim = ifa, *prim1 = NULL;
1197         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1198         __be32 any = ifa->ifa_address & ifa->ifa_mask;
1199 #define LOCAL_OK        1
1200 #define BRD_OK          2
1201 #define BRD0_OK         4
1202 #define BRD1_OK         8
1203         unsigned int ok = 0;
1204         int subnet = 0;         /* Primary network */
1205         int gone = 1;           /* Address is missing */
1206         int same_prefsrc = 0;   /* Another primary with same IP */
1207
1208         if (ifa->ifa_flags & IFA_F_SECONDARY) {
1209                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1210                 if (!prim) {
1211                         /* if the device has been deleted, we don't perform
1212                          * address promotion
1213                          */
1214                         if (!in_dev->dead)
1215                                 pr_warn("%s: bug: prim == NULL\n", __func__);
1216                         return;
1217                 }
1218                 if (iprim && iprim != prim) {
1219                         pr_warn("%s: bug: iprim != prim\n", __func__);
1220                         return;
1221                 }
1222         } else if (!ipv4_is_zeronet(any) &&
1223                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1224                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1225                         fib_magic(RTM_DELROUTE,
1226                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1227                                   any, ifa->ifa_prefixlen, prim, 0);
1228                 subnet = 1;
1229         }
1230
1231         if (in_dev->dead)
1232                 goto no_promotions;
1233
1234         /* Deletion is more complicated than add.
1235          * We should take care of not to delete too much :-)
1236          *
1237          * Scan address list to be sure that addresses are really gone.
1238          */
1239         rcu_read_lock();
1240         in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1241                 if (ifa1 == ifa) {
1242                         /* promotion, keep the IP */
1243                         gone = 0;
1244                         continue;
1245                 }
1246                 /* Ignore IFAs from our subnet */
1247                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1248                     inet_ifa_match(ifa1->ifa_address, iprim))
1249                         continue;
1250
1251                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
1252                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1253                         /* Another address from our subnet? */
1254                         if (ifa1->ifa_mask == prim->ifa_mask &&
1255                             inet_ifa_match(ifa1->ifa_address, prim))
1256                                 prim1 = prim;
1257                         else {
1258                                 /* We reached the secondaries, so
1259                                  * same_prefsrc should be determined.
1260                                  */
1261                                 if (!same_prefsrc)
1262                                         continue;
1263                                 /* Search new prim1 if ifa1 is not
1264                                  * using the current prim1
1265                                  */
1266                                 if (!prim1 ||
1267                                     ifa1->ifa_mask != prim1->ifa_mask ||
1268                                     !inet_ifa_match(ifa1->ifa_address, prim1))
1269                                         prim1 = inet_ifa_byprefix(in_dev,
1270                                                         ifa1->ifa_address,
1271                                                         ifa1->ifa_mask);
1272                                 if (!prim1)
1273                                         continue;
1274                                 if (prim1->ifa_local != prim->ifa_local)
1275                                         continue;
1276                         }
1277                 } else {
1278                         if (prim->ifa_local != ifa1->ifa_local)
1279                                 continue;
1280                         prim1 = ifa1;
1281                         if (prim != prim1)
1282                                 same_prefsrc = 1;
1283                 }
1284                 if (ifa->ifa_local == ifa1->ifa_local)
1285                         ok |= LOCAL_OK;
1286                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1287                         ok |= BRD_OK;
1288                 if (brd == ifa1->ifa_broadcast)
1289                         ok |= BRD1_OK;
1290                 if (any == ifa1->ifa_broadcast)
1291                         ok |= BRD0_OK;
1292                 /* primary has network specific broadcasts */
1293                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1294                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1295                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1296
1297                         if (!ipv4_is_zeronet(any1)) {
1298                                 if (ifa->ifa_broadcast == brd1 ||
1299                                     ifa->ifa_broadcast == any1)
1300                                         ok |= BRD_OK;
1301                                 if (brd == brd1 || brd == any1)
1302                                         ok |= BRD1_OK;
1303                                 if (any == brd1 || any == any1)
1304                                         ok |= BRD0_OK;
1305                         }
1306                 }
1307         }
1308         rcu_read_unlock();
1309
1310 no_promotions:
1311         if (!(ok & BRD_OK))
1312                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1313                           prim, 0);
1314         if (subnet && ifa->ifa_prefixlen < 31) {
1315                 if (!(ok & BRD1_OK))
1316                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1317                                   prim, 0);
1318                 if (!(ok & BRD0_OK))
1319                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1320                                   prim, 0);
1321         }
1322         if (!(ok & LOCAL_OK)) {
1323                 unsigned int addr_type;
1324
1325                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1326
1327                 /* Check, that this local address finally disappeared. */
1328                 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1329                                                      ifa->ifa_local);
1330                 if (gone && addr_type != RTN_LOCAL) {
1331                         /* And the last, but not the least thing.
1332                          * We must flush stray FIB entries.
1333                          *
1334                          * First of all, we scan fib_info list searching
1335                          * for stray nexthop entries, then ignite fib_flush.
1336                          */
1337                         if (fib_sync_down_addr(dev, ifa->ifa_local))
1338                                 fib_flush(dev_net(dev));
1339                 }
1340         }
1341 #undef LOCAL_OK
1342 #undef BRD_OK
1343 #undef BRD0_OK
1344 #undef BRD1_OK
1345 }
1346
1347 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1348 {
1349
1350         struct fib_result       res;
1351         struct flowi4           fl4 = {
1352                 .flowi4_mark = frn->fl_mark,
1353                 .daddr = frn->fl_addr,
1354                 .flowi4_tos = frn->fl_tos & INET_DSCP_MASK,
1355                 .flowi4_scope = frn->fl_scope,
1356         };
1357         struct fib_table *tb;
1358
1359         rcu_read_lock();
1360
1361         tb = fib_get_table(net, frn->tb_id_in);
1362
1363         frn->err = -ENOENT;
1364         if (tb) {
1365                 local_bh_disable();
1366
1367                 frn->tb_id = tb->tb_id;
1368                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1369
1370                 if (!frn->err) {
1371                         frn->prefixlen = res.prefixlen;
1372                         frn->nh_sel = res.nh_sel;
1373                         frn->type = res.type;
1374                         frn->scope = res.scope;
1375                 }
1376                 local_bh_enable();
1377         }
1378
1379         rcu_read_unlock();
1380 }
1381
1382 static void nl_fib_input(struct sk_buff *skb)
1383 {
1384         struct net *net;
1385         struct fib_result_nl *frn;
1386         struct nlmsghdr *nlh;
1387         u32 portid;
1388
1389         net = sock_net(skb->sk);
1390         nlh = nlmsg_hdr(skb);
1391         if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1392             skb->len < nlh->nlmsg_len ||
1393             nlmsg_len(nlh) < sizeof(*frn))
1394                 return;
1395
1396         skb = netlink_skb_clone(skb, GFP_KERNEL);
1397         if (!skb)
1398                 return;
1399         nlh = nlmsg_hdr(skb);
1400
1401         frn = nlmsg_data(nlh);
1402         nl_fib_lookup(net, frn);
1403
1404         portid = NETLINK_CB(skb).portid;      /* netlink portid */
1405         NETLINK_CB(skb).portid = 0;        /* from kernel */
1406         NETLINK_CB(skb).dst_group = 0;  /* unicast */
1407         nlmsg_unicast(net->ipv4.fibnl, skb, portid);
1408 }
1409
1410 static int __net_init nl_fib_lookup_init(struct net *net)
1411 {
1412         struct sock *sk;
1413         struct netlink_kernel_cfg cfg = {
1414                 .input  = nl_fib_input,
1415         };
1416
1417         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1418         if (!sk)
1419                 return -EAFNOSUPPORT;
1420         net->ipv4.fibnl = sk;
1421         return 0;
1422 }
1423
1424 static void nl_fib_lookup_exit(struct net *net)
1425 {
1426         netlink_kernel_release(net->ipv4.fibnl);
1427         net->ipv4.fibnl = NULL;
1428 }
1429
1430 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1431                            bool force)
1432 {
1433         if (fib_sync_down_dev(dev, event, force))
1434                 fib_flush(dev_net(dev));
1435         else
1436                 rt_cache_flush(dev_net(dev));
1437         arp_ifdown(dev);
1438 }
1439
1440 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1441 {
1442         struct in_ifaddr *ifa = ptr;
1443         struct net_device *dev = ifa->ifa_dev->dev;
1444         struct net *net = dev_net(dev);
1445
1446         switch (event) {
1447         case NETDEV_UP:
1448                 fib_add_ifaddr(ifa);
1449 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1450                 fib_sync_up(dev, RTNH_F_DEAD);
1451 #endif
1452                 atomic_inc(&net->ipv4.dev_addr_genid);
1453                 rt_cache_flush(dev_net(dev));
1454                 break;
1455         case NETDEV_DOWN:
1456                 fib_del_ifaddr(ifa, NULL);
1457                 atomic_inc(&net->ipv4.dev_addr_genid);
1458                 if (!ifa->ifa_dev->ifa_list) {
1459                         /* Last address was deleted from this interface.
1460                          * Disable IP.
1461                          */
1462                         fib_disable_ip(dev, event, true);
1463                 } else {
1464                         rt_cache_flush(dev_net(dev));
1465                 }
1466                 break;
1467         }
1468         return NOTIFY_DONE;
1469 }
1470
1471 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1472 {
1473         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1474         struct netdev_notifier_changeupper_info *upper_info = ptr;
1475         struct netdev_notifier_info_ext *info_ext = ptr;
1476         struct in_device *in_dev;
1477         struct net *net = dev_net(dev);
1478         struct in_ifaddr *ifa;
1479         unsigned int flags;
1480
1481         if (event == NETDEV_UNREGISTER) {
1482                 fib_disable_ip(dev, event, true);
1483                 rt_flush_dev(dev);
1484                 return NOTIFY_DONE;
1485         }
1486
1487         in_dev = __in_dev_get_rtnl(dev);
1488         if (!in_dev)
1489                 return NOTIFY_DONE;
1490
1491         switch (event) {
1492         case NETDEV_UP:
1493                 in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1494                         fib_add_ifaddr(ifa);
1495                 }
1496 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1497                 fib_sync_up(dev, RTNH_F_DEAD);
1498 #endif
1499                 atomic_inc(&net->ipv4.dev_addr_genid);
1500                 rt_cache_flush(net);
1501                 break;
1502         case NETDEV_DOWN:
1503                 fib_disable_ip(dev, event, false);
1504                 break;
1505         case NETDEV_CHANGE:
1506                 flags = dev_get_flags(dev);
1507                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1508                         fib_sync_up(dev, RTNH_F_LINKDOWN);
1509                 else
1510                         fib_sync_down_dev(dev, event, false);
1511                 rt_cache_flush(net);
1512                 break;
1513         case NETDEV_CHANGEMTU:
1514                 fib_sync_mtu(dev, info_ext->ext.mtu);
1515                 rt_cache_flush(net);
1516                 break;
1517         case NETDEV_CHANGEUPPER:
1518                 upper_info = ptr;
1519                 /* flush all routes if dev is linked to or unlinked from
1520                  * an L3 master device (e.g., VRF)
1521                  */
1522                 if (upper_info->upper_dev &&
1523                     netif_is_l3_master(upper_info->upper_dev))
1524                         fib_disable_ip(dev, NETDEV_DOWN, true);
1525                 break;
1526         }
1527         return NOTIFY_DONE;
1528 }
1529
1530 static struct notifier_block fib_inetaddr_notifier = {
1531         .notifier_call = fib_inetaddr_event,
1532 };
1533
1534 static struct notifier_block fib_netdev_notifier = {
1535         .notifier_call = fib_netdev_event,
1536 };
1537
1538 static int __net_init ip_fib_net_init(struct net *net)
1539 {
1540         int err;
1541         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1542
1543         err = fib4_notifier_init(net);
1544         if (err)
1545                 return err;
1546
1547 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1548         /* Default to 3-tuple */
1549         net->ipv4.sysctl_fib_multipath_hash_fields =
1550                 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK;
1551 #endif
1552
1553         /* Avoid false sharing : Use at least a full cache line */
1554         size = max_t(size_t, size, L1_CACHE_BYTES);
1555
1556         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1557         if (!net->ipv4.fib_table_hash) {
1558                 err = -ENOMEM;
1559                 goto err_table_hash_alloc;
1560         }
1561
1562         err = fib4_rules_init(net);
1563         if (err < 0)
1564                 goto err_rules_init;
1565         return 0;
1566
1567 err_rules_init:
1568         kfree(net->ipv4.fib_table_hash);
1569 err_table_hash_alloc:
1570         fib4_notifier_exit(net);
1571         return err;
1572 }
1573
1574 static void ip_fib_net_exit(struct net *net)
1575 {
1576         int i;
1577
1578         ASSERT_RTNL();
1579 #ifdef CONFIG_IP_MULTIPLE_TABLES
1580         RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1581         RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1582 #endif
1583         /* Destroy the tables in reverse order to guarantee that the
1584          * local table, ID 255, is destroyed before the main table, ID
1585          * 254. This is necessary as the local table may contain
1586          * references to data contained in the main table.
1587          */
1588         for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1589                 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1590                 struct hlist_node *tmp;
1591                 struct fib_table *tb;
1592
1593                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1594                         hlist_del(&tb->tb_hlist);
1595                         fib_table_flush(net, tb, true);
1596                         fib_free_table(tb);
1597                 }
1598         }
1599
1600 #ifdef CONFIG_IP_MULTIPLE_TABLES
1601         fib4_rules_exit(net);
1602 #endif
1603
1604         kfree(net->ipv4.fib_table_hash);
1605         fib4_notifier_exit(net);
1606 }
1607
1608 static int __net_init fib_net_init(struct net *net)
1609 {
1610         int error;
1611
1612 #ifdef CONFIG_IP_ROUTE_CLASSID
1613         atomic_set(&net->ipv4.fib_num_tclassid_users, 0);
1614 #endif
1615         error = ip_fib_net_init(net);
1616         if (error < 0)
1617                 goto out;
1618         error = nl_fib_lookup_init(net);
1619         if (error < 0)
1620                 goto out_nlfl;
1621         error = fib_proc_init(net);
1622         if (error < 0)
1623                 goto out_proc;
1624 out:
1625         return error;
1626
1627 out_proc:
1628         nl_fib_lookup_exit(net);
1629 out_nlfl:
1630         rtnl_lock();
1631         ip_fib_net_exit(net);
1632         rtnl_unlock();
1633         goto out;
1634 }
1635
1636 static void __net_exit fib_net_exit(struct net *net)
1637 {
1638         fib_proc_exit(net);
1639         nl_fib_lookup_exit(net);
1640 }
1641
1642 static void __net_exit fib_net_exit_batch(struct list_head *net_list)
1643 {
1644         struct net *net;
1645
1646         rtnl_lock();
1647         list_for_each_entry(net, net_list, exit_list)
1648                 ip_fib_net_exit(net);
1649
1650         rtnl_unlock();
1651 }
1652
1653 static struct pernet_operations fib_net_ops = {
1654         .init = fib_net_init,
1655         .exit = fib_net_exit,
1656         .exit_batch = fib_net_exit_batch,
1657 };
1658
1659 static const struct rtnl_msg_handler fib_rtnl_msg_handlers[] __initconst = {
1660         {.protocol = PF_INET, .msgtype = RTM_NEWROUTE,
1661          .doit = inet_rtm_newroute},
1662         {.protocol = PF_INET, .msgtype = RTM_DELROUTE,
1663          .doit = inet_rtm_delroute},
1664         {.protocol = PF_INET, .msgtype = RTM_GETROUTE, .dumpit = inet_dump_fib,
1665          .flags = RTNL_FLAG_DUMP_UNLOCKED | RTNL_FLAG_DUMP_SPLIT_NLM_DONE},
1666 };
1667
1668 void __init ip_fib_init(void)
1669 {
1670         fib_trie_init();
1671
1672         register_pernet_subsys(&fib_net_ops);
1673
1674         register_netdevice_notifier(&fib_netdev_notifier);
1675         register_inetaddr_notifier(&fib_inetaddr_notifier);
1676
1677         rtnl_register_many(fib_rtnl_msg_handlers);
1678 }