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/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 = ip_hdr(skb)->tos & IPTOS_RT_MASK,
 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                                  u8 tos, int oif, struct net_device *dev,
 346                                  int rpf, struct in_device *idev, u32 *itag)
 347 {
 348         struct net *net = dev_net(dev);
 349         struct flow_keys flkeys;
 350         int ret, no_addr;
 351         struct fib_result res;
 352         struct flowi4 fl4;
 353         bool dev_match;
 354
 355         fl4.flowi4_oif = 0;
 356         fl4.flowi4_l3mdev = l3mdev_master_ifindex_rcu(dev);
 357         fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
 358         fl4.daddr = src;
 359         fl4.saddr = dst;
 360         fl4.flowi4_tos = tos;
 361         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 362         fl4.flowi4_tun_key.tun_id = 0;
 363         fl4.flowi4_flags = 0;
 364         fl4.flowi4_uid = sock_net_uid(net, NULL);
 365         fl4.flowi4_multipath_hash = 0;
 366
 367         no_addr = idev->ifa_list == NULL;
 368
 369         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
 370         if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
 371                 fl4.flowi4_proto = 0;
 372                 fl4.fl4_sport = 0;
 373                 fl4.fl4_dport = 0;
 374         } else {
 375                 swap(fl4.fl4_sport, fl4.fl4_dport);
 376         }
 377
 378         if (fib_lookup(net, &fl4, &res, 0))
 379                 goto last_resort;
 380         if (res.type != RTN_UNICAST &&
 381             (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
 382                 goto e_inval;
 383         fib_combine_itag(itag, &res);
 384
 385         dev_match = fib_info_nh_uses_dev(res.fi, dev);
 386         /* This is not common, loopback packets retain skb_dst so normally they
 387          * would not even hit this slow path.
 388          */
 389         dev_match = dev_match || (res.type == RTN_LOCAL &&
 390                                   dev == net->loopback_dev);
 391         if (dev_match) {
 392                 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
 393                 return ret;
 394         }
 395         if (no_addr)
 396                 goto last_resort;
 397         if (rpf == 1)
 398                 goto e_rpf;
 399         fl4.flowi4_oif = dev->ifindex;
 400
 401         ret = 0;
 402         if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
 403                 if (res.type == RTN_UNICAST)
 404                         ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
 405         }
 406         return ret;
 407
 408 last_resort:
 409         if (rpf)
 410                 goto e_rpf;
 411         *itag = 0;
 412         return 0;
 413
 414 e_inval:
 415         return -EINVAL;
 416 e_rpf:
 417         return -EXDEV;
 418 }
 419
 420 /* Ignore rp_filter for packets protected by IPsec. */
 421 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 422                         u8 tos, int oif, struct net_device *dev,
 423                         struct in_device *idev, u32 *itag)
 424 {
 425         int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
 426         struct net *net = dev_net(dev);
 427
 428         if (!r && !fib_num_tclassid_users(net) &&
 429             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
 430                 if (IN_DEV_ACCEPT_LOCAL(idev))
 431                         goto ok;
 432                 /* with custom local routes in place, checking local addresses
 433                  * only will be too optimistic, with custom rules, checking
 434                  * local addresses only can be too strict, e.g. due to vrf
 435                  */
 436                 if (net->ipv4.fib_has_custom_local_routes ||
 437                     fib4_has_custom_rules(net))
 438                         goto full_check;
 439                 /* Within the same container, it is regarded as a martian source,
 440                  * and the same host but different containers are not.
 441                  */
 442                 if (inet_lookup_ifaddr_rcu(net, src))
 443                         return -EINVAL;
 444
 445 ok:
 446                 *itag = 0;
 447                 return 0;
 448         }
 449
 450 full_check:
 451         return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
 452 }
 453
 454 static inline __be32 sk_extract_addr(struct sockaddr *addr)
 455 {
 456         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 457 }
 458
 459 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
 460 {
 461         struct nlattr *nla;
 462
 463         nla = (struct nlattr *) ((char *) mx + len);
 464         nla->nla_type = type;
 465         nla->nla_len = nla_attr_size(4);
 466         *(u32 *) nla_data(nla) = value;
 467
 468         return len + nla_total_size(4);
 469 }
 470
 471 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
 472                                  struct fib_config *cfg)
 473 {
 474         __be32 addr;
 475         int plen;
 476
 477         memset(cfg, 0, sizeof(*cfg));
 478         cfg->fc_nlinfo.nl_net = net;
 479
 480         if (rt->rt_dst.sa_family != AF_INET)
 481                 return -EAFNOSUPPORT;
 482
 483         /*
 484          * Check mask for validity:
 485          * a) it must be contiguous.
 486          * b) destination must have all host bits clear.
 487          * c) if application forgot to set correct family (AF_INET),
 488          *    reject request unless it is absolutely clear i.e.
 489          *    both family and mask are zero.
 490          */
 491         plen = 32;
 492         addr = sk_extract_addr(&rt->rt_dst);
 493         if (!(rt->rt_flags & RTF_HOST)) {
 494                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
 495
 496                 if (rt->rt_genmask.sa_family != AF_INET) {
 497                         if (mask || rt->rt_genmask.sa_family)
 498                                 return -EAFNOSUPPORT;
 499                 }
 500
 501                 if (bad_mask(mask, addr))
 502                         return -EINVAL;
 503
 504                 plen = inet_mask_len(mask);
 505         }
 506
 507         cfg->fc_dst_len = plen;
 508         cfg->fc_dst = addr;
 509
 510         if (cmd != SIOCDELRT) {
 511                 cfg->fc_nlflags = NLM_F_CREATE;
 512                 cfg->fc_protocol = RTPROT_BOOT;
 513         }
 514
 515         if (rt->rt_metric)
 516                 cfg->fc_priority = rt->rt_metric - 1;
 517
 518         if (rt->rt_flags & RTF_REJECT) {
 519                 cfg->fc_scope = RT_SCOPE_HOST;
 520                 cfg->fc_type = RTN_UNREACHABLE;
 521                 return 0;
 522         }
 523
 524         cfg->fc_scope = RT_SCOPE_NOWHERE;
 525         cfg->fc_type = RTN_UNICAST;
 526
 527         if (rt->rt_dev) {
 528                 char *colon;
 529                 struct net_device *dev;
 530                 char devname[IFNAMSIZ];
 531
 532                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
 533                         return -EFAULT;
 534
 535                 devname[IFNAMSIZ-1] = 0;
 536                 colon = strchr(devname, ':');
 537                 if (colon)
 538                         *colon = 0;
 539                 dev = __dev_get_by_name(net, devname);
 540                 if (!dev)
 541                         return -ENODEV;
 542                 cfg->fc_oif = dev->ifindex;
 543                 cfg->fc_table = l3mdev_fib_table(dev);
 544                 if (colon) {
 545                         const struct in_ifaddr *ifa;
 546                         struct in_device *in_dev;
 547
 548                         in_dev = __in_dev_get_rtnl(dev);
 549                         if (!in_dev)
 550                                 return -ENODEV;
 551
 552                         *colon = ':';
 553
 554                         rcu_read_lock();
 555                         in_dev_for_each_ifa_rcu(ifa, in_dev) {
 556                                 if (strcmp(ifa->ifa_label, devname) == 0)
 557                                         break;
 558                         }
 559                         rcu_read_unlock();
 560
 561                         if (!ifa)
 562                                 return -ENODEV;
 563                         cfg->fc_prefsrc = ifa->ifa_local;
 564                 }
 565         }
 566
 567         addr = sk_extract_addr(&rt->rt_gateway);
 568         if (rt->rt_gateway.sa_family == AF_INET && addr) {
 569                 unsigned int addr_type;
 570
 571                 cfg->fc_gw4 = addr;
 572                 cfg->fc_gw_family = AF_INET;
 573                 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
 574                 if (rt->rt_flags & RTF_GATEWAY &&
 575                     addr_type == RTN_UNICAST)
 576                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
 577         }
 578
 579         if (cmd == SIOCDELRT)
 580                 return 0;
 581
 582         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
 583                 return -EINVAL;
 584
 585         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 586                 cfg->fc_scope = RT_SCOPE_LINK;
 587
 588         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 589                 struct nlattr *mx;
 590                 int len = 0;
 591
 592                 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
 593                 if (!mx)
 594                         return -ENOMEM;
 595
 596                 if (rt->rt_flags & RTF_MTU)
 597                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 598
 599                 if (rt->rt_flags & RTF_WINDOW)
 600                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 601
 602                 if (rt->rt_flags & RTF_IRTT)
 603                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 604
 605                 cfg->fc_mx = mx;
 606                 cfg->fc_mx_len = len;
 607         }
 608
 609         return 0;
 610 }
 611
 612 /*
 613  * Handle IP routing ioctl calls.
 614  * These are used to manipulate the routing tables
 615  */
 616 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
 617 {
 618         struct fib_config cfg;
 619         int err;
 620
 621         switch (cmd) {
 622         case SIOCADDRT:         /* Add a route */
 623         case SIOCDELRT:         /* Delete a route */
 624                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 625                         return -EPERM;
 626
 627                 rtnl_lock();
 628                 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
 629                 if (err == 0) {
 630                         struct fib_table *tb;
 631
 632                         if (cmd == SIOCDELRT) {
 633                                 tb = fib_get_table(net, cfg.fc_table);
 634                                 if (tb)
 635                                         err = fib_table_delete(net, tb, &cfg,
 636                                                                NULL);
 637                                 else
 638                                         err = -ESRCH;
 639                         } else {
 640                                 tb = fib_new_table(net, cfg.fc_table);
 641                                 if (tb)
 642                                         err = fib_table_insert(net, tb,
 643                                                                &cfg, NULL);
 644                                 else
 645                                         err = -ENOBUFS;
 646                         }
 647
 648                         /* allocated by rtentry_to_fib_config() */
 649                         kfree(cfg.fc_mx);
 650                 }
 651                 rtnl_unlock();
 652                 return err;
 653         }
 654         return -EINVAL;
 655 }
 656
 657 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 658         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
 659         [RTA_DST]               = { .type = NLA_U32 },
 660         [RTA_SRC]               = { .type = NLA_U32 },
 661         [RTA_IIF]               = { .type = NLA_U32 },
 662         [RTA_OIF]               = { .type = NLA_U32 },
 663         [RTA_GATEWAY]           = { .type = NLA_U32 },
 664         [RTA_PRIORITY]          = { .type = NLA_U32 },
 665         [RTA_PREFSRC]           = { .type = NLA_U32 },
 666         [RTA_METRICS]           = { .type = NLA_NESTED },
 667         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 668         [RTA_FLOW]              = { .type = NLA_U32 },
 669         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
 670         [RTA_ENCAP]             = { .type = NLA_NESTED },
 671         [RTA_UID]               = { .type = NLA_U32 },
 672         [RTA_MARK]              = { .type = NLA_U32 },
 673         [RTA_TABLE]             = { .type = NLA_U32 },
 674         [RTA_IP_PROTO]          = { .type = NLA_U8 },
 675         [RTA_SPORT]             = { .type = NLA_U16 },
 676         [RTA_DPORT]             = { .type = NLA_U16 },
 677         [RTA_NH_ID]             = { .type = NLA_U32 },
 678 };
 679
 680 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
 681                     struct netlink_ext_ack *extack)
 682 {
 683         struct rtvia *via;
 684         int alen;
 685
 686         if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
 687                 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
 688                 return -EINVAL;
 689         }
 690
 691         via = nla_data(nla);
 692         alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
 693
 694         switch (via->rtvia_family) {
 695         case AF_INET:
 696                 if (alen != sizeof(__be32)) {
 697                         NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
 698                         return -EINVAL;
 699                 }
 700                 cfg->fc_gw_family = AF_INET;
 701                 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
 702                 break;
 703         case AF_INET6:
 704 #if IS_ENABLED(CONFIG_IPV6)
 705                 if (alen != sizeof(struct in6_addr)) {
 706                         NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
 707                         return -EINVAL;
 708                 }
 709                 cfg->fc_gw_family = AF_INET6;
 710                 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
 711 #else
 712                 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
 713                 return -EINVAL;
 714 #endif
 715                 break;
 716         default:
 717                 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
 718                 return -EINVAL;
 719         }
 720
 721         return 0;
 722 }
 723
 724 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 725                              struct nlmsghdr *nlh, struct fib_config *cfg,
 726                              struct netlink_ext_ack *extack)
 727 {
 728         bool has_gw = false, has_via = false;
 729         struct nlattr *attr;
 730         int err, remaining;
 731         struct rtmsg *rtm;
 732
 733         err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
 734                                         rtm_ipv4_policy, extack);
 735         if (err < 0)
 736                 goto errout;
 737
 738         memset(cfg, 0, sizeof(*cfg));
 739
 740         rtm = nlmsg_data(nlh);
 741
 742         if (!inet_validate_dscp(rtm->rtm_tos)) {
 743                 NL_SET_ERR_MSG(extack,
 744                                "Invalid dsfield (tos): ECN bits must be 0");
 745                 err = -EINVAL;
 746                 goto errout;
 747         }
 748         cfg->fc_dscp = inet_dsfield_to_dscp(rtm->rtm_tos);
 749
 750         cfg->fc_dst_len = rtm->rtm_dst_len;
 751         cfg->fc_table = rtm->rtm_table;
 752         cfg->fc_protocol = rtm->rtm_protocol;
 753         cfg->fc_scope = rtm->rtm_scope;
 754         cfg->fc_type = rtm->rtm_type;
 755         cfg->fc_flags = rtm->rtm_flags;
 756         cfg->fc_nlflags = nlh->nlmsg_flags;
 757
 758         cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
 759         cfg->fc_nlinfo.nlh = nlh;
 760         cfg->fc_nlinfo.nl_net = net;
 761
 762         if (cfg->fc_type > RTN_MAX) {
 763                 NL_SET_ERR_MSG(extack, "Invalid route type");
 764                 err = -EINVAL;
 765                 goto errout;
 766         }
 767
 768         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 769                 switch (nla_type(attr)) {
 770                 case RTA_DST:
 771                         cfg->fc_dst = nla_get_be32(attr);
 772                         break;
 773                 case RTA_OIF:
 774                         cfg->fc_oif = nla_get_u32(attr);
 775                         break;
 776                 case RTA_GATEWAY:
 777                         has_gw = true;
 778                         cfg->fc_gw4 = nla_get_be32(attr);
 779                         if (cfg->fc_gw4)
 780                                 cfg->fc_gw_family = AF_INET;
 781                         break;
 782                 case RTA_VIA:
 783                         has_via = true;
 784                         err = fib_gw_from_via(cfg, attr, extack);
 785                         if (err)
 786                                 goto errout;
 787                         break;
 788                 case RTA_PRIORITY:
 789                         cfg->fc_priority = nla_get_u32(attr);
 790                         break;
 791                 case RTA_PREFSRC:
 792                         cfg->fc_prefsrc = nla_get_be32(attr);
 793                         break;
 794                 case RTA_METRICS:
 795                         cfg->fc_mx = nla_data(attr);
 796                         cfg->fc_mx_len = nla_len(attr);
 797                         break;
 798                 case RTA_MULTIPATH:
 799                         err = lwtunnel_valid_encap_type_attr(nla_data(attr),
 800                                                              nla_len(attr),
 801                                                              extack);
 802                         if (err < 0)
 803                                 goto errout;
 804                         cfg->fc_mp = nla_data(attr);
 805                         cfg->fc_mp_len = nla_len(attr);
 806                         break;
 807                 case RTA_FLOW:
 808                         cfg->fc_flow = nla_get_u32(attr);
 809                         break;
 810                 case RTA_TABLE:
 811                         cfg->fc_table = nla_get_u32(attr);
 812                         break;
 813                 case RTA_ENCAP:
 814                         cfg->fc_encap = attr;
 815                         break;
 816                 case RTA_ENCAP_TYPE:
 817                         cfg->fc_encap_type = nla_get_u16(attr);
 818                         err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
 819                                                         extack);
 820                         if (err < 0)
 821                                 goto errout;
 822                         break;
 823                 case RTA_NH_ID:
 824                         cfg->fc_nh_id = nla_get_u32(attr);
 825                         break;
 826                 }
 827         }
 828
 829         if (cfg->fc_nh_id) {
 830                 if (cfg->fc_oif || cfg->fc_gw_family ||
 831                     cfg->fc_encap || cfg->fc_mp) {
 832                         NL_SET_ERR_MSG(extack,
 833                                        "Nexthop specification and nexthop id are mutually exclusive");
 834                         return -EINVAL;
 835                 }
 836         }
 837
 838         if (has_gw && has_via) {
 839                 NL_SET_ERR_MSG(extack,
 840                                "Nexthop configuration can not contain both GATEWAY and VIA");
 841                 return -EINVAL;
 842         }
 843
 844         if (!cfg->fc_table)
 845                 cfg->fc_table = RT_TABLE_MAIN;
 846
 847         return 0;
 848 errout:
 849         return err;
 850 }
 851
 852 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 853                              struct netlink_ext_ack *extack)
 854 {
 855         struct net *net = sock_net(skb->sk);
 856         struct fib_config cfg;
 857         struct fib_table *tb;
 858         int err;
 859
 860         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 861         if (err < 0)
 862                 goto errout;
 863
 864         if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
 865                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
 866                 err = -EINVAL;
 867                 goto errout;
 868         }
 869
 870         tb = fib_get_table(net, cfg.fc_table);
 871         if (!tb) {
 872                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
 873                 err = -ESRCH;
 874                 goto errout;
 875         }
 876
 877         err = fib_table_delete(net, tb, &cfg, extack);
 878 errout:
 879         return err;
 880 }
 881
 882 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
 883                              struct netlink_ext_ack *extack)
 884 {
 885         struct net *net = sock_net(skb->sk);
 886         struct fib_config cfg;
 887         struct fib_table *tb;
 888         int err;
 889
 890         err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
 891         if (err < 0)
 892                 goto errout;
 893
 894         tb = fib_new_table(net, cfg.fc_table);
 895         if (!tb) {
 896                 err = -ENOBUFS;
 897                 goto errout;
 898         }
 899
 900         err = fib_table_insert(net, tb, &cfg, extack);
 901         if (!err && cfg.fc_type == RTN_LOCAL)
 902                 net->ipv4.fib_has_custom_local_routes = true;
 903 errout:
 904         return err;
 905 }
 906
 907 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
 908                           struct fib_dump_filter *filter,
 909                           struct netlink_callback *cb)
 910 {
 911         struct netlink_ext_ack *extack = cb->extack;
 912         struct nlattr *tb[RTA_MAX + 1];
 913         struct rtmsg *rtm;
 914         int err, i;
 915
 916         ASSERT_RTNL();
 917
 918         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
 919                 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
 920                 return -EINVAL;
 921         }
 922
 923         rtm = nlmsg_data(nlh);
 924         if (rtm->rtm_dst_len || rtm->rtm_src_len  || rtm->rtm_tos   ||
 925             rtm->rtm_scope) {
 926                 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
 927                 return -EINVAL;
 928         }
 929
 930         if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
 931                 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
 932                 return -EINVAL;
 933         }
 934         if (rtm->rtm_flags & RTM_F_CLONED)
 935                 filter->dump_routes = false;
 936         else
 937                 filter->dump_exceptions = false;
 938
 939         filter->flags    = rtm->rtm_flags;
 940         filter->protocol = rtm->rtm_protocol;
 941         filter->rt_type  = rtm->rtm_type;
 942         filter->table_id = rtm->rtm_table;
 943
 944         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
 945                                             rtm_ipv4_policy, extack);
 946         if (err < 0)
 947                 return err;
 948
 949         for (i = 0; i <= RTA_MAX; ++i) {
 950                 int ifindex;
 951
 952                 if (!tb[i])
 953                         continue;
 954
 955                 switch (i) {
 956                 case RTA_TABLE:
 957                         filter->table_id = nla_get_u32(tb[i]);
 958                         break;
 959                 case RTA_OIF:
 960                         ifindex = nla_get_u32(tb[i]);
 961                         filter->dev = __dev_get_by_index(net, ifindex);
 962                         if (!filter->dev)
 963                                 return -ENODEV;
 964                         break;
 965                 default:
 966                         NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
 967                         return -EINVAL;
 968                 }
 969         }
 970
 971         if (filter->flags || filter->protocol || filter->rt_type ||
 972             filter->table_id || filter->dev) {
 973                 filter->filter_set = 1;
 974                 cb->answer_flags = NLM_F_DUMP_FILTERED;
 975         }
 976
 977         return 0;
 978 }
 979 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
 980
 981 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 982 {
 983         struct fib_dump_filter filter = { .dump_routes = true,
 984                                           .dump_exceptions = true };
 985         const struct nlmsghdr *nlh = cb->nlh;
 986         struct net *net = sock_net(skb->sk);
 987         unsigned int h, s_h;
 988         unsigned int e = 0, s_e;
 989         struct fib_table *tb;
 990         struct hlist_head *head;
 991         int dumped = 0, err;
 992
 993         if (cb->strict_check) {
 994                 err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
 995                 if (err < 0)
 996                         return err;
 997         } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
 998                 struct rtmsg *rtm = nlmsg_data(nlh);
 999
1000                 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
1001         }
1002
1003         /* ipv4 does not use prefix flag */
1004         if (filter.flags & RTM_F_PREFIX)
1005                 return skb->len;
1006
1007         if (filter.table_id) {
1008                 tb = fib_get_table(net, filter.table_id);
1009                 if (!tb) {
1010                         if (rtnl_msg_family(cb->nlh) != PF_INET)
1011                                 return skb->len;
1012
1013                         NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1014                         return -ENOENT;
1015                 }
1016
1017                 rcu_read_lock();
1018                 err = fib_table_dump(tb, skb, cb, &filter);
1019                 rcu_read_unlock();
1020                 return skb->len ? : err;
1021         }
1022
1023         s_h = cb->args[0];
1024         s_e = cb->args[1];
1025
1026         rcu_read_lock();
1027
1028         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1029                 e = 0;
1030                 head = &net->ipv4.fib_table_hash[h];
1031                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1032                         if (e < s_e)
1033                                 goto next;
1034                         if (dumped)
1035                                 memset(&cb->args[2], 0, sizeof(cb->args) -
1036                                                  2 * sizeof(cb->args[0]));
1037                         err = fib_table_dump(tb, skb, cb, &filter);
1038                         if (err < 0) {
1039                                 if (likely(skb->len))
1040                                         goto out;
1041
1042                                 goto out_err;
1043                         }
1044                         dumped = 1;
1045 next:
1046                         e++;
1047                 }
1048         }
1049 out:
1050         err = skb->len;
1051 out_err:
1052         rcu_read_unlock();
1053
1054         cb->args[1] = e;
1055         cb->args[0] = h;
1056
1057         return err;
1058 }
1059
1060 /* Prepare and feed intra-kernel routing request.
1061  * Really, it should be netlink message, but :-( netlink
1062  * can be not configured, so that we feed it directly
1063  * to fib engine. It is legal, because all events occur
1064  * only when netlink is already locked.
1065  */
1066 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1067                       struct in_ifaddr *ifa, u32 rt_priority)
1068 {
1069         struct net *net = dev_net(ifa->ifa_dev->dev);
1070         u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1071         struct fib_table *tb;
1072         struct fib_config cfg = {
1073                 .fc_protocol = RTPROT_KERNEL,
1074                 .fc_type = type,
1075                 .fc_dst = dst,
1076                 .fc_dst_len = dst_len,
1077                 .fc_priority = rt_priority,
1078                 .fc_prefsrc = ifa->ifa_local,
1079                 .fc_oif = ifa->ifa_dev->dev->ifindex,
1080                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1081                 .fc_nlinfo = {
1082                         .nl_net = net,
1083                 },
1084         };
1085
1086         if (!tb_id)
1087                 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1088
1089         tb = fib_new_table(net, tb_id);
1090         if (!tb)
1091                 return;
1092
1093         cfg.fc_table = tb->tb_id;
1094
1095         if (type != RTN_LOCAL)
1096                 cfg.fc_scope = RT_SCOPE_LINK;
1097         else
1098                 cfg.fc_scope = RT_SCOPE_HOST;
1099
1100         if (cmd == RTM_NEWROUTE)
1101                 fib_table_insert(net, tb, &cfg, NULL);
1102         else
1103                 fib_table_delete(net, tb, &cfg, NULL);
1104 }
1105
1106 void fib_add_ifaddr(struct in_ifaddr *ifa)
1107 {
1108         struct in_device *in_dev = ifa->ifa_dev;
1109         struct net_device *dev = in_dev->dev;
1110         struct in_ifaddr *prim = ifa;
1111         __be32 mask = ifa->ifa_mask;
1112         __be32 addr = ifa->ifa_local;
1113         __be32 prefix = ifa->ifa_address & mask;
1114
1115         if (ifa->ifa_flags & IFA_F_SECONDARY) {
1116                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
1117                 if (!prim) {
1118                         pr_warn("%s: bug: prim == NULL\n", __func__);
1119                         return;
1120                 }
1121         }
1122
1123         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1124
1125         if (!(dev->flags & IFF_UP))
1126                 return;
1127
1128         /* Add broadcast address, if it is explicitly assigned. */
1129         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) {
1130                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1131                           prim, 0);
1132                 arp_invalidate(dev, ifa->ifa_broadcast, false);
1133         }
1134
1135         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1136             (prefix != addr || ifa->ifa_prefixlen < 32)) {
1137                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1138                         fib_magic(RTM_NEWROUTE,
1139                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1140                                   prefix, ifa->ifa_prefixlen, prim,
1141                                   ifa->ifa_rt_priority);
1142
1143                 /* Add the network broadcast address, when it makes sense */
1144                 if (ifa->ifa_prefixlen < 31) {
1145                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1146                                   32, prim, 0);
1147                         arp_invalidate(dev, prefix | ~mask, false);
1148                 }
1149         }
1150 }
1151
1152 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1153 {
1154         __be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1155         struct in_device *in_dev = ifa->ifa_dev;
1156         struct net_device *dev = in_dev->dev;
1157
1158         if (!(dev->flags & IFF_UP) ||
1159             ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1160             ipv4_is_zeronet(prefix) ||
1161             (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32))
1162                 return;
1163
1164         /* add the new */
1165         fib_magic(RTM_NEWROUTE,
1166                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1167                   prefix, ifa->ifa_prefixlen, ifa, new_metric);
1168
1169         /* delete the old */
1170         fib_magic(RTM_DELROUTE,
1171                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1172                   prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1173 }
1174
1175 /* Delete primary or secondary address.
1176  * Optionally, on secondary address promotion consider the addresses
1177  * from subnet iprim as deleted, even if they are in device list.
1178  * In this case the secondary ifa can be in device list.
1179  */
1180 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1181 {
1182         struct in_device *in_dev = ifa->ifa_dev;
1183         struct net_device *dev = in_dev->dev;
1184         struct in_ifaddr *ifa1;
1185         struct in_ifaddr *prim = ifa, *prim1 = NULL;
1186         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1187         __be32 any = ifa->ifa_address & ifa->ifa_mask;
1188 #define LOCAL_OK        1
1189 #define BRD_OK          2
1190 #define BRD0_OK         4
1191 #define BRD1_OK         8
1192         unsigned int ok = 0;
1193         int subnet = 0;         /* Primary network */
1194         int gone = 1;           /* Address is missing */
1195         int same_prefsrc = 0;   /* Another primary with same IP */
1196
1197         if (ifa->ifa_flags & IFA_F_SECONDARY) {
1198                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1199                 if (!prim) {
1200                         /* if the device has been deleted, we don't perform
1201                          * address promotion
1202                          */
1203                         if (!in_dev->dead)
1204                                 pr_warn("%s: bug: prim == NULL\n", __func__);
1205                         return;
1206                 }
1207                 if (iprim && iprim != prim) {
1208                         pr_warn("%s: bug: iprim != prim\n", __func__);
1209                         return;
1210                 }
1211         } else if (!ipv4_is_zeronet(any) &&
1212                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1213                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1214                         fib_magic(RTM_DELROUTE,
1215                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1216                                   any, ifa->ifa_prefixlen, prim, 0);
1217                 subnet = 1;
1218         }
1219
1220         if (in_dev->dead)
1221                 goto no_promotions;
1222
1223         /* Deletion is more complicated than add.
1224          * We should take care of not to delete too much :-)
1225          *
1226          * Scan address list to be sure that addresses are really gone.
1227          */
1228         rcu_read_lock();
1229         in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1230                 if (ifa1 == ifa) {
1231                         /* promotion, keep the IP */
1232                         gone = 0;
1233                         continue;
1234                 }
1235                 /* Ignore IFAs from our subnet */
1236                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1237                     inet_ifa_match(ifa1->ifa_address, iprim))
1238                         continue;
1239
1240                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
1241                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1242                         /* Another address from our subnet? */
1243                         if (ifa1->ifa_mask == prim->ifa_mask &&
1244                             inet_ifa_match(ifa1->ifa_address, prim))
1245                                 prim1 = prim;
1246                         else {
1247                                 /* We reached the secondaries, so
1248                                  * same_prefsrc should be determined.
1249                                  */
1250                                 if (!same_prefsrc)
1251                                         continue;
1252                                 /* Search new prim1 if ifa1 is not
1253                                  * using the current prim1
1254                                  */
1255                                 if (!prim1 ||
1256                                     ifa1->ifa_mask != prim1->ifa_mask ||
1257                                     !inet_ifa_match(ifa1->ifa_address, prim1))
1258                                         prim1 = inet_ifa_byprefix(in_dev,
1259                                                         ifa1->ifa_address,
1260                                                         ifa1->ifa_mask);
1261                                 if (!prim1)
1262                                         continue;
1263                                 if (prim1->ifa_local != prim->ifa_local)
1264                                         continue;
1265                         }
1266                 } else {
1267                         if (prim->ifa_local != ifa1->ifa_local)
1268                                 continue;
1269                         prim1 = ifa1;
1270                         if (prim != prim1)
1271                                 same_prefsrc = 1;
1272                 }
1273                 if (ifa->ifa_local == ifa1->ifa_local)
1274                         ok |= LOCAL_OK;
1275                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1276                         ok |= BRD_OK;
1277                 if (brd == ifa1->ifa_broadcast)
1278                         ok |= BRD1_OK;
1279                 if (any == ifa1->ifa_broadcast)
1280                         ok |= BRD0_OK;
1281                 /* primary has network specific broadcasts */
1282                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1283                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1284                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1285
1286                         if (!ipv4_is_zeronet(any1)) {
1287                                 if (ifa->ifa_broadcast == brd1 ||
1288                                     ifa->ifa_broadcast == any1)
1289                                         ok |= BRD_OK;
1290                                 if (brd == brd1 || brd == any1)
1291                                         ok |= BRD1_OK;
1292                                 if (any == brd1 || any == any1)
1293                                         ok |= BRD0_OK;
1294                         }
1295                 }
1296         }
1297         rcu_read_unlock();
1298
1299 no_promotions:
1300         if (!(ok & BRD_OK))
1301                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1302                           prim, 0);
1303         if (subnet && ifa->ifa_prefixlen < 31) {
1304                 if (!(ok & BRD1_OK))
1305                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1306                                   prim, 0);
1307                 if (!(ok & BRD0_OK))
1308                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1309                                   prim, 0);
1310         }
1311         if (!(ok & LOCAL_OK)) {
1312                 unsigned int addr_type;
1313
1314                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1315
1316                 /* Check, that this local address finally disappeared. */
1317                 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1318                                                      ifa->ifa_local);
1319                 if (gone && addr_type != RTN_LOCAL) {
1320                         /* And the last, but not the least thing.
1321                          * We must flush stray FIB entries.
1322                          *
1323                          * First of all, we scan fib_info list searching
1324                          * for stray nexthop entries, then ignite fib_flush.
1325                          */
1326                         if (fib_sync_down_addr(dev, ifa->ifa_local))
1327                                 fib_flush(dev_net(dev));
1328                 }
1329         }
1330 #undef LOCAL_OK
1331 #undef BRD_OK
1332 #undef BRD0_OK
1333 #undef BRD1_OK
1334 }
1335
1336 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1337 {
1338
1339         struct fib_result       res;
1340         struct flowi4           fl4 = {
1341                 .flowi4_mark = frn->fl_mark,
1342                 .daddr = frn->fl_addr,
1343                 .flowi4_tos = frn->fl_tos,
1344                 .flowi4_scope = frn->fl_scope,
1345         };
1346         struct fib_table *tb;
1347
1348         rcu_read_lock();
1349
1350         tb = fib_get_table(net, frn->tb_id_in);
1351
1352         frn->err = -ENOENT;
1353         if (tb) {
1354                 local_bh_disable();
1355
1356                 frn->tb_id = tb->tb_id;
1357                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1358
1359                 if (!frn->err) {
1360                         frn->prefixlen = res.prefixlen;
1361                         frn->nh_sel = res.nh_sel;
1362                         frn->type = res.type;
1363                         frn->scope = res.scope;
1364                 }
1365                 local_bh_enable();
1366         }
1367
1368         rcu_read_unlock();
1369 }
1370
1371 static void nl_fib_input(struct sk_buff *skb)
1372 {
1373         struct net *net;
1374         struct fib_result_nl *frn;
1375         struct nlmsghdr *nlh;
1376         u32 portid;
1377
1378         net = sock_net(skb->sk);
1379         nlh = nlmsg_hdr(skb);
1380         if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1381             skb->len < nlh->nlmsg_len ||
1382             nlmsg_len(nlh) < sizeof(*frn))
1383                 return;
1384
1385         skb = netlink_skb_clone(skb, GFP_KERNEL);
1386         if (!skb)
1387                 return;
1388         nlh = nlmsg_hdr(skb);
1389
1390         frn = nlmsg_data(nlh);
1391         nl_fib_lookup(net, frn);
1392
1393         portid = NETLINK_CB(skb).portid;      /* netlink portid */
1394         NETLINK_CB(skb).portid = 0;        /* from kernel */
1395         NETLINK_CB(skb).dst_group = 0;  /* unicast */
1396         nlmsg_unicast(net->ipv4.fibnl, skb, portid);
1397 }
1398
1399 static int __net_init nl_fib_lookup_init(struct net *net)
1400 {
1401         struct sock *sk;
1402         struct netlink_kernel_cfg cfg = {
1403                 .input  = nl_fib_input,
1404         };
1405
1406         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1407         if (!sk)
1408                 return -EAFNOSUPPORT;
1409         net->ipv4.fibnl = sk;
1410         return 0;
1411 }
1412
1413 static void nl_fib_lookup_exit(struct net *net)
1414 {
1415         netlink_kernel_release(net->ipv4.fibnl);
1416         net->ipv4.fibnl = NULL;
1417 }
1418
1419 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1420                            bool force)
1421 {
1422         if (fib_sync_down_dev(dev, event, force))
1423                 fib_flush(dev_net(dev));
1424         else
1425                 rt_cache_flush(dev_net(dev));
1426         arp_ifdown(dev);
1427 }
1428
1429 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1430 {
1431         struct in_ifaddr *ifa = ptr;
1432         struct net_device *dev = ifa->ifa_dev->dev;
1433         struct net *net = dev_net(dev);
1434
1435         switch (event) {
1436         case NETDEV_UP:
1437                 fib_add_ifaddr(ifa);
1438 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1439                 fib_sync_up(dev, RTNH_F_DEAD);
1440 #endif
1441                 atomic_inc(&net->ipv4.dev_addr_genid);
1442                 rt_cache_flush(dev_net(dev));
1443                 break;
1444         case NETDEV_DOWN:
1445                 fib_del_ifaddr(ifa, NULL);
1446                 atomic_inc(&net->ipv4.dev_addr_genid);
1447                 if (!ifa->ifa_dev->ifa_list) {
1448                         /* Last address was deleted from this interface.
1449                          * Disable IP.
1450                          */
1451                         fib_disable_ip(dev, event, true);
1452                 } else {
1453                         rt_cache_flush(dev_net(dev));
1454                 }
1455                 break;
1456         }
1457         return NOTIFY_DONE;
1458 }
1459
1460 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1461 {
1462         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1463         struct netdev_notifier_changeupper_info *upper_info = ptr;
1464         struct netdev_notifier_info_ext *info_ext = ptr;
1465         struct in_device *in_dev;
1466         struct net *net = dev_net(dev);
1467         struct in_ifaddr *ifa;
1468         unsigned int flags;
1469
1470         if (event == NETDEV_UNREGISTER) {
1471                 fib_disable_ip(dev, event, true);
1472                 rt_flush_dev(dev);
1473                 return NOTIFY_DONE;
1474         }
1475
1476         in_dev = __in_dev_get_rtnl(dev);
1477         if (!in_dev)
1478                 return NOTIFY_DONE;
1479
1480         switch (event) {
1481         case NETDEV_UP:
1482                 in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1483                         fib_add_ifaddr(ifa);
1484                 }
1485 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1486                 fib_sync_up(dev, RTNH_F_DEAD);
1487 #endif
1488                 atomic_inc(&net->ipv4.dev_addr_genid);
1489                 rt_cache_flush(net);
1490                 break;
1491         case NETDEV_DOWN:
1492                 fib_disable_ip(dev, event, false);
1493                 break;
1494         case NETDEV_CHANGE:
1495                 flags = dev_get_flags(dev);
1496                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1497                         fib_sync_up(dev, RTNH_F_LINKDOWN);
1498                 else
1499                         fib_sync_down_dev(dev, event, false);
1500                 rt_cache_flush(net);
1501                 break;
1502         case NETDEV_CHANGEMTU:
1503                 fib_sync_mtu(dev, info_ext->ext.mtu);
1504                 rt_cache_flush(net);
1505                 break;
1506         case NETDEV_CHANGEUPPER:
1507                 upper_info = ptr;
1508                 /* flush all routes if dev is linked to or unlinked from
1509                  * an L3 master device (e.g., VRF)
1510                  */
1511                 if (upper_info->upper_dev &&
1512                     netif_is_l3_master(upper_info->upper_dev))
1513                         fib_disable_ip(dev, NETDEV_DOWN, true);
1514                 break;
1515         }
1516         return NOTIFY_DONE;
1517 }
1518
1519 static struct notifier_block fib_inetaddr_notifier = {
1520         .notifier_call = fib_inetaddr_event,
1521 };
1522
1523 static struct notifier_block fib_netdev_notifier = {
1524         .notifier_call = fib_netdev_event,
1525 };
1526
1527 static int __net_init ip_fib_net_init(struct net *net)
1528 {
1529         int err;
1530         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1531
1532         err = fib4_notifier_init(net);
1533         if (err)
1534                 return err;
1535
1536 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1537         /* Default to 3-tuple */
1538         net->ipv4.sysctl_fib_multipath_hash_fields =
1539                 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK;
1540 #endif
1541
1542         /* Avoid false sharing : Use at least a full cache line */
1543         size = max_t(size_t, size, L1_CACHE_BYTES);
1544
1545         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1546         if (!net->ipv4.fib_table_hash) {
1547                 err = -ENOMEM;
1548                 goto err_table_hash_alloc;
1549         }
1550
1551         err = fib4_rules_init(net);
1552         if (err < 0)
1553                 goto err_rules_init;
1554         return 0;
1555
1556 err_rules_init:
1557         kfree(net->ipv4.fib_table_hash);
1558 err_table_hash_alloc:
1559         fib4_notifier_exit(net);
1560         return err;
1561 }
1562
1563 static void ip_fib_net_exit(struct net *net)
1564 {
1565         int i;
1566
1567         ASSERT_RTNL();
1568 #ifdef CONFIG_IP_MULTIPLE_TABLES
1569         RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1570         RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1571 #endif
1572         /* Destroy the tables in reverse order to guarantee that the
1573          * local table, ID 255, is destroyed before the main table, ID
1574          * 254. This is necessary as the local table may contain
1575          * references to data contained in the main table.
1576          */
1577         for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1578                 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1579                 struct hlist_node *tmp;
1580                 struct fib_table *tb;
1581
1582                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1583                         hlist_del(&tb->tb_hlist);
1584                         fib_table_flush(net, tb, true);
1585                         fib_free_table(tb);
1586                 }
1587         }
1588
1589 #ifdef CONFIG_IP_MULTIPLE_TABLES
1590         fib4_rules_exit(net);
1591 #endif
1592
1593         kfree(net->ipv4.fib_table_hash);
1594         fib4_notifier_exit(net);
1595 }
1596
1597 static int __net_init fib_net_init(struct net *net)
1598 {
1599         int error;
1600
1601 #ifdef CONFIG_IP_ROUTE_CLASSID
1602         atomic_set(&net->ipv4.fib_num_tclassid_users, 0);
1603 #endif
1604         error = ip_fib_net_init(net);
1605         if (error < 0)
1606                 goto out;
1607         error = nl_fib_lookup_init(net);
1608         if (error < 0)
1609                 goto out_nlfl;
1610         error = fib_proc_init(net);
1611         if (error < 0)
1612                 goto out_proc;
1613 out:
1614         return error;
1615
1616 out_proc:
1617         nl_fib_lookup_exit(net);
1618 out_nlfl:
1619         rtnl_lock();
1620         ip_fib_net_exit(net);
1621         rtnl_unlock();
1622         goto out;
1623 }
1624
1625 static void __net_exit fib_net_exit(struct net *net)
1626 {
1627         fib_proc_exit(net);
1628         nl_fib_lookup_exit(net);
1629 }
1630
1631 static void __net_exit fib_net_exit_batch(struct list_head *net_list)
1632 {
1633         struct net *net;
1634
1635         rtnl_lock();
1636         list_for_each_entry(net, net_list, exit_list)
1637                 ip_fib_net_exit(net);
1638
1639         rtnl_unlock();
1640 }
1641
1642 static struct pernet_operations fib_net_ops = {
1643         .init = fib_net_init,
1644         .exit = fib_net_exit,
1645         .exit_batch = fib_net_exit_batch,
1646 };
1647
1648 void __init ip_fib_init(void)
1649 {
1650         fib_trie_init();
1651
1652         register_pernet_subsys(&fib_net_ops);
1653
1654         register_netdevice_notifier(&fib_netdev_notifier);
1655         register_inetaddr_notifier(&fib_inetaddr_notifier);
1656
1657         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1658         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1659         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);
1660 }