net/ipv4/fib_frontend.c

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