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