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 <asm/uaccess.h>
  18 #include <asm/system.h>
  19 #include <linux/bitops.h>
  20 #include <linux/capability.h>
  21 #include <linux/types.h>
  22 #include <linux/kernel.h>
  23 #include <linux/mm.h>
  24 #include <linux/string.h>
  25 #include <linux/socket.h>
  26 #include <linux/sockios.h>
  27 #include <linux/errno.h>
  28 #include <linux/in.h>
  29 #include <linux/inet.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/netdevice.h>
  32 #include <linux/if_addr.h>
  33 #include <linux/if_arp.h>
  34 #include <linux/skbuff.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
  49 #ifndef CONFIG_IP_MULTIPLE_TABLES
  50
  51 static int __net_init fib4_rules_init(struct net *net)
  52 {
  53         struct fib_table *local_table, *main_table;
  54
  55         local_table = fib_trie_table(RT_TABLE_LOCAL);
  56         if (local_table == NULL)
  57                 return -ENOMEM;
  58
  59         main_table  = fib_trie_table(RT_TABLE_MAIN);
  60         if (main_table == NULL)
  61                 goto fail;
  62
  63         hlist_add_head_rcu(&local_table->tb_hlist,
  64                                 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
  65         hlist_add_head_rcu(&main_table->tb_hlist,
  66                                 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
  67         return 0;
  68
  69 fail:
  70         kfree(local_table);
  71         return -ENOMEM;
  72 }
  73 #else
  74
  75 struct fib_table *fib_new_table(struct net *net, u32 id)
  76 {
  77         struct fib_table *tb;
  78         unsigned int h;
  79
  80         if (id == 0)
  81                 id = RT_TABLE_MAIN;
  82         tb = fib_get_table(net, id);
  83         if (tb)
  84                 return tb;
  85
  86         tb = fib_trie_table(id);
  87         if (!tb)
  88                 return NULL;
  89         h = id & (FIB_TABLE_HASHSZ - 1);
  90         hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
  91         return tb;
  92 }
  93
  94 struct fib_table *fib_get_table(struct net *net, u32 id)
  95 {
  96         struct fib_table *tb;
  97         struct hlist_node *node;
  98         struct hlist_head *head;
  99         unsigned int h;
 100
 101         if (id == 0)
 102                 id = RT_TABLE_MAIN;
 103         h = id & (FIB_TABLE_HASHSZ - 1);
 104
 105         rcu_read_lock();
 106         head = &net->ipv4.fib_table_hash[h];
 107         hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
 108                 if (tb->tb_id == id) {
 109                         rcu_read_unlock();
 110                         return tb;
 111                 }
 112         }
 113         rcu_read_unlock();
 114         return NULL;
 115 }
 116 #endif /* CONFIG_IP_MULTIPLE_TABLES */
 117
 118 static void fib_flush(struct net *net)
 119 {
 120         int flushed = 0;
 121         struct fib_table *tb;
 122         struct hlist_node *node;
 123         struct hlist_head *head;
 124         unsigned int h;
 125
 126         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
 127                 head = &net->ipv4.fib_table_hash[h];
 128                 hlist_for_each_entry(tb, node, head, tb_hlist)
 129                         flushed += fib_table_flush(tb);
 130         }
 131
 132         if (flushed)
 133                 rt_cache_flush(net, -1);
 134 }
 135
 136 /*
 137  * Find address type as if only "dev" was present in the system. If
 138  * on_dev is NULL then all interfaces are taken into consideration.
 139  */
 140 static inline unsigned __inet_dev_addr_type(struct net *net,
 141                                             const struct net_device *dev,
 142                                             __be32 addr)
 143 {
 144         struct flowi4           fl4 = { .daddr = addr };
 145         struct fib_result       res;
 146         unsigned ret = RTN_BROADCAST;
 147         struct fib_table *local_table;
 148
 149         if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
 150                 return RTN_BROADCAST;
 151         if (ipv4_is_multicast(addr))
 152                 return RTN_MULTICAST;
 153
 154 #ifdef CONFIG_IP_MULTIPLE_TABLES
 155         res.r = NULL;
 156 #endif
 157
 158         local_table = fib_get_table(net, RT_TABLE_LOCAL);
 159         if (local_table) {
 160                 ret = RTN_UNICAST;
 161                 rcu_read_lock();
 162                 if (!fib_table_lookup(local_table, &fl4, &res, FIB_LOOKUP_NOREF)) {
 163                         if (!dev || dev == res.fi->fib_dev)
 164                                 ret = res.type;
 165                 }
 166                 rcu_read_unlock();
 167         }
 168         return ret;
 169 }
 170
 171 unsigned int inet_addr_type(struct net *net, __be32 addr)
 172 {
 173         return __inet_dev_addr_type(net, NULL, addr);
 174 }
 175 EXPORT_SYMBOL(inet_addr_type);
 176
 177 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
 178                                 __be32 addr)
 179 {
 180         return __inet_dev_addr_type(net, dev, addr);
 181 }
 182 EXPORT_SYMBOL(inet_dev_addr_type);
 183
 184 /* Given (packet source, input interface) and optional (dst, oif, tos):
 185  * - (main) check, that source is valid i.e. not broadcast or our local
 186  *   address.
 187  * - figure out what "logical" interface this packet arrived
 188  *   and calculate "specific destination" address.
 189  * - check, that packet arrived from expected physical interface.
 190  * called with rcu_read_lock()
 191  */
 192 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, u8 tos,
 193                         int oif, struct net_device *dev, __be32 *spec_dst,
 194                         u32 *itag)
 195 {
 196         struct in_device *in_dev;
 197         struct flowi4 fl4;
 198         struct fib_result res;
 199         int no_addr, rpf, accept_local;
 200         bool dev_match;
 201         int ret;
 202         struct net *net;
 203
 204         fl4.flowi4_oif = 0;
 205         fl4.flowi4_iif = oif;
 206         fl4.daddr = src;
 207         fl4.saddr = dst;
 208         fl4.flowi4_tos = tos;
 209         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 210
 211         no_addr = rpf = accept_local = 0;
 212         in_dev = __in_dev_get_rcu(dev);
 213         if (in_dev) {
 214                 no_addr = in_dev->ifa_list == NULL;
 215
 216                 /* Ignore rp_filter for packets protected by IPsec. */
 217                 rpf = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(in_dev);
 218
 219                 accept_local = IN_DEV_ACCEPT_LOCAL(in_dev);
 220                 fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0;
 221         }
 222
 223         if (in_dev == NULL)
 224                 goto e_inval;
 225
 226         net = dev_net(dev);
 227         if (fib_lookup(net, &fl4, &res))
 228                 goto last_resort;
 229         if (res.type != RTN_UNICAST) {
 230                 if (res.type != RTN_LOCAL || !accept_local)
 231                         goto e_inval;
 232         }
 233         *spec_dst = FIB_RES_PREFSRC(net, res);
 234         fib_combine_itag(itag, &res);
 235         dev_match = false;
 236
 237 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 238         for (ret = 0; ret < res.fi->fib_nhs; ret++) {
 239                 struct fib_nh *nh = &res.fi->fib_nh[ret];
 240
 241                 if (nh->nh_dev == dev) {
 242                         dev_match = true;
 243                         break;
 244                 }
 245         }
 246 #else
 247         if (FIB_RES_DEV(res) == dev)
 248                 dev_match = true;
 249 #endif
 250         if (dev_match) {
 251                 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 252                 return ret;
 253         }
 254         if (no_addr)
 255                 goto last_resort;
 256         if (rpf == 1)
 257                 goto e_rpf;
 258         fl4.flowi4_oif = dev->ifindex;
 259
 260         ret = 0;
 261         if (fib_lookup(net, &fl4, &res) == 0) {
 262                 if (res.type == RTN_UNICAST) {
 263                         *spec_dst = FIB_RES_PREFSRC(net, res);
 264                         ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 265                 }
 266         }
 267         return ret;
 268
 269 last_resort:
 270         if (rpf)
 271                 goto e_rpf;
 272         *spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
 273         *itag = 0;
 274         return 0;
 275
 276 e_inval:
 277         return -EINVAL;
 278 e_rpf:
 279         return -EXDEV;
 280 }
 281
 282 static inline __be32 sk_extract_addr(struct sockaddr *addr)
 283 {
 284         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 285 }
 286
 287 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
 288 {
 289         struct nlattr *nla;
 290
 291         nla = (struct nlattr *) ((char *) mx + len);
 292         nla->nla_type = type;
 293         nla->nla_len = nla_attr_size(4);
 294         *(u32 *) nla_data(nla) = value;
 295
 296         return len + nla_total_size(4);
 297 }
 298
 299 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
 300                                  struct fib_config *cfg)
 301 {
 302         __be32 addr;
 303         int plen;
 304
 305         memset(cfg, 0, sizeof(*cfg));
 306         cfg->fc_nlinfo.nl_net = net;
 307
 308         if (rt->rt_dst.sa_family != AF_INET)
 309                 return -EAFNOSUPPORT;
 310
 311         /*
 312          * Check mask for validity:
 313          * a) it must be contiguous.
 314          * b) destination must have all host bits clear.
 315          * c) if application forgot to set correct family (AF_INET),
 316          *    reject request unless it is absolutely clear i.e.
 317          *    both family and mask are zero.
 318          */
 319         plen = 32;
 320         addr = sk_extract_addr(&rt->rt_dst);
 321         if (!(rt->rt_flags & RTF_HOST)) {
 322                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
 323
 324                 if (rt->rt_genmask.sa_family != AF_INET) {
 325                         if (mask || rt->rt_genmask.sa_family)
 326                                 return -EAFNOSUPPORT;
 327                 }
 328
 329                 if (bad_mask(mask, addr))
 330                         return -EINVAL;
 331
 332                 plen = inet_mask_len(mask);
 333         }
 334
 335         cfg->fc_dst_len = plen;
 336         cfg->fc_dst = addr;
 337
 338         if (cmd != SIOCDELRT) {
 339                 cfg->fc_nlflags = NLM_F_CREATE;
 340                 cfg->fc_protocol = RTPROT_BOOT;
 341         }
 342
 343         if (rt->rt_metric)
 344                 cfg->fc_priority = rt->rt_metric - 1;
 345
 346         if (rt->rt_flags & RTF_REJECT) {
 347                 cfg->fc_scope = RT_SCOPE_HOST;
 348                 cfg->fc_type = RTN_UNREACHABLE;
 349                 return 0;
 350         }
 351
 352         cfg->fc_scope = RT_SCOPE_NOWHERE;
 353         cfg->fc_type = RTN_UNICAST;
 354
 355         if (rt->rt_dev) {
 356                 char *colon;
 357                 struct net_device *dev;
 358                 char devname[IFNAMSIZ];
 359
 360                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
 361                         return -EFAULT;
 362
 363                 devname[IFNAMSIZ-1] = 0;
 364                 colon = strchr(devname, ':');
 365                 if (colon)
 366                         *colon = 0;
 367                 dev = __dev_get_by_name(net, devname);
 368                 if (!dev)
 369                         return -ENODEV;
 370                 cfg->fc_oif = dev->ifindex;
 371                 if (colon) {
 372                         struct in_ifaddr *ifa;
 373                         struct in_device *in_dev = __in_dev_get_rtnl(dev);
 374                         if (!in_dev)
 375                                 return -ENODEV;
 376                         *colon = ':';
 377                         for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
 378                                 if (strcmp(ifa->ifa_label, devname) == 0)
 379                                         break;
 380                         if (ifa == NULL)
 381                                 return -ENODEV;
 382                         cfg->fc_prefsrc = ifa->ifa_local;
 383                 }
 384         }
 385
 386         addr = sk_extract_addr(&rt->rt_gateway);
 387         if (rt->rt_gateway.sa_family == AF_INET && addr) {
 388                 cfg->fc_gw = addr;
 389                 if (rt->rt_flags & RTF_GATEWAY &&
 390                     inet_addr_type(net, addr) == RTN_UNICAST)
 391                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
 392         }
 393
 394         if (cmd == SIOCDELRT)
 395                 return 0;
 396
 397         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
 398                 return -EINVAL;
 399
 400         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 401                 cfg->fc_scope = RT_SCOPE_LINK;
 402
 403         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 404                 struct nlattr *mx;
 405                 int len = 0;
 406
 407                 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
 408                 if (mx == NULL)
 409                         return -ENOMEM;
 410
 411                 if (rt->rt_flags & RTF_MTU)
 412                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 413
 414                 if (rt->rt_flags & RTF_WINDOW)
 415                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 416
 417                 if (rt->rt_flags & RTF_IRTT)
 418                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 419
 420                 cfg->fc_mx = mx;
 421                 cfg->fc_mx_len = len;
 422         }
 423
 424         return 0;
 425 }
 426
 427 /*
 428  * Handle IP routing ioctl calls.
 429  * These are used to manipulate the routing tables
 430  */
 431 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
 432 {
 433         struct fib_config cfg;
 434         struct rtentry rt;
 435         int err;
 436
 437         switch (cmd) {
 438         case SIOCADDRT:         /* Add a route */
 439         case SIOCDELRT:         /* Delete a route */
 440                 if (!capable(CAP_NET_ADMIN))
 441                         return -EPERM;
 442
 443                 if (copy_from_user(&rt, arg, sizeof(rt)))
 444                         return -EFAULT;
 445
 446                 rtnl_lock();
 447                 err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
 448                 if (err == 0) {
 449                         struct fib_table *tb;
 450
 451                         if (cmd == SIOCDELRT) {
 452                                 tb = fib_get_table(net, cfg.fc_table);
 453                                 if (tb)
 454                                         err = fib_table_delete(tb, &cfg);
 455                                 else
 456                                         err = -ESRCH;
 457                         } else {
 458                                 tb = fib_new_table(net, cfg.fc_table);
 459                                 if (tb)
 460                                         err = fib_table_insert(tb, &cfg);
 461                                 else
 462                                         err = -ENOBUFS;
 463                         }
 464
 465                         /* allocated by rtentry_to_fib_config() */
 466                         kfree(cfg.fc_mx);
 467                 }
 468                 rtnl_unlock();
 469                 return err;
 470         }
 471         return -EINVAL;
 472 }
 473
 474 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 475         [RTA_DST]               = { .type = NLA_U32 },
 476         [RTA_SRC]               = { .type = NLA_U32 },
 477         [RTA_IIF]               = { .type = NLA_U32 },
 478         [RTA_OIF]               = { .type = NLA_U32 },
 479         [RTA_GATEWAY]           = { .type = NLA_U32 },
 480         [RTA_PRIORITY]          = { .type = NLA_U32 },
 481         [RTA_PREFSRC]           = { .type = NLA_U32 },
 482         [RTA_METRICS]           = { .type = NLA_NESTED },
 483         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 484         [RTA_FLOW]              = { .type = NLA_U32 },
 485 };
 486
 487 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 488                              struct nlmsghdr *nlh, struct fib_config *cfg)
 489 {
 490         struct nlattr *attr;
 491         int err, remaining;
 492         struct rtmsg *rtm;
 493
 494         err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
 495         if (err < 0)
 496                 goto errout;
 497
 498         memset(cfg, 0, sizeof(*cfg));
 499
 500         rtm = nlmsg_data(nlh);
 501         cfg->fc_dst_len = rtm->rtm_dst_len;
 502         cfg->fc_tos = rtm->rtm_tos;
 503         cfg->fc_table = rtm->rtm_table;
 504         cfg->fc_protocol = rtm->rtm_protocol;
 505         cfg->fc_scope = rtm->rtm_scope;
 506         cfg->fc_type = rtm->rtm_type;
 507         cfg->fc_flags = rtm->rtm_flags;
 508         cfg->fc_nlflags = nlh->nlmsg_flags;
 509
 510         cfg->fc_nlinfo.pid = NETLINK_CB(skb).pid;
 511         cfg->fc_nlinfo.nlh = nlh;
 512         cfg->fc_nlinfo.nl_net = net;
 513
 514         if (cfg->fc_type > RTN_MAX) {
 515                 err = -EINVAL;
 516                 goto errout;
 517         }
 518
 519         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 520                 switch (nla_type(attr)) {
 521                 case RTA_DST:
 522                         cfg->fc_dst = nla_get_be32(attr);
 523                         break;
 524                 case RTA_OIF:
 525                         cfg->fc_oif = nla_get_u32(attr);
 526                         break;
 527                 case RTA_GATEWAY:
 528                         cfg->fc_gw = nla_get_be32(attr);
 529                         break;
 530                 case RTA_PRIORITY:
 531                         cfg->fc_priority = nla_get_u32(attr);
 532                         break;
 533                 case RTA_PREFSRC:
 534                         cfg->fc_prefsrc = nla_get_be32(attr);
 535                         break;
 536                 case RTA_METRICS:
 537                         cfg->fc_mx = nla_data(attr);
 538                         cfg->fc_mx_len = nla_len(attr);
 539                         break;
 540                 case RTA_MULTIPATH:
 541                         cfg->fc_mp = nla_data(attr);
 542                         cfg->fc_mp_len = nla_len(attr);
 543                         break;
 544                 case RTA_FLOW:
 545                         cfg->fc_flow = nla_get_u32(attr);
 546                         break;
 547                 case RTA_TABLE:
 548                         cfg->fc_table = nla_get_u32(attr);
 549                         break;
 550                 }
 551         }
 552
 553         return 0;
 554 errout:
 555         return err;
 556 }
 557
 558 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 559 {
 560         struct net *net = sock_net(skb->sk);
 561         struct fib_config cfg;
 562         struct fib_table *tb;
 563         int err;
 564
 565         err = rtm_to_fib_config(net, skb, nlh, &cfg);
 566         if (err < 0)
 567                 goto errout;
 568
 569         tb = fib_get_table(net, cfg.fc_table);
 570         if (tb == NULL) {
 571                 err = -ESRCH;
 572                 goto errout;
 573         }
 574
 575         err = fib_table_delete(tb, &cfg);
 576 errout:
 577         return err;
 578 }
 579
 580 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 581 {
 582         struct net *net = sock_net(skb->sk);
 583         struct fib_config cfg;
 584         struct fib_table *tb;
 585         int err;
 586
 587         err = rtm_to_fib_config(net, skb, nlh, &cfg);
 588         if (err < 0)
 589                 goto errout;
 590
 591         tb = fib_new_table(net, cfg.fc_table);
 592         if (tb == NULL) {
 593                 err = -ENOBUFS;
 594                 goto errout;
 595         }
 596
 597         err = fib_table_insert(tb, &cfg);
 598 errout:
 599         return err;
 600 }
 601
 602 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 603 {
 604         struct net *net = sock_net(skb->sk);
 605         unsigned int h, s_h;
 606         unsigned int e = 0, s_e;
 607         struct fib_table *tb;
 608         struct hlist_node *node;
 609         struct hlist_head *head;
 610         int dumped = 0;
 611
 612         if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
 613             ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
 614                 return ip_rt_dump(skb, cb);
 615
 616         s_h = cb->args[0];
 617         s_e = cb->args[1];
 618
 619         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
 620                 e = 0;
 621                 head = &net->ipv4.fib_table_hash[h];
 622                 hlist_for_each_entry(tb, node, head, tb_hlist) {
 623                         if (e < s_e)
 624                                 goto next;
 625                         if (dumped)
 626                                 memset(&cb->args[2], 0, sizeof(cb->args) -
 627                                                  2 * sizeof(cb->args[0]));
 628                         if (fib_table_dump(tb, skb, cb) < 0)
 629                                 goto out;
 630                         dumped = 1;
 631 next:
 632                         e++;
 633                 }
 634         }
 635 out:
 636         cb->args[1] = e;
 637         cb->args[0] = h;
 638
 639         return skb->len;
 640 }
 641
 642 /* Prepare and feed intra-kernel routing request.
 643  * Really, it should be netlink message, but :-( netlink
 644  * can be not configured, so that we feed it directly
 645  * to fib engine. It is legal, because all events occur
 646  * only when netlink is already locked.
 647  */
 648 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
 649 {
 650         struct net *net = dev_net(ifa->ifa_dev->dev);
 651         struct fib_table *tb;
 652         struct fib_config cfg = {
 653                 .fc_protocol = RTPROT_KERNEL,
 654                 .fc_type = type,
 655                 .fc_dst = dst,
 656                 .fc_dst_len = dst_len,
 657                 .fc_prefsrc = ifa->ifa_local,
 658                 .fc_oif = ifa->ifa_dev->dev->ifindex,
 659                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
 660                 .fc_nlinfo = {
 661                         .nl_net = net,
 662                 },
 663         };
 664
 665         if (type == RTN_UNICAST)
 666                 tb = fib_new_table(net, RT_TABLE_MAIN);
 667         else
 668                 tb = fib_new_table(net, RT_TABLE_LOCAL);
 669
 670         if (tb == NULL)
 671                 return;
 672
 673         cfg.fc_table = tb->tb_id;
 674
 675         if (type != RTN_LOCAL)
 676                 cfg.fc_scope = RT_SCOPE_LINK;
 677         else
 678                 cfg.fc_scope = RT_SCOPE_HOST;
 679
 680         if (cmd == RTM_NEWROUTE)
 681                 fib_table_insert(tb, &cfg);
 682         else
 683                 fib_table_delete(tb, &cfg);
 684 }
 685
 686 void fib_add_ifaddr(struct in_ifaddr *ifa)
 687 {
 688         struct in_device *in_dev = ifa->ifa_dev;
 689         struct net_device *dev = in_dev->dev;
 690         struct in_ifaddr *prim = ifa;
 691         __be32 mask = ifa->ifa_mask;
 692         __be32 addr = ifa->ifa_local;
 693         __be32 prefix = ifa->ifa_address & mask;
 694
 695         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 696                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
 697                 if (prim == NULL) {
 698                         printk(KERN_WARNING "fib_add_ifaddr: bug: prim == NULL\n");
 699                         return;
 700                 }
 701         }
 702
 703         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
 704
 705         if (!(dev->flags & IFF_UP))
 706                 return;
 707
 708         /* Add broadcast address, if it is explicitly assigned. */
 709         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
 710                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
 711
 712         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
 713             (prefix != addr || ifa->ifa_prefixlen < 32)) {
 714                 fib_magic(RTM_NEWROUTE,
 715                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 716                           prefix, ifa->ifa_prefixlen, prim);
 717
 718                 /* Add network specific broadcasts, when it takes a sense */
 719                 if (ifa->ifa_prefixlen < 31) {
 720                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
 721                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
 722                                   32, prim);
 723                 }
 724         }
 725 }
 726
 727 /* Delete primary or secondary address.
 728  * Optionally, on secondary address promotion consider the addresses
 729  * from subnet iprim as deleted, even if they are in device list.
 730  * In this case the secondary ifa can be in device list.
 731  */
 732 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
 733 {
 734         struct in_device *in_dev = ifa->ifa_dev;
 735         struct net_device *dev = in_dev->dev;
 736         struct in_ifaddr *ifa1;
 737         struct in_ifaddr *prim = ifa, *prim1 = NULL;
 738         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
 739         __be32 any = ifa->ifa_address & ifa->ifa_mask;
 740 #define LOCAL_OK        1
 741 #define BRD_OK          2
 742 #define BRD0_OK         4
 743 #define BRD1_OK         8
 744         unsigned ok = 0;
 745         int subnet = 0;         /* Primary network */
 746         int gone = 1;           /* Address is missing */
 747         int same_prefsrc = 0;   /* Another primary with same IP */
 748
 749         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 750                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
 751                 if (prim == NULL) {
 752                         printk(KERN_WARNING "fib_del_ifaddr: bug: prim == NULL\n");
 753                         return;
 754                 }
 755                 if (iprim && iprim != prim) {
 756                         printk(KERN_WARNING "fib_del_ifaddr: bug: iprim != prim\n");
 757                         return;
 758                 }
 759         } else if (!ipv4_is_zeronet(any) &&
 760                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
 761                 fib_magic(RTM_DELROUTE,
 762                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 763                           any, ifa->ifa_prefixlen, prim);
 764                 subnet = 1;
 765         }
 766
 767         /* Deletion is more complicated than add.
 768          * We should take care of not to delete too much :-)
 769          *
 770          * Scan address list to be sure that addresses are really gone.
 771          */
 772
 773         for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
 774                 if (ifa1 == ifa) {
 775                         /* promotion, keep the IP */
 776                         gone = 0;
 777                         continue;
 778                 }
 779                 /* Ignore IFAs from our subnet */
 780                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
 781                     inet_ifa_match(ifa1->ifa_address, iprim))
 782                         continue;
 783
 784                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
 785                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
 786                         /* Another address from our subnet? */
 787                         if (ifa1->ifa_mask == prim->ifa_mask &&
 788                             inet_ifa_match(ifa1->ifa_address, prim))
 789                                 prim1 = prim;
 790                         else {
 791                                 /* We reached the secondaries, so
 792                                  * same_prefsrc should be determined.
 793                                  */
 794                                 if (!same_prefsrc)
 795                                         continue;
 796                                 /* Search new prim1 if ifa1 is not
 797                                  * using the current prim1
 798                                  */
 799                                 if (!prim1 ||
 800                                     ifa1->ifa_mask != prim1->ifa_mask ||
 801                                     !inet_ifa_match(ifa1->ifa_address, prim1))
 802                                         prim1 = inet_ifa_byprefix(in_dev,
 803                                                         ifa1->ifa_address,
 804                                                         ifa1->ifa_mask);
 805                                 if (!prim1)
 806                                         continue;
 807                                 if (prim1->ifa_local != prim->ifa_local)
 808                                         continue;
 809                         }
 810                 } else {
 811                         if (prim->ifa_local != ifa1->ifa_local)
 812                                 continue;
 813                         prim1 = ifa1;
 814                         if (prim != prim1)
 815                                 same_prefsrc = 1;
 816                 }
 817                 if (ifa->ifa_local == ifa1->ifa_local)
 818                         ok |= LOCAL_OK;
 819                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
 820                         ok |= BRD_OK;
 821                 if (brd == ifa1->ifa_broadcast)
 822                         ok |= BRD1_OK;
 823                 if (any == ifa1->ifa_broadcast)
 824                         ok |= BRD0_OK;
 825                 /* primary has network specific broadcasts */
 826                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
 827                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
 828                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
 829
 830                         if (!ipv4_is_zeronet(any1)) {
 831                                 if (ifa->ifa_broadcast == brd1 ||
 832                                     ifa->ifa_broadcast == any1)
 833                                         ok |= BRD_OK;
 834                                 if (brd == brd1 || brd == any1)
 835                                         ok |= BRD1_OK;
 836                                 if (any == brd1 || any == any1)
 837                                         ok |= BRD0_OK;
 838                         }
 839                 }
 840         }
 841
 842         if (!(ok & BRD_OK))
 843                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
 844         if (subnet && ifa->ifa_prefixlen < 31) {
 845                 if (!(ok & BRD1_OK))
 846                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
 847                 if (!(ok & BRD0_OK))
 848                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
 849         }
 850         if (!(ok & LOCAL_OK)) {
 851                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
 852
 853                 /* Check, that this local address finally disappeared. */
 854                 if (gone &&
 855                     inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) {
 856                         /* And the last, but not the least thing.
 857                          * We must flush stray FIB entries.
 858                          *
 859                          * First of all, we scan fib_info list searching
 860                          * for stray nexthop entries, then ignite fib_flush.
 861                          */
 862                         if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
 863                                 fib_flush(dev_net(dev));
 864                 }
 865         }
 866 #undef LOCAL_OK
 867 #undef BRD_OK
 868 #undef BRD0_OK
 869 #undef BRD1_OK
 870 }
 871
 872 static void nl_fib_lookup(struct fib_result_nl *frn, struct fib_table *tb)
 873 {
 874
 875         struct fib_result       res;
 876         struct flowi4           fl4 = {
 877                 .flowi4_mark = frn->fl_mark,
 878                 .daddr = frn->fl_addr,
 879                 .flowi4_tos = frn->fl_tos,
 880                 .flowi4_scope = frn->fl_scope,
 881         };
 882
 883 #ifdef CONFIG_IP_MULTIPLE_TABLES
 884         res.r = NULL;
 885 #endif
 886
 887         frn->err = -ENOENT;
 888         if (tb) {
 889                 local_bh_disable();
 890
 891                 frn->tb_id = tb->tb_id;
 892                 rcu_read_lock();
 893                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
 894
 895                 if (!frn->err) {
 896                         frn->prefixlen = res.prefixlen;
 897                         frn->nh_sel = res.nh_sel;
 898                         frn->type = res.type;
 899                         frn->scope = res.scope;
 900                 }
 901                 rcu_read_unlock();
 902                 local_bh_enable();
 903         }
 904 }
 905
 906 static void nl_fib_input(struct sk_buff *skb)
 907 {
 908         struct net *net;
 909         struct fib_result_nl *frn;
 910         struct nlmsghdr *nlh;
 911         struct fib_table *tb;
 912         u32 pid;
 913
 914         net = sock_net(skb->sk);
 915         nlh = nlmsg_hdr(skb);
 916         if (skb->len < NLMSG_SPACE(0) || skb->len < nlh->nlmsg_len ||
 917             nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*frn)))
 918                 return;
 919
 920         skb = skb_clone(skb, GFP_KERNEL);
 921         if (skb == NULL)
 922                 return;
 923         nlh = nlmsg_hdr(skb);
 924
 925         frn = (struct fib_result_nl *) NLMSG_DATA(nlh);
 926         tb = fib_get_table(net, frn->tb_id_in);
 927
 928         nl_fib_lookup(frn, tb);
 929
 930         pid = NETLINK_CB(skb).pid;      /* pid of sending process */
 931         NETLINK_CB(skb).pid = 0;        /* from kernel */
 932         NETLINK_CB(skb).dst_group = 0;  /* unicast */
 933         netlink_unicast(net->ipv4.fibnl, skb, pid, MSG_DONTWAIT);
 934 }
 935
 936 static int __net_init nl_fib_lookup_init(struct net *net)
 937 {
 938         struct sock *sk;
 939         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, 0,
 940                                    nl_fib_input, NULL, THIS_MODULE);
 941         if (sk == NULL)
 942                 return -EAFNOSUPPORT;
 943         net->ipv4.fibnl = sk;
 944         return 0;
 945 }
 946
 947 static void nl_fib_lookup_exit(struct net *net)
 948 {
 949         netlink_kernel_release(net->ipv4.fibnl);
 950         net->ipv4.fibnl = NULL;
 951 }
 952
 953 static void fib_disable_ip(struct net_device *dev, int force, int delay)
 954 {
 955         if (fib_sync_down_dev(dev, force))
 956                 fib_flush(dev_net(dev));
 957         rt_cache_flush(dev_net(dev), delay);
 958         arp_ifdown(dev);
 959 }
 960
 961 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
 962 {
 963         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
 964         struct net_device *dev = ifa->ifa_dev->dev;
 965         struct net *net = dev_net(dev);
 966
 967         switch (event) {
 968         case NETDEV_UP:
 969                 fib_add_ifaddr(ifa);
 970 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 971                 fib_sync_up(dev);
 972 #endif
 973                 atomic_inc(&net->ipv4.dev_addr_genid);
 974                 rt_cache_flush(dev_net(dev), -1);
 975                 break;
 976         case NETDEV_DOWN:
 977                 fib_del_ifaddr(ifa, NULL);
 978                 atomic_inc(&net->ipv4.dev_addr_genid);
 979                 if (ifa->ifa_dev->ifa_list == NULL) {
 980                         /* Last address was deleted from this interface.
 981                          * Disable IP.
 982                          */
 983                         fib_disable_ip(dev, 1, 0);
 984                 } else {
 985                         rt_cache_flush(dev_net(dev), -1);
 986                 }
 987                 break;
 988         }
 989         return NOTIFY_DONE;
 990 }
 991
 992 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
 993 {
 994         struct net_device *dev = ptr;
 995         struct in_device *in_dev = __in_dev_get_rtnl(dev);
 996         struct net *net = dev_net(dev);
 997
 998         if (event == NETDEV_UNREGISTER) {
 999                 fib_disable_ip(dev, 2, -1);
1000                 return NOTIFY_DONE;
1001         }
1002
1003         if (!in_dev)
1004                 return NOTIFY_DONE;
1005
1006         switch (event) {
1007         case NETDEV_UP:
1008                 for_ifa(in_dev) {
1009                         fib_add_ifaddr(ifa);
1010                 } endfor_ifa(in_dev);
1011 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1012                 fib_sync_up(dev);
1013 #endif
1014                 atomic_inc(&net->ipv4.dev_addr_genid);
1015                 rt_cache_flush(dev_net(dev), -1);
1016                 break;
1017         case NETDEV_DOWN:
1018                 fib_disable_ip(dev, 0, 0);
1019                 break;
1020         case NETDEV_CHANGEMTU:
1021         case NETDEV_CHANGE:
1022                 rt_cache_flush(dev_net(dev), 0);
1023                 break;
1024         case NETDEV_UNREGISTER_BATCH:
1025                 /* The batch unregister is only called on the first
1026                  * device in the list of devices being unregistered.
1027                  * Therefore we should not pass dev_net(dev) in here.
1028                  */
1029                 rt_cache_flush_batch(NULL);
1030                 break;
1031         }
1032         return NOTIFY_DONE;
1033 }
1034
1035 static struct notifier_block fib_inetaddr_notifier = {
1036         .notifier_call = fib_inetaddr_event,
1037 };
1038
1039 static struct notifier_block fib_netdev_notifier = {
1040         .notifier_call = fib_netdev_event,
1041 };
1042
1043 static int __net_init ip_fib_net_init(struct net *net)
1044 {
1045         int err;
1046         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1047
1048         /* Avoid false sharing : Use at least a full cache line */
1049         size = max_t(size_t, size, L1_CACHE_BYTES);
1050
1051         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1052         if (net->ipv4.fib_table_hash == NULL)
1053                 return -ENOMEM;
1054
1055         err = fib4_rules_init(net);
1056         if (err < 0)
1057                 goto fail;
1058         return 0;
1059
1060 fail:
1061         kfree(net->ipv4.fib_table_hash);
1062         return err;
1063 }
1064
1065 static void ip_fib_net_exit(struct net *net)
1066 {
1067         unsigned int i;
1068
1069 #ifdef CONFIG_IP_MULTIPLE_TABLES
1070         fib4_rules_exit(net);
1071 #endif
1072
1073         rtnl_lock();
1074         for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1075                 struct fib_table *tb;
1076                 struct hlist_head *head;
1077                 struct hlist_node *node, *tmp;
1078
1079                 head = &net->ipv4.fib_table_hash[i];
1080                 hlist_for_each_entry_safe(tb, node, tmp, head, tb_hlist) {
1081                         hlist_del(node);
1082                         fib_table_flush(tb);
1083                         fib_free_table(tb);
1084                 }
1085         }
1086         rtnl_unlock();
1087         kfree(net->ipv4.fib_table_hash);
1088 }
1089
1090 static int __net_init fib_net_init(struct net *net)
1091 {
1092         int error;
1093
1094         error = ip_fib_net_init(net);
1095         if (error < 0)
1096                 goto out;
1097         error = nl_fib_lookup_init(net);
1098         if (error < 0)
1099                 goto out_nlfl;
1100         error = fib_proc_init(net);
1101         if (error < 0)
1102                 goto out_proc;
1103 out:
1104         return error;
1105
1106 out_proc:
1107         nl_fib_lookup_exit(net);
1108 out_nlfl:
1109         ip_fib_net_exit(net);
1110         goto out;
1111 }
1112
1113 static void __net_exit fib_net_exit(struct net *net)
1114 {
1115         fib_proc_exit(net);
1116         nl_fib_lookup_exit(net);
1117         ip_fib_net_exit(net);
1118 }
1119
1120 static struct pernet_operations fib_net_ops = {
1121         .init = fib_net_init,
1122         .exit = fib_net_exit,
1123 };
1124
1125 void __init ip_fib_init(void)
1126 {
1127         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1128         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1129         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1130
1131         register_pernet_subsys(&fib_net_ops);
1132         register_netdevice_notifier(&fib_netdev_notifier);
1133         register_inetaddr_notifier(&fib_inetaddr_notifier);
1134
1135         fib_trie_init();
1136 }