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