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 <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/init.h>
  35 #include <linux/list.h>
  36 #include <linux/slab.h>
  37
  38 #include <net/ip.h>
  39 #include <net/protocol.h>
  40 #include <net/route.h>
  41 #include <net/tcp.h>
  42 #include <net/sock.h>
  43 #include <net/arp.h>
  44 #include <net/ip_fib.h>
  45 #include <net/rtnetlink.h>
  46 #include <net/xfrm.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 int __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 int 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(struct sk_buff *skb, __be32 src, __be32 dst, u8 tos,
 192                         int oif, struct net_device *dev, __be32 *spec_dst,
 193                         u32 *itag)
 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.daddr = src;
 206         fl4.saddr = dst;
 207         fl4.flowi4_tos = tos;
 208         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 209
 210         no_addr = rpf = accept_local = 0;
 211         in_dev = __in_dev_get_rcu(dev);
 212         if (in_dev) {
 213                 no_addr = in_dev->ifa_list == NULL;
 214
 215                 /* Ignore rp_filter for packets protected by IPsec. */
 216                 rpf = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(in_dev);
 217
 218                 accept_local = IN_DEV_ACCEPT_LOCAL(in_dev);
 219                 fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0;
 220         }
 221
 222         if (in_dev == NULL)
 223                 goto e_inval;
 224
 225         net = dev_net(dev);
 226         if (fib_lookup(net, &fl4, &res))
 227                 goto last_resort;
 228         if (res.type != RTN_UNICAST) {
 229                 if (res.type != RTN_LOCAL || !accept_local)
 230                         goto e_inval;
 231         }
 232         *spec_dst = FIB_RES_PREFSRC(net, res);
 233         fib_combine_itag(itag, &res);
 234         dev_match = false;
 235
 236 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 237         for (ret = 0; ret < res.fi->fib_nhs; ret++) {
 238                 struct fib_nh *nh = &res.fi->fib_nh[ret];
 239
 240                 if (nh->nh_dev == dev) {
 241                         dev_match = true;
 242                         break;
 243                 }
 244         }
 245 #else
 246         if (FIB_RES_DEV(res) == dev)
 247                 dev_match = true;
 248 #endif
 249         if (dev_match) {
 250                 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 251                 return ret;
 252         }
 253         if (no_addr)
 254                 goto last_resort;
 255         if (rpf == 1)
 256                 goto e_rpf;
 257         fl4.flowi4_oif = dev->ifindex;
 258
 259         ret = 0;
 260         if (fib_lookup(net, &fl4, &res) == 0) {
 261                 if (res.type == RTN_UNICAST) {
 262                         *spec_dst = FIB_RES_PREFSRC(net, res);
 263                         ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 264                 }
 265         }
 266         return ret;
 267
 268 last_resort:
 269         if (rpf)
 270                 goto e_rpf;
 271         *spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
 272         *itag = 0;
 273         return 0;
 274
 275 e_inval:
 276         return -EINVAL;
 277 e_rpf:
 278         return -EXDEV;
 279 }
 280
 281 static inline __be32 sk_extract_addr(struct sockaddr *addr)
 282 {
 283         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 284 }
 285
 286 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
 287 {
 288         struct nlattr *nla;
 289
 290         nla = (struct nlattr *) ((char *) mx + len);
 291         nla->nla_type = type;
 292         nla->nla_len = nla_attr_size(4);
 293         *(u32 *) nla_data(nla) = value;
 294
 295         return len + nla_total_size(4);
 296 }
 297
 298 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
 299                                  struct fib_config *cfg)
 300 {
 301         __be32 addr;
 302         int plen;
 303
 304         memset(cfg, 0, sizeof(*cfg));
 305         cfg->fc_nlinfo.nl_net = net;
 306
 307         if (rt->rt_dst.sa_family != AF_INET)
 308                 return -EAFNOSUPPORT;
 309
 310         /*
 311          * Check mask for validity:
 312          * a) it must be contiguous.
 313          * b) destination must have all host bits clear.
 314          * c) if application forgot to set correct family (AF_INET),
 315          *    reject request unless it is absolutely clear i.e.
 316          *    both family and mask are zero.
 317          */
 318         plen = 32;
 319         addr = sk_extract_addr(&rt->rt_dst);
 320         if (!(rt->rt_flags & RTF_HOST)) {
 321                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
 322
 323                 if (rt->rt_genmask.sa_family != AF_INET) {
 324                         if (mask || rt->rt_genmask.sa_family)
 325                                 return -EAFNOSUPPORT;
 326                 }
 327
 328                 if (bad_mask(mask, addr))
 329                         return -EINVAL;
 330
 331                 plen = inet_mask_len(mask);
 332         }
 333
 334         cfg->fc_dst_len = plen;
 335         cfg->fc_dst = addr;
 336
 337         if (cmd != SIOCDELRT) {
 338                 cfg->fc_nlflags = NLM_F_CREATE;
 339                 cfg->fc_protocol = RTPROT_BOOT;
 340         }
 341
 342         if (rt->rt_metric)
 343                 cfg->fc_priority = rt->rt_metric - 1;
 344
 345         if (rt->rt_flags & RTF_REJECT) {
 346                 cfg->fc_scope = RT_SCOPE_HOST;
 347                 cfg->fc_type = RTN_UNREACHABLE;
 348                 return 0;
 349         }
 350
 351         cfg->fc_scope = RT_SCOPE_NOWHERE;
 352         cfg->fc_type = RTN_UNICAST;
 353
 354         if (rt->rt_dev) {
 355                 char *colon;
 356                 struct net_device *dev;
 357                 char devname[IFNAMSIZ];
 358
 359                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
 360                         return -EFAULT;
 361
 362                 devname[IFNAMSIZ-1] = 0;
 363                 colon = strchr(devname, ':');
 364                 if (colon)
 365                         *colon = 0;
 366                 dev = __dev_get_by_name(net, devname);
 367                 if (!dev)
 368                         return -ENODEV;
 369                 cfg->fc_oif = dev->ifindex;
 370                 if (colon) {
 371                         struct in_ifaddr *ifa;
 372                         struct in_device *in_dev = __in_dev_get_rtnl(dev);
 373                         if (!in_dev)
 374                                 return -ENODEV;
 375                         *colon = ':';
 376                         for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
 377                                 if (strcmp(ifa->ifa_label, devname) == 0)
 378                                         break;
 379                         if (ifa == NULL)
 380                                 return -ENODEV;
 381                         cfg->fc_prefsrc = ifa->ifa_local;
 382                 }
 383         }
 384
 385         addr = sk_extract_addr(&rt->rt_gateway);
 386         if (rt->rt_gateway.sa_family == AF_INET && addr) {
 387                 cfg->fc_gw = addr;
 388                 if (rt->rt_flags & RTF_GATEWAY &&
 389                     inet_addr_type(net, addr) == RTN_UNICAST)
 390                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
 391         }
 392
 393         if (cmd == SIOCDELRT)
 394                 return 0;
 395
 396         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
 397                 return -EINVAL;
 398
 399         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 400                 cfg->fc_scope = RT_SCOPE_LINK;
 401
 402         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 403                 struct nlattr *mx;
 404                 int len = 0;
 405
 406                 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
 407                 if (mx == NULL)
 408                         return -ENOMEM;
 409
 410                 if (rt->rt_flags & RTF_MTU)
 411                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 412
 413                 if (rt->rt_flags & RTF_WINDOW)
 414                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 415
 416                 if (rt->rt_flags & RTF_IRTT)
 417                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 418
 419                 cfg->fc_mx = mx;
 420                 cfg->fc_mx_len = len;
 421         }
 422
 423         return 0;
 424 }
 425
 426 /*
 427  * Handle IP routing ioctl calls.
 428  * These are used to manipulate the routing tables
 429  */
 430 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
 431 {
 432         struct fib_config cfg;
 433         struct rtentry rt;
 434         int err;
 435
 436         switch (cmd) {
 437         case SIOCADDRT:         /* Add a route */
 438         case SIOCDELRT:         /* Delete a route */
 439                 if (!capable(CAP_NET_ADMIN))
 440                         return -EPERM;
 441
 442                 if (copy_from_user(&rt, arg, sizeof(rt)))
 443                         return -EFAULT;
 444
 445                 rtnl_lock();
 446                 err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
 447                 if (err == 0) {
 448                         struct fib_table *tb;
 449
 450                         if (cmd == SIOCDELRT) {
 451                                 tb = fib_get_table(net, cfg.fc_table);
 452                                 if (tb)
 453                                         err = fib_table_delete(tb, &cfg);
 454                                 else
 455                                         err = -ESRCH;
 456                         } else {
 457                                 tb = fib_new_table(net, cfg.fc_table);
 458                                 if (tb)
 459                                         err = fib_table_insert(tb, &cfg);
 460                                 else
 461                                         err = -ENOBUFS;
 462                         }
 463
 464                         /* allocated by rtentry_to_fib_config() */
 465                         kfree(cfg.fc_mx);
 466                 }
 467                 rtnl_unlock();
 468                 return err;
 469         }
 470         return -EINVAL;
 471 }
 472
 473 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 474         [RTA_DST]               = { .type = NLA_U32 },
 475         [RTA_SRC]               = { .type = NLA_U32 },
 476         [RTA_IIF]               = { .type = NLA_U32 },
 477         [RTA_OIF]               = { .type = NLA_U32 },
 478         [RTA_GATEWAY]           = { .type = NLA_U32 },
 479         [RTA_PRIORITY]          = { .type = NLA_U32 },
 480         [RTA_PREFSRC]           = { .type = NLA_U32 },
 481         [RTA_METRICS]           = { .type = NLA_NESTED },
 482         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 483         [RTA_FLOW]              = { .type = NLA_U32 },
 484 };
 485
 486 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 487                              struct nlmsghdr *nlh, struct fib_config *cfg)
 488 {
 489         struct nlattr *attr;
 490         int err, remaining;
 491         struct rtmsg *rtm;
 492
 493         err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
 494         if (err < 0)
 495                 goto errout;
 496
 497         memset(cfg, 0, sizeof(*cfg));
 498
 499         rtm = nlmsg_data(nlh);
 500         cfg->fc_dst_len = rtm->rtm_dst_len;
 501         cfg->fc_tos = rtm->rtm_tos;
 502         cfg->fc_table = rtm->rtm_table;
 503         cfg->fc_protocol = rtm->rtm_protocol;
 504         cfg->fc_scope = rtm->rtm_scope;
 505         cfg->fc_type = rtm->rtm_type;
 506         cfg->fc_flags = rtm->rtm_flags;
 507         cfg->fc_nlflags = nlh->nlmsg_flags;
 508
 509         cfg->fc_nlinfo.pid = NETLINK_CB(skb).pid;
 510         cfg->fc_nlinfo.nlh = nlh;
 511         cfg->fc_nlinfo.nl_net = net;
 512
 513         if (cfg->fc_type > RTN_MAX) {
 514                 err = -EINVAL;
 515                 goto errout;
 516         }
 517
 518         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 519                 switch (nla_type(attr)) {
 520                 case RTA_DST:
 521                         cfg->fc_dst = nla_get_be32(attr);
 522                         break;
 523                 case RTA_OIF:
 524                         cfg->fc_oif = nla_get_u32(attr);
 525                         break;
 526                 case RTA_GATEWAY:
 527                         cfg->fc_gw = nla_get_be32(attr);
 528                         break;
 529                 case RTA_PRIORITY:
 530                         cfg->fc_priority = nla_get_u32(attr);
 531                         break;
 532                 case RTA_PREFSRC:
 533                         cfg->fc_prefsrc = nla_get_be32(attr);
 534                         break;
 535                 case RTA_METRICS:
 536                         cfg->fc_mx = nla_data(attr);
 537                         cfg->fc_mx_len = nla_len(attr);
 538                         break;
 539                 case RTA_MULTIPATH:
 540                         cfg->fc_mp = nla_data(attr);
 541                         cfg->fc_mp_len = nla_len(attr);
 542                         break;
 543                 case RTA_FLOW:
 544                         cfg->fc_flow = nla_get_u32(attr);
 545                         break;
 546                 case RTA_TABLE:
 547                         cfg->fc_table = nla_get_u32(attr);
 548                         break;
 549                 }
 550         }
 551
 552         return 0;
 553 errout:
 554         return err;
 555 }
 556
 557 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 558 {
 559         struct net *net = sock_net(skb->sk);
 560         struct fib_config cfg;
 561         struct fib_table *tb;
 562         int err;
 563
 564         err = rtm_to_fib_config(net, skb, nlh, &cfg);
 565         if (err < 0)
 566                 goto errout;
 567
 568         tb = fib_get_table(net, cfg.fc_table);
 569         if (tb == NULL) {
 570                 err = -ESRCH;
 571                 goto errout;
 572         }
 573
 574         err = fib_table_delete(tb, &cfg);
 575 errout:
 576         return err;
 577 }
 578
 579 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 580 {
 581         struct net *net = sock_net(skb->sk);
 582         struct fib_config cfg;
 583         struct fib_table *tb;
 584         int err;
 585
 586         err = rtm_to_fib_config(net, skb, nlh, &cfg);
 587         if (err < 0)
 588                 goto errout;
 589
 590         tb = fib_new_table(net, cfg.fc_table);
 591         if (tb == NULL) {
 592                 err = -ENOBUFS;
 593                 goto errout;
 594         }
 595
 596         err = fib_table_insert(tb, &cfg);
 597 errout:
 598         return err;
 599 }
 600
 601 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 602 {
 603         struct net *net = sock_net(skb->sk);
 604         unsigned int h, s_h;
 605         unsigned int e = 0, s_e;
 606         struct fib_table *tb;
 607         struct hlist_node *node;
 608         struct hlist_head *head;
 609         int dumped = 0;
 610
 611         if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
 612             ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
 613                 return ip_rt_dump(skb, cb);
 614
 615         s_h = cb->args[0];
 616         s_e = cb->args[1];
 617
 618         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
 619                 e = 0;
 620                 head = &net->ipv4.fib_table_hash[h];
 621                 hlist_for_each_entry(tb, node, head, tb_hlist) {
 622                         if (e < s_e)
 623                                 goto next;
 624                         if (dumped)
 625                                 memset(&cb->args[2], 0, sizeof(cb->args) -
 626                                                  2 * sizeof(cb->args[0]));
 627                         if (fib_table_dump(tb, skb, cb) < 0)
 628                                 goto out;
 629                         dumped = 1;
 630 next:
 631                         e++;
 632                 }
 633         }
 634 out:
 635         cb->args[1] = e;
 636         cb->args[0] = h;
 637
 638         return skb->len;
 639 }
 640
 641 /* Prepare and feed intra-kernel routing request.
 642  * Really, it should be netlink message, but :-( netlink
 643  * can be not configured, so that we feed it directly
 644  * to fib engine. It is legal, because all events occur
 645  * only when netlink is already locked.
 646  */
 647 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
 648 {
 649         struct net *net = dev_net(ifa->ifa_dev->dev);
 650         struct fib_table *tb;
 651         struct fib_config cfg = {
 652                 .fc_protocol = RTPROT_KERNEL,
 653                 .fc_type = type,
 654                 .fc_dst = dst,
 655                 .fc_dst_len = dst_len,
 656                 .fc_prefsrc = ifa->ifa_local,
 657                 .fc_oif = ifa->ifa_dev->dev->ifindex,
 658                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
 659                 .fc_nlinfo = {
 660                         .nl_net = net,
 661                 },
 662         };
 663
 664         if (type == RTN_UNICAST)
 665                 tb = fib_new_table(net, RT_TABLE_MAIN);
 666         else
 667                 tb = fib_new_table(net, RT_TABLE_LOCAL);
 668
 669         if (tb == NULL)
 670                 return;
 671
 672         cfg.fc_table = tb->tb_id;
 673
 674         if (type != RTN_LOCAL)
 675                 cfg.fc_scope = RT_SCOPE_LINK;
 676         else
 677                 cfg.fc_scope = RT_SCOPE_HOST;
 678
 679         if (cmd == RTM_NEWROUTE)
 680                 fib_table_insert(tb, &cfg);
 681         else
 682                 fib_table_delete(tb, &cfg);
 683 }
 684
 685 void fib_add_ifaddr(struct in_ifaddr *ifa)
 686 {
 687         struct in_device *in_dev = ifa->ifa_dev;
 688         struct net_device *dev = in_dev->dev;
 689         struct in_ifaddr *prim = ifa;
 690         __be32 mask = ifa->ifa_mask;
 691         __be32 addr = ifa->ifa_local;
 692         __be32 prefix = ifa->ifa_address & mask;
 693
 694         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 695                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
 696                 if (prim == NULL) {
 697                         pr_warn("%s: bug: prim == NULL\n", __func__);
 698                         return;
 699                 }
 700         }
 701
 702         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
 703
 704         if (!(dev->flags & IFF_UP))
 705                 return;
 706
 707         /* Add broadcast address, if it is explicitly assigned. */
 708         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
 709                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
 710
 711         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
 712             (prefix != addr || ifa->ifa_prefixlen < 32)) {
 713                 fib_magic(RTM_NEWROUTE,
 714                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 715                           prefix, ifa->ifa_prefixlen, prim);
 716
 717                 /* Add network specific broadcasts, when it takes a sense */
 718                 if (ifa->ifa_prefixlen < 31) {
 719                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
 720                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
 721                                   32, prim);
 722                 }
 723         }
 724 }
 725
 726 /* Delete primary or secondary address.
 727  * Optionally, on secondary address promotion consider the addresses
 728  * from subnet iprim as deleted, even if they are in device list.
 729  * In this case the secondary ifa can be in device list.
 730  */
 731 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
 732 {
 733         struct in_device *in_dev = ifa->ifa_dev;
 734         struct net_device *dev = in_dev->dev;
 735         struct in_ifaddr *ifa1;
 736         struct in_ifaddr *prim = ifa, *prim1 = NULL;
 737         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
 738         __be32 any = ifa->ifa_address & ifa->ifa_mask;
 739 #define LOCAL_OK        1
 740 #define BRD_OK          2
 741 #define BRD0_OK         4
 742 #define BRD1_OK         8
 743         unsigned int ok = 0;
 744         int subnet = 0;         /* Primary network */
 745         int gone = 1;           /* Address is missing */
 746         int same_prefsrc = 0;   /* Another primary with same IP */
 747
 748         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 749                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
 750                 if (prim == NULL) {
 751                         pr_warn("%s: bug: prim == NULL\n", __func__);
 752                         return;
 753                 }
 754                 if (iprim && iprim != prim) {
 755                         pr_warn("%s: bug: iprim != prim\n", __func__);
 756                         return;
 757                 }
 758         } else if (!ipv4_is_zeronet(any) &&
 759                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
 760                 fib_magic(RTM_DELROUTE,
 761                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 762                           any, ifa->ifa_prefixlen, prim);
 763                 subnet = 1;
 764         }
 765
 766         /* Deletion is more complicated than add.
 767          * We should take care of not to delete too much :-)
 768          *
 769          * Scan address list to be sure that addresses are really gone.
 770          */
 771
 772         for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
 773                 if (ifa1 == ifa) {
 774                         /* promotion, keep the IP */
 775                         gone = 0;
 776                         continue;
 777                 }
 778                 /* Ignore IFAs from our subnet */
 779                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
 780                     inet_ifa_match(ifa1->ifa_address, iprim))
 781                         continue;
 782
 783                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
 784                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
 785                         /* Another address from our subnet? */
 786                         if (ifa1->ifa_mask == prim->ifa_mask &&
 787                             inet_ifa_match(ifa1->ifa_address, prim))
 788                                 prim1 = prim;
 789                         else {
 790                                 /* We reached the secondaries, so
 791                                  * same_prefsrc should be determined.
 792                                  */
 793                                 if (!same_prefsrc)
 794                                         continue;
 795                                 /* Search new prim1 if ifa1 is not
 796                                  * using the current prim1
 797                                  */
 798                                 if (!prim1 ||
 799                                     ifa1->ifa_mask != prim1->ifa_mask ||
 800                                     !inet_ifa_match(ifa1->ifa_address, prim1))
 801                                         prim1 = inet_ifa_byprefix(in_dev,
 802                                                         ifa1->ifa_address,
 803                                                         ifa1->ifa_mask);
 804                                 if (!prim1)
 805                                         continue;
 806                                 if (prim1->ifa_local != prim->ifa_local)
 807                                         continue;
 808                         }
 809                 } else {
 810                         if (prim->ifa_local != ifa1->ifa_local)
 811                                 continue;
 812                         prim1 = ifa1;
 813                         if (prim != prim1)
 814                                 same_prefsrc = 1;
 815                 }
 816                 if (ifa->ifa_local == ifa1->ifa_local)
 817                         ok |= LOCAL_OK;
 818                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
 819                         ok |= BRD_OK;
 820                 if (brd == ifa1->ifa_broadcast)
 821                         ok |= BRD1_OK;
 822                 if (any == ifa1->ifa_broadcast)
 823                         ok |= BRD0_OK;
 824                 /* primary has network specific broadcasts */
 825                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
 826                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
 827                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
 828
 829                         if (!ipv4_is_zeronet(any1)) {
 830                                 if (ifa->ifa_broadcast == brd1 ||
 831                                     ifa->ifa_broadcast == any1)
 832                                         ok |= BRD_OK;
 833                                 if (brd == brd1 || brd == any1)
 834                                         ok |= BRD1_OK;
 835                                 if (any == brd1 || any == any1)
 836                                         ok |= BRD0_OK;
 837                         }
 838                 }
 839         }
 840
 841         if (!(ok & BRD_OK))
 842                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
 843         if (subnet && ifa->ifa_prefixlen < 31) {
 844                 if (!(ok & BRD1_OK))
 845                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
 846                 if (!(ok & BRD0_OK))
 847                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
 848         }
 849         if (!(ok & LOCAL_OK)) {
 850                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
 851
 852                 /* Check, that this local address finally disappeared. */
 853                 if (gone &&
 854                     inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) {
 855                         /* And the last, but not the least thing.
 856                          * We must flush stray FIB entries.
 857                          *
 858                          * First of all, we scan fib_info list searching
 859                          * for stray nexthop entries, then ignite fib_flush.
 860                          */
 861                         if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
 862                                 fib_flush(dev_net(dev));
 863                 }
 864         }
 865 #undef LOCAL_OK
 866 #undef BRD_OK
 867 #undef BRD0_OK
 868 #undef BRD1_OK
 869 }
 870
 871 static void nl_fib_lookup(struct fib_result_nl *frn, struct fib_table *tb)
 872 {
 873
 874         struct fib_result       res;
 875         struct flowi4           fl4 = {
 876                 .flowi4_mark = frn->fl_mark,
 877                 .daddr = frn->fl_addr,
 878                 .flowi4_tos = frn->fl_tos,
 879                 .flowi4_scope = frn->fl_scope,
 880         };
 881
 882 #ifdef CONFIG_IP_MULTIPLE_TABLES
 883         res.r = NULL;
 884 #endif
 885
 886         frn->err = -ENOENT;
 887         if (tb) {
 888                 local_bh_disable();
 889
 890                 frn->tb_id = tb->tb_id;
 891                 rcu_read_lock();
 892                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
 893
 894                 if (!frn->err) {
 895                         frn->prefixlen = res.prefixlen;
 896                         frn->nh_sel = res.nh_sel;
 897                         frn->type = res.type;
 898                         frn->scope = res.scope;
 899                 }
 900                 rcu_read_unlock();
 901                 local_bh_enable();
 902         }
 903 }
 904
 905 static void nl_fib_input(struct sk_buff *skb)
 906 {
 907         struct net *net;
 908         struct fib_result_nl *frn;
 909         struct nlmsghdr *nlh;
 910         struct fib_table *tb;
 911         u32 pid;
 912
 913         net = sock_net(skb->sk);
 914         nlh = nlmsg_hdr(skb);
 915         if (skb->len < NLMSG_SPACE(0) || skb->len < nlh->nlmsg_len ||
 916             nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*frn)))
 917                 return;
 918
 919         skb = skb_clone(skb, GFP_KERNEL);
 920         if (skb == NULL)
 921                 return;
 922         nlh = nlmsg_hdr(skb);
 923
 924         frn = (struct fib_result_nl *) NLMSG_DATA(nlh);
 925         tb = fib_get_table(net, frn->tb_id_in);
 926
 927         nl_fib_lookup(frn, tb);
 928
 929         pid = NETLINK_CB(skb).pid;      /* pid of sending process */
 930         NETLINK_CB(skb).pid = 0;        /* from kernel */
 931         NETLINK_CB(skb).dst_group = 0;  /* unicast */
 932         netlink_unicast(net->ipv4.fibnl, skb, pid, MSG_DONTWAIT);
 933 }
 934
 935 static int __net_init nl_fib_lookup_init(struct net *net)
 936 {
 937         struct sock *sk;
 938         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, 0,
 939                                    nl_fib_input, NULL, THIS_MODULE);
 940         if (sk == NULL)
 941                 return -EAFNOSUPPORT;
 942         net->ipv4.fibnl = sk;
 943         return 0;
 944 }
 945
 946 static void nl_fib_lookup_exit(struct net *net)
 947 {
 948         netlink_kernel_release(net->ipv4.fibnl);
 949         net->ipv4.fibnl = NULL;
 950 }
 951
 952 static void fib_disable_ip(struct net_device *dev, int force, int delay)
 953 {
 954         if (fib_sync_down_dev(dev, force))
 955                 fib_flush(dev_net(dev));
 956         rt_cache_flush(dev_net(dev), delay);
 957         arp_ifdown(dev);
 958 }
 959
 960 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
 961 {
 962         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
 963         struct net_device *dev = ifa->ifa_dev->dev;
 964         struct net *net = dev_net(dev);
 965
 966         switch (event) {
 967         case NETDEV_UP:
 968                 fib_add_ifaddr(ifa);
 969 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 970                 fib_sync_up(dev);
 971 #endif
 972                 atomic_inc(&net->ipv4.dev_addr_genid);
 973                 rt_cache_flush(dev_net(dev), -1);
 974                 break;
 975         case NETDEV_DOWN:
 976                 fib_del_ifaddr(ifa, NULL);
 977                 atomic_inc(&net->ipv4.dev_addr_genid);
 978                 if (ifa->ifa_dev->ifa_list == NULL) {
 979                         /* Last address was deleted from this interface.
 980                          * Disable IP.
 981                          */
 982                         fib_disable_ip(dev, 1, 0);
 983                 } else {
 984                         rt_cache_flush(dev_net(dev), -1);
 985                 }
 986                 break;
 987         }
 988         return NOTIFY_DONE;
 989 }
 990
 991 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
 992 {
 993         struct net_device *dev = ptr;
 994         struct in_device *in_dev = __in_dev_get_rtnl(dev);
 995         struct net *net = dev_net(dev);
 996
 997         if (event == NETDEV_UNREGISTER) {
 998                 fib_disable_ip(dev, 2, -1);
 999                 return NOTIFY_DONE;
1000         }
1001
1002         if (!in_dev)
1003                 return NOTIFY_DONE;
1004
1005         switch (event) {
1006         case NETDEV_UP:
1007                 for_ifa(in_dev) {
1008                         fib_add_ifaddr(ifa);
1009                 } endfor_ifa(in_dev);
1010 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1011                 fib_sync_up(dev);
1012 #endif
1013                 atomic_inc(&net->ipv4.dev_addr_genid);
1014                 rt_cache_flush(dev_net(dev), -1);
1015                 break;
1016         case NETDEV_DOWN:
1017                 fib_disable_ip(dev, 0, 0);
1018                 break;
1019         case NETDEV_CHANGEMTU:
1020         case NETDEV_CHANGE:
1021                 rt_cache_flush(dev_net(dev), 0);
1022                 break;
1023         case NETDEV_UNREGISTER_BATCH:
1024                 /* The batch unregister is only called on the first
1025                  * device in the list of devices being unregistered.
1026                  * Therefore we should not pass dev_net(dev) in here.
1027                  */
1028                 rt_cache_flush_batch(NULL);
1029                 break;
1030         }
1031         return NOTIFY_DONE;
1032 }
1033
1034 static struct notifier_block fib_inetaddr_notifier = {
1035         .notifier_call = fib_inetaddr_event,
1036 };
1037
1038 static struct notifier_block fib_netdev_notifier = {
1039         .notifier_call = fib_netdev_event,
1040 };
1041
1042 static int __net_init ip_fib_net_init(struct net *net)
1043 {
1044         int err;
1045         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1046
1047         /* Avoid false sharing : Use at least a full cache line */
1048         size = max_t(size_t, size, L1_CACHE_BYTES);
1049
1050         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1051         if (net->ipv4.fib_table_hash == NULL)
1052                 return -ENOMEM;
1053
1054         err = fib4_rules_init(net);
1055         if (err < 0)
1056                 goto fail;
1057         return 0;
1058
1059 fail:
1060         kfree(net->ipv4.fib_table_hash);
1061         return err;
1062 }
1063
1064 static void ip_fib_net_exit(struct net *net)
1065 {
1066         unsigned int i;
1067
1068 #ifdef CONFIG_IP_MULTIPLE_TABLES
1069         fib4_rules_exit(net);
1070 #endif
1071
1072         rtnl_lock();
1073         for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1074                 struct fib_table *tb;
1075                 struct hlist_head *head;
1076                 struct hlist_node *node, *tmp;
1077
1078                 head = &net->ipv4.fib_table_hash[i];
1079                 hlist_for_each_entry_safe(tb, node, tmp, head, tb_hlist) {
1080                         hlist_del(node);
1081                         fib_table_flush(tb);
1082                         fib_free_table(tb);
1083                 }
1084         }
1085         rtnl_unlock();
1086         kfree(net->ipv4.fib_table_hash);
1087 }
1088
1089 static int __net_init fib_net_init(struct net *net)
1090 {
1091         int error;
1092
1093         error = ip_fib_net_init(net);
1094         if (error < 0)
1095                 goto out;
1096         error = nl_fib_lookup_init(net);
1097         if (error < 0)
1098                 goto out_nlfl;
1099         error = fib_proc_init(net);
1100         if (error < 0)
1101                 goto out_proc;
1102 out:
1103         return error;
1104
1105 out_proc:
1106         nl_fib_lookup_exit(net);
1107 out_nlfl:
1108         ip_fib_net_exit(net);
1109         goto out;
1110 }
1111
1112 static void __net_exit fib_net_exit(struct net *net)
1113 {
1114         fib_proc_exit(net);
1115         nl_fib_lookup_exit(net);
1116         ip_fib_net_exit(net);
1117 }
1118
1119 static struct pernet_operations fib_net_ops = {
1120         .init = fib_net_init,
1121         .exit = fib_net_exit,
1122 };
1123
1124 void __init ip_fib_init(void)
1125 {
1126         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1127         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1128         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1129
1130         register_pernet_subsys(&fib_net_ops);
1131         register_netdevice_notifier(&fib_netdev_notifier);
1132         register_inetaddr_notifier(&fib_inetaddr_notifier);
1133
1134         fib_trie_init();
1135 }