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 __be32 fib_compute_spec_dst(struct sk_buff *skb)
 184 {
 185         struct net_device *dev = skb->dev;
 186         struct in_device *in_dev;
 187         struct fib_result res;
 188         struct flowi4 fl4;
 189         struct net *net;
 190
 191         if (skb->pkt_type != PACKET_BROADCAST &&
 192             skb->pkt_type != PACKET_MULTICAST)
 193                 return ip_hdr(skb)->daddr;
 194
 195         in_dev = __in_dev_get_rcu(dev);
 196         BUG_ON(!in_dev);
 197         fl4.flowi4_oif = 0;
 198         fl4.flowi4_iif = 0;
 199         fl4.daddr = ip_hdr(skb)->saddr;
 200         fl4.saddr = ip_hdr(skb)->daddr;
 201         fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
 202         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 203         fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0;
 204
 205         net = dev_net(dev);
 206         if (!fib_lookup(net, &fl4, &res))
 207                 return FIB_RES_PREFSRC(net, res);
 208         else
 209                 return inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);
 210 }
 211
 212 /* Given (packet source, input interface) and optional (dst, oif, tos):
 213  * - (main) check, that source is valid i.e. not broadcast or our local
 214  *   address.
 215  * - figure out what "logical" interface this packet arrived
 216  *   and calculate "specific destination" address.
 217  * - check, that packet arrived from expected physical interface.
 218  * called with rcu_read_lock()
 219  */
 220 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, u8 tos,
 221                         int oif, struct net_device *dev, u32 *itag)
 222 {
 223         struct in_device *in_dev;
 224         struct flowi4 fl4;
 225         struct fib_result res;
 226         int no_addr, rpf, accept_local;
 227         bool dev_match;
 228         int ret;
 229         struct net *net;
 230
 231         fl4.flowi4_oif = 0;
 232         fl4.flowi4_iif = oif;
 233         fl4.daddr = src;
 234         fl4.saddr = dst;
 235         fl4.flowi4_tos = tos;
 236         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 237
 238         no_addr = rpf = accept_local = 0;
 239         in_dev = __in_dev_get_rcu(dev);
 240         if (in_dev) {
 241                 no_addr = in_dev->ifa_list == NULL;
 242
 243                 /* Ignore rp_filter for packets protected by IPsec. */
 244                 rpf = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(in_dev);
 245
 246                 accept_local = IN_DEV_ACCEPT_LOCAL(in_dev);
 247                 fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0;
 248         }
 249
 250         if (in_dev == NULL)
 251                 goto e_inval;
 252
 253         net = dev_net(dev);
 254         if (fib_lookup(net, &fl4, &res))
 255                 goto last_resort;
 256         if (res.type != RTN_UNICAST) {
 257                 if (res.type != RTN_LOCAL || !accept_local)
 258                         goto e_inval;
 259         }
 260         fib_combine_itag(itag, &res);
 261         dev_match = false;
 262
 263 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 264         for (ret = 0; ret < res.fi->fib_nhs; ret++) {
 265                 struct fib_nh *nh = &res.fi->fib_nh[ret];
 266
 267                 if (nh->nh_dev == dev) {
 268                         dev_match = true;
 269                         break;
 270                 }
 271         }
 272 #else
 273         if (FIB_RES_DEV(res) == dev)
 274                 dev_match = true;
 275 #endif
 276         if (dev_match) {
 277                 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 278                 return ret;
 279         }
 280         if (no_addr)
 281                 goto last_resort;
 282         if (rpf == 1)
 283                 goto e_rpf;
 284         fl4.flowi4_oif = dev->ifindex;
 285
 286         ret = 0;
 287         if (fib_lookup(net, &fl4, &res) == 0) {
 288                 if (res.type == RTN_UNICAST)
 289                         ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 290         }
 291         return ret;
 292
 293 last_resort:
 294         if (rpf)
 295                 goto e_rpf;
 296         *itag = 0;
 297         return 0;
 298
 299 e_inval:
 300         return -EINVAL;
 301 e_rpf:
 302         return -EXDEV;
 303 }
 304
 305 static inline __be32 sk_extract_addr(struct sockaddr *addr)
 306 {
 307         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 308 }
 309
 310 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
 311 {
 312         struct nlattr *nla;
 313
 314         nla = (struct nlattr *) ((char *) mx + len);
 315         nla->nla_type = type;
 316         nla->nla_len = nla_attr_size(4);
 317         *(u32 *) nla_data(nla) = value;
 318
 319         return len + nla_total_size(4);
 320 }
 321
 322 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
 323                                  struct fib_config *cfg)
 324 {
 325         __be32 addr;
 326         int plen;
 327
 328         memset(cfg, 0, sizeof(*cfg));
 329         cfg->fc_nlinfo.nl_net = net;
 330
 331         if (rt->rt_dst.sa_family != AF_INET)
 332                 return -EAFNOSUPPORT;
 333
 334         /*
 335          * Check mask for validity:
 336          * a) it must be contiguous.
 337          * b) destination must have all host bits clear.
 338          * c) if application forgot to set correct family (AF_INET),
 339          *    reject request unless it is absolutely clear i.e.
 340          *    both family and mask are zero.
 341          */
 342         plen = 32;
 343         addr = sk_extract_addr(&rt->rt_dst);
 344         if (!(rt->rt_flags & RTF_HOST)) {
 345                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
 346
 347                 if (rt->rt_genmask.sa_family != AF_INET) {
 348                         if (mask || rt->rt_genmask.sa_family)
 349                                 return -EAFNOSUPPORT;
 350                 }
 351
 352                 if (bad_mask(mask, addr))
 353                         return -EINVAL;
 354
 355                 plen = inet_mask_len(mask);
 356         }
 357
 358         cfg->fc_dst_len = plen;
 359         cfg->fc_dst = addr;
 360
 361         if (cmd != SIOCDELRT) {
 362                 cfg->fc_nlflags = NLM_F_CREATE;
 363                 cfg->fc_protocol = RTPROT_BOOT;
 364         }
 365
 366         if (rt->rt_metric)
 367                 cfg->fc_priority = rt->rt_metric - 1;
 368
 369         if (rt->rt_flags & RTF_REJECT) {
 370                 cfg->fc_scope = RT_SCOPE_HOST;
 371                 cfg->fc_type = RTN_UNREACHABLE;
 372                 return 0;
 373         }
 374
 375         cfg->fc_scope = RT_SCOPE_NOWHERE;
 376         cfg->fc_type = RTN_UNICAST;
 377
 378         if (rt->rt_dev) {
 379                 char *colon;
 380                 struct net_device *dev;
 381                 char devname[IFNAMSIZ];
 382
 383                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
 384                         return -EFAULT;
 385
 386                 devname[IFNAMSIZ-1] = 0;
 387                 colon = strchr(devname, ':');
 388                 if (colon)
 389                         *colon = 0;
 390                 dev = __dev_get_by_name(net, devname);
 391                 if (!dev)
 392                         return -ENODEV;
 393                 cfg->fc_oif = dev->ifindex;
 394                 if (colon) {
 395                         struct in_ifaddr *ifa;
 396                         struct in_device *in_dev = __in_dev_get_rtnl(dev);
 397                         if (!in_dev)
 398                                 return -ENODEV;
 399                         *colon = ':';
 400                         for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
 401                                 if (strcmp(ifa->ifa_label, devname) == 0)
 402                                         break;
 403                         if (ifa == NULL)
 404                                 return -ENODEV;
 405                         cfg->fc_prefsrc = ifa->ifa_local;
 406                 }
 407         }
 408
 409         addr = sk_extract_addr(&rt->rt_gateway);
 410         if (rt->rt_gateway.sa_family == AF_INET && addr) {
 411                 cfg->fc_gw = addr;
 412                 if (rt->rt_flags & RTF_GATEWAY &&
 413                     inet_addr_type(net, addr) == RTN_UNICAST)
 414                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
 415         }
 416
 417         if (cmd == SIOCDELRT)
 418                 return 0;
 419
 420         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
 421                 return -EINVAL;
 422
 423         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 424                 cfg->fc_scope = RT_SCOPE_LINK;
 425
 426         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 427                 struct nlattr *mx;
 428                 int len = 0;
 429
 430                 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
 431                 if (mx == NULL)
 432                         return -ENOMEM;
 433
 434                 if (rt->rt_flags & RTF_MTU)
 435                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 436
 437                 if (rt->rt_flags & RTF_WINDOW)
 438                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 439
 440                 if (rt->rt_flags & RTF_IRTT)
 441                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 442
 443                 cfg->fc_mx = mx;
 444                 cfg->fc_mx_len = len;
 445         }
 446
 447         return 0;
 448 }
 449
 450 /*
 451  * Handle IP routing ioctl calls.
 452  * These are used to manipulate the routing tables
 453  */
 454 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
 455 {
 456         struct fib_config cfg;
 457         struct rtentry rt;
 458         int err;
 459
 460         switch (cmd) {
 461         case SIOCADDRT:         /* Add a route */
 462         case SIOCDELRT:         /* Delete a route */
 463                 if (!capable(CAP_NET_ADMIN))
 464                         return -EPERM;
 465
 466                 if (copy_from_user(&rt, arg, sizeof(rt)))
 467                         return -EFAULT;
 468
 469                 rtnl_lock();
 470                 err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
 471                 if (err == 0) {
 472                         struct fib_table *tb;
 473
 474                         if (cmd == SIOCDELRT) {
 475                                 tb = fib_get_table(net, cfg.fc_table);
 476                                 if (tb)
 477                                         err = fib_table_delete(tb, &cfg);
 478                                 else
 479                                         err = -ESRCH;
 480                         } else {
 481                                 tb = fib_new_table(net, cfg.fc_table);
 482                                 if (tb)
 483                                         err = fib_table_insert(tb, &cfg);
 484                                 else
 485                                         err = -ENOBUFS;
 486                         }
 487
 488                         /* allocated by rtentry_to_fib_config() */
 489                         kfree(cfg.fc_mx);
 490                 }
 491                 rtnl_unlock();
 492                 return err;
 493         }
 494         return -EINVAL;
 495 }
 496
 497 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 498         [RTA_DST]               = { .type = NLA_U32 },
 499         [RTA_SRC]               = { .type = NLA_U32 },
 500         [RTA_IIF]               = { .type = NLA_U32 },
 501         [RTA_OIF]               = { .type = NLA_U32 },
 502         [RTA_GATEWAY]           = { .type = NLA_U32 },
 503         [RTA_PRIORITY]          = { .type = NLA_U32 },
 504         [RTA_PREFSRC]           = { .type = NLA_U32 },
 505         [RTA_METRICS]           = { .type = NLA_NESTED },
 506         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 507         [RTA_FLOW]              = { .type = NLA_U32 },
 508 };
 509
 510 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 511                              struct nlmsghdr *nlh, struct fib_config *cfg)
 512 {
 513         struct nlattr *attr;
 514         int err, remaining;
 515         struct rtmsg *rtm;
 516
 517         err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
 518         if (err < 0)
 519                 goto errout;
 520
 521         memset(cfg, 0, sizeof(*cfg));
 522
 523         rtm = nlmsg_data(nlh);
 524         cfg->fc_dst_len = rtm->rtm_dst_len;
 525         cfg->fc_tos = rtm->rtm_tos;
 526         cfg->fc_table = rtm->rtm_table;
 527         cfg->fc_protocol = rtm->rtm_protocol;
 528         cfg->fc_scope = rtm->rtm_scope;
 529         cfg->fc_type = rtm->rtm_type;
 530         cfg->fc_flags = rtm->rtm_flags;
 531         cfg->fc_nlflags = nlh->nlmsg_flags;
 532
 533         cfg->fc_nlinfo.pid = NETLINK_CB(skb).pid;
 534         cfg->fc_nlinfo.nlh = nlh;
 535         cfg->fc_nlinfo.nl_net = net;
 536
 537         if (cfg->fc_type > RTN_MAX) {
 538                 err = -EINVAL;
 539                 goto errout;
 540         }
 541
 542         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 543                 switch (nla_type(attr)) {
 544                 case RTA_DST:
 545                         cfg->fc_dst = nla_get_be32(attr);
 546                         break;
 547                 case RTA_OIF:
 548                         cfg->fc_oif = nla_get_u32(attr);
 549                         break;
 550                 case RTA_GATEWAY:
 551                         cfg->fc_gw = nla_get_be32(attr);
 552                         break;
 553                 case RTA_PRIORITY:
 554                         cfg->fc_priority = nla_get_u32(attr);
 555                         break;
 556                 case RTA_PREFSRC:
 557                         cfg->fc_prefsrc = nla_get_be32(attr);
 558                         break;
 559                 case RTA_METRICS:
 560                         cfg->fc_mx = nla_data(attr);
 561                         cfg->fc_mx_len = nla_len(attr);
 562                         break;
 563                 case RTA_MULTIPATH:
 564                         cfg->fc_mp = nla_data(attr);
 565                         cfg->fc_mp_len = nla_len(attr);
 566                         break;
 567                 case RTA_FLOW:
 568                         cfg->fc_flow = nla_get_u32(attr);
 569                         break;
 570                 case RTA_TABLE:
 571                         cfg->fc_table = nla_get_u32(attr);
 572                         break;
 573                 }
 574         }
 575
 576         return 0;
 577 errout:
 578         return err;
 579 }
 580
 581 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 582 {
 583         struct net *net = sock_net(skb->sk);
 584         struct fib_config cfg;
 585         struct fib_table *tb;
 586         int err;
 587
 588         err = rtm_to_fib_config(net, skb, nlh, &cfg);
 589         if (err < 0)
 590                 goto errout;
 591
 592         tb = fib_get_table(net, cfg.fc_table);
 593         if (tb == NULL) {
 594                 err = -ESRCH;
 595                 goto errout;
 596         }
 597
 598         err = fib_table_delete(tb, &cfg);
 599 errout:
 600         return err;
 601 }
 602
 603 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 604 {
 605         struct net *net = sock_net(skb->sk);
 606         struct fib_config cfg;
 607         struct fib_table *tb;
 608         int err;
 609
 610         err = rtm_to_fib_config(net, skb, nlh, &cfg);
 611         if (err < 0)
 612                 goto errout;
 613
 614         tb = fib_new_table(net, cfg.fc_table);
 615         if (tb == NULL) {
 616                 err = -ENOBUFS;
 617                 goto errout;
 618         }
 619
 620         err = fib_table_insert(tb, &cfg);
 621 errout:
 622         return err;
 623 }
 624
 625 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 626 {
 627         struct net *net = sock_net(skb->sk);
 628         unsigned int h, s_h;
 629         unsigned int e = 0, s_e;
 630         struct fib_table *tb;
 631         struct hlist_node *node;
 632         struct hlist_head *head;
 633         int dumped = 0;
 634
 635         if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
 636             ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
 637                 return ip_rt_dump(skb, cb);
 638
 639         s_h = cb->args[0];
 640         s_e = cb->args[1];
 641
 642         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
 643                 e = 0;
 644                 head = &net->ipv4.fib_table_hash[h];
 645                 hlist_for_each_entry(tb, node, head, tb_hlist) {
 646                         if (e < s_e)
 647                                 goto next;
 648                         if (dumped)
 649                                 memset(&cb->args[2], 0, sizeof(cb->args) -
 650                                                  2 * sizeof(cb->args[0]));
 651                         if (fib_table_dump(tb, skb, cb) < 0)
 652                                 goto out;
 653                         dumped = 1;
 654 next:
 655                         e++;
 656                 }
 657         }
 658 out:
 659         cb->args[1] = e;
 660         cb->args[0] = h;
 661
 662         return skb->len;
 663 }
 664
 665 /* Prepare and feed intra-kernel routing request.
 666  * Really, it should be netlink message, but :-( netlink
 667  * can be not configured, so that we feed it directly
 668  * to fib engine. It is legal, because all events occur
 669  * only when netlink is already locked.
 670  */
 671 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
 672 {
 673         struct net *net = dev_net(ifa->ifa_dev->dev);
 674         struct fib_table *tb;
 675         struct fib_config cfg = {
 676                 .fc_protocol = RTPROT_KERNEL,
 677                 .fc_type = type,
 678                 .fc_dst = dst,
 679                 .fc_dst_len = dst_len,
 680                 .fc_prefsrc = ifa->ifa_local,
 681                 .fc_oif = ifa->ifa_dev->dev->ifindex,
 682                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
 683                 .fc_nlinfo = {
 684                         .nl_net = net,
 685                 },
 686         };
 687
 688         if (type == RTN_UNICAST)
 689                 tb = fib_new_table(net, RT_TABLE_MAIN);
 690         else
 691                 tb = fib_new_table(net, RT_TABLE_LOCAL);
 692
 693         if (tb == NULL)
 694                 return;
 695
 696         cfg.fc_table = tb->tb_id;
 697
 698         if (type != RTN_LOCAL)
 699                 cfg.fc_scope = RT_SCOPE_LINK;
 700         else
 701                 cfg.fc_scope = RT_SCOPE_HOST;
 702
 703         if (cmd == RTM_NEWROUTE)
 704                 fib_table_insert(tb, &cfg);
 705         else
 706                 fib_table_delete(tb, &cfg);
 707 }
 708
 709 void fib_add_ifaddr(struct in_ifaddr *ifa)
 710 {
 711         struct in_device *in_dev = ifa->ifa_dev;
 712         struct net_device *dev = in_dev->dev;
 713         struct in_ifaddr *prim = ifa;
 714         __be32 mask = ifa->ifa_mask;
 715         __be32 addr = ifa->ifa_local;
 716         __be32 prefix = ifa->ifa_address & mask;
 717
 718         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 719                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
 720                 if (prim == NULL) {
 721                         pr_warn("%s: bug: prim == NULL\n", __func__);
 722                         return;
 723                 }
 724         }
 725
 726         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
 727
 728         if (!(dev->flags & IFF_UP))
 729                 return;
 730
 731         /* Add broadcast address, if it is explicitly assigned. */
 732         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
 733                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
 734
 735         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
 736             (prefix != addr || ifa->ifa_prefixlen < 32)) {
 737                 fib_magic(RTM_NEWROUTE,
 738                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 739                           prefix, ifa->ifa_prefixlen, prim);
 740
 741                 /* Add network specific broadcasts, when it takes a sense */
 742                 if (ifa->ifa_prefixlen < 31) {
 743                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
 744                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
 745                                   32, prim);
 746                 }
 747         }
 748 }
 749
 750 /* Delete primary or secondary address.
 751  * Optionally, on secondary address promotion consider the addresses
 752  * from subnet iprim as deleted, even if they are in device list.
 753  * In this case the secondary ifa can be in device list.
 754  */
 755 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
 756 {
 757         struct in_device *in_dev = ifa->ifa_dev;
 758         struct net_device *dev = in_dev->dev;
 759         struct in_ifaddr *ifa1;
 760         struct in_ifaddr *prim = ifa, *prim1 = NULL;
 761         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
 762         __be32 any = ifa->ifa_address & ifa->ifa_mask;
 763 #define LOCAL_OK        1
 764 #define BRD_OK          2
 765 #define BRD0_OK         4
 766 #define BRD1_OK         8
 767         unsigned int ok = 0;
 768         int subnet = 0;         /* Primary network */
 769         int gone = 1;           /* Address is missing */
 770         int same_prefsrc = 0;   /* Another primary with same IP */
 771
 772         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 773                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
 774                 if (prim == NULL) {
 775                         pr_warn("%s: bug: prim == NULL\n", __func__);
 776                         return;
 777                 }
 778                 if (iprim && iprim != prim) {
 779                         pr_warn("%s: bug: iprim != prim\n", __func__);
 780                         return;
 781                 }
 782         } else if (!ipv4_is_zeronet(any) &&
 783                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
 784                 fib_magic(RTM_DELROUTE,
 785                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 786                           any, ifa->ifa_prefixlen, prim);
 787                 subnet = 1;
 788         }
 789
 790         /* Deletion is more complicated than add.
 791          * We should take care of not to delete too much :-)
 792          *
 793          * Scan address list to be sure that addresses are really gone.
 794          */
 795
 796         for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
 797                 if (ifa1 == ifa) {
 798                         /* promotion, keep the IP */
 799                         gone = 0;
 800                         continue;
 801                 }
 802                 /* Ignore IFAs from our subnet */
 803                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
 804                     inet_ifa_match(ifa1->ifa_address, iprim))
 805                         continue;
 806
 807                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
 808                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
 809                         /* Another address from our subnet? */
 810                         if (ifa1->ifa_mask == prim->ifa_mask &&
 811                             inet_ifa_match(ifa1->ifa_address, prim))
 812                                 prim1 = prim;
 813                         else {
 814                                 /* We reached the secondaries, so
 815                                  * same_prefsrc should be determined.
 816                                  */
 817                                 if (!same_prefsrc)
 818                                         continue;
 819                                 /* Search new prim1 if ifa1 is not
 820                                  * using the current prim1
 821                                  */
 822                                 if (!prim1 ||
 823                                     ifa1->ifa_mask != prim1->ifa_mask ||
 824                                     !inet_ifa_match(ifa1->ifa_address, prim1))
 825                                         prim1 = inet_ifa_byprefix(in_dev,
 826                                                         ifa1->ifa_address,
 827                                                         ifa1->ifa_mask);
 828                                 if (!prim1)
 829                                         continue;
 830                                 if (prim1->ifa_local != prim->ifa_local)
 831                                         continue;
 832                         }
 833                 } else {
 834                         if (prim->ifa_local != ifa1->ifa_local)
 835                                 continue;
 836                         prim1 = ifa1;
 837                         if (prim != prim1)
 838                                 same_prefsrc = 1;
 839                 }
 840                 if (ifa->ifa_local == ifa1->ifa_local)
 841                         ok |= LOCAL_OK;
 842                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
 843                         ok |= BRD_OK;
 844                 if (brd == ifa1->ifa_broadcast)
 845                         ok |= BRD1_OK;
 846                 if (any == ifa1->ifa_broadcast)
 847                         ok |= BRD0_OK;
 848                 /* primary has network specific broadcasts */
 849                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
 850                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
 851                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
 852
 853                         if (!ipv4_is_zeronet(any1)) {
 854                                 if (ifa->ifa_broadcast == brd1 ||
 855                                     ifa->ifa_broadcast == any1)
 856                                         ok |= BRD_OK;
 857                                 if (brd == brd1 || brd == any1)
 858                                         ok |= BRD1_OK;
 859                                 if (any == brd1 || any == any1)
 860                                         ok |= BRD0_OK;
 861                         }
 862                 }
 863         }
 864
 865         if (!(ok & BRD_OK))
 866                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
 867         if (subnet && ifa->ifa_prefixlen < 31) {
 868                 if (!(ok & BRD1_OK))
 869                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
 870                 if (!(ok & BRD0_OK))
 871                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
 872         }
 873         if (!(ok & LOCAL_OK)) {
 874                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
 875
 876                 /* Check, that this local address finally disappeared. */
 877                 if (gone &&
 878                     inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) {
 879                         /* And the last, but not the least thing.
 880                          * We must flush stray FIB entries.
 881                          *
 882                          * First of all, we scan fib_info list searching
 883                          * for stray nexthop entries, then ignite fib_flush.
 884                          */
 885                         if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
 886                                 fib_flush(dev_net(dev));
 887                 }
 888         }
 889 #undef LOCAL_OK
 890 #undef BRD_OK
 891 #undef BRD0_OK
 892 #undef BRD1_OK
 893 }
 894
 895 static void nl_fib_lookup(struct fib_result_nl *frn, struct fib_table *tb)
 896 {
 897
 898         struct fib_result       res;
 899         struct flowi4           fl4 = {
 900                 .flowi4_mark = frn->fl_mark,
 901                 .daddr = frn->fl_addr,
 902                 .flowi4_tos = frn->fl_tos,
 903                 .flowi4_scope = frn->fl_scope,
 904         };
 905
 906 #ifdef CONFIG_IP_MULTIPLE_TABLES
 907         res.r = NULL;
 908 #endif
 909
 910         frn->err = -ENOENT;
 911         if (tb) {
 912                 local_bh_disable();
 913
 914                 frn->tb_id = tb->tb_id;
 915                 rcu_read_lock();
 916                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
 917
 918                 if (!frn->err) {
 919                         frn->prefixlen = res.prefixlen;
 920                         frn->nh_sel = res.nh_sel;
 921                         frn->type = res.type;
 922                         frn->scope = res.scope;
 923                 }
 924                 rcu_read_unlock();
 925                 local_bh_enable();
 926         }
 927 }
 928
 929 static void nl_fib_input(struct sk_buff *skb)
 930 {
 931         struct net *net;
 932         struct fib_result_nl *frn;
 933         struct nlmsghdr *nlh;
 934         struct fib_table *tb;
 935         u32 pid;
 936
 937         net = sock_net(skb->sk);
 938         nlh = nlmsg_hdr(skb);
 939         if (skb->len < NLMSG_SPACE(0) || skb->len < nlh->nlmsg_len ||
 940             nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*frn)))
 941                 return;
 942
 943         skb = skb_clone(skb, GFP_KERNEL);
 944         if (skb == NULL)
 945                 return;
 946         nlh = nlmsg_hdr(skb);
 947
 948         frn = (struct fib_result_nl *) NLMSG_DATA(nlh);
 949         tb = fib_get_table(net, frn->tb_id_in);
 950
 951         nl_fib_lookup(frn, tb);
 952
 953         pid = NETLINK_CB(skb).pid;      /* pid of sending process */
 954         NETLINK_CB(skb).pid = 0;        /* from kernel */
 955         NETLINK_CB(skb).dst_group = 0;  /* unicast */
 956         netlink_unicast(net->ipv4.fibnl, skb, pid, MSG_DONTWAIT);
 957 }
 958
 959 static int __net_init nl_fib_lookup_init(struct net *net)
 960 {
 961         struct sock *sk;
 962         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, 0,
 963                                    nl_fib_input, NULL, THIS_MODULE);
 964         if (sk == NULL)
 965                 return -EAFNOSUPPORT;
 966         net->ipv4.fibnl = sk;
 967         return 0;
 968 }
 969
 970 static void nl_fib_lookup_exit(struct net *net)
 971 {
 972         netlink_kernel_release(net->ipv4.fibnl);
 973         net->ipv4.fibnl = NULL;
 974 }
 975
 976 static void fib_disable_ip(struct net_device *dev, int force, int delay)
 977 {
 978         if (fib_sync_down_dev(dev, force))
 979                 fib_flush(dev_net(dev));
 980         rt_cache_flush(dev_net(dev), delay);
 981         arp_ifdown(dev);
 982 }
 983
 984 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
 985 {
 986         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
 987         struct net_device *dev = ifa->ifa_dev->dev;
 988         struct net *net = dev_net(dev);
 989
 990         switch (event) {
 991         case NETDEV_UP:
 992                 fib_add_ifaddr(ifa);
 993 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 994                 fib_sync_up(dev);
 995 #endif
 996                 atomic_inc(&net->ipv4.dev_addr_genid);
 997                 rt_cache_flush(dev_net(dev), -1);
 998                 break;
 999         case NETDEV_DOWN:
1000                 fib_del_ifaddr(ifa, NULL);
1001                 atomic_inc(&net->ipv4.dev_addr_genid);
1002                 if (ifa->ifa_dev->ifa_list == NULL) {
1003                         /* Last address was deleted from this interface.
1004                          * Disable IP.
1005                          */
1006                         fib_disable_ip(dev, 1, 0);
1007                 } else {
1008                         rt_cache_flush(dev_net(dev), -1);
1009                 }
1010                 break;
1011         }
1012         return NOTIFY_DONE;
1013 }
1014
1015 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1016 {
1017         struct net_device *dev = ptr;
1018         struct in_device *in_dev = __in_dev_get_rtnl(dev);
1019         struct net *net = dev_net(dev);
1020
1021         if (event == NETDEV_UNREGISTER) {
1022                 fib_disable_ip(dev, 2, -1);
1023                 return NOTIFY_DONE;
1024         }
1025
1026         if (!in_dev)
1027                 return NOTIFY_DONE;
1028
1029         switch (event) {
1030         case NETDEV_UP:
1031                 for_ifa(in_dev) {
1032                         fib_add_ifaddr(ifa);
1033                 } endfor_ifa(in_dev);
1034 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1035                 fib_sync_up(dev);
1036 #endif
1037                 atomic_inc(&net->ipv4.dev_addr_genid);
1038                 rt_cache_flush(dev_net(dev), -1);
1039                 break;
1040         case NETDEV_DOWN:
1041                 fib_disable_ip(dev, 0, 0);
1042                 break;
1043         case NETDEV_CHANGEMTU:
1044         case NETDEV_CHANGE:
1045                 rt_cache_flush(dev_net(dev), 0);
1046                 break;
1047         case NETDEV_UNREGISTER_BATCH:
1048                 /* The batch unregister is only called on the first
1049                  * device in the list of devices being unregistered.
1050                  * Therefore we should not pass dev_net(dev) in here.
1051                  */
1052                 rt_cache_flush_batch(NULL);
1053                 break;
1054         }
1055         return NOTIFY_DONE;
1056 }
1057
1058 static struct notifier_block fib_inetaddr_notifier = {
1059         .notifier_call = fib_inetaddr_event,
1060 };
1061
1062 static struct notifier_block fib_netdev_notifier = {
1063         .notifier_call = fib_netdev_event,
1064 };
1065
1066 static int __net_init ip_fib_net_init(struct net *net)
1067 {
1068         int err;
1069         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1070
1071         /* Avoid false sharing : Use at least a full cache line */
1072         size = max_t(size_t, size, L1_CACHE_BYTES);
1073
1074         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1075         if (net->ipv4.fib_table_hash == NULL)
1076                 return -ENOMEM;
1077
1078         err = fib4_rules_init(net);
1079         if (err < 0)
1080                 goto fail;
1081         return 0;
1082
1083 fail:
1084         kfree(net->ipv4.fib_table_hash);
1085         return err;
1086 }
1087
1088 static void ip_fib_net_exit(struct net *net)
1089 {
1090         unsigned int i;
1091
1092 #ifdef CONFIG_IP_MULTIPLE_TABLES
1093         fib4_rules_exit(net);
1094 #endif
1095
1096         rtnl_lock();
1097         for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1098                 struct fib_table *tb;
1099                 struct hlist_head *head;
1100                 struct hlist_node *node, *tmp;
1101
1102                 head = &net->ipv4.fib_table_hash[i];
1103                 hlist_for_each_entry_safe(tb, node, tmp, head, tb_hlist) {
1104                         hlist_del(node);
1105                         fib_table_flush(tb);
1106                         fib_free_table(tb);
1107                 }
1108         }
1109         rtnl_unlock();
1110         kfree(net->ipv4.fib_table_hash);
1111 }
1112
1113 static int __net_init fib_net_init(struct net *net)
1114 {
1115         int error;
1116
1117         error = ip_fib_net_init(net);
1118         if (error < 0)
1119                 goto out;
1120         error = nl_fib_lookup_init(net);
1121         if (error < 0)
1122                 goto out_nlfl;
1123         error = fib_proc_init(net);
1124         if (error < 0)
1125                 goto out_proc;
1126 out:
1127         return error;
1128
1129 out_proc:
1130         nl_fib_lookup_exit(net);
1131 out_nlfl:
1132         ip_fib_net_exit(net);
1133         goto out;
1134 }
1135
1136 static void __net_exit fib_net_exit(struct net *net)
1137 {
1138         fib_proc_exit(net);
1139         nl_fib_lookup_exit(net);
1140         ip_fib_net_exit(net);
1141 }
1142
1143 static struct pernet_operations fib_net_ops = {
1144         .init = fib_net_init,
1145         .exit = fib_net_exit,
1146 };
1147
1148 void __init ip_fib_init(void)
1149 {
1150         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1151         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1152         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1153
1154         register_pernet_subsys(&fib_net_ops);
1155         register_netdevice_notifier(&fib_netdev_notifier);
1156         register_inetaddr_notifier(&fib_inetaddr_notifier);
1157
1158         fib_trie_init();
1159 }