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 rtable *rt;
 189         struct flowi4 fl4;
 190         struct net *net;
 191         int scope;
 192
 193         rt = skb_rtable(skb);
 194         if (!(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)))
 195                 return ip_hdr(skb)->daddr;
 196
 197         in_dev = __in_dev_get_rcu(dev);
 198         BUG_ON(!in_dev);
 199
 200         net = dev_net(dev);
 201
 202         scope = RT_SCOPE_UNIVERSE;
 203         if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
 204                 fl4.flowi4_oif = 0;
 205                 fl4.flowi4_iif = net->loopback_dev->ifindex;
 206                 fl4.daddr = ip_hdr(skb)->saddr;
 207                 fl4.saddr = 0;
 208                 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
 209                 fl4.flowi4_scope = scope;
 210                 fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0;
 211                 if (!fib_lookup(net, &fl4, &res))
 212                         return FIB_RES_PREFSRC(net, res);
 213         } else {
 214                 scope = RT_SCOPE_LINK;
 215         }
 216
 217         return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
 218 }
 219
 220 /* Given (packet source, input interface) and optional (dst, oif, tos):
 221  * - (main) check, that source is valid i.e. not broadcast or our local
 222  *   address.
 223  * - figure out what "logical" interface this packet arrived
 224  *   and calculate "specific destination" address.
 225  * - check, that packet arrived from expected physical interface.
 226  * called with rcu_read_lock()
 227  */
 228 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst, u8 tos,
 229                         int oif, struct net_device *dev, struct in_device *idev,
 230                         u32 *itag)
 231 {
 232         int ret, no_addr, rpf, accept_local;
 233         struct fib_result res;
 234         struct flowi4 fl4;
 235         struct net *net;
 236         bool dev_match;
 237
 238         fl4.flowi4_oif = 0;
 239         fl4.flowi4_iif = oif;
 240         fl4.daddr = src;
 241         fl4.saddr = dst;
 242         fl4.flowi4_tos = tos;
 243         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 244
 245         no_addr = rpf = accept_local = 0;
 246         no_addr = idev->ifa_list == NULL;
 247
 248         /* Ignore rp_filter for packets protected by IPsec. */
 249         rpf = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
 250
 251         accept_local = IN_DEV_ACCEPT_LOCAL(idev);
 252         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
 253
 254         net = dev_net(dev);
 255         if (fib_lookup(net, &fl4, &res))
 256                 goto last_resort;
 257         if (res.type != RTN_UNICAST) {
 258                 if (res.type != RTN_LOCAL || !accept_local)
 259                         goto e_inval;
 260         }
 261         fib_combine_itag(itag, &res);
 262         dev_match = false;
 263
 264 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 265         for (ret = 0; ret < res.fi->fib_nhs; ret++) {
 266                 struct fib_nh *nh = &res.fi->fib_nh[ret];
 267
 268                 if (nh->nh_dev == dev) {
 269                         dev_match = true;
 270                         break;
 271                 }
 272         }
 273 #else
 274         if (FIB_RES_DEV(res) == dev)
 275                 dev_match = true;
 276 #endif
 277         if (dev_match) {
 278                 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 279                 return ret;
 280         }
 281         if (no_addr)
 282                 goto last_resort;
 283         if (rpf == 1)
 284                 goto e_rpf;
 285         fl4.flowi4_oif = dev->ifindex;
 286
 287         ret = 0;
 288         if (fib_lookup(net, &fl4, &res) == 0) {
 289                 if (res.type == RTN_UNICAST)
 290                         ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 291         }
 292         return ret;
 293
 294 last_resort:
 295         if (rpf)
 296                 goto e_rpf;
 297         *itag = 0;
 298         return 0;
 299
 300 e_inval:
 301         return -EINVAL;
 302 e_rpf:
 303         return -EXDEV;
 304 }
 305
 306 static inline __be32 sk_extract_addr(struct sockaddr *addr)
 307 {
 308         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 309 }
 310
 311 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
 312 {
 313         struct nlattr *nla;
 314
 315         nla = (struct nlattr *) ((char *) mx + len);
 316         nla->nla_type = type;
 317         nla->nla_len = nla_attr_size(4);
 318         *(u32 *) nla_data(nla) = value;
 319
 320         return len + nla_total_size(4);
 321 }
 322
 323 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
 324                                  struct fib_config *cfg)
 325 {
 326         __be32 addr;
 327         int plen;
 328
 329         memset(cfg, 0, sizeof(*cfg));
 330         cfg->fc_nlinfo.nl_net = net;
 331
 332         if (rt->rt_dst.sa_family != AF_INET)
 333                 return -EAFNOSUPPORT;
 334
 335         /*
 336          * Check mask for validity:
 337          * a) it must be contiguous.
 338          * b) destination must have all host bits clear.
 339          * c) if application forgot to set correct family (AF_INET),
 340          *    reject request unless it is absolutely clear i.e.
 341          *    both family and mask are zero.
 342          */
 343         plen = 32;
 344         addr = sk_extract_addr(&rt->rt_dst);
 345         if (!(rt->rt_flags & RTF_HOST)) {
 346                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
 347
 348                 if (rt->rt_genmask.sa_family != AF_INET) {
 349                         if (mask || rt->rt_genmask.sa_family)
 350                                 return -EAFNOSUPPORT;
 351                 }
 352
 353                 if (bad_mask(mask, addr))
 354                         return -EINVAL;
 355
 356                 plen = inet_mask_len(mask);
 357         }
 358
 359         cfg->fc_dst_len = plen;
 360         cfg->fc_dst = addr;
 361
 362         if (cmd != SIOCDELRT) {
 363                 cfg->fc_nlflags = NLM_F_CREATE;
 364                 cfg->fc_protocol = RTPROT_BOOT;
 365         }
 366
 367         if (rt->rt_metric)
 368                 cfg->fc_priority = rt->rt_metric - 1;
 369
 370         if (rt->rt_flags & RTF_REJECT) {
 371                 cfg->fc_scope = RT_SCOPE_HOST;
 372                 cfg->fc_type = RTN_UNREACHABLE;
 373                 return 0;
 374         }
 375
 376         cfg->fc_scope = RT_SCOPE_NOWHERE;
 377         cfg->fc_type = RTN_UNICAST;
 378
 379         if (rt->rt_dev) {
 380                 char *colon;
 381                 struct net_device *dev;
 382                 char devname[IFNAMSIZ];
 383
 384                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
 385                         return -EFAULT;
 386
 387                 devname[IFNAMSIZ-1] = 0;
 388                 colon = strchr(devname, ':');
 389                 if (colon)
 390                         *colon = 0;
 391                 dev = __dev_get_by_name(net, devname);
 392                 if (!dev)
 393                         return -ENODEV;
 394                 cfg->fc_oif = dev->ifindex;
 395                 if (colon) {
 396                         struct in_ifaddr *ifa;
 397                         struct in_device *in_dev = __in_dev_get_rtnl(dev);
 398                         if (!in_dev)
 399                                 return -ENODEV;
 400                         *colon = ':';
 401                         for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
 402                                 if (strcmp(ifa->ifa_label, devname) == 0)
 403                                         break;
 404                         if (ifa == NULL)
 405                                 return -ENODEV;
 406                         cfg->fc_prefsrc = ifa->ifa_local;
 407                 }
 408         }
 409
 410         addr = sk_extract_addr(&rt->rt_gateway);
 411         if (rt->rt_gateway.sa_family == AF_INET && addr) {
 412                 cfg->fc_gw = addr;
 413                 if (rt->rt_flags & RTF_GATEWAY &&
 414                     inet_addr_type(net, addr) == RTN_UNICAST)
 415                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
 416         }
 417
 418         if (cmd == SIOCDELRT)
 419                 return 0;
 420
 421         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
 422                 return -EINVAL;
 423
 424         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 425                 cfg->fc_scope = RT_SCOPE_LINK;
 426
 427         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 428                 struct nlattr *mx;
 429                 int len = 0;
 430
 431                 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
 432                 if (mx == NULL)
 433                         return -ENOMEM;
 434
 435                 if (rt->rt_flags & RTF_MTU)
 436                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 437
 438                 if (rt->rt_flags & RTF_WINDOW)
 439                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 440
 441                 if (rt->rt_flags & RTF_IRTT)
 442                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 443
 444                 cfg->fc_mx = mx;
 445                 cfg->fc_mx_len = len;
 446         }
 447
 448         return 0;
 449 }
 450
 451 /*
 452  * Handle IP routing ioctl calls.
 453  * These are used to manipulate the routing tables
 454  */
 455 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
 456 {
 457         struct fib_config cfg;
 458         struct rtentry rt;
 459         int err;
 460
 461         switch (cmd) {
 462         case SIOCADDRT:         /* Add a route */
 463         case SIOCDELRT:         /* Delete a route */
 464                 if (!capable(CAP_NET_ADMIN))
 465                         return -EPERM;
 466
 467                 if (copy_from_user(&rt, arg, sizeof(rt)))
 468                         return -EFAULT;
 469
 470                 rtnl_lock();
 471                 err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
 472                 if (err == 0) {
 473                         struct fib_table *tb;
 474
 475                         if (cmd == SIOCDELRT) {
 476                                 tb = fib_get_table(net, cfg.fc_table);
 477                                 if (tb)
 478                                         err = fib_table_delete(tb, &cfg);
 479                                 else
 480                                         err = -ESRCH;
 481                         } else {
 482                                 tb = fib_new_table(net, cfg.fc_table);
 483                                 if (tb)
 484                                         err = fib_table_insert(tb, &cfg);
 485                                 else
 486                                         err = -ENOBUFS;
 487                         }
 488
 489                         /* allocated by rtentry_to_fib_config() */
 490                         kfree(cfg.fc_mx);
 491                 }
 492                 rtnl_unlock();
 493                 return err;
 494         }
 495         return -EINVAL;
 496 }
 497
 498 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 499         [RTA_DST]               = { .type = NLA_U32 },
 500         [RTA_SRC]               = { .type = NLA_U32 },
 501         [RTA_IIF]               = { .type = NLA_U32 },
 502         [RTA_OIF]               = { .type = NLA_U32 },
 503         [RTA_GATEWAY]           = { .type = NLA_U32 },
 504         [RTA_PRIORITY]          = { .type = NLA_U32 },
 505         [RTA_PREFSRC]           = { .type = NLA_U32 },
 506         [RTA_METRICS]           = { .type = NLA_NESTED },
 507         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 508         [RTA_FLOW]              = { .type = NLA_U32 },
 509 };
 510
 511 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 512                              struct nlmsghdr *nlh, struct fib_config *cfg)
 513 {
 514         struct nlattr *attr;
 515         int err, remaining;
 516         struct rtmsg *rtm;
 517
 518         err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
 519         if (err < 0)
 520                 goto errout;
 521
 522         memset(cfg, 0, sizeof(*cfg));
 523
 524         rtm = nlmsg_data(nlh);
 525         cfg->fc_dst_len = rtm->rtm_dst_len;
 526         cfg->fc_tos = rtm->rtm_tos;
 527         cfg->fc_table = rtm->rtm_table;
 528         cfg->fc_protocol = rtm->rtm_protocol;
 529         cfg->fc_scope = rtm->rtm_scope;
 530         cfg->fc_type = rtm->rtm_type;
 531         cfg->fc_flags = rtm->rtm_flags;
 532         cfg->fc_nlflags = nlh->nlmsg_flags;
 533
 534         cfg->fc_nlinfo.pid = NETLINK_CB(skb).pid;
 535         cfg->fc_nlinfo.nlh = nlh;
 536         cfg->fc_nlinfo.nl_net = net;
 537
 538         if (cfg->fc_type > RTN_MAX) {
 539                 err = -EINVAL;
 540                 goto errout;
 541         }
 542
 543         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 544                 switch (nla_type(attr)) {
 545                 case RTA_DST:
 546                         cfg->fc_dst = nla_get_be32(attr);
 547                         break;
 548                 case RTA_OIF:
 549                         cfg->fc_oif = nla_get_u32(attr);
 550                         break;
 551                 case RTA_GATEWAY:
 552                         cfg->fc_gw = nla_get_be32(attr);
 553                         break;
 554                 case RTA_PRIORITY:
 555                         cfg->fc_priority = nla_get_u32(attr);
 556                         break;
 557                 case RTA_PREFSRC:
 558                         cfg->fc_prefsrc = nla_get_be32(attr);
 559                         break;
 560                 case RTA_METRICS:
 561                         cfg->fc_mx = nla_data(attr);
 562                         cfg->fc_mx_len = nla_len(attr);
 563                         break;
 564                 case RTA_MULTIPATH:
 565                         cfg->fc_mp = nla_data(attr);
 566                         cfg->fc_mp_len = nla_len(attr);
 567                         break;
 568                 case RTA_FLOW:
 569                         cfg->fc_flow = nla_get_u32(attr);
 570                         break;
 571                 case RTA_TABLE:
 572                         cfg->fc_table = nla_get_u32(attr);
 573                         break;
 574                 }
 575         }
 576
 577         return 0;
 578 errout:
 579         return err;
 580 }
 581
 582 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 583 {
 584         struct net *net = sock_net(skb->sk);
 585         struct fib_config cfg;
 586         struct fib_table *tb;
 587         int err;
 588
 589         err = rtm_to_fib_config(net, skb, nlh, &cfg);
 590         if (err < 0)
 591                 goto errout;
 592
 593         tb = fib_get_table(net, cfg.fc_table);
 594         if (tb == NULL) {
 595                 err = -ESRCH;
 596                 goto errout;
 597         }
 598
 599         err = fib_table_delete(tb, &cfg);
 600 errout:
 601         return err;
 602 }
 603
 604 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 605 {
 606         struct net *net = sock_net(skb->sk);
 607         struct fib_config cfg;
 608         struct fib_table *tb;
 609         int err;
 610
 611         err = rtm_to_fib_config(net, skb, nlh, &cfg);
 612         if (err < 0)
 613                 goto errout;
 614
 615         tb = fib_new_table(net, cfg.fc_table);
 616         if (tb == NULL) {
 617                 err = -ENOBUFS;
 618                 goto errout;
 619         }
 620
 621         err = fib_table_insert(tb, &cfg);
 622 errout:
 623         return err;
 624 }
 625
 626 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 627 {
 628         struct net *net = sock_net(skb->sk);
 629         unsigned int h, s_h;
 630         unsigned int e = 0, s_e;
 631         struct fib_table *tb;
 632         struct hlist_node *node;
 633         struct hlist_head *head;
 634         int dumped = 0;
 635
 636         if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
 637             ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
 638                 return ip_rt_dump(skb, cb);
 639
 640         s_h = cb->args[0];
 641         s_e = cb->args[1];
 642
 643         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
 644                 e = 0;
 645                 head = &net->ipv4.fib_table_hash[h];
 646                 hlist_for_each_entry(tb, node, head, tb_hlist) {
 647                         if (e < s_e)
 648                                 goto next;
 649                         if (dumped)
 650                                 memset(&cb->args[2], 0, sizeof(cb->args) -
 651                                                  2 * sizeof(cb->args[0]));
 652                         if (fib_table_dump(tb, skb, cb) < 0)
 653                                 goto out;
 654                         dumped = 1;
 655 next:
 656                         e++;
 657                 }
 658         }
 659 out:
 660         cb->args[1] = e;
 661         cb->args[0] = h;
 662
 663         return skb->len;
 664 }
 665
 666 /* Prepare and feed intra-kernel routing request.
 667  * Really, it should be netlink message, but :-( netlink
 668  * can be not configured, so that we feed it directly
 669  * to fib engine. It is legal, because all events occur
 670  * only when netlink is already locked.
 671  */
 672 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
 673 {
 674         struct net *net = dev_net(ifa->ifa_dev->dev);
 675         struct fib_table *tb;
 676         struct fib_config cfg = {
 677                 .fc_protocol = RTPROT_KERNEL,
 678                 .fc_type = type,
 679                 .fc_dst = dst,
 680                 .fc_dst_len = dst_len,
 681                 .fc_prefsrc = ifa->ifa_local,
 682                 .fc_oif = ifa->ifa_dev->dev->ifindex,
 683                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
 684                 .fc_nlinfo = {
 685                         .nl_net = net,
 686                 },
 687         };
 688
 689         if (type == RTN_UNICAST)
 690                 tb = fib_new_table(net, RT_TABLE_MAIN);
 691         else
 692                 tb = fib_new_table(net, RT_TABLE_LOCAL);
 693
 694         if (tb == NULL)
 695                 return;
 696
 697         cfg.fc_table = tb->tb_id;
 698
 699         if (type != RTN_LOCAL)
 700                 cfg.fc_scope = RT_SCOPE_LINK;
 701         else
 702                 cfg.fc_scope = RT_SCOPE_HOST;
 703
 704         if (cmd == RTM_NEWROUTE)
 705                 fib_table_insert(tb, &cfg);
 706         else
 707                 fib_table_delete(tb, &cfg);
 708 }
 709
 710 void fib_add_ifaddr(struct in_ifaddr *ifa)
 711 {
 712         struct in_device *in_dev = ifa->ifa_dev;
 713         struct net_device *dev = in_dev->dev;
 714         struct in_ifaddr *prim = ifa;
 715         __be32 mask = ifa->ifa_mask;
 716         __be32 addr = ifa->ifa_local;
 717         __be32 prefix = ifa->ifa_address & mask;
 718
 719         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 720                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
 721                 if (prim == NULL) {
 722                         pr_warn("%s: bug: prim == NULL\n", __func__);
 723                         return;
 724                 }
 725         }
 726
 727         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
 728
 729         if (!(dev->flags & IFF_UP))
 730                 return;
 731
 732         /* Add broadcast address, if it is explicitly assigned. */
 733         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
 734                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
 735
 736         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
 737             (prefix != addr || ifa->ifa_prefixlen < 32)) {
 738                 fib_magic(RTM_NEWROUTE,
 739                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 740                           prefix, ifa->ifa_prefixlen, prim);
 741
 742                 /* Add network specific broadcasts, when it takes a sense */
 743                 if (ifa->ifa_prefixlen < 31) {
 744                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
 745                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
 746                                   32, prim);
 747                 }
 748         }
 749 }
 750
 751 /* Delete primary or secondary address.
 752  * Optionally, on secondary address promotion consider the addresses
 753  * from subnet iprim as deleted, even if they are in device list.
 754  * In this case the secondary ifa can be in device list.
 755  */
 756 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
 757 {
 758         struct in_device *in_dev = ifa->ifa_dev;
 759         struct net_device *dev = in_dev->dev;
 760         struct in_ifaddr *ifa1;
 761         struct in_ifaddr *prim = ifa, *prim1 = NULL;
 762         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
 763         __be32 any = ifa->ifa_address & ifa->ifa_mask;
 764 #define LOCAL_OK        1
 765 #define BRD_OK          2
 766 #define BRD0_OK         4
 767 #define BRD1_OK         8
 768         unsigned int ok = 0;
 769         int subnet = 0;         /* Primary network */
 770         int gone = 1;           /* Address is missing */
 771         int same_prefsrc = 0;   /* Another primary with same IP */
 772
 773         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 774                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
 775                 if (prim == NULL) {
 776                         pr_warn("%s: bug: prim == NULL\n", __func__);
 777                         return;
 778                 }
 779                 if (iprim && iprim != prim) {
 780                         pr_warn("%s: bug: iprim != prim\n", __func__);
 781                         return;
 782                 }
 783         } else if (!ipv4_is_zeronet(any) &&
 784                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
 785                 fib_magic(RTM_DELROUTE,
 786                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 787                           any, ifa->ifa_prefixlen, prim);
 788                 subnet = 1;
 789         }
 790
 791         /* Deletion is more complicated than add.
 792          * We should take care of not to delete too much :-)
 793          *
 794          * Scan address list to be sure that addresses are really gone.
 795          */
 796
 797         for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
 798                 if (ifa1 == ifa) {
 799                         /* promotion, keep the IP */
 800                         gone = 0;
 801                         continue;
 802                 }
 803                 /* Ignore IFAs from our subnet */
 804                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
 805                     inet_ifa_match(ifa1->ifa_address, iprim))
 806                         continue;
 807
 808                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
 809                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
 810                         /* Another address from our subnet? */
 811                         if (ifa1->ifa_mask == prim->ifa_mask &&
 812                             inet_ifa_match(ifa1->ifa_address, prim))
 813                                 prim1 = prim;
 814                         else {
 815                                 /* We reached the secondaries, so
 816                                  * same_prefsrc should be determined.
 817                                  */
 818                                 if (!same_prefsrc)
 819                                         continue;
 820                                 /* Search new prim1 if ifa1 is not
 821                                  * using the current prim1
 822                                  */
 823                                 if (!prim1 ||
 824                                     ifa1->ifa_mask != prim1->ifa_mask ||
 825                                     !inet_ifa_match(ifa1->ifa_address, prim1))
 826                                         prim1 = inet_ifa_byprefix(in_dev,
 827                                                         ifa1->ifa_address,
 828                                                         ifa1->ifa_mask);
 829                                 if (!prim1)
 830                                         continue;
 831                                 if (prim1->ifa_local != prim->ifa_local)
 832                                         continue;
 833                         }
 834                 } else {
 835                         if (prim->ifa_local != ifa1->ifa_local)
 836                                 continue;
 837                         prim1 = ifa1;
 838                         if (prim != prim1)
 839                                 same_prefsrc = 1;
 840                 }
 841                 if (ifa->ifa_local == ifa1->ifa_local)
 842                         ok |= LOCAL_OK;
 843                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
 844                         ok |= BRD_OK;
 845                 if (brd == ifa1->ifa_broadcast)
 846                         ok |= BRD1_OK;
 847                 if (any == ifa1->ifa_broadcast)
 848                         ok |= BRD0_OK;
 849                 /* primary has network specific broadcasts */
 850                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
 851                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
 852                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
 853
 854                         if (!ipv4_is_zeronet(any1)) {
 855                                 if (ifa->ifa_broadcast == brd1 ||
 856                                     ifa->ifa_broadcast == any1)
 857                                         ok |= BRD_OK;
 858                                 if (brd == brd1 || brd == any1)
 859                                         ok |= BRD1_OK;
 860                                 if (any == brd1 || any == any1)
 861                                         ok |= BRD0_OK;
 862                         }
 863                 }
 864         }
 865
 866         if (!(ok & BRD_OK))
 867                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
 868         if (subnet && ifa->ifa_prefixlen < 31) {
 869                 if (!(ok & BRD1_OK))
 870                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
 871                 if (!(ok & BRD0_OK))
 872                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
 873         }
 874         if (!(ok & LOCAL_OK)) {
 875                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
 876
 877                 /* Check, that this local address finally disappeared. */
 878                 if (gone &&
 879                     inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) {
 880                         /* And the last, but not the least thing.
 881                          * We must flush stray FIB entries.
 882                          *
 883                          * First of all, we scan fib_info list searching
 884                          * for stray nexthop entries, then ignite fib_flush.
 885                          */
 886                         if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
 887                                 fib_flush(dev_net(dev));
 888                 }
 889         }
 890 #undef LOCAL_OK
 891 #undef BRD_OK
 892 #undef BRD0_OK
 893 #undef BRD1_OK
 894 }
 895
 896 static void nl_fib_lookup(struct fib_result_nl *frn, struct fib_table *tb)
 897 {
 898
 899         struct fib_result       res;
 900         struct flowi4           fl4 = {
 901                 .flowi4_mark = frn->fl_mark,
 902                 .daddr = frn->fl_addr,
 903                 .flowi4_tos = frn->fl_tos,
 904                 .flowi4_scope = frn->fl_scope,
 905         };
 906
 907 #ifdef CONFIG_IP_MULTIPLE_TABLES
 908         res.r = NULL;
 909 #endif
 910
 911         frn->err = -ENOENT;
 912         if (tb) {
 913                 local_bh_disable();
 914
 915                 frn->tb_id = tb->tb_id;
 916                 rcu_read_lock();
 917                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
 918
 919                 if (!frn->err) {
 920                         frn->prefixlen = res.prefixlen;
 921                         frn->nh_sel = res.nh_sel;
 922                         frn->type = res.type;
 923                         frn->scope = res.scope;
 924                 }
 925                 rcu_read_unlock();
 926                 local_bh_enable();
 927         }
 928 }
 929
 930 static void nl_fib_input(struct sk_buff *skb)
 931 {
 932         struct net *net;
 933         struct fib_result_nl *frn;
 934         struct nlmsghdr *nlh;
 935         struct fib_table *tb;
 936         u32 pid;
 937
 938         net = sock_net(skb->sk);
 939         nlh = nlmsg_hdr(skb);
 940         if (skb->len < NLMSG_SPACE(0) || skb->len < nlh->nlmsg_len ||
 941             nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*frn)))
 942                 return;
 943
 944         skb = skb_clone(skb, GFP_KERNEL);
 945         if (skb == NULL)
 946                 return;
 947         nlh = nlmsg_hdr(skb);
 948
 949         frn = (struct fib_result_nl *) NLMSG_DATA(nlh);
 950         tb = fib_get_table(net, frn->tb_id_in);
 951
 952         nl_fib_lookup(frn, tb);
 953
 954         pid = NETLINK_CB(skb).pid;      /* pid of sending process */
 955         NETLINK_CB(skb).pid = 0;        /* from kernel */
 956         NETLINK_CB(skb).dst_group = 0;  /* unicast */
 957         netlink_unicast(net->ipv4.fibnl, skb, pid, MSG_DONTWAIT);
 958 }
 959
 960 static int __net_init nl_fib_lookup_init(struct net *net)
 961 {
 962         struct sock *sk;
 963         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, 0,
 964                                    nl_fib_input, NULL, THIS_MODULE);
 965         if (sk == NULL)
 966                 return -EAFNOSUPPORT;
 967         net->ipv4.fibnl = sk;
 968         return 0;
 969 }
 970
 971 static void nl_fib_lookup_exit(struct net *net)
 972 {
 973         netlink_kernel_release(net->ipv4.fibnl);
 974         net->ipv4.fibnl = NULL;
 975 }
 976
 977 static void fib_disable_ip(struct net_device *dev, int force, int delay)
 978 {
 979         if (fib_sync_down_dev(dev, force))
 980                 fib_flush(dev_net(dev));
 981         rt_cache_flush(dev_net(dev), delay);
 982         arp_ifdown(dev);
 983 }
 984
 985 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
 986 {
 987         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
 988         struct net_device *dev = ifa->ifa_dev->dev;
 989         struct net *net = dev_net(dev);
 990
 991         switch (event) {
 992         case NETDEV_UP:
 993                 fib_add_ifaddr(ifa);
 994 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 995                 fib_sync_up(dev);
 996 #endif
 997                 atomic_inc(&net->ipv4.dev_addr_genid);
 998                 rt_cache_flush(dev_net(dev), -1);
 999                 break;
1000         case NETDEV_DOWN:
1001                 fib_del_ifaddr(ifa, NULL);
1002                 atomic_inc(&net->ipv4.dev_addr_genid);
1003                 if (ifa->ifa_dev->ifa_list == NULL) {
1004                         /* Last address was deleted from this interface.
1005                          * Disable IP.
1006                          */
1007                         fib_disable_ip(dev, 1, 0);
1008                 } else {
1009                         rt_cache_flush(dev_net(dev), -1);
1010                 }
1011                 break;
1012         }
1013         return NOTIFY_DONE;
1014 }
1015
1016 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1017 {
1018         struct net_device *dev = ptr;
1019         struct in_device *in_dev = __in_dev_get_rtnl(dev);
1020         struct net *net = dev_net(dev);
1021
1022         if (event == NETDEV_UNREGISTER) {
1023                 fib_disable_ip(dev, 2, -1);
1024                 return NOTIFY_DONE;
1025         }
1026
1027         if (!in_dev)
1028                 return NOTIFY_DONE;
1029
1030         switch (event) {
1031         case NETDEV_UP:
1032                 for_ifa(in_dev) {
1033                         fib_add_ifaddr(ifa);
1034                 } endfor_ifa(in_dev);
1035 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1036                 fib_sync_up(dev);
1037 #endif
1038                 atomic_inc(&net->ipv4.dev_addr_genid);
1039                 rt_cache_flush(dev_net(dev), -1);
1040                 break;
1041         case NETDEV_DOWN:
1042                 fib_disable_ip(dev, 0, 0);
1043                 break;
1044         case NETDEV_CHANGEMTU:
1045         case NETDEV_CHANGE:
1046                 rt_cache_flush(dev_net(dev), 0);
1047                 break;
1048         case NETDEV_UNREGISTER_BATCH:
1049                 /* The batch unregister is only called on the first
1050                  * device in the list of devices being unregistered.
1051                  * Therefore we should not pass dev_net(dev) in here.
1052                  */
1053                 rt_cache_flush_batch(NULL);
1054                 break;
1055         }
1056         return NOTIFY_DONE;
1057 }
1058
1059 static struct notifier_block fib_inetaddr_notifier = {
1060         .notifier_call = fib_inetaddr_event,
1061 };
1062
1063 static struct notifier_block fib_netdev_notifier = {
1064         .notifier_call = fib_netdev_event,
1065 };
1066
1067 static int __net_init ip_fib_net_init(struct net *net)
1068 {
1069         int err;
1070         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1071
1072         /* Avoid false sharing : Use at least a full cache line */
1073         size = max_t(size_t, size, L1_CACHE_BYTES);
1074
1075         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1076         if (net->ipv4.fib_table_hash == NULL)
1077                 return -ENOMEM;
1078
1079         err = fib4_rules_init(net);
1080         if (err < 0)
1081                 goto fail;
1082         return 0;
1083
1084 fail:
1085         kfree(net->ipv4.fib_table_hash);
1086         return err;
1087 }
1088
1089 static void ip_fib_net_exit(struct net *net)
1090 {
1091         unsigned int i;
1092
1093 #ifdef CONFIG_IP_MULTIPLE_TABLES
1094         fib4_rules_exit(net);
1095 #endif
1096
1097         rtnl_lock();
1098         for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1099                 struct fib_table *tb;
1100                 struct hlist_head *head;
1101                 struct hlist_node *node, *tmp;
1102
1103                 head = &net->ipv4.fib_table_hash[i];
1104                 hlist_for_each_entry_safe(tb, node, tmp, head, tb_hlist) {
1105                         hlist_del(node);
1106                         fib_table_flush(tb);
1107                         fib_free_table(tb);
1108                 }
1109         }
1110         rtnl_unlock();
1111         kfree(net->ipv4.fib_table_hash);
1112 }
1113
1114 static int __net_init fib_net_init(struct net *net)
1115 {
1116         int error;
1117
1118         error = ip_fib_net_init(net);
1119         if (error < 0)
1120                 goto out;
1121         error = nl_fib_lookup_init(net);
1122         if (error < 0)
1123                 goto out_nlfl;
1124         error = fib_proc_init(net);
1125         if (error < 0)
1126                 goto out_proc;
1127 out:
1128         return error;
1129
1130 out_proc:
1131         nl_fib_lookup_exit(net);
1132 out_nlfl:
1133         ip_fib_net_exit(net);
1134         goto out;
1135 }
1136
1137 static void __net_exit fib_net_exit(struct net *net)
1138 {
1139         fib_proc_exit(net);
1140         nl_fib_lookup_exit(net);
1141         ip_fib_net_exit(net);
1142 }
1143
1144 static struct pernet_operations fib_net_ops = {
1145         .init = fib_net_init,
1146         .exit = fib_net_exit,
1147 };
1148
1149 void __init ip_fib_init(void)
1150 {
1151         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1152         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1153         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1154
1155         register_pernet_subsys(&fib_net_ops);
1156         register_netdevice_notifier(&fib_netdev_notifier);
1157         register_inetaddr_notifier(&fib_inetaddr_notifier);
1158
1159         fib_trie_init();
1160 }