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/cache.h>
  35 #include <linux/init.h>
  36 #include <linux/list.h>
  37 #include <linux/slab.h>
  38
  39 #include <net/ip.h>
  40 #include <net/protocol.h>
  41 #include <net/route.h>
  42 #include <net/tcp.h>
  43 #include <net/sock.h>
  44 #include <net/arp.h>
  45 #include <net/ip_fib.h>
  46 #include <net/rtnetlink.h>
  47 #include <net/xfrm.h>
  48
  49 #ifndef CONFIG_IP_MULTIPLE_TABLES
  50
  51 static int __net_init fib4_rules_init(struct net *net)
  52 {
  53         struct fib_table *local_table, *main_table;
  54
  55         local_table = fib_trie_table(RT_TABLE_LOCAL);
  56         if (local_table == NULL)
  57                 return -ENOMEM;
  58
  59         main_table  = fib_trie_table(RT_TABLE_MAIN);
  60         if (main_table == NULL)
  61                 goto fail;
  62
  63         hlist_add_head_rcu(&local_table->tb_hlist,
  64                                 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
  65         hlist_add_head_rcu(&main_table->tb_hlist,
  66                                 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
  67         return 0;
  68
  69 fail:
  70         kfree(local_table);
  71         return -ENOMEM;
  72 }
  73 #else
  74
  75 struct fib_table *fib_new_table(struct net *net, u32 id)
  76 {
  77         struct fib_table *tb;
  78         unsigned int h;
  79
  80         if (id == 0)
  81                 id = RT_TABLE_MAIN;
  82         tb = fib_get_table(net, id);
  83         if (tb)
  84                 return tb;
  85
  86         tb = fib_trie_table(id);
  87         if (!tb)
  88                 return NULL;
  89         h = id & (FIB_TABLE_HASHSZ - 1);
  90         hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
  91         return tb;
  92 }
  93
  94 struct fib_table *fib_get_table(struct net *net, u32 id)
  95 {
  96         struct fib_table *tb;
  97         struct hlist_node *node;
  98         struct hlist_head *head;
  99         unsigned int h;
 100
 101         if (id == 0)
 102                 id = RT_TABLE_MAIN;
 103         h = id & (FIB_TABLE_HASHSZ - 1);
 104
 105         rcu_read_lock();
 106         head = &net->ipv4.fib_table_hash[h];
 107         hlist_for_each_entry_rcu(tb, node, head, tb_hlist) {
 108                 if (tb->tb_id == id) {
 109                         rcu_read_unlock();
 110                         return tb;
 111                 }
 112         }
 113         rcu_read_unlock();
 114         return NULL;
 115 }
 116 #endif /* CONFIG_IP_MULTIPLE_TABLES */
 117
 118 static void fib_flush(struct net *net)
 119 {
 120         int flushed = 0;
 121         struct fib_table *tb;
 122         struct hlist_node *node;
 123         struct hlist_head *head;
 124         unsigned int h;
 125
 126         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
 127                 head = &net->ipv4.fib_table_hash[h];
 128                 hlist_for_each_entry(tb, node, head, tb_hlist)
 129                         flushed += fib_table_flush(tb);
 130         }
 131
 132         if (flushed)
 133                 rt_cache_flush(net, -1);
 134 }
 135
 136 /*
 137  * Find address type as if only "dev" was present in the system. If
 138  * on_dev is NULL then all interfaces are taken into consideration.
 139  */
 140 static inline unsigned int __inet_dev_addr_type(struct net *net,
 141                                                 const struct net_device *dev,
 142                                                 __be32 addr)
 143 {
 144         struct flowi4           fl4 = { .daddr = addr };
 145         struct fib_result       res;
 146         unsigned int ret = RTN_BROADCAST;
 147         struct fib_table *local_table;
 148
 149         if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
 150                 return RTN_BROADCAST;
 151         if (ipv4_is_multicast(addr))
 152                 return RTN_MULTICAST;
 153
 154 #ifdef CONFIG_IP_MULTIPLE_TABLES
 155         res.r = NULL;
 156 #endif
 157
 158         local_table = fib_get_table(net, RT_TABLE_LOCAL);
 159         if (local_table) {
 160                 ret = RTN_UNICAST;
 161                 rcu_read_lock();
 162                 if (!fib_table_lookup(local_table, &fl4, &res, FIB_LOOKUP_NOREF)) {
 163                         if (!dev || dev == res.fi->fib_dev)
 164                                 ret = res.type;
 165                 }
 166                 rcu_read_unlock();
 167         }
 168         return ret;
 169 }
 170
 171 unsigned int inet_addr_type(struct net *net, __be32 addr)
 172 {
 173         return __inet_dev_addr_type(net, NULL, addr);
 174 }
 175 EXPORT_SYMBOL(inet_addr_type);
 176
 177 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
 178                                 __be32 addr)
 179 {
 180         return __inet_dev_addr_type(net, dev, addr);
 181 }
 182 EXPORT_SYMBOL(inet_dev_addr_type);
 183
 184 __be32 fib_compute_spec_dst(struct sk_buff *skb)
 185 {
 186         struct net_device *dev = skb->dev;
 187         struct in_device *in_dev;
 188         struct fib_result res;
 189         struct rtable *rt;
 190         struct flowi4 fl4;
 191         struct net *net;
 192         int scope;
 193
 194         rt = skb_rtable(skb);
 195         if (!(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)))
 196                 return ip_hdr(skb)->daddr;
 197
 198         in_dev = __in_dev_get_rcu(dev);
 199         BUG_ON(!in_dev);
 200
 201         net = dev_net(dev);
 202
 203         scope = RT_SCOPE_UNIVERSE;
 204         if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
 205                 fl4.flowi4_oif = 0;
 206                 fl4.flowi4_iif = net->loopback_dev->ifindex;
 207                 fl4.daddr = ip_hdr(skb)->saddr;
 208                 fl4.saddr = 0;
 209                 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
 210                 fl4.flowi4_scope = scope;
 211                 fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0;
 212                 if (!fib_lookup(net, &fl4, &res))
 213                         return FIB_RES_PREFSRC(net, res);
 214         } else {
 215                 scope = RT_SCOPE_LINK;
 216         }
 217
 218         return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
 219 }
 220
 221 #ifdef CONFIG_IP_ROUTE_CLASSID
 222 int fib_num_tclassid_users __read_mostly;
 223 #endif
 224
 225 /* Given (packet source, input interface) and optional (dst, oif, tos):
 226  * - (main) check, that source is valid i.e. not broadcast or our local
 227  *   address.
 228  * - figure out what "logical" interface this packet arrived
 229  *   and calculate "specific destination" address.
 230  * - check, that packet arrived from expected physical interface.
 231  * called with rcu_read_lock()
 232  */
 233 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 234                                  u8 tos, int oif, struct net_device *dev,
 235                                  int rpf, struct in_device *idev, u32 *itag)
 236 {
 237         int ret, no_addr, accept_local;
 238         struct fib_result res;
 239         struct flowi4 fl4;
 240         struct net *net;
 241         bool dev_match;
 242
 243         fl4.flowi4_oif = 0;
 244         fl4.flowi4_iif = oif;
 245         fl4.daddr = src;
 246         fl4.saddr = dst;
 247         fl4.flowi4_tos = tos;
 248         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
 249
 250         no_addr = accept_local = 0;
 251         no_addr = idev->ifa_list == NULL;
 252
 253         accept_local = IN_DEV_ACCEPT_LOCAL(idev);
 254         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
 255
 256         net = dev_net(dev);
 257         if (fib_lookup(net, &fl4, &res))
 258                 goto last_resort;
 259         if (res.type != RTN_UNICAST) {
 260                 if (res.type != RTN_LOCAL || !accept_local)
 261                         goto e_inval;
 262         }
 263         fib_combine_itag(itag, &res);
 264         dev_match = false;
 265
 266 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 267         for (ret = 0; ret < res.fi->fib_nhs; ret++) {
 268                 struct fib_nh *nh = &res.fi->fib_nh[ret];
 269
 270                 if (nh->nh_dev == dev) {
 271                         dev_match = true;
 272                         break;
 273                 }
 274         }
 275 #else
 276         if (FIB_RES_DEV(res) == dev)
 277                 dev_match = true;
 278 #endif
 279         if (dev_match) {
 280                 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 281                 return ret;
 282         }
 283         if (no_addr)
 284                 goto last_resort;
 285         if (rpf == 1)
 286                 goto e_rpf;
 287         fl4.flowi4_oif = dev->ifindex;
 288
 289         ret = 0;
 290         if (fib_lookup(net, &fl4, &res) == 0) {
 291                 if (res.type == RTN_UNICAST)
 292                         ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
 293         }
 294         return ret;
 295
 296 last_resort:
 297         if (rpf)
 298                 goto e_rpf;
 299         *itag = 0;
 300         return 0;
 301
 302 e_inval:
 303         return -EINVAL;
 304 e_rpf:
 305         return -EXDEV;
 306 }
 307
 308 /* Ignore rp_filter for packets protected by IPsec. */
 309 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
 310                         u8 tos, int oif, struct net_device *dev,
 311                         struct in_device *idev, u32 *itag)
 312 {
 313         int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
 314
 315         if (!r && !fib_num_tclassid_users) {
 316                 *itag = 0;
 317                 return 0;
 318         }
 319         return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
 320 }
 321
 322 static inline __be32 sk_extract_addr(struct sockaddr *addr)
 323 {
 324         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
 325 }
 326
 327 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
 328 {
 329         struct nlattr *nla;
 330
 331         nla = (struct nlattr *) ((char *) mx + len);
 332         nla->nla_type = type;
 333         nla->nla_len = nla_attr_size(4);
 334         *(u32 *) nla_data(nla) = value;
 335
 336         return len + nla_total_size(4);
 337 }
 338
 339 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
 340                                  struct fib_config *cfg)
 341 {
 342         __be32 addr;
 343         int plen;
 344
 345         memset(cfg, 0, sizeof(*cfg));
 346         cfg->fc_nlinfo.nl_net = net;
 347
 348         if (rt->rt_dst.sa_family != AF_INET)
 349                 return -EAFNOSUPPORT;
 350
 351         /*
 352          * Check mask for validity:
 353          * a) it must be contiguous.
 354          * b) destination must have all host bits clear.
 355          * c) if application forgot to set correct family (AF_INET),
 356          *    reject request unless it is absolutely clear i.e.
 357          *    both family and mask are zero.
 358          */
 359         plen = 32;
 360         addr = sk_extract_addr(&rt->rt_dst);
 361         if (!(rt->rt_flags & RTF_HOST)) {
 362                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
 363
 364                 if (rt->rt_genmask.sa_family != AF_INET) {
 365                         if (mask || rt->rt_genmask.sa_family)
 366                                 return -EAFNOSUPPORT;
 367                 }
 368
 369                 if (bad_mask(mask, addr))
 370                         return -EINVAL;
 371
 372                 plen = inet_mask_len(mask);
 373         }
 374
 375         cfg->fc_dst_len = plen;
 376         cfg->fc_dst = addr;
 377
 378         if (cmd != SIOCDELRT) {
 379                 cfg->fc_nlflags = NLM_F_CREATE;
 380                 cfg->fc_protocol = RTPROT_BOOT;
 381         }
 382
 383         if (rt->rt_metric)
 384                 cfg->fc_priority = rt->rt_metric - 1;
 385
 386         if (rt->rt_flags & RTF_REJECT) {
 387                 cfg->fc_scope = RT_SCOPE_HOST;
 388                 cfg->fc_type = RTN_UNREACHABLE;
 389                 return 0;
 390         }
 391
 392         cfg->fc_scope = RT_SCOPE_NOWHERE;
 393         cfg->fc_type = RTN_UNICAST;
 394
 395         if (rt->rt_dev) {
 396                 char *colon;
 397                 struct net_device *dev;
 398                 char devname[IFNAMSIZ];
 399
 400                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
 401                         return -EFAULT;
 402
 403                 devname[IFNAMSIZ-1] = 0;
 404                 colon = strchr(devname, ':');
 405                 if (colon)
 406                         *colon = 0;
 407                 dev = __dev_get_by_name(net, devname);
 408                 if (!dev)
 409                         return -ENODEV;
 410                 cfg->fc_oif = dev->ifindex;
 411                 if (colon) {
 412                         struct in_ifaddr *ifa;
 413                         struct in_device *in_dev = __in_dev_get_rtnl(dev);
 414                         if (!in_dev)
 415                                 return -ENODEV;
 416                         *colon = ':';
 417                         for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
 418                                 if (strcmp(ifa->ifa_label, devname) == 0)
 419                                         break;
 420                         if (ifa == NULL)
 421                                 return -ENODEV;
 422                         cfg->fc_prefsrc = ifa->ifa_local;
 423                 }
 424         }
 425
 426         addr = sk_extract_addr(&rt->rt_gateway);
 427         if (rt->rt_gateway.sa_family == AF_INET && addr) {
 428                 cfg->fc_gw = addr;
 429                 if (rt->rt_flags & RTF_GATEWAY &&
 430                     inet_addr_type(net, addr) == RTN_UNICAST)
 431                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
 432         }
 433
 434         if (cmd == SIOCDELRT)
 435                 return 0;
 436
 437         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
 438                 return -EINVAL;
 439
 440         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
 441                 cfg->fc_scope = RT_SCOPE_LINK;
 442
 443         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
 444                 struct nlattr *mx;
 445                 int len = 0;
 446
 447                 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
 448                 if (mx == NULL)
 449                         return -ENOMEM;
 450
 451                 if (rt->rt_flags & RTF_MTU)
 452                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
 453
 454                 if (rt->rt_flags & RTF_WINDOW)
 455                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
 456
 457                 if (rt->rt_flags & RTF_IRTT)
 458                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
 459
 460                 cfg->fc_mx = mx;
 461                 cfg->fc_mx_len = len;
 462         }
 463
 464         return 0;
 465 }
 466
 467 /*
 468  * Handle IP routing ioctl calls.
 469  * These are used to manipulate the routing tables
 470  */
 471 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
 472 {
 473         struct fib_config cfg;
 474         struct rtentry rt;
 475         int err;
 476
 477         switch (cmd) {
 478         case SIOCADDRT:         /* Add a route */
 479         case SIOCDELRT:         /* Delete a route */
 480                 if (!capable(CAP_NET_ADMIN))
 481                         return -EPERM;
 482
 483                 if (copy_from_user(&rt, arg, sizeof(rt)))
 484                         return -EFAULT;
 485
 486                 rtnl_lock();
 487                 err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
 488                 if (err == 0) {
 489                         struct fib_table *tb;
 490
 491                         if (cmd == SIOCDELRT) {
 492                                 tb = fib_get_table(net, cfg.fc_table);
 493                                 if (tb)
 494                                         err = fib_table_delete(tb, &cfg);
 495                                 else
 496                                         err = -ESRCH;
 497                         } else {
 498                                 tb = fib_new_table(net, cfg.fc_table);
 499                                 if (tb)
 500                                         err = fib_table_insert(tb, &cfg);
 501                                 else
 502                                         err = -ENOBUFS;
 503                         }
 504
 505                         /* allocated by rtentry_to_fib_config() */
 506                         kfree(cfg.fc_mx);
 507                 }
 508                 rtnl_unlock();
 509                 return err;
 510         }
 511         return -EINVAL;
 512 }
 513
 514 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
 515         [RTA_DST]               = { .type = NLA_U32 },
 516         [RTA_SRC]               = { .type = NLA_U32 },
 517         [RTA_IIF]               = { .type = NLA_U32 },
 518         [RTA_OIF]               = { .type = NLA_U32 },
 519         [RTA_GATEWAY]           = { .type = NLA_U32 },
 520         [RTA_PRIORITY]          = { .type = NLA_U32 },
 521         [RTA_PREFSRC]           = { .type = NLA_U32 },
 522         [RTA_METRICS]           = { .type = NLA_NESTED },
 523         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
 524         [RTA_FLOW]              = { .type = NLA_U32 },
 525 };
 526
 527 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
 528                              struct nlmsghdr *nlh, struct fib_config *cfg)
 529 {
 530         struct nlattr *attr;
 531         int err, remaining;
 532         struct rtmsg *rtm;
 533
 534         err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
 535         if (err < 0)
 536                 goto errout;
 537
 538         memset(cfg, 0, sizeof(*cfg));
 539
 540         rtm = nlmsg_data(nlh);
 541         cfg->fc_dst_len = rtm->rtm_dst_len;
 542         cfg->fc_tos = rtm->rtm_tos;
 543         cfg->fc_table = rtm->rtm_table;
 544         cfg->fc_protocol = rtm->rtm_protocol;
 545         cfg->fc_scope = rtm->rtm_scope;
 546         cfg->fc_type = rtm->rtm_type;
 547         cfg->fc_flags = rtm->rtm_flags;
 548         cfg->fc_nlflags = nlh->nlmsg_flags;
 549
 550         cfg->fc_nlinfo.pid = NETLINK_CB(skb).pid;
 551         cfg->fc_nlinfo.nlh = nlh;
 552         cfg->fc_nlinfo.nl_net = net;
 553
 554         if (cfg->fc_type > RTN_MAX) {
 555                 err = -EINVAL;
 556                 goto errout;
 557         }
 558
 559         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
 560                 switch (nla_type(attr)) {
 561                 case RTA_DST:
 562                         cfg->fc_dst = nla_get_be32(attr);
 563                         break;
 564                 case RTA_OIF:
 565                         cfg->fc_oif = nla_get_u32(attr);
 566                         break;
 567                 case RTA_GATEWAY:
 568                         cfg->fc_gw = nla_get_be32(attr);
 569                         break;
 570                 case RTA_PRIORITY:
 571                         cfg->fc_priority = nla_get_u32(attr);
 572                         break;
 573                 case RTA_PREFSRC:
 574                         cfg->fc_prefsrc = nla_get_be32(attr);
 575                         break;
 576                 case RTA_METRICS:
 577                         cfg->fc_mx = nla_data(attr);
 578                         cfg->fc_mx_len = nla_len(attr);
 579                         break;
 580                 case RTA_MULTIPATH:
 581                         cfg->fc_mp = nla_data(attr);
 582                         cfg->fc_mp_len = nla_len(attr);
 583                         break;
 584                 case RTA_FLOW:
 585                         cfg->fc_flow = nla_get_u32(attr);
 586                         break;
 587                 case RTA_TABLE:
 588                         cfg->fc_table = nla_get_u32(attr);
 589                         break;
 590                 }
 591         }
 592
 593         return 0;
 594 errout:
 595         return err;
 596 }
 597
 598 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 599 {
 600         struct net *net = sock_net(skb->sk);
 601         struct fib_config cfg;
 602         struct fib_table *tb;
 603         int err;
 604
 605         err = rtm_to_fib_config(net, skb, nlh, &cfg);
 606         if (err < 0)
 607                 goto errout;
 608
 609         tb = fib_get_table(net, cfg.fc_table);
 610         if (tb == NULL) {
 611                 err = -ESRCH;
 612                 goto errout;
 613         }
 614
 615         err = fib_table_delete(tb, &cfg);
 616 errout:
 617         return err;
 618 }
 619
 620 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh, void *arg)
 621 {
 622         struct net *net = sock_net(skb->sk);
 623         struct fib_config cfg;
 624         struct fib_table *tb;
 625         int err;
 626
 627         err = rtm_to_fib_config(net, skb, nlh, &cfg);
 628         if (err < 0)
 629                 goto errout;
 630
 631         tb = fib_new_table(net, cfg.fc_table);
 632         if (tb == NULL) {
 633                 err = -ENOBUFS;
 634                 goto errout;
 635         }
 636
 637         err = fib_table_insert(tb, &cfg);
 638 errout:
 639         return err;
 640 }
 641
 642 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 643 {
 644         struct net *net = sock_net(skb->sk);
 645         unsigned int h, s_h;
 646         unsigned int e = 0, s_e;
 647         struct fib_table *tb;
 648         struct hlist_node *node;
 649         struct hlist_head *head;
 650         int dumped = 0;
 651
 652         if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
 653             ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
 654                 return ip_rt_dump(skb, cb);
 655
 656         s_h = cb->args[0];
 657         s_e = cb->args[1];
 658
 659         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
 660                 e = 0;
 661                 head = &net->ipv4.fib_table_hash[h];
 662                 hlist_for_each_entry(tb, node, head, tb_hlist) {
 663                         if (e < s_e)
 664                                 goto next;
 665                         if (dumped)
 666                                 memset(&cb->args[2], 0, sizeof(cb->args) -
 667                                                  2 * sizeof(cb->args[0]));
 668                         if (fib_table_dump(tb, skb, cb) < 0)
 669                                 goto out;
 670                         dumped = 1;
 671 next:
 672                         e++;
 673                 }
 674         }
 675 out:
 676         cb->args[1] = e;
 677         cb->args[0] = h;
 678
 679         return skb->len;
 680 }
 681
 682 /* Prepare and feed intra-kernel routing request.
 683  * Really, it should be netlink message, but :-( netlink
 684  * can be not configured, so that we feed it directly
 685  * to fib engine. It is legal, because all events occur
 686  * only when netlink is already locked.
 687  */
 688 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
 689 {
 690         struct net *net = dev_net(ifa->ifa_dev->dev);
 691         struct fib_table *tb;
 692         struct fib_config cfg = {
 693                 .fc_protocol = RTPROT_KERNEL,
 694                 .fc_type = type,
 695                 .fc_dst = dst,
 696                 .fc_dst_len = dst_len,
 697                 .fc_prefsrc = ifa->ifa_local,
 698                 .fc_oif = ifa->ifa_dev->dev->ifindex,
 699                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
 700                 .fc_nlinfo = {
 701                         .nl_net = net,
 702                 },
 703         };
 704
 705         if (type == RTN_UNICAST)
 706                 tb = fib_new_table(net, RT_TABLE_MAIN);
 707         else
 708                 tb = fib_new_table(net, RT_TABLE_LOCAL);
 709
 710         if (tb == NULL)
 711                 return;
 712
 713         cfg.fc_table = tb->tb_id;
 714
 715         if (type != RTN_LOCAL)
 716                 cfg.fc_scope = RT_SCOPE_LINK;
 717         else
 718                 cfg.fc_scope = RT_SCOPE_HOST;
 719
 720         if (cmd == RTM_NEWROUTE)
 721                 fib_table_insert(tb, &cfg);
 722         else
 723                 fib_table_delete(tb, &cfg);
 724 }
 725
 726 void fib_add_ifaddr(struct in_ifaddr *ifa)
 727 {
 728         struct in_device *in_dev = ifa->ifa_dev;
 729         struct net_device *dev = in_dev->dev;
 730         struct in_ifaddr *prim = ifa;
 731         __be32 mask = ifa->ifa_mask;
 732         __be32 addr = ifa->ifa_local;
 733         __be32 prefix = ifa->ifa_address & mask;
 734
 735         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 736                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
 737                 if (prim == NULL) {
 738                         pr_warn("%s: bug: prim == NULL\n", __func__);
 739                         return;
 740                 }
 741         }
 742
 743         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
 744
 745         if (!(dev->flags & IFF_UP))
 746                 return;
 747
 748         /* Add broadcast address, if it is explicitly assigned. */
 749         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
 750                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
 751
 752         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
 753             (prefix != addr || ifa->ifa_prefixlen < 32)) {
 754                 fib_magic(RTM_NEWROUTE,
 755                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 756                           prefix, ifa->ifa_prefixlen, prim);
 757
 758                 /* Add network specific broadcasts, when it takes a sense */
 759                 if (ifa->ifa_prefixlen < 31) {
 760                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
 761                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
 762                                   32, prim);
 763                 }
 764         }
 765 }
 766
 767 /* Delete primary or secondary address.
 768  * Optionally, on secondary address promotion consider the addresses
 769  * from subnet iprim as deleted, even if they are in device list.
 770  * In this case the secondary ifa can be in device list.
 771  */
 772 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
 773 {
 774         struct in_device *in_dev = ifa->ifa_dev;
 775         struct net_device *dev = in_dev->dev;
 776         struct in_ifaddr *ifa1;
 777         struct in_ifaddr *prim = ifa, *prim1 = NULL;
 778         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
 779         __be32 any = ifa->ifa_address & ifa->ifa_mask;
 780 #define LOCAL_OK        1
 781 #define BRD_OK          2
 782 #define BRD0_OK         4
 783 #define BRD1_OK         8
 784         unsigned int ok = 0;
 785         int subnet = 0;         /* Primary network */
 786         int gone = 1;           /* Address is missing */
 787         int same_prefsrc = 0;   /* Another primary with same IP */
 788
 789         if (ifa->ifa_flags & IFA_F_SECONDARY) {
 790                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
 791                 if (prim == NULL) {
 792                         pr_warn("%s: bug: prim == NULL\n", __func__);
 793                         return;
 794                 }
 795                 if (iprim && iprim != prim) {
 796                         pr_warn("%s: bug: iprim != prim\n", __func__);
 797                         return;
 798                 }
 799         } else if (!ipv4_is_zeronet(any) &&
 800                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
 801                 fib_magic(RTM_DELROUTE,
 802                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
 803                           any, ifa->ifa_prefixlen, prim);
 804                 subnet = 1;
 805         }
 806
 807         /* Deletion is more complicated than add.
 808          * We should take care of not to delete too much :-)
 809          *
 810          * Scan address list to be sure that addresses are really gone.
 811          */
 812
 813         for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
 814                 if (ifa1 == ifa) {
 815                         /* promotion, keep the IP */
 816                         gone = 0;
 817                         continue;
 818                 }
 819                 /* Ignore IFAs from our subnet */
 820                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
 821                     inet_ifa_match(ifa1->ifa_address, iprim))
 822                         continue;
 823
 824                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
 825                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
 826                         /* Another address from our subnet? */
 827                         if (ifa1->ifa_mask == prim->ifa_mask &&
 828                             inet_ifa_match(ifa1->ifa_address, prim))
 829                                 prim1 = prim;
 830                         else {
 831                                 /* We reached the secondaries, so
 832                                  * same_prefsrc should be determined.
 833                                  */
 834                                 if (!same_prefsrc)
 835                                         continue;
 836                                 /* Search new prim1 if ifa1 is not
 837                                  * using the current prim1
 838                                  */
 839                                 if (!prim1 ||
 840                                     ifa1->ifa_mask != prim1->ifa_mask ||
 841                                     !inet_ifa_match(ifa1->ifa_address, prim1))
 842                                         prim1 = inet_ifa_byprefix(in_dev,
 843                                                         ifa1->ifa_address,
 844                                                         ifa1->ifa_mask);
 845                                 if (!prim1)
 846                                         continue;
 847                                 if (prim1->ifa_local != prim->ifa_local)
 848                                         continue;
 849                         }
 850                 } else {
 851                         if (prim->ifa_local != ifa1->ifa_local)
 852                                 continue;
 853                         prim1 = ifa1;
 854                         if (prim != prim1)
 855                                 same_prefsrc = 1;
 856                 }
 857                 if (ifa->ifa_local == ifa1->ifa_local)
 858                         ok |= LOCAL_OK;
 859                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
 860                         ok |= BRD_OK;
 861                 if (brd == ifa1->ifa_broadcast)
 862                         ok |= BRD1_OK;
 863                 if (any == ifa1->ifa_broadcast)
 864                         ok |= BRD0_OK;
 865                 /* primary has network specific broadcasts */
 866                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
 867                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
 868                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
 869
 870                         if (!ipv4_is_zeronet(any1)) {
 871                                 if (ifa->ifa_broadcast == brd1 ||
 872                                     ifa->ifa_broadcast == any1)
 873                                         ok |= BRD_OK;
 874                                 if (brd == brd1 || brd == any1)
 875                                         ok |= BRD1_OK;
 876                                 if (any == brd1 || any == any1)
 877                                         ok |= BRD0_OK;
 878                         }
 879                 }
 880         }
 881
 882         if (!(ok & BRD_OK))
 883                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
 884         if (subnet && ifa->ifa_prefixlen < 31) {
 885                 if (!(ok & BRD1_OK))
 886                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
 887                 if (!(ok & BRD0_OK))
 888                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
 889         }
 890         if (!(ok & LOCAL_OK)) {
 891                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
 892
 893                 /* Check, that this local address finally disappeared. */
 894                 if (gone &&
 895                     inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) {
 896                         /* And the last, but not the least thing.
 897                          * We must flush stray FIB entries.
 898                          *
 899                          * First of all, we scan fib_info list searching
 900                          * for stray nexthop entries, then ignite fib_flush.
 901                          */
 902                         if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
 903                                 fib_flush(dev_net(dev));
 904                 }
 905         }
 906 #undef LOCAL_OK
 907 #undef BRD_OK
 908 #undef BRD0_OK
 909 #undef BRD1_OK
 910 }
 911
 912 static void nl_fib_lookup(struct fib_result_nl *frn, struct fib_table *tb)
 913 {
 914
 915         struct fib_result       res;
 916         struct flowi4           fl4 = {
 917                 .flowi4_mark = frn->fl_mark,
 918                 .daddr = frn->fl_addr,
 919                 .flowi4_tos = frn->fl_tos,
 920                 .flowi4_scope = frn->fl_scope,
 921         };
 922
 923 #ifdef CONFIG_IP_MULTIPLE_TABLES
 924         res.r = NULL;
 925 #endif
 926
 927         frn->err = -ENOENT;
 928         if (tb) {
 929                 local_bh_disable();
 930
 931                 frn->tb_id = tb->tb_id;
 932                 rcu_read_lock();
 933                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
 934
 935                 if (!frn->err) {
 936                         frn->prefixlen = res.prefixlen;
 937                         frn->nh_sel = res.nh_sel;
 938                         frn->type = res.type;
 939                         frn->scope = res.scope;
 940                 }
 941                 rcu_read_unlock();
 942                 local_bh_enable();
 943         }
 944 }
 945
 946 static void nl_fib_input(struct sk_buff *skb)
 947 {
 948         struct net *net;
 949         struct fib_result_nl *frn;
 950         struct nlmsghdr *nlh;
 951         struct fib_table *tb;
 952         u32 pid;
 953
 954         net = sock_net(skb->sk);
 955         nlh = nlmsg_hdr(skb);
 956         if (skb->len < NLMSG_SPACE(0) || skb->len < nlh->nlmsg_len ||
 957             nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*frn)))
 958                 return;
 959
 960         skb = skb_clone(skb, GFP_KERNEL);
 961         if (skb == NULL)
 962                 return;
 963         nlh = nlmsg_hdr(skb);
 964
 965         frn = (struct fib_result_nl *) NLMSG_DATA(nlh);
 966         tb = fib_get_table(net, frn->tb_id_in);
 967
 968         nl_fib_lookup(frn, tb);
 969
 970         pid = NETLINK_CB(skb).pid;      /* pid of sending process */
 971         NETLINK_CB(skb).pid = 0;        /* from kernel */
 972         NETLINK_CB(skb).dst_group = 0;  /* unicast */
 973         netlink_unicast(net->ipv4.fibnl, skb, pid, MSG_DONTWAIT);
 974 }
 975
 976 static int __net_init nl_fib_lookup_init(struct net *net)
 977 {
 978         struct sock *sk;
 979         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, 0,
 980                                    nl_fib_input, NULL, THIS_MODULE);
 981         if (sk == NULL)
 982                 return -EAFNOSUPPORT;
 983         net->ipv4.fibnl = sk;
 984         return 0;
 985 }
 986
 987 static void nl_fib_lookup_exit(struct net *net)
 988 {
 989         netlink_kernel_release(net->ipv4.fibnl);
 990         net->ipv4.fibnl = NULL;
 991 }
 992
 993 static void fib_disable_ip(struct net_device *dev, int force, int delay)
 994 {
 995         if (fib_sync_down_dev(dev, force))
 996                 fib_flush(dev_net(dev));
 997         rt_cache_flush(dev_net(dev), delay);
 998         arp_ifdown(dev);
 999 }
1000
1001 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1002 {
1003         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1004         struct net_device *dev = ifa->ifa_dev->dev;
1005         struct net *net = dev_net(dev);
1006
1007         switch (event) {
1008         case NETDEV_UP:
1009                 fib_add_ifaddr(ifa);
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_del_ifaddr(ifa, NULL);
1018                 atomic_inc(&net->ipv4.dev_addr_genid);
1019                 if (ifa->ifa_dev->ifa_list == NULL) {
1020                         /* Last address was deleted from this interface.
1021                          * Disable IP.
1022                          */
1023                         fib_disable_ip(dev, 1, 0);
1024                 } else {
1025                         rt_cache_flush(dev_net(dev), -1);
1026                 }
1027                 break;
1028         }
1029         return NOTIFY_DONE;
1030 }
1031
1032 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1033 {
1034         struct net_device *dev = ptr;
1035         struct in_device *in_dev = __in_dev_get_rtnl(dev);
1036         struct net *net = dev_net(dev);
1037
1038         if (event == NETDEV_UNREGISTER) {
1039                 fib_disable_ip(dev, 2, -1);
1040                 return NOTIFY_DONE;
1041         }
1042
1043         if (!in_dev)
1044                 return NOTIFY_DONE;
1045
1046         switch (event) {
1047         case NETDEV_UP:
1048                 for_ifa(in_dev) {
1049                         fib_add_ifaddr(ifa);
1050                 } endfor_ifa(in_dev);
1051 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1052                 fib_sync_up(dev);
1053 #endif
1054                 atomic_inc(&net->ipv4.dev_addr_genid);
1055                 rt_cache_flush(dev_net(dev), -1);
1056                 break;
1057         case NETDEV_DOWN:
1058                 fib_disable_ip(dev, 0, 0);
1059                 break;
1060         case NETDEV_CHANGEMTU:
1061         case NETDEV_CHANGE:
1062                 rt_cache_flush(dev_net(dev), 0);
1063                 break;
1064         case NETDEV_UNREGISTER_BATCH:
1065                 /* The batch unregister is only called on the first
1066                  * device in the list of devices being unregistered.
1067                  * Therefore we should not pass dev_net(dev) in here.
1068                  */
1069                 rt_cache_flush_batch(NULL);
1070                 break;
1071         }
1072         return NOTIFY_DONE;
1073 }
1074
1075 static struct notifier_block fib_inetaddr_notifier = {
1076         .notifier_call = fib_inetaddr_event,
1077 };
1078
1079 static struct notifier_block fib_netdev_notifier = {
1080         .notifier_call = fib_netdev_event,
1081 };
1082
1083 static int __net_init ip_fib_net_init(struct net *net)
1084 {
1085         int err;
1086         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1087
1088         /* Avoid false sharing : Use at least a full cache line */
1089         size = max_t(size_t, size, L1_CACHE_BYTES);
1090
1091         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1092         if (net->ipv4.fib_table_hash == NULL)
1093                 return -ENOMEM;
1094
1095         err = fib4_rules_init(net);
1096         if (err < 0)
1097                 goto fail;
1098         return 0;
1099
1100 fail:
1101         kfree(net->ipv4.fib_table_hash);
1102         return err;
1103 }
1104
1105 static void ip_fib_net_exit(struct net *net)
1106 {
1107         unsigned int i;
1108
1109 #ifdef CONFIG_IP_MULTIPLE_TABLES
1110         fib4_rules_exit(net);
1111 #endif
1112
1113         rtnl_lock();
1114         for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1115                 struct fib_table *tb;
1116                 struct hlist_head *head;
1117                 struct hlist_node *node, *tmp;
1118
1119                 head = &net->ipv4.fib_table_hash[i];
1120                 hlist_for_each_entry_safe(tb, node, tmp, head, tb_hlist) {
1121                         hlist_del(node);
1122                         fib_table_flush(tb);
1123                         fib_free_table(tb);
1124                 }
1125         }
1126         rtnl_unlock();
1127         kfree(net->ipv4.fib_table_hash);
1128 }
1129
1130 static int __net_init fib_net_init(struct net *net)
1131 {
1132         int error;
1133
1134         error = ip_fib_net_init(net);
1135         if (error < 0)
1136                 goto out;
1137         error = nl_fib_lookup_init(net);
1138         if (error < 0)
1139                 goto out_nlfl;
1140         error = fib_proc_init(net);
1141         if (error < 0)
1142                 goto out_proc;
1143 out:
1144         return error;
1145
1146 out_proc:
1147         nl_fib_lookup_exit(net);
1148 out_nlfl:
1149         ip_fib_net_exit(net);
1150         goto out;
1151 }
1152
1153 static void __net_exit fib_net_exit(struct net *net)
1154 {
1155         fib_proc_exit(net);
1156         nl_fib_lookup_exit(net);
1157         ip_fib_net_exit(net);
1158 }
1159
1160 static struct pernet_operations fib_net_ops = {
1161         .init = fib_net_init,
1162         .exit = fib_net_exit,
1163 };
1164
1165 void __init ip_fib_init(void)
1166 {
1167         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1168         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1169         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1170
1171         register_pernet_subsys(&fib_net_ops);
1172         register_netdevice_notifier(&fib_netdev_notifier);
1173         register_inetaddr_notifier(&fib_inetaddr_notifier);
1174
1175         fib_trie_init();
1176 }