net/ipv4/fib_semantics.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: semantics.
   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 <asm/uaccess.h>
  17 #include <asm/system.h>
  18 #include <linux/bitops.h>
  19 #include <linux/types.h>
  20 #include <linux/kernel.h>
  21 #include <linux/jiffies.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_arp.h>
  32 #include <linux/proc_fs.h>
  33 #include <linux/skbuff.h>
  34 #include <linux/init.h>
  35 #include <linux/slab.h>
  36
  37 #include <net/arp.h>
  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/ip_fib.h>
  44 #include <net/netlink.h>
  45 #include <net/nexthop.h>
  46
  47 #include "fib_lookup.h"
  48
  49 static DEFINE_SPINLOCK(fib_info_lock);
  50 static struct hlist_head *fib_info_hash;
  51 static struct hlist_head *fib_info_laddrhash;
  52 static unsigned int fib_info_hash_size;
  53 static unsigned int fib_info_cnt;
  54
  55 #define DEVINDEX_HASHBITS 8
  56 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
  57 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
  58
  59 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  60
  61 static DEFINE_SPINLOCK(fib_multipath_lock);
  62
  63 #define for_nexthops(fi) {                                              \
  64         int nhsel; const struct fib_nh *nh;                             \
  65         for (nhsel = 0, nh = (fi)->fib_nh;                              \
  66              nhsel < (fi)->fib_nhs;                                     \
  67              nh++, nhsel++)
  68
  69 #define change_nexthops(fi) {                                           \
  70         int nhsel; struct fib_nh *nexthop_nh;                           \
  71         for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh);   \
  72              nhsel < (fi)->fib_nhs;                                     \
  73              nexthop_nh++, nhsel++)
  74
  75 #else /* CONFIG_IP_ROUTE_MULTIPATH */
  76
  77 /* Hope, that gcc will optimize it to get rid of dummy loop */
  78
  79 #define for_nexthops(fi) {                                              \
  80         int nhsel; const struct fib_nh *nh = (fi)->fib_nh;              \
  81         for (nhsel = 0; nhsel < 1; nhsel++)
  82
  83 #define change_nexthops(fi) {                                           \
  84         int nhsel;                                                      \
  85         struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);    \
  86         for (nhsel = 0; nhsel < 1; nhsel++)
  87
  88 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
  89
  90 #define endfor_nexthops(fi) }
  91
  92
  93 const struct fib_prop fib_props[RTN_MAX + 1] = {
  94         [RTN_UNSPEC] = {
  95                 .error  = 0,
  96                 .scope  = RT_SCOPE_NOWHERE,
  97         },
  98         [RTN_UNICAST] = {
  99                 .error  = 0,
 100                 .scope  = RT_SCOPE_UNIVERSE,
 101         },
 102         [RTN_LOCAL] = {
 103                 .error  = 0,
 104                 .scope  = RT_SCOPE_HOST,
 105         },
 106         [RTN_BROADCAST] = {
 107                 .error  = 0,
 108                 .scope  = RT_SCOPE_LINK,
 109         },
 110         [RTN_ANYCAST] = {
 111                 .error  = 0,
 112                 .scope  = RT_SCOPE_LINK,
 113         },
 114         [RTN_MULTICAST] = {
 115                 .error  = 0,
 116                 .scope  = RT_SCOPE_UNIVERSE,
 117         },
 118         [RTN_BLACKHOLE] = {
 119                 .error  = -EINVAL,
 120                 .scope  = RT_SCOPE_UNIVERSE,
 121         },
 122         [RTN_UNREACHABLE] = {
 123                 .error  = -EHOSTUNREACH,
 124                 .scope  = RT_SCOPE_UNIVERSE,
 125         },
 126         [RTN_PROHIBIT] = {
 127                 .error  = -EACCES,
 128                 .scope  = RT_SCOPE_UNIVERSE,
 129         },
 130         [RTN_THROW] = {
 131                 .error  = -EAGAIN,
 132                 .scope  = RT_SCOPE_UNIVERSE,
 133         },
 134         [RTN_NAT] = {
 135                 .error  = -EINVAL,
 136                 .scope  = RT_SCOPE_NOWHERE,
 137         },
 138         [RTN_XRESOLVE] = {
 139                 .error  = -EINVAL,
 140                 .scope  = RT_SCOPE_NOWHERE,
 141         },
 142 };
 143
 144
 145 /* Release a nexthop info record */
 146
 147 static void free_fib_info_rcu(struct rcu_head *head)
 148 {
 149         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 150
 151         if (fi->fib_metrics != (u32 *) dst_default_metrics)
 152                 kfree(fi->fib_metrics);
 153         kfree(fi);
 154 }
 155
 156 void free_fib_info(struct fib_info *fi)
 157 {
 158         if (fi->fib_dead == 0) {
 159                 pr_warning("Freeing alive fib_info %p\n", fi);
 160                 return;
 161         }
 162         change_nexthops(fi) {
 163                 if (nexthop_nh->nh_dev)
 164                         dev_put(nexthop_nh->nh_dev);
 165                 nexthop_nh->nh_dev = NULL;
 166         } endfor_nexthops(fi);
 167         fib_info_cnt--;
 168         release_net(fi->fib_net);
 169         call_rcu(&fi->rcu, free_fib_info_rcu);
 170 }
 171
 172 void fib_release_info(struct fib_info *fi)
 173 {
 174         spin_lock_bh(&fib_info_lock);
 175         if (fi && --fi->fib_treeref == 0) {
 176                 hlist_del(&fi->fib_hash);
 177                 if (fi->fib_prefsrc)
 178                         hlist_del(&fi->fib_lhash);
 179                 change_nexthops(fi) {
 180                         if (!nexthop_nh->nh_dev)
 181                                 continue;
 182                         hlist_del(&nexthop_nh->nh_hash);
 183                 } endfor_nexthops(fi)
 184                 fi->fib_dead = 1;
 185                 fib_info_put(fi);
 186         }
 187         spin_unlock_bh(&fib_info_lock);
 188 }
 189
 190 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
 191 {
 192         const struct fib_nh *onh = ofi->fib_nh;
 193
 194         for_nexthops(fi) {
 195                 if (nh->nh_oif != onh->nh_oif ||
 196                     nh->nh_gw  != onh->nh_gw ||
 197                     nh->nh_scope != onh->nh_scope ||
 198 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 199                     nh->nh_weight != onh->nh_weight ||
 200 #endif
 201 #ifdef CONFIG_IP_ROUTE_CLASSID
 202                     nh->nh_tclassid != onh->nh_tclassid ||
 203 #endif
 204                     ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
 205                         return -1;
 206                 onh++;
 207         } endfor_nexthops(fi);
 208         return 0;
 209 }
 210
 211 static inline unsigned int fib_devindex_hashfn(unsigned int val)
 212 {
 213         unsigned int mask = DEVINDEX_HASHSIZE - 1;
 214
 215         return (val ^
 216                 (val >> DEVINDEX_HASHBITS) ^
 217                 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
 218 }
 219
 220 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
 221 {
 222         unsigned int mask = (fib_info_hash_size - 1);
 223         unsigned int val = fi->fib_nhs;
 224
 225         val ^= (fi->fib_protocol << 8) | fi->fib_scope;
 226         val ^= (__force u32)fi->fib_prefsrc;
 227         val ^= fi->fib_priority;
 228         for_nexthops(fi) {
 229                 val ^= fib_devindex_hashfn(nh->nh_oif);
 230         } endfor_nexthops(fi)
 231
 232         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 233 }
 234
 235 static struct fib_info *fib_find_info(const struct fib_info *nfi)
 236 {
 237         struct hlist_head *head;
 238         struct hlist_node *node;
 239         struct fib_info *fi;
 240         unsigned int hash;
 241
 242         hash = fib_info_hashfn(nfi);
 243         head = &fib_info_hash[hash];
 244
 245         hlist_for_each_entry(fi, node, head, fib_hash) {
 246                 if (!net_eq(fi->fib_net, nfi->fib_net))
 247                         continue;
 248                 if (fi->fib_nhs != nfi->fib_nhs)
 249                         continue;
 250                 if (nfi->fib_protocol == fi->fib_protocol &&
 251                     nfi->fib_scope == fi->fib_scope &&
 252                     nfi->fib_prefsrc == fi->fib_prefsrc &&
 253                     nfi->fib_priority == fi->fib_priority &&
 254                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 255                            sizeof(u32) * RTAX_MAX) == 0 &&
 256                     ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
 257                     (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
 258                         return fi;
 259         }
 260
 261         return NULL;
 262 }
 263
 264 /* Check, that the gateway is already configured.
 265  * Used only by redirect accept routine.
 266  */
 267 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 268 {
 269         struct hlist_head *head;
 270         struct hlist_node *node;
 271         struct fib_nh *nh;
 272         unsigned int hash;
 273
 274         spin_lock(&fib_info_lock);
 275
 276         hash = fib_devindex_hashfn(dev->ifindex);
 277         head = &fib_info_devhash[hash];
 278         hlist_for_each_entry(nh, node, head, nh_hash) {
 279                 if (nh->nh_dev == dev &&
 280                     nh->nh_gw == gw &&
 281                     !(nh->nh_flags & RTNH_F_DEAD)) {
 282                         spin_unlock(&fib_info_lock);
 283                         return 0;
 284                 }
 285         }
 286
 287         spin_unlock(&fib_info_lock);
 288
 289         return -1;
 290 }
 291
 292 static inline size_t fib_nlmsg_size(struct fib_info *fi)
 293 {
 294         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 295                          + nla_total_size(4) /* RTA_TABLE */
 296                          + nla_total_size(4) /* RTA_DST */
 297                          + nla_total_size(4) /* RTA_PRIORITY */
 298                          + nla_total_size(4); /* RTA_PREFSRC */
 299
 300         /* space for nested metrics */
 301         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 302
 303         if (fi->fib_nhs) {
 304                 /* Also handles the special case fib_nhs == 1 */
 305
 306                 /* each nexthop is packed in an attribute */
 307                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 308
 309                 /* may contain flow and gateway attribute */
 310                 nhsize += 2 * nla_total_size(4);
 311
 312                 /* all nexthops are packed in a nested attribute */
 313                 payload += nla_total_size(fi->fib_nhs * nhsize);
 314         }
 315
 316         return payload;
 317 }
 318
 319 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 320                int dst_len, u32 tb_id, struct nl_info *info,
 321                unsigned int nlm_flags)
 322 {
 323         struct sk_buff *skb;
 324         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 325         int err = -ENOBUFS;
 326
 327         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 328         if (skb == NULL)
 329                 goto errout;
 330
 331         err = fib_dump_info(skb, info->pid, seq, event, tb_id,
 332                             fa->fa_type, key, dst_len,
 333                             fa->fa_tos, fa->fa_info, nlm_flags);
 334         if (err < 0) {
 335                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 336                 WARN_ON(err == -EMSGSIZE);
 337                 kfree_skb(skb);
 338                 goto errout;
 339         }
 340         rtnl_notify(skb, info->nl_net, info->pid, RTNLGRP_IPV4_ROUTE,
 341                     info->nlh, GFP_KERNEL);
 342         return;
 343 errout:
 344         if (err < 0)
 345                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 346 }
 347
 348 /* Return the first fib alias matching TOS with
 349  * priority less than or equal to PRIO.
 350  */
 351 struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
 352 {
 353         if (fah) {
 354                 struct fib_alias *fa;
 355                 list_for_each_entry(fa, fah, fa_list) {
 356                         if (fa->fa_tos > tos)
 357                                 continue;
 358                         if (fa->fa_info->fib_priority >= prio ||
 359                             fa->fa_tos < tos)
 360                                 return fa;
 361                 }
 362         }
 363         return NULL;
 364 }
 365
 366 int fib_detect_death(struct fib_info *fi, int order,
 367                      struct fib_info **last_resort, int *last_idx, int dflt)
 368 {
 369         struct neighbour *n;
 370         int state = NUD_NONE;
 371
 372         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 373         if (n) {
 374                 state = n->nud_state;
 375                 neigh_release(n);
 376         }
 377         if (state == NUD_REACHABLE)
 378                 return 0;
 379         if ((state & NUD_VALID) && order != dflt)
 380                 return 0;
 381         if ((state & NUD_VALID) ||
 382             (*last_idx < 0 && order > dflt)) {
 383                 *last_resort = fi;
 384                 *last_idx = order;
 385         }
 386         return 1;
 387 }
 388
 389 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 390
 391 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
 392 {
 393         int nhs = 0;
 394
 395         while (rtnh_ok(rtnh, remaining)) {
 396                 nhs++;
 397                 rtnh = rtnh_next(rtnh, &remaining);
 398         }
 399
 400         /* leftover implies invalid nexthop configuration, discard it */
 401         return remaining > 0 ? 0 : nhs;
 402 }
 403
 404 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 405                        int remaining, struct fib_config *cfg)
 406 {
 407         change_nexthops(fi) {
 408                 int attrlen;
 409
 410                 if (!rtnh_ok(rtnh, remaining))
 411                         return -EINVAL;
 412
 413                 nexthop_nh->nh_flags =
 414                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 415                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
 416                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
 417
 418                 attrlen = rtnh_attrlen(rtnh);
 419                 if (attrlen > 0) {
 420                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 421
 422                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 423                         nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
 424 #ifdef CONFIG_IP_ROUTE_CLASSID
 425                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 426                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 427 #endif
 428                 }
 429
 430                 rtnh = rtnh_next(rtnh, &remaining);
 431         } endfor_nexthops(fi);
 432
 433         return 0;
 434 }
 435
 436 #endif
 437
 438 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
 439 {
 440 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 441         struct rtnexthop *rtnh;
 442         int remaining;
 443 #endif
 444
 445         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 446                 return 1;
 447
 448         if (cfg->fc_oif || cfg->fc_gw) {
 449                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 450                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 451                         return 0;
 452                 return 1;
 453         }
 454
 455 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 456         if (cfg->fc_mp == NULL)
 457                 return 0;
 458
 459         rtnh = cfg->fc_mp;
 460         remaining = cfg->fc_mp_len;
 461
 462         for_nexthops(fi) {
 463                 int attrlen;
 464
 465                 if (!rtnh_ok(rtnh, remaining))
 466                         return -EINVAL;
 467
 468                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 469                         return 1;
 470
 471                 attrlen = rtnh_attrlen(rtnh);
 472                 if (attrlen < 0) {
 473                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 474
 475                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 476                         if (nla && nla_get_be32(nla) != nh->nh_gw)
 477                                 return 1;
 478 #ifdef CONFIG_IP_ROUTE_CLASSID
 479                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 480                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 481                                 return 1;
 482 #endif
 483                 }
 484
 485                 rtnh = rtnh_next(rtnh, &remaining);
 486         } endfor_nexthops(fi);
 487 #endif
 488         return 0;
 489 }
 490
 491
 492 /*
 493  * Picture
 494  * -------
 495  *
 496  * Semantics of nexthop is very messy by historical reasons.
 497  * We have to take into account, that:
 498  * a) gateway can be actually local interface address,
 499  *    so that gatewayed route is direct.
 500  * b) gateway must be on-link address, possibly
 501  *    described not by an ifaddr, but also by a direct route.
 502  * c) If both gateway and interface are specified, they should not
 503  *    contradict.
 504  * d) If we use tunnel routes, gateway could be not on-link.
 505  *
 506  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 507  * results in pretty ugly and hairy code with obscure logic.
 508  *
 509  * I chose to generalized it instead, so that the size
 510  * of code does not increase practically, but it becomes
 511  * much more general.
 512  * Every prefix is assigned a "scope" value: "host" is local address,
 513  * "link" is direct route,
 514  * [ ... "site" ... "interior" ... ]
 515  * and "universe" is true gateway route with global meaning.
 516  *
 517  * Every prefix refers to a set of "nexthop"s (gw, oif),
 518  * where gw must have narrower scope. This recursion stops
 519  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 520  * which means that gw is forced to be on link.
 521  *
 522  * Code is still hairy, but now it is apparently logically
 523  * consistent and very flexible. F.e. as by-product it allows
 524  * to co-exists in peace independent exterior and interior
 525  * routing processes.
 526  *
 527  * Normally it looks as following.
 528  *
 529  * {universe prefix}  -> (gw, oif) [scope link]
 530  *                |
 531  *                |-> {link prefix} -> (gw, oif) [scope local]
 532  *                                      |
 533  *                                      |-> {local prefix} (terminal node)
 534  */
 535 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
 536                         struct fib_nh *nh)
 537 {
 538         int err;
 539         struct net *net;
 540         struct net_device *dev;
 541
 542         net = cfg->fc_nlinfo.nl_net;
 543         if (nh->nh_gw) {
 544                 struct fib_result res;
 545
 546                 if (nh->nh_flags & RTNH_F_ONLINK) {
 547
 548                         if (cfg->fc_scope >= RT_SCOPE_LINK)
 549                                 return -EINVAL;
 550                         if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
 551                                 return -EINVAL;
 552                         dev = __dev_get_by_index(net, nh->nh_oif);
 553                         if (!dev)
 554                                 return -ENODEV;
 555                         if (!(dev->flags & IFF_UP))
 556                                 return -ENETDOWN;
 557                         nh->nh_dev = dev;
 558                         dev_hold(dev);
 559                         nh->nh_scope = RT_SCOPE_LINK;
 560                         return 0;
 561                 }
 562                 rcu_read_lock();
 563                 {
 564                         struct flowi4 fl4 = {
 565                                 .daddr = nh->nh_gw,
 566                                 .flowi4_scope = cfg->fc_scope + 1,
 567                                 .flowi4_oif = nh->nh_oif,
 568                         };
 569
 570                         /* It is not necessary, but requires a bit of thinking */
 571                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
 572                                 fl4.flowi4_scope = RT_SCOPE_LINK;
 573                         err = fib_lookup(net, &fl4, &res);
 574                         if (err) {
 575                                 rcu_read_unlock();
 576                                 return err;
 577                         }
 578                 }
 579                 err = -EINVAL;
 580                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 581                         goto out;
 582                 nh->nh_scope = res.scope;
 583                 nh->nh_oif = FIB_RES_OIF(res);
 584                 nh->nh_dev = dev = FIB_RES_DEV(res);
 585                 if (!dev)
 586                         goto out;
 587                 dev_hold(dev);
 588                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 589         } else {
 590                 struct in_device *in_dev;
 591
 592                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 593                         return -EINVAL;
 594
 595                 rcu_read_lock();
 596                 err = -ENODEV;
 597                 in_dev = inetdev_by_index(net, nh->nh_oif);
 598                 if (in_dev == NULL)
 599                         goto out;
 600                 err = -ENETDOWN;
 601                 if (!(in_dev->dev->flags & IFF_UP))
 602                         goto out;
 603                 nh->nh_dev = in_dev->dev;
 604                 dev_hold(nh->nh_dev);
 605                 nh->nh_scope = RT_SCOPE_HOST;
 606                 err = 0;
 607         }
 608 out:
 609         rcu_read_unlock();
 610         return err;
 611 }
 612
 613 static inline unsigned int fib_laddr_hashfn(__be32 val)
 614 {
 615         unsigned int mask = (fib_info_hash_size - 1);
 616
 617         return ((__force u32)val ^
 618                 ((__force u32)val >> 7) ^
 619                 ((__force u32)val >> 14)) & mask;
 620 }
 621
 622 static struct hlist_head *fib_info_hash_alloc(int bytes)
 623 {
 624         if (bytes <= PAGE_SIZE)
 625                 return kzalloc(bytes, GFP_KERNEL);
 626         else
 627                 return (struct hlist_head *)
 628                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 629                                          get_order(bytes));
 630 }
 631
 632 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 633 {
 634         if (!hash)
 635                 return;
 636
 637         if (bytes <= PAGE_SIZE)
 638                 kfree(hash);
 639         else
 640                 free_pages((unsigned long) hash, get_order(bytes));
 641 }
 642
 643 static void fib_info_hash_move(struct hlist_head *new_info_hash,
 644                                struct hlist_head *new_laddrhash,
 645                                unsigned int new_size)
 646 {
 647         struct hlist_head *old_info_hash, *old_laddrhash;
 648         unsigned int old_size = fib_info_hash_size;
 649         unsigned int i, bytes;
 650
 651         spin_lock_bh(&fib_info_lock);
 652         old_info_hash = fib_info_hash;
 653         old_laddrhash = fib_info_laddrhash;
 654         fib_info_hash_size = new_size;
 655
 656         for (i = 0; i < old_size; i++) {
 657                 struct hlist_head *head = &fib_info_hash[i];
 658                 struct hlist_node *node, *n;
 659                 struct fib_info *fi;
 660
 661                 hlist_for_each_entry_safe(fi, node, n, head, fib_hash) {
 662                         struct hlist_head *dest;
 663                         unsigned int new_hash;
 664
 665                         hlist_del(&fi->fib_hash);
 666
 667                         new_hash = fib_info_hashfn(fi);
 668                         dest = &new_info_hash[new_hash];
 669                         hlist_add_head(&fi->fib_hash, dest);
 670                 }
 671         }
 672         fib_info_hash = new_info_hash;
 673
 674         for (i = 0; i < old_size; i++) {
 675                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 676                 struct hlist_node *node, *n;
 677                 struct fib_info *fi;
 678
 679                 hlist_for_each_entry_safe(fi, node, n, lhead, fib_lhash) {
 680                         struct hlist_head *ldest;
 681                         unsigned int new_hash;
 682
 683                         hlist_del(&fi->fib_lhash);
 684
 685                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 686                         ldest = &new_laddrhash[new_hash];
 687                         hlist_add_head(&fi->fib_lhash, ldest);
 688                 }
 689         }
 690         fib_info_laddrhash = new_laddrhash;
 691
 692         spin_unlock_bh(&fib_info_lock);
 693
 694         bytes = old_size * sizeof(struct hlist_head *);
 695         fib_info_hash_free(old_info_hash, bytes);
 696         fib_info_hash_free(old_laddrhash, bytes);
 697 }
 698
 699 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 700 {
 701         nh->nh_saddr = inet_select_addr(nh->nh_dev,
 702                                         nh->nh_gw,
 703                                         nh->nh_parent->fib_scope);
 704         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 705
 706         return nh->nh_saddr;
 707 }
 708
 709 struct fib_info *fib_create_info(struct fib_config *cfg)
 710 {
 711         int err;
 712         struct fib_info *fi = NULL;
 713         struct fib_info *ofi;
 714         int nhs = 1;
 715         struct net *net = cfg->fc_nlinfo.nl_net;
 716
 717         if (cfg->fc_type > RTN_MAX)
 718                 goto err_inval;
 719
 720         /* Fast check to catch the most weird cases */
 721         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
 722                 goto err_inval;
 723
 724 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 725         if (cfg->fc_mp) {
 726                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
 727                 if (nhs == 0)
 728                         goto err_inval;
 729         }
 730 #endif
 731
 732         err = -ENOBUFS;
 733         if (fib_info_cnt >= fib_info_hash_size) {
 734                 unsigned int new_size = fib_info_hash_size << 1;
 735                 struct hlist_head *new_info_hash;
 736                 struct hlist_head *new_laddrhash;
 737                 unsigned int bytes;
 738
 739                 if (!new_size)
 740                         new_size = 1;
 741                 bytes = new_size * sizeof(struct hlist_head *);
 742                 new_info_hash = fib_info_hash_alloc(bytes);
 743                 new_laddrhash = fib_info_hash_alloc(bytes);
 744                 if (!new_info_hash || !new_laddrhash) {
 745                         fib_info_hash_free(new_info_hash, bytes);
 746                         fib_info_hash_free(new_laddrhash, bytes);
 747                 } else
 748                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
 749
 750                 if (!fib_info_hash_size)
 751                         goto failure;
 752         }
 753
 754         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
 755         if (fi == NULL)
 756                 goto failure;
 757         if (cfg->fc_mx) {
 758                 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
 759                 if (!fi->fib_metrics)
 760                         goto failure;
 761         } else
 762                 fi->fib_metrics = (u32 *) dst_default_metrics;
 763         fib_info_cnt++;
 764
 765         fi->fib_net = hold_net(net);
 766         fi->fib_protocol = cfg->fc_protocol;
 767         fi->fib_scope = cfg->fc_scope;
 768         fi->fib_flags = cfg->fc_flags;
 769         fi->fib_priority = cfg->fc_priority;
 770         fi->fib_prefsrc = cfg->fc_prefsrc;
 771
 772         fi->fib_nhs = nhs;
 773         change_nexthops(fi) {
 774                 nexthop_nh->nh_parent = fi;
 775         } endfor_nexthops(fi)
 776
 777         if (cfg->fc_mx) {
 778                 struct nlattr *nla;
 779                 int remaining;
 780
 781                 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 782                         int type = nla_type(nla);
 783
 784                         if (type) {
 785                                 if (type > RTAX_MAX)
 786                                         goto err_inval;
 787                                 fi->fib_metrics[type - 1] = nla_get_u32(nla);
 788                         }
 789                 }
 790         }
 791
 792         if (cfg->fc_mp) {
 793 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 794                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
 795                 if (err != 0)
 796                         goto failure;
 797                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
 798                         goto err_inval;
 799                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
 800                         goto err_inval;
 801 #ifdef CONFIG_IP_ROUTE_CLASSID
 802                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
 803                         goto err_inval;
 804 #endif
 805 #else
 806                 goto err_inval;
 807 #endif
 808         } else {
 809                 struct fib_nh *nh = fi->fib_nh;
 810
 811                 nh->nh_oif = cfg->fc_oif;
 812                 nh->nh_gw = cfg->fc_gw;
 813                 nh->nh_flags = cfg->fc_flags;
 814 #ifdef CONFIG_IP_ROUTE_CLASSID
 815                 nh->nh_tclassid = cfg->fc_flow;
 816 #endif
 817 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 818                 nh->nh_weight = 1;
 819 #endif
 820         }
 821
 822         if (fib_props[cfg->fc_type].error) {
 823                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
 824                         goto err_inval;
 825                 goto link_it;
 826         } else {
 827                 switch (cfg->fc_type) {
 828                 case RTN_UNICAST:
 829                 case RTN_LOCAL:
 830                 case RTN_BROADCAST:
 831                 case RTN_ANYCAST:
 832                 case RTN_MULTICAST:
 833                         break;
 834                 default:
 835                         goto err_inval;
 836                 }
 837         }
 838
 839         if (cfg->fc_scope > RT_SCOPE_HOST)
 840                 goto err_inval;
 841
 842         if (cfg->fc_scope == RT_SCOPE_HOST) {
 843                 struct fib_nh *nh = fi->fib_nh;
 844
 845                 /* Local address is added. */
 846                 if (nhs != 1 || nh->nh_gw)
 847                         goto err_inval;
 848                 nh->nh_scope = RT_SCOPE_NOWHERE;
 849                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
 850                 err = -ENODEV;
 851                 if (nh->nh_dev == NULL)
 852                         goto failure;
 853         } else {
 854                 change_nexthops(fi) {
 855                         err = fib_check_nh(cfg, fi, nexthop_nh);
 856                         if (err != 0)
 857                                 goto failure;
 858                 } endfor_nexthops(fi)
 859         }
 860
 861         if (fi->fib_prefsrc) {
 862                 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 863                     fi->fib_prefsrc != cfg->fc_dst)
 864                         if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
 865                                 goto err_inval;
 866         }
 867
 868         change_nexthops(fi) {
 869                 fib_info_update_nh_saddr(net, nexthop_nh);
 870         } endfor_nexthops(fi)
 871
 872 link_it:
 873         ofi = fib_find_info(fi);
 874         if (ofi) {
 875                 fi->fib_dead = 1;
 876                 free_fib_info(fi);
 877                 ofi->fib_treeref++;
 878                 return ofi;
 879         }
 880
 881         fi->fib_treeref++;
 882         atomic_inc(&fi->fib_clntref);
 883         spin_lock_bh(&fib_info_lock);
 884         hlist_add_head(&fi->fib_hash,
 885                        &fib_info_hash[fib_info_hashfn(fi)]);
 886         if (fi->fib_prefsrc) {
 887                 struct hlist_head *head;
 888
 889                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
 890                 hlist_add_head(&fi->fib_lhash, head);
 891         }
 892         change_nexthops(fi) {
 893                 struct hlist_head *head;
 894                 unsigned int hash;
 895
 896                 if (!nexthop_nh->nh_dev)
 897                         continue;
 898                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
 899                 head = &fib_info_devhash[hash];
 900                 hlist_add_head(&nexthop_nh->nh_hash, head);
 901         } endfor_nexthops(fi)
 902         spin_unlock_bh(&fib_info_lock);
 903         return fi;
 904
 905 err_inval:
 906         err = -EINVAL;
 907
 908 failure:
 909         if (fi) {
 910                 fi->fib_dead = 1;
 911                 free_fib_info(fi);
 912         }
 913
 914         return ERR_PTR(err);
 915 }
 916
 917 int fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
 918                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
 919                   struct fib_info *fi, unsigned int flags)
 920 {
 921         struct nlmsghdr *nlh;
 922         struct rtmsg *rtm;
 923
 924         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*rtm), flags);
 925         if (nlh == NULL)
 926                 return -EMSGSIZE;
 927
 928         rtm = nlmsg_data(nlh);
 929         rtm->rtm_family = AF_INET;
 930         rtm->rtm_dst_len = dst_len;
 931         rtm->rtm_src_len = 0;
 932         rtm->rtm_tos = tos;
 933         if (tb_id < 256)
 934                 rtm->rtm_table = tb_id;
 935         else
 936                 rtm->rtm_table = RT_TABLE_COMPAT;
 937         NLA_PUT_U32(skb, RTA_TABLE, tb_id);
 938         rtm->rtm_type = type;
 939         rtm->rtm_flags = fi->fib_flags;
 940         rtm->rtm_scope = fi->fib_scope;
 941         rtm->rtm_protocol = fi->fib_protocol;
 942
 943         if (rtm->rtm_dst_len)
 944                 NLA_PUT_BE32(skb, RTA_DST, dst);
 945
 946         if (fi->fib_priority)
 947                 NLA_PUT_U32(skb, RTA_PRIORITY, fi->fib_priority);
 948
 949         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
 950                 goto nla_put_failure;
 951
 952         if (fi->fib_prefsrc)
 953                 NLA_PUT_BE32(skb, RTA_PREFSRC, fi->fib_prefsrc);
 954
 955         if (fi->fib_nhs == 1) {
 956                 if (fi->fib_nh->nh_gw)
 957                         NLA_PUT_BE32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw);
 958
 959                 if (fi->fib_nh->nh_oif)
 960                         NLA_PUT_U32(skb, RTA_OIF, fi->fib_nh->nh_oif);
 961 #ifdef CONFIG_IP_ROUTE_CLASSID
 962                 if (fi->fib_nh[0].nh_tclassid)
 963                         NLA_PUT_U32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid);
 964 #endif
 965         }
 966 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 967         if (fi->fib_nhs > 1) {
 968                 struct rtnexthop *rtnh;
 969                 struct nlattr *mp;
 970
 971                 mp = nla_nest_start(skb, RTA_MULTIPATH);
 972                 if (mp == NULL)
 973                         goto nla_put_failure;
 974
 975                 for_nexthops(fi) {
 976                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
 977                         if (rtnh == NULL)
 978                                 goto nla_put_failure;
 979
 980                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
 981                         rtnh->rtnh_hops = nh->nh_weight - 1;
 982                         rtnh->rtnh_ifindex = nh->nh_oif;
 983
 984                         if (nh->nh_gw)
 985                                 NLA_PUT_BE32(skb, RTA_GATEWAY, nh->nh_gw);
 986 #ifdef CONFIG_IP_ROUTE_CLASSID
 987                         if (nh->nh_tclassid)
 988                                 NLA_PUT_U32(skb, RTA_FLOW, nh->nh_tclassid);
 989 #endif
 990                         /* length of rtnetlink header + attributes */
 991                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
 992                 } endfor_nexthops(fi);
 993
 994                 nla_nest_end(skb, mp);
 995         }
 996 #endif
 997         return nlmsg_end(skb, nlh);
 998
 999 nla_put_failure:
1000         nlmsg_cancel(skb, nlh);
1001         return -EMSGSIZE;
1002 }
1003
1004 /*
1005  * Update FIB if:
1006  * - local address disappeared -> we must delete all the entries
1007  *   referring to it.
1008  * - device went down -> we must shutdown all nexthops going via it.
1009  */
1010 int fib_sync_down_addr(struct net *net, __be32 local)
1011 {
1012         int ret = 0;
1013         unsigned int hash = fib_laddr_hashfn(local);
1014         struct hlist_head *head = &fib_info_laddrhash[hash];
1015         struct hlist_node *node;
1016         struct fib_info *fi;
1017
1018         if (fib_info_laddrhash == NULL || local == 0)
1019                 return 0;
1020
1021         hlist_for_each_entry(fi, node, head, fib_lhash) {
1022                 if (!net_eq(fi->fib_net, net))
1023                         continue;
1024                 if (fi->fib_prefsrc == local) {
1025                         fi->fib_flags |= RTNH_F_DEAD;
1026                         ret++;
1027                 }
1028         }
1029         return ret;
1030 }
1031
1032 int fib_sync_down_dev(struct net_device *dev, int force)
1033 {
1034         int ret = 0;
1035         int scope = RT_SCOPE_NOWHERE;
1036         struct fib_info *prev_fi = NULL;
1037         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1038         struct hlist_head *head = &fib_info_devhash[hash];
1039         struct hlist_node *node;
1040         struct fib_nh *nh;
1041
1042         if (force)
1043                 scope = -1;
1044
1045         hlist_for_each_entry(nh, node, head, nh_hash) {
1046                 struct fib_info *fi = nh->nh_parent;
1047                 int dead;
1048
1049                 BUG_ON(!fi->fib_nhs);
1050                 if (nh->nh_dev != dev || fi == prev_fi)
1051                         continue;
1052                 prev_fi = fi;
1053                 dead = 0;
1054                 change_nexthops(fi) {
1055                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1056                                 dead++;
1057                         else if (nexthop_nh->nh_dev == dev &&
1058                                  nexthop_nh->nh_scope != scope) {
1059                                 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1060 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1061                                 spin_lock_bh(&fib_multipath_lock);
1062                                 fi->fib_power -= nexthop_nh->nh_power;
1063                                 nexthop_nh->nh_power = 0;
1064                                 spin_unlock_bh(&fib_multipath_lock);
1065 #endif
1066                                 dead++;
1067                         }
1068 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1069                         if (force > 1 && nexthop_nh->nh_dev == dev) {
1070                                 dead = fi->fib_nhs;
1071                                 break;
1072                         }
1073 #endif
1074                 } endfor_nexthops(fi)
1075                 if (dead == fi->fib_nhs) {
1076                         fi->fib_flags |= RTNH_F_DEAD;
1077                         ret++;
1078                 }
1079         }
1080
1081         return ret;
1082 }
1083
1084 /* Must be invoked inside of an RCU protected region.  */
1085 void fib_select_default(struct fib_result *res)
1086 {
1087         struct fib_info *fi = NULL, *last_resort = NULL;
1088         struct list_head *fa_head = res->fa_head;
1089         struct fib_table *tb = res->table;
1090         int order = -1, last_idx = -1;
1091         struct fib_alias *fa;
1092
1093         list_for_each_entry_rcu(fa, fa_head, fa_list) {
1094                 struct fib_info *next_fi = fa->fa_info;
1095
1096                 if (next_fi->fib_scope != res->scope ||
1097                     fa->fa_type != RTN_UNICAST)
1098                         continue;
1099
1100                 if (next_fi->fib_priority > res->fi->fib_priority)
1101                         break;
1102                 if (!next_fi->fib_nh[0].nh_gw ||
1103                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1104                         continue;
1105
1106                 fib_alias_accessed(fa);
1107
1108                 if (fi == NULL) {
1109                         if (next_fi != res->fi)
1110                                 break;
1111                 } else if (!fib_detect_death(fi, order, &last_resort,
1112                                              &last_idx, tb->tb_default)) {
1113                         fib_result_assign(res, fi);
1114                         tb->tb_default = order;
1115                         goto out;
1116                 }
1117                 fi = next_fi;
1118                 order++;
1119         }
1120
1121         if (order <= 0 || fi == NULL) {
1122                 tb->tb_default = -1;
1123                 goto out;
1124         }
1125
1126         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1127                                 tb->tb_default)) {
1128                 fib_result_assign(res, fi);
1129                 tb->tb_default = order;
1130                 goto out;
1131         }
1132
1133         if (last_idx >= 0)
1134                 fib_result_assign(res, last_resort);
1135         tb->tb_default = last_idx;
1136 out:
1137         return;
1138 }
1139
1140 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1141
1142 /*
1143  * Dead device goes up. We wake up dead nexthops.
1144  * It takes sense only on multipath routes.
1145  */
1146 int fib_sync_up(struct net_device *dev)
1147 {
1148         struct fib_info *prev_fi;
1149         unsigned int hash;
1150         struct hlist_head *head;
1151         struct hlist_node *node;
1152         struct fib_nh *nh;
1153         int ret;
1154
1155         if (!(dev->flags & IFF_UP))
1156                 return 0;
1157
1158         prev_fi = NULL;
1159         hash = fib_devindex_hashfn(dev->ifindex);
1160         head = &fib_info_devhash[hash];
1161         ret = 0;
1162
1163         hlist_for_each_entry(nh, node, head, nh_hash) {
1164                 struct fib_info *fi = nh->nh_parent;
1165                 int alive;
1166
1167                 BUG_ON(!fi->fib_nhs);
1168                 if (nh->nh_dev != dev || fi == prev_fi)
1169                         continue;
1170
1171                 prev_fi = fi;
1172                 alive = 0;
1173                 change_nexthops(fi) {
1174                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1175                                 alive++;
1176                                 continue;
1177                         }
1178                         if (nexthop_nh->nh_dev == NULL ||
1179                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1180                                 continue;
1181                         if (nexthop_nh->nh_dev != dev ||
1182                             !__in_dev_get_rtnl(dev))
1183                                 continue;
1184                         alive++;
1185                         spin_lock_bh(&fib_multipath_lock);
1186                         nexthop_nh->nh_power = 0;
1187                         nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1188                         spin_unlock_bh(&fib_multipath_lock);
1189                 } endfor_nexthops(fi)
1190
1191                 if (alive > 0) {
1192                         fi->fib_flags &= ~RTNH_F_DEAD;
1193                         ret++;
1194                 }
1195         }
1196
1197         return ret;
1198 }
1199
1200 /*
1201  * The algorithm is suboptimal, but it provides really
1202  * fair weighted route distribution.
1203  */
1204 void fib_select_multipath(struct fib_result *res)
1205 {
1206         struct fib_info *fi = res->fi;
1207         int w;
1208
1209         spin_lock_bh(&fib_multipath_lock);
1210         if (fi->fib_power <= 0) {
1211                 int power = 0;
1212                 change_nexthops(fi) {
1213                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1214                                 power += nexthop_nh->nh_weight;
1215                                 nexthop_nh->nh_power = nexthop_nh->nh_weight;
1216                         }
1217                 } endfor_nexthops(fi);
1218                 fi->fib_power = power;
1219                 if (power <= 0) {
1220                         spin_unlock_bh(&fib_multipath_lock);
1221                         /* Race condition: route has just become dead. */
1222                         res->nh_sel = 0;
1223                         return;
1224                 }
1225         }
1226
1227
1228         /* w should be random number [0..fi->fib_power-1],
1229          * it is pretty bad approximation.
1230          */
1231
1232         w = jiffies % fi->fib_power;
1233
1234         change_nexthops(fi) {
1235                 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1236                     nexthop_nh->nh_power) {
1237                         w -= nexthop_nh->nh_power;
1238                         if (w <= 0) {
1239                                 nexthop_nh->nh_power--;
1240                                 fi->fib_power--;
1241                                 res->nh_sel = nhsel;
1242                                 spin_unlock_bh(&fib_multipath_lock);
1243                                 return;
1244                         }
1245                 }
1246         } endfor_nexthops(fi);
1247
1248         /* Race condition: route has just become dead. */
1249         res->nh_sel = 0;
1250         spin_unlock_bh(&fib_multipath_lock);
1251 }
1252 #endif