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 <linux/bitops.h>
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/jiffies.h>
  21 #include <linux/mm.h>
  22 #include <linux/string.h>
  23 #include <linux/socket.h>
  24 #include <linux/sockios.h>
  25 #include <linux/errno.h>
  26 #include <linux/in.h>
  27 #include <linux/inet.h>
  28 #include <linux/inetdevice.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/if_arp.h>
  31 #include <linux/proc_fs.h>
  32 #include <linux/skbuff.h>
  33 #include <linux/init.h>
  34 #include <linux/slab.h>
  35
  36 #include <net/arp.h>
  37 #include <net/ip.h>
  38 #include <net/protocol.h>
  39 #include <net/route.h>
  40 #include <net/tcp.h>
  41 #include <net/sock.h>
  42 #include <net/ip_fib.h>
  43 #include <net/netlink.h>
  44 #include <net/nexthop.h>
  45
  46 #include "fib_lookup.h"
  47
  48 static DEFINE_SPINLOCK(fib_info_lock);
  49 static struct hlist_head *fib_info_hash;
  50 static struct hlist_head *fib_info_laddrhash;
  51 static unsigned int fib_info_hash_size;
  52 static unsigned int fib_info_cnt;
  53
  54 #define DEVINDEX_HASHBITS 8
  55 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
  56 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
  57
  58 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  59
  60 static DEFINE_SPINLOCK(fib_multipath_lock);
  61
  62 #define for_nexthops(fi) {                                              \
  63         int nhsel; const struct fib_nh *nh;                             \
  64         for (nhsel = 0, nh = (fi)->fib_nh;                              \
  65              nhsel < (fi)->fib_nhs;                                     \
  66              nh++, nhsel++)
  67
  68 #define change_nexthops(fi) {                                           \
  69         int nhsel; struct fib_nh *nexthop_nh;                           \
  70         for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh);   \
  71              nhsel < (fi)->fib_nhs;                                     \
  72              nexthop_nh++, nhsel++)
  73
  74 #else /* CONFIG_IP_ROUTE_MULTIPATH */
  75
  76 /* Hope, that gcc will optimize it to get rid of dummy loop */
  77
  78 #define for_nexthops(fi) {                                              \
  79         int nhsel; const struct fib_nh *nh = (fi)->fib_nh;              \
  80         for (nhsel = 0; nhsel < 1; nhsel++)
  81
  82 #define change_nexthops(fi) {                                           \
  83         int nhsel;                                                      \
  84         struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);    \
  85         for (nhsel = 0; nhsel < 1; nhsel++)
  86
  87 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
  88
  89 #define endfor_nexthops(fi) }
  90
  91
  92 const struct fib_prop fib_props[RTN_MAX + 1] = {
  93         [RTN_UNSPEC] = {
  94                 .error  = 0,
  95                 .scope  = RT_SCOPE_NOWHERE,
  96         },
  97         [RTN_UNICAST] = {
  98                 .error  = 0,
  99                 .scope  = RT_SCOPE_UNIVERSE,
 100         },
 101         [RTN_LOCAL] = {
 102                 .error  = 0,
 103                 .scope  = RT_SCOPE_HOST,
 104         },
 105         [RTN_BROADCAST] = {
 106                 .error  = 0,
 107                 .scope  = RT_SCOPE_LINK,
 108         },
 109         [RTN_ANYCAST] = {
 110                 .error  = 0,
 111                 .scope  = RT_SCOPE_LINK,
 112         },
 113         [RTN_MULTICAST] = {
 114                 .error  = 0,
 115                 .scope  = RT_SCOPE_UNIVERSE,
 116         },
 117         [RTN_BLACKHOLE] = {
 118                 .error  = -EINVAL,
 119                 .scope  = RT_SCOPE_UNIVERSE,
 120         },
 121         [RTN_UNREACHABLE] = {
 122                 .error  = -EHOSTUNREACH,
 123                 .scope  = RT_SCOPE_UNIVERSE,
 124         },
 125         [RTN_PROHIBIT] = {
 126                 .error  = -EACCES,
 127                 .scope  = RT_SCOPE_UNIVERSE,
 128         },
 129         [RTN_THROW] = {
 130                 .error  = -EAGAIN,
 131                 .scope  = RT_SCOPE_UNIVERSE,
 132         },
 133         [RTN_NAT] = {
 134                 .error  = -EINVAL,
 135                 .scope  = RT_SCOPE_NOWHERE,
 136         },
 137         [RTN_XRESOLVE] = {
 138                 .error  = -EINVAL,
 139                 .scope  = RT_SCOPE_NOWHERE,
 140         },
 141 };
 142
 143 static void free_nh_exceptions(struct fib_nh *nh)
 144 {
 145         struct fnhe_hash_bucket *hash = nh->nh_exceptions;
 146         int i;
 147
 148         for (i = 0; i < FNHE_HASH_SIZE; i++) {
 149                 struct fib_nh_exception *fnhe;
 150
 151                 fnhe = rcu_dereference_protected(hash[i].chain, 1);
 152                 while (fnhe) {
 153                         struct fib_nh_exception *next;
 154
 155                         next = rcu_dereference_protected(fnhe->fnhe_next, 1);
 156                         kfree(fnhe);
 157
 158                         fnhe = next;
 159                 }
 160         }
 161         kfree(hash);
 162 }
 163
 164 static void rt_nexthop_free(struct rtable __rcu **rtp)
 165 {
 166         struct rtable *rt = rcu_dereference_protected(*rtp, 1);
 167
 168         if (!rt)
 169                 return;
 170
 171         /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
 172          * because we waited an RCU grace period before calling
 173          * free_fib_info_rcu()
 174          */
 175
 176         dst_free(&rt->dst);
 177 }
 178
 179 static void rt_nexthop_free_cpus(struct rtable __rcu * __percpu *rtp)
 180 {
 181         int cpu;
 182
 183         if (!rtp)
 184                 return;
 185
 186         for_each_possible_cpu(cpu) {
 187                 struct rtable *rt;
 188
 189                 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
 190                 if (rt)
 191                         dst_free(&rt->dst);
 192         }
 193         free_percpu(rtp);
 194 }
 195
 196 /* Release a nexthop info record */
 197 static void free_fib_info_rcu(struct rcu_head *head)
 198 {
 199         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 200
 201         change_nexthops(fi) {
 202                 if (nexthop_nh->nh_dev)
 203                         dev_put(nexthop_nh->nh_dev);
 204                 if (nexthop_nh->nh_exceptions)
 205                         free_nh_exceptions(nexthop_nh);
 206                 rt_nexthop_free_cpus(nexthop_nh->nh_pcpu_rth_output);
 207                 rt_nexthop_free(&nexthop_nh->nh_rth_input);
 208         } endfor_nexthops(fi);
 209
 210         release_net(fi->fib_net);
 211         if (fi->fib_metrics != (u32 *) dst_default_metrics)
 212                 kfree(fi->fib_metrics);
 213         kfree(fi);
 214 }
 215
 216 void free_fib_info(struct fib_info *fi)
 217 {
 218         if (fi->fib_dead == 0) {
 219                 pr_warn("Freeing alive fib_info %p\n", fi);
 220                 return;
 221         }
 222         fib_info_cnt--;
 223 #ifdef CONFIG_IP_ROUTE_CLASSID
 224         change_nexthops(fi) {
 225                 if (nexthop_nh->nh_tclassid)
 226                         fi->fib_net->ipv4.fib_num_tclassid_users--;
 227         } endfor_nexthops(fi);
 228 #endif
 229         call_rcu(&fi->rcu, free_fib_info_rcu);
 230 }
 231
 232 void fib_release_info(struct fib_info *fi)
 233 {
 234         spin_lock_bh(&fib_info_lock);
 235         if (fi && --fi->fib_treeref == 0) {
 236                 hlist_del(&fi->fib_hash);
 237                 if (fi->fib_prefsrc)
 238                         hlist_del(&fi->fib_lhash);
 239                 change_nexthops(fi) {
 240                         if (!nexthop_nh->nh_dev)
 241                                 continue;
 242                         hlist_del(&nexthop_nh->nh_hash);
 243                 } endfor_nexthops(fi)
 244                 fi->fib_dead = 1;
 245                 fib_info_put(fi);
 246         }
 247         spin_unlock_bh(&fib_info_lock);
 248 }
 249
 250 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
 251 {
 252         const struct fib_nh *onh = ofi->fib_nh;
 253
 254         for_nexthops(fi) {
 255                 if (nh->nh_oif != onh->nh_oif ||
 256                     nh->nh_gw  != onh->nh_gw ||
 257                     nh->nh_scope != onh->nh_scope ||
 258 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 259                     nh->nh_weight != onh->nh_weight ||
 260 #endif
 261 #ifdef CONFIG_IP_ROUTE_CLASSID
 262                     nh->nh_tclassid != onh->nh_tclassid ||
 263 #endif
 264                     ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
 265                         return -1;
 266                 onh++;
 267         } endfor_nexthops(fi);
 268         return 0;
 269 }
 270
 271 static inline unsigned int fib_devindex_hashfn(unsigned int val)
 272 {
 273         unsigned int mask = DEVINDEX_HASHSIZE - 1;
 274
 275         return (val ^
 276                 (val >> DEVINDEX_HASHBITS) ^
 277                 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
 278 }
 279
 280 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
 281 {
 282         unsigned int mask = (fib_info_hash_size - 1);
 283         unsigned int val = fi->fib_nhs;
 284
 285         val ^= (fi->fib_protocol << 8) | fi->fib_scope;
 286         val ^= (__force u32)fi->fib_prefsrc;
 287         val ^= fi->fib_priority;
 288         for_nexthops(fi) {
 289                 val ^= fib_devindex_hashfn(nh->nh_oif);
 290         } endfor_nexthops(fi)
 291
 292         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 293 }
 294
 295 static struct fib_info *fib_find_info(const struct fib_info *nfi)
 296 {
 297         struct hlist_head *head;
 298         struct hlist_node *node;
 299         struct fib_info *fi;
 300         unsigned int hash;
 301
 302         hash = fib_info_hashfn(nfi);
 303         head = &fib_info_hash[hash];
 304
 305         hlist_for_each_entry(fi, node, head, fib_hash) {
 306                 if (!net_eq(fi->fib_net, nfi->fib_net))
 307                         continue;
 308                 if (fi->fib_nhs != nfi->fib_nhs)
 309                         continue;
 310                 if (nfi->fib_protocol == fi->fib_protocol &&
 311                     nfi->fib_scope == fi->fib_scope &&
 312                     nfi->fib_prefsrc == fi->fib_prefsrc &&
 313                     nfi->fib_priority == fi->fib_priority &&
 314                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 315                            sizeof(u32) * RTAX_MAX) == 0 &&
 316                     ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
 317                     (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
 318                         return fi;
 319         }
 320
 321         return NULL;
 322 }
 323
 324 /* Check, that the gateway is already configured.
 325  * Used only by redirect accept routine.
 326  */
 327 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 328 {
 329         struct hlist_head *head;
 330         struct hlist_node *node;
 331         struct fib_nh *nh;
 332         unsigned int hash;
 333
 334         spin_lock(&fib_info_lock);
 335
 336         hash = fib_devindex_hashfn(dev->ifindex);
 337         head = &fib_info_devhash[hash];
 338         hlist_for_each_entry(nh, node, head, nh_hash) {
 339                 if (nh->nh_dev == dev &&
 340                     nh->nh_gw == gw &&
 341                     !(nh->nh_flags & RTNH_F_DEAD)) {
 342                         spin_unlock(&fib_info_lock);
 343                         return 0;
 344                 }
 345         }
 346
 347         spin_unlock(&fib_info_lock);
 348
 349         return -1;
 350 }
 351
 352 static inline size_t fib_nlmsg_size(struct fib_info *fi)
 353 {
 354         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 355                          + nla_total_size(4) /* RTA_TABLE */
 356                          + nla_total_size(4) /* RTA_DST */
 357                          + nla_total_size(4) /* RTA_PRIORITY */
 358                          + nla_total_size(4); /* RTA_PREFSRC */
 359
 360         /* space for nested metrics */
 361         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 362
 363         if (fi->fib_nhs) {
 364                 /* Also handles the special case fib_nhs == 1 */
 365
 366                 /* each nexthop is packed in an attribute */
 367                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 368
 369                 /* may contain flow and gateway attribute */
 370                 nhsize += 2 * nla_total_size(4);
 371
 372                 /* all nexthops are packed in a nested attribute */
 373                 payload += nla_total_size(fi->fib_nhs * nhsize);
 374         }
 375
 376         return payload;
 377 }
 378
 379 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 380                int dst_len, u32 tb_id, struct nl_info *info,
 381                unsigned int nlm_flags)
 382 {
 383         struct sk_buff *skb;
 384         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 385         int err = -ENOBUFS;
 386
 387         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 388         if (skb == NULL)
 389                 goto errout;
 390
 391         err = fib_dump_info(skb, info->pid, seq, event, tb_id,
 392                             fa->fa_type, key, dst_len,
 393                             fa->fa_tos, fa->fa_info, nlm_flags);
 394         if (err < 0) {
 395                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 396                 WARN_ON(err == -EMSGSIZE);
 397                 kfree_skb(skb);
 398                 goto errout;
 399         }
 400         rtnl_notify(skb, info->nl_net, info->pid, RTNLGRP_IPV4_ROUTE,
 401                     info->nlh, GFP_KERNEL);
 402         return;
 403 errout:
 404         if (err < 0)
 405                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 406 }
 407
 408 /* Return the first fib alias matching TOS with
 409  * priority less than or equal to PRIO.
 410  */
 411 struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
 412 {
 413         if (fah) {
 414                 struct fib_alias *fa;
 415                 list_for_each_entry(fa, fah, fa_list) {
 416                         if (fa->fa_tos > tos)
 417                                 continue;
 418                         if (fa->fa_info->fib_priority >= prio ||
 419                             fa->fa_tos < tos)
 420                                 return fa;
 421                 }
 422         }
 423         return NULL;
 424 }
 425
 426 int fib_detect_death(struct fib_info *fi, int order,
 427                      struct fib_info **last_resort, int *last_idx, int dflt)
 428 {
 429         struct neighbour *n;
 430         int state = NUD_NONE;
 431
 432         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 433         if (n) {
 434                 state = n->nud_state;
 435                 neigh_release(n);
 436         }
 437         if (state == NUD_REACHABLE)
 438                 return 0;
 439         if ((state & NUD_VALID) && order != dflt)
 440                 return 0;
 441         if ((state & NUD_VALID) ||
 442             (*last_idx < 0 && order > dflt)) {
 443                 *last_resort = fi;
 444                 *last_idx = order;
 445         }
 446         return 1;
 447 }
 448
 449 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 450
 451 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
 452 {
 453         int nhs = 0;
 454
 455         while (rtnh_ok(rtnh, remaining)) {
 456                 nhs++;
 457                 rtnh = rtnh_next(rtnh, &remaining);
 458         }
 459
 460         /* leftover implies invalid nexthop configuration, discard it */
 461         return remaining > 0 ? 0 : nhs;
 462 }
 463
 464 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 465                        int remaining, struct fib_config *cfg)
 466 {
 467         change_nexthops(fi) {
 468                 int attrlen;
 469
 470                 if (!rtnh_ok(rtnh, remaining))
 471                         return -EINVAL;
 472
 473                 nexthop_nh->nh_flags =
 474                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 475                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
 476                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
 477
 478                 attrlen = rtnh_attrlen(rtnh);
 479                 if (attrlen > 0) {
 480                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 481
 482                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 483                         nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
 484 #ifdef CONFIG_IP_ROUTE_CLASSID
 485                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 486                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 487                         if (nexthop_nh->nh_tclassid)
 488                                 fi->fib_net->ipv4.fib_num_tclassid_users++;
 489 #endif
 490                 }
 491
 492                 rtnh = rtnh_next(rtnh, &remaining);
 493         } endfor_nexthops(fi);
 494
 495         return 0;
 496 }
 497
 498 #endif
 499
 500 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
 501 {
 502 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 503         struct rtnexthop *rtnh;
 504         int remaining;
 505 #endif
 506
 507         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 508                 return 1;
 509
 510         if (cfg->fc_oif || cfg->fc_gw) {
 511                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 512                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 513                         return 0;
 514                 return 1;
 515         }
 516
 517 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 518         if (cfg->fc_mp == NULL)
 519                 return 0;
 520
 521         rtnh = cfg->fc_mp;
 522         remaining = cfg->fc_mp_len;
 523
 524         for_nexthops(fi) {
 525                 int attrlen;
 526
 527                 if (!rtnh_ok(rtnh, remaining))
 528                         return -EINVAL;
 529
 530                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 531                         return 1;
 532
 533                 attrlen = rtnh_attrlen(rtnh);
 534                 if (attrlen < 0) {
 535                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 536
 537                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 538                         if (nla && nla_get_be32(nla) != nh->nh_gw)
 539                                 return 1;
 540 #ifdef CONFIG_IP_ROUTE_CLASSID
 541                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 542                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 543                                 return 1;
 544 #endif
 545                 }
 546
 547                 rtnh = rtnh_next(rtnh, &remaining);
 548         } endfor_nexthops(fi);
 549 #endif
 550         return 0;
 551 }
 552
 553
 554 /*
 555  * Picture
 556  * -------
 557  *
 558  * Semantics of nexthop is very messy by historical reasons.
 559  * We have to take into account, that:
 560  * a) gateway can be actually local interface address,
 561  *    so that gatewayed route is direct.
 562  * b) gateway must be on-link address, possibly
 563  *    described not by an ifaddr, but also by a direct route.
 564  * c) If both gateway and interface are specified, they should not
 565  *    contradict.
 566  * d) If we use tunnel routes, gateway could be not on-link.
 567  *
 568  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 569  * results in pretty ugly and hairy code with obscure logic.
 570  *
 571  * I chose to generalized it instead, so that the size
 572  * of code does not increase practically, but it becomes
 573  * much more general.
 574  * Every prefix is assigned a "scope" value: "host" is local address,
 575  * "link" is direct route,
 576  * [ ... "site" ... "interior" ... ]
 577  * and "universe" is true gateway route with global meaning.
 578  *
 579  * Every prefix refers to a set of "nexthop"s (gw, oif),
 580  * where gw must have narrower scope. This recursion stops
 581  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 582  * which means that gw is forced to be on link.
 583  *
 584  * Code is still hairy, but now it is apparently logically
 585  * consistent and very flexible. F.e. as by-product it allows
 586  * to co-exists in peace independent exterior and interior
 587  * routing processes.
 588  *
 589  * Normally it looks as following.
 590  *
 591  * {universe prefix}  -> (gw, oif) [scope link]
 592  *                |
 593  *                |-> {link prefix} -> (gw, oif) [scope local]
 594  *                                      |
 595  *                                      |-> {local prefix} (terminal node)
 596  */
 597 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
 598                         struct fib_nh *nh)
 599 {
 600         int err;
 601         struct net *net;
 602         struct net_device *dev;
 603
 604         net = cfg->fc_nlinfo.nl_net;
 605         if (nh->nh_gw) {
 606                 struct fib_result res;
 607
 608                 if (nh->nh_flags & RTNH_F_ONLINK) {
 609
 610                         if (cfg->fc_scope >= RT_SCOPE_LINK)
 611                                 return -EINVAL;
 612                         if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
 613                                 return -EINVAL;
 614                         dev = __dev_get_by_index(net, nh->nh_oif);
 615                         if (!dev)
 616                                 return -ENODEV;
 617                         if (!(dev->flags & IFF_UP))
 618                                 return -ENETDOWN;
 619                         nh->nh_dev = dev;
 620                         dev_hold(dev);
 621                         nh->nh_scope = RT_SCOPE_LINK;
 622                         return 0;
 623                 }
 624                 rcu_read_lock();
 625                 {
 626                         struct flowi4 fl4 = {
 627                                 .daddr = nh->nh_gw,
 628                                 .flowi4_scope = cfg->fc_scope + 1,
 629                                 .flowi4_oif = nh->nh_oif,
 630                         };
 631
 632                         /* It is not necessary, but requires a bit of thinking */
 633                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
 634                                 fl4.flowi4_scope = RT_SCOPE_LINK;
 635                         err = fib_lookup(net, &fl4, &res);
 636                         if (err) {
 637                                 rcu_read_unlock();
 638                                 return err;
 639                         }
 640                 }
 641                 err = -EINVAL;
 642                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 643                         goto out;
 644                 nh->nh_scope = res.scope;
 645                 nh->nh_oif = FIB_RES_OIF(res);
 646                 nh->nh_dev = dev = FIB_RES_DEV(res);
 647                 if (!dev)
 648                         goto out;
 649                 dev_hold(dev);
 650                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 651         } else {
 652                 struct in_device *in_dev;
 653
 654                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 655                         return -EINVAL;
 656
 657                 rcu_read_lock();
 658                 err = -ENODEV;
 659                 in_dev = inetdev_by_index(net, nh->nh_oif);
 660                 if (in_dev == NULL)
 661                         goto out;
 662                 err = -ENETDOWN;
 663                 if (!(in_dev->dev->flags & IFF_UP))
 664                         goto out;
 665                 nh->nh_dev = in_dev->dev;
 666                 dev_hold(nh->nh_dev);
 667                 nh->nh_scope = RT_SCOPE_HOST;
 668                 err = 0;
 669         }
 670 out:
 671         rcu_read_unlock();
 672         return err;
 673 }
 674
 675 static inline unsigned int fib_laddr_hashfn(__be32 val)
 676 {
 677         unsigned int mask = (fib_info_hash_size - 1);
 678
 679         return ((__force u32)val ^
 680                 ((__force u32)val >> 7) ^
 681                 ((__force u32)val >> 14)) & mask;
 682 }
 683
 684 static struct hlist_head *fib_info_hash_alloc(int bytes)
 685 {
 686         if (bytes <= PAGE_SIZE)
 687                 return kzalloc(bytes, GFP_KERNEL);
 688         else
 689                 return (struct hlist_head *)
 690                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 691                                          get_order(bytes));
 692 }
 693
 694 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 695 {
 696         if (!hash)
 697                 return;
 698
 699         if (bytes <= PAGE_SIZE)
 700                 kfree(hash);
 701         else
 702                 free_pages((unsigned long) hash, get_order(bytes));
 703 }
 704
 705 static void fib_info_hash_move(struct hlist_head *new_info_hash,
 706                                struct hlist_head *new_laddrhash,
 707                                unsigned int new_size)
 708 {
 709         struct hlist_head *old_info_hash, *old_laddrhash;
 710         unsigned int old_size = fib_info_hash_size;
 711         unsigned int i, bytes;
 712
 713         spin_lock_bh(&fib_info_lock);
 714         old_info_hash = fib_info_hash;
 715         old_laddrhash = fib_info_laddrhash;
 716         fib_info_hash_size = new_size;
 717
 718         for (i = 0; i < old_size; i++) {
 719                 struct hlist_head *head = &fib_info_hash[i];
 720                 struct hlist_node *node, *n;
 721                 struct fib_info *fi;
 722
 723                 hlist_for_each_entry_safe(fi, node, n, head, fib_hash) {
 724                         struct hlist_head *dest;
 725                         unsigned int new_hash;
 726
 727                         hlist_del(&fi->fib_hash);
 728
 729                         new_hash = fib_info_hashfn(fi);
 730                         dest = &new_info_hash[new_hash];
 731                         hlist_add_head(&fi->fib_hash, dest);
 732                 }
 733         }
 734         fib_info_hash = new_info_hash;
 735
 736         for (i = 0; i < old_size; i++) {
 737                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 738                 struct hlist_node *node, *n;
 739                 struct fib_info *fi;
 740
 741                 hlist_for_each_entry_safe(fi, node, n, lhead, fib_lhash) {
 742                         struct hlist_head *ldest;
 743                         unsigned int new_hash;
 744
 745                         hlist_del(&fi->fib_lhash);
 746
 747                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 748                         ldest = &new_laddrhash[new_hash];
 749                         hlist_add_head(&fi->fib_lhash, ldest);
 750                 }
 751         }
 752         fib_info_laddrhash = new_laddrhash;
 753
 754         spin_unlock_bh(&fib_info_lock);
 755
 756         bytes = old_size * sizeof(struct hlist_head *);
 757         fib_info_hash_free(old_info_hash, bytes);
 758         fib_info_hash_free(old_laddrhash, bytes);
 759 }
 760
 761 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 762 {
 763         nh->nh_saddr = inet_select_addr(nh->nh_dev,
 764                                         nh->nh_gw,
 765                                         nh->nh_parent->fib_scope);
 766         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 767
 768         return nh->nh_saddr;
 769 }
 770
 771 struct fib_info *fib_create_info(struct fib_config *cfg)
 772 {
 773         int err;
 774         struct fib_info *fi = NULL;
 775         struct fib_info *ofi;
 776         int nhs = 1;
 777         struct net *net = cfg->fc_nlinfo.nl_net;
 778
 779         if (cfg->fc_type > RTN_MAX)
 780                 goto err_inval;
 781
 782         /* Fast check to catch the most weird cases */
 783         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
 784                 goto err_inval;
 785
 786 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 787         if (cfg->fc_mp) {
 788                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
 789                 if (nhs == 0)
 790                         goto err_inval;
 791         }
 792 #endif
 793
 794         err = -ENOBUFS;
 795         if (fib_info_cnt >= fib_info_hash_size) {
 796                 unsigned int new_size = fib_info_hash_size << 1;
 797                 struct hlist_head *new_info_hash;
 798                 struct hlist_head *new_laddrhash;
 799                 unsigned int bytes;
 800
 801                 if (!new_size)
 802                         new_size = 1;
 803                 bytes = new_size * sizeof(struct hlist_head *);
 804                 new_info_hash = fib_info_hash_alloc(bytes);
 805                 new_laddrhash = fib_info_hash_alloc(bytes);
 806                 if (!new_info_hash || !new_laddrhash) {
 807                         fib_info_hash_free(new_info_hash, bytes);
 808                         fib_info_hash_free(new_laddrhash, bytes);
 809                 } else
 810                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
 811
 812                 if (!fib_info_hash_size)
 813                         goto failure;
 814         }
 815
 816         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
 817         if (fi == NULL)
 818                 goto failure;
 819         if (cfg->fc_mx) {
 820                 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
 821                 if (!fi->fib_metrics)
 822                         goto failure;
 823         } else
 824                 fi->fib_metrics = (u32 *) dst_default_metrics;
 825         fib_info_cnt++;
 826
 827         fi->fib_net = hold_net(net);
 828         fi->fib_protocol = cfg->fc_protocol;
 829         fi->fib_scope = cfg->fc_scope;
 830         fi->fib_flags = cfg->fc_flags;
 831         fi->fib_priority = cfg->fc_priority;
 832         fi->fib_prefsrc = cfg->fc_prefsrc;
 833
 834         fi->fib_nhs = nhs;
 835         change_nexthops(fi) {
 836                 nexthop_nh->nh_parent = fi;
 837                 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
 838         } endfor_nexthops(fi)
 839
 840         if (cfg->fc_mx) {
 841                 struct nlattr *nla;
 842                 int remaining;
 843
 844                 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 845                         int type = nla_type(nla);
 846
 847                         if (type) {
 848                                 u32 val;
 849
 850                                 if (type > RTAX_MAX)
 851                                         goto err_inval;
 852                                 val = nla_get_u32(nla);
 853                                 if (type == RTAX_ADVMSS && val > 65535 - 40)
 854                                         val = 65535 - 40;
 855                                 if (type == RTAX_MTU && val > 65535 - 15)
 856                                         val = 65535 - 15;
 857                                 fi->fib_metrics[type - 1] = val;
 858                         }
 859                 }
 860         }
 861
 862         if (cfg->fc_mp) {
 863 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 864                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
 865                 if (err != 0)
 866                         goto failure;
 867                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
 868                         goto err_inval;
 869                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
 870                         goto err_inval;
 871 #ifdef CONFIG_IP_ROUTE_CLASSID
 872                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
 873                         goto err_inval;
 874 #endif
 875 #else
 876                 goto err_inval;
 877 #endif
 878         } else {
 879                 struct fib_nh *nh = fi->fib_nh;
 880
 881                 nh->nh_oif = cfg->fc_oif;
 882                 nh->nh_gw = cfg->fc_gw;
 883                 nh->nh_flags = cfg->fc_flags;
 884 #ifdef CONFIG_IP_ROUTE_CLASSID
 885                 nh->nh_tclassid = cfg->fc_flow;
 886                 if (nh->nh_tclassid)
 887                         fi->fib_net->ipv4.fib_num_tclassid_users++;
 888 #endif
 889 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 890                 nh->nh_weight = 1;
 891 #endif
 892         }
 893
 894         if (fib_props[cfg->fc_type].error) {
 895                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
 896                         goto err_inval;
 897                 goto link_it;
 898         } else {
 899                 switch (cfg->fc_type) {
 900                 case RTN_UNICAST:
 901                 case RTN_LOCAL:
 902                 case RTN_BROADCAST:
 903                 case RTN_ANYCAST:
 904                 case RTN_MULTICAST:
 905                         break;
 906                 default:
 907                         goto err_inval;
 908                 }
 909         }
 910
 911         if (cfg->fc_scope > RT_SCOPE_HOST)
 912                 goto err_inval;
 913
 914         if (cfg->fc_scope == RT_SCOPE_HOST) {
 915                 struct fib_nh *nh = fi->fib_nh;
 916
 917                 /* Local address is added. */
 918                 if (nhs != 1 || nh->nh_gw)
 919                         goto err_inval;
 920                 nh->nh_scope = RT_SCOPE_NOWHERE;
 921                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
 922                 err = -ENODEV;
 923                 if (nh->nh_dev == NULL)
 924                         goto failure;
 925         } else {
 926                 change_nexthops(fi) {
 927                         err = fib_check_nh(cfg, fi, nexthop_nh);
 928                         if (err != 0)
 929                                 goto failure;
 930                 } endfor_nexthops(fi)
 931         }
 932
 933         if (fi->fib_prefsrc) {
 934                 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 935                     fi->fib_prefsrc != cfg->fc_dst)
 936                         if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
 937                                 goto err_inval;
 938         }
 939
 940         change_nexthops(fi) {
 941                 fib_info_update_nh_saddr(net, nexthop_nh);
 942         } endfor_nexthops(fi)
 943
 944 link_it:
 945         ofi = fib_find_info(fi);
 946         if (ofi) {
 947                 fi->fib_dead = 1;
 948                 free_fib_info(fi);
 949                 ofi->fib_treeref++;
 950                 return ofi;
 951         }
 952
 953         fi->fib_treeref++;
 954         atomic_inc(&fi->fib_clntref);
 955         spin_lock_bh(&fib_info_lock);
 956         hlist_add_head(&fi->fib_hash,
 957                        &fib_info_hash[fib_info_hashfn(fi)]);
 958         if (fi->fib_prefsrc) {
 959                 struct hlist_head *head;
 960
 961                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
 962                 hlist_add_head(&fi->fib_lhash, head);
 963         }
 964         change_nexthops(fi) {
 965                 struct hlist_head *head;
 966                 unsigned int hash;
 967
 968                 if (!nexthop_nh->nh_dev)
 969                         continue;
 970                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
 971                 head = &fib_info_devhash[hash];
 972                 hlist_add_head(&nexthop_nh->nh_hash, head);
 973         } endfor_nexthops(fi)
 974         spin_unlock_bh(&fib_info_lock);
 975         return fi;
 976
 977 err_inval:
 978         err = -EINVAL;
 979
 980 failure:
 981         if (fi) {
 982                 fi->fib_dead = 1;
 983                 free_fib_info(fi);
 984         }
 985
 986         return ERR_PTR(err);
 987 }
 988
 989 int fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
 990                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
 991                   struct fib_info *fi, unsigned int flags)
 992 {
 993         struct nlmsghdr *nlh;
 994         struct rtmsg *rtm;
 995
 996         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*rtm), flags);
 997         if (nlh == NULL)
 998                 return -EMSGSIZE;
 999
1000         rtm = nlmsg_data(nlh);
1001         rtm->rtm_family = AF_INET;
1002         rtm->rtm_dst_len = dst_len;
1003         rtm->rtm_src_len = 0;
1004         rtm->rtm_tos = tos;
1005         if (tb_id < 256)
1006                 rtm->rtm_table = tb_id;
1007         else
1008                 rtm->rtm_table = RT_TABLE_COMPAT;
1009         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1010                 goto nla_put_failure;
1011         rtm->rtm_type = type;
1012         rtm->rtm_flags = fi->fib_flags;
1013         rtm->rtm_scope = fi->fib_scope;
1014         rtm->rtm_protocol = fi->fib_protocol;
1015
1016         if (rtm->rtm_dst_len &&
1017             nla_put_be32(skb, RTA_DST, dst))
1018                 goto nla_put_failure;
1019         if (fi->fib_priority &&
1020             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1021                 goto nla_put_failure;
1022         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1023                 goto nla_put_failure;
1024
1025         if (fi->fib_prefsrc &&
1026             nla_put_be32(skb, RTA_PREFSRC, fi->fib_prefsrc))
1027                 goto nla_put_failure;
1028         if (fi->fib_nhs == 1) {
1029                 if (fi->fib_nh->nh_gw &&
1030                     nla_put_be32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1031                         goto nla_put_failure;
1032                 if (fi->fib_nh->nh_oif &&
1033                     nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1034                         goto nla_put_failure;
1035 #ifdef CONFIG_IP_ROUTE_CLASSID
1036                 if (fi->fib_nh[0].nh_tclassid &&
1037                     nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1038                         goto nla_put_failure;
1039 #endif
1040         }
1041 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1042         if (fi->fib_nhs > 1) {
1043                 struct rtnexthop *rtnh;
1044                 struct nlattr *mp;
1045
1046                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1047                 if (mp == NULL)
1048                         goto nla_put_failure;
1049
1050                 for_nexthops(fi) {
1051                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1052                         if (rtnh == NULL)
1053                                 goto nla_put_failure;
1054
1055                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1056                         rtnh->rtnh_hops = nh->nh_weight - 1;
1057                         rtnh->rtnh_ifindex = nh->nh_oif;
1058
1059                         if (nh->nh_gw &&
1060                             nla_put_be32(skb, RTA_GATEWAY, nh->nh_gw))
1061                                 goto nla_put_failure;
1062 #ifdef CONFIG_IP_ROUTE_CLASSID
1063                         if (nh->nh_tclassid &&
1064                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1065                                 goto nla_put_failure;
1066 #endif
1067                         /* length of rtnetlink header + attributes */
1068                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1069                 } endfor_nexthops(fi);
1070
1071                 nla_nest_end(skb, mp);
1072         }
1073 #endif
1074         return nlmsg_end(skb, nlh);
1075
1076 nla_put_failure:
1077         nlmsg_cancel(skb, nlh);
1078         return -EMSGSIZE;
1079 }
1080
1081 /*
1082  * Update FIB if:
1083  * - local address disappeared -> we must delete all the entries
1084  *   referring to it.
1085  * - device went down -> we must shutdown all nexthops going via it.
1086  */
1087 int fib_sync_down_addr(struct net *net, __be32 local)
1088 {
1089         int ret = 0;
1090         unsigned int hash = fib_laddr_hashfn(local);
1091         struct hlist_head *head = &fib_info_laddrhash[hash];
1092         struct hlist_node *node;
1093         struct fib_info *fi;
1094
1095         if (fib_info_laddrhash == NULL || local == 0)
1096                 return 0;
1097
1098         hlist_for_each_entry(fi, node, head, fib_lhash) {
1099                 if (!net_eq(fi->fib_net, net))
1100                         continue;
1101                 if (fi->fib_prefsrc == local) {
1102                         fi->fib_flags |= RTNH_F_DEAD;
1103                         ret++;
1104                 }
1105         }
1106         return ret;
1107 }
1108
1109 int fib_sync_down_dev(struct net_device *dev, int force)
1110 {
1111         int ret = 0;
1112         int scope = RT_SCOPE_NOWHERE;
1113         struct fib_info *prev_fi = NULL;
1114         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1115         struct hlist_head *head = &fib_info_devhash[hash];
1116         struct hlist_node *node;
1117         struct fib_nh *nh;
1118
1119         if (force)
1120                 scope = -1;
1121
1122         hlist_for_each_entry(nh, node, head, nh_hash) {
1123                 struct fib_info *fi = nh->nh_parent;
1124                 int dead;
1125
1126                 BUG_ON(!fi->fib_nhs);
1127                 if (nh->nh_dev != dev || fi == prev_fi)
1128                         continue;
1129                 prev_fi = fi;
1130                 dead = 0;
1131                 change_nexthops(fi) {
1132                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1133                                 dead++;
1134                         else if (nexthop_nh->nh_dev == dev &&
1135                                  nexthop_nh->nh_scope != scope) {
1136                                 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1137 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1138                                 spin_lock_bh(&fib_multipath_lock);
1139                                 fi->fib_power -= nexthop_nh->nh_power;
1140                                 nexthop_nh->nh_power = 0;
1141                                 spin_unlock_bh(&fib_multipath_lock);
1142 #endif
1143                                 dead++;
1144                         }
1145 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1146                         if (force > 1 && nexthop_nh->nh_dev == dev) {
1147                                 dead = fi->fib_nhs;
1148                                 break;
1149                         }
1150 #endif
1151                 } endfor_nexthops(fi)
1152                 if (dead == fi->fib_nhs) {
1153                         fi->fib_flags |= RTNH_F_DEAD;
1154                         ret++;
1155                 }
1156         }
1157
1158         return ret;
1159 }
1160
1161 /* Must be invoked inside of an RCU protected region.  */
1162 void fib_select_default(struct fib_result *res)
1163 {
1164         struct fib_info *fi = NULL, *last_resort = NULL;
1165         struct list_head *fa_head = res->fa_head;
1166         struct fib_table *tb = res->table;
1167         int order = -1, last_idx = -1;
1168         struct fib_alias *fa;
1169
1170         list_for_each_entry_rcu(fa, fa_head, fa_list) {
1171                 struct fib_info *next_fi = fa->fa_info;
1172
1173                 if (next_fi->fib_scope != res->scope ||
1174                     fa->fa_type != RTN_UNICAST)
1175                         continue;
1176
1177                 if (next_fi->fib_priority > res->fi->fib_priority)
1178                         break;
1179                 if (!next_fi->fib_nh[0].nh_gw ||
1180                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1181                         continue;
1182
1183                 fib_alias_accessed(fa);
1184
1185                 if (fi == NULL) {
1186                         if (next_fi != res->fi)
1187                                 break;
1188                 } else if (!fib_detect_death(fi, order, &last_resort,
1189                                              &last_idx, tb->tb_default)) {
1190                         fib_result_assign(res, fi);
1191                         tb->tb_default = order;
1192                         goto out;
1193                 }
1194                 fi = next_fi;
1195                 order++;
1196         }
1197
1198         if (order <= 0 || fi == NULL) {
1199                 tb->tb_default = -1;
1200                 goto out;
1201         }
1202
1203         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1204                                 tb->tb_default)) {
1205                 fib_result_assign(res, fi);
1206                 tb->tb_default = order;
1207                 goto out;
1208         }
1209
1210         if (last_idx >= 0)
1211                 fib_result_assign(res, last_resort);
1212         tb->tb_default = last_idx;
1213 out:
1214         return;
1215 }
1216
1217 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1218
1219 /*
1220  * Dead device goes up. We wake up dead nexthops.
1221  * It takes sense only on multipath routes.
1222  */
1223 int fib_sync_up(struct net_device *dev)
1224 {
1225         struct fib_info *prev_fi;
1226         unsigned int hash;
1227         struct hlist_head *head;
1228         struct hlist_node *node;
1229         struct fib_nh *nh;
1230         int ret;
1231
1232         if (!(dev->flags & IFF_UP))
1233                 return 0;
1234
1235         prev_fi = NULL;
1236         hash = fib_devindex_hashfn(dev->ifindex);
1237         head = &fib_info_devhash[hash];
1238         ret = 0;
1239
1240         hlist_for_each_entry(nh, node, head, nh_hash) {
1241                 struct fib_info *fi = nh->nh_parent;
1242                 int alive;
1243
1244                 BUG_ON(!fi->fib_nhs);
1245                 if (nh->nh_dev != dev || fi == prev_fi)
1246                         continue;
1247
1248                 prev_fi = fi;
1249                 alive = 0;
1250                 change_nexthops(fi) {
1251                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1252                                 alive++;
1253                                 continue;
1254                         }
1255                         if (nexthop_nh->nh_dev == NULL ||
1256                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1257                                 continue;
1258                         if (nexthop_nh->nh_dev != dev ||
1259                             !__in_dev_get_rtnl(dev))
1260                                 continue;
1261                         alive++;
1262                         spin_lock_bh(&fib_multipath_lock);
1263                         nexthop_nh->nh_power = 0;
1264                         nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1265                         spin_unlock_bh(&fib_multipath_lock);
1266                 } endfor_nexthops(fi)
1267
1268                 if (alive > 0) {
1269                         fi->fib_flags &= ~RTNH_F_DEAD;
1270                         ret++;
1271                 }
1272         }
1273
1274         return ret;
1275 }
1276
1277 /*
1278  * The algorithm is suboptimal, but it provides really
1279  * fair weighted route distribution.
1280  */
1281 void fib_select_multipath(struct fib_result *res)
1282 {
1283         struct fib_info *fi = res->fi;
1284         int w;
1285
1286         spin_lock_bh(&fib_multipath_lock);
1287         if (fi->fib_power <= 0) {
1288                 int power = 0;
1289                 change_nexthops(fi) {
1290                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1291                                 power += nexthop_nh->nh_weight;
1292                                 nexthop_nh->nh_power = nexthop_nh->nh_weight;
1293                         }
1294                 } endfor_nexthops(fi);
1295                 fi->fib_power = power;
1296                 if (power <= 0) {
1297                         spin_unlock_bh(&fib_multipath_lock);
1298                         /* Race condition: route has just become dead. */
1299                         res->nh_sel = 0;
1300                         return;
1301                 }
1302         }
1303
1304
1305         /* w should be random number [0..fi->fib_power-1],
1306          * it is pretty bad approximation.
1307          */
1308
1309         w = jiffies % fi->fib_power;
1310
1311         change_nexthops(fi) {
1312                 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1313                     nexthop_nh->nh_power) {
1314                         w -= nexthop_nh->nh_power;
1315                         if (w <= 0) {
1316                                 nexthop_nh->nh_power--;
1317                                 fi->fib_power--;
1318                                 res->nh_sel = nhsel;
1319                                 spin_unlock_bh(&fib_multipath_lock);
1320                                 return;
1321                         }
1322                 }
1323         } endfor_nexthops(fi);
1324
1325         /* Race condition: route has just become dead. */
1326         res->nh_sel = 0;
1327         spin_unlock_bh(&fib_multipath_lock);
1328 }
1329 #endif