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 rt_fibinfo_free(struct rtable __rcu **rtp)
 144 {
 145         struct rtable *rt = rcu_dereference_protected(*rtp, 1);
 146
 147         if (!rt)
 148                 return;
 149
 150         /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
 151          * because we waited an RCU grace period before calling
 152          * free_fib_info_rcu()
 153          */
 154
 155         dst_free(&rt->dst);
 156 }
 157
 158 static void free_nh_exceptions(struct fib_nh *nh)
 159 {
 160         struct fnhe_hash_bucket *hash = nh->nh_exceptions;
 161         int i;
 162
 163         for (i = 0; i < FNHE_HASH_SIZE; i++) {
 164                 struct fib_nh_exception *fnhe;
 165
 166                 fnhe = rcu_dereference_protected(hash[i].chain, 1);
 167                 while (fnhe) {
 168                         struct fib_nh_exception *next;
 169
 170                         next = rcu_dereference_protected(fnhe->fnhe_next, 1);
 171
 172                         rt_fibinfo_free(&fnhe->fnhe_rth_input);
 173                         rt_fibinfo_free(&fnhe->fnhe_rth_output);
 174
 175                         kfree(fnhe);
 176
 177                         fnhe = next;
 178                 }
 179         }
 180         kfree(hash);
 181 }
 182
 183 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
 184 {
 185         int cpu;
 186
 187         if (!rtp)
 188                 return;
 189
 190         for_each_possible_cpu(cpu) {
 191                 struct rtable *rt;
 192
 193                 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
 194                 if (rt)
 195                         dst_free(&rt->dst);
 196         }
 197         free_percpu(rtp);
 198 }
 199
 200 /* Release a nexthop info record */
 201 static void free_fib_info_rcu(struct rcu_head *head)
 202 {
 203         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 204
 205         change_nexthops(fi) {
 206                 if (nexthop_nh->nh_dev)
 207                         dev_put(nexthop_nh->nh_dev);
 208                 if (nexthop_nh->nh_exceptions)
 209                         free_nh_exceptions(nexthop_nh);
 210                 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
 211                 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
 212         } endfor_nexthops(fi);
 213
 214         release_net(fi->fib_net);
 215         if (fi->fib_metrics != (u32 *) dst_default_metrics)
 216                 kfree(fi->fib_metrics);
 217         kfree(fi);
 218 }
 219
 220 void free_fib_info(struct fib_info *fi)
 221 {
 222         if (fi->fib_dead == 0) {
 223                 pr_warn("Freeing alive fib_info %p\n", fi);
 224                 return;
 225         }
 226         fib_info_cnt--;
 227 #ifdef CONFIG_IP_ROUTE_CLASSID
 228         change_nexthops(fi) {
 229                 if (nexthop_nh->nh_tclassid)
 230                         fi->fib_net->ipv4.fib_num_tclassid_users--;
 231         } endfor_nexthops(fi);
 232 #endif
 233         call_rcu(&fi->rcu, free_fib_info_rcu);
 234 }
 235
 236 void fib_release_info(struct fib_info *fi)
 237 {
 238         spin_lock_bh(&fib_info_lock);
 239         if (fi && --fi->fib_treeref == 0) {
 240                 hlist_del(&fi->fib_hash);
 241                 if (fi->fib_prefsrc)
 242                         hlist_del(&fi->fib_lhash);
 243                 change_nexthops(fi) {
 244                         if (!nexthop_nh->nh_dev)
 245                                 continue;
 246                         hlist_del(&nexthop_nh->nh_hash);
 247                 } endfor_nexthops(fi)
 248                 fi->fib_dead = 1;
 249                 fib_info_put(fi);
 250         }
 251         spin_unlock_bh(&fib_info_lock);
 252 }
 253
 254 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
 255 {
 256         const struct fib_nh *onh = ofi->fib_nh;
 257
 258         for_nexthops(fi) {
 259                 if (nh->nh_oif != onh->nh_oif ||
 260                     nh->nh_gw  != onh->nh_gw ||
 261                     nh->nh_scope != onh->nh_scope ||
 262 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 263                     nh->nh_weight != onh->nh_weight ||
 264 #endif
 265 #ifdef CONFIG_IP_ROUTE_CLASSID
 266                     nh->nh_tclassid != onh->nh_tclassid ||
 267 #endif
 268                     ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
 269                         return -1;
 270                 onh++;
 271         } endfor_nexthops(fi);
 272         return 0;
 273 }
 274
 275 static inline unsigned int fib_devindex_hashfn(unsigned int val)
 276 {
 277         unsigned int mask = DEVINDEX_HASHSIZE - 1;
 278
 279         return (val ^
 280                 (val >> DEVINDEX_HASHBITS) ^
 281                 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
 282 }
 283
 284 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
 285 {
 286         unsigned int mask = (fib_info_hash_size - 1);
 287         unsigned int val = fi->fib_nhs;
 288
 289         val ^= (fi->fib_protocol << 8) | fi->fib_scope;
 290         val ^= (__force u32)fi->fib_prefsrc;
 291         val ^= fi->fib_priority;
 292         for_nexthops(fi) {
 293                 val ^= fib_devindex_hashfn(nh->nh_oif);
 294         } endfor_nexthops(fi)
 295
 296         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 297 }
 298
 299 static struct fib_info *fib_find_info(const struct fib_info *nfi)
 300 {
 301         struct hlist_head *head;
 302         struct fib_info *fi;
 303         unsigned int hash;
 304
 305         hash = fib_info_hashfn(nfi);
 306         head = &fib_info_hash[hash];
 307
 308         hlist_for_each_entry(fi, head, fib_hash) {
 309                 if (!net_eq(fi->fib_net, nfi->fib_net))
 310                         continue;
 311                 if (fi->fib_nhs != nfi->fib_nhs)
 312                         continue;
 313                 if (nfi->fib_protocol == fi->fib_protocol &&
 314                     nfi->fib_scope == fi->fib_scope &&
 315                     nfi->fib_prefsrc == fi->fib_prefsrc &&
 316                     nfi->fib_priority == fi->fib_priority &&
 317                     nfi->fib_type == fi->fib_type &&
 318                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 319                            sizeof(u32) * RTAX_MAX) == 0 &&
 320                     ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
 321                     (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
 322                         return fi;
 323         }
 324
 325         return NULL;
 326 }
 327
 328 /* Check, that the gateway is already configured.
 329  * Used only by redirect accept routine.
 330  */
 331 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 332 {
 333         struct hlist_head *head;
 334         struct fib_nh *nh;
 335         unsigned int hash;
 336
 337         spin_lock(&fib_info_lock);
 338
 339         hash = fib_devindex_hashfn(dev->ifindex);
 340         head = &fib_info_devhash[hash];
 341         hlist_for_each_entry(nh, head, nh_hash) {
 342                 if (nh->nh_dev == dev &&
 343                     nh->nh_gw == gw &&
 344                     !(nh->nh_flags & RTNH_F_DEAD)) {
 345                         spin_unlock(&fib_info_lock);
 346                         return 0;
 347                 }
 348         }
 349
 350         spin_unlock(&fib_info_lock);
 351
 352         return -1;
 353 }
 354
 355 static inline size_t fib_nlmsg_size(struct fib_info *fi)
 356 {
 357         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 358                          + nla_total_size(4) /* RTA_TABLE */
 359                          + nla_total_size(4) /* RTA_DST */
 360                          + nla_total_size(4) /* RTA_PRIORITY */
 361                          + nla_total_size(4); /* RTA_PREFSRC */
 362
 363         /* space for nested metrics */
 364         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 365
 366         if (fi->fib_nhs) {
 367                 /* Also handles the special case fib_nhs == 1 */
 368
 369                 /* each nexthop is packed in an attribute */
 370                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 371
 372                 /* may contain flow and gateway attribute */
 373                 nhsize += 2 * nla_total_size(4);
 374
 375                 /* all nexthops are packed in a nested attribute */
 376                 payload += nla_total_size(fi->fib_nhs * nhsize);
 377         }
 378
 379         return payload;
 380 }
 381
 382 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 383                int dst_len, u32 tb_id, const struct nl_info *info,
 384                unsigned int nlm_flags)
 385 {
 386         struct sk_buff *skb;
 387         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 388         int err = -ENOBUFS;
 389
 390         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 391         if (skb == NULL)
 392                 goto errout;
 393
 394         err = fib_dump_info(skb, info->portid, seq, event, tb_id,
 395                             fa->fa_type, key, dst_len,
 396                             fa->fa_tos, fa->fa_info, nlm_flags);
 397         if (err < 0) {
 398                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 399                 WARN_ON(err == -EMSGSIZE);
 400                 kfree_skb(skb);
 401                 goto errout;
 402         }
 403         rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
 404                     info->nlh, GFP_KERNEL);
 405         return;
 406 errout:
 407         if (err < 0)
 408                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 409 }
 410
 411 /* Return the first fib alias matching TOS with
 412  * priority less than or equal to PRIO.
 413  */
 414 struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
 415 {
 416         if (fah) {
 417                 struct fib_alias *fa;
 418                 list_for_each_entry(fa, fah, fa_list) {
 419                         if (fa->fa_tos > tos)
 420                                 continue;
 421                         if (fa->fa_info->fib_priority >= prio ||
 422                             fa->fa_tos < tos)
 423                                 return fa;
 424                 }
 425         }
 426         return NULL;
 427 }
 428
 429 static int fib_detect_death(struct fib_info *fi, int order,
 430                             struct fib_info **last_resort, int *last_idx,
 431                             int dflt)
 432 {
 433         struct neighbour *n;
 434         int state = NUD_NONE;
 435
 436         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 437         if (n) {
 438                 state = n->nud_state;
 439                 neigh_release(n);
 440         }
 441         if (state == NUD_REACHABLE)
 442                 return 0;
 443         if ((state & NUD_VALID) && order != dflt)
 444                 return 0;
 445         if ((state & NUD_VALID) ||
 446             (*last_idx < 0 && order > dflt)) {
 447                 *last_resort = fi;
 448                 *last_idx = order;
 449         }
 450         return 1;
 451 }
 452
 453 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 454
 455 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
 456 {
 457         int nhs = 0;
 458
 459         while (rtnh_ok(rtnh, remaining)) {
 460                 nhs++;
 461                 rtnh = rtnh_next(rtnh, &remaining);
 462         }
 463
 464         /* leftover implies invalid nexthop configuration, discard it */
 465         return remaining > 0 ? 0 : nhs;
 466 }
 467
 468 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 469                        int remaining, struct fib_config *cfg)
 470 {
 471         change_nexthops(fi) {
 472                 int attrlen;
 473
 474                 if (!rtnh_ok(rtnh, remaining))
 475                         return -EINVAL;
 476
 477                 nexthop_nh->nh_flags =
 478                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 479                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
 480                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
 481
 482                 attrlen = rtnh_attrlen(rtnh);
 483                 if (attrlen > 0) {
 484                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 485
 486                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 487                         nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
 488 #ifdef CONFIG_IP_ROUTE_CLASSID
 489                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 490                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 491                         if (nexthop_nh->nh_tclassid)
 492                                 fi->fib_net->ipv4.fib_num_tclassid_users++;
 493 #endif
 494                 }
 495
 496                 rtnh = rtnh_next(rtnh, &remaining);
 497         } endfor_nexthops(fi);
 498
 499         return 0;
 500 }
 501
 502 #endif
 503
 504 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
 505 {
 506 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 507         struct rtnexthop *rtnh;
 508         int remaining;
 509 #endif
 510
 511         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 512                 return 1;
 513
 514         if (cfg->fc_oif || cfg->fc_gw) {
 515                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 516                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 517                         return 0;
 518                 return 1;
 519         }
 520
 521 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 522         if (cfg->fc_mp == NULL)
 523                 return 0;
 524
 525         rtnh = cfg->fc_mp;
 526         remaining = cfg->fc_mp_len;
 527
 528         for_nexthops(fi) {
 529                 int attrlen;
 530
 531                 if (!rtnh_ok(rtnh, remaining))
 532                         return -EINVAL;
 533
 534                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 535                         return 1;
 536
 537                 attrlen = rtnh_attrlen(rtnh);
 538                 if (attrlen < 0) {
 539                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 540
 541                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 542                         if (nla && nla_get_be32(nla) != nh->nh_gw)
 543                                 return 1;
 544 #ifdef CONFIG_IP_ROUTE_CLASSID
 545                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 546                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 547                                 return 1;
 548 #endif
 549                 }
 550
 551                 rtnh = rtnh_next(rtnh, &remaining);
 552         } endfor_nexthops(fi);
 553 #endif
 554         return 0;
 555 }
 556
 557
 558 /*
 559  * Picture
 560  * -------
 561  *
 562  * Semantics of nexthop is very messy by historical reasons.
 563  * We have to take into account, that:
 564  * a) gateway can be actually local interface address,
 565  *    so that gatewayed route is direct.
 566  * b) gateway must be on-link address, possibly
 567  *    described not by an ifaddr, but also by a direct route.
 568  * c) If both gateway and interface are specified, they should not
 569  *    contradict.
 570  * d) If we use tunnel routes, gateway could be not on-link.
 571  *
 572  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 573  * results in pretty ugly and hairy code with obscure logic.
 574  *
 575  * I chose to generalized it instead, so that the size
 576  * of code does not increase practically, but it becomes
 577  * much more general.
 578  * Every prefix is assigned a "scope" value: "host" is local address,
 579  * "link" is direct route,
 580  * [ ... "site" ... "interior" ... ]
 581  * and "universe" is true gateway route with global meaning.
 582  *
 583  * Every prefix refers to a set of "nexthop"s (gw, oif),
 584  * where gw must have narrower scope. This recursion stops
 585  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 586  * which means that gw is forced to be on link.
 587  *
 588  * Code is still hairy, but now it is apparently logically
 589  * consistent and very flexible. F.e. as by-product it allows
 590  * to co-exists in peace independent exterior and interior
 591  * routing processes.
 592  *
 593  * Normally it looks as following.
 594  *
 595  * {universe prefix}  -> (gw, oif) [scope link]
 596  *                |
 597  *                |-> {link prefix} -> (gw, oif) [scope local]
 598  *                                      |
 599  *                                      |-> {local prefix} (terminal node)
 600  */
 601 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
 602                         struct fib_nh *nh)
 603 {
 604         int err;
 605         struct net *net;
 606         struct net_device *dev;
 607
 608         net = cfg->fc_nlinfo.nl_net;
 609         if (nh->nh_gw) {
 610                 struct fib_result res;
 611
 612                 if (nh->nh_flags & RTNH_F_ONLINK) {
 613
 614                         if (cfg->fc_scope >= RT_SCOPE_LINK)
 615                                 return -EINVAL;
 616                         if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
 617                                 return -EINVAL;
 618                         dev = __dev_get_by_index(net, nh->nh_oif);
 619                         if (!dev)
 620                                 return -ENODEV;
 621                         if (!(dev->flags & IFF_UP))
 622                                 return -ENETDOWN;
 623                         nh->nh_dev = dev;
 624                         dev_hold(dev);
 625                         nh->nh_scope = RT_SCOPE_LINK;
 626                         return 0;
 627                 }
 628                 rcu_read_lock();
 629                 {
 630                         struct flowi4 fl4 = {
 631                                 .daddr = nh->nh_gw,
 632                                 .flowi4_scope = cfg->fc_scope + 1,
 633                                 .flowi4_oif = nh->nh_oif,
 634                         };
 635
 636                         /* It is not necessary, but requires a bit of thinking */
 637                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
 638                                 fl4.flowi4_scope = RT_SCOPE_LINK;
 639                         err = fib_lookup(net, &fl4, &res);
 640                         if (err) {
 641                                 rcu_read_unlock();
 642                                 return err;
 643                         }
 644                 }
 645                 err = -EINVAL;
 646                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 647                         goto out;
 648                 nh->nh_scope = res.scope;
 649                 nh->nh_oif = FIB_RES_OIF(res);
 650                 nh->nh_dev = dev = FIB_RES_DEV(res);
 651                 if (!dev)
 652                         goto out;
 653                 dev_hold(dev);
 654                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 655         } else {
 656                 struct in_device *in_dev;
 657
 658                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 659                         return -EINVAL;
 660
 661                 rcu_read_lock();
 662                 err = -ENODEV;
 663                 in_dev = inetdev_by_index(net, nh->nh_oif);
 664                 if (in_dev == NULL)
 665                         goto out;
 666                 err = -ENETDOWN;
 667                 if (!(in_dev->dev->flags & IFF_UP))
 668                         goto out;
 669                 nh->nh_dev = in_dev->dev;
 670                 dev_hold(nh->nh_dev);
 671                 nh->nh_scope = RT_SCOPE_HOST;
 672                 err = 0;
 673         }
 674 out:
 675         rcu_read_unlock();
 676         return err;
 677 }
 678
 679 static inline unsigned int fib_laddr_hashfn(__be32 val)
 680 {
 681         unsigned int mask = (fib_info_hash_size - 1);
 682
 683         return ((__force u32)val ^
 684                 ((__force u32)val >> 7) ^
 685                 ((__force u32)val >> 14)) & mask;
 686 }
 687
 688 static struct hlist_head *fib_info_hash_alloc(int bytes)
 689 {
 690         if (bytes <= PAGE_SIZE)
 691                 return kzalloc(bytes, GFP_KERNEL);
 692         else
 693                 return (struct hlist_head *)
 694                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 695                                          get_order(bytes));
 696 }
 697
 698 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 699 {
 700         if (!hash)
 701                 return;
 702
 703         if (bytes <= PAGE_SIZE)
 704                 kfree(hash);
 705         else
 706                 free_pages((unsigned long) hash, get_order(bytes));
 707 }
 708
 709 static void fib_info_hash_move(struct hlist_head *new_info_hash,
 710                                struct hlist_head *new_laddrhash,
 711                                unsigned int new_size)
 712 {
 713         struct hlist_head *old_info_hash, *old_laddrhash;
 714         unsigned int old_size = fib_info_hash_size;
 715         unsigned int i, bytes;
 716
 717         spin_lock_bh(&fib_info_lock);
 718         old_info_hash = fib_info_hash;
 719         old_laddrhash = fib_info_laddrhash;
 720         fib_info_hash_size = new_size;
 721
 722         for (i = 0; i < old_size; i++) {
 723                 struct hlist_head *head = &fib_info_hash[i];
 724                 struct hlist_node *n;
 725                 struct fib_info *fi;
 726
 727                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
 728                         struct hlist_head *dest;
 729                         unsigned int new_hash;
 730
 731                         hlist_del(&fi->fib_hash);
 732
 733                         new_hash = fib_info_hashfn(fi);
 734                         dest = &new_info_hash[new_hash];
 735                         hlist_add_head(&fi->fib_hash, dest);
 736                 }
 737         }
 738         fib_info_hash = new_info_hash;
 739
 740         for (i = 0; i < old_size; i++) {
 741                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 742                 struct hlist_node *n;
 743                 struct fib_info *fi;
 744
 745                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
 746                         struct hlist_head *ldest;
 747                         unsigned int new_hash;
 748
 749                         hlist_del(&fi->fib_lhash);
 750
 751                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 752                         ldest = &new_laddrhash[new_hash];
 753                         hlist_add_head(&fi->fib_lhash, ldest);
 754                 }
 755         }
 756         fib_info_laddrhash = new_laddrhash;
 757
 758         spin_unlock_bh(&fib_info_lock);
 759
 760         bytes = old_size * sizeof(struct hlist_head *);
 761         fib_info_hash_free(old_info_hash, bytes);
 762         fib_info_hash_free(old_laddrhash, bytes);
 763 }
 764
 765 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 766 {
 767         nh->nh_saddr = inet_select_addr(nh->nh_dev,
 768                                         nh->nh_gw,
 769                                         nh->nh_parent->fib_scope);
 770         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 771
 772         return nh->nh_saddr;
 773 }
 774
 775 struct fib_info *fib_create_info(struct fib_config *cfg)
 776 {
 777         int err;
 778         struct fib_info *fi = NULL;
 779         struct fib_info *ofi;
 780         int nhs = 1;
 781         struct net *net = cfg->fc_nlinfo.nl_net;
 782
 783         if (cfg->fc_type > RTN_MAX)
 784                 goto err_inval;
 785
 786         /* Fast check to catch the most weird cases */
 787         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
 788                 goto err_inval;
 789
 790 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 791         if (cfg->fc_mp) {
 792                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
 793                 if (nhs == 0)
 794                         goto err_inval;
 795         }
 796 #endif
 797
 798         err = -ENOBUFS;
 799         if (fib_info_cnt >= fib_info_hash_size) {
 800                 unsigned int new_size = fib_info_hash_size << 1;
 801                 struct hlist_head *new_info_hash;
 802                 struct hlist_head *new_laddrhash;
 803                 unsigned int bytes;
 804
 805                 if (!new_size)
 806                         new_size = 16;
 807                 bytes = new_size * sizeof(struct hlist_head *);
 808                 new_info_hash = fib_info_hash_alloc(bytes);
 809                 new_laddrhash = fib_info_hash_alloc(bytes);
 810                 if (!new_info_hash || !new_laddrhash) {
 811                         fib_info_hash_free(new_info_hash, bytes);
 812                         fib_info_hash_free(new_laddrhash, bytes);
 813                 } else
 814                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
 815
 816                 if (!fib_info_hash_size)
 817                         goto failure;
 818         }
 819
 820         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
 821         if (fi == NULL)
 822                 goto failure;
 823         if (cfg->fc_mx) {
 824                 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
 825                 if (!fi->fib_metrics)
 826                         goto failure;
 827         } else
 828                 fi->fib_metrics = (u32 *) dst_default_metrics;
 829         fib_info_cnt++;
 830
 831         fi->fib_net = hold_net(net);
 832         fi->fib_protocol = cfg->fc_protocol;
 833         fi->fib_scope = cfg->fc_scope;
 834         fi->fib_flags = cfg->fc_flags;
 835         fi->fib_priority = cfg->fc_priority;
 836         fi->fib_prefsrc = cfg->fc_prefsrc;
 837         fi->fib_type = cfg->fc_type;
 838
 839         fi->fib_nhs = nhs;
 840         change_nexthops(fi) {
 841                 nexthop_nh->nh_parent = fi;
 842                 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
 843                 if (!nexthop_nh->nh_pcpu_rth_output)
 844                         goto failure;
 845         } endfor_nexthops(fi)
 846
 847         if (cfg->fc_mx) {
 848                 struct nlattr *nla;
 849                 int remaining;
 850
 851                 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 852                         int type = nla_type(nla);
 853
 854                         if (type) {
 855                                 u32 val;
 856
 857                                 if (type > RTAX_MAX)
 858                                         goto err_inval;
 859                                 val = nla_get_u32(nla);
 860                                 if (type == RTAX_ADVMSS && val > 65535 - 40)
 861                                         val = 65535 - 40;
 862                                 if (type == RTAX_MTU && val > 65535 - 15)
 863                                         val = 65535 - 15;
 864                                 fi->fib_metrics[type - 1] = val;
 865                         }
 866                 }
 867         }
 868
 869         if (cfg->fc_mp) {
 870 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 871                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
 872                 if (err != 0)
 873                         goto failure;
 874                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
 875                         goto err_inval;
 876                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
 877                         goto err_inval;
 878 #ifdef CONFIG_IP_ROUTE_CLASSID
 879                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
 880                         goto err_inval;
 881 #endif
 882 #else
 883                 goto err_inval;
 884 #endif
 885         } else {
 886                 struct fib_nh *nh = fi->fib_nh;
 887
 888                 nh->nh_oif = cfg->fc_oif;
 889                 nh->nh_gw = cfg->fc_gw;
 890                 nh->nh_flags = cfg->fc_flags;
 891 #ifdef CONFIG_IP_ROUTE_CLASSID
 892                 nh->nh_tclassid = cfg->fc_flow;
 893                 if (nh->nh_tclassid)
 894                         fi->fib_net->ipv4.fib_num_tclassid_users++;
 895 #endif
 896 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 897                 nh->nh_weight = 1;
 898 #endif
 899         }
 900
 901         if (fib_props[cfg->fc_type].error) {
 902                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
 903                         goto err_inval;
 904                 goto link_it;
 905         } else {
 906                 switch (cfg->fc_type) {
 907                 case RTN_UNICAST:
 908                 case RTN_LOCAL:
 909                 case RTN_BROADCAST:
 910                 case RTN_ANYCAST:
 911                 case RTN_MULTICAST:
 912                         break;
 913                 default:
 914                         goto err_inval;
 915                 }
 916         }
 917
 918         if (cfg->fc_scope > RT_SCOPE_HOST)
 919                 goto err_inval;
 920
 921         if (cfg->fc_scope == RT_SCOPE_HOST) {
 922                 struct fib_nh *nh = fi->fib_nh;
 923
 924                 /* Local address is added. */
 925                 if (nhs != 1 || nh->nh_gw)
 926                         goto err_inval;
 927                 nh->nh_scope = RT_SCOPE_NOWHERE;
 928                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
 929                 err = -ENODEV;
 930                 if (nh->nh_dev == NULL)
 931                         goto failure;
 932         } else {
 933                 change_nexthops(fi) {
 934                         err = fib_check_nh(cfg, fi, nexthop_nh);
 935                         if (err != 0)
 936                                 goto failure;
 937                 } endfor_nexthops(fi)
 938         }
 939
 940         if (fi->fib_prefsrc) {
 941                 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 942                     fi->fib_prefsrc != cfg->fc_dst)
 943                         if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
 944                                 goto err_inval;
 945         }
 946
 947         change_nexthops(fi) {
 948                 fib_info_update_nh_saddr(net, nexthop_nh);
 949         } endfor_nexthops(fi)
 950
 951 link_it:
 952         ofi = fib_find_info(fi);
 953         if (ofi) {
 954                 fi->fib_dead = 1;
 955                 free_fib_info(fi);
 956                 ofi->fib_treeref++;
 957                 return ofi;
 958         }
 959
 960         fi->fib_treeref++;
 961         atomic_inc(&fi->fib_clntref);
 962         spin_lock_bh(&fib_info_lock);
 963         hlist_add_head(&fi->fib_hash,
 964                        &fib_info_hash[fib_info_hashfn(fi)]);
 965         if (fi->fib_prefsrc) {
 966                 struct hlist_head *head;
 967
 968                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
 969                 hlist_add_head(&fi->fib_lhash, head);
 970         }
 971         change_nexthops(fi) {
 972                 struct hlist_head *head;
 973                 unsigned int hash;
 974
 975                 if (!nexthop_nh->nh_dev)
 976                         continue;
 977                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
 978                 head = &fib_info_devhash[hash];
 979                 hlist_add_head(&nexthop_nh->nh_hash, head);
 980         } endfor_nexthops(fi)
 981         spin_unlock_bh(&fib_info_lock);
 982         return fi;
 983
 984 err_inval:
 985         err = -EINVAL;
 986
 987 failure:
 988         if (fi) {
 989                 fi->fib_dead = 1;
 990                 free_fib_info(fi);
 991         }
 992
 993         return ERR_PTR(err);
 994 }
 995
 996 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
 997                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
 998                   struct fib_info *fi, unsigned int flags)
 999 {
1000         struct nlmsghdr *nlh;
1001         struct rtmsg *rtm;
1002
1003         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1004         if (nlh == NULL)
1005                 return -EMSGSIZE;
1006
1007         rtm = nlmsg_data(nlh);
1008         rtm->rtm_family = AF_INET;
1009         rtm->rtm_dst_len = dst_len;
1010         rtm->rtm_src_len = 0;
1011         rtm->rtm_tos = tos;
1012         if (tb_id < 256)
1013                 rtm->rtm_table = tb_id;
1014         else
1015                 rtm->rtm_table = RT_TABLE_COMPAT;
1016         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1017                 goto nla_put_failure;
1018         rtm->rtm_type = type;
1019         rtm->rtm_flags = fi->fib_flags;
1020         rtm->rtm_scope = fi->fib_scope;
1021         rtm->rtm_protocol = fi->fib_protocol;
1022
1023         if (rtm->rtm_dst_len &&
1024             nla_put_be32(skb, RTA_DST, dst))
1025                 goto nla_put_failure;
1026         if (fi->fib_priority &&
1027             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1028                 goto nla_put_failure;
1029         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1030                 goto nla_put_failure;
1031
1032         if (fi->fib_prefsrc &&
1033             nla_put_be32(skb, RTA_PREFSRC, fi->fib_prefsrc))
1034                 goto nla_put_failure;
1035         if (fi->fib_nhs == 1) {
1036                 if (fi->fib_nh->nh_gw &&
1037                     nla_put_be32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1038                         goto nla_put_failure;
1039                 if (fi->fib_nh->nh_oif &&
1040                     nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1041                         goto nla_put_failure;
1042 #ifdef CONFIG_IP_ROUTE_CLASSID
1043                 if (fi->fib_nh[0].nh_tclassid &&
1044                     nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1045                         goto nla_put_failure;
1046 #endif
1047         }
1048 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1049         if (fi->fib_nhs > 1) {
1050                 struct rtnexthop *rtnh;
1051                 struct nlattr *mp;
1052
1053                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1054                 if (mp == NULL)
1055                         goto nla_put_failure;
1056
1057                 for_nexthops(fi) {
1058                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1059                         if (rtnh == NULL)
1060                                 goto nla_put_failure;
1061
1062                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1063                         rtnh->rtnh_hops = nh->nh_weight - 1;
1064                         rtnh->rtnh_ifindex = nh->nh_oif;
1065
1066                         if (nh->nh_gw &&
1067                             nla_put_be32(skb, RTA_GATEWAY, nh->nh_gw))
1068                                 goto nla_put_failure;
1069 #ifdef CONFIG_IP_ROUTE_CLASSID
1070                         if (nh->nh_tclassid &&
1071                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1072                                 goto nla_put_failure;
1073 #endif
1074                         /* length of rtnetlink header + attributes */
1075                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1076                 } endfor_nexthops(fi);
1077
1078                 nla_nest_end(skb, mp);
1079         }
1080 #endif
1081         return nlmsg_end(skb, nlh);
1082
1083 nla_put_failure:
1084         nlmsg_cancel(skb, nlh);
1085         return -EMSGSIZE;
1086 }
1087
1088 /*
1089  * Update FIB if:
1090  * - local address disappeared -> we must delete all the entries
1091  *   referring to it.
1092  * - device went down -> we must shutdown all nexthops going via it.
1093  */
1094 int fib_sync_down_addr(struct net *net, __be32 local)
1095 {
1096         int ret = 0;
1097         unsigned int hash = fib_laddr_hashfn(local);
1098         struct hlist_head *head = &fib_info_laddrhash[hash];
1099         struct fib_info *fi;
1100
1101         if (fib_info_laddrhash == NULL || local == 0)
1102                 return 0;
1103
1104         hlist_for_each_entry(fi, head, fib_lhash) {
1105                 if (!net_eq(fi->fib_net, net))
1106                         continue;
1107                 if (fi->fib_prefsrc == local) {
1108                         fi->fib_flags |= RTNH_F_DEAD;
1109                         ret++;
1110                 }
1111         }
1112         return ret;
1113 }
1114
1115 int fib_sync_down_dev(struct net_device *dev, int force)
1116 {
1117         int ret = 0;
1118         int scope = RT_SCOPE_NOWHERE;
1119         struct fib_info *prev_fi = NULL;
1120         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1121         struct hlist_head *head = &fib_info_devhash[hash];
1122         struct fib_nh *nh;
1123
1124         if (force)
1125                 scope = -1;
1126
1127         hlist_for_each_entry(nh, head, nh_hash) {
1128                 struct fib_info *fi = nh->nh_parent;
1129                 int dead;
1130
1131                 BUG_ON(!fi->fib_nhs);
1132                 if (nh->nh_dev != dev || fi == prev_fi)
1133                         continue;
1134                 prev_fi = fi;
1135                 dead = 0;
1136                 change_nexthops(fi) {
1137                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1138                                 dead++;
1139                         else if (nexthop_nh->nh_dev == dev &&
1140                                  nexthop_nh->nh_scope != scope) {
1141                                 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1142 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1143                                 spin_lock_bh(&fib_multipath_lock);
1144                                 fi->fib_power -= nexthop_nh->nh_power;
1145                                 nexthop_nh->nh_power = 0;
1146                                 spin_unlock_bh(&fib_multipath_lock);
1147 #endif
1148                                 dead++;
1149                         }
1150 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1151                         if (force > 1 && nexthop_nh->nh_dev == dev) {
1152                                 dead = fi->fib_nhs;
1153                                 break;
1154                         }
1155 #endif
1156                 } endfor_nexthops(fi)
1157                 if (dead == fi->fib_nhs) {
1158                         fi->fib_flags |= RTNH_F_DEAD;
1159                         ret++;
1160                 }
1161         }
1162
1163         return ret;
1164 }
1165
1166 /* Must be invoked inside of an RCU protected region.  */
1167 void fib_select_default(struct fib_result *res)
1168 {
1169         struct fib_info *fi = NULL, *last_resort = NULL;
1170         struct list_head *fa_head = res->fa_head;
1171         struct fib_table *tb = res->table;
1172         int order = -1, last_idx = -1;
1173         struct fib_alias *fa;
1174
1175         list_for_each_entry_rcu(fa, fa_head, fa_list) {
1176                 struct fib_info *next_fi = fa->fa_info;
1177
1178                 if (next_fi->fib_scope != res->scope ||
1179                     fa->fa_type != RTN_UNICAST)
1180                         continue;
1181
1182                 if (next_fi->fib_priority > res->fi->fib_priority)
1183                         break;
1184                 if (!next_fi->fib_nh[0].nh_gw ||
1185                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1186                         continue;
1187
1188                 fib_alias_accessed(fa);
1189
1190                 if (fi == NULL) {
1191                         if (next_fi != res->fi)
1192                                 break;
1193                 } else if (!fib_detect_death(fi, order, &last_resort,
1194                                              &last_idx, tb->tb_default)) {
1195                         fib_result_assign(res, fi);
1196                         tb->tb_default = order;
1197                         goto out;
1198                 }
1199                 fi = next_fi;
1200                 order++;
1201         }
1202
1203         if (order <= 0 || fi == NULL) {
1204                 tb->tb_default = -1;
1205                 goto out;
1206         }
1207
1208         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1209                                 tb->tb_default)) {
1210                 fib_result_assign(res, fi);
1211                 tb->tb_default = order;
1212                 goto out;
1213         }
1214
1215         if (last_idx >= 0)
1216                 fib_result_assign(res, last_resort);
1217         tb->tb_default = last_idx;
1218 out:
1219         return;
1220 }
1221
1222 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1223
1224 /*
1225  * Dead device goes up. We wake up dead nexthops.
1226  * It takes sense only on multipath routes.
1227  */
1228 int fib_sync_up(struct net_device *dev)
1229 {
1230         struct fib_info *prev_fi;
1231         unsigned int hash;
1232         struct hlist_head *head;
1233         struct fib_nh *nh;
1234         int ret;
1235
1236         if (!(dev->flags & IFF_UP))
1237                 return 0;
1238
1239         prev_fi = NULL;
1240         hash = fib_devindex_hashfn(dev->ifindex);
1241         head = &fib_info_devhash[hash];
1242         ret = 0;
1243
1244         hlist_for_each_entry(nh, head, nh_hash) {
1245                 struct fib_info *fi = nh->nh_parent;
1246                 int alive;
1247
1248                 BUG_ON(!fi->fib_nhs);
1249                 if (nh->nh_dev != dev || fi == prev_fi)
1250                         continue;
1251
1252                 prev_fi = fi;
1253                 alive = 0;
1254                 change_nexthops(fi) {
1255                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1256                                 alive++;
1257                                 continue;
1258                         }
1259                         if (nexthop_nh->nh_dev == NULL ||
1260                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1261                                 continue;
1262                         if (nexthop_nh->nh_dev != dev ||
1263                             !__in_dev_get_rtnl(dev))
1264                                 continue;
1265                         alive++;
1266                         spin_lock_bh(&fib_multipath_lock);
1267                         nexthop_nh->nh_power = 0;
1268                         nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1269                         spin_unlock_bh(&fib_multipath_lock);
1270                 } endfor_nexthops(fi)
1271
1272                 if (alive > 0) {
1273                         fi->fib_flags &= ~RTNH_F_DEAD;
1274                         ret++;
1275                 }
1276         }
1277
1278         return ret;
1279 }
1280
1281 /*
1282  * The algorithm is suboptimal, but it provides really
1283  * fair weighted route distribution.
1284  */
1285 void fib_select_multipath(struct fib_result *res)
1286 {
1287         struct fib_info *fi = res->fi;
1288         int w;
1289
1290         spin_lock_bh(&fib_multipath_lock);
1291         if (fi->fib_power <= 0) {
1292                 int power = 0;
1293                 change_nexthops(fi) {
1294                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1295                                 power += nexthop_nh->nh_weight;
1296                                 nexthop_nh->nh_power = nexthop_nh->nh_weight;
1297                         }
1298                 } endfor_nexthops(fi);
1299                 fi->fib_power = power;
1300                 if (power <= 0) {
1301                         spin_unlock_bh(&fib_multipath_lock);
1302                         /* Race condition: route has just become dead. */
1303                         res->nh_sel = 0;
1304                         return;
1305                 }
1306         }
1307
1308
1309         /* w should be random number [0..fi->fib_power-1],
1310          * it is pretty bad approximation.
1311          */
1312
1313         w = jiffies % fi->fib_power;
1314
1315         change_nexthops(fi) {
1316                 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1317                     nexthop_nh->nh_power) {
1318                         w -= nexthop_nh->nh_power;
1319                         if (w <= 0) {
1320                                 nexthop_nh->nh_power--;
1321                                 fi->fib_power--;
1322                                 res->nh_sel = nhsel;
1323                                 spin_unlock_bh(&fib_multipath_lock);
1324                                 return;
1325                         }
1326                 }
1327         } endfor_nexthops(fi);
1328
1329         /* Race condition: route has just become dead. */
1330         res->nh_sel = 0;
1331         spin_unlock_bh(&fib_multipath_lock);
1332 }
1333 #endif