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;
 161         int i;
 162
 163         hash = rcu_dereference_protected(nh->nh_exceptions, 1);
 164         if (!hash)
 165                 return;
 166         for (i = 0; i < FNHE_HASH_SIZE; i++) {
 167                 struct fib_nh_exception *fnhe;
 168
 169                 fnhe = rcu_dereference_protected(hash[i].chain, 1);
 170                 while (fnhe) {
 171                         struct fib_nh_exception *next;
 172
 173                         next = rcu_dereference_protected(fnhe->fnhe_next, 1);
 174
 175                         rt_fibinfo_free(&fnhe->fnhe_rth_input);
 176                         rt_fibinfo_free(&fnhe->fnhe_rth_output);
 177
 178                         kfree(fnhe);
 179
 180                         fnhe = next;
 181                 }
 182         }
 183         kfree(hash);
 184 }
 185
 186 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
 187 {
 188         int cpu;
 189
 190         if (!rtp)
 191                 return;
 192
 193         for_each_possible_cpu(cpu) {
 194                 struct rtable *rt;
 195
 196                 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
 197                 if (rt)
 198                         dst_free(&rt->dst);
 199         }
 200         free_percpu(rtp);
 201 }
 202
 203 /* Release a nexthop info record */
 204 static void free_fib_info_rcu(struct rcu_head *head)
 205 {
 206         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 207
 208         change_nexthops(fi) {
 209                 if (nexthop_nh->nh_dev)
 210                         dev_put(nexthop_nh->nh_dev);
 211                 free_nh_exceptions(nexthop_nh);
 212                 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
 213                 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
 214         } endfor_nexthops(fi);
 215
 216         release_net(fi->fib_net);
 217         if (fi->fib_metrics != (u32 *) dst_default_metrics)
 218                 kfree(fi->fib_metrics);
 219         kfree(fi);
 220 }
 221
 222 void free_fib_info(struct fib_info *fi)
 223 {
 224         if (fi->fib_dead == 0) {
 225                 pr_warn("Freeing alive fib_info %p\n", fi);
 226                 return;
 227         }
 228         fib_info_cnt--;
 229 #ifdef CONFIG_IP_ROUTE_CLASSID
 230         change_nexthops(fi) {
 231                 if (nexthop_nh->nh_tclassid)
 232                         fi->fib_net->ipv4.fib_num_tclassid_users--;
 233         } endfor_nexthops(fi);
 234 #endif
 235         call_rcu(&fi->rcu, free_fib_info_rcu);
 236 }
 237
 238 void fib_release_info(struct fib_info *fi)
 239 {
 240         spin_lock_bh(&fib_info_lock);
 241         if (fi && --fi->fib_treeref == 0) {
 242                 hlist_del(&fi->fib_hash);
 243                 if (fi->fib_prefsrc)
 244                         hlist_del(&fi->fib_lhash);
 245                 change_nexthops(fi) {
 246                         if (!nexthop_nh->nh_dev)
 247                                 continue;
 248                         hlist_del(&nexthop_nh->nh_hash);
 249                 } endfor_nexthops(fi)
 250                 fi->fib_dead = 1;
 251                 fib_info_put(fi);
 252         }
 253         spin_unlock_bh(&fib_info_lock);
 254 }
 255
 256 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
 257 {
 258         const struct fib_nh *onh = ofi->fib_nh;
 259
 260         for_nexthops(fi) {
 261                 if (nh->nh_oif != onh->nh_oif ||
 262                     nh->nh_gw  != onh->nh_gw ||
 263                     nh->nh_scope != onh->nh_scope ||
 264 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 265                     nh->nh_weight != onh->nh_weight ||
 266 #endif
 267 #ifdef CONFIG_IP_ROUTE_CLASSID
 268                     nh->nh_tclassid != onh->nh_tclassid ||
 269 #endif
 270                     ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
 271                         return -1;
 272                 onh++;
 273         } endfor_nexthops(fi);
 274         return 0;
 275 }
 276
 277 static inline unsigned int fib_devindex_hashfn(unsigned int val)
 278 {
 279         unsigned int mask = DEVINDEX_HASHSIZE - 1;
 280
 281         return (val ^
 282                 (val >> DEVINDEX_HASHBITS) ^
 283                 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
 284 }
 285
 286 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
 287 {
 288         unsigned int mask = (fib_info_hash_size - 1);
 289         unsigned int val = fi->fib_nhs;
 290
 291         val ^= (fi->fib_protocol << 8) | fi->fib_scope;
 292         val ^= (__force u32)fi->fib_prefsrc;
 293         val ^= fi->fib_priority;
 294         for_nexthops(fi) {
 295                 val ^= fib_devindex_hashfn(nh->nh_oif);
 296         } endfor_nexthops(fi)
 297
 298         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 299 }
 300
 301 static struct fib_info *fib_find_info(const struct fib_info *nfi)
 302 {
 303         struct hlist_head *head;
 304         struct fib_info *fi;
 305         unsigned int hash;
 306
 307         hash = fib_info_hashfn(nfi);
 308         head = &fib_info_hash[hash];
 309
 310         hlist_for_each_entry(fi, head, fib_hash) {
 311                 if (!net_eq(fi->fib_net, nfi->fib_net))
 312                         continue;
 313                 if (fi->fib_nhs != nfi->fib_nhs)
 314                         continue;
 315                 if (nfi->fib_protocol == fi->fib_protocol &&
 316                     nfi->fib_scope == fi->fib_scope &&
 317                     nfi->fib_prefsrc == fi->fib_prefsrc &&
 318                     nfi->fib_priority == fi->fib_priority &&
 319                     nfi->fib_type == fi->fib_type &&
 320                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 321                            sizeof(u32) * RTAX_MAX) == 0 &&
 322                     ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
 323                     (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
 324                         return fi;
 325         }
 326
 327         return NULL;
 328 }
 329
 330 /* Check, that the gateway is already configured.
 331  * Used only by redirect accept routine.
 332  */
 333 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 334 {
 335         struct hlist_head *head;
 336         struct fib_nh *nh;
 337         unsigned int hash;
 338
 339         spin_lock(&fib_info_lock);
 340
 341         hash = fib_devindex_hashfn(dev->ifindex);
 342         head = &fib_info_devhash[hash];
 343         hlist_for_each_entry(nh, head, nh_hash) {
 344                 if (nh->nh_dev == dev &&
 345                     nh->nh_gw == gw &&
 346                     !(nh->nh_flags & RTNH_F_DEAD)) {
 347                         spin_unlock(&fib_info_lock);
 348                         return 0;
 349                 }
 350         }
 351
 352         spin_unlock(&fib_info_lock);
 353
 354         return -1;
 355 }
 356
 357 static inline size_t fib_nlmsg_size(struct fib_info *fi)
 358 {
 359         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 360                          + nla_total_size(4) /* RTA_TABLE */
 361                          + nla_total_size(4) /* RTA_DST */
 362                          + nla_total_size(4) /* RTA_PRIORITY */
 363                          + nla_total_size(4) /* RTA_PREFSRC */
 364                          + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
 365
 366         /* space for nested metrics */
 367         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 368
 369         if (fi->fib_nhs) {
 370                 /* Also handles the special case fib_nhs == 1 */
 371
 372                 /* each nexthop is packed in an attribute */
 373                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 374
 375                 /* may contain flow and gateway attribute */
 376                 nhsize += 2 * nla_total_size(4);
 377
 378                 /* all nexthops are packed in a nested attribute */
 379                 payload += nla_total_size(fi->fib_nhs * nhsize);
 380         }
 381
 382         return payload;
 383 }
 384
 385 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 386                int dst_len, u32 tb_id, const struct nl_info *info,
 387                unsigned int nlm_flags)
 388 {
 389         struct sk_buff *skb;
 390         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 391         int err = -ENOBUFS;
 392
 393         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 394         if (skb == NULL)
 395                 goto errout;
 396
 397         err = fib_dump_info(skb, info->portid, seq, event, tb_id,
 398                             fa->fa_type, key, dst_len,
 399                             fa->fa_tos, fa->fa_info, nlm_flags);
 400         if (err < 0) {
 401                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 402                 WARN_ON(err == -EMSGSIZE);
 403                 kfree_skb(skb);
 404                 goto errout;
 405         }
 406         rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
 407                     info->nlh, GFP_KERNEL);
 408         return;
 409 errout:
 410         if (err < 0)
 411                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 412 }
 413
 414 /* Return the first fib alias matching TOS with
 415  * priority less than or equal to PRIO.
 416  */
 417 struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
 418 {
 419         if (fah) {
 420                 struct fib_alias *fa;
 421                 list_for_each_entry(fa, fah, fa_list) {
 422                         if (fa->fa_tos > tos)
 423                                 continue;
 424                         if (fa->fa_info->fib_priority >= prio ||
 425                             fa->fa_tos < tos)
 426                                 return fa;
 427                 }
 428         }
 429         return NULL;
 430 }
 431
 432 static int fib_detect_death(struct fib_info *fi, int order,
 433                             struct fib_info **last_resort, int *last_idx,
 434                             int dflt)
 435 {
 436         struct neighbour *n;
 437         int state = NUD_NONE;
 438
 439         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 440         if (n) {
 441                 state = n->nud_state;
 442                 neigh_release(n);
 443         }
 444         if (state == NUD_REACHABLE)
 445                 return 0;
 446         if ((state & NUD_VALID) && order != dflt)
 447                 return 0;
 448         if ((state & NUD_VALID) ||
 449             (*last_idx < 0 && order > dflt)) {
 450                 *last_resort = fi;
 451                 *last_idx = order;
 452         }
 453         return 1;
 454 }
 455
 456 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 457
 458 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
 459 {
 460         int nhs = 0;
 461
 462         while (rtnh_ok(rtnh, remaining)) {
 463                 nhs++;
 464                 rtnh = rtnh_next(rtnh, &remaining);
 465         }
 466
 467         /* leftover implies invalid nexthop configuration, discard it */
 468         return remaining > 0 ? 0 : nhs;
 469 }
 470
 471 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 472                        int remaining, struct fib_config *cfg)
 473 {
 474         change_nexthops(fi) {
 475                 int attrlen;
 476
 477                 if (!rtnh_ok(rtnh, remaining))
 478                         return -EINVAL;
 479
 480                 nexthop_nh->nh_flags =
 481                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 482                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
 483                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
 484
 485                 attrlen = rtnh_attrlen(rtnh);
 486                 if (attrlen > 0) {
 487                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 488
 489                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 490                         nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
 491 #ifdef CONFIG_IP_ROUTE_CLASSID
 492                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 493                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 494                         if (nexthop_nh->nh_tclassid)
 495                                 fi->fib_net->ipv4.fib_num_tclassid_users++;
 496 #endif
 497                 }
 498
 499                 rtnh = rtnh_next(rtnh, &remaining);
 500         } endfor_nexthops(fi);
 501
 502         return 0;
 503 }
 504
 505 #endif
 506
 507 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
 508 {
 509 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 510         struct rtnexthop *rtnh;
 511         int remaining;
 512 #endif
 513
 514         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 515                 return 1;
 516
 517         if (cfg->fc_oif || cfg->fc_gw) {
 518                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 519                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 520                         return 0;
 521                 return 1;
 522         }
 523
 524 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 525         if (cfg->fc_mp == NULL)
 526                 return 0;
 527
 528         rtnh = cfg->fc_mp;
 529         remaining = cfg->fc_mp_len;
 530
 531         for_nexthops(fi) {
 532                 int attrlen;
 533
 534                 if (!rtnh_ok(rtnh, remaining))
 535                         return -EINVAL;
 536
 537                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 538                         return 1;
 539
 540                 attrlen = rtnh_attrlen(rtnh);
 541                 if (attrlen > 0) {
 542                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 543
 544                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 545                         if (nla && nla_get_be32(nla) != nh->nh_gw)
 546                                 return 1;
 547 #ifdef CONFIG_IP_ROUTE_CLASSID
 548                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 549                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 550                                 return 1;
 551 #endif
 552                 }
 553
 554                 rtnh = rtnh_next(rtnh, &remaining);
 555         } endfor_nexthops(fi);
 556 #endif
 557         return 0;
 558 }
 559
 560
 561 /*
 562  * Picture
 563  * -------
 564  *
 565  * Semantics of nexthop is very messy by historical reasons.
 566  * We have to take into account, that:
 567  * a) gateway can be actually local interface address,
 568  *    so that gatewayed route is direct.
 569  * b) gateway must be on-link address, possibly
 570  *    described not by an ifaddr, but also by a direct route.
 571  * c) If both gateway and interface are specified, they should not
 572  *    contradict.
 573  * d) If we use tunnel routes, gateway could be not on-link.
 574  *
 575  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 576  * results in pretty ugly and hairy code with obscure logic.
 577  *
 578  * I chose to generalized it instead, so that the size
 579  * of code does not increase practically, but it becomes
 580  * much more general.
 581  * Every prefix is assigned a "scope" value: "host" is local address,
 582  * "link" is direct route,
 583  * [ ... "site" ... "interior" ... ]
 584  * and "universe" is true gateway route with global meaning.
 585  *
 586  * Every prefix refers to a set of "nexthop"s (gw, oif),
 587  * where gw must have narrower scope. This recursion stops
 588  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 589  * which means that gw is forced to be on link.
 590  *
 591  * Code is still hairy, but now it is apparently logically
 592  * consistent and very flexible. F.e. as by-product it allows
 593  * to co-exists in peace independent exterior and interior
 594  * routing processes.
 595  *
 596  * Normally it looks as following.
 597  *
 598  * {universe prefix}  -> (gw, oif) [scope link]
 599  *                |
 600  *                |-> {link prefix} -> (gw, oif) [scope local]
 601  *                                      |
 602  *                                      |-> {local prefix} (terminal node)
 603  */
 604 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
 605                         struct fib_nh *nh)
 606 {
 607         int err;
 608         struct net *net;
 609         struct net_device *dev;
 610
 611         net = cfg->fc_nlinfo.nl_net;
 612         if (nh->nh_gw) {
 613                 struct fib_result res;
 614
 615                 if (nh->nh_flags & RTNH_F_ONLINK) {
 616
 617                         if (cfg->fc_scope >= RT_SCOPE_LINK)
 618                                 return -EINVAL;
 619                         if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
 620                                 return -EINVAL;
 621                         dev = __dev_get_by_index(net, nh->nh_oif);
 622                         if (!dev)
 623                                 return -ENODEV;
 624                         if (!(dev->flags & IFF_UP))
 625                                 return -ENETDOWN;
 626                         nh->nh_dev = dev;
 627                         dev_hold(dev);
 628                         nh->nh_scope = RT_SCOPE_LINK;
 629                         return 0;
 630                 }
 631                 rcu_read_lock();
 632                 {
 633                         struct flowi4 fl4 = {
 634                                 .daddr = nh->nh_gw,
 635                                 .flowi4_scope = cfg->fc_scope + 1,
 636                                 .flowi4_oif = nh->nh_oif,
 637                                 .flowi4_iif = LOOPBACK_IFINDEX,
 638                         };
 639
 640                         /* It is not necessary, but requires a bit of thinking */
 641                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
 642                                 fl4.flowi4_scope = RT_SCOPE_LINK;
 643                         err = fib_lookup(net, &fl4, &res);
 644                         if (err) {
 645                                 rcu_read_unlock();
 646                                 return err;
 647                         }
 648                 }
 649                 err = -EINVAL;
 650                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 651                         goto out;
 652                 nh->nh_scope = res.scope;
 653                 nh->nh_oif = FIB_RES_OIF(res);
 654                 nh->nh_dev = dev = FIB_RES_DEV(res);
 655                 if (!dev)
 656                         goto out;
 657                 dev_hold(dev);
 658                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 659         } else {
 660                 struct in_device *in_dev;
 661
 662                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 663                         return -EINVAL;
 664
 665                 rcu_read_lock();
 666                 err = -ENODEV;
 667                 in_dev = inetdev_by_index(net, nh->nh_oif);
 668                 if (in_dev == NULL)
 669                         goto out;
 670                 err = -ENETDOWN;
 671                 if (!(in_dev->dev->flags & IFF_UP))
 672                         goto out;
 673                 nh->nh_dev = in_dev->dev;
 674                 dev_hold(nh->nh_dev);
 675                 nh->nh_scope = RT_SCOPE_HOST;
 676                 err = 0;
 677         }
 678 out:
 679         rcu_read_unlock();
 680         return err;
 681 }
 682
 683 static inline unsigned int fib_laddr_hashfn(__be32 val)
 684 {
 685         unsigned int mask = (fib_info_hash_size - 1);
 686
 687         return ((__force u32)val ^
 688                 ((__force u32)val >> 7) ^
 689                 ((__force u32)val >> 14)) & mask;
 690 }
 691
 692 static struct hlist_head *fib_info_hash_alloc(int bytes)
 693 {
 694         if (bytes <= PAGE_SIZE)
 695                 return kzalloc(bytes, GFP_KERNEL);
 696         else
 697                 return (struct hlist_head *)
 698                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 699                                          get_order(bytes));
 700 }
 701
 702 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 703 {
 704         if (!hash)
 705                 return;
 706
 707         if (bytes <= PAGE_SIZE)
 708                 kfree(hash);
 709         else
 710                 free_pages((unsigned long) hash, get_order(bytes));
 711 }
 712
 713 static void fib_info_hash_move(struct hlist_head *new_info_hash,
 714                                struct hlist_head *new_laddrhash,
 715                                unsigned int new_size)
 716 {
 717         struct hlist_head *old_info_hash, *old_laddrhash;
 718         unsigned int old_size = fib_info_hash_size;
 719         unsigned int i, bytes;
 720
 721         spin_lock_bh(&fib_info_lock);
 722         old_info_hash = fib_info_hash;
 723         old_laddrhash = fib_info_laddrhash;
 724         fib_info_hash_size = new_size;
 725
 726         for (i = 0; i < old_size; i++) {
 727                 struct hlist_head *head = &fib_info_hash[i];
 728                 struct hlist_node *n;
 729                 struct fib_info *fi;
 730
 731                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
 732                         struct hlist_head *dest;
 733                         unsigned int new_hash;
 734
 735                         hlist_del(&fi->fib_hash);
 736
 737                         new_hash = fib_info_hashfn(fi);
 738                         dest = &new_info_hash[new_hash];
 739                         hlist_add_head(&fi->fib_hash, dest);
 740                 }
 741         }
 742         fib_info_hash = new_info_hash;
 743
 744         for (i = 0; i < old_size; i++) {
 745                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 746                 struct hlist_node *n;
 747                 struct fib_info *fi;
 748
 749                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
 750                         struct hlist_head *ldest;
 751                         unsigned int new_hash;
 752
 753                         hlist_del(&fi->fib_lhash);
 754
 755                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 756                         ldest = &new_laddrhash[new_hash];
 757                         hlist_add_head(&fi->fib_lhash, ldest);
 758                 }
 759         }
 760         fib_info_laddrhash = new_laddrhash;
 761
 762         spin_unlock_bh(&fib_info_lock);
 763
 764         bytes = old_size * sizeof(struct hlist_head *);
 765         fib_info_hash_free(old_info_hash, bytes);
 766         fib_info_hash_free(old_laddrhash, bytes);
 767 }
 768
 769 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 770 {
 771         nh->nh_saddr = inet_select_addr(nh->nh_dev,
 772                                         nh->nh_gw,
 773                                         nh->nh_parent->fib_scope);
 774         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 775
 776         return nh->nh_saddr;
 777 }
 778
 779 struct fib_info *fib_create_info(struct fib_config *cfg)
 780 {
 781         int err;
 782         struct fib_info *fi = NULL;
 783         struct fib_info *ofi;
 784         int nhs = 1;
 785         struct net *net = cfg->fc_nlinfo.nl_net;
 786
 787         if (cfg->fc_type > RTN_MAX)
 788                 goto err_inval;
 789
 790         /* Fast check to catch the most weird cases */
 791         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
 792                 goto err_inval;
 793
 794 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 795         if (cfg->fc_mp) {
 796                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
 797                 if (nhs == 0)
 798                         goto err_inval;
 799         }
 800 #endif
 801
 802         err = -ENOBUFS;
 803         if (fib_info_cnt >= fib_info_hash_size) {
 804                 unsigned int new_size = fib_info_hash_size << 1;
 805                 struct hlist_head *new_info_hash;
 806                 struct hlist_head *new_laddrhash;
 807                 unsigned int bytes;
 808
 809                 if (!new_size)
 810                         new_size = 16;
 811                 bytes = new_size * sizeof(struct hlist_head *);
 812                 new_info_hash = fib_info_hash_alloc(bytes);
 813                 new_laddrhash = fib_info_hash_alloc(bytes);
 814                 if (!new_info_hash || !new_laddrhash) {
 815                         fib_info_hash_free(new_info_hash, bytes);
 816                         fib_info_hash_free(new_laddrhash, bytes);
 817                 } else
 818                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
 819
 820                 if (!fib_info_hash_size)
 821                         goto failure;
 822         }
 823
 824         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
 825         if (fi == NULL)
 826                 goto failure;
 827         fib_info_cnt++;
 828         if (cfg->fc_mx) {
 829                 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
 830                 if (!fi->fib_metrics)
 831                         goto failure;
 832         } else
 833                 fi->fib_metrics = (u32 *) dst_default_metrics;
 834
 835         fi->fib_net = hold_net(net);
 836         fi->fib_protocol = cfg->fc_protocol;
 837         fi->fib_scope = cfg->fc_scope;
 838         fi->fib_flags = cfg->fc_flags;
 839         fi->fib_priority = cfg->fc_priority;
 840         fi->fib_prefsrc = cfg->fc_prefsrc;
 841         fi->fib_type = cfg->fc_type;
 842
 843         fi->fib_nhs = nhs;
 844         change_nexthops(fi) {
 845                 nexthop_nh->nh_parent = fi;
 846                 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
 847                 if (!nexthop_nh->nh_pcpu_rth_output)
 848                         goto failure;
 849         } endfor_nexthops(fi)
 850
 851         if (cfg->fc_mx) {
 852                 struct nlattr *nla;
 853                 int remaining;
 854
 855                 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 856                         int type = nla_type(nla);
 857
 858                         if (type) {
 859                                 u32 val;
 860
 861                                 if (type > RTAX_MAX)
 862                                         goto err_inval;
 863                                 if (type == RTAX_CC_ALGO) {
 864                                         char tmp[TCP_CA_NAME_MAX];
 865
 866                                         nla_strlcpy(tmp, nla, sizeof(tmp));
 867                                         val = tcp_ca_get_key_by_name(tmp);
 868                                         if (val == TCP_CA_UNSPEC)
 869                                                 goto err_inval;
 870                                 } else {
 871                                         val = nla_get_u32(nla);
 872                                 }
 873                                 if (type == RTAX_ADVMSS && val > 65535 - 40)
 874                                         val = 65535 - 40;
 875                                 if (type == RTAX_MTU && val > 65535 - 15)
 876                                         val = 65535 - 15;
 877                                 fi->fib_metrics[type - 1] = val;
 878                         }
 879                 }
 880         }
 881
 882         if (cfg->fc_mp) {
 883 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 884                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
 885                 if (err != 0)
 886                         goto failure;
 887                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
 888                         goto err_inval;
 889                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
 890                         goto err_inval;
 891 #ifdef CONFIG_IP_ROUTE_CLASSID
 892                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
 893                         goto err_inval;
 894 #endif
 895 #else
 896                 goto err_inval;
 897 #endif
 898         } else {
 899                 struct fib_nh *nh = fi->fib_nh;
 900
 901                 nh->nh_oif = cfg->fc_oif;
 902                 nh->nh_gw = cfg->fc_gw;
 903                 nh->nh_flags = cfg->fc_flags;
 904 #ifdef CONFIG_IP_ROUTE_CLASSID
 905                 nh->nh_tclassid = cfg->fc_flow;
 906                 if (nh->nh_tclassid)
 907                         fi->fib_net->ipv4.fib_num_tclassid_users++;
 908 #endif
 909 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 910                 nh->nh_weight = 1;
 911 #endif
 912         }
 913
 914         if (fib_props[cfg->fc_type].error) {
 915                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
 916                         goto err_inval;
 917                 goto link_it;
 918         } else {
 919                 switch (cfg->fc_type) {
 920                 case RTN_UNICAST:
 921                 case RTN_LOCAL:
 922                 case RTN_BROADCAST:
 923                 case RTN_ANYCAST:
 924                 case RTN_MULTICAST:
 925                         break;
 926                 default:
 927                         goto err_inval;
 928                 }
 929         }
 930
 931         if (cfg->fc_scope > RT_SCOPE_HOST)
 932                 goto err_inval;
 933
 934         if (cfg->fc_scope == RT_SCOPE_HOST) {
 935                 struct fib_nh *nh = fi->fib_nh;
 936
 937                 /* Local address is added. */
 938                 if (nhs != 1 || nh->nh_gw)
 939                         goto err_inval;
 940                 nh->nh_scope = RT_SCOPE_NOWHERE;
 941                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
 942                 err = -ENODEV;
 943                 if (nh->nh_dev == NULL)
 944                         goto failure;
 945         } else {
 946                 change_nexthops(fi) {
 947                         err = fib_check_nh(cfg, fi, nexthop_nh);
 948                         if (err != 0)
 949                                 goto failure;
 950                 } endfor_nexthops(fi)
 951         }
 952
 953         if (fi->fib_prefsrc) {
 954                 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 955                     fi->fib_prefsrc != cfg->fc_dst)
 956                         if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
 957                                 goto err_inval;
 958         }
 959
 960         change_nexthops(fi) {
 961                 fib_info_update_nh_saddr(net, nexthop_nh);
 962         } endfor_nexthops(fi)
 963
 964 link_it:
 965         ofi = fib_find_info(fi);
 966         if (ofi) {
 967                 fi->fib_dead = 1;
 968                 free_fib_info(fi);
 969                 ofi->fib_treeref++;
 970                 return ofi;
 971         }
 972
 973         fi->fib_treeref++;
 974         atomic_inc(&fi->fib_clntref);
 975         spin_lock_bh(&fib_info_lock);
 976         hlist_add_head(&fi->fib_hash,
 977                        &fib_info_hash[fib_info_hashfn(fi)]);
 978         if (fi->fib_prefsrc) {
 979                 struct hlist_head *head;
 980
 981                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
 982                 hlist_add_head(&fi->fib_lhash, head);
 983         }
 984         change_nexthops(fi) {
 985                 struct hlist_head *head;
 986                 unsigned int hash;
 987
 988                 if (!nexthop_nh->nh_dev)
 989                         continue;
 990                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
 991                 head = &fib_info_devhash[hash];
 992                 hlist_add_head(&nexthop_nh->nh_hash, head);
 993         } endfor_nexthops(fi)
 994         spin_unlock_bh(&fib_info_lock);
 995         return fi;
 996
 997 err_inval:
 998         err = -EINVAL;
 999
1000 failure:
1001         if (fi) {
1002                 fi->fib_dead = 1;
1003                 free_fib_info(fi);
1004         }
1005
1006         return ERR_PTR(err);
1007 }
1008
1009 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1010                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1011                   struct fib_info *fi, unsigned int flags)
1012 {
1013         struct nlmsghdr *nlh;
1014         struct rtmsg *rtm;
1015
1016         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1017         if (nlh == NULL)
1018                 return -EMSGSIZE;
1019
1020         rtm = nlmsg_data(nlh);
1021         rtm->rtm_family = AF_INET;
1022         rtm->rtm_dst_len = dst_len;
1023         rtm->rtm_src_len = 0;
1024         rtm->rtm_tos = tos;
1025         if (tb_id < 256)
1026                 rtm->rtm_table = tb_id;
1027         else
1028                 rtm->rtm_table = RT_TABLE_COMPAT;
1029         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1030                 goto nla_put_failure;
1031         rtm->rtm_type = type;
1032         rtm->rtm_flags = fi->fib_flags;
1033         rtm->rtm_scope = fi->fib_scope;
1034         rtm->rtm_protocol = fi->fib_protocol;
1035
1036         if (rtm->rtm_dst_len &&
1037             nla_put_be32(skb, RTA_DST, dst))
1038                 goto nla_put_failure;
1039         if (fi->fib_priority &&
1040             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1041                 goto nla_put_failure;
1042         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1043                 goto nla_put_failure;
1044
1045         if (fi->fib_prefsrc &&
1046             nla_put_be32(skb, RTA_PREFSRC, fi->fib_prefsrc))
1047                 goto nla_put_failure;
1048         if (fi->fib_nhs == 1) {
1049                 if (fi->fib_nh->nh_gw &&
1050                     nla_put_be32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1051                         goto nla_put_failure;
1052                 if (fi->fib_nh->nh_oif &&
1053                     nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1054                         goto nla_put_failure;
1055 #ifdef CONFIG_IP_ROUTE_CLASSID
1056                 if (fi->fib_nh[0].nh_tclassid &&
1057                     nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1058                         goto nla_put_failure;
1059 #endif
1060         }
1061 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1062         if (fi->fib_nhs > 1) {
1063                 struct rtnexthop *rtnh;
1064                 struct nlattr *mp;
1065
1066                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1067                 if (mp == NULL)
1068                         goto nla_put_failure;
1069
1070                 for_nexthops(fi) {
1071                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1072                         if (rtnh == NULL)
1073                                 goto nla_put_failure;
1074
1075                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1076                         rtnh->rtnh_hops = nh->nh_weight - 1;
1077                         rtnh->rtnh_ifindex = nh->nh_oif;
1078
1079                         if (nh->nh_gw &&
1080                             nla_put_be32(skb, RTA_GATEWAY, nh->nh_gw))
1081                                 goto nla_put_failure;
1082 #ifdef CONFIG_IP_ROUTE_CLASSID
1083                         if (nh->nh_tclassid &&
1084                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1085                                 goto nla_put_failure;
1086 #endif
1087                         /* length of rtnetlink header + attributes */
1088                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1089                 } endfor_nexthops(fi);
1090
1091                 nla_nest_end(skb, mp);
1092         }
1093 #endif
1094         nlmsg_end(skb, nlh);
1095         return 0;
1096
1097 nla_put_failure:
1098         nlmsg_cancel(skb, nlh);
1099         return -EMSGSIZE;
1100 }
1101
1102 /*
1103  * Update FIB if:
1104  * - local address disappeared -> we must delete all the entries
1105  *   referring to it.
1106  * - device went down -> we must shutdown all nexthops going via it.
1107  */
1108 int fib_sync_down_addr(struct net *net, __be32 local)
1109 {
1110         int ret = 0;
1111         unsigned int hash = fib_laddr_hashfn(local);
1112         struct hlist_head *head = &fib_info_laddrhash[hash];
1113         struct fib_info *fi;
1114
1115         if (fib_info_laddrhash == NULL || local == 0)
1116                 return 0;
1117
1118         hlist_for_each_entry(fi, head, fib_lhash) {
1119                 if (!net_eq(fi->fib_net, net))
1120                         continue;
1121                 if (fi->fib_prefsrc == local) {
1122                         fi->fib_flags |= RTNH_F_DEAD;
1123                         ret++;
1124                 }
1125         }
1126         return ret;
1127 }
1128
1129 int fib_sync_down_dev(struct net_device *dev, int force)
1130 {
1131         int ret = 0;
1132         int scope = RT_SCOPE_NOWHERE;
1133         struct fib_info *prev_fi = NULL;
1134         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1135         struct hlist_head *head = &fib_info_devhash[hash];
1136         struct fib_nh *nh;
1137
1138         if (force)
1139                 scope = -1;
1140
1141         hlist_for_each_entry(nh, head, nh_hash) {
1142                 struct fib_info *fi = nh->nh_parent;
1143                 int dead;
1144
1145                 BUG_ON(!fi->fib_nhs);
1146                 if (nh->nh_dev != dev || fi == prev_fi)
1147                         continue;
1148                 prev_fi = fi;
1149                 dead = 0;
1150                 change_nexthops(fi) {
1151                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1152                                 dead++;
1153                         else if (nexthop_nh->nh_dev == dev &&
1154                                  nexthop_nh->nh_scope != scope) {
1155                                 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1156 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1157                                 spin_lock_bh(&fib_multipath_lock);
1158                                 fi->fib_power -= nexthop_nh->nh_power;
1159                                 nexthop_nh->nh_power = 0;
1160                                 spin_unlock_bh(&fib_multipath_lock);
1161 #endif
1162                                 dead++;
1163                         }
1164 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1165                         if (force > 1 && nexthop_nh->nh_dev == dev) {
1166                                 dead = fi->fib_nhs;
1167                                 break;
1168                         }
1169 #endif
1170                 } endfor_nexthops(fi)
1171                 if (dead == fi->fib_nhs) {
1172                         fi->fib_flags |= RTNH_F_DEAD;
1173                         ret++;
1174                 }
1175         }
1176
1177         return ret;
1178 }
1179
1180 /* Must be invoked inside of an RCU protected region.  */
1181 void fib_select_default(struct fib_result *res)
1182 {
1183         struct fib_info *fi = NULL, *last_resort = NULL;
1184         struct list_head *fa_head = res->fa_head;
1185         struct fib_table *tb = res->table;
1186         int order = -1, last_idx = -1;
1187         struct fib_alias *fa;
1188
1189         list_for_each_entry_rcu(fa, fa_head, fa_list) {
1190                 struct fib_info *next_fi = fa->fa_info;
1191
1192                 if (next_fi->fib_scope != res->scope ||
1193                     fa->fa_type != RTN_UNICAST)
1194                         continue;
1195
1196                 if (next_fi->fib_priority > res->fi->fib_priority)
1197                         break;
1198                 if (!next_fi->fib_nh[0].nh_gw ||
1199                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1200                         continue;
1201
1202                 fib_alias_accessed(fa);
1203
1204                 if (fi == NULL) {
1205                         if (next_fi != res->fi)
1206                                 break;
1207                 } else if (!fib_detect_death(fi, order, &last_resort,
1208                                              &last_idx, tb->tb_default)) {
1209                         fib_result_assign(res, fi);
1210                         tb->tb_default = order;
1211                         goto out;
1212                 }
1213                 fi = next_fi;
1214                 order++;
1215         }
1216
1217         if (order <= 0 || fi == NULL) {
1218                 tb->tb_default = -1;
1219                 goto out;
1220         }
1221
1222         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1223                                 tb->tb_default)) {
1224                 fib_result_assign(res, fi);
1225                 tb->tb_default = order;
1226                 goto out;
1227         }
1228
1229         if (last_idx >= 0)
1230                 fib_result_assign(res, last_resort);
1231         tb->tb_default = last_idx;
1232 out:
1233         return;
1234 }
1235
1236 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1237
1238 /*
1239  * Dead device goes up. We wake up dead nexthops.
1240  * It takes sense only on multipath routes.
1241  */
1242 int fib_sync_up(struct net_device *dev)
1243 {
1244         struct fib_info *prev_fi;
1245         unsigned int hash;
1246         struct hlist_head *head;
1247         struct fib_nh *nh;
1248         int ret;
1249
1250         if (!(dev->flags & IFF_UP))
1251                 return 0;
1252
1253         prev_fi = NULL;
1254         hash = fib_devindex_hashfn(dev->ifindex);
1255         head = &fib_info_devhash[hash];
1256         ret = 0;
1257
1258         hlist_for_each_entry(nh, head, nh_hash) {
1259                 struct fib_info *fi = nh->nh_parent;
1260                 int alive;
1261
1262                 BUG_ON(!fi->fib_nhs);
1263                 if (nh->nh_dev != dev || fi == prev_fi)
1264                         continue;
1265
1266                 prev_fi = fi;
1267                 alive = 0;
1268                 change_nexthops(fi) {
1269                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1270                                 alive++;
1271                                 continue;
1272                         }
1273                         if (nexthop_nh->nh_dev == NULL ||
1274                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1275                                 continue;
1276                         if (nexthop_nh->nh_dev != dev ||
1277                             !__in_dev_get_rtnl(dev))
1278                                 continue;
1279                         alive++;
1280                         spin_lock_bh(&fib_multipath_lock);
1281                         nexthop_nh->nh_power = 0;
1282                         nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1283                         spin_unlock_bh(&fib_multipath_lock);
1284                 } endfor_nexthops(fi)
1285
1286                 if (alive > 0) {
1287                         fi->fib_flags &= ~RTNH_F_DEAD;
1288                         ret++;
1289                 }
1290         }
1291
1292         return ret;
1293 }
1294
1295 /*
1296  * The algorithm is suboptimal, but it provides really
1297  * fair weighted route distribution.
1298  */
1299 void fib_select_multipath(struct fib_result *res)
1300 {
1301         struct fib_info *fi = res->fi;
1302         int w;
1303
1304         spin_lock_bh(&fib_multipath_lock);
1305         if (fi->fib_power <= 0) {
1306                 int power = 0;
1307                 change_nexthops(fi) {
1308                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1309                                 power += nexthop_nh->nh_weight;
1310                                 nexthop_nh->nh_power = nexthop_nh->nh_weight;
1311                         }
1312                 } endfor_nexthops(fi);
1313                 fi->fib_power = power;
1314                 if (power <= 0) {
1315                         spin_unlock_bh(&fib_multipath_lock);
1316                         /* Race condition: route has just become dead. */
1317                         res->nh_sel = 0;
1318                         return;
1319                 }
1320         }
1321
1322
1323         /* w should be random number [0..fi->fib_power-1],
1324          * it is pretty bad approximation.
1325          */
1326
1327         w = jiffies % fi->fib_power;
1328
1329         change_nexthops(fi) {
1330                 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1331                     nexthop_nh->nh_power) {
1332                         w -= nexthop_nh->nh_power;
1333                         if (w <= 0) {
1334                                 nexthop_nh->nh_power--;
1335                                 fi->fib_power--;
1336                                 res->nh_sel = nhsel;
1337                                 spin_unlock_bh(&fib_multipath_lock);
1338                                 return;
1339                         }
1340                 }
1341         } endfor_nexthops(fi);
1342
1343         /* Race condition: route has just become dead. */
1344         res->nh_sel = 0;
1345         spin_unlock_bh(&fib_multipath_lock);
1346 }
1347 #endif