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, 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 int fib_detect_death(struct fib_info *fi, int order,
 430                      struct fib_info **last_resort, int *last_idx, int dflt)
 431 {
 432         struct neighbour *n;
 433         int state = NUD_NONE;
 434
 435         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 436         if (n) {
 437                 state = n->nud_state;
 438                 neigh_release(n);
 439         }
 440         if (state == NUD_REACHABLE)
 441                 return 0;
 442         if ((state & NUD_VALID) && order != dflt)
 443                 return 0;
 444         if ((state & NUD_VALID) ||
 445             (*last_idx < 0 && order > dflt)) {
 446                 *last_resort = fi;
 447                 *last_idx = order;
 448         }
 449         return 1;
 450 }
 451
 452 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 453
 454 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
 455 {
 456         int nhs = 0;
 457
 458         while (rtnh_ok(rtnh, remaining)) {
 459                 nhs++;
 460                 rtnh = rtnh_next(rtnh, &remaining);
 461         }
 462
 463         /* leftover implies invalid nexthop configuration, discard it */
 464         return remaining > 0 ? 0 : nhs;
 465 }
 466
 467 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 468                        int remaining, struct fib_config *cfg)
 469 {
 470         change_nexthops(fi) {
 471                 int attrlen;
 472
 473                 if (!rtnh_ok(rtnh, remaining))
 474                         return -EINVAL;
 475
 476                 nexthop_nh->nh_flags =
 477                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 478                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
 479                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
 480
 481                 attrlen = rtnh_attrlen(rtnh);
 482                 if (attrlen > 0) {
 483                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 484
 485                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 486                         nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
 487 #ifdef CONFIG_IP_ROUTE_CLASSID
 488                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 489                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 490                         if (nexthop_nh->nh_tclassid)
 491                                 fi->fib_net->ipv4.fib_num_tclassid_users++;
 492 #endif
 493                 }
 494
 495                 rtnh = rtnh_next(rtnh, &remaining);
 496         } endfor_nexthops(fi);
 497
 498         return 0;
 499 }
 500
 501 #endif
 502
 503 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
 504 {
 505 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 506         struct rtnexthop *rtnh;
 507         int remaining;
 508 #endif
 509
 510         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 511                 return 1;
 512
 513         if (cfg->fc_oif || cfg->fc_gw) {
 514                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 515                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 516                         return 0;
 517                 return 1;
 518         }
 519
 520 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 521         if (cfg->fc_mp == NULL)
 522                 return 0;
 523
 524         rtnh = cfg->fc_mp;
 525         remaining = cfg->fc_mp_len;
 526
 527         for_nexthops(fi) {
 528                 int attrlen;
 529
 530                 if (!rtnh_ok(rtnh, remaining))
 531                         return -EINVAL;
 532
 533                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 534                         return 1;
 535
 536                 attrlen = rtnh_attrlen(rtnh);
 537                 if (attrlen < 0) {
 538                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 539
 540                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 541                         if (nla && nla_get_be32(nla) != nh->nh_gw)
 542                                 return 1;
 543 #ifdef CONFIG_IP_ROUTE_CLASSID
 544                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 545                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 546                                 return 1;
 547 #endif
 548                 }
 549
 550                 rtnh = rtnh_next(rtnh, &remaining);
 551         } endfor_nexthops(fi);
 552 #endif
 553         return 0;
 554 }
 555
 556
 557 /*
 558  * Picture
 559  * -------
 560  *
 561  * Semantics of nexthop is very messy by historical reasons.
 562  * We have to take into account, that:
 563  * a) gateway can be actually local interface address,
 564  *    so that gatewayed route is direct.
 565  * b) gateway must be on-link address, possibly
 566  *    described not by an ifaddr, but also by a direct route.
 567  * c) If both gateway and interface are specified, they should not
 568  *    contradict.
 569  * d) If we use tunnel routes, gateway could be not on-link.
 570  *
 571  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 572  * results in pretty ugly and hairy code with obscure logic.
 573  *
 574  * I chose to generalized it instead, so that the size
 575  * of code does not increase practically, but it becomes
 576  * much more general.
 577  * Every prefix is assigned a "scope" value: "host" is local address,
 578  * "link" is direct route,
 579  * [ ... "site" ... "interior" ... ]
 580  * and "universe" is true gateway route with global meaning.
 581  *
 582  * Every prefix refers to a set of "nexthop"s (gw, oif),
 583  * where gw must have narrower scope. This recursion stops
 584  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 585  * which means that gw is forced to be on link.
 586  *
 587  * Code is still hairy, but now it is apparently logically
 588  * consistent and very flexible. F.e. as by-product it allows
 589  * to co-exists in peace independent exterior and interior
 590  * routing processes.
 591  *
 592  * Normally it looks as following.
 593  *
 594  * {universe prefix}  -> (gw, oif) [scope link]
 595  *                |
 596  *                |-> {link prefix} -> (gw, oif) [scope local]
 597  *                                      |
 598  *                                      |-> {local prefix} (terminal node)
 599  */
 600 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
 601                         struct fib_nh *nh)
 602 {
 603         int err;
 604         struct net *net;
 605         struct net_device *dev;
 606
 607         net = cfg->fc_nlinfo.nl_net;
 608         if (nh->nh_gw) {
 609                 struct fib_result res;
 610
 611                 if (nh->nh_flags & RTNH_F_ONLINK) {
 612
 613                         if (cfg->fc_scope >= RT_SCOPE_LINK)
 614                                 return -EINVAL;
 615                         if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
 616                                 return -EINVAL;
 617                         dev = __dev_get_by_index(net, nh->nh_oif);
 618                         if (!dev)
 619                                 return -ENODEV;
 620                         if (!(dev->flags & IFF_UP))
 621                                 return -ENETDOWN;
 622                         nh->nh_dev = dev;
 623                         dev_hold(dev);
 624                         nh->nh_scope = RT_SCOPE_LINK;
 625                         return 0;
 626                 }
 627                 rcu_read_lock();
 628                 {
 629                         struct flowi4 fl4 = {
 630                                 .daddr = nh->nh_gw,
 631                                 .flowi4_scope = cfg->fc_scope + 1,
 632                                 .flowi4_oif = nh->nh_oif,
 633                         };
 634
 635                         /* It is not necessary, but requires a bit of thinking */
 636                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
 637                                 fl4.flowi4_scope = RT_SCOPE_LINK;
 638                         err = fib_lookup(net, &fl4, &res);
 639                         if (err) {
 640                                 rcu_read_unlock();
 641                                 return err;
 642                         }
 643                 }
 644                 err = -EINVAL;
 645                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 646                         goto out;
 647                 nh->nh_scope = res.scope;
 648                 nh->nh_oif = FIB_RES_OIF(res);
 649                 nh->nh_dev = dev = FIB_RES_DEV(res);
 650                 if (!dev)
 651                         goto out;
 652                 dev_hold(dev);
 653                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 654         } else {
 655                 struct in_device *in_dev;
 656
 657                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 658                         return -EINVAL;
 659
 660                 rcu_read_lock();
 661                 err = -ENODEV;
 662                 in_dev = inetdev_by_index(net, nh->nh_oif);
 663                 if (in_dev == NULL)
 664                         goto out;
 665                 err = -ENETDOWN;
 666                 if (!(in_dev->dev->flags & IFF_UP))
 667                         goto out;
 668                 nh->nh_dev = in_dev->dev;
 669                 dev_hold(nh->nh_dev);
 670                 nh->nh_scope = RT_SCOPE_HOST;
 671                 err = 0;
 672         }
 673 out:
 674         rcu_read_unlock();
 675         return err;
 676 }
 677
 678 static inline unsigned int fib_laddr_hashfn(__be32 val)
 679 {
 680         unsigned int mask = (fib_info_hash_size - 1);
 681
 682         return ((__force u32)val ^
 683                 ((__force u32)val >> 7) ^
 684                 ((__force u32)val >> 14)) & mask;
 685 }
 686
 687 static struct hlist_head *fib_info_hash_alloc(int bytes)
 688 {
 689         if (bytes <= PAGE_SIZE)
 690                 return kzalloc(bytes, GFP_KERNEL);
 691         else
 692                 return (struct hlist_head *)
 693                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 694                                          get_order(bytes));
 695 }
 696
 697 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 698 {
 699         if (!hash)
 700                 return;
 701
 702         if (bytes <= PAGE_SIZE)
 703                 kfree(hash);
 704         else
 705                 free_pages((unsigned long) hash, get_order(bytes));
 706 }
 707
 708 static void fib_info_hash_move(struct hlist_head *new_info_hash,
 709                                struct hlist_head *new_laddrhash,
 710                                unsigned int new_size)
 711 {
 712         struct hlist_head *old_info_hash, *old_laddrhash;
 713         unsigned int old_size = fib_info_hash_size;
 714         unsigned int i, bytes;
 715
 716         spin_lock_bh(&fib_info_lock);
 717         old_info_hash = fib_info_hash;
 718         old_laddrhash = fib_info_laddrhash;
 719         fib_info_hash_size = new_size;
 720
 721         for (i = 0; i < old_size; i++) {
 722                 struct hlist_head *head = &fib_info_hash[i];
 723                 struct hlist_node *n;
 724                 struct fib_info *fi;
 725
 726                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
 727                         struct hlist_head *dest;
 728                         unsigned int new_hash;
 729
 730                         hlist_del(&fi->fib_hash);
 731
 732                         new_hash = fib_info_hashfn(fi);
 733                         dest = &new_info_hash[new_hash];
 734                         hlist_add_head(&fi->fib_hash, dest);
 735                 }
 736         }
 737         fib_info_hash = new_info_hash;
 738
 739         for (i = 0; i < old_size; i++) {
 740                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 741                 struct hlist_node *n;
 742                 struct fib_info *fi;
 743
 744                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
 745                         struct hlist_head *ldest;
 746                         unsigned int new_hash;
 747
 748                         hlist_del(&fi->fib_lhash);
 749
 750                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 751                         ldest = &new_laddrhash[new_hash];
 752                         hlist_add_head(&fi->fib_lhash, ldest);
 753                 }
 754         }
 755         fib_info_laddrhash = new_laddrhash;
 756
 757         spin_unlock_bh(&fib_info_lock);
 758
 759         bytes = old_size * sizeof(struct hlist_head *);
 760         fib_info_hash_free(old_info_hash, bytes);
 761         fib_info_hash_free(old_laddrhash, bytes);
 762 }
 763
 764 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 765 {
 766         nh->nh_saddr = inet_select_addr(nh->nh_dev,
 767                                         nh->nh_gw,
 768                                         nh->nh_parent->fib_scope);
 769         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 770
 771         return nh->nh_saddr;
 772 }
 773
 774 struct fib_info *fib_create_info(struct fib_config *cfg)
 775 {
 776         int err;
 777         struct fib_info *fi = NULL;
 778         struct fib_info *ofi;
 779         int nhs = 1;
 780         struct net *net = cfg->fc_nlinfo.nl_net;
 781
 782         if (cfg->fc_type > RTN_MAX)
 783                 goto err_inval;
 784
 785         /* Fast check to catch the most weird cases */
 786         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
 787                 goto err_inval;
 788
 789 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 790         if (cfg->fc_mp) {
 791                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
 792                 if (nhs == 0)
 793                         goto err_inval;
 794         }
 795 #endif
 796
 797         err = -ENOBUFS;
 798         if (fib_info_cnt >= fib_info_hash_size) {
 799                 unsigned int new_size = fib_info_hash_size << 1;
 800                 struct hlist_head *new_info_hash;
 801                 struct hlist_head *new_laddrhash;
 802                 unsigned int bytes;
 803
 804                 if (!new_size)
 805                         new_size = 16;
 806                 bytes = new_size * sizeof(struct hlist_head *);
 807                 new_info_hash = fib_info_hash_alloc(bytes);
 808                 new_laddrhash = fib_info_hash_alloc(bytes);
 809                 if (!new_info_hash || !new_laddrhash) {
 810                         fib_info_hash_free(new_info_hash, bytes);
 811                         fib_info_hash_free(new_laddrhash, bytes);
 812                 } else
 813                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
 814
 815                 if (!fib_info_hash_size)
 816                         goto failure;
 817         }
 818
 819         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
 820         if (fi == NULL)
 821                 goto failure;
 822         if (cfg->fc_mx) {
 823                 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
 824                 if (!fi->fib_metrics)
 825                         goto failure;
 826         } else
 827                 fi->fib_metrics = (u32 *) dst_default_metrics;
 828         fib_info_cnt++;
 829
 830         fi->fib_net = hold_net(net);
 831         fi->fib_protocol = cfg->fc_protocol;
 832         fi->fib_scope = cfg->fc_scope;
 833         fi->fib_flags = cfg->fc_flags;
 834         fi->fib_priority = cfg->fc_priority;
 835         fi->fib_prefsrc = cfg->fc_prefsrc;
 836         fi->fib_type = cfg->fc_type;
 837
 838         fi->fib_nhs = nhs;
 839         change_nexthops(fi) {
 840                 nexthop_nh->nh_parent = fi;
 841                 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
 842                 if (!nexthop_nh->nh_pcpu_rth_output)
 843                         goto failure;
 844         } endfor_nexthops(fi)
 845
 846         if (cfg->fc_mx) {
 847                 struct nlattr *nla;
 848                 int remaining;
 849
 850                 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 851                         int type = nla_type(nla);
 852
 853                         if (type) {
 854                                 u32 val;
 855
 856                                 if (type > RTAX_MAX)
 857                                         goto err_inval;
 858                                 val = nla_get_u32(nla);
 859                                 if (type == RTAX_ADVMSS && val > 65535 - 40)
 860                                         val = 65535 - 40;
 861                                 if (type == RTAX_MTU && val > 65535 - 15)
 862                                         val = 65535 - 15;
 863                                 fi->fib_metrics[type - 1] = val;
 864                         }
 865                 }
 866         }
 867
 868         if (cfg->fc_mp) {
 869 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 870                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
 871                 if (err != 0)
 872                         goto failure;
 873                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
 874                         goto err_inval;
 875                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
 876                         goto err_inval;
 877 #ifdef CONFIG_IP_ROUTE_CLASSID
 878                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
 879                         goto err_inval;
 880 #endif
 881 #else
 882                 goto err_inval;
 883 #endif
 884         } else {
 885                 struct fib_nh *nh = fi->fib_nh;
 886
 887                 nh->nh_oif = cfg->fc_oif;
 888                 nh->nh_gw = cfg->fc_gw;
 889                 nh->nh_flags = cfg->fc_flags;
 890 #ifdef CONFIG_IP_ROUTE_CLASSID
 891                 nh->nh_tclassid = cfg->fc_flow;
 892                 if (nh->nh_tclassid)
 893                         fi->fib_net->ipv4.fib_num_tclassid_users++;
 894 #endif
 895 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 896                 nh->nh_weight = 1;
 897 #endif
 898         }
 899
 900         if (fib_props[cfg->fc_type].error) {
 901                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
 902                         goto err_inval;
 903                 goto link_it;
 904         } else {
 905                 switch (cfg->fc_type) {
 906                 case RTN_UNICAST:
 907                 case RTN_LOCAL:
 908                 case RTN_BROADCAST:
 909                 case RTN_ANYCAST:
 910                 case RTN_MULTICAST:
 911                         break;
 912                 default:
 913                         goto err_inval;
 914                 }
 915         }
 916
 917         if (cfg->fc_scope > RT_SCOPE_HOST)
 918                 goto err_inval;
 919
 920         if (cfg->fc_scope == RT_SCOPE_HOST) {
 921                 struct fib_nh *nh = fi->fib_nh;
 922
 923                 /* Local address is added. */
 924                 if (nhs != 1 || nh->nh_gw)
 925                         goto err_inval;
 926                 nh->nh_scope = RT_SCOPE_NOWHERE;
 927                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
 928                 err = -ENODEV;
 929                 if (nh->nh_dev == NULL)
 930                         goto failure;
 931         } else {
 932                 change_nexthops(fi) {
 933                         err = fib_check_nh(cfg, fi, nexthop_nh);
 934                         if (err != 0)
 935                                 goto failure;
 936                 } endfor_nexthops(fi)
 937         }
 938
 939         if (fi->fib_prefsrc) {
 940                 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 941                     fi->fib_prefsrc != cfg->fc_dst)
 942                         if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
 943                                 goto err_inval;
 944         }
 945
 946         change_nexthops(fi) {
 947                 fib_info_update_nh_saddr(net, nexthop_nh);
 948         } endfor_nexthops(fi)
 949
 950 link_it:
 951         ofi = fib_find_info(fi);
 952         if (ofi) {
 953                 fi->fib_dead = 1;
 954                 free_fib_info(fi);
 955                 ofi->fib_treeref++;
 956                 return ofi;
 957         }
 958
 959         fi->fib_treeref++;
 960         atomic_inc(&fi->fib_clntref);
 961         spin_lock_bh(&fib_info_lock);
 962         hlist_add_head(&fi->fib_hash,
 963                        &fib_info_hash[fib_info_hashfn(fi)]);
 964         if (fi->fib_prefsrc) {
 965                 struct hlist_head *head;
 966
 967                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
 968                 hlist_add_head(&fi->fib_lhash, head);
 969         }
 970         change_nexthops(fi) {
 971                 struct hlist_head *head;
 972                 unsigned int hash;
 973
 974                 if (!nexthop_nh->nh_dev)
 975                         continue;
 976                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
 977                 head = &fib_info_devhash[hash];
 978                 hlist_add_head(&nexthop_nh->nh_hash, head);
 979         } endfor_nexthops(fi)
 980         spin_unlock_bh(&fib_info_lock);
 981         return fi;
 982
 983 err_inval:
 984         err = -EINVAL;
 985
 986 failure:
 987         if (fi) {
 988                 fi->fib_dead = 1;
 989                 free_fib_info(fi);
 990         }
 991
 992         return ERR_PTR(err);
 993 }
 994
 995 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
 996                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
 997                   struct fib_info *fi, unsigned int flags)
 998 {
 999         struct nlmsghdr *nlh;
1000         struct rtmsg *rtm;
1001
1002         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1003         if (nlh == NULL)
1004                 return -EMSGSIZE;
1005
1006         rtm = nlmsg_data(nlh);
1007         rtm->rtm_family = AF_INET;
1008         rtm->rtm_dst_len = dst_len;
1009         rtm->rtm_src_len = 0;
1010         rtm->rtm_tos = tos;
1011         if (tb_id < 256)
1012                 rtm->rtm_table = tb_id;
1013         else
1014                 rtm->rtm_table = RT_TABLE_COMPAT;
1015         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1016                 goto nla_put_failure;
1017         rtm->rtm_type = type;
1018         rtm->rtm_flags = fi->fib_flags;
1019         rtm->rtm_scope = fi->fib_scope;
1020         rtm->rtm_protocol = fi->fib_protocol;
1021
1022         if (rtm->rtm_dst_len &&
1023             nla_put_be32(skb, RTA_DST, dst))
1024                 goto nla_put_failure;
1025         if (fi->fib_priority &&
1026             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1027                 goto nla_put_failure;
1028         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1029                 goto nla_put_failure;
1030
1031         if (fi->fib_prefsrc &&
1032             nla_put_be32(skb, RTA_PREFSRC, fi->fib_prefsrc))
1033                 goto nla_put_failure;
1034         if (fi->fib_nhs == 1) {
1035                 if (fi->fib_nh->nh_gw &&
1036                     nla_put_be32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1037                         goto nla_put_failure;
1038                 if (fi->fib_nh->nh_oif &&
1039                     nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1040                         goto nla_put_failure;
1041 #ifdef CONFIG_IP_ROUTE_CLASSID
1042                 if (fi->fib_nh[0].nh_tclassid &&
1043                     nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1044                         goto nla_put_failure;
1045 #endif
1046         }
1047 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1048         if (fi->fib_nhs > 1) {
1049                 struct rtnexthop *rtnh;
1050                 struct nlattr *mp;
1051
1052                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1053                 if (mp == NULL)
1054                         goto nla_put_failure;
1055
1056                 for_nexthops(fi) {
1057                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1058                         if (rtnh == NULL)
1059                                 goto nla_put_failure;
1060
1061                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1062                         rtnh->rtnh_hops = nh->nh_weight - 1;
1063                         rtnh->rtnh_ifindex = nh->nh_oif;
1064
1065                         if (nh->nh_gw &&
1066                             nla_put_be32(skb, RTA_GATEWAY, nh->nh_gw))
1067                                 goto nla_put_failure;
1068 #ifdef CONFIG_IP_ROUTE_CLASSID
1069                         if (nh->nh_tclassid &&
1070                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1071                                 goto nla_put_failure;
1072 #endif
1073                         /* length of rtnetlink header + attributes */
1074                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1075                 } endfor_nexthops(fi);
1076
1077                 nla_nest_end(skb, mp);
1078         }
1079 #endif
1080         return nlmsg_end(skb, nlh);
1081
1082 nla_put_failure:
1083         nlmsg_cancel(skb, nlh);
1084         return -EMSGSIZE;
1085 }
1086
1087 /*
1088  * Update FIB if:
1089  * - local address disappeared -> we must delete all the entries
1090  *   referring to it.
1091  * - device went down -> we must shutdown all nexthops going via it.
1092  */
1093 int fib_sync_down_addr(struct net *net, __be32 local)
1094 {
1095         int ret = 0;
1096         unsigned int hash = fib_laddr_hashfn(local);
1097         struct hlist_head *head = &fib_info_laddrhash[hash];
1098         struct fib_info *fi;
1099
1100         if (fib_info_laddrhash == NULL || local == 0)
1101                 return 0;
1102
1103         hlist_for_each_entry(fi, head, fib_lhash) {
1104                 if (!net_eq(fi->fib_net, net))
1105                         continue;
1106                 if (fi->fib_prefsrc == local) {
1107                         fi->fib_flags |= RTNH_F_DEAD;
1108                         ret++;
1109                 }
1110         }
1111         return ret;
1112 }
1113
1114 int fib_sync_down_dev(struct net_device *dev, int force)
1115 {
1116         int ret = 0;
1117         int scope = RT_SCOPE_NOWHERE;
1118         struct fib_info *prev_fi = NULL;
1119         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1120         struct hlist_head *head = &fib_info_devhash[hash];
1121         struct fib_nh *nh;
1122
1123         if (force)
1124                 scope = -1;
1125
1126         hlist_for_each_entry(nh, head, nh_hash) {
1127                 struct fib_info *fi = nh->nh_parent;
1128                 int dead;
1129
1130                 BUG_ON(!fi->fib_nhs);
1131                 if (nh->nh_dev != dev || fi == prev_fi)
1132                         continue;
1133                 prev_fi = fi;
1134                 dead = 0;
1135                 change_nexthops(fi) {
1136                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1137                                 dead++;
1138                         else if (nexthop_nh->nh_dev == dev &&
1139                                  nexthop_nh->nh_scope != scope) {
1140                                 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1141 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1142                                 spin_lock_bh(&fib_multipath_lock);
1143                                 fi->fib_power -= nexthop_nh->nh_power;
1144                                 nexthop_nh->nh_power = 0;
1145                                 spin_unlock_bh(&fib_multipath_lock);
1146 #endif
1147                                 dead++;
1148                         }
1149 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1150                         if (force > 1 && nexthop_nh->nh_dev == dev) {
1151                                 dead = fi->fib_nhs;
1152                                 break;
1153                         }
1154 #endif
1155                 } endfor_nexthops(fi)
1156                 if (dead == fi->fib_nhs) {
1157                         fi->fib_flags |= RTNH_F_DEAD;
1158                         ret++;
1159                 }
1160         }
1161
1162         return ret;
1163 }
1164
1165 /* Must be invoked inside of an RCU protected region.  */
1166 void fib_select_default(struct fib_result *res)
1167 {
1168         struct fib_info *fi = NULL, *last_resort = NULL;
1169         struct list_head *fa_head = res->fa_head;
1170         struct fib_table *tb = res->table;
1171         int order = -1, last_idx = -1;
1172         struct fib_alias *fa;
1173
1174         list_for_each_entry_rcu(fa, fa_head, fa_list) {
1175                 struct fib_info *next_fi = fa->fa_info;
1176
1177                 if (next_fi->fib_scope != res->scope ||
1178                     fa->fa_type != RTN_UNICAST)
1179                         continue;
1180
1181                 if (next_fi->fib_priority > res->fi->fib_priority)
1182                         break;
1183                 if (!next_fi->fib_nh[0].nh_gw ||
1184                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1185                         continue;
1186
1187                 fib_alias_accessed(fa);
1188
1189                 if (fi == NULL) {
1190                         if (next_fi != res->fi)
1191                                 break;
1192                 } else if (!fib_detect_death(fi, order, &last_resort,
1193                                              &last_idx, tb->tb_default)) {
1194                         fib_result_assign(res, fi);
1195                         tb->tb_default = order;
1196                         goto out;
1197                 }
1198                 fi = next_fi;
1199                 order++;
1200         }
1201
1202         if (order <= 0 || fi == NULL) {
1203                 tb->tb_default = -1;
1204                 goto out;
1205         }
1206
1207         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1208                                 tb->tb_default)) {
1209                 fib_result_assign(res, fi);
1210                 tb->tb_default = order;
1211                 goto out;
1212         }
1213
1214         if (last_idx >= 0)
1215                 fib_result_assign(res, last_resort);
1216         tb->tb_default = last_idx;
1217 out:
1218         return;
1219 }
1220
1221 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1222
1223 /*
1224  * Dead device goes up. We wake up dead nexthops.
1225  * It takes sense only on multipath routes.
1226  */
1227 int fib_sync_up(struct net_device *dev)
1228 {
1229         struct fib_info *prev_fi;
1230         unsigned int hash;
1231         struct hlist_head *head;
1232         struct fib_nh *nh;
1233         int ret;
1234
1235         if (!(dev->flags & IFF_UP))
1236                 return 0;
1237
1238         prev_fi = NULL;
1239         hash = fib_devindex_hashfn(dev->ifindex);
1240         head = &fib_info_devhash[hash];
1241         ret = 0;
1242
1243         hlist_for_each_entry(nh, head, nh_hash) {
1244                 struct fib_info *fi = nh->nh_parent;
1245                 int alive;
1246
1247                 BUG_ON(!fi->fib_nhs);
1248                 if (nh->nh_dev != dev || fi == prev_fi)
1249                         continue;
1250
1251                 prev_fi = fi;
1252                 alive = 0;
1253                 change_nexthops(fi) {
1254                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1255                                 alive++;
1256                                 continue;
1257                         }
1258                         if (nexthop_nh->nh_dev == NULL ||
1259                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1260                                 continue;
1261                         if (nexthop_nh->nh_dev != dev ||
1262                             !__in_dev_get_rtnl(dev))
1263                                 continue;
1264                         alive++;
1265                         spin_lock_bh(&fib_multipath_lock);
1266                         nexthop_nh->nh_power = 0;
1267                         nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1268                         spin_unlock_bh(&fib_multipath_lock);
1269                 } endfor_nexthops(fi)
1270
1271                 if (alive > 0) {
1272                         fi->fib_flags &= ~RTNH_F_DEAD;
1273                         ret++;
1274                 }
1275         }
1276
1277         return ret;
1278 }
1279
1280 /*
1281  * The algorithm is suboptimal, but it provides really
1282  * fair weighted route distribution.
1283  */
1284 void fib_select_multipath(struct fib_result *res)
1285 {
1286         struct fib_info *fi = res->fi;
1287         int w;
1288
1289         spin_lock_bh(&fib_multipath_lock);
1290         if (fi->fib_power <= 0) {
1291                 int power = 0;
1292                 change_nexthops(fi) {
1293                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1294                                 power += nexthop_nh->nh_weight;
1295                                 nexthop_nh->nh_power = nexthop_nh->nh_weight;
1296                         }
1297                 } endfor_nexthops(fi);
1298                 fi->fib_power = power;
1299                 if (power <= 0) {
1300                         spin_unlock_bh(&fib_multipath_lock);
1301                         /* Race condition: route has just become dead. */
1302                         res->nh_sel = 0;
1303                         return;
1304                 }
1305         }
1306
1307
1308         /* w should be random number [0..fi->fib_power-1],
1309          * it is pretty bad approximation.
1310          */
1311
1312         w = jiffies % fi->fib_power;
1313
1314         change_nexthops(fi) {
1315                 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1316                     nexthop_nh->nh_power) {
1317                         w -= nexthop_nh->nh_power;
1318                         if (w <= 0) {
1319                                 nexthop_nh->nh_power--;
1320                                 fi->fib_power--;
1321                                 res->nh_sel = nhsel;
1322                                 spin_unlock_bh(&fib_multipath_lock);
1323                                 return;
1324                         }
1325                 }
1326         } endfor_nexthops(fi);
1327
1328         /* Race condition: route has just become dead. */
1329         res->nh_sel = 0;
1330         spin_unlock_bh(&fib_multipath_lock);
1331 }
1332 #endif