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 #include <net/lwtunnel.h>
  46
  47 #include "fib_lookup.h"
  48
  49 static DEFINE_SPINLOCK(fib_info_lock);
  50 static struct hlist_head *fib_info_hash;
  51 static struct hlist_head *fib_info_laddrhash;
  52 static unsigned int fib_info_hash_size;
  53 static unsigned int fib_info_cnt;
  54
  55 #define DEVINDEX_HASHBITS 8
  56 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
  57 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
  58
  59 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  60
  61 static DEFINE_SPINLOCK(fib_multipath_lock);
  62
  63 #define for_nexthops(fi) {                                              \
  64         int nhsel; const struct fib_nh *nh;                             \
  65         for (nhsel = 0, nh = (fi)->fib_nh;                              \
  66              nhsel < (fi)->fib_nhs;                                     \
  67              nh++, nhsel++)
  68
  69 #define change_nexthops(fi) {                                           \
  70         int nhsel; struct fib_nh *nexthop_nh;                           \
  71         for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh);   \
  72              nhsel < (fi)->fib_nhs;                                     \
  73              nexthop_nh++, nhsel++)
  74
  75 #else /* CONFIG_IP_ROUTE_MULTIPATH */
  76
  77 /* Hope, that gcc will optimize it to get rid of dummy loop */
  78
  79 #define for_nexthops(fi) {                                              \
  80         int nhsel; const struct fib_nh *nh = (fi)->fib_nh;              \
  81         for (nhsel = 0; nhsel < 1; nhsel++)
  82
  83 #define change_nexthops(fi) {                                           \
  84         int nhsel;                                                      \
  85         struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);    \
  86         for (nhsel = 0; nhsel < 1; nhsel++)
  87
  88 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
  89
  90 #define endfor_nexthops(fi) }
  91
  92
  93 const struct fib_prop fib_props[RTN_MAX + 1] = {
  94         [RTN_UNSPEC] = {
  95                 .error  = 0,
  96                 .scope  = RT_SCOPE_NOWHERE,
  97         },
  98         [RTN_UNICAST] = {
  99                 .error  = 0,
 100                 .scope  = RT_SCOPE_UNIVERSE,
 101         },
 102         [RTN_LOCAL] = {
 103                 .error  = 0,
 104                 .scope  = RT_SCOPE_HOST,
 105         },
 106         [RTN_BROADCAST] = {
 107                 .error  = 0,
 108                 .scope  = RT_SCOPE_LINK,
 109         },
 110         [RTN_ANYCAST] = {
 111                 .error  = 0,
 112                 .scope  = RT_SCOPE_LINK,
 113         },
 114         [RTN_MULTICAST] = {
 115                 .error  = 0,
 116                 .scope  = RT_SCOPE_UNIVERSE,
 117         },
 118         [RTN_BLACKHOLE] = {
 119                 .error  = -EINVAL,
 120                 .scope  = RT_SCOPE_UNIVERSE,
 121         },
 122         [RTN_UNREACHABLE] = {
 123                 .error  = -EHOSTUNREACH,
 124                 .scope  = RT_SCOPE_UNIVERSE,
 125         },
 126         [RTN_PROHIBIT] = {
 127                 .error  = -EACCES,
 128                 .scope  = RT_SCOPE_UNIVERSE,
 129         },
 130         [RTN_THROW] = {
 131                 .error  = -EAGAIN,
 132                 .scope  = RT_SCOPE_UNIVERSE,
 133         },
 134         [RTN_NAT] = {
 135                 .error  = -EINVAL,
 136                 .scope  = RT_SCOPE_NOWHERE,
 137         },
 138         [RTN_XRESOLVE] = {
 139                 .error  = -EINVAL,
 140                 .scope  = RT_SCOPE_NOWHERE,
 141         },
 142 };
 143
 144 static void rt_fibinfo_free(struct rtable __rcu **rtp)
 145 {
 146         struct rtable *rt = rcu_dereference_protected(*rtp, 1);
 147
 148         if (!rt)
 149                 return;
 150
 151         /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
 152          * because we waited an RCU grace period before calling
 153          * free_fib_info_rcu()
 154          */
 155
 156         dst_free(&rt->dst);
 157 }
 158
 159 static void free_nh_exceptions(struct fib_nh *nh)
 160 {
 161         struct fnhe_hash_bucket *hash;
 162         int i;
 163
 164         hash = rcu_dereference_protected(nh->nh_exceptions, 1);
 165         if (!hash)
 166                 return;
 167         for (i = 0; i < FNHE_HASH_SIZE; i++) {
 168                 struct fib_nh_exception *fnhe;
 169
 170                 fnhe = rcu_dereference_protected(hash[i].chain, 1);
 171                 while (fnhe) {
 172                         struct fib_nh_exception *next;
 173
 174                         next = rcu_dereference_protected(fnhe->fnhe_next, 1);
 175
 176                         rt_fibinfo_free(&fnhe->fnhe_rth_input);
 177                         rt_fibinfo_free(&fnhe->fnhe_rth_output);
 178
 179                         kfree(fnhe);
 180
 181                         fnhe = next;
 182                 }
 183         }
 184         kfree(hash);
 185 }
 186
 187 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
 188 {
 189         int cpu;
 190
 191         if (!rtp)
 192                 return;
 193
 194         for_each_possible_cpu(cpu) {
 195                 struct rtable *rt;
 196
 197                 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
 198                 if (rt)
 199                         dst_free(&rt->dst);
 200         }
 201         free_percpu(rtp);
 202 }
 203
 204 /* Release a nexthop info record */
 205 static void free_fib_info_rcu(struct rcu_head *head)
 206 {
 207         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 208
 209         change_nexthops(fi) {
 210                 if (nexthop_nh->nh_dev)
 211                         dev_put(nexthop_nh->nh_dev);
 212                 lwtunnel_state_put(nexthop_nh->nh_lwtstate);
 213                 free_nh_exceptions(nexthop_nh);
 214                 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
 215                 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
 216         } endfor_nexthops(fi);
 217
 218         if (fi->fib_metrics != (u32 *) dst_default_metrics)
 219                 kfree(fi->fib_metrics);
 220         kfree(fi);
 221 }
 222
 223 void free_fib_info(struct fib_info *fi)
 224 {
 225         if (fi->fib_dead == 0) {
 226                 pr_warn("Freeing alive fib_info %p\n", fi);
 227                 return;
 228         }
 229         fib_info_cnt--;
 230 #ifdef CONFIG_IP_ROUTE_CLASSID
 231         change_nexthops(fi) {
 232                 if (nexthop_nh->nh_tclassid)
 233                         fi->fib_net->ipv4.fib_num_tclassid_users--;
 234         } endfor_nexthops(fi);
 235 #endif
 236         call_rcu(&fi->rcu, free_fib_info_rcu);
 237 }
 238
 239 void fib_release_info(struct fib_info *fi)
 240 {
 241         spin_lock_bh(&fib_info_lock);
 242         if (fi && --fi->fib_treeref == 0) {
 243                 hlist_del(&fi->fib_hash);
 244                 if (fi->fib_prefsrc)
 245                         hlist_del(&fi->fib_lhash);
 246                 change_nexthops(fi) {
 247                         if (!nexthop_nh->nh_dev)
 248                                 continue;
 249                         hlist_del(&nexthop_nh->nh_hash);
 250                 } endfor_nexthops(fi)
 251                 fi->fib_dead = 1;
 252                 fib_info_put(fi);
 253         }
 254         spin_unlock_bh(&fib_info_lock);
 255 }
 256
 257 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
 258 {
 259         const struct fib_nh *onh = ofi->fib_nh;
 260
 261         for_nexthops(fi) {
 262                 if (nh->nh_oif != onh->nh_oif ||
 263                     nh->nh_gw  != onh->nh_gw ||
 264                     nh->nh_scope != onh->nh_scope ||
 265 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 266                     nh->nh_weight != onh->nh_weight ||
 267 #endif
 268 #ifdef CONFIG_IP_ROUTE_CLASSID
 269                     nh->nh_tclassid != onh->nh_tclassid ||
 270 #endif
 271                     lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
 272                     ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
 273                         return -1;
 274                 onh++;
 275         } endfor_nexthops(fi);
 276         return 0;
 277 }
 278
 279 static inline unsigned int fib_devindex_hashfn(unsigned int val)
 280 {
 281         unsigned int mask = DEVINDEX_HASHSIZE - 1;
 282
 283         return (val ^
 284                 (val >> DEVINDEX_HASHBITS) ^
 285                 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
 286 }
 287
 288 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
 289 {
 290         unsigned int mask = (fib_info_hash_size - 1);
 291         unsigned int val = fi->fib_nhs;
 292
 293         val ^= (fi->fib_protocol << 8) | fi->fib_scope;
 294         val ^= (__force u32)fi->fib_prefsrc;
 295         val ^= fi->fib_priority;
 296         for_nexthops(fi) {
 297                 val ^= fib_devindex_hashfn(nh->nh_oif);
 298         } endfor_nexthops(fi)
 299
 300         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 301 }
 302
 303 static struct fib_info *fib_find_info(const struct fib_info *nfi)
 304 {
 305         struct hlist_head *head;
 306         struct fib_info *fi;
 307         unsigned int hash;
 308
 309         hash = fib_info_hashfn(nfi);
 310         head = &fib_info_hash[hash];
 311
 312         hlist_for_each_entry(fi, head, fib_hash) {
 313                 if (!net_eq(fi->fib_net, nfi->fib_net))
 314                         continue;
 315                 if (fi->fib_nhs != nfi->fib_nhs)
 316                         continue;
 317                 if (nfi->fib_protocol == fi->fib_protocol &&
 318                     nfi->fib_scope == fi->fib_scope &&
 319                     nfi->fib_prefsrc == fi->fib_prefsrc &&
 320                     nfi->fib_priority == fi->fib_priority &&
 321                     nfi->fib_type == fi->fib_type &&
 322                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 323                            sizeof(u32) * RTAX_MAX) == 0 &&
 324                     !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
 325                     (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
 326                         return fi;
 327         }
 328
 329         return NULL;
 330 }
 331
 332 /* Check, that the gateway is already configured.
 333  * Used only by redirect accept routine.
 334  */
 335 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 336 {
 337         struct hlist_head *head;
 338         struct fib_nh *nh;
 339         unsigned int hash;
 340
 341         spin_lock(&fib_info_lock);
 342
 343         hash = fib_devindex_hashfn(dev->ifindex);
 344         head = &fib_info_devhash[hash];
 345         hlist_for_each_entry(nh, head, nh_hash) {
 346                 if (nh->nh_dev == dev &&
 347                     nh->nh_gw == gw &&
 348                     !(nh->nh_flags & RTNH_F_DEAD)) {
 349                         spin_unlock(&fib_info_lock);
 350                         return 0;
 351                 }
 352         }
 353
 354         spin_unlock(&fib_info_lock);
 355
 356         return -1;
 357 }
 358
 359 static inline size_t fib_nlmsg_size(struct fib_info *fi)
 360 {
 361         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 362                          + nla_total_size(4) /* RTA_TABLE */
 363                          + nla_total_size(4) /* RTA_DST */
 364                          + nla_total_size(4) /* RTA_PRIORITY */
 365                          + nla_total_size(4) /* RTA_PREFSRC */
 366                          + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
 367
 368         /* space for nested metrics */
 369         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 370
 371         if (fi->fib_nhs) {
 372                 size_t nh_encapsize = 0;
 373                 /* Also handles the special case fib_nhs == 1 */
 374
 375                 /* each nexthop is packed in an attribute */
 376                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 377
 378                 /* may contain flow and gateway attribute */
 379                 nhsize += 2 * nla_total_size(4);
 380
 381                 /* grab encap info */
 382                 for_nexthops(fi) {
 383                         if (nh->nh_lwtstate) {
 384                                 /* RTA_ENCAP_TYPE */
 385                                 nh_encapsize += lwtunnel_get_encap_size(
 386                                                 nh->nh_lwtstate);
 387                                 /* RTA_ENCAP */
 388                                 nh_encapsize +=  nla_total_size(2);
 389                         }
 390                 } endfor_nexthops(fi);
 391
 392                 /* all nexthops are packed in a nested attribute */
 393                 payload += nla_total_size((fi->fib_nhs * nhsize) +
 394                                           nh_encapsize);
 395
 396         }
 397
 398         return payload;
 399 }
 400
 401 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 402                int dst_len, u32 tb_id, const struct nl_info *info,
 403                unsigned int nlm_flags)
 404 {
 405         struct sk_buff *skb;
 406         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 407         int err = -ENOBUFS;
 408
 409         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 410         if (!skb)
 411                 goto errout;
 412
 413         err = fib_dump_info(skb, info->portid, seq, event, tb_id,
 414                             fa->fa_type, key, dst_len,
 415                             fa->fa_tos, fa->fa_info, nlm_flags);
 416         if (err < 0) {
 417                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 418                 WARN_ON(err == -EMSGSIZE);
 419                 kfree_skb(skb);
 420                 goto errout;
 421         }
 422         rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
 423                     info->nlh, GFP_KERNEL);
 424         return;
 425 errout:
 426         if (err < 0)
 427                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 428 }
 429
 430 static int fib_detect_death(struct fib_info *fi, int order,
 431                             struct fib_info **last_resort, int *last_idx,
 432                             int dflt)
 433 {
 434         struct neighbour *n;
 435         int state = NUD_NONE;
 436
 437         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 438         if (n) {
 439                 state = n->nud_state;
 440                 neigh_release(n);
 441         }
 442         if (state == NUD_REACHABLE)
 443                 return 0;
 444         if ((state & NUD_VALID) && order != dflt)
 445                 return 0;
 446         if ((state & NUD_VALID) ||
 447             (*last_idx < 0 && order > dflt)) {
 448                 *last_resort = fi;
 449                 *last_idx = order;
 450         }
 451         return 1;
 452 }
 453
 454 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 455
 456 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
 457 {
 458         int nhs = 0;
 459
 460         while (rtnh_ok(rtnh, remaining)) {
 461                 nhs++;
 462                 rtnh = rtnh_next(rtnh, &remaining);
 463         }
 464
 465         /* leftover implies invalid nexthop configuration, discard it */
 466         return remaining > 0 ? 0 : nhs;
 467 }
 468
 469 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 470                        int remaining, struct fib_config *cfg)
 471 {
 472         struct net *net = cfg->fc_nlinfo.nl_net;
 473         int ret;
 474
 475         change_nexthops(fi) {
 476                 int attrlen;
 477
 478                 if (!rtnh_ok(rtnh, remaining))
 479                         return -EINVAL;
 480
 481                 nexthop_nh->nh_flags =
 482                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 483                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
 484                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
 485
 486                 attrlen = rtnh_attrlen(rtnh);
 487                 if (attrlen > 0) {
 488                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 489
 490                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 491                         nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
 492 #ifdef CONFIG_IP_ROUTE_CLASSID
 493                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 494                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 495                         if (nexthop_nh->nh_tclassid)
 496                                 fi->fib_net->ipv4.fib_num_tclassid_users++;
 497 #endif
 498                         nla = nla_find(attrs, attrlen, RTA_ENCAP);
 499                         if (nla) {
 500                                 struct lwtunnel_state *lwtstate;
 501                                 struct net_device *dev = NULL;
 502                                 struct nlattr *nla_entype;
 503
 504                                 nla_entype = nla_find(attrs, attrlen,
 505                                                       RTA_ENCAP_TYPE);
 506                                 if (!nla_entype)
 507                                         goto err_inval;
 508                                 if (cfg->fc_oif)
 509                                         dev = __dev_get_by_index(net, cfg->fc_oif);
 510                                 ret = lwtunnel_build_state(dev, nla_get_u16(
 511                                                            nla_entype),
 512                                                            nla, &lwtstate);
 513                                 if (ret)
 514                                         goto errout;
 515                                 lwtunnel_state_get(lwtstate);
 516                                 nexthop_nh->nh_lwtstate = lwtstate;
 517                         }
 518                 }
 519
 520                 rtnh = rtnh_next(rtnh, &remaining);
 521         } endfor_nexthops(fi);
 522
 523         return 0;
 524
 525 err_inval:
 526         ret = -EINVAL;
 527
 528 errout:
 529         return ret;
 530 }
 531
 532 #endif
 533
 534 int fib_encap_match(struct net *net, u16 encap_type,
 535                     struct nlattr *encap,
 536                     int oif, const struct fib_nh *nh)
 537 {
 538         struct lwtunnel_state *lwtstate;
 539         struct net_device *dev = NULL;
 540         int ret;
 541
 542         if (encap_type == LWTUNNEL_ENCAP_NONE)
 543                 return 0;
 544
 545         if (oif)
 546                 dev = __dev_get_by_index(net, oif);
 547         ret = lwtunnel_build_state(dev, encap_type,
 548                                    encap, &lwtstate);
 549         if (!ret)
 550                 return lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
 551
 552         return 0;
 553 }
 554
 555 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
 556 {
 557         struct net *net = cfg->fc_nlinfo.nl_net;
 558 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 559         struct rtnexthop *rtnh;
 560         int remaining;
 561 #endif
 562
 563         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 564                 return 1;
 565
 566         if (cfg->fc_oif || cfg->fc_gw) {
 567                 if (cfg->fc_encap) {
 568                         if (fib_encap_match(net, cfg->fc_encap_type,
 569                                             cfg->fc_encap, cfg->fc_oif,
 570                                             fi->fib_nh))
 571                             return 1;
 572                 }
 573                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 574                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 575                         return 0;
 576                 return 1;
 577         }
 578
 579 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 580         if (!cfg->fc_mp)
 581                 return 0;
 582
 583         rtnh = cfg->fc_mp;
 584         remaining = cfg->fc_mp_len;
 585
 586         for_nexthops(fi) {
 587                 int attrlen;
 588
 589                 if (!rtnh_ok(rtnh, remaining))
 590                         return -EINVAL;
 591
 592                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 593                         return 1;
 594
 595                 attrlen = rtnh_attrlen(rtnh);
 596                 if (attrlen > 0) {
 597                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 598
 599                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 600                         if (nla && nla_get_in_addr(nla) != nh->nh_gw)
 601                                 return 1;
 602 #ifdef CONFIG_IP_ROUTE_CLASSID
 603                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 604                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 605                                 return 1;
 606 #endif
 607                 }
 608
 609                 rtnh = rtnh_next(rtnh, &remaining);
 610         } endfor_nexthops(fi);
 611 #endif
 612         return 0;
 613 }
 614
 615
 616 /*
 617  * Picture
 618  * -------
 619  *
 620  * Semantics of nexthop is very messy by historical reasons.
 621  * We have to take into account, that:
 622  * a) gateway can be actually local interface address,
 623  *    so that gatewayed route is direct.
 624  * b) gateway must be on-link address, possibly
 625  *    described not by an ifaddr, but also by a direct route.
 626  * c) If both gateway and interface are specified, they should not
 627  *    contradict.
 628  * d) If we use tunnel routes, gateway could be not on-link.
 629  *
 630  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 631  * results in pretty ugly and hairy code with obscure logic.
 632  *
 633  * I chose to generalized it instead, so that the size
 634  * of code does not increase practically, but it becomes
 635  * much more general.
 636  * Every prefix is assigned a "scope" value: "host" is local address,
 637  * "link" is direct route,
 638  * [ ... "site" ... "interior" ... ]
 639  * and "universe" is true gateway route with global meaning.
 640  *
 641  * Every prefix refers to a set of "nexthop"s (gw, oif),
 642  * where gw must have narrower scope. This recursion stops
 643  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 644  * which means that gw is forced to be on link.
 645  *
 646  * Code is still hairy, but now it is apparently logically
 647  * consistent and very flexible. F.e. as by-product it allows
 648  * to co-exists in peace independent exterior and interior
 649  * routing processes.
 650  *
 651  * Normally it looks as following.
 652  *
 653  * {universe prefix}  -> (gw, oif) [scope link]
 654  *                |
 655  *                |-> {link prefix} -> (gw, oif) [scope local]
 656  *                                      |
 657  *                                      |-> {local prefix} (terminal node)
 658  */
 659 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
 660                         struct fib_nh *nh)
 661 {
 662         int err;
 663         struct net *net;
 664         struct net_device *dev;
 665
 666         net = cfg->fc_nlinfo.nl_net;
 667         if (nh->nh_gw) {
 668                 struct fib_result res;
 669
 670                 if (nh->nh_flags & RTNH_F_ONLINK) {
 671
 672                         if (cfg->fc_scope >= RT_SCOPE_LINK)
 673                                 return -EINVAL;
 674                         if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
 675                                 return -EINVAL;
 676                         dev = __dev_get_by_index(net, nh->nh_oif);
 677                         if (!dev)
 678                                 return -ENODEV;
 679                         if (!(dev->flags & IFF_UP))
 680                                 return -ENETDOWN;
 681                         if (!netif_carrier_ok(dev))
 682                                 nh->nh_flags |= RTNH_F_LINKDOWN;
 683                         nh->nh_dev = dev;
 684                         dev_hold(dev);
 685                         nh->nh_scope = RT_SCOPE_LINK;
 686                         return 0;
 687                 }
 688                 rcu_read_lock();
 689                 {
 690                         struct flowi4 fl4 = {
 691                                 .daddr = nh->nh_gw,
 692                                 .flowi4_scope = cfg->fc_scope + 1,
 693                                 .flowi4_oif = nh->nh_oif,
 694                                 .flowi4_iif = LOOPBACK_IFINDEX,
 695                         };
 696
 697                         /* It is not necessary, but requires a bit of thinking */
 698                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
 699                                 fl4.flowi4_scope = RT_SCOPE_LINK;
 700                         err = fib_lookup(net, &fl4, &res,
 701                                          FIB_LOOKUP_IGNORE_LINKSTATE);
 702                         if (err) {
 703                                 rcu_read_unlock();
 704                                 return err;
 705                         }
 706                 }
 707                 err = -EINVAL;
 708                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 709                         goto out;
 710                 nh->nh_scope = res.scope;
 711                 nh->nh_oif = FIB_RES_OIF(res);
 712                 nh->nh_dev = dev = FIB_RES_DEV(res);
 713                 if (!dev)
 714                         goto out;
 715                 dev_hold(dev);
 716                 if (!netif_carrier_ok(dev))
 717                         nh->nh_flags |= RTNH_F_LINKDOWN;
 718                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 719         } else {
 720                 struct in_device *in_dev;
 721
 722                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 723                         return -EINVAL;
 724
 725                 rcu_read_lock();
 726                 err = -ENODEV;
 727                 in_dev = inetdev_by_index(net, nh->nh_oif);
 728                 if (!in_dev)
 729                         goto out;
 730                 err = -ENETDOWN;
 731                 if (!(in_dev->dev->flags & IFF_UP))
 732                         goto out;
 733                 nh->nh_dev = in_dev->dev;
 734                 dev_hold(nh->nh_dev);
 735                 nh->nh_scope = RT_SCOPE_HOST;
 736                 if (!netif_carrier_ok(nh->nh_dev))
 737                         nh->nh_flags |= RTNH_F_LINKDOWN;
 738                 err = 0;
 739         }
 740 out:
 741         rcu_read_unlock();
 742         return err;
 743 }
 744
 745 static inline unsigned int fib_laddr_hashfn(__be32 val)
 746 {
 747         unsigned int mask = (fib_info_hash_size - 1);
 748
 749         return ((__force u32)val ^
 750                 ((__force u32)val >> 7) ^
 751                 ((__force u32)val >> 14)) & mask;
 752 }
 753
 754 static struct hlist_head *fib_info_hash_alloc(int bytes)
 755 {
 756         if (bytes <= PAGE_SIZE)
 757                 return kzalloc(bytes, GFP_KERNEL);
 758         else
 759                 return (struct hlist_head *)
 760                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 761                                          get_order(bytes));
 762 }
 763
 764 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 765 {
 766         if (!hash)
 767                 return;
 768
 769         if (bytes <= PAGE_SIZE)
 770                 kfree(hash);
 771         else
 772                 free_pages((unsigned long) hash, get_order(bytes));
 773 }
 774
 775 static void fib_info_hash_move(struct hlist_head *new_info_hash,
 776                                struct hlist_head *new_laddrhash,
 777                                unsigned int new_size)
 778 {
 779         struct hlist_head *old_info_hash, *old_laddrhash;
 780         unsigned int old_size = fib_info_hash_size;
 781         unsigned int i, bytes;
 782
 783         spin_lock_bh(&fib_info_lock);
 784         old_info_hash = fib_info_hash;
 785         old_laddrhash = fib_info_laddrhash;
 786         fib_info_hash_size = new_size;
 787
 788         for (i = 0; i < old_size; i++) {
 789                 struct hlist_head *head = &fib_info_hash[i];
 790                 struct hlist_node *n;
 791                 struct fib_info *fi;
 792
 793                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
 794                         struct hlist_head *dest;
 795                         unsigned int new_hash;
 796
 797                         new_hash = fib_info_hashfn(fi);
 798                         dest = &new_info_hash[new_hash];
 799                         hlist_add_head(&fi->fib_hash, dest);
 800                 }
 801         }
 802         fib_info_hash = new_info_hash;
 803
 804         for (i = 0; i < old_size; i++) {
 805                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 806                 struct hlist_node *n;
 807                 struct fib_info *fi;
 808
 809                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
 810                         struct hlist_head *ldest;
 811                         unsigned int new_hash;
 812
 813                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 814                         ldest = &new_laddrhash[new_hash];
 815                         hlist_add_head(&fi->fib_lhash, ldest);
 816                 }
 817         }
 818         fib_info_laddrhash = new_laddrhash;
 819
 820         spin_unlock_bh(&fib_info_lock);
 821
 822         bytes = old_size * sizeof(struct hlist_head *);
 823         fib_info_hash_free(old_info_hash, bytes);
 824         fib_info_hash_free(old_laddrhash, bytes);
 825 }
 826
 827 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 828 {
 829         nh->nh_saddr = inet_select_addr(nh->nh_dev,
 830                                         nh->nh_gw,
 831                                         nh->nh_parent->fib_scope);
 832         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 833
 834         return nh->nh_saddr;
 835 }
 836
 837 struct fib_info *fib_create_info(struct fib_config *cfg)
 838 {
 839         int err;
 840         struct fib_info *fi = NULL;
 841         struct fib_info *ofi;
 842         int nhs = 1;
 843         struct net *net = cfg->fc_nlinfo.nl_net;
 844
 845         if (cfg->fc_type > RTN_MAX)
 846                 goto err_inval;
 847
 848         /* Fast check to catch the most weird cases */
 849         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
 850                 goto err_inval;
 851
 852 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 853         if (cfg->fc_mp) {
 854                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
 855                 if (nhs == 0)
 856                         goto err_inval;
 857         }
 858 #endif
 859
 860         err = -ENOBUFS;
 861         if (fib_info_cnt >= fib_info_hash_size) {
 862                 unsigned int new_size = fib_info_hash_size << 1;
 863                 struct hlist_head *new_info_hash;
 864                 struct hlist_head *new_laddrhash;
 865                 unsigned int bytes;
 866
 867                 if (!new_size)
 868                         new_size = 16;
 869                 bytes = new_size * sizeof(struct hlist_head *);
 870                 new_info_hash = fib_info_hash_alloc(bytes);
 871                 new_laddrhash = fib_info_hash_alloc(bytes);
 872                 if (!new_info_hash || !new_laddrhash) {
 873                         fib_info_hash_free(new_info_hash, bytes);
 874                         fib_info_hash_free(new_laddrhash, bytes);
 875                 } else
 876                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
 877
 878                 if (!fib_info_hash_size)
 879                         goto failure;
 880         }
 881
 882         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
 883         if (!fi)
 884                 goto failure;
 885         fib_info_cnt++;
 886         if (cfg->fc_mx) {
 887                 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
 888                 if (!fi->fib_metrics)
 889                         goto failure;
 890         } else
 891                 fi->fib_metrics = (u32 *) dst_default_metrics;
 892
 893         fi->fib_net = net;
 894         fi->fib_protocol = cfg->fc_protocol;
 895         fi->fib_scope = cfg->fc_scope;
 896         fi->fib_flags = cfg->fc_flags;
 897         fi->fib_priority = cfg->fc_priority;
 898         fi->fib_prefsrc = cfg->fc_prefsrc;
 899         fi->fib_type = cfg->fc_type;
 900
 901         fi->fib_nhs = nhs;
 902         change_nexthops(fi) {
 903                 nexthop_nh->nh_parent = fi;
 904                 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
 905                 if (!nexthop_nh->nh_pcpu_rth_output)
 906                         goto failure;
 907         } endfor_nexthops(fi)
 908
 909         if (cfg->fc_mx) {
 910                 struct nlattr *nla;
 911                 int remaining;
 912
 913                 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 914                         int type = nla_type(nla);
 915
 916                         if (type) {
 917                                 u32 val;
 918
 919                                 if (type > RTAX_MAX)
 920                                         goto err_inval;
 921                                 if (type == RTAX_CC_ALGO) {
 922                                         char tmp[TCP_CA_NAME_MAX];
 923
 924                                         nla_strlcpy(tmp, nla, sizeof(tmp));
 925                                         val = tcp_ca_get_key_by_name(tmp);
 926                                         if (val == TCP_CA_UNSPEC)
 927                                                 goto err_inval;
 928                                 } else {
 929                                         val = nla_get_u32(nla);
 930                                 }
 931                                 if (type == RTAX_ADVMSS && val > 65535 - 40)
 932                                         val = 65535 - 40;
 933                                 if (type == RTAX_MTU && val > 65535 - 15)
 934                                         val = 65535 - 15;
 935                                 fi->fib_metrics[type - 1] = val;
 936                         }
 937                 }
 938         }
 939
 940         if (cfg->fc_mp) {
 941 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 942                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
 943                 if (err != 0)
 944                         goto failure;
 945                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
 946                         goto err_inval;
 947                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
 948                         goto err_inval;
 949 #ifdef CONFIG_IP_ROUTE_CLASSID
 950                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
 951                         goto err_inval;
 952 #endif
 953 #else
 954                 goto err_inval;
 955 #endif
 956         } else {
 957                 struct fib_nh *nh = fi->fib_nh;
 958
 959                 if (cfg->fc_encap) {
 960                         struct lwtunnel_state *lwtstate;
 961                         struct net_device *dev = NULL;
 962
 963                         if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
 964                                 goto err_inval;
 965                         if (cfg->fc_oif)
 966                                 dev = __dev_get_by_index(net, cfg->fc_oif);
 967                         err = lwtunnel_build_state(dev, cfg->fc_encap_type,
 968                                                    cfg->fc_encap, &lwtstate);
 969                         if (err)
 970                                 goto failure;
 971
 972                         lwtunnel_state_get(lwtstate);
 973                         nh->nh_lwtstate = lwtstate;
 974                 }
 975                 nh->nh_oif = cfg->fc_oif;
 976                 nh->nh_gw = cfg->fc_gw;
 977                 nh->nh_flags = cfg->fc_flags;
 978 #ifdef CONFIG_IP_ROUTE_CLASSID
 979                 nh->nh_tclassid = cfg->fc_flow;
 980                 if (nh->nh_tclassid)
 981                         fi->fib_net->ipv4.fib_num_tclassid_users++;
 982 #endif
 983 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 984                 nh->nh_weight = 1;
 985 #endif
 986         }
 987
 988         if (fib_props[cfg->fc_type].error) {
 989                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
 990                         goto err_inval;
 991                 goto link_it;
 992         } else {
 993                 switch (cfg->fc_type) {
 994                 case RTN_UNICAST:
 995                 case RTN_LOCAL:
 996                 case RTN_BROADCAST:
 997                 case RTN_ANYCAST:
 998                 case RTN_MULTICAST:
 999                         break;
1000                 default:
1001                         goto err_inval;
1002                 }
1003         }
1004
1005         if (cfg->fc_scope > RT_SCOPE_HOST)
1006                 goto err_inval;
1007
1008         if (cfg->fc_scope == RT_SCOPE_HOST) {
1009                 struct fib_nh *nh = fi->fib_nh;
1010
1011                 /* Local address is added. */
1012                 if (nhs != 1 || nh->nh_gw)
1013                         goto err_inval;
1014                 nh->nh_scope = RT_SCOPE_NOWHERE;
1015                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1016                 err = -ENODEV;
1017                 if (!nh->nh_dev)
1018                         goto failure;
1019         } else {
1020                 int linkdown = 0;
1021
1022                 change_nexthops(fi) {
1023                         err = fib_check_nh(cfg, fi, nexthop_nh);
1024                         if (err != 0)
1025                                 goto failure;
1026                         if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1027                                 linkdown++;
1028                 } endfor_nexthops(fi)
1029                 if (linkdown == fi->fib_nhs)
1030                         fi->fib_flags |= RTNH_F_LINKDOWN;
1031         }
1032
1033         if (fi->fib_prefsrc) {
1034                 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1035                     fi->fib_prefsrc != cfg->fc_dst)
1036                         if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
1037                                 goto err_inval;
1038         }
1039
1040         change_nexthops(fi) {
1041                 fib_info_update_nh_saddr(net, nexthop_nh);
1042         } endfor_nexthops(fi)
1043
1044 link_it:
1045         ofi = fib_find_info(fi);
1046         if (ofi) {
1047                 fi->fib_dead = 1;
1048                 free_fib_info(fi);
1049                 ofi->fib_treeref++;
1050                 return ofi;
1051         }
1052
1053         fi->fib_treeref++;
1054         atomic_inc(&fi->fib_clntref);
1055         spin_lock_bh(&fib_info_lock);
1056         hlist_add_head(&fi->fib_hash,
1057                        &fib_info_hash[fib_info_hashfn(fi)]);
1058         if (fi->fib_prefsrc) {
1059                 struct hlist_head *head;
1060
1061                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1062                 hlist_add_head(&fi->fib_lhash, head);
1063         }
1064         change_nexthops(fi) {
1065                 struct hlist_head *head;
1066                 unsigned int hash;
1067
1068                 if (!nexthop_nh->nh_dev)
1069                         continue;
1070                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1071                 head = &fib_info_devhash[hash];
1072                 hlist_add_head(&nexthop_nh->nh_hash, head);
1073         } endfor_nexthops(fi)
1074         spin_unlock_bh(&fib_info_lock);
1075         return fi;
1076
1077 err_inval:
1078         err = -EINVAL;
1079
1080 failure:
1081         if (fi) {
1082                 fi->fib_dead = 1;
1083                 free_fib_info(fi);
1084         }
1085
1086         return ERR_PTR(err);
1087 }
1088
1089 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1090                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1091                   struct fib_info *fi, unsigned int flags)
1092 {
1093         struct nlmsghdr *nlh;
1094         struct rtmsg *rtm;
1095
1096         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1097         if (!nlh)
1098                 return -EMSGSIZE;
1099
1100         rtm = nlmsg_data(nlh);
1101         rtm->rtm_family = AF_INET;
1102         rtm->rtm_dst_len = dst_len;
1103         rtm->rtm_src_len = 0;
1104         rtm->rtm_tos = tos;
1105         if (tb_id < 256)
1106                 rtm->rtm_table = tb_id;
1107         else
1108                 rtm->rtm_table = RT_TABLE_COMPAT;
1109         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1110                 goto nla_put_failure;
1111         rtm->rtm_type = type;
1112         rtm->rtm_flags = fi->fib_flags;
1113         rtm->rtm_scope = fi->fib_scope;
1114         rtm->rtm_protocol = fi->fib_protocol;
1115
1116         if (rtm->rtm_dst_len &&
1117             nla_put_in_addr(skb, RTA_DST, dst))
1118                 goto nla_put_failure;
1119         if (fi->fib_priority &&
1120             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1121                 goto nla_put_failure;
1122         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1123                 goto nla_put_failure;
1124
1125         if (fi->fib_prefsrc &&
1126             nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1127                 goto nla_put_failure;
1128         if (fi->fib_nhs == 1) {
1129                 struct in_device *in_dev;
1130
1131                 if (fi->fib_nh->nh_gw &&
1132                     nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1133                         goto nla_put_failure;
1134                 if (fi->fib_nh->nh_oif &&
1135                     nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1136                         goto nla_put_failure;
1137                 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1138                         in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1139                         if (in_dev &&
1140                             IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1141                                 rtm->rtm_flags |= RTNH_F_DEAD;
1142                 }
1143 #ifdef CONFIG_IP_ROUTE_CLASSID
1144                 if (fi->fib_nh[0].nh_tclassid &&
1145                     nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1146                         goto nla_put_failure;
1147 #endif
1148                 if (fi->fib_nh->nh_lwtstate)
1149                         lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate);
1150         }
1151 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1152         if (fi->fib_nhs > 1) {
1153                 struct rtnexthop *rtnh;
1154                 struct nlattr *mp;
1155
1156                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1157                 if (!mp)
1158                         goto nla_put_failure;
1159
1160                 for_nexthops(fi) {
1161                         struct in_device *in_dev;
1162
1163                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1164                         if (!rtnh)
1165                                 goto nla_put_failure;
1166
1167                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1168                         if (nh->nh_flags & RTNH_F_LINKDOWN) {
1169                                 in_dev = __in_dev_get_rtnl(nh->nh_dev);
1170                                 if (in_dev &&
1171                                     IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1172                                         rtnh->rtnh_flags |= RTNH_F_DEAD;
1173                         }
1174                         rtnh->rtnh_hops = nh->nh_weight - 1;
1175                         rtnh->rtnh_ifindex = nh->nh_oif;
1176
1177                         if (nh->nh_gw &&
1178                             nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1179                                 goto nla_put_failure;
1180 #ifdef CONFIG_IP_ROUTE_CLASSID
1181                         if (nh->nh_tclassid &&
1182                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1183                                 goto nla_put_failure;
1184 #endif
1185                         if (nh->nh_lwtstate)
1186                                 lwtunnel_fill_encap(skb, nh->nh_lwtstate);
1187                         /* length of rtnetlink header + attributes */
1188                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1189                 } endfor_nexthops(fi);
1190
1191                 nla_nest_end(skb, mp);
1192         }
1193 #endif
1194         nlmsg_end(skb, nlh);
1195         return 0;
1196
1197 nla_put_failure:
1198         nlmsg_cancel(skb, nlh);
1199         return -EMSGSIZE;
1200 }
1201
1202 /*
1203  * Update FIB if:
1204  * - local address disappeared -> we must delete all the entries
1205  *   referring to it.
1206  * - device went down -> we must shutdown all nexthops going via it.
1207  */
1208 int fib_sync_down_addr(struct net *net, __be32 local)
1209 {
1210         int ret = 0;
1211         unsigned int hash = fib_laddr_hashfn(local);
1212         struct hlist_head *head = &fib_info_laddrhash[hash];
1213         struct fib_info *fi;
1214
1215         if (!fib_info_laddrhash || local == 0)
1216                 return 0;
1217
1218         hlist_for_each_entry(fi, head, fib_lhash) {
1219                 if (!net_eq(fi->fib_net, net))
1220                         continue;
1221                 if (fi->fib_prefsrc == local) {
1222                         fi->fib_flags |= RTNH_F_DEAD;
1223                         ret++;
1224                 }
1225         }
1226         return ret;
1227 }
1228
1229 int fib_sync_down_dev(struct net_device *dev, unsigned long event)
1230 {
1231         int ret = 0;
1232         int scope = RT_SCOPE_NOWHERE;
1233         struct fib_info *prev_fi = NULL;
1234         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1235         struct hlist_head *head = &fib_info_devhash[hash];
1236         struct fib_nh *nh;
1237
1238         if (event == NETDEV_UNREGISTER ||
1239             event == NETDEV_DOWN)
1240                 scope = -1;
1241
1242         hlist_for_each_entry(nh, head, nh_hash) {
1243                 struct fib_info *fi = nh->nh_parent;
1244                 int dead;
1245
1246                 BUG_ON(!fi->fib_nhs);
1247                 if (nh->nh_dev != dev || fi == prev_fi)
1248                         continue;
1249                 prev_fi = fi;
1250                 dead = 0;
1251                 change_nexthops(fi) {
1252                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1253                                 dead++;
1254                         else if (nexthop_nh->nh_dev == dev &&
1255                                  nexthop_nh->nh_scope != scope) {
1256                                 switch (event) {
1257                                 case NETDEV_DOWN:
1258                                 case NETDEV_UNREGISTER:
1259                                         nexthop_nh->nh_flags |= RTNH_F_DEAD;
1260                                         /* fall through */
1261                                 case NETDEV_CHANGE:
1262                                         nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1263                                         break;
1264                                 }
1265 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1266                                 spin_lock_bh(&fib_multipath_lock);
1267                                 fi->fib_power -= nexthop_nh->nh_power;
1268                                 nexthop_nh->nh_power = 0;
1269                                 spin_unlock_bh(&fib_multipath_lock);
1270 #endif
1271                                 dead++;
1272                         }
1273 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1274                         if (event == NETDEV_UNREGISTER &&
1275                             nexthop_nh->nh_dev == dev) {
1276                                 dead = fi->fib_nhs;
1277                                 break;
1278                         }
1279 #endif
1280                 } endfor_nexthops(fi)
1281                 if (dead == fi->fib_nhs) {
1282                         switch (event) {
1283                         case NETDEV_DOWN:
1284                         case NETDEV_UNREGISTER:
1285                                 fi->fib_flags |= RTNH_F_DEAD;
1286                                 /* fall through */
1287                         case NETDEV_CHANGE:
1288                                 fi->fib_flags |= RTNH_F_LINKDOWN;
1289                                 break;
1290                         }
1291                         ret++;
1292                 }
1293         }
1294
1295         return ret;
1296 }
1297
1298 /* Must be invoked inside of an RCU protected region.  */
1299 void fib_select_default(struct fib_result *res)
1300 {
1301         struct fib_info *fi = NULL, *last_resort = NULL;
1302         struct hlist_head *fa_head = res->fa_head;
1303         struct fib_table *tb = res->table;
1304         int order = -1, last_idx = -1;
1305         struct fib_alias *fa;
1306
1307         hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1308                 struct fib_info *next_fi = fa->fa_info;
1309
1310                 if (next_fi->fib_scope != res->scope ||
1311                     fa->fa_type != RTN_UNICAST)
1312                         continue;
1313
1314                 if (next_fi->fib_priority > res->fi->fib_priority)
1315                         break;
1316                 if (!next_fi->fib_nh[0].nh_gw ||
1317                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1318                         continue;
1319
1320                 fib_alias_accessed(fa);
1321
1322                 if (!fi) {
1323                         if (next_fi != res->fi)
1324                                 break;
1325                 } else if (!fib_detect_death(fi, order, &last_resort,
1326                                              &last_idx, tb->tb_default)) {
1327                         fib_result_assign(res, fi);
1328                         tb->tb_default = order;
1329                         goto out;
1330                 }
1331                 fi = next_fi;
1332                 order++;
1333         }
1334
1335         if (order <= 0 || !fi) {
1336                 tb->tb_default = -1;
1337                 goto out;
1338         }
1339
1340         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1341                                 tb->tb_default)) {
1342                 fib_result_assign(res, fi);
1343                 tb->tb_default = order;
1344                 goto out;
1345         }
1346
1347         if (last_idx >= 0)
1348                 fib_result_assign(res, last_resort);
1349         tb->tb_default = last_idx;
1350 out:
1351         return;
1352 }
1353
1354 /*
1355  * Dead device goes up. We wake up dead nexthops.
1356  * It takes sense only on multipath routes.
1357  */
1358 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1359 {
1360         struct fib_info *prev_fi;
1361         unsigned int hash;
1362         struct hlist_head *head;
1363         struct fib_nh *nh;
1364         int ret;
1365
1366         if (!(dev->flags & IFF_UP))
1367                 return 0;
1368
1369         prev_fi = NULL;
1370         hash = fib_devindex_hashfn(dev->ifindex);
1371         head = &fib_info_devhash[hash];
1372         ret = 0;
1373
1374         hlist_for_each_entry(nh, head, nh_hash) {
1375                 struct fib_info *fi = nh->nh_parent;
1376                 int alive;
1377
1378                 BUG_ON(!fi->fib_nhs);
1379                 if (nh->nh_dev != dev || fi == prev_fi)
1380                         continue;
1381
1382                 prev_fi = fi;
1383                 alive = 0;
1384                 change_nexthops(fi) {
1385                         if (!(nexthop_nh->nh_flags & nh_flags)) {
1386                                 alive++;
1387                                 continue;
1388                         }
1389                         if (!nexthop_nh->nh_dev ||
1390                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1391                                 continue;
1392                         if (nexthop_nh->nh_dev != dev ||
1393                             !__in_dev_get_rtnl(dev))
1394                                 continue;
1395                         alive++;
1396 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1397                         spin_lock_bh(&fib_multipath_lock);
1398                         nexthop_nh->nh_power = 0;
1399                         nexthop_nh->nh_flags &= ~nh_flags;
1400                         spin_unlock_bh(&fib_multipath_lock);
1401 #else
1402                         nexthop_nh->nh_flags &= ~nh_flags;
1403 #endif
1404                 } endfor_nexthops(fi)
1405
1406                 if (alive > 0) {
1407                         fi->fib_flags &= ~nh_flags;
1408                         ret++;
1409                 }
1410         }
1411
1412         return ret;
1413 }
1414
1415 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1416
1417 /*
1418  * The algorithm is suboptimal, but it provides really
1419  * fair weighted route distribution.
1420  */
1421 void fib_select_multipath(struct fib_result *res)
1422 {
1423         struct fib_info *fi = res->fi;
1424         struct in_device *in_dev;
1425         int w;
1426
1427         spin_lock_bh(&fib_multipath_lock);
1428         if (fi->fib_power <= 0) {
1429                 int power = 0;
1430                 change_nexthops(fi) {
1431                         in_dev = __in_dev_get_rcu(nexthop_nh->nh_dev);
1432                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1433                                 continue;
1434                         if (in_dev &&
1435                             IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1436                             nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1437                                 continue;
1438                         power += nexthop_nh->nh_weight;
1439                         nexthop_nh->nh_power = nexthop_nh->nh_weight;
1440                 } endfor_nexthops(fi);
1441                 fi->fib_power = power;
1442                 if (power <= 0) {
1443                         spin_unlock_bh(&fib_multipath_lock);
1444                         /* Race condition: route has just become dead. */
1445                         res->nh_sel = 0;
1446                         return;
1447                 }
1448         }
1449
1450
1451         /* w should be random number [0..fi->fib_power-1],
1452          * it is pretty bad approximation.
1453          */
1454
1455         w = jiffies % fi->fib_power;
1456
1457         change_nexthops(fi) {
1458                 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1459                     nexthop_nh->nh_power) {
1460                         w -= nexthop_nh->nh_power;
1461                         if (w <= 0) {
1462                                 nexthop_nh->nh_power--;
1463                                 fi->fib_power--;
1464                                 res->nh_sel = nhsel;
1465                                 spin_unlock_bh(&fib_multipath_lock);
1466                                 return;
1467                         }
1468                 }
1469         } endfor_nexthops(fi);
1470
1471         /* Race condition: route has just become dead. */
1472         res->nh_sel = 0;
1473         spin_unlock_bh(&fib_multipath_lock);
1474 }
1475 #endif