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                 lwtstate_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         } else {
 442                 return 0;
 443         }
 444         if (state == NUD_REACHABLE)
 445                 return 0;
 446         if ((state & NUD_VALID) && order != dflt)
 447                 return 0;
 448         if ((state & NUD_VALID) ||
 449             (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
 450                 *last_resort = fi;
 451                 *last_idx = order;
 452         }
 453         return 1;
 454 }
 455
 456 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 457
 458 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
 459 {
 460         int nhs = 0;
 461
 462         while (rtnh_ok(rtnh, remaining)) {
 463                 nhs++;
 464                 rtnh = rtnh_next(rtnh, &remaining);
 465         }
 466
 467         /* leftover implies invalid nexthop configuration, discard it */
 468         return remaining > 0 ? 0 : nhs;
 469 }
 470
 471 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 472                        int remaining, struct fib_config *cfg)
 473 {
 474         struct net *net = cfg->fc_nlinfo.nl_net;
 475         int ret;
 476
 477         change_nexthops(fi) {
 478                 int attrlen;
 479
 480                 if (!rtnh_ok(rtnh, remaining))
 481                         return -EINVAL;
 482
 483                 nexthop_nh->nh_flags =
 484                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 485                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
 486                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
 487
 488                 attrlen = rtnh_attrlen(rtnh);
 489                 if (attrlen > 0) {
 490                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 491
 492                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 493                         nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
 494 #ifdef CONFIG_IP_ROUTE_CLASSID
 495                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 496                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 497                         if (nexthop_nh->nh_tclassid)
 498                                 fi->fib_net->ipv4.fib_num_tclassid_users++;
 499 #endif
 500                         nla = nla_find(attrs, attrlen, RTA_ENCAP);
 501                         if (nla) {
 502                                 struct lwtunnel_state *lwtstate;
 503                                 struct net_device *dev = NULL;
 504                                 struct nlattr *nla_entype;
 505
 506                                 nla_entype = nla_find(attrs, attrlen,
 507                                                       RTA_ENCAP_TYPE);
 508                                 if (!nla_entype)
 509                                         goto err_inval;
 510                                 if (cfg->fc_oif)
 511                                         dev = __dev_get_by_index(net, cfg->fc_oif);
 512                                 ret = lwtunnel_build_state(dev, nla_get_u16(
 513                                                            nla_entype),
 514                                                            nla, &lwtstate);
 515                                 if (ret)
 516                                         goto errout;
 517                                 nexthop_nh->nh_lwtstate =
 518                                         lwtstate_get(lwtstate);
 519                         }
 520                 }
 521
 522                 rtnh = rtnh_next(rtnh, &remaining);
 523         } endfor_nexthops(fi);
 524
 525         return 0;
 526
 527 err_inval:
 528         ret = -EINVAL;
 529
 530 errout:
 531         return ret;
 532 }
 533
 534 #endif
 535
 536 int fib_encap_match(struct net *net, u16 encap_type,
 537                     struct nlattr *encap,
 538                     int oif, const struct fib_nh *nh)
 539 {
 540         struct lwtunnel_state *lwtstate;
 541         struct net_device *dev = NULL;
 542         int ret;
 543
 544         if (encap_type == LWTUNNEL_ENCAP_NONE)
 545                 return 0;
 546
 547         if (oif)
 548                 dev = __dev_get_by_index(net, oif);
 549         ret = lwtunnel_build_state(dev, encap_type,
 550                                    encap, &lwtstate);
 551         if (!ret)
 552                 return lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
 553
 554         return 0;
 555 }
 556
 557 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
 558 {
 559         struct net *net = cfg->fc_nlinfo.nl_net;
 560 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 561         struct rtnexthop *rtnh;
 562         int remaining;
 563 #endif
 564
 565         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 566                 return 1;
 567
 568         if (cfg->fc_oif || cfg->fc_gw) {
 569                 if (cfg->fc_encap) {
 570                         if (fib_encap_match(net, cfg->fc_encap_type,
 571                                             cfg->fc_encap, cfg->fc_oif,
 572                                             fi->fib_nh))
 573                             return 1;
 574                 }
 575                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 576                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 577                         return 0;
 578                 return 1;
 579         }
 580
 581 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 582         if (!cfg->fc_mp)
 583                 return 0;
 584
 585         rtnh = cfg->fc_mp;
 586         remaining = cfg->fc_mp_len;
 587
 588         for_nexthops(fi) {
 589                 int attrlen;
 590
 591                 if (!rtnh_ok(rtnh, remaining))
 592                         return -EINVAL;
 593
 594                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 595                         return 1;
 596
 597                 attrlen = rtnh_attrlen(rtnh);
 598                 if (attrlen > 0) {
 599                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 600
 601                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 602                         if (nla && nla_get_in_addr(nla) != nh->nh_gw)
 603                                 return 1;
 604 #ifdef CONFIG_IP_ROUTE_CLASSID
 605                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 606                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 607                                 return 1;
 608 #endif
 609                 }
 610
 611                 rtnh = rtnh_next(rtnh, &remaining);
 612         } endfor_nexthops(fi);
 613 #endif
 614         return 0;
 615 }
 616
 617
 618 /*
 619  * Picture
 620  * -------
 621  *
 622  * Semantics of nexthop is very messy by historical reasons.
 623  * We have to take into account, that:
 624  * a) gateway can be actually local interface address,
 625  *    so that gatewayed route is direct.
 626  * b) gateway must be on-link address, possibly
 627  *    described not by an ifaddr, but also by a direct route.
 628  * c) If both gateway and interface are specified, they should not
 629  *    contradict.
 630  * d) If we use tunnel routes, gateway could be not on-link.
 631  *
 632  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 633  * results in pretty ugly and hairy code with obscure logic.
 634  *
 635  * I chose to generalized it instead, so that the size
 636  * of code does not increase practically, but it becomes
 637  * much more general.
 638  * Every prefix is assigned a "scope" value: "host" is local address,
 639  * "link" is direct route,
 640  * [ ... "site" ... "interior" ... ]
 641  * and "universe" is true gateway route with global meaning.
 642  *
 643  * Every prefix refers to a set of "nexthop"s (gw, oif),
 644  * where gw must have narrower scope. This recursion stops
 645  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 646  * which means that gw is forced to be on link.
 647  *
 648  * Code is still hairy, but now it is apparently logically
 649  * consistent and very flexible. F.e. as by-product it allows
 650  * to co-exists in peace independent exterior and interior
 651  * routing processes.
 652  *
 653  * Normally it looks as following.
 654  *
 655  * {universe prefix}  -> (gw, oif) [scope link]
 656  *                |
 657  *                |-> {link prefix} -> (gw, oif) [scope local]
 658  *                                      |
 659  *                                      |-> {local prefix} (terminal node)
 660  */
 661 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
 662                         struct fib_nh *nh)
 663 {
 664         int err;
 665         struct net *net;
 666         struct net_device *dev;
 667
 668         net = cfg->fc_nlinfo.nl_net;
 669         if (nh->nh_gw) {
 670                 struct fib_result res;
 671
 672                 if (nh->nh_flags & RTNH_F_ONLINK) {
 673                         unsigned int addr_type;
 674
 675                         if (cfg->fc_scope >= RT_SCOPE_LINK)
 676                                 return -EINVAL;
 677                         dev = __dev_get_by_index(net, nh->nh_oif);
 678                         if (!dev)
 679                                 return -ENODEV;
 680                         if (!(dev->flags & IFF_UP))
 681                                 return -ENETDOWN;
 682                         addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
 683                         if (addr_type != RTN_UNICAST)
 684                                 return -EINVAL;
 685                         if (!netif_carrier_ok(dev))
 686                                 nh->nh_flags |= RTNH_F_LINKDOWN;
 687                         nh->nh_dev = dev;
 688                         dev_hold(dev);
 689                         nh->nh_scope = RT_SCOPE_LINK;
 690                         return 0;
 691                 }
 692                 rcu_read_lock();
 693                 {
 694                         struct flowi4 fl4 = {
 695                                 .daddr = nh->nh_gw,
 696                                 .flowi4_scope = cfg->fc_scope + 1,
 697                                 .flowi4_oif = nh->nh_oif,
 698                                 .flowi4_iif = LOOPBACK_IFINDEX,
 699                         };
 700
 701                         /* It is not necessary, but requires a bit of thinking */
 702                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
 703                                 fl4.flowi4_scope = RT_SCOPE_LINK;
 704                         err = fib_lookup(net, &fl4, &res,
 705                                          FIB_LOOKUP_IGNORE_LINKSTATE);
 706                         if (err) {
 707                                 rcu_read_unlock();
 708                                 return err;
 709                         }
 710                 }
 711                 err = -EINVAL;
 712                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 713                         goto out;
 714                 nh->nh_scope = res.scope;
 715                 nh->nh_oif = FIB_RES_OIF(res);
 716                 nh->nh_dev = dev = FIB_RES_DEV(res);
 717                 if (!dev)
 718                         goto out;
 719                 dev_hold(dev);
 720                 if (!netif_carrier_ok(dev))
 721                         nh->nh_flags |= RTNH_F_LINKDOWN;
 722                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 723         } else {
 724                 struct in_device *in_dev;
 725
 726                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 727                         return -EINVAL;
 728
 729                 rcu_read_lock();
 730                 err = -ENODEV;
 731                 in_dev = inetdev_by_index(net, nh->nh_oif);
 732                 if (!in_dev)
 733                         goto out;
 734                 err = -ENETDOWN;
 735                 if (!(in_dev->dev->flags & IFF_UP))
 736                         goto out;
 737                 nh->nh_dev = in_dev->dev;
 738                 dev_hold(nh->nh_dev);
 739                 nh->nh_scope = RT_SCOPE_HOST;
 740                 if (!netif_carrier_ok(nh->nh_dev))
 741                         nh->nh_flags |= RTNH_F_LINKDOWN;
 742                 err = 0;
 743         }
 744 out:
 745         rcu_read_unlock();
 746         return err;
 747 }
 748
 749 static inline unsigned int fib_laddr_hashfn(__be32 val)
 750 {
 751         unsigned int mask = (fib_info_hash_size - 1);
 752
 753         return ((__force u32)val ^
 754                 ((__force u32)val >> 7) ^
 755                 ((__force u32)val >> 14)) & mask;
 756 }
 757
 758 static struct hlist_head *fib_info_hash_alloc(int bytes)
 759 {
 760         if (bytes <= PAGE_SIZE)
 761                 return kzalloc(bytes, GFP_KERNEL);
 762         else
 763                 return (struct hlist_head *)
 764                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 765                                          get_order(bytes));
 766 }
 767
 768 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 769 {
 770         if (!hash)
 771                 return;
 772
 773         if (bytes <= PAGE_SIZE)
 774                 kfree(hash);
 775         else
 776                 free_pages((unsigned long) hash, get_order(bytes));
 777 }
 778
 779 static void fib_info_hash_move(struct hlist_head *new_info_hash,
 780                                struct hlist_head *new_laddrhash,
 781                                unsigned int new_size)
 782 {
 783         struct hlist_head *old_info_hash, *old_laddrhash;
 784         unsigned int old_size = fib_info_hash_size;
 785         unsigned int i, bytes;
 786
 787         spin_lock_bh(&fib_info_lock);
 788         old_info_hash = fib_info_hash;
 789         old_laddrhash = fib_info_laddrhash;
 790         fib_info_hash_size = new_size;
 791
 792         for (i = 0; i < old_size; i++) {
 793                 struct hlist_head *head = &fib_info_hash[i];
 794                 struct hlist_node *n;
 795                 struct fib_info *fi;
 796
 797                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
 798                         struct hlist_head *dest;
 799                         unsigned int new_hash;
 800
 801                         new_hash = fib_info_hashfn(fi);
 802                         dest = &new_info_hash[new_hash];
 803                         hlist_add_head(&fi->fib_hash, dest);
 804                 }
 805         }
 806         fib_info_hash = new_info_hash;
 807
 808         for (i = 0; i < old_size; i++) {
 809                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 810                 struct hlist_node *n;
 811                 struct fib_info *fi;
 812
 813                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
 814                         struct hlist_head *ldest;
 815                         unsigned int new_hash;
 816
 817                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 818                         ldest = &new_laddrhash[new_hash];
 819                         hlist_add_head(&fi->fib_lhash, ldest);
 820                 }
 821         }
 822         fib_info_laddrhash = new_laddrhash;
 823
 824         spin_unlock_bh(&fib_info_lock);
 825
 826         bytes = old_size * sizeof(struct hlist_head *);
 827         fib_info_hash_free(old_info_hash, bytes);
 828         fib_info_hash_free(old_laddrhash, bytes);
 829 }
 830
 831 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 832 {
 833         nh->nh_saddr = inet_select_addr(nh->nh_dev,
 834                                         nh->nh_gw,
 835                                         nh->nh_parent->fib_scope);
 836         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 837
 838         return nh->nh_saddr;
 839 }
 840
 841 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
 842 {
 843         if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 844             fib_prefsrc != cfg->fc_dst) {
 845                 int tb_id = cfg->fc_table;
 846
 847                 if (tb_id == RT_TABLE_MAIN)
 848                         tb_id = RT_TABLE_LOCAL;
 849
 850                 if (inet_addr_type_table(cfg->fc_nlinfo.nl_net,
 851                                          fib_prefsrc, tb_id) != RTN_LOCAL) {
 852                         return false;
 853                 }
 854         }
 855         return true;
 856 }
 857
 858 struct fib_info *fib_create_info(struct fib_config *cfg)
 859 {
 860         int err;
 861         struct fib_info *fi = NULL;
 862         struct fib_info *ofi;
 863         int nhs = 1;
 864         struct net *net = cfg->fc_nlinfo.nl_net;
 865
 866         if (cfg->fc_type > RTN_MAX)
 867                 goto err_inval;
 868
 869         /* Fast check to catch the most weird cases */
 870         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
 871                 goto err_inval;
 872
 873 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 874         if (cfg->fc_mp) {
 875                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
 876                 if (nhs == 0)
 877                         goto err_inval;
 878         }
 879 #endif
 880
 881         err = -ENOBUFS;
 882         if (fib_info_cnt >= fib_info_hash_size) {
 883                 unsigned int new_size = fib_info_hash_size << 1;
 884                 struct hlist_head *new_info_hash;
 885                 struct hlist_head *new_laddrhash;
 886                 unsigned int bytes;
 887
 888                 if (!new_size)
 889                         new_size = 16;
 890                 bytes = new_size * sizeof(struct hlist_head *);
 891                 new_info_hash = fib_info_hash_alloc(bytes);
 892                 new_laddrhash = fib_info_hash_alloc(bytes);
 893                 if (!new_info_hash || !new_laddrhash) {
 894                         fib_info_hash_free(new_info_hash, bytes);
 895                         fib_info_hash_free(new_laddrhash, bytes);
 896                 } else
 897                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
 898
 899                 if (!fib_info_hash_size)
 900                         goto failure;
 901         }
 902
 903         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
 904         if (!fi)
 905                 goto failure;
 906         fib_info_cnt++;
 907         if (cfg->fc_mx) {
 908                 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
 909                 if (!fi->fib_metrics)
 910                         goto failure;
 911         } else
 912                 fi->fib_metrics = (u32 *) dst_default_metrics;
 913
 914         fi->fib_net = net;
 915         fi->fib_protocol = cfg->fc_protocol;
 916         fi->fib_scope = cfg->fc_scope;
 917         fi->fib_flags = cfg->fc_flags;
 918         fi->fib_priority = cfg->fc_priority;
 919         fi->fib_prefsrc = cfg->fc_prefsrc;
 920         fi->fib_type = cfg->fc_type;
 921
 922         fi->fib_nhs = nhs;
 923         change_nexthops(fi) {
 924                 nexthop_nh->nh_parent = fi;
 925                 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
 926                 if (!nexthop_nh->nh_pcpu_rth_output)
 927                         goto failure;
 928         } endfor_nexthops(fi)
 929
 930         if (cfg->fc_mx) {
 931                 struct nlattr *nla;
 932                 int remaining;
 933
 934                 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 935                         int type = nla_type(nla);
 936
 937                         if (type) {
 938                                 u32 val;
 939
 940                                 if (type > RTAX_MAX)
 941                                         goto err_inval;
 942                                 if (type == RTAX_CC_ALGO) {
 943                                         char tmp[TCP_CA_NAME_MAX];
 944
 945                                         nla_strlcpy(tmp, nla, sizeof(tmp));
 946                                         val = tcp_ca_get_key_by_name(tmp);
 947                                         if (val == TCP_CA_UNSPEC)
 948                                                 goto err_inval;
 949                                 } else {
 950                                         val = nla_get_u32(nla);
 951                                 }
 952                                 if (type == RTAX_ADVMSS && val > 65535 - 40)
 953                                         val = 65535 - 40;
 954                                 if (type == RTAX_MTU && val > 65535 - 15)
 955                                         val = 65535 - 15;
 956                                 fi->fib_metrics[type - 1] = val;
 957                         }
 958                 }
 959         }
 960
 961         if (cfg->fc_mp) {
 962 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 963                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
 964                 if (err != 0)
 965                         goto failure;
 966                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
 967                         goto err_inval;
 968                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
 969                         goto err_inval;
 970 #ifdef CONFIG_IP_ROUTE_CLASSID
 971                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
 972                         goto err_inval;
 973 #endif
 974 #else
 975                 goto err_inval;
 976 #endif
 977         } else {
 978                 struct fib_nh *nh = fi->fib_nh;
 979
 980                 if (cfg->fc_encap) {
 981                         struct lwtunnel_state *lwtstate;
 982                         struct net_device *dev = NULL;
 983
 984                         if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
 985                                 goto err_inval;
 986                         if (cfg->fc_oif)
 987                                 dev = __dev_get_by_index(net, cfg->fc_oif);
 988                         err = lwtunnel_build_state(dev, cfg->fc_encap_type,
 989                                                    cfg->fc_encap, &lwtstate);
 990                         if (err)
 991                                 goto failure;
 992
 993                         nh->nh_lwtstate = lwtstate_get(lwtstate);
 994                 }
 995                 nh->nh_oif = cfg->fc_oif;
 996                 nh->nh_gw = cfg->fc_gw;
 997                 nh->nh_flags = cfg->fc_flags;
 998 #ifdef CONFIG_IP_ROUTE_CLASSID
 999                 nh->nh_tclassid = cfg->fc_flow;
1000                 if (nh->nh_tclassid)
1001                         fi->fib_net->ipv4.fib_num_tclassid_users++;
1002 #endif
1003 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1004                 nh->nh_weight = 1;
1005 #endif
1006         }
1007
1008         if (fib_props[cfg->fc_type].error) {
1009                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
1010                         goto err_inval;
1011                 goto link_it;
1012         } else {
1013                 switch (cfg->fc_type) {
1014                 case RTN_UNICAST:
1015                 case RTN_LOCAL:
1016                 case RTN_BROADCAST:
1017                 case RTN_ANYCAST:
1018                 case RTN_MULTICAST:
1019                         break;
1020                 default:
1021                         goto err_inval;
1022                 }
1023         }
1024
1025         if (cfg->fc_scope > RT_SCOPE_HOST)
1026                 goto err_inval;
1027
1028         if (cfg->fc_scope == RT_SCOPE_HOST) {
1029                 struct fib_nh *nh = fi->fib_nh;
1030
1031                 /* Local address is added. */
1032                 if (nhs != 1 || nh->nh_gw)
1033                         goto err_inval;
1034                 nh->nh_scope = RT_SCOPE_NOWHERE;
1035                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1036                 err = -ENODEV;
1037                 if (!nh->nh_dev)
1038                         goto failure;
1039         } else {
1040                 int linkdown = 0;
1041
1042                 change_nexthops(fi) {
1043                         err = fib_check_nh(cfg, fi, nexthop_nh);
1044                         if (err != 0)
1045                                 goto failure;
1046                         if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1047                                 linkdown++;
1048                 } endfor_nexthops(fi)
1049                 if (linkdown == fi->fib_nhs)
1050                         fi->fib_flags |= RTNH_F_LINKDOWN;
1051         }
1052
1053         if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc))
1054                 goto err_inval;
1055
1056         change_nexthops(fi) {
1057                 fib_info_update_nh_saddr(net, nexthop_nh);
1058         } endfor_nexthops(fi)
1059
1060 link_it:
1061         ofi = fib_find_info(fi);
1062         if (ofi) {
1063                 fi->fib_dead = 1;
1064                 free_fib_info(fi);
1065                 ofi->fib_treeref++;
1066                 return ofi;
1067         }
1068
1069         fi->fib_treeref++;
1070         atomic_inc(&fi->fib_clntref);
1071         spin_lock_bh(&fib_info_lock);
1072         hlist_add_head(&fi->fib_hash,
1073                        &fib_info_hash[fib_info_hashfn(fi)]);
1074         if (fi->fib_prefsrc) {
1075                 struct hlist_head *head;
1076
1077                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1078                 hlist_add_head(&fi->fib_lhash, head);
1079         }
1080         change_nexthops(fi) {
1081                 struct hlist_head *head;
1082                 unsigned int hash;
1083
1084                 if (!nexthop_nh->nh_dev)
1085                         continue;
1086                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1087                 head = &fib_info_devhash[hash];
1088                 hlist_add_head(&nexthop_nh->nh_hash, head);
1089         } endfor_nexthops(fi)
1090         spin_unlock_bh(&fib_info_lock);
1091         return fi;
1092
1093 err_inval:
1094         err = -EINVAL;
1095
1096 failure:
1097         if (fi) {
1098                 fi->fib_dead = 1;
1099                 free_fib_info(fi);
1100         }
1101
1102         return ERR_PTR(err);
1103 }
1104
1105 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1106                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1107                   struct fib_info *fi, unsigned int flags)
1108 {
1109         struct nlmsghdr *nlh;
1110         struct rtmsg *rtm;
1111
1112         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1113         if (!nlh)
1114                 return -EMSGSIZE;
1115
1116         rtm = nlmsg_data(nlh);
1117         rtm->rtm_family = AF_INET;
1118         rtm->rtm_dst_len = dst_len;
1119         rtm->rtm_src_len = 0;
1120         rtm->rtm_tos = tos;
1121         if (tb_id < 256)
1122                 rtm->rtm_table = tb_id;
1123         else
1124                 rtm->rtm_table = RT_TABLE_COMPAT;
1125         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1126                 goto nla_put_failure;
1127         rtm->rtm_type = type;
1128         rtm->rtm_flags = fi->fib_flags;
1129         rtm->rtm_scope = fi->fib_scope;
1130         rtm->rtm_protocol = fi->fib_protocol;
1131
1132         if (rtm->rtm_dst_len &&
1133             nla_put_in_addr(skb, RTA_DST, dst))
1134                 goto nla_put_failure;
1135         if (fi->fib_priority &&
1136             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1137                 goto nla_put_failure;
1138         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1139                 goto nla_put_failure;
1140
1141         if (fi->fib_prefsrc &&
1142             nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1143                 goto nla_put_failure;
1144         if (fi->fib_nhs == 1) {
1145                 struct in_device *in_dev;
1146
1147                 if (fi->fib_nh->nh_gw &&
1148                     nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1149                         goto nla_put_failure;
1150                 if (fi->fib_nh->nh_oif &&
1151                     nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1152                         goto nla_put_failure;
1153                 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1154                         in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1155                         if (in_dev &&
1156                             IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1157                                 rtm->rtm_flags |= RTNH_F_DEAD;
1158                 }
1159 #ifdef CONFIG_IP_ROUTE_CLASSID
1160                 if (fi->fib_nh[0].nh_tclassid &&
1161                     nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1162                         goto nla_put_failure;
1163 #endif
1164                 if (fi->fib_nh->nh_lwtstate)
1165                         lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate);
1166         }
1167 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1168         if (fi->fib_nhs > 1) {
1169                 struct rtnexthop *rtnh;
1170                 struct nlattr *mp;
1171
1172                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1173                 if (!mp)
1174                         goto nla_put_failure;
1175
1176                 for_nexthops(fi) {
1177                         struct in_device *in_dev;
1178
1179                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1180                         if (!rtnh)
1181                                 goto nla_put_failure;
1182
1183                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1184                         if (nh->nh_flags & RTNH_F_LINKDOWN) {
1185                                 in_dev = __in_dev_get_rtnl(nh->nh_dev);
1186                                 if (in_dev &&
1187                                     IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1188                                         rtnh->rtnh_flags |= RTNH_F_DEAD;
1189                         }
1190                         rtnh->rtnh_hops = nh->nh_weight - 1;
1191                         rtnh->rtnh_ifindex = nh->nh_oif;
1192
1193                         if (nh->nh_gw &&
1194                             nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1195                                 goto nla_put_failure;
1196 #ifdef CONFIG_IP_ROUTE_CLASSID
1197                         if (nh->nh_tclassid &&
1198                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1199                                 goto nla_put_failure;
1200 #endif
1201                         if (nh->nh_lwtstate)
1202                                 lwtunnel_fill_encap(skb, nh->nh_lwtstate);
1203                         /* length of rtnetlink header + attributes */
1204                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1205                 } endfor_nexthops(fi);
1206
1207                 nla_nest_end(skb, mp);
1208         }
1209 #endif
1210         nlmsg_end(skb, nlh);
1211         return 0;
1212
1213 nla_put_failure:
1214         nlmsg_cancel(skb, nlh);
1215         return -EMSGSIZE;
1216 }
1217
1218 /*
1219  * Update FIB if:
1220  * - local address disappeared -> we must delete all the entries
1221  *   referring to it.
1222  * - device went down -> we must shutdown all nexthops going via it.
1223  */
1224 int fib_sync_down_addr(struct net *net, __be32 local)
1225 {
1226         int ret = 0;
1227         unsigned int hash = fib_laddr_hashfn(local);
1228         struct hlist_head *head = &fib_info_laddrhash[hash];
1229         struct fib_info *fi;
1230
1231         if (!fib_info_laddrhash || local == 0)
1232                 return 0;
1233
1234         hlist_for_each_entry(fi, head, fib_lhash) {
1235                 if (!net_eq(fi->fib_net, net))
1236                         continue;
1237                 if (fi->fib_prefsrc == local) {
1238                         fi->fib_flags |= RTNH_F_DEAD;
1239                         ret++;
1240                 }
1241         }
1242         return ret;
1243 }
1244
1245 int fib_sync_down_dev(struct net_device *dev, unsigned long event)
1246 {
1247         int ret = 0;
1248         int scope = RT_SCOPE_NOWHERE;
1249         struct fib_info *prev_fi = NULL;
1250         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1251         struct hlist_head *head = &fib_info_devhash[hash];
1252         struct fib_nh *nh;
1253
1254         if (event == NETDEV_UNREGISTER ||
1255             event == NETDEV_DOWN)
1256                 scope = -1;
1257
1258         hlist_for_each_entry(nh, head, nh_hash) {
1259                 struct fib_info *fi = nh->nh_parent;
1260                 int dead;
1261
1262                 BUG_ON(!fi->fib_nhs);
1263                 if (nh->nh_dev != dev || fi == prev_fi)
1264                         continue;
1265                 prev_fi = fi;
1266                 dead = 0;
1267                 change_nexthops(fi) {
1268                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1269                                 dead++;
1270                         else if (nexthop_nh->nh_dev == dev &&
1271                                  nexthop_nh->nh_scope != scope) {
1272                                 switch (event) {
1273                                 case NETDEV_DOWN:
1274                                 case NETDEV_UNREGISTER:
1275                                         nexthop_nh->nh_flags |= RTNH_F_DEAD;
1276                                         /* fall through */
1277                                 case NETDEV_CHANGE:
1278                                         nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1279                                         break;
1280                                 }
1281 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1282                                 spin_lock_bh(&fib_multipath_lock);
1283                                 fi->fib_power -= nexthop_nh->nh_power;
1284                                 nexthop_nh->nh_power = 0;
1285                                 spin_unlock_bh(&fib_multipath_lock);
1286 #endif
1287                                 dead++;
1288                         }
1289 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1290                         if (event == NETDEV_UNREGISTER &&
1291                             nexthop_nh->nh_dev == dev) {
1292                                 dead = fi->fib_nhs;
1293                                 break;
1294                         }
1295 #endif
1296                 } endfor_nexthops(fi)
1297                 if (dead == fi->fib_nhs) {
1298                         switch (event) {
1299                         case NETDEV_DOWN:
1300                         case NETDEV_UNREGISTER:
1301                                 fi->fib_flags |= RTNH_F_DEAD;
1302                                 /* fall through */
1303                         case NETDEV_CHANGE:
1304                                 fi->fib_flags |= RTNH_F_LINKDOWN;
1305                                 break;
1306                         }
1307                         ret++;
1308                 }
1309         }
1310
1311         return ret;
1312 }
1313
1314 /* Must be invoked inside of an RCU protected region.  */
1315 void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1316 {
1317         struct fib_info *fi = NULL, *last_resort = NULL;
1318         struct hlist_head *fa_head = res->fa_head;
1319         struct fib_table *tb = res->table;
1320         u8 slen = 32 - res->prefixlen;
1321         int order = -1, last_idx = -1;
1322         struct fib_alias *fa, *fa1 = NULL;
1323         u32 last_prio = res->fi->fib_priority;
1324         u8 last_tos = 0;
1325
1326         hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1327                 struct fib_info *next_fi = fa->fa_info;
1328
1329                 if (fa->fa_slen != slen)
1330                         continue;
1331                 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1332                         continue;
1333                 if (fa->tb_id != tb->tb_id)
1334                         continue;
1335                 if (next_fi->fib_priority > last_prio &&
1336                     fa->fa_tos == last_tos) {
1337                         if (last_tos)
1338                                 continue;
1339                         break;
1340                 }
1341                 if (next_fi->fib_flags & RTNH_F_DEAD)
1342                         continue;
1343                 last_tos = fa->fa_tos;
1344                 last_prio = next_fi->fib_priority;
1345
1346                 if (next_fi->fib_scope != res->scope ||
1347                     fa->fa_type != RTN_UNICAST)
1348                         continue;
1349                 if (!next_fi->fib_nh[0].nh_gw ||
1350                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1351                         continue;
1352
1353                 fib_alias_accessed(fa);
1354
1355                 if (!fi) {
1356                         if (next_fi != res->fi)
1357                                 break;
1358                         fa1 = fa;
1359                 } else if (!fib_detect_death(fi, order, &last_resort,
1360                                              &last_idx, fa1->fa_default)) {
1361                         fib_result_assign(res, fi);
1362                         fa1->fa_default = order;
1363                         goto out;
1364                 }
1365                 fi = next_fi;
1366                 order++;
1367         }
1368
1369         if (order <= 0 || !fi) {
1370                 if (fa1)
1371                         fa1->fa_default = -1;
1372                 goto out;
1373         }
1374
1375         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1376                               fa1->fa_default)) {
1377                 fib_result_assign(res, fi);
1378                 fa1->fa_default = order;
1379                 goto out;
1380         }
1381
1382         if (last_idx >= 0)
1383                 fib_result_assign(res, last_resort);
1384         fa1->fa_default = last_idx;
1385 out:
1386         return;
1387 }
1388
1389 /*
1390  * Dead device goes up. We wake up dead nexthops.
1391  * It takes sense only on multipath routes.
1392  */
1393 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1394 {
1395         struct fib_info *prev_fi;
1396         unsigned int hash;
1397         struct hlist_head *head;
1398         struct fib_nh *nh;
1399         int ret;
1400
1401         if (!(dev->flags & IFF_UP))
1402                 return 0;
1403
1404         prev_fi = NULL;
1405         hash = fib_devindex_hashfn(dev->ifindex);
1406         head = &fib_info_devhash[hash];
1407         ret = 0;
1408
1409         hlist_for_each_entry(nh, head, nh_hash) {
1410                 struct fib_info *fi = nh->nh_parent;
1411                 int alive;
1412
1413                 BUG_ON(!fi->fib_nhs);
1414                 if (nh->nh_dev != dev || fi == prev_fi)
1415                         continue;
1416
1417                 prev_fi = fi;
1418                 alive = 0;
1419                 change_nexthops(fi) {
1420                         if (!(nexthop_nh->nh_flags & nh_flags)) {
1421                                 alive++;
1422                                 continue;
1423                         }
1424                         if (!nexthop_nh->nh_dev ||
1425                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1426                                 continue;
1427                         if (nexthop_nh->nh_dev != dev ||
1428                             !__in_dev_get_rtnl(dev))
1429                                 continue;
1430                         alive++;
1431 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1432                         spin_lock_bh(&fib_multipath_lock);
1433                         nexthop_nh->nh_power = 0;
1434                         nexthop_nh->nh_flags &= ~nh_flags;
1435                         spin_unlock_bh(&fib_multipath_lock);
1436 #else
1437                         nexthop_nh->nh_flags &= ~nh_flags;
1438 #endif
1439                 } endfor_nexthops(fi)
1440
1441                 if (alive > 0) {
1442                         fi->fib_flags &= ~nh_flags;
1443                         ret++;
1444                 }
1445         }
1446
1447         return ret;
1448 }
1449
1450 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1451
1452 /*
1453  * The algorithm is suboptimal, but it provides really
1454  * fair weighted route distribution.
1455  */
1456 void fib_select_multipath(struct fib_result *res)
1457 {
1458         struct fib_info *fi = res->fi;
1459         struct in_device *in_dev;
1460         int w;
1461
1462         spin_lock_bh(&fib_multipath_lock);
1463         if (fi->fib_power <= 0) {
1464                 int power = 0;
1465                 change_nexthops(fi) {
1466                         in_dev = __in_dev_get_rcu(nexthop_nh->nh_dev);
1467                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1468                                 continue;
1469                         if (in_dev &&
1470                             IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1471                             nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1472                                 continue;
1473                         power += nexthop_nh->nh_weight;
1474                         nexthop_nh->nh_power = nexthop_nh->nh_weight;
1475                 } endfor_nexthops(fi);
1476                 fi->fib_power = power;
1477                 if (power <= 0) {
1478                         spin_unlock_bh(&fib_multipath_lock);
1479                         /* Race condition: route has just become dead. */
1480                         res->nh_sel = 0;
1481                         return;
1482                 }
1483         }
1484
1485
1486         /* w should be random number [0..fi->fib_power-1],
1487          * it is pretty bad approximation.
1488          */
1489
1490         w = jiffies % fi->fib_power;
1491
1492         change_nexthops(fi) {
1493                 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1494                     nexthop_nh->nh_power) {
1495                         w -= nexthop_nh->nh_power;
1496                         if (w <= 0) {
1497                                 nexthop_nh->nh_power--;
1498                                 fi->fib_power--;
1499                                 res->nh_sel = nhsel;
1500                                 spin_unlock_bh(&fib_multipath_lock);
1501                                 return;
1502                         }
1503                 }
1504         } endfor_nexthops(fi);
1505
1506         /* Race condition: route has just become dead. */
1507         res->nh_sel = 0;
1508         spin_unlock_bh(&fib_multipath_lock);
1509 }
1510 #endif