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 fib_table *tbl = NULL;
 695                         struct flowi4 fl4 = {
 696                                 .daddr = nh->nh_gw,
 697                                 .flowi4_scope = cfg->fc_scope + 1,
 698                                 .flowi4_oif = nh->nh_oif,
 699                                 .flowi4_iif = LOOPBACK_IFINDEX,
 700                         };
 701
 702                         /* It is not necessary, but requires a bit of thinking */
 703                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
 704                                 fl4.flowi4_scope = RT_SCOPE_LINK;
 705
 706                         if (cfg->fc_table)
 707                                 tbl = fib_get_table(net, cfg->fc_table);
 708
 709                         if (tbl)
 710                                 err = fib_table_lookup(tbl, &fl4, &res,
 711                                                        FIB_LOOKUP_IGNORE_LINKSTATE |
 712                                                        FIB_LOOKUP_NOREF);
 713                         else
 714                                 err = fib_lookup(net, &fl4, &res,
 715                                                  FIB_LOOKUP_IGNORE_LINKSTATE);
 716                         if (err) {
 717                                 rcu_read_unlock();
 718                                 return err;
 719                         }
 720                 }
 721                 err = -EINVAL;
 722                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 723                         goto out;
 724                 nh->nh_scope = res.scope;
 725                 nh->nh_oif = FIB_RES_OIF(res);
 726                 nh->nh_dev = dev = FIB_RES_DEV(res);
 727                 if (!dev)
 728                         goto out;
 729                 dev_hold(dev);
 730                 if (!netif_carrier_ok(dev))
 731                         nh->nh_flags |= RTNH_F_LINKDOWN;
 732                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 733         } else {
 734                 struct in_device *in_dev;
 735
 736                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 737                         return -EINVAL;
 738
 739                 rcu_read_lock();
 740                 err = -ENODEV;
 741                 in_dev = inetdev_by_index(net, nh->nh_oif);
 742                 if (!in_dev)
 743                         goto out;
 744                 err = -ENETDOWN;
 745                 if (!(in_dev->dev->flags & IFF_UP))
 746                         goto out;
 747                 nh->nh_dev = in_dev->dev;
 748                 dev_hold(nh->nh_dev);
 749                 nh->nh_scope = RT_SCOPE_HOST;
 750                 if (!netif_carrier_ok(nh->nh_dev))
 751                         nh->nh_flags |= RTNH_F_LINKDOWN;
 752                 err = 0;
 753         }
 754 out:
 755         rcu_read_unlock();
 756         return err;
 757 }
 758
 759 static inline unsigned int fib_laddr_hashfn(__be32 val)
 760 {
 761         unsigned int mask = (fib_info_hash_size - 1);
 762
 763         return ((__force u32)val ^
 764                 ((__force u32)val >> 7) ^
 765                 ((__force u32)val >> 14)) & mask;
 766 }
 767
 768 static struct hlist_head *fib_info_hash_alloc(int bytes)
 769 {
 770         if (bytes <= PAGE_SIZE)
 771                 return kzalloc(bytes, GFP_KERNEL);
 772         else
 773                 return (struct hlist_head *)
 774                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 775                                          get_order(bytes));
 776 }
 777
 778 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 779 {
 780         if (!hash)
 781                 return;
 782
 783         if (bytes <= PAGE_SIZE)
 784                 kfree(hash);
 785         else
 786                 free_pages((unsigned long) hash, get_order(bytes));
 787 }
 788
 789 static void fib_info_hash_move(struct hlist_head *new_info_hash,
 790                                struct hlist_head *new_laddrhash,
 791                                unsigned int new_size)
 792 {
 793         struct hlist_head *old_info_hash, *old_laddrhash;
 794         unsigned int old_size = fib_info_hash_size;
 795         unsigned int i, bytes;
 796
 797         spin_lock_bh(&fib_info_lock);
 798         old_info_hash = fib_info_hash;
 799         old_laddrhash = fib_info_laddrhash;
 800         fib_info_hash_size = new_size;
 801
 802         for (i = 0; i < old_size; i++) {
 803                 struct hlist_head *head = &fib_info_hash[i];
 804                 struct hlist_node *n;
 805                 struct fib_info *fi;
 806
 807                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
 808                         struct hlist_head *dest;
 809                         unsigned int new_hash;
 810
 811                         new_hash = fib_info_hashfn(fi);
 812                         dest = &new_info_hash[new_hash];
 813                         hlist_add_head(&fi->fib_hash, dest);
 814                 }
 815         }
 816         fib_info_hash = new_info_hash;
 817
 818         for (i = 0; i < old_size; i++) {
 819                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 820                 struct hlist_node *n;
 821                 struct fib_info *fi;
 822
 823                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
 824                         struct hlist_head *ldest;
 825                         unsigned int new_hash;
 826
 827                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 828                         ldest = &new_laddrhash[new_hash];
 829                         hlist_add_head(&fi->fib_lhash, ldest);
 830                 }
 831         }
 832         fib_info_laddrhash = new_laddrhash;
 833
 834         spin_unlock_bh(&fib_info_lock);
 835
 836         bytes = old_size * sizeof(struct hlist_head *);
 837         fib_info_hash_free(old_info_hash, bytes);
 838         fib_info_hash_free(old_laddrhash, bytes);
 839 }
 840
 841 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 842 {
 843         nh->nh_saddr = inet_select_addr(nh->nh_dev,
 844                                         nh->nh_gw,
 845                                         nh->nh_parent->fib_scope);
 846         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 847
 848         return nh->nh_saddr;
 849 }
 850
 851 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
 852 {
 853         if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 854             fib_prefsrc != cfg->fc_dst) {
 855                 int tb_id = cfg->fc_table;
 856
 857                 if (tb_id == RT_TABLE_MAIN)
 858                         tb_id = RT_TABLE_LOCAL;
 859
 860                 if (inet_addr_type_table(cfg->fc_nlinfo.nl_net,
 861                                          fib_prefsrc, tb_id) != RTN_LOCAL) {
 862                         return false;
 863                 }
 864         }
 865         return true;
 866 }
 867
 868 struct fib_info *fib_create_info(struct fib_config *cfg)
 869 {
 870         int err;
 871         struct fib_info *fi = NULL;
 872         struct fib_info *ofi;
 873         int nhs = 1;
 874         struct net *net = cfg->fc_nlinfo.nl_net;
 875
 876         if (cfg->fc_type > RTN_MAX)
 877                 goto err_inval;
 878
 879         /* Fast check to catch the most weird cases */
 880         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
 881                 goto err_inval;
 882
 883 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 884         if (cfg->fc_mp) {
 885                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
 886                 if (nhs == 0)
 887                         goto err_inval;
 888         }
 889 #endif
 890
 891         err = -ENOBUFS;
 892         if (fib_info_cnt >= fib_info_hash_size) {
 893                 unsigned int new_size = fib_info_hash_size << 1;
 894                 struct hlist_head *new_info_hash;
 895                 struct hlist_head *new_laddrhash;
 896                 unsigned int bytes;
 897
 898                 if (!new_size)
 899                         new_size = 16;
 900                 bytes = new_size * sizeof(struct hlist_head *);
 901                 new_info_hash = fib_info_hash_alloc(bytes);
 902                 new_laddrhash = fib_info_hash_alloc(bytes);
 903                 if (!new_info_hash || !new_laddrhash) {
 904                         fib_info_hash_free(new_info_hash, bytes);
 905                         fib_info_hash_free(new_laddrhash, bytes);
 906                 } else
 907                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
 908
 909                 if (!fib_info_hash_size)
 910                         goto failure;
 911         }
 912
 913         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
 914         if (!fi)
 915                 goto failure;
 916         fib_info_cnt++;
 917         if (cfg->fc_mx) {
 918                 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
 919                 if (!fi->fib_metrics)
 920                         goto failure;
 921         } else
 922                 fi->fib_metrics = (u32 *) dst_default_metrics;
 923
 924         fi->fib_net = net;
 925         fi->fib_protocol = cfg->fc_protocol;
 926         fi->fib_scope = cfg->fc_scope;
 927         fi->fib_flags = cfg->fc_flags;
 928         fi->fib_priority = cfg->fc_priority;
 929         fi->fib_prefsrc = cfg->fc_prefsrc;
 930         fi->fib_type = cfg->fc_type;
 931
 932         fi->fib_nhs = nhs;
 933         change_nexthops(fi) {
 934                 nexthop_nh->nh_parent = fi;
 935                 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
 936                 if (!nexthop_nh->nh_pcpu_rth_output)
 937                         goto failure;
 938         } endfor_nexthops(fi)
 939
 940         if (cfg->fc_mx) {
 941                 struct nlattr *nla;
 942                 int remaining;
 943
 944                 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 945                         int type = nla_type(nla);
 946
 947                         if (type) {
 948                                 u32 val;
 949
 950                                 if (type > RTAX_MAX)
 951                                         goto err_inval;
 952                                 if (type == RTAX_CC_ALGO) {
 953                                         char tmp[TCP_CA_NAME_MAX];
 954
 955                                         nla_strlcpy(tmp, nla, sizeof(tmp));
 956                                         val = tcp_ca_get_key_by_name(tmp);
 957                                         if (val == TCP_CA_UNSPEC)
 958                                                 goto err_inval;
 959                                 } else {
 960                                         val = nla_get_u32(nla);
 961                                 }
 962                                 if (type == RTAX_ADVMSS && val > 65535 - 40)
 963                                         val = 65535 - 40;
 964                                 if (type == RTAX_MTU && val > 65535 - 15)
 965                                         val = 65535 - 15;
 966                                 fi->fib_metrics[type - 1] = val;
 967                         }
 968                 }
 969         }
 970
 971         if (cfg->fc_mp) {
 972 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 973                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
 974                 if (err != 0)
 975                         goto failure;
 976                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
 977                         goto err_inval;
 978                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
 979                         goto err_inval;
 980 #ifdef CONFIG_IP_ROUTE_CLASSID
 981                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
 982                         goto err_inval;
 983 #endif
 984 #else
 985                 goto err_inval;
 986 #endif
 987         } else {
 988                 struct fib_nh *nh = fi->fib_nh;
 989
 990                 if (cfg->fc_encap) {
 991                         struct lwtunnel_state *lwtstate;
 992                         struct net_device *dev = NULL;
 993
 994                         if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
 995                                 goto err_inval;
 996                         if (cfg->fc_oif)
 997                                 dev = __dev_get_by_index(net, cfg->fc_oif);
 998                         err = lwtunnel_build_state(dev, cfg->fc_encap_type,
 999                                                    cfg->fc_encap, &lwtstate);
1000                         if (err)
1001                                 goto failure;
1002
1003                         nh->nh_lwtstate = lwtstate_get(lwtstate);
1004                 }
1005                 nh->nh_oif = cfg->fc_oif;
1006                 nh->nh_gw = cfg->fc_gw;
1007                 nh->nh_flags = cfg->fc_flags;
1008 #ifdef CONFIG_IP_ROUTE_CLASSID
1009                 nh->nh_tclassid = cfg->fc_flow;
1010                 if (nh->nh_tclassid)
1011                         fi->fib_net->ipv4.fib_num_tclassid_users++;
1012 #endif
1013 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1014                 nh->nh_weight = 1;
1015 #endif
1016         }
1017
1018         if (fib_props[cfg->fc_type].error) {
1019                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
1020                         goto err_inval;
1021                 goto link_it;
1022         } else {
1023                 switch (cfg->fc_type) {
1024                 case RTN_UNICAST:
1025                 case RTN_LOCAL:
1026                 case RTN_BROADCAST:
1027                 case RTN_ANYCAST:
1028                 case RTN_MULTICAST:
1029                         break;
1030                 default:
1031                         goto err_inval;
1032                 }
1033         }
1034
1035         if (cfg->fc_scope > RT_SCOPE_HOST)
1036                 goto err_inval;
1037
1038         if (cfg->fc_scope == RT_SCOPE_HOST) {
1039                 struct fib_nh *nh = fi->fib_nh;
1040
1041                 /* Local address is added. */
1042                 if (nhs != 1 || nh->nh_gw)
1043                         goto err_inval;
1044                 nh->nh_scope = RT_SCOPE_NOWHERE;
1045                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1046                 err = -ENODEV;
1047                 if (!nh->nh_dev)
1048                         goto failure;
1049         } else {
1050                 int linkdown = 0;
1051
1052                 change_nexthops(fi) {
1053                         err = fib_check_nh(cfg, fi, nexthop_nh);
1054                         if (err != 0)
1055                                 goto failure;
1056                         if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1057                                 linkdown++;
1058                 } endfor_nexthops(fi)
1059                 if (linkdown == fi->fib_nhs)
1060                         fi->fib_flags |= RTNH_F_LINKDOWN;
1061         }
1062
1063         if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc))
1064                 goto err_inval;
1065
1066         change_nexthops(fi) {
1067                 fib_info_update_nh_saddr(net, nexthop_nh);
1068         } endfor_nexthops(fi)
1069
1070 link_it:
1071         ofi = fib_find_info(fi);
1072         if (ofi) {
1073                 fi->fib_dead = 1;
1074                 free_fib_info(fi);
1075                 ofi->fib_treeref++;
1076                 return ofi;
1077         }
1078
1079         fi->fib_treeref++;
1080         atomic_inc(&fi->fib_clntref);
1081         spin_lock_bh(&fib_info_lock);
1082         hlist_add_head(&fi->fib_hash,
1083                        &fib_info_hash[fib_info_hashfn(fi)]);
1084         if (fi->fib_prefsrc) {
1085                 struct hlist_head *head;
1086
1087                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1088                 hlist_add_head(&fi->fib_lhash, head);
1089         }
1090         change_nexthops(fi) {
1091                 struct hlist_head *head;
1092                 unsigned int hash;
1093
1094                 if (!nexthop_nh->nh_dev)
1095                         continue;
1096                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1097                 head = &fib_info_devhash[hash];
1098                 hlist_add_head(&nexthop_nh->nh_hash, head);
1099         } endfor_nexthops(fi)
1100         spin_unlock_bh(&fib_info_lock);
1101         return fi;
1102
1103 err_inval:
1104         err = -EINVAL;
1105
1106 failure:
1107         if (fi) {
1108                 fi->fib_dead = 1;
1109                 free_fib_info(fi);
1110         }
1111
1112         return ERR_PTR(err);
1113 }
1114
1115 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1116                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1117                   struct fib_info *fi, unsigned int flags)
1118 {
1119         struct nlmsghdr *nlh;
1120         struct rtmsg *rtm;
1121
1122         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1123         if (!nlh)
1124                 return -EMSGSIZE;
1125
1126         rtm = nlmsg_data(nlh);
1127         rtm->rtm_family = AF_INET;
1128         rtm->rtm_dst_len = dst_len;
1129         rtm->rtm_src_len = 0;
1130         rtm->rtm_tos = tos;
1131         if (tb_id < 256)
1132                 rtm->rtm_table = tb_id;
1133         else
1134                 rtm->rtm_table = RT_TABLE_COMPAT;
1135         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1136                 goto nla_put_failure;
1137         rtm->rtm_type = type;
1138         rtm->rtm_flags = fi->fib_flags;
1139         rtm->rtm_scope = fi->fib_scope;
1140         rtm->rtm_protocol = fi->fib_protocol;
1141
1142         if (rtm->rtm_dst_len &&
1143             nla_put_in_addr(skb, RTA_DST, dst))
1144                 goto nla_put_failure;
1145         if (fi->fib_priority &&
1146             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1147                 goto nla_put_failure;
1148         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1149                 goto nla_put_failure;
1150
1151         if (fi->fib_prefsrc &&
1152             nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1153                 goto nla_put_failure;
1154         if (fi->fib_nhs == 1) {
1155                 struct in_device *in_dev;
1156
1157                 if (fi->fib_nh->nh_gw &&
1158                     nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1159                         goto nla_put_failure;
1160                 if (fi->fib_nh->nh_oif &&
1161                     nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1162                         goto nla_put_failure;
1163                 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1164                         in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1165                         if (in_dev &&
1166                             IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1167                                 rtm->rtm_flags |= RTNH_F_DEAD;
1168                 }
1169 #ifdef CONFIG_IP_ROUTE_CLASSID
1170                 if (fi->fib_nh[0].nh_tclassid &&
1171                     nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1172                         goto nla_put_failure;
1173 #endif
1174                 if (fi->fib_nh->nh_lwtstate)
1175                         lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate);
1176         }
1177 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1178         if (fi->fib_nhs > 1) {
1179                 struct rtnexthop *rtnh;
1180                 struct nlattr *mp;
1181
1182                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1183                 if (!mp)
1184                         goto nla_put_failure;
1185
1186                 for_nexthops(fi) {
1187                         struct in_device *in_dev;
1188
1189                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1190                         if (!rtnh)
1191                                 goto nla_put_failure;
1192
1193                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1194                         if (nh->nh_flags & RTNH_F_LINKDOWN) {
1195                                 in_dev = __in_dev_get_rtnl(nh->nh_dev);
1196                                 if (in_dev &&
1197                                     IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1198                                         rtnh->rtnh_flags |= RTNH_F_DEAD;
1199                         }
1200                         rtnh->rtnh_hops = nh->nh_weight - 1;
1201                         rtnh->rtnh_ifindex = nh->nh_oif;
1202
1203                         if (nh->nh_gw &&
1204                             nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1205                                 goto nla_put_failure;
1206 #ifdef CONFIG_IP_ROUTE_CLASSID
1207                         if (nh->nh_tclassid &&
1208                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1209                                 goto nla_put_failure;
1210 #endif
1211                         if (nh->nh_lwtstate)
1212                                 lwtunnel_fill_encap(skb, nh->nh_lwtstate);
1213                         /* length of rtnetlink header + attributes */
1214                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1215                 } endfor_nexthops(fi);
1216
1217                 nla_nest_end(skb, mp);
1218         }
1219 #endif
1220         nlmsg_end(skb, nlh);
1221         return 0;
1222
1223 nla_put_failure:
1224         nlmsg_cancel(skb, nlh);
1225         return -EMSGSIZE;
1226 }
1227
1228 /*
1229  * Update FIB if:
1230  * - local address disappeared -> we must delete all the entries
1231  *   referring to it.
1232  * - device went down -> we must shutdown all nexthops going via it.
1233  */
1234 int fib_sync_down_addr(struct net *net, __be32 local)
1235 {
1236         int ret = 0;
1237         unsigned int hash = fib_laddr_hashfn(local);
1238         struct hlist_head *head = &fib_info_laddrhash[hash];
1239         struct fib_info *fi;
1240
1241         if (!fib_info_laddrhash || local == 0)
1242                 return 0;
1243
1244         hlist_for_each_entry(fi, head, fib_lhash) {
1245                 if (!net_eq(fi->fib_net, net))
1246                         continue;
1247                 if (fi->fib_prefsrc == local) {
1248                         fi->fib_flags |= RTNH_F_DEAD;
1249                         ret++;
1250                 }
1251         }
1252         return ret;
1253 }
1254
1255 int fib_sync_down_dev(struct net_device *dev, unsigned long event)
1256 {
1257         int ret = 0;
1258         int scope = RT_SCOPE_NOWHERE;
1259         struct fib_info *prev_fi = NULL;
1260         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1261         struct hlist_head *head = &fib_info_devhash[hash];
1262         struct fib_nh *nh;
1263
1264         if (event == NETDEV_UNREGISTER ||
1265             event == NETDEV_DOWN)
1266                 scope = -1;
1267
1268         hlist_for_each_entry(nh, head, nh_hash) {
1269                 struct fib_info *fi = nh->nh_parent;
1270                 int dead;
1271
1272                 BUG_ON(!fi->fib_nhs);
1273                 if (nh->nh_dev != dev || fi == prev_fi)
1274                         continue;
1275                 prev_fi = fi;
1276                 dead = 0;
1277                 change_nexthops(fi) {
1278                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1279                                 dead++;
1280                         else if (nexthop_nh->nh_dev == dev &&
1281                                  nexthop_nh->nh_scope != scope) {
1282                                 switch (event) {
1283                                 case NETDEV_DOWN:
1284                                 case NETDEV_UNREGISTER:
1285                                         nexthop_nh->nh_flags |= RTNH_F_DEAD;
1286                                         /* fall through */
1287                                 case NETDEV_CHANGE:
1288                                         nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1289                                         break;
1290                                 }
1291 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1292                                 spin_lock_bh(&fib_multipath_lock);
1293                                 fi->fib_power -= nexthop_nh->nh_power;
1294                                 nexthop_nh->nh_power = 0;
1295                                 spin_unlock_bh(&fib_multipath_lock);
1296 #endif
1297                                 dead++;
1298                         }
1299 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1300                         if (event == NETDEV_UNREGISTER &&
1301                             nexthop_nh->nh_dev == dev) {
1302                                 dead = fi->fib_nhs;
1303                                 break;
1304                         }
1305 #endif
1306                 } endfor_nexthops(fi)
1307                 if (dead == fi->fib_nhs) {
1308                         switch (event) {
1309                         case NETDEV_DOWN:
1310                         case NETDEV_UNREGISTER:
1311                                 fi->fib_flags |= RTNH_F_DEAD;
1312                                 /* fall through */
1313                         case NETDEV_CHANGE:
1314                                 fi->fib_flags |= RTNH_F_LINKDOWN;
1315                                 break;
1316                         }
1317                         ret++;
1318                 }
1319         }
1320
1321         return ret;
1322 }
1323
1324 /* Must be invoked inside of an RCU protected region.  */
1325 void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1326 {
1327         struct fib_info *fi = NULL, *last_resort = NULL;
1328         struct hlist_head *fa_head = res->fa_head;
1329         struct fib_table *tb = res->table;
1330         u8 slen = 32 - res->prefixlen;
1331         int order = -1, last_idx = -1;
1332         struct fib_alias *fa, *fa1 = NULL;
1333         u32 last_prio = res->fi->fib_priority;
1334         u8 last_tos = 0;
1335
1336         hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1337                 struct fib_info *next_fi = fa->fa_info;
1338
1339                 if (fa->fa_slen != slen)
1340                         continue;
1341                 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1342                         continue;
1343                 if (fa->tb_id != tb->tb_id)
1344                         continue;
1345                 if (next_fi->fib_priority > last_prio &&
1346                     fa->fa_tos == last_tos) {
1347                         if (last_tos)
1348                                 continue;
1349                         break;
1350                 }
1351                 if (next_fi->fib_flags & RTNH_F_DEAD)
1352                         continue;
1353                 last_tos = fa->fa_tos;
1354                 last_prio = next_fi->fib_priority;
1355
1356                 if (next_fi->fib_scope != res->scope ||
1357                     fa->fa_type != RTN_UNICAST)
1358                         continue;
1359                 if (!next_fi->fib_nh[0].nh_gw ||
1360                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1361                         continue;
1362
1363                 fib_alias_accessed(fa);
1364
1365                 if (!fi) {
1366                         if (next_fi != res->fi)
1367                                 break;
1368                         fa1 = fa;
1369                 } else if (!fib_detect_death(fi, order, &last_resort,
1370                                              &last_idx, fa1->fa_default)) {
1371                         fib_result_assign(res, fi);
1372                         fa1->fa_default = order;
1373                         goto out;
1374                 }
1375                 fi = next_fi;
1376                 order++;
1377         }
1378
1379         if (order <= 0 || !fi) {
1380                 if (fa1)
1381                         fa1->fa_default = -1;
1382                 goto out;
1383         }
1384
1385         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1386                               fa1->fa_default)) {
1387                 fib_result_assign(res, fi);
1388                 fa1->fa_default = order;
1389                 goto out;
1390         }
1391
1392         if (last_idx >= 0)
1393                 fib_result_assign(res, last_resort);
1394         fa1->fa_default = last_idx;
1395 out:
1396         return;
1397 }
1398
1399 /*
1400  * Dead device goes up. We wake up dead nexthops.
1401  * It takes sense only on multipath routes.
1402  */
1403 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1404 {
1405         struct fib_info *prev_fi;
1406         unsigned int hash;
1407         struct hlist_head *head;
1408         struct fib_nh *nh;
1409         int ret;
1410
1411         if (!(dev->flags & IFF_UP))
1412                 return 0;
1413
1414         prev_fi = NULL;
1415         hash = fib_devindex_hashfn(dev->ifindex);
1416         head = &fib_info_devhash[hash];
1417         ret = 0;
1418
1419         hlist_for_each_entry(nh, head, nh_hash) {
1420                 struct fib_info *fi = nh->nh_parent;
1421                 int alive;
1422
1423                 BUG_ON(!fi->fib_nhs);
1424                 if (nh->nh_dev != dev || fi == prev_fi)
1425                         continue;
1426
1427                 prev_fi = fi;
1428                 alive = 0;
1429                 change_nexthops(fi) {
1430                         if (!(nexthop_nh->nh_flags & nh_flags)) {
1431                                 alive++;
1432                                 continue;
1433                         }
1434                         if (!nexthop_nh->nh_dev ||
1435                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1436                                 continue;
1437                         if (nexthop_nh->nh_dev != dev ||
1438                             !__in_dev_get_rtnl(dev))
1439                                 continue;
1440                         alive++;
1441 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1442                         spin_lock_bh(&fib_multipath_lock);
1443                         nexthop_nh->nh_power = 0;
1444                         nexthop_nh->nh_flags &= ~nh_flags;
1445                         spin_unlock_bh(&fib_multipath_lock);
1446 #else
1447                         nexthop_nh->nh_flags &= ~nh_flags;
1448 #endif
1449                 } endfor_nexthops(fi)
1450
1451                 if (alive > 0) {
1452                         fi->fib_flags &= ~nh_flags;
1453                         ret++;
1454                 }
1455         }
1456
1457         return ret;
1458 }
1459
1460 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1461
1462 /*
1463  * The algorithm is suboptimal, but it provides really
1464  * fair weighted route distribution.
1465  */
1466 void fib_select_multipath(struct fib_result *res)
1467 {
1468         struct fib_info *fi = res->fi;
1469         struct in_device *in_dev;
1470         int w;
1471
1472         spin_lock_bh(&fib_multipath_lock);
1473         if (fi->fib_power <= 0) {
1474                 int power = 0;
1475                 change_nexthops(fi) {
1476                         in_dev = __in_dev_get_rcu(nexthop_nh->nh_dev);
1477                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1478                                 continue;
1479                         if (in_dev &&
1480                             IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1481                             nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1482                                 continue;
1483                         power += nexthop_nh->nh_weight;
1484                         nexthop_nh->nh_power = nexthop_nh->nh_weight;
1485                 } endfor_nexthops(fi);
1486                 fi->fib_power = power;
1487                 if (power <= 0) {
1488                         spin_unlock_bh(&fib_multipath_lock);
1489                         /* Race condition: route has just become dead. */
1490                         res->nh_sel = 0;
1491                         return;
1492                 }
1493         }
1494
1495
1496         /* w should be random number [0..fi->fib_power-1],
1497          * it is pretty bad approximation.
1498          */
1499
1500         w = jiffies % fi->fib_power;
1501
1502         change_nexthops(fi) {
1503                 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1504                     nexthop_nh->nh_power) {
1505                         w -= nexthop_nh->nh_power;
1506                         if (w <= 0) {
1507                                 nexthop_nh->nh_power--;
1508                                 fi->fib_power--;
1509                                 res->nh_sel = nhsel;
1510                                 spin_unlock_bh(&fib_multipath_lock);
1511                                 return;
1512                         }
1513                 }
1514         } endfor_nexthops(fi);
1515
1516         /* Race condition: route has just become dead. */
1517         res->nh_sel = 0;
1518         spin_unlock_bh(&fib_multipath_lock);
1519 }
1520 #endif