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 u32 fib_multipath_secret __read_mostly;
  61
  62 #define for_nexthops(fi) {                                              \
  63         int nhsel; const struct fib_nh *nh;                             \
  64         for (nhsel = 0, nh = (fi)->fib_nh;                              \
  65              nhsel < (fi)->fib_nhs;                                     \
  66              nh++, nhsel++)
  67
  68 #define change_nexthops(fi) {                                           \
  69         int nhsel; struct fib_nh *nexthop_nh;                           \
  70         for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh);   \
  71              nhsel < (fi)->fib_nhs;                                     \
  72              nexthop_nh++, nhsel++)
  73
  74 #else /* CONFIG_IP_ROUTE_MULTIPATH */
  75
  76 /* Hope, that gcc will optimize it to get rid of dummy loop */
  77
  78 #define for_nexthops(fi) {                                              \
  79         int nhsel; const struct fib_nh *nh = (fi)->fib_nh;              \
  80         for (nhsel = 0; nhsel < 1; nhsel++)
  81
  82 #define change_nexthops(fi) {                                           \
  83         int nhsel;                                                      \
  84         struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);    \
  85         for (nhsel = 0; nhsel < 1; nhsel++)
  86
  87 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
  88
  89 #define endfor_nexthops(fi) }
  90
  91
  92 const struct fib_prop fib_props[RTN_MAX + 1] = {
  93         [RTN_UNSPEC] = {
  94                 .error  = 0,
  95                 .scope  = RT_SCOPE_NOWHERE,
  96         },
  97         [RTN_UNICAST] = {
  98                 .error  = 0,
  99                 .scope  = RT_SCOPE_UNIVERSE,
 100         },
 101         [RTN_LOCAL] = {
 102                 .error  = 0,
 103                 .scope  = RT_SCOPE_HOST,
 104         },
 105         [RTN_BROADCAST] = {
 106                 .error  = 0,
 107                 .scope  = RT_SCOPE_LINK,
 108         },
 109         [RTN_ANYCAST] = {
 110                 .error  = 0,
 111                 .scope  = RT_SCOPE_LINK,
 112         },
 113         [RTN_MULTICAST] = {
 114                 .error  = 0,
 115                 .scope  = RT_SCOPE_UNIVERSE,
 116         },
 117         [RTN_BLACKHOLE] = {
 118                 .error  = -EINVAL,
 119                 .scope  = RT_SCOPE_UNIVERSE,
 120         },
 121         [RTN_UNREACHABLE] = {
 122                 .error  = -EHOSTUNREACH,
 123                 .scope  = RT_SCOPE_UNIVERSE,
 124         },
 125         [RTN_PROHIBIT] = {
 126                 .error  = -EACCES,
 127                 .scope  = RT_SCOPE_UNIVERSE,
 128         },
 129         [RTN_THROW] = {
 130                 .error  = -EAGAIN,
 131                 .scope  = RT_SCOPE_UNIVERSE,
 132         },
 133         [RTN_NAT] = {
 134                 .error  = -EINVAL,
 135                 .scope  = RT_SCOPE_NOWHERE,
 136         },
 137         [RTN_XRESOLVE] = {
 138                 .error  = -EINVAL,
 139                 .scope  = RT_SCOPE_NOWHERE,
 140         },
 141 };
 142
 143 static void rt_fibinfo_free(struct rtable __rcu **rtp)
 144 {
 145         struct rtable *rt = rcu_dereference_protected(*rtp, 1);
 146
 147         if (!rt)
 148                 return;
 149
 150         /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
 151          * because we waited an RCU grace period before calling
 152          * free_fib_info_rcu()
 153          */
 154
 155         dst_free(&rt->dst);
 156 }
 157
 158 static void free_nh_exceptions(struct fib_nh *nh)
 159 {
 160         struct fnhe_hash_bucket *hash;
 161         int i;
 162
 163         hash = rcu_dereference_protected(nh->nh_exceptions, 1);
 164         if (!hash)
 165                 return;
 166         for (i = 0; i < FNHE_HASH_SIZE; i++) {
 167                 struct fib_nh_exception *fnhe;
 168
 169                 fnhe = rcu_dereference_protected(hash[i].chain, 1);
 170                 while (fnhe) {
 171                         struct fib_nh_exception *next;
 172
 173                         next = rcu_dereference_protected(fnhe->fnhe_next, 1);
 174
 175                         rt_fibinfo_free(&fnhe->fnhe_rth_input);
 176                         rt_fibinfo_free(&fnhe->fnhe_rth_output);
 177
 178                         kfree(fnhe);
 179
 180                         fnhe = next;
 181                 }
 182         }
 183         kfree(hash);
 184 }
 185
 186 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
 187 {
 188         int cpu;
 189
 190         if (!rtp)
 191                 return;
 192
 193         for_each_possible_cpu(cpu) {
 194                 struct rtable *rt;
 195
 196                 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
 197                 if (rt)
 198                         dst_free(&rt->dst);
 199         }
 200         free_percpu(rtp);
 201 }
 202
 203 /* Release a nexthop info record */
 204 static void free_fib_info_rcu(struct rcu_head *head)
 205 {
 206         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 207
 208         change_nexthops(fi) {
 209                 if (nexthop_nh->nh_dev)
 210                         dev_put(nexthop_nh->nh_dev);
 211                 lwtstate_put(nexthop_nh->nh_lwtstate);
 212                 free_nh_exceptions(nexthop_nh);
 213                 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
 214                 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
 215         } endfor_nexthops(fi);
 216
 217         if (fi->fib_metrics != (u32 *) dst_default_metrics)
 218                 kfree(fi->fib_metrics);
 219         kfree(fi);
 220 }
 221
 222 void free_fib_info(struct fib_info *fi)
 223 {
 224         if (fi->fib_dead == 0) {
 225                 pr_warn("Freeing alive fib_info %p\n", fi);
 226                 return;
 227         }
 228         fib_info_cnt--;
 229 #ifdef CONFIG_IP_ROUTE_CLASSID
 230         change_nexthops(fi) {
 231                 if (nexthop_nh->nh_tclassid)
 232                         fi->fib_net->ipv4.fib_num_tclassid_users--;
 233         } endfor_nexthops(fi);
 234 #endif
 235         call_rcu(&fi->rcu, free_fib_info_rcu);
 236 }
 237
 238 void fib_release_info(struct fib_info *fi)
 239 {
 240         spin_lock_bh(&fib_info_lock);
 241         if (fi && --fi->fib_treeref == 0) {
 242                 hlist_del(&fi->fib_hash);
 243                 if (fi->fib_prefsrc)
 244                         hlist_del(&fi->fib_lhash);
 245                 change_nexthops(fi) {
 246                         if (!nexthop_nh->nh_dev)
 247                                 continue;
 248                         hlist_del(&nexthop_nh->nh_hash);
 249                 } endfor_nexthops(fi)
 250                 fi->fib_dead = 1;
 251                 fib_info_put(fi);
 252         }
 253         spin_unlock_bh(&fib_info_lock);
 254 }
 255
 256 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
 257 {
 258         const struct fib_nh *onh = ofi->fib_nh;
 259
 260         for_nexthops(fi) {
 261                 if (nh->nh_oif != onh->nh_oif ||
 262                     nh->nh_gw  != onh->nh_gw ||
 263                     nh->nh_scope != onh->nh_scope ||
 264 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 265                     nh->nh_weight != onh->nh_weight ||
 266 #endif
 267 #ifdef CONFIG_IP_ROUTE_CLASSID
 268                     nh->nh_tclassid != onh->nh_tclassid ||
 269 #endif
 270                     lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
 271                     ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
 272                         return -1;
 273                 onh++;
 274         } endfor_nexthops(fi);
 275         return 0;
 276 }
 277
 278 static inline unsigned int fib_devindex_hashfn(unsigned int val)
 279 {
 280         unsigned int mask = DEVINDEX_HASHSIZE - 1;
 281
 282         return (val ^
 283                 (val >> DEVINDEX_HASHBITS) ^
 284                 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
 285 }
 286
 287 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
 288 {
 289         unsigned int mask = (fib_info_hash_size - 1);
 290         unsigned int val = fi->fib_nhs;
 291
 292         val ^= (fi->fib_protocol << 8) | fi->fib_scope;
 293         val ^= (__force u32)fi->fib_prefsrc;
 294         val ^= fi->fib_priority;
 295         for_nexthops(fi) {
 296                 val ^= fib_devindex_hashfn(nh->nh_oif);
 297         } endfor_nexthops(fi)
 298
 299         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 300 }
 301
 302 static struct fib_info *fib_find_info(const struct fib_info *nfi)
 303 {
 304         struct hlist_head *head;
 305         struct fib_info *fi;
 306         unsigned int hash;
 307
 308         hash = fib_info_hashfn(nfi);
 309         head = &fib_info_hash[hash];
 310
 311         hlist_for_each_entry(fi, head, fib_hash) {
 312                 if (!net_eq(fi->fib_net, nfi->fib_net))
 313                         continue;
 314                 if (fi->fib_nhs != nfi->fib_nhs)
 315                         continue;
 316                 if (nfi->fib_protocol == fi->fib_protocol &&
 317                     nfi->fib_scope == fi->fib_scope &&
 318                     nfi->fib_prefsrc == fi->fib_prefsrc &&
 319                     nfi->fib_priority == fi->fib_priority &&
 320                     nfi->fib_type == fi->fib_type &&
 321                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 322                            sizeof(u32) * RTAX_MAX) == 0 &&
 323                     !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
 324                     (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
 325                         return fi;
 326         }
 327
 328         return NULL;
 329 }
 330
 331 /* Check, that the gateway is already configured.
 332  * Used only by redirect accept routine.
 333  */
 334 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 335 {
 336         struct hlist_head *head;
 337         struct fib_nh *nh;
 338         unsigned int hash;
 339
 340         spin_lock(&fib_info_lock);
 341
 342         hash = fib_devindex_hashfn(dev->ifindex);
 343         head = &fib_info_devhash[hash];
 344         hlist_for_each_entry(nh, head, nh_hash) {
 345                 if (nh->nh_dev == dev &&
 346                     nh->nh_gw == gw &&
 347                     !(nh->nh_flags & RTNH_F_DEAD)) {
 348                         spin_unlock(&fib_info_lock);
 349                         return 0;
 350                 }
 351         }
 352
 353         spin_unlock(&fib_info_lock);
 354
 355         return -1;
 356 }
 357
 358 static inline size_t fib_nlmsg_size(struct fib_info *fi)
 359 {
 360         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 361                          + nla_total_size(4) /* RTA_TABLE */
 362                          + nla_total_size(4) /* RTA_DST */
 363                          + nla_total_size(4) /* RTA_PRIORITY */
 364                          + nla_total_size(4) /* RTA_PREFSRC */
 365                          + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
 366
 367         /* space for nested metrics */
 368         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 369
 370         if (fi->fib_nhs) {
 371                 size_t nh_encapsize = 0;
 372                 /* Also handles the special case fib_nhs == 1 */
 373
 374                 /* each nexthop is packed in an attribute */
 375                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 376
 377                 /* may contain flow and gateway attribute */
 378                 nhsize += 2 * nla_total_size(4);
 379
 380                 /* grab encap info */
 381                 for_nexthops(fi) {
 382                         if (nh->nh_lwtstate) {
 383                                 /* RTA_ENCAP_TYPE */
 384                                 nh_encapsize += lwtunnel_get_encap_size(
 385                                                 nh->nh_lwtstate);
 386                                 /* RTA_ENCAP */
 387                                 nh_encapsize +=  nla_total_size(2);
 388                         }
 389                 } endfor_nexthops(fi);
 390
 391                 /* all nexthops are packed in a nested attribute */
 392                 payload += nla_total_size((fi->fib_nhs * nhsize) +
 393                                           nh_encapsize);
 394
 395         }
 396
 397         return payload;
 398 }
 399
 400 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 401                int dst_len, u32 tb_id, const struct nl_info *info,
 402                unsigned int nlm_flags)
 403 {
 404         struct sk_buff *skb;
 405         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 406         int err = -ENOBUFS;
 407
 408         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 409         if (!skb)
 410                 goto errout;
 411
 412         err = fib_dump_info(skb, info->portid, seq, event, tb_id,
 413                             fa->fa_type, key, dst_len,
 414                             fa->fa_tos, fa->fa_info, nlm_flags);
 415         if (err < 0) {
 416                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 417                 WARN_ON(err == -EMSGSIZE);
 418                 kfree_skb(skb);
 419                 goto errout;
 420         }
 421         rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
 422                     info->nlh, GFP_KERNEL);
 423         return;
 424 errout:
 425         if (err < 0)
 426                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 427 }
 428
 429 static int fib_detect_death(struct fib_info *fi, int order,
 430                             struct fib_info **last_resort, int *last_idx,
 431                             int dflt)
 432 {
 433         struct neighbour *n;
 434         int state = NUD_NONE;
 435
 436         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 437         if (n) {
 438                 state = n->nud_state;
 439                 neigh_release(n);
 440         } else {
 441                 return 0;
 442         }
 443         if (state == NUD_REACHABLE)
 444                 return 0;
 445         if ((state & NUD_VALID) && order != dflt)
 446                 return 0;
 447         if ((state & NUD_VALID) ||
 448             (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
 449                 *last_resort = fi;
 450                 *last_idx = order;
 451         }
 452         return 1;
 453 }
 454
 455 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 456
 457 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
 458 {
 459         int nhs = 0;
 460
 461         while (rtnh_ok(rtnh, remaining)) {
 462                 nhs++;
 463                 rtnh = rtnh_next(rtnh, &remaining);
 464         }
 465
 466         /* leftover implies invalid nexthop configuration, discard it */
 467         return remaining > 0 ? 0 : nhs;
 468 }
 469
 470 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 471                        int remaining, struct fib_config *cfg)
 472 {
 473         struct net *net = cfg->fc_nlinfo.nl_net;
 474         int ret;
 475
 476         change_nexthops(fi) {
 477                 int attrlen;
 478
 479                 if (!rtnh_ok(rtnh, remaining))
 480                         return -EINVAL;
 481
 482                 nexthop_nh->nh_flags =
 483                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 484                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
 485                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
 486
 487                 attrlen = rtnh_attrlen(rtnh);
 488                 if (attrlen > 0) {
 489                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 490
 491                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 492                         nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
 493 #ifdef CONFIG_IP_ROUTE_CLASSID
 494                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 495                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 496                         if (nexthop_nh->nh_tclassid)
 497                                 fi->fib_net->ipv4.fib_num_tclassid_users++;
 498 #endif
 499                         nla = nla_find(attrs, attrlen, RTA_ENCAP);
 500                         if (nla) {
 501                                 struct lwtunnel_state *lwtstate;
 502                                 struct net_device *dev = NULL;
 503                                 struct nlattr *nla_entype;
 504
 505                                 nla_entype = nla_find(attrs, attrlen,
 506                                                       RTA_ENCAP_TYPE);
 507                                 if (!nla_entype)
 508                                         goto err_inval;
 509                                 if (cfg->fc_oif)
 510                                         dev = __dev_get_by_index(net, cfg->fc_oif);
 511                                 ret = lwtunnel_build_state(dev, nla_get_u16(
 512                                                            nla_entype),
 513                                                            nla,  AF_INET, cfg,
 514                                                            &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 static void fib_rebalance(struct fib_info *fi)
 535 {
 536         int total;
 537         int w;
 538         struct in_device *in_dev;
 539
 540         if (fi->fib_nhs < 2)
 541                 return;
 542
 543         total = 0;
 544         for_nexthops(fi) {
 545                 if (nh->nh_flags & RTNH_F_DEAD)
 546                         continue;
 547
 548                 in_dev = __in_dev_get_rtnl(nh->nh_dev);
 549
 550                 if (in_dev &&
 551                     IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
 552                     nh->nh_flags & RTNH_F_LINKDOWN)
 553                         continue;
 554
 555                 total += nh->nh_weight;
 556         } endfor_nexthops(fi);
 557
 558         w = 0;
 559         change_nexthops(fi) {
 560                 int upper_bound;
 561
 562                 in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
 563
 564                 if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
 565                         upper_bound = -1;
 566                 } else if (in_dev &&
 567                            IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
 568                            nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
 569                         upper_bound = -1;
 570                 } else {
 571                         w += nexthop_nh->nh_weight;
 572                         upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
 573                                                             total) - 1;
 574                 }
 575
 576                 atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
 577         } endfor_nexthops(fi);
 578
 579         net_get_random_once(&fib_multipath_secret,
 580                             sizeof(fib_multipath_secret));
 581 }
 582
 583 static inline void fib_add_weight(struct fib_info *fi,
 584                                   const struct fib_nh *nh)
 585 {
 586         fi->fib_weight += nh->nh_weight;
 587 }
 588
 589 #else /* CONFIG_IP_ROUTE_MULTIPATH */
 590
 591 #define fib_rebalance(fi) do { } while (0)
 592 #define fib_add_weight(fi, nh) do { } while (0)
 593
 594 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
 595
 596 static int fib_encap_match(struct net *net, u16 encap_type,
 597                            struct nlattr *encap,
 598                            int oif, const struct fib_nh *nh,
 599                            const struct fib_config *cfg)
 600 {
 601         struct lwtunnel_state *lwtstate;
 602         struct net_device *dev = NULL;
 603         int ret, result = 0;
 604
 605         if (encap_type == LWTUNNEL_ENCAP_NONE)
 606                 return 0;
 607
 608         if (oif)
 609                 dev = __dev_get_by_index(net, oif);
 610         ret = lwtunnel_build_state(dev, encap_type, encap,
 611                                    AF_INET, cfg, &lwtstate);
 612         if (!ret) {
 613                 result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
 614                 lwtstate_free(lwtstate);
 615         }
 616
 617         return result;
 618 }
 619
 620 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
 621 {
 622         struct net *net = cfg->fc_nlinfo.nl_net;
 623 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 624         struct rtnexthop *rtnh;
 625         int remaining;
 626 #endif
 627
 628         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 629                 return 1;
 630
 631         if (cfg->fc_oif || cfg->fc_gw) {
 632                 if (cfg->fc_encap) {
 633                         if (fib_encap_match(net, cfg->fc_encap_type,
 634                                             cfg->fc_encap, cfg->fc_oif,
 635                                             fi->fib_nh, cfg))
 636                             return 1;
 637                 }
 638                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 639                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 640                         return 0;
 641                 return 1;
 642         }
 643
 644 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 645         if (!cfg->fc_mp)
 646                 return 0;
 647
 648         rtnh = cfg->fc_mp;
 649         remaining = cfg->fc_mp_len;
 650
 651         for_nexthops(fi) {
 652                 int attrlen;
 653
 654                 if (!rtnh_ok(rtnh, remaining))
 655                         return -EINVAL;
 656
 657                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 658                         return 1;
 659
 660                 attrlen = rtnh_attrlen(rtnh);
 661                 if (attrlen > 0) {
 662                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 663
 664                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 665                         if (nla && nla_get_in_addr(nla) != nh->nh_gw)
 666                                 return 1;
 667 #ifdef CONFIG_IP_ROUTE_CLASSID
 668                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 669                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 670                                 return 1;
 671 #endif
 672                 }
 673
 674                 rtnh = rtnh_next(rtnh, &remaining);
 675         } endfor_nexthops(fi);
 676 #endif
 677         return 0;
 678 }
 679
 680
 681 /*
 682  * Picture
 683  * -------
 684  *
 685  * Semantics of nexthop is very messy by historical reasons.
 686  * We have to take into account, that:
 687  * a) gateway can be actually local interface address,
 688  *    so that gatewayed route is direct.
 689  * b) gateway must be on-link address, possibly
 690  *    described not by an ifaddr, but also by a direct route.
 691  * c) If both gateway and interface are specified, they should not
 692  *    contradict.
 693  * d) If we use tunnel routes, gateway could be not on-link.
 694  *
 695  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 696  * results in pretty ugly and hairy code with obscure logic.
 697  *
 698  * I chose to generalized it instead, so that the size
 699  * of code does not increase practically, but it becomes
 700  * much more general.
 701  * Every prefix is assigned a "scope" value: "host" is local address,
 702  * "link" is direct route,
 703  * [ ... "site" ... "interior" ... ]
 704  * and "universe" is true gateway route with global meaning.
 705  *
 706  * Every prefix refers to a set of "nexthop"s (gw, oif),
 707  * where gw must have narrower scope. This recursion stops
 708  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 709  * which means that gw is forced to be on link.
 710  *
 711  * Code is still hairy, but now it is apparently logically
 712  * consistent and very flexible. F.e. as by-product it allows
 713  * to co-exists in peace independent exterior and interior
 714  * routing processes.
 715  *
 716  * Normally it looks as following.
 717  *
 718  * {universe prefix}  -> (gw, oif) [scope link]
 719  *                |
 720  *                |-> {link prefix} -> (gw, oif) [scope local]
 721  *                                      |
 722  *                                      |-> {local prefix} (terminal node)
 723  */
 724 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
 725                         struct fib_nh *nh)
 726 {
 727         int err = 0;
 728         struct net *net;
 729         struct net_device *dev;
 730
 731         net = cfg->fc_nlinfo.nl_net;
 732         if (nh->nh_gw) {
 733                 struct fib_result res;
 734
 735                 if (nh->nh_flags & RTNH_F_ONLINK) {
 736                         unsigned int addr_type;
 737
 738                         if (cfg->fc_scope >= RT_SCOPE_LINK)
 739                                 return -EINVAL;
 740                         dev = __dev_get_by_index(net, nh->nh_oif);
 741                         if (!dev)
 742                                 return -ENODEV;
 743                         if (!(dev->flags & IFF_UP))
 744                                 return -ENETDOWN;
 745                         addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
 746                         if (addr_type != RTN_UNICAST)
 747                                 return -EINVAL;
 748                         if (!netif_carrier_ok(dev))
 749                                 nh->nh_flags |= RTNH_F_LINKDOWN;
 750                         nh->nh_dev = dev;
 751                         dev_hold(dev);
 752                         nh->nh_scope = RT_SCOPE_LINK;
 753                         return 0;
 754                 }
 755                 rcu_read_lock();
 756                 {
 757                         struct fib_table *tbl = NULL;
 758                         struct flowi4 fl4 = {
 759                                 .daddr = nh->nh_gw,
 760                                 .flowi4_scope = cfg->fc_scope + 1,
 761                                 .flowi4_oif = nh->nh_oif,
 762                                 .flowi4_iif = LOOPBACK_IFINDEX,
 763                         };
 764
 765                         /* It is not necessary, but requires a bit of thinking */
 766                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
 767                                 fl4.flowi4_scope = RT_SCOPE_LINK;
 768
 769                         if (cfg->fc_table)
 770                                 tbl = fib_get_table(net, cfg->fc_table);
 771
 772                         if (tbl)
 773                                 err = fib_table_lookup(tbl, &fl4, &res,
 774                                                        FIB_LOOKUP_IGNORE_LINKSTATE |
 775                                                        FIB_LOOKUP_NOREF);
 776
 777                         /* on error or if no table given do full lookup. This
 778                          * is needed for example when nexthops are in the local
 779                          * table rather than the given table
 780                          */
 781                         if (!tbl || err) {
 782                                 err = fib_lookup(net, &fl4, &res,
 783                                                  FIB_LOOKUP_IGNORE_LINKSTATE);
 784                         }
 785
 786                         if (err) {
 787                                 rcu_read_unlock();
 788                                 return err;
 789                         }
 790                 }
 791                 err = -EINVAL;
 792                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 793                         goto out;
 794                 nh->nh_scope = res.scope;
 795                 nh->nh_oif = FIB_RES_OIF(res);
 796                 nh->nh_dev = dev = FIB_RES_DEV(res);
 797                 if (!dev)
 798                         goto out;
 799                 dev_hold(dev);
 800                 if (!netif_carrier_ok(dev))
 801                         nh->nh_flags |= RTNH_F_LINKDOWN;
 802                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 803         } else {
 804                 struct in_device *in_dev;
 805
 806                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 807                         return -EINVAL;
 808
 809                 rcu_read_lock();
 810                 err = -ENODEV;
 811                 in_dev = inetdev_by_index(net, nh->nh_oif);
 812                 if (!in_dev)
 813                         goto out;
 814                 err = -ENETDOWN;
 815                 if (!(in_dev->dev->flags & IFF_UP))
 816                         goto out;
 817                 nh->nh_dev = in_dev->dev;
 818                 dev_hold(nh->nh_dev);
 819                 nh->nh_scope = RT_SCOPE_HOST;
 820                 if (!netif_carrier_ok(nh->nh_dev))
 821                         nh->nh_flags |= RTNH_F_LINKDOWN;
 822                 err = 0;
 823         }
 824 out:
 825         rcu_read_unlock();
 826         return err;
 827 }
 828
 829 static inline unsigned int fib_laddr_hashfn(__be32 val)
 830 {
 831         unsigned int mask = (fib_info_hash_size - 1);
 832
 833         return ((__force u32)val ^
 834                 ((__force u32)val >> 7) ^
 835                 ((__force u32)val >> 14)) & mask;
 836 }
 837
 838 static struct hlist_head *fib_info_hash_alloc(int bytes)
 839 {
 840         if (bytes <= PAGE_SIZE)
 841                 return kzalloc(bytes, GFP_KERNEL);
 842         else
 843                 return (struct hlist_head *)
 844                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 845                                          get_order(bytes));
 846 }
 847
 848 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 849 {
 850         if (!hash)
 851                 return;
 852
 853         if (bytes <= PAGE_SIZE)
 854                 kfree(hash);
 855         else
 856                 free_pages((unsigned long) hash, get_order(bytes));
 857 }
 858
 859 static void fib_info_hash_move(struct hlist_head *new_info_hash,
 860                                struct hlist_head *new_laddrhash,
 861                                unsigned int new_size)
 862 {
 863         struct hlist_head *old_info_hash, *old_laddrhash;
 864         unsigned int old_size = fib_info_hash_size;
 865         unsigned int i, bytes;
 866
 867         spin_lock_bh(&fib_info_lock);
 868         old_info_hash = fib_info_hash;
 869         old_laddrhash = fib_info_laddrhash;
 870         fib_info_hash_size = new_size;
 871
 872         for (i = 0; i < old_size; i++) {
 873                 struct hlist_head *head = &fib_info_hash[i];
 874                 struct hlist_node *n;
 875                 struct fib_info *fi;
 876
 877                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
 878                         struct hlist_head *dest;
 879                         unsigned int new_hash;
 880
 881                         new_hash = fib_info_hashfn(fi);
 882                         dest = &new_info_hash[new_hash];
 883                         hlist_add_head(&fi->fib_hash, dest);
 884                 }
 885         }
 886         fib_info_hash = new_info_hash;
 887
 888         for (i = 0; i < old_size; i++) {
 889                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 890                 struct hlist_node *n;
 891                 struct fib_info *fi;
 892
 893                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
 894                         struct hlist_head *ldest;
 895                         unsigned int new_hash;
 896
 897                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 898                         ldest = &new_laddrhash[new_hash];
 899                         hlist_add_head(&fi->fib_lhash, ldest);
 900                 }
 901         }
 902         fib_info_laddrhash = new_laddrhash;
 903
 904         spin_unlock_bh(&fib_info_lock);
 905
 906         bytes = old_size * sizeof(struct hlist_head *);
 907         fib_info_hash_free(old_info_hash, bytes);
 908         fib_info_hash_free(old_laddrhash, bytes);
 909 }
 910
 911 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 912 {
 913         nh->nh_saddr = inet_select_addr(nh->nh_dev,
 914                                         nh->nh_gw,
 915                                         nh->nh_parent->fib_scope);
 916         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 917
 918         return nh->nh_saddr;
 919 }
 920
 921 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
 922 {
 923         if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 924             fib_prefsrc != cfg->fc_dst) {
 925                 u32 tb_id = cfg->fc_table;
 926
 927                 if (tb_id == RT_TABLE_MAIN)
 928                         tb_id = RT_TABLE_LOCAL;
 929
 930                 if (inet_addr_type_table(cfg->fc_nlinfo.nl_net,
 931                                          fib_prefsrc, tb_id) != RTN_LOCAL) {
 932                         return false;
 933                 }
 934         }
 935         return true;
 936 }
 937
 938 static int
 939 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
 940 {
 941         bool ecn_ca = false;
 942         struct nlattr *nla;
 943         int remaining;
 944
 945         if (!cfg->fc_mx)
 946                 return 0;
 947
 948         nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 949                 int type = nla_type(nla);
 950                 u32 val;
 951
 952                 if (!type)
 953                         continue;
 954                 if (type > RTAX_MAX)
 955                         return -EINVAL;
 956
 957                 if (type == RTAX_CC_ALGO) {
 958                         char tmp[TCP_CA_NAME_MAX];
 959
 960                         nla_strlcpy(tmp, nla, sizeof(tmp));
 961                         val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
 962                         if (val == TCP_CA_UNSPEC)
 963                                 return -EINVAL;
 964                 } else {
 965                         val = nla_get_u32(nla);
 966                 }
 967                 if (type == RTAX_ADVMSS && val > 65535 - 40)
 968                         val = 65535 - 40;
 969                 if (type == RTAX_MTU && val > 65535 - 15)
 970                         val = 65535 - 15;
 971                 if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
 972                         return -EINVAL;
 973                 fi->fib_metrics[type - 1] = val;
 974         }
 975
 976         if (ecn_ca)
 977                 fi->fib_metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
 978
 979         return 0;
 980 }
 981
 982 struct fib_info *fib_create_info(struct fib_config *cfg)
 983 {
 984         int err;
 985         struct fib_info *fi = NULL;
 986         struct fib_info *ofi;
 987         int nhs = 1;
 988         struct net *net = cfg->fc_nlinfo.nl_net;
 989
 990         if (cfg->fc_type > RTN_MAX)
 991                 goto err_inval;
 992
 993         /* Fast check to catch the most weird cases */
 994         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
 995                 goto err_inval;
 996
 997 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 998         if (cfg->fc_mp) {
 999                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
1000                 if (nhs == 0)
1001                         goto err_inval;
1002         }
1003 #endif
1004
1005         err = -ENOBUFS;
1006         if (fib_info_cnt >= fib_info_hash_size) {
1007                 unsigned int new_size = fib_info_hash_size << 1;
1008                 struct hlist_head *new_info_hash;
1009                 struct hlist_head *new_laddrhash;
1010                 unsigned int bytes;
1011
1012                 if (!new_size)
1013                         new_size = 16;
1014                 bytes = new_size * sizeof(struct hlist_head *);
1015                 new_info_hash = fib_info_hash_alloc(bytes);
1016                 new_laddrhash = fib_info_hash_alloc(bytes);
1017                 if (!new_info_hash || !new_laddrhash) {
1018                         fib_info_hash_free(new_info_hash, bytes);
1019                         fib_info_hash_free(new_laddrhash, bytes);
1020                 } else
1021                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1022
1023                 if (!fib_info_hash_size)
1024                         goto failure;
1025         }
1026
1027         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1028         if (!fi)
1029                 goto failure;
1030         fib_info_cnt++;
1031         if (cfg->fc_mx) {
1032                 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
1033                 if (!fi->fib_metrics)
1034                         goto failure;
1035         } else
1036                 fi->fib_metrics = (u32 *) dst_default_metrics;
1037
1038         fi->fib_net = net;
1039         fi->fib_protocol = cfg->fc_protocol;
1040         fi->fib_scope = cfg->fc_scope;
1041         fi->fib_flags = cfg->fc_flags;
1042         fi->fib_priority = cfg->fc_priority;
1043         fi->fib_prefsrc = cfg->fc_prefsrc;
1044         fi->fib_type = cfg->fc_type;
1045
1046         fi->fib_nhs = nhs;
1047         change_nexthops(fi) {
1048                 nexthop_nh->nh_parent = fi;
1049                 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1050                 if (!nexthop_nh->nh_pcpu_rth_output)
1051                         goto failure;
1052         } endfor_nexthops(fi)
1053
1054         err = fib_convert_metrics(fi, cfg);
1055         if (err)
1056                 goto failure;
1057
1058         if (cfg->fc_mp) {
1059 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1060                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
1061                 if (err != 0)
1062                         goto failure;
1063                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
1064                         goto err_inval;
1065                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
1066                         goto err_inval;
1067 #ifdef CONFIG_IP_ROUTE_CLASSID
1068                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
1069                         goto err_inval;
1070 #endif
1071 #else
1072                 goto err_inval;
1073 #endif
1074         } else {
1075                 struct fib_nh *nh = fi->fib_nh;
1076
1077                 if (cfg->fc_encap) {
1078                         struct lwtunnel_state *lwtstate;
1079                         struct net_device *dev = NULL;
1080
1081                         if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
1082                                 goto err_inval;
1083                         if (cfg->fc_oif)
1084                                 dev = __dev_get_by_index(net, cfg->fc_oif);
1085                         err = lwtunnel_build_state(dev, cfg->fc_encap_type,
1086                                                    cfg->fc_encap, AF_INET, cfg,
1087                                                    &lwtstate);
1088                         if (err)
1089                                 goto failure;
1090
1091                         nh->nh_lwtstate = lwtstate_get(lwtstate);
1092                 }
1093                 nh->nh_oif = cfg->fc_oif;
1094                 nh->nh_gw = cfg->fc_gw;
1095                 nh->nh_flags = cfg->fc_flags;
1096 #ifdef CONFIG_IP_ROUTE_CLASSID
1097                 nh->nh_tclassid = cfg->fc_flow;
1098                 if (nh->nh_tclassid)
1099                         fi->fib_net->ipv4.fib_num_tclassid_users++;
1100 #endif
1101 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1102                 nh->nh_weight = 1;
1103 #endif
1104         }
1105
1106         if (fib_props[cfg->fc_type].error) {
1107                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
1108                         goto err_inval;
1109                 goto link_it;
1110         } else {
1111                 switch (cfg->fc_type) {
1112                 case RTN_UNICAST:
1113                 case RTN_LOCAL:
1114                 case RTN_BROADCAST:
1115                 case RTN_ANYCAST:
1116                 case RTN_MULTICAST:
1117                         break;
1118                 default:
1119                         goto err_inval;
1120                 }
1121         }
1122
1123         if (cfg->fc_scope > RT_SCOPE_HOST)
1124                 goto err_inval;
1125
1126         if (cfg->fc_scope == RT_SCOPE_HOST) {
1127                 struct fib_nh *nh = fi->fib_nh;
1128
1129                 /* Local address is added. */
1130                 if (nhs != 1 || nh->nh_gw)
1131                         goto err_inval;
1132                 nh->nh_scope = RT_SCOPE_NOWHERE;
1133                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1134                 err = -ENODEV;
1135                 if (!nh->nh_dev)
1136                         goto failure;
1137         } else {
1138                 int linkdown = 0;
1139
1140                 change_nexthops(fi) {
1141                         err = fib_check_nh(cfg, fi, nexthop_nh);
1142                         if (err != 0)
1143                                 goto failure;
1144                         if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1145                                 linkdown++;
1146                 } endfor_nexthops(fi)
1147                 if (linkdown == fi->fib_nhs)
1148                         fi->fib_flags |= RTNH_F_LINKDOWN;
1149         }
1150
1151         if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc))
1152                 goto err_inval;
1153
1154         change_nexthops(fi) {
1155                 fib_info_update_nh_saddr(net, nexthop_nh);
1156                 fib_add_weight(fi, nexthop_nh);
1157         } endfor_nexthops(fi)
1158
1159         fib_rebalance(fi);
1160
1161 link_it:
1162         ofi = fib_find_info(fi);
1163         if (ofi) {
1164                 fi->fib_dead = 1;
1165                 free_fib_info(fi);
1166                 ofi->fib_treeref++;
1167                 return ofi;
1168         }
1169
1170         fi->fib_treeref++;
1171         atomic_inc(&fi->fib_clntref);
1172         spin_lock_bh(&fib_info_lock);
1173         hlist_add_head(&fi->fib_hash,
1174                        &fib_info_hash[fib_info_hashfn(fi)]);
1175         if (fi->fib_prefsrc) {
1176                 struct hlist_head *head;
1177
1178                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1179                 hlist_add_head(&fi->fib_lhash, head);
1180         }
1181         change_nexthops(fi) {
1182                 struct hlist_head *head;
1183                 unsigned int hash;
1184
1185                 if (!nexthop_nh->nh_dev)
1186                         continue;
1187                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1188                 head = &fib_info_devhash[hash];
1189                 hlist_add_head(&nexthop_nh->nh_hash, head);
1190         } endfor_nexthops(fi)
1191         spin_unlock_bh(&fib_info_lock);
1192         return fi;
1193
1194 err_inval:
1195         err = -EINVAL;
1196
1197 failure:
1198         if (fi) {
1199                 fi->fib_dead = 1;
1200                 free_fib_info(fi);
1201         }
1202
1203         return ERR_PTR(err);
1204 }
1205
1206 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1207                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1208                   struct fib_info *fi, unsigned int flags)
1209 {
1210         struct nlmsghdr *nlh;
1211         struct rtmsg *rtm;
1212
1213         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1214         if (!nlh)
1215                 return -EMSGSIZE;
1216
1217         rtm = nlmsg_data(nlh);
1218         rtm->rtm_family = AF_INET;
1219         rtm->rtm_dst_len = dst_len;
1220         rtm->rtm_src_len = 0;
1221         rtm->rtm_tos = tos;
1222         if (tb_id < 256)
1223                 rtm->rtm_table = tb_id;
1224         else
1225                 rtm->rtm_table = RT_TABLE_COMPAT;
1226         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1227                 goto nla_put_failure;
1228         rtm->rtm_type = type;
1229         rtm->rtm_flags = fi->fib_flags;
1230         rtm->rtm_scope = fi->fib_scope;
1231         rtm->rtm_protocol = fi->fib_protocol;
1232
1233         if (rtm->rtm_dst_len &&
1234             nla_put_in_addr(skb, RTA_DST, dst))
1235                 goto nla_put_failure;
1236         if (fi->fib_priority &&
1237             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1238                 goto nla_put_failure;
1239         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1240                 goto nla_put_failure;
1241
1242         if (fi->fib_prefsrc &&
1243             nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1244                 goto nla_put_failure;
1245         if (fi->fib_nhs == 1) {
1246                 struct in_device *in_dev;
1247
1248                 if (fi->fib_nh->nh_gw &&
1249                     nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1250                         goto nla_put_failure;
1251                 if (fi->fib_nh->nh_oif &&
1252                     nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1253                         goto nla_put_failure;
1254                 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1255                         in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1256                         if (in_dev &&
1257                             IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1258                                 rtm->rtm_flags |= RTNH_F_DEAD;
1259                 }
1260 #ifdef CONFIG_IP_ROUTE_CLASSID
1261                 if (fi->fib_nh[0].nh_tclassid &&
1262                     nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1263                         goto nla_put_failure;
1264 #endif
1265                 if (fi->fib_nh->nh_lwtstate)
1266                         lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate);
1267         }
1268 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1269         if (fi->fib_nhs > 1) {
1270                 struct rtnexthop *rtnh;
1271                 struct nlattr *mp;
1272
1273                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1274                 if (!mp)
1275                         goto nla_put_failure;
1276
1277                 for_nexthops(fi) {
1278                         struct in_device *in_dev;
1279
1280                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1281                         if (!rtnh)
1282                                 goto nla_put_failure;
1283
1284                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1285                         if (nh->nh_flags & RTNH_F_LINKDOWN) {
1286                                 in_dev = __in_dev_get_rtnl(nh->nh_dev);
1287                                 if (in_dev &&
1288                                     IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1289                                         rtnh->rtnh_flags |= RTNH_F_DEAD;
1290                         }
1291                         rtnh->rtnh_hops = nh->nh_weight - 1;
1292                         rtnh->rtnh_ifindex = nh->nh_oif;
1293
1294                         if (nh->nh_gw &&
1295                             nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1296                                 goto nla_put_failure;
1297 #ifdef CONFIG_IP_ROUTE_CLASSID
1298                         if (nh->nh_tclassid &&
1299                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1300                                 goto nla_put_failure;
1301 #endif
1302                         if (nh->nh_lwtstate)
1303                                 lwtunnel_fill_encap(skb, nh->nh_lwtstate);
1304                         /* length of rtnetlink header + attributes */
1305                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1306                 } endfor_nexthops(fi);
1307
1308                 nla_nest_end(skb, mp);
1309         }
1310 #endif
1311         nlmsg_end(skb, nlh);
1312         return 0;
1313
1314 nla_put_failure:
1315         nlmsg_cancel(skb, nlh);
1316         return -EMSGSIZE;
1317 }
1318
1319 /*
1320  * Update FIB if:
1321  * - local address disappeared -> we must delete all the entries
1322  *   referring to it.
1323  * - device went down -> we must shutdown all nexthops going via it.
1324  */
1325 int fib_sync_down_addr(struct net *net, __be32 local)
1326 {
1327         int ret = 0;
1328         unsigned int hash = fib_laddr_hashfn(local);
1329         struct hlist_head *head = &fib_info_laddrhash[hash];
1330         struct fib_info *fi;
1331
1332         if (!fib_info_laddrhash || local == 0)
1333                 return 0;
1334
1335         hlist_for_each_entry(fi, head, fib_lhash) {
1336                 if (!net_eq(fi->fib_net, net))
1337                         continue;
1338                 if (fi->fib_prefsrc == local) {
1339                         fi->fib_flags |= RTNH_F_DEAD;
1340                         ret++;
1341                 }
1342         }
1343         return ret;
1344 }
1345
1346 /* Event              force Flags           Description
1347  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1348  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1349  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1350  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1351  */
1352 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1353 {
1354         int ret = 0;
1355         int scope = RT_SCOPE_NOWHERE;
1356         struct fib_info *prev_fi = NULL;
1357         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1358         struct hlist_head *head = &fib_info_devhash[hash];
1359         struct fib_nh *nh;
1360
1361         if (force)
1362                 scope = -1;
1363
1364         hlist_for_each_entry(nh, head, nh_hash) {
1365                 struct fib_info *fi = nh->nh_parent;
1366                 int dead;
1367
1368                 BUG_ON(!fi->fib_nhs);
1369                 if (nh->nh_dev != dev || fi == prev_fi)
1370                         continue;
1371                 prev_fi = fi;
1372                 dead = 0;
1373                 change_nexthops(fi) {
1374                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1375                                 dead++;
1376                         else if (nexthop_nh->nh_dev == dev &&
1377                                  nexthop_nh->nh_scope != scope) {
1378                                 switch (event) {
1379                                 case NETDEV_DOWN:
1380                                 case NETDEV_UNREGISTER:
1381                                         nexthop_nh->nh_flags |= RTNH_F_DEAD;
1382                                         /* fall through */
1383                                 case NETDEV_CHANGE:
1384                                         nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1385                                         break;
1386                                 }
1387                                 dead++;
1388                         }
1389 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1390                         if (event == NETDEV_UNREGISTER &&
1391                             nexthop_nh->nh_dev == dev) {
1392                                 dead = fi->fib_nhs;
1393                                 break;
1394                         }
1395 #endif
1396                 } endfor_nexthops(fi)
1397                 if (dead == fi->fib_nhs) {
1398                         switch (event) {
1399                         case NETDEV_DOWN:
1400                         case NETDEV_UNREGISTER:
1401                                 fi->fib_flags |= RTNH_F_DEAD;
1402                                 /* fall through */
1403                         case NETDEV_CHANGE:
1404                                 fi->fib_flags |= RTNH_F_LINKDOWN;
1405                                 break;
1406                         }
1407                         ret++;
1408                 }
1409
1410                 fib_rebalance(fi);
1411         }
1412
1413         return ret;
1414 }
1415
1416 /* Must be invoked inside of an RCU protected region.  */
1417 void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1418 {
1419         struct fib_info *fi = NULL, *last_resort = NULL;
1420         struct hlist_head *fa_head = res->fa_head;
1421         struct fib_table *tb = res->table;
1422         u8 slen = 32 - res->prefixlen;
1423         int order = -1, last_idx = -1;
1424         struct fib_alias *fa, *fa1 = NULL;
1425         u32 last_prio = res->fi->fib_priority;
1426         u8 last_tos = 0;
1427
1428         hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1429                 struct fib_info *next_fi = fa->fa_info;
1430
1431                 if (fa->fa_slen != slen)
1432                         continue;
1433                 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1434                         continue;
1435                 if (fa->tb_id != tb->tb_id)
1436                         continue;
1437                 if (next_fi->fib_priority > last_prio &&
1438                     fa->fa_tos == last_tos) {
1439                         if (last_tos)
1440                                 continue;
1441                         break;
1442                 }
1443                 if (next_fi->fib_flags & RTNH_F_DEAD)
1444                         continue;
1445                 last_tos = fa->fa_tos;
1446                 last_prio = next_fi->fib_priority;
1447
1448                 if (next_fi->fib_scope != res->scope ||
1449                     fa->fa_type != RTN_UNICAST)
1450                         continue;
1451                 if (!next_fi->fib_nh[0].nh_gw ||
1452                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1453                         continue;
1454
1455                 fib_alias_accessed(fa);
1456
1457                 if (!fi) {
1458                         if (next_fi != res->fi)
1459                                 break;
1460                         fa1 = fa;
1461                 } else if (!fib_detect_death(fi, order, &last_resort,
1462                                              &last_idx, fa1->fa_default)) {
1463                         fib_result_assign(res, fi);
1464                         fa1->fa_default = order;
1465                         goto out;
1466                 }
1467                 fi = next_fi;
1468                 order++;
1469         }
1470
1471         if (order <= 0 || !fi) {
1472                 if (fa1)
1473                         fa1->fa_default = -1;
1474                 goto out;
1475         }
1476
1477         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1478                               fa1->fa_default)) {
1479                 fib_result_assign(res, fi);
1480                 fa1->fa_default = order;
1481                 goto out;
1482         }
1483
1484         if (last_idx >= 0)
1485                 fib_result_assign(res, last_resort);
1486         fa1->fa_default = last_idx;
1487 out:
1488         return;
1489 }
1490
1491 /*
1492  * Dead device goes up. We wake up dead nexthops.
1493  * It takes sense only on multipath routes.
1494  */
1495 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1496 {
1497         struct fib_info *prev_fi;
1498         unsigned int hash;
1499         struct hlist_head *head;
1500         struct fib_nh *nh;
1501         int ret;
1502
1503         if (!(dev->flags & IFF_UP))
1504                 return 0;
1505
1506         if (nh_flags & RTNH_F_DEAD) {
1507                 unsigned int flags = dev_get_flags(dev);
1508
1509                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1510                         nh_flags |= RTNH_F_LINKDOWN;
1511         }
1512
1513         prev_fi = NULL;
1514         hash = fib_devindex_hashfn(dev->ifindex);
1515         head = &fib_info_devhash[hash];
1516         ret = 0;
1517
1518         hlist_for_each_entry(nh, head, nh_hash) {
1519                 struct fib_info *fi = nh->nh_parent;
1520                 int alive;
1521
1522                 BUG_ON(!fi->fib_nhs);
1523                 if (nh->nh_dev != dev || fi == prev_fi)
1524                         continue;
1525
1526                 prev_fi = fi;
1527                 alive = 0;
1528                 change_nexthops(fi) {
1529                         if (!(nexthop_nh->nh_flags & nh_flags)) {
1530                                 alive++;
1531                                 continue;
1532                         }
1533                         if (!nexthop_nh->nh_dev ||
1534                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1535                                 continue;
1536                         if (nexthop_nh->nh_dev != dev ||
1537                             !__in_dev_get_rtnl(dev))
1538                                 continue;
1539                         alive++;
1540                         nexthop_nh->nh_flags &= ~nh_flags;
1541                 } endfor_nexthops(fi)
1542
1543                 if (alive > 0) {
1544                         fi->fib_flags &= ~nh_flags;
1545                         ret++;
1546                 }
1547
1548                 fib_rebalance(fi);
1549         }
1550
1551         return ret;
1552 }
1553
1554 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1555
1556 void fib_select_multipath(struct fib_result *res, int hash)
1557 {
1558         struct fib_info *fi = res->fi;
1559
1560         for_nexthops(fi) {
1561                 if (hash > atomic_read(&nh->nh_upper_bound))
1562                         continue;
1563
1564                 res->nh_sel = nhsel;
1565                 return;
1566         } endfor_nexthops(fi);
1567
1568         /* Race condition: route has just become dead. */
1569         res->nh_sel = 0;
1570 }
1571 #endif
1572
1573 void fib_select_path(struct net *net, struct fib_result *res,
1574                      struct flowi4 *fl4, int mp_hash)
1575 {
1576 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1577         if (res->fi->fib_nhs > 1 && fl4->flowi4_oif == 0) {
1578                 if (mp_hash < 0)
1579                         mp_hash = get_hash_from_flowi4(fl4) >> 1;
1580
1581                 fib_select_multipath(res, mp_hash);
1582         }
1583         else
1584 #endif
1585         if (!res->prefixlen &&
1586             res->table->tb_num_default > 1 &&
1587             res->type == RTN_UNICAST && !fl4->flowi4_oif)
1588                 fib_select_default(fl4, res);
1589
1590         if (!fl4->saddr)
1591                 fl4->saddr = FIB_RES_PREFSRC(net, *res);
1592 }
1593 EXPORT_SYMBOL_GPL(fib_select_path);