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 <linux/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 EXPORT_SYMBOL_GPL(free_fib_info);
 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                 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
 484                         return -EINVAL;
 485
 486                 nexthop_nh->nh_flags =
 487                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 488                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
 489                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
 490
 491                 attrlen = rtnh_attrlen(rtnh);
 492                 if (attrlen > 0) {
 493                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 494
 495                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 496                         nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
 497 #ifdef CONFIG_IP_ROUTE_CLASSID
 498                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 499                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 500                         if (nexthop_nh->nh_tclassid)
 501                                 fi->fib_net->ipv4.fib_num_tclassid_users++;
 502 #endif
 503                         nla = nla_find(attrs, attrlen, RTA_ENCAP);
 504                         if (nla) {
 505                                 struct lwtunnel_state *lwtstate;
 506                                 struct net_device *dev = NULL;
 507                                 struct nlattr *nla_entype;
 508
 509                                 nla_entype = nla_find(attrs, attrlen,
 510                                                       RTA_ENCAP_TYPE);
 511                                 if (!nla_entype)
 512                                         goto err_inval;
 513                                 if (cfg->fc_oif)
 514                                         dev = __dev_get_by_index(net, cfg->fc_oif);
 515                                 ret = lwtunnel_build_state(dev, nla_get_u16(
 516                                                            nla_entype),
 517                                                            nla,  AF_INET, cfg,
 518                                                            &lwtstate);
 519                                 if (ret)
 520                                         goto errout;
 521                                 nexthop_nh->nh_lwtstate =
 522                                         lwtstate_get(lwtstate);
 523                         }
 524                 }
 525
 526                 rtnh = rtnh_next(rtnh, &remaining);
 527         } endfor_nexthops(fi);
 528
 529         return 0;
 530
 531 err_inval:
 532         ret = -EINVAL;
 533
 534 errout:
 535         return ret;
 536 }
 537
 538 static void fib_rebalance(struct fib_info *fi)
 539 {
 540         int total;
 541         int w;
 542         struct in_device *in_dev;
 543
 544         if (fi->fib_nhs < 2)
 545                 return;
 546
 547         total = 0;
 548         for_nexthops(fi) {
 549                 if (nh->nh_flags & RTNH_F_DEAD)
 550                         continue;
 551
 552                 in_dev = __in_dev_get_rtnl(nh->nh_dev);
 553
 554                 if (in_dev &&
 555                     IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
 556                     nh->nh_flags & RTNH_F_LINKDOWN)
 557                         continue;
 558
 559                 total += nh->nh_weight;
 560         } endfor_nexthops(fi);
 561
 562         w = 0;
 563         change_nexthops(fi) {
 564                 int upper_bound;
 565
 566                 in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
 567
 568                 if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
 569                         upper_bound = -1;
 570                 } else if (in_dev &&
 571                            IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
 572                            nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
 573                         upper_bound = -1;
 574                 } else {
 575                         w += nexthop_nh->nh_weight;
 576                         upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
 577                                                             total) - 1;
 578                 }
 579
 580                 atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
 581         } endfor_nexthops(fi);
 582
 583         net_get_random_once(&fib_multipath_secret,
 584                             sizeof(fib_multipath_secret));
 585 }
 586
 587 static inline void fib_add_weight(struct fib_info *fi,
 588                                   const struct fib_nh *nh)
 589 {
 590         fi->fib_weight += nh->nh_weight;
 591 }
 592
 593 #else /* CONFIG_IP_ROUTE_MULTIPATH */
 594
 595 #define fib_rebalance(fi) do { } while (0)
 596 #define fib_add_weight(fi, nh) do { } while (0)
 597
 598 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
 599
 600 static int fib_encap_match(struct net *net, u16 encap_type,
 601                            struct nlattr *encap,
 602                            int oif, const struct fib_nh *nh,
 603                            const struct fib_config *cfg)
 604 {
 605         struct lwtunnel_state *lwtstate;
 606         struct net_device *dev = NULL;
 607         int ret, result = 0;
 608
 609         if (encap_type == LWTUNNEL_ENCAP_NONE)
 610                 return 0;
 611
 612         if (oif)
 613                 dev = __dev_get_by_index(net, oif);
 614         ret = lwtunnel_build_state(dev, encap_type, encap,
 615                                    AF_INET, cfg, &lwtstate);
 616         if (!ret) {
 617                 result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
 618                 lwtstate_free(lwtstate);
 619         }
 620
 621         return result;
 622 }
 623
 624 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
 625 {
 626         struct net *net = cfg->fc_nlinfo.nl_net;
 627 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 628         struct rtnexthop *rtnh;
 629         int remaining;
 630 #endif
 631
 632         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 633                 return 1;
 634
 635         if (cfg->fc_oif || cfg->fc_gw) {
 636                 if (cfg->fc_encap) {
 637                         if (fib_encap_match(net, cfg->fc_encap_type,
 638                                             cfg->fc_encap, cfg->fc_oif,
 639                                             fi->fib_nh, cfg))
 640                             return 1;
 641                 }
 642                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 643                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 644                         return 0;
 645                 return 1;
 646         }
 647
 648 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 649         if (!cfg->fc_mp)
 650                 return 0;
 651
 652         rtnh = cfg->fc_mp;
 653         remaining = cfg->fc_mp_len;
 654
 655         for_nexthops(fi) {
 656                 int attrlen;
 657
 658                 if (!rtnh_ok(rtnh, remaining))
 659                         return -EINVAL;
 660
 661                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 662                         return 1;
 663
 664                 attrlen = rtnh_attrlen(rtnh);
 665                 if (attrlen > 0) {
 666                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 667
 668                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 669                         if (nla && nla_get_in_addr(nla) != nh->nh_gw)
 670                                 return 1;
 671 #ifdef CONFIG_IP_ROUTE_CLASSID
 672                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 673                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 674                                 return 1;
 675 #endif
 676                 }
 677
 678                 rtnh = rtnh_next(rtnh, &remaining);
 679         } endfor_nexthops(fi);
 680 #endif
 681         return 0;
 682 }
 683
 684
 685 /*
 686  * Picture
 687  * -------
 688  *
 689  * Semantics of nexthop is very messy by historical reasons.
 690  * We have to take into account, that:
 691  * a) gateway can be actually local interface address,
 692  *    so that gatewayed route is direct.
 693  * b) gateway must be on-link address, possibly
 694  *    described not by an ifaddr, but also by a direct route.
 695  * c) If both gateway and interface are specified, they should not
 696  *    contradict.
 697  * d) If we use tunnel routes, gateway could be not on-link.
 698  *
 699  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 700  * results in pretty ugly and hairy code with obscure logic.
 701  *
 702  * I chose to generalized it instead, so that the size
 703  * of code does not increase practically, but it becomes
 704  * much more general.
 705  * Every prefix is assigned a "scope" value: "host" is local address,
 706  * "link" is direct route,
 707  * [ ... "site" ... "interior" ... ]
 708  * and "universe" is true gateway route with global meaning.
 709  *
 710  * Every prefix refers to a set of "nexthop"s (gw, oif),
 711  * where gw must have narrower scope. This recursion stops
 712  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 713  * which means that gw is forced to be on link.
 714  *
 715  * Code is still hairy, but now it is apparently logically
 716  * consistent and very flexible. F.e. as by-product it allows
 717  * to co-exists in peace independent exterior and interior
 718  * routing processes.
 719  *
 720  * Normally it looks as following.
 721  *
 722  * {universe prefix}  -> (gw, oif) [scope link]
 723  *                |
 724  *                |-> {link prefix} -> (gw, oif) [scope local]
 725  *                                      |
 726  *                                      |-> {local prefix} (terminal node)
 727  */
 728 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
 729                         struct fib_nh *nh)
 730 {
 731         int err = 0;
 732         struct net *net;
 733         struct net_device *dev;
 734
 735         net = cfg->fc_nlinfo.nl_net;
 736         if (nh->nh_gw) {
 737                 struct fib_result res;
 738
 739                 if (nh->nh_flags & RTNH_F_ONLINK) {
 740                         unsigned int addr_type;
 741
 742                         if (cfg->fc_scope >= RT_SCOPE_LINK)
 743                                 return -EINVAL;
 744                         dev = __dev_get_by_index(net, nh->nh_oif);
 745                         if (!dev)
 746                                 return -ENODEV;
 747                         if (!(dev->flags & IFF_UP))
 748                                 return -ENETDOWN;
 749                         addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
 750                         if (addr_type != RTN_UNICAST)
 751                                 return -EINVAL;
 752                         if (!netif_carrier_ok(dev))
 753                                 nh->nh_flags |= RTNH_F_LINKDOWN;
 754                         nh->nh_dev = dev;
 755                         dev_hold(dev);
 756                         nh->nh_scope = RT_SCOPE_LINK;
 757                         return 0;
 758                 }
 759                 rcu_read_lock();
 760                 {
 761                         struct fib_table *tbl = NULL;
 762                         struct flowi4 fl4 = {
 763                                 .daddr = nh->nh_gw,
 764                                 .flowi4_scope = cfg->fc_scope + 1,
 765                                 .flowi4_oif = nh->nh_oif,
 766                                 .flowi4_iif = LOOPBACK_IFINDEX,
 767                         };
 768
 769                         /* It is not necessary, but requires a bit of thinking */
 770                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
 771                                 fl4.flowi4_scope = RT_SCOPE_LINK;
 772
 773                         if (cfg->fc_table)
 774                                 tbl = fib_get_table(net, cfg->fc_table);
 775
 776                         if (tbl)
 777                                 err = fib_table_lookup(tbl, &fl4, &res,
 778                                                        FIB_LOOKUP_IGNORE_LINKSTATE |
 779                                                        FIB_LOOKUP_NOREF);
 780
 781                         /* on error or if no table given do full lookup. This
 782                          * is needed for example when nexthops are in the local
 783                          * table rather than the given table
 784                          */
 785                         if (!tbl || err) {
 786                                 err = fib_lookup(net, &fl4, &res,
 787                                                  FIB_LOOKUP_IGNORE_LINKSTATE);
 788                         }
 789
 790                         if (err) {
 791                                 rcu_read_unlock();
 792                                 return err;
 793                         }
 794                 }
 795                 err = -EINVAL;
 796                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 797                         goto out;
 798                 nh->nh_scope = res.scope;
 799                 nh->nh_oif = FIB_RES_OIF(res);
 800                 nh->nh_dev = dev = FIB_RES_DEV(res);
 801                 if (!dev)
 802                         goto out;
 803                 dev_hold(dev);
 804                 if (!netif_carrier_ok(dev))
 805                         nh->nh_flags |= RTNH_F_LINKDOWN;
 806                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 807         } else {
 808                 struct in_device *in_dev;
 809
 810                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 811                         return -EINVAL;
 812
 813                 rcu_read_lock();
 814                 err = -ENODEV;
 815                 in_dev = inetdev_by_index(net, nh->nh_oif);
 816                 if (!in_dev)
 817                         goto out;
 818                 err = -ENETDOWN;
 819                 if (!(in_dev->dev->flags & IFF_UP))
 820                         goto out;
 821                 nh->nh_dev = in_dev->dev;
 822                 dev_hold(nh->nh_dev);
 823                 nh->nh_scope = RT_SCOPE_HOST;
 824                 if (!netif_carrier_ok(nh->nh_dev))
 825                         nh->nh_flags |= RTNH_F_LINKDOWN;
 826                 err = 0;
 827         }
 828 out:
 829         rcu_read_unlock();
 830         return err;
 831 }
 832
 833 static inline unsigned int fib_laddr_hashfn(__be32 val)
 834 {
 835         unsigned int mask = (fib_info_hash_size - 1);
 836
 837         return ((__force u32)val ^
 838                 ((__force u32)val >> 7) ^
 839                 ((__force u32)val >> 14)) & mask;
 840 }
 841
 842 static struct hlist_head *fib_info_hash_alloc(int bytes)
 843 {
 844         if (bytes <= PAGE_SIZE)
 845                 return kzalloc(bytes, GFP_KERNEL);
 846         else
 847                 return (struct hlist_head *)
 848                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 849                                          get_order(bytes));
 850 }
 851
 852 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 853 {
 854         if (!hash)
 855                 return;
 856
 857         if (bytes <= PAGE_SIZE)
 858                 kfree(hash);
 859         else
 860                 free_pages((unsigned long) hash, get_order(bytes));
 861 }
 862
 863 static void fib_info_hash_move(struct hlist_head *new_info_hash,
 864                                struct hlist_head *new_laddrhash,
 865                                unsigned int new_size)
 866 {
 867         struct hlist_head *old_info_hash, *old_laddrhash;
 868         unsigned int old_size = fib_info_hash_size;
 869         unsigned int i, bytes;
 870
 871         spin_lock_bh(&fib_info_lock);
 872         old_info_hash = fib_info_hash;
 873         old_laddrhash = fib_info_laddrhash;
 874         fib_info_hash_size = new_size;
 875
 876         for (i = 0; i < old_size; i++) {
 877                 struct hlist_head *head = &fib_info_hash[i];
 878                 struct hlist_node *n;
 879                 struct fib_info *fi;
 880
 881                 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
 882                         struct hlist_head *dest;
 883                         unsigned int new_hash;
 884
 885                         new_hash = fib_info_hashfn(fi);
 886                         dest = &new_info_hash[new_hash];
 887                         hlist_add_head(&fi->fib_hash, dest);
 888                 }
 889         }
 890         fib_info_hash = new_info_hash;
 891
 892         for (i = 0; i < old_size; i++) {
 893                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 894                 struct hlist_node *n;
 895                 struct fib_info *fi;
 896
 897                 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
 898                         struct hlist_head *ldest;
 899                         unsigned int new_hash;
 900
 901                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 902                         ldest = &new_laddrhash[new_hash];
 903                         hlist_add_head(&fi->fib_lhash, ldest);
 904                 }
 905         }
 906         fib_info_laddrhash = new_laddrhash;
 907
 908         spin_unlock_bh(&fib_info_lock);
 909
 910         bytes = old_size * sizeof(struct hlist_head *);
 911         fib_info_hash_free(old_info_hash, bytes);
 912         fib_info_hash_free(old_laddrhash, bytes);
 913 }
 914
 915 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 916 {
 917         nh->nh_saddr = inet_select_addr(nh->nh_dev,
 918                                         nh->nh_gw,
 919                                         nh->nh_parent->fib_scope);
 920         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 921
 922         return nh->nh_saddr;
 923 }
 924
 925 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
 926 {
 927         if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 928             fib_prefsrc != cfg->fc_dst) {
 929                 u32 tb_id = cfg->fc_table;
 930                 int rc;
 931
 932                 if (tb_id == RT_TABLE_MAIN)
 933                         tb_id = RT_TABLE_LOCAL;
 934
 935                 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
 936                                           fib_prefsrc, tb_id);
 937
 938                 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
 939                         rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
 940                                                   fib_prefsrc, RT_TABLE_LOCAL);
 941                 }
 942
 943                 if (rc != RTN_LOCAL)
 944                         return false;
 945         }
 946         return true;
 947 }
 948
 949 static int
 950 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
 951 {
 952         bool ecn_ca = false;
 953         struct nlattr *nla;
 954         int remaining;
 955
 956         if (!cfg->fc_mx)
 957                 return 0;
 958
 959         nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 960                 int type = nla_type(nla);
 961                 u32 val;
 962
 963                 if (!type)
 964                         continue;
 965                 if (type > RTAX_MAX)
 966                         return -EINVAL;
 967
 968                 if (type == RTAX_CC_ALGO) {
 969                         char tmp[TCP_CA_NAME_MAX];
 970
 971                         nla_strlcpy(tmp, nla, sizeof(tmp));
 972                         val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
 973                         if (val == TCP_CA_UNSPEC)
 974                                 return -EINVAL;
 975                 } else {
 976                         val = nla_get_u32(nla);
 977                 }
 978                 if (type == RTAX_ADVMSS && val > 65535 - 40)
 979                         val = 65535 - 40;
 980                 if (type == RTAX_MTU && val > 65535 - 15)
 981                         val = 65535 - 15;
 982                 if (type == RTAX_HOPLIMIT && val > 255)
 983                         val = 255;
 984                 if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
 985                         return -EINVAL;
 986                 fi->fib_metrics[type - 1] = val;
 987         }
 988
 989         if (ecn_ca)
 990                 fi->fib_metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
 991
 992         return 0;
 993 }
 994
 995 struct fib_info *fib_create_info(struct fib_config *cfg)
 996 {
 997         int err;
 998         struct fib_info *fi = NULL;
 999         struct fib_info *ofi;
1000         int nhs = 1;
1001         struct net *net = cfg->fc_nlinfo.nl_net;
1002
1003         if (cfg->fc_type > RTN_MAX)
1004                 goto err_inval;
1005
1006         /* Fast check to catch the most weird cases */
1007         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
1008                 goto err_inval;
1009
1010         if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
1011                 goto err_inval;
1012
1013 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1014         if (cfg->fc_mp) {
1015                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
1016                 if (nhs == 0)
1017                         goto err_inval;
1018         }
1019 #endif
1020
1021         err = -ENOBUFS;
1022         if (fib_info_cnt >= fib_info_hash_size) {
1023                 unsigned int new_size = fib_info_hash_size << 1;
1024                 struct hlist_head *new_info_hash;
1025                 struct hlist_head *new_laddrhash;
1026                 unsigned int bytes;
1027
1028                 if (!new_size)
1029                         new_size = 16;
1030                 bytes = new_size * sizeof(struct hlist_head *);
1031                 new_info_hash = fib_info_hash_alloc(bytes);
1032                 new_laddrhash = fib_info_hash_alloc(bytes);
1033                 if (!new_info_hash || !new_laddrhash) {
1034                         fib_info_hash_free(new_info_hash, bytes);
1035                         fib_info_hash_free(new_laddrhash, bytes);
1036                 } else
1037                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1038
1039                 if (!fib_info_hash_size)
1040                         goto failure;
1041         }
1042
1043         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1044         if (!fi)
1045                 goto failure;
1046         fib_info_cnt++;
1047         if (cfg->fc_mx) {
1048                 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
1049                 if (!fi->fib_metrics)
1050                         goto failure;
1051         } else
1052                 fi->fib_metrics = (u32 *) dst_default_metrics;
1053
1054         fi->fib_net = net;
1055         fi->fib_protocol = cfg->fc_protocol;
1056         fi->fib_scope = cfg->fc_scope;
1057         fi->fib_flags = cfg->fc_flags;
1058         fi->fib_priority = cfg->fc_priority;
1059         fi->fib_prefsrc = cfg->fc_prefsrc;
1060         fi->fib_type = cfg->fc_type;
1061         fi->fib_tb_id = cfg->fc_table;
1062
1063         fi->fib_nhs = nhs;
1064         change_nexthops(fi) {
1065                 nexthop_nh->nh_parent = fi;
1066                 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1067                 if (!nexthop_nh->nh_pcpu_rth_output)
1068                         goto failure;
1069         } endfor_nexthops(fi)
1070
1071         err = fib_convert_metrics(fi, cfg);
1072         if (err)
1073                 goto failure;
1074
1075         if (cfg->fc_mp) {
1076 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1077                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
1078                 if (err != 0)
1079                         goto failure;
1080                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
1081                         goto err_inval;
1082                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
1083                         goto err_inval;
1084 #ifdef CONFIG_IP_ROUTE_CLASSID
1085                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
1086                         goto err_inval;
1087 #endif
1088 #else
1089                 goto err_inval;
1090 #endif
1091         } else {
1092                 struct fib_nh *nh = fi->fib_nh;
1093
1094                 if (cfg->fc_encap) {
1095                         struct lwtunnel_state *lwtstate;
1096                         struct net_device *dev = NULL;
1097
1098                         if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
1099                                 goto err_inval;
1100                         if (cfg->fc_oif)
1101                                 dev = __dev_get_by_index(net, cfg->fc_oif);
1102                         err = lwtunnel_build_state(dev, cfg->fc_encap_type,
1103                                                    cfg->fc_encap, AF_INET, cfg,
1104                                                    &lwtstate);
1105                         if (err)
1106                                 goto failure;
1107
1108                         nh->nh_lwtstate = lwtstate_get(lwtstate);
1109                 }
1110                 nh->nh_oif = cfg->fc_oif;
1111                 nh->nh_gw = cfg->fc_gw;
1112                 nh->nh_flags = cfg->fc_flags;
1113 #ifdef CONFIG_IP_ROUTE_CLASSID
1114                 nh->nh_tclassid = cfg->fc_flow;
1115                 if (nh->nh_tclassid)
1116                         fi->fib_net->ipv4.fib_num_tclassid_users++;
1117 #endif
1118 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1119                 nh->nh_weight = 1;
1120 #endif
1121         }
1122
1123         if (fib_props[cfg->fc_type].error) {
1124                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
1125                         goto err_inval;
1126                 goto link_it;
1127         } else {
1128                 switch (cfg->fc_type) {
1129                 case RTN_UNICAST:
1130                 case RTN_LOCAL:
1131                 case RTN_BROADCAST:
1132                 case RTN_ANYCAST:
1133                 case RTN_MULTICAST:
1134                         break;
1135                 default:
1136                         goto err_inval;
1137                 }
1138         }
1139
1140         if (cfg->fc_scope > RT_SCOPE_HOST)
1141                 goto err_inval;
1142
1143         if (cfg->fc_scope == RT_SCOPE_HOST) {
1144                 struct fib_nh *nh = fi->fib_nh;
1145
1146                 /* Local address is added. */
1147                 if (nhs != 1 || nh->nh_gw)
1148                         goto err_inval;
1149                 nh->nh_scope = RT_SCOPE_NOWHERE;
1150                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1151                 err = -ENODEV;
1152                 if (!nh->nh_dev)
1153                         goto failure;
1154         } else {
1155                 int linkdown = 0;
1156
1157                 change_nexthops(fi) {
1158                         err = fib_check_nh(cfg, fi, nexthop_nh);
1159                         if (err != 0)
1160                                 goto failure;
1161                         if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1162                                 linkdown++;
1163                 } endfor_nexthops(fi)
1164                 if (linkdown == fi->fib_nhs)
1165                         fi->fib_flags |= RTNH_F_LINKDOWN;
1166         }
1167
1168         if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc))
1169                 goto err_inval;
1170
1171         change_nexthops(fi) {
1172                 fib_info_update_nh_saddr(net, nexthop_nh);
1173                 fib_add_weight(fi, nexthop_nh);
1174         } endfor_nexthops(fi)
1175
1176         fib_rebalance(fi);
1177
1178 link_it:
1179         ofi = fib_find_info(fi);
1180         if (ofi) {
1181                 fi->fib_dead = 1;
1182                 free_fib_info(fi);
1183                 ofi->fib_treeref++;
1184                 return ofi;
1185         }
1186
1187         fi->fib_treeref++;
1188         atomic_inc(&fi->fib_clntref);
1189         spin_lock_bh(&fib_info_lock);
1190         hlist_add_head(&fi->fib_hash,
1191                        &fib_info_hash[fib_info_hashfn(fi)]);
1192         if (fi->fib_prefsrc) {
1193                 struct hlist_head *head;
1194
1195                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1196                 hlist_add_head(&fi->fib_lhash, head);
1197         }
1198         change_nexthops(fi) {
1199                 struct hlist_head *head;
1200                 unsigned int hash;
1201
1202                 if (!nexthop_nh->nh_dev)
1203                         continue;
1204                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1205                 head = &fib_info_devhash[hash];
1206                 hlist_add_head(&nexthop_nh->nh_hash, head);
1207         } endfor_nexthops(fi)
1208         spin_unlock_bh(&fib_info_lock);
1209         return fi;
1210
1211 err_inval:
1212         err = -EINVAL;
1213
1214 failure:
1215         if (fi) {
1216                 fi->fib_dead = 1;
1217                 free_fib_info(fi);
1218         }
1219
1220         return ERR_PTR(err);
1221 }
1222
1223 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1224                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1225                   struct fib_info *fi, unsigned int flags)
1226 {
1227         struct nlmsghdr *nlh;
1228         struct rtmsg *rtm;
1229
1230         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1231         if (!nlh)
1232                 return -EMSGSIZE;
1233
1234         rtm = nlmsg_data(nlh);
1235         rtm->rtm_family = AF_INET;
1236         rtm->rtm_dst_len = dst_len;
1237         rtm->rtm_src_len = 0;
1238         rtm->rtm_tos = tos;
1239         if (tb_id < 256)
1240                 rtm->rtm_table = tb_id;
1241         else
1242                 rtm->rtm_table = RT_TABLE_COMPAT;
1243         if (nla_put_u32(skb, RTA_TABLE, tb_id))
1244                 goto nla_put_failure;
1245         rtm->rtm_type = type;
1246         rtm->rtm_flags = fi->fib_flags;
1247         rtm->rtm_scope = fi->fib_scope;
1248         rtm->rtm_protocol = fi->fib_protocol;
1249
1250         if (rtm->rtm_dst_len &&
1251             nla_put_in_addr(skb, RTA_DST, dst))
1252                 goto nla_put_failure;
1253         if (fi->fib_priority &&
1254             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1255                 goto nla_put_failure;
1256         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1257                 goto nla_put_failure;
1258
1259         if (fi->fib_prefsrc &&
1260             nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1261                 goto nla_put_failure;
1262         if (fi->fib_nhs == 1) {
1263                 struct in_device *in_dev;
1264
1265                 if (fi->fib_nh->nh_gw &&
1266                     nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1267                         goto nla_put_failure;
1268                 if (fi->fib_nh->nh_oif &&
1269                     nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1270                         goto nla_put_failure;
1271                 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1272                         in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1273                         if (in_dev &&
1274                             IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1275                                 rtm->rtm_flags |= RTNH_F_DEAD;
1276                 }
1277 #ifdef CONFIG_IP_ROUTE_CLASSID
1278                 if (fi->fib_nh[0].nh_tclassid &&
1279                     nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1280                         goto nla_put_failure;
1281 #endif
1282                 if (fi->fib_nh->nh_lwtstate)
1283                         lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate);
1284         }
1285 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1286         if (fi->fib_nhs > 1) {
1287                 struct rtnexthop *rtnh;
1288                 struct nlattr *mp;
1289
1290                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1291                 if (!mp)
1292                         goto nla_put_failure;
1293
1294                 for_nexthops(fi) {
1295                         struct in_device *in_dev;
1296
1297                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1298                         if (!rtnh)
1299                                 goto nla_put_failure;
1300
1301                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1302                         if (nh->nh_flags & RTNH_F_LINKDOWN) {
1303                                 in_dev = __in_dev_get_rtnl(nh->nh_dev);
1304                                 if (in_dev &&
1305                                     IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1306                                         rtnh->rtnh_flags |= RTNH_F_DEAD;
1307                         }
1308                         rtnh->rtnh_hops = nh->nh_weight - 1;
1309                         rtnh->rtnh_ifindex = nh->nh_oif;
1310
1311                         if (nh->nh_gw &&
1312                             nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1313                                 goto nla_put_failure;
1314 #ifdef CONFIG_IP_ROUTE_CLASSID
1315                         if (nh->nh_tclassid &&
1316                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1317                                 goto nla_put_failure;
1318 #endif
1319                         if (nh->nh_lwtstate)
1320                                 lwtunnel_fill_encap(skb, nh->nh_lwtstate);
1321                         /* length of rtnetlink header + attributes */
1322                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1323                 } endfor_nexthops(fi);
1324
1325                 nla_nest_end(skb, mp);
1326         }
1327 #endif
1328         nlmsg_end(skb, nlh);
1329         return 0;
1330
1331 nla_put_failure:
1332         nlmsg_cancel(skb, nlh);
1333         return -EMSGSIZE;
1334 }
1335
1336 /*
1337  * Update FIB if:
1338  * - local address disappeared -> we must delete all the entries
1339  *   referring to it.
1340  * - device went down -> we must shutdown all nexthops going via it.
1341  */
1342 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1343 {
1344         int ret = 0;
1345         unsigned int hash = fib_laddr_hashfn(local);
1346         struct hlist_head *head = &fib_info_laddrhash[hash];
1347         struct net *net = dev_net(dev);
1348         int tb_id = l3mdev_fib_table(dev);
1349         struct fib_info *fi;
1350
1351         if (!fib_info_laddrhash || local == 0)
1352                 return 0;
1353
1354         hlist_for_each_entry(fi, head, fib_lhash) {
1355                 if (!net_eq(fi->fib_net, net) ||
1356                     fi->fib_tb_id != tb_id)
1357                         continue;
1358                 if (fi->fib_prefsrc == local) {
1359                         fi->fib_flags |= RTNH_F_DEAD;
1360                         ret++;
1361                 }
1362         }
1363         return ret;
1364 }
1365
1366 /* Event              force Flags           Description
1367  * NETDEV_CHANGE      0     LINKDOWN        Carrier OFF, not for scope host
1368  * NETDEV_DOWN        0     LINKDOWN|DEAD   Link down, not for scope host
1369  * NETDEV_DOWN        1     LINKDOWN|DEAD   Last address removed
1370  * NETDEV_UNREGISTER  1     LINKDOWN|DEAD   Device removed
1371  */
1372 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1373 {
1374         int ret = 0;
1375         int scope = RT_SCOPE_NOWHERE;
1376         struct fib_info *prev_fi = NULL;
1377         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1378         struct hlist_head *head = &fib_info_devhash[hash];
1379         struct fib_nh *nh;
1380
1381         if (force)
1382                 scope = -1;
1383
1384         hlist_for_each_entry(nh, head, nh_hash) {
1385                 struct fib_info *fi = nh->nh_parent;
1386                 int dead;
1387
1388                 BUG_ON(!fi->fib_nhs);
1389                 if (nh->nh_dev != dev || fi == prev_fi)
1390                         continue;
1391                 prev_fi = fi;
1392                 dead = 0;
1393                 change_nexthops(fi) {
1394                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1395                                 dead++;
1396                         else if (nexthop_nh->nh_dev == dev &&
1397                                  nexthop_nh->nh_scope != scope) {
1398                                 switch (event) {
1399                                 case NETDEV_DOWN:
1400                                 case NETDEV_UNREGISTER:
1401                                         nexthop_nh->nh_flags |= RTNH_F_DEAD;
1402                                         /* fall through */
1403                                 case NETDEV_CHANGE:
1404                                         nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1405                                         break;
1406                                 }
1407                                 dead++;
1408                         }
1409 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1410                         if (event == NETDEV_UNREGISTER &&
1411                             nexthop_nh->nh_dev == dev) {
1412                                 dead = fi->fib_nhs;
1413                                 break;
1414                         }
1415 #endif
1416                 } endfor_nexthops(fi)
1417                 if (dead == fi->fib_nhs) {
1418                         switch (event) {
1419                         case NETDEV_DOWN:
1420                         case NETDEV_UNREGISTER:
1421                                 fi->fib_flags |= RTNH_F_DEAD;
1422                                 /* fall through */
1423                         case NETDEV_CHANGE:
1424                                 fi->fib_flags |= RTNH_F_LINKDOWN;
1425                                 break;
1426                         }
1427                         ret++;
1428                 }
1429
1430                 fib_rebalance(fi);
1431         }
1432
1433         return ret;
1434 }
1435
1436 /* Must be invoked inside of an RCU protected region.  */
1437 void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1438 {
1439         struct fib_info *fi = NULL, *last_resort = NULL;
1440         struct hlist_head *fa_head = res->fa_head;
1441         struct fib_table *tb = res->table;
1442         u8 slen = 32 - res->prefixlen;
1443         int order = -1, last_idx = -1;
1444         struct fib_alias *fa, *fa1 = NULL;
1445         u32 last_prio = res->fi->fib_priority;
1446         u8 last_tos = 0;
1447
1448         hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1449                 struct fib_info *next_fi = fa->fa_info;
1450
1451                 if (fa->fa_slen != slen)
1452                         continue;
1453                 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1454                         continue;
1455                 if (fa->tb_id != tb->tb_id)
1456                         continue;
1457                 if (next_fi->fib_priority > last_prio &&
1458                     fa->fa_tos == last_tos) {
1459                         if (last_tos)
1460                                 continue;
1461                         break;
1462                 }
1463                 if (next_fi->fib_flags & RTNH_F_DEAD)
1464                         continue;
1465                 last_tos = fa->fa_tos;
1466                 last_prio = next_fi->fib_priority;
1467
1468                 if (next_fi->fib_scope != res->scope ||
1469                     fa->fa_type != RTN_UNICAST)
1470                         continue;
1471                 if (!next_fi->fib_nh[0].nh_gw ||
1472                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1473                         continue;
1474
1475                 fib_alias_accessed(fa);
1476
1477                 if (!fi) {
1478                         if (next_fi != res->fi)
1479                                 break;
1480                         fa1 = fa;
1481                 } else if (!fib_detect_death(fi, order, &last_resort,
1482                                              &last_idx, fa1->fa_default)) {
1483                         fib_result_assign(res, fi);
1484                         fa1->fa_default = order;
1485                         goto out;
1486                 }
1487                 fi = next_fi;
1488                 order++;
1489         }
1490
1491         if (order <= 0 || !fi) {
1492                 if (fa1)
1493                         fa1->fa_default = -1;
1494                 goto out;
1495         }
1496
1497         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1498                               fa1->fa_default)) {
1499                 fib_result_assign(res, fi);
1500                 fa1->fa_default = order;
1501                 goto out;
1502         }
1503
1504         if (last_idx >= 0)
1505                 fib_result_assign(res, last_resort);
1506         fa1->fa_default = last_idx;
1507 out:
1508         return;
1509 }
1510
1511 /*
1512  * Dead device goes up. We wake up dead nexthops.
1513  * It takes sense only on multipath routes.
1514  */
1515 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1516 {
1517         struct fib_info *prev_fi;
1518         unsigned int hash;
1519         struct hlist_head *head;
1520         struct fib_nh *nh;
1521         int ret;
1522
1523         if (!(dev->flags & IFF_UP))
1524                 return 0;
1525
1526         if (nh_flags & RTNH_F_DEAD) {
1527                 unsigned int flags = dev_get_flags(dev);
1528
1529                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1530                         nh_flags |= RTNH_F_LINKDOWN;
1531         }
1532
1533         prev_fi = NULL;
1534         hash = fib_devindex_hashfn(dev->ifindex);
1535         head = &fib_info_devhash[hash];
1536         ret = 0;
1537
1538         hlist_for_each_entry(nh, head, nh_hash) {
1539                 struct fib_info *fi = nh->nh_parent;
1540                 int alive;
1541
1542                 BUG_ON(!fi->fib_nhs);
1543                 if (nh->nh_dev != dev || fi == prev_fi)
1544                         continue;
1545
1546                 prev_fi = fi;
1547                 alive = 0;
1548                 change_nexthops(fi) {
1549                         if (!(nexthop_nh->nh_flags & nh_flags)) {
1550                                 alive++;
1551                                 continue;
1552                         }
1553                         if (!nexthop_nh->nh_dev ||
1554                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1555                                 continue;
1556                         if (nexthop_nh->nh_dev != dev ||
1557                             !__in_dev_get_rtnl(dev))
1558                                 continue;
1559                         alive++;
1560                         nexthop_nh->nh_flags &= ~nh_flags;
1561                 } endfor_nexthops(fi)
1562
1563                 if (alive > 0) {
1564                         fi->fib_flags &= ~nh_flags;
1565                         ret++;
1566                 }
1567
1568                 fib_rebalance(fi);
1569         }
1570
1571         return ret;
1572 }
1573
1574 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1575 static bool fib_good_nh(const struct fib_nh *nh)
1576 {
1577         int state = NUD_REACHABLE;
1578
1579         if (nh->nh_scope == RT_SCOPE_LINK) {
1580                 struct neighbour *n;
1581
1582                 rcu_read_lock_bh();
1583
1584                 n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1585                                               (__force u32)nh->nh_gw);
1586                 if (n)
1587                         state = n->nud_state;
1588
1589                 rcu_read_unlock_bh();
1590         }
1591
1592         return !!(state & NUD_VALID);
1593 }
1594
1595 void fib_select_multipath(struct fib_result *res, int hash)
1596 {
1597         struct fib_info *fi = res->fi;
1598         struct net *net = fi->fib_net;
1599         bool first = false;
1600
1601         for_nexthops(fi) {
1602                 if (hash > atomic_read(&nh->nh_upper_bound))
1603                         continue;
1604
1605                 if (!net->ipv4.sysctl_fib_multipath_use_neigh ||
1606                     fib_good_nh(nh)) {
1607                         res->nh_sel = nhsel;
1608                         return;
1609                 }
1610                 if (!first) {
1611                         res->nh_sel = nhsel;
1612                         first = true;
1613                 }
1614         } endfor_nexthops(fi);
1615 }
1616 #endif
1617
1618 void fib_select_path(struct net *net, struct fib_result *res,
1619                      struct flowi4 *fl4, int mp_hash)
1620 {
1621 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1622         if (res->fi->fib_nhs > 1 && fl4->flowi4_oif == 0) {
1623                 if (mp_hash < 0)
1624                         mp_hash = get_hash_from_flowi4(fl4) >> 1;
1625
1626                 fib_select_multipath(res, mp_hash);
1627         }
1628         else
1629 #endif
1630         if (!res->prefixlen &&
1631             res->table->tb_num_default > 1 &&
1632             res->type == RTN_UNICAST && !fl4->flowi4_oif)
1633                 fib_select_default(fl4, res);
1634
1635         if (!fl4->saddr)
1636                 fl4->saddr = FIB_RES_PREFSRC(net, *res);
1637 }
1638 EXPORT_SYMBOL_GPL(fib_select_path);