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 <asm/system.h>
  18 #include <linux/bitops.h>
  19 #include <linux/types.h>
  20 #include <linux/kernel.h>
  21 #include <linux/jiffies.h>
  22 #include <linux/mm.h>
  23 #include <linux/string.h>
  24 #include <linux/socket.h>
  25 #include <linux/sockios.h>
  26 #include <linux/errno.h>
  27 #include <linux/in.h>
  28 #include <linux/inet.h>
  29 #include <linux/inetdevice.h>
  30 #include <linux/netdevice.h>
  31 #include <linux/if_arp.h>
  32 #include <linux/proc_fs.h>
  33 #include <linux/skbuff.h>
  34 #include <linux/init.h>
  35 #include <linux/slab.h>
  36
  37 #include <net/arp.h>
  38 #include <net/ip.h>
  39 #include <net/protocol.h>
  40 #include <net/route.h>
  41 #include <net/tcp.h>
  42 #include <net/sock.h>
  43 #include <net/ip_fib.h>
  44 #include <net/netlink.h>
  45 #include <net/nexthop.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_hash_size;
  53 static unsigned int fib_info_cnt;
  54
  55 #define DEVINDEX_HASHBITS 8
  56 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
  57 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
  58
  59 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  60
  61 static DEFINE_SPINLOCK(fib_multipath_lock);
  62
  63 #define for_nexthops(fi) {                                              \
  64         int nhsel; const struct fib_nh *nh;                             \
  65         for (nhsel = 0, nh = (fi)->fib_nh;                              \
  66              nhsel < (fi)->fib_nhs;                                     \
  67              nh++, nhsel++)
  68
  69 #define change_nexthops(fi) {                                           \
  70         int nhsel; struct fib_nh *nexthop_nh;                           \
  71         for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh);   \
  72              nhsel < (fi)->fib_nhs;                                     \
  73              nexthop_nh++, nhsel++)
  74
  75 #else /* CONFIG_IP_ROUTE_MULTIPATH */
  76
  77 /* Hope, that gcc will optimize it to get rid of dummy loop */
  78
  79 #define for_nexthops(fi) {                                              \
  80         int nhsel; const struct fib_nh *nh = (fi)->fib_nh;              \
  81         for (nhsel = 0; nhsel < 1; nhsel++)
  82
  83 #define change_nexthops(fi) {                                           \
  84         int nhsel;                                                      \
  85         struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);    \
  86         for (nhsel = 0; nhsel < 1; nhsel++)
  87
  88 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
  89
  90 #define endfor_nexthops(fi) }
  91
  92
  93 static const struct
  94 {
  95         int     error;
  96         u8      scope;
  97 } fib_props[RTN_MAX + 1] = {
  98         [RTN_UNSPEC] = {
  99                 .error  = 0,
 100                 .scope  = RT_SCOPE_NOWHERE,
 101         },
 102         [RTN_UNICAST] = {
 103                 .error  = 0,
 104                 .scope  = RT_SCOPE_UNIVERSE,
 105         },
 106         [RTN_LOCAL] = {
 107                 .error  = 0,
 108                 .scope  = RT_SCOPE_HOST,
 109         },
 110         [RTN_BROADCAST] = {
 111                 .error  = 0,
 112                 .scope  = RT_SCOPE_LINK,
 113         },
 114         [RTN_ANYCAST] = {
 115                 .error  = 0,
 116                 .scope  = RT_SCOPE_LINK,
 117         },
 118         [RTN_MULTICAST] = {
 119                 .error  = 0,
 120                 .scope  = RT_SCOPE_UNIVERSE,
 121         },
 122         [RTN_BLACKHOLE] = {
 123                 .error  = -EINVAL,
 124                 .scope  = RT_SCOPE_UNIVERSE,
 125         },
 126         [RTN_UNREACHABLE] = {
 127                 .error  = -EHOSTUNREACH,
 128                 .scope  = RT_SCOPE_UNIVERSE,
 129         },
 130         [RTN_PROHIBIT] = {
 131                 .error  = -EACCES,
 132                 .scope  = RT_SCOPE_UNIVERSE,
 133         },
 134         [RTN_THROW] = {
 135                 .error  = -EAGAIN,
 136                 .scope  = RT_SCOPE_UNIVERSE,
 137         },
 138         [RTN_NAT] = {
 139                 .error  = -EINVAL,
 140                 .scope  = RT_SCOPE_NOWHERE,
 141         },
 142         [RTN_XRESOLVE] = {
 143                 .error  = -EINVAL,
 144                 .scope  = RT_SCOPE_NOWHERE,
 145         },
 146 };
 147
 148
 149 /* Release a nexthop info record */
 150
 151 static void free_fib_info_rcu(struct rcu_head *head)
 152 {
 153         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 154
 155         kfree(fi);
 156 }
 157
 158 void free_fib_info(struct fib_info *fi)
 159 {
 160         if (fi->fib_dead == 0) {
 161                 pr_warning("Freeing alive fib_info %p\n", fi);
 162                 return;
 163         }
 164         change_nexthops(fi) {
 165                 if (nexthop_nh->nh_dev)
 166                         dev_put(nexthop_nh->nh_dev);
 167                 nexthop_nh->nh_dev = NULL;
 168         } endfor_nexthops(fi);
 169         fib_info_cnt--;
 170         release_net(fi->fib_net);
 171         call_rcu(&fi->rcu, free_fib_info_rcu);
 172 }
 173
 174 void fib_release_info(struct fib_info *fi)
 175 {
 176         spin_lock_bh(&fib_info_lock);
 177         if (fi && --fi->fib_treeref == 0) {
 178                 hlist_del(&fi->fib_hash);
 179                 if (fi->fib_prefsrc)
 180                         hlist_del(&fi->fib_lhash);
 181                 change_nexthops(fi) {
 182                         if (!nexthop_nh->nh_dev)
 183                                 continue;
 184                         hlist_del(&nexthop_nh->nh_hash);
 185                 } endfor_nexthops(fi)
 186                 fi->fib_dead = 1;
 187                 fib_info_put(fi);
 188         }
 189         spin_unlock_bh(&fib_info_lock);
 190 }
 191
 192 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
 193 {
 194         const struct fib_nh *onh = ofi->fib_nh;
 195
 196         for_nexthops(fi) {
 197                 if (nh->nh_oif != onh->nh_oif ||
 198                     nh->nh_gw  != onh->nh_gw ||
 199                     nh->nh_scope != onh->nh_scope ||
 200 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 201                     nh->nh_weight != onh->nh_weight ||
 202 #endif
 203 #ifdef CONFIG_NET_CLS_ROUTE
 204                     nh->nh_tclassid != onh->nh_tclassid ||
 205 #endif
 206                     ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
 207                         return -1;
 208                 onh++;
 209         } endfor_nexthops(fi);
 210         return 0;
 211 }
 212
 213 static inline unsigned int fib_devindex_hashfn(unsigned int val)
 214 {
 215         unsigned int mask = DEVINDEX_HASHSIZE - 1;
 216
 217         return (val ^
 218                 (val >> DEVINDEX_HASHBITS) ^
 219                 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
 220 }
 221
 222 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
 223 {
 224         unsigned int mask = (fib_hash_size - 1);
 225         unsigned int val = fi->fib_nhs;
 226
 227         val ^= fi->fib_protocol;
 228         val ^= (__force u32)fi->fib_prefsrc;
 229         val ^= fi->fib_priority;
 230         for_nexthops(fi) {
 231                 val ^= fib_devindex_hashfn(nh->nh_oif);
 232         } endfor_nexthops(fi)
 233
 234         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 235 }
 236
 237 static struct fib_info *fib_find_info(const struct fib_info *nfi)
 238 {
 239         struct hlist_head *head;
 240         struct hlist_node *node;
 241         struct fib_info *fi;
 242         unsigned int hash;
 243
 244         hash = fib_info_hashfn(nfi);
 245         head = &fib_info_hash[hash];
 246
 247         hlist_for_each_entry(fi, node, head, fib_hash) {
 248                 if (!net_eq(fi->fib_net, nfi->fib_net))
 249                         continue;
 250                 if (fi->fib_nhs != nfi->fib_nhs)
 251                         continue;
 252                 if (nfi->fib_protocol == fi->fib_protocol &&
 253                     nfi->fib_prefsrc == fi->fib_prefsrc &&
 254                     nfi->fib_priority == fi->fib_priority &&
 255                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 256                            sizeof(fi->fib_metrics)) == 0 &&
 257                     ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
 258                     (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
 259                         return fi;
 260         }
 261
 262         return NULL;
 263 }
 264
 265 /* Check, that the gateway is already configured.
 266  * Used only by redirect accept routine.
 267  */
 268 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 269 {
 270         struct hlist_head *head;
 271         struct hlist_node *node;
 272         struct fib_nh *nh;
 273         unsigned int hash;
 274
 275         spin_lock(&fib_info_lock);
 276
 277         hash = fib_devindex_hashfn(dev->ifindex);
 278         head = &fib_info_devhash[hash];
 279         hlist_for_each_entry(nh, node, head, nh_hash) {
 280                 if (nh->nh_dev == dev &&
 281                     nh->nh_gw == gw &&
 282                     !(nh->nh_flags & RTNH_F_DEAD)) {
 283                         spin_unlock(&fib_info_lock);
 284                         return 0;
 285                 }
 286         }
 287
 288         spin_unlock(&fib_info_lock);
 289
 290         return -1;
 291 }
 292
 293 static inline size_t fib_nlmsg_size(struct fib_info *fi)
 294 {
 295         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 296                          + nla_total_size(4) /* RTA_TABLE */
 297                          + nla_total_size(4) /* RTA_DST */
 298                          + nla_total_size(4) /* RTA_PRIORITY */
 299                          + nla_total_size(4); /* RTA_PREFSRC */
 300
 301         /* space for nested metrics */
 302         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 303
 304         if (fi->fib_nhs) {
 305                 /* Also handles the special case fib_nhs == 1 */
 306
 307                 /* each nexthop is packed in an attribute */
 308                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 309
 310                 /* may contain flow and gateway attribute */
 311                 nhsize += 2 * nla_total_size(4);
 312
 313                 /* all nexthops are packed in a nested attribute */
 314                 payload += nla_total_size(fi->fib_nhs * nhsize);
 315         }
 316
 317         return payload;
 318 }
 319
 320 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 321                int dst_len, u32 tb_id, struct nl_info *info,
 322                unsigned int nlm_flags)
 323 {
 324         struct sk_buff *skb;
 325         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 326         int err = -ENOBUFS;
 327
 328         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 329         if (skb == NULL)
 330                 goto errout;
 331
 332         err = fib_dump_info(skb, info->pid, seq, event, tb_id,
 333                             fa->fa_type, fa->fa_scope, key, dst_len,
 334                             fa->fa_tos, fa->fa_info, nlm_flags);
 335         if (err < 0) {
 336                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 337                 WARN_ON(err == -EMSGSIZE);
 338                 kfree_skb(skb);
 339                 goto errout;
 340         }
 341         rtnl_notify(skb, info->nl_net, info->pid, RTNLGRP_IPV4_ROUTE,
 342                     info->nlh, GFP_KERNEL);
 343         return;
 344 errout:
 345         if (err < 0)
 346                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 347 }
 348
 349 /* Return the first fib alias matching TOS with
 350  * priority less than or equal to PRIO.
 351  */
 352 struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
 353 {
 354         if (fah) {
 355                 struct fib_alias *fa;
 356                 list_for_each_entry(fa, fah, fa_list) {
 357                         if (fa->fa_tos > tos)
 358                                 continue;
 359                         if (fa->fa_info->fib_priority >= prio ||
 360                             fa->fa_tos < tos)
 361                                 return fa;
 362                 }
 363         }
 364         return NULL;
 365 }
 366
 367 int fib_detect_death(struct fib_info *fi, int order,
 368                      struct fib_info **last_resort, int *last_idx, int dflt)
 369 {
 370         struct neighbour *n;
 371         int state = NUD_NONE;
 372
 373         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 374         if (n) {
 375                 state = n->nud_state;
 376                 neigh_release(n);
 377         }
 378         if (state == NUD_REACHABLE)
 379                 return 0;
 380         if ((state & NUD_VALID) && order != dflt)
 381                 return 0;
 382         if ((state & NUD_VALID) ||
 383             (*last_idx < 0 && order > dflt)) {
 384                 *last_resort = fi;
 385                 *last_idx = order;
 386         }
 387         return 1;
 388 }
 389
 390 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 391
 392 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
 393 {
 394         int nhs = 0;
 395
 396         while (rtnh_ok(rtnh, remaining)) {
 397                 nhs++;
 398                 rtnh = rtnh_next(rtnh, &remaining);
 399         }
 400
 401         /* leftover implies invalid nexthop configuration, discard it */
 402         return remaining > 0 ? 0 : nhs;
 403 }
 404
 405 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 406                        int remaining, struct fib_config *cfg)
 407 {
 408         change_nexthops(fi) {
 409                 int attrlen;
 410
 411                 if (!rtnh_ok(rtnh, remaining))
 412                         return -EINVAL;
 413
 414                 nexthop_nh->nh_flags =
 415                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 416                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
 417                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
 418
 419                 attrlen = rtnh_attrlen(rtnh);
 420                 if (attrlen > 0) {
 421                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 422
 423                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 424                         nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
 425 #ifdef CONFIG_NET_CLS_ROUTE
 426                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 427                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 428 #endif
 429                 }
 430
 431                 rtnh = rtnh_next(rtnh, &remaining);
 432         } endfor_nexthops(fi);
 433
 434         return 0;
 435 }
 436
 437 #endif
 438
 439 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
 440 {
 441 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 442         struct rtnexthop *rtnh;
 443         int remaining;
 444 #endif
 445
 446         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 447                 return 1;
 448
 449         if (cfg->fc_oif || cfg->fc_gw) {
 450                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 451                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 452                         return 0;
 453                 return 1;
 454         }
 455
 456 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 457         if (cfg->fc_mp == NULL)
 458                 return 0;
 459
 460         rtnh = cfg->fc_mp;
 461         remaining = cfg->fc_mp_len;
 462
 463         for_nexthops(fi) {
 464                 int attrlen;
 465
 466                 if (!rtnh_ok(rtnh, remaining))
 467                         return -EINVAL;
 468
 469                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 470                         return 1;
 471
 472                 attrlen = rtnh_attrlen(rtnh);
 473                 if (attrlen < 0) {
 474                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 475
 476                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 477                         if (nla && nla_get_be32(nla) != nh->nh_gw)
 478                                 return 1;
 479 #ifdef CONFIG_NET_CLS_ROUTE
 480                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 481                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 482                                 return 1;
 483 #endif
 484                 }
 485
 486                 rtnh = rtnh_next(rtnh, &remaining);
 487         } endfor_nexthops(fi);
 488 #endif
 489         return 0;
 490 }
 491
 492
 493 /*
 494  * Picture
 495  * -------
 496  *
 497  * Semantics of nexthop is very messy by historical reasons.
 498  * We have to take into account, that:
 499  * a) gateway can be actually local interface address,
 500  *    so that gatewayed route is direct.
 501  * b) gateway must be on-link address, possibly
 502  *    described not by an ifaddr, but also by a direct route.
 503  * c) If both gateway and interface are specified, they should not
 504  *    contradict.
 505  * d) If we use tunnel routes, gateway could be not on-link.
 506  *
 507  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 508  * results in pretty ugly and hairy code with obscure logic.
 509  *
 510  * I chose to generalized it instead, so that the size
 511  * of code does not increase practically, but it becomes
 512  * much more general.
 513  * Every prefix is assigned a "scope" value: "host" is local address,
 514  * "link" is direct route,
 515  * [ ... "site" ... "interior" ... ]
 516  * and "universe" is true gateway route with global meaning.
 517  *
 518  * Every prefix refers to a set of "nexthop"s (gw, oif),
 519  * where gw must have narrower scope. This recursion stops
 520  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 521  * which means that gw is forced to be on link.
 522  *
 523  * Code is still hairy, but now it is apparently logically
 524  * consistent and very flexible. F.e. as by-product it allows
 525  * to co-exists in peace independent exterior and interior
 526  * routing processes.
 527  *
 528  * Normally it looks as following.
 529  *
 530  * {universe prefix}  -> (gw, oif) [scope link]
 531  *                |
 532  *                |-> {link prefix} -> (gw, oif) [scope local]
 533  *                                      |
 534  *                                      |-> {local prefix} (terminal node)
 535  */
 536 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
 537                         struct fib_nh *nh)
 538 {
 539         int err;
 540         struct net *net;
 541         struct net_device *dev;
 542
 543         net = cfg->fc_nlinfo.nl_net;
 544         if (nh->nh_gw) {
 545                 struct fib_result res;
 546
 547                 if (nh->nh_flags & RTNH_F_ONLINK) {
 548
 549                         if (cfg->fc_scope >= RT_SCOPE_LINK)
 550                                 return -EINVAL;
 551                         if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
 552                                 return -EINVAL;
 553                         dev = __dev_get_by_index(net, nh->nh_oif);
 554                         if (!dev)
 555                                 return -ENODEV;
 556                         if (!(dev->flags & IFF_UP))
 557                                 return -ENETDOWN;
 558                         nh->nh_dev = dev;
 559                         dev_hold(dev);
 560                         nh->nh_scope = RT_SCOPE_LINK;
 561                         return 0;
 562                 }
 563                 rcu_read_lock();
 564                 {
 565                         struct flowi fl = {
 566                                 .nl_u = {
 567                                         .ip4_u = {
 568                                                 .daddr = nh->nh_gw,
 569                                                 .scope = cfg->fc_scope + 1,
 570                                         },
 571                                 },
 572                                 .oif = nh->nh_oif,
 573                         };
 574
 575                         /* It is not necessary, but requires a bit of thinking */
 576                         if (fl.fl4_scope < RT_SCOPE_LINK)
 577                                 fl.fl4_scope = RT_SCOPE_LINK;
 578                         err = fib_lookup(net, &fl, &res);
 579                         if (err) {
 580                                 rcu_read_unlock();
 581                                 return err;
 582                         }
 583                 }
 584                 err = -EINVAL;
 585                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 586                         goto out;
 587                 nh->nh_scope = res.scope;
 588                 nh->nh_oif = FIB_RES_OIF(res);
 589                 nh->nh_dev = dev = FIB_RES_DEV(res);
 590                 if (!dev)
 591                         goto out;
 592                 dev_hold(dev);
 593                 err = -ENETDOWN;
 594                 if (!(dev->flags & IFF_UP))
 595                         goto out;
 596                 err = 0;
 597 out:
 598                 rcu_read_unlock();
 599                 return err;
 600         } else {
 601                 struct in_device *in_dev;
 602
 603                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 604                         return -EINVAL;
 605
 606                 in_dev = inetdev_by_index(net, nh->nh_oif);
 607                 if (in_dev == NULL)
 608                         return -ENODEV;
 609                 if (!(in_dev->dev->flags & IFF_UP)) {
 610                         in_dev_put(in_dev);
 611                         return -ENETDOWN;
 612                 }
 613                 nh->nh_dev = in_dev->dev;
 614                 dev_hold(nh->nh_dev);
 615                 nh->nh_scope = RT_SCOPE_HOST;
 616                 in_dev_put(in_dev);
 617         }
 618         return 0;
 619 }
 620
 621 static inline unsigned int fib_laddr_hashfn(__be32 val)
 622 {
 623         unsigned int mask = (fib_hash_size - 1);
 624
 625         return ((__force u32)val ^
 626                 ((__force u32)val >> 7) ^
 627                 ((__force u32)val >> 14)) & mask;
 628 }
 629
 630 static struct hlist_head *fib_hash_alloc(int bytes)
 631 {
 632         if (bytes <= PAGE_SIZE)
 633                 return kzalloc(bytes, GFP_KERNEL);
 634         else
 635                 return (struct hlist_head *)
 636                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 637                                          get_order(bytes));
 638 }
 639
 640 static void fib_hash_free(struct hlist_head *hash, int bytes)
 641 {
 642         if (!hash)
 643                 return;
 644
 645         if (bytes <= PAGE_SIZE)
 646                 kfree(hash);
 647         else
 648                 free_pages((unsigned long) hash, get_order(bytes));
 649 }
 650
 651 static void fib_hash_move(struct hlist_head *new_info_hash,
 652                           struct hlist_head *new_laddrhash,
 653                           unsigned int new_size)
 654 {
 655         struct hlist_head *old_info_hash, *old_laddrhash;
 656         unsigned int old_size = fib_hash_size;
 657         unsigned int i, bytes;
 658
 659         spin_lock_bh(&fib_info_lock);
 660         old_info_hash = fib_info_hash;
 661         old_laddrhash = fib_info_laddrhash;
 662         fib_hash_size = new_size;
 663
 664         for (i = 0; i < old_size; i++) {
 665                 struct hlist_head *head = &fib_info_hash[i];
 666                 struct hlist_node *node, *n;
 667                 struct fib_info *fi;
 668
 669                 hlist_for_each_entry_safe(fi, node, n, head, fib_hash) {
 670                         struct hlist_head *dest;
 671                         unsigned int new_hash;
 672
 673                         hlist_del(&fi->fib_hash);
 674
 675                         new_hash = fib_info_hashfn(fi);
 676                         dest = &new_info_hash[new_hash];
 677                         hlist_add_head(&fi->fib_hash, dest);
 678                 }
 679         }
 680         fib_info_hash = new_info_hash;
 681
 682         for (i = 0; i < old_size; i++) {
 683                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 684                 struct hlist_node *node, *n;
 685                 struct fib_info *fi;
 686
 687                 hlist_for_each_entry_safe(fi, node, n, lhead, fib_lhash) {
 688                         struct hlist_head *ldest;
 689                         unsigned int new_hash;
 690
 691                         hlist_del(&fi->fib_lhash);
 692
 693                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 694                         ldest = &new_laddrhash[new_hash];
 695                         hlist_add_head(&fi->fib_lhash, ldest);
 696                 }
 697         }
 698         fib_info_laddrhash = new_laddrhash;
 699
 700         spin_unlock_bh(&fib_info_lock);
 701
 702         bytes = old_size * sizeof(struct hlist_head *);
 703         fib_hash_free(old_info_hash, bytes);
 704         fib_hash_free(old_laddrhash, bytes);
 705 }
 706
 707 struct fib_info *fib_create_info(struct fib_config *cfg)
 708 {
 709         int err;
 710         struct fib_info *fi = NULL;
 711         struct fib_info *ofi;
 712         int nhs = 1;
 713         struct net *net = cfg->fc_nlinfo.nl_net;
 714
 715         /* Fast check to catch the most weird cases */
 716         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
 717                 goto err_inval;
 718
 719 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 720         if (cfg->fc_mp) {
 721                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
 722                 if (nhs == 0)
 723                         goto err_inval;
 724         }
 725 #endif
 726
 727         err = -ENOBUFS;
 728         if (fib_info_cnt >= fib_hash_size) {
 729                 unsigned int new_size = fib_hash_size << 1;
 730                 struct hlist_head *new_info_hash;
 731                 struct hlist_head *new_laddrhash;
 732                 unsigned int bytes;
 733
 734                 if (!new_size)
 735                         new_size = 1;
 736                 bytes = new_size * sizeof(struct hlist_head *);
 737                 new_info_hash = fib_hash_alloc(bytes);
 738                 new_laddrhash = fib_hash_alloc(bytes);
 739                 if (!new_info_hash || !new_laddrhash) {
 740                         fib_hash_free(new_info_hash, bytes);
 741                         fib_hash_free(new_laddrhash, bytes);
 742                 } else
 743                         fib_hash_move(new_info_hash, new_laddrhash, new_size);
 744
 745                 if (!fib_hash_size)
 746                         goto failure;
 747         }
 748
 749         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
 750         if (fi == NULL)
 751                 goto failure;
 752         fib_info_cnt++;
 753
 754         fi->fib_net = hold_net(net);
 755         fi->fib_protocol = cfg->fc_protocol;
 756         fi->fib_flags = cfg->fc_flags;
 757         fi->fib_priority = cfg->fc_priority;
 758         fi->fib_prefsrc = cfg->fc_prefsrc;
 759
 760         fi->fib_nhs = nhs;
 761         change_nexthops(fi) {
 762                 nexthop_nh->nh_parent = fi;
 763         } endfor_nexthops(fi)
 764
 765         if (cfg->fc_mx) {
 766                 struct nlattr *nla;
 767                 int remaining;
 768
 769                 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 770                         int type = nla_type(nla);
 771
 772                         if (type) {
 773                                 if (type > RTAX_MAX)
 774                                         goto err_inval;
 775                                 fi->fib_metrics[type - 1] = nla_get_u32(nla);
 776                         }
 777                 }
 778         }
 779
 780         if (cfg->fc_mp) {
 781 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 782                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
 783                 if (err != 0)
 784                         goto failure;
 785                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
 786                         goto err_inval;
 787                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
 788                         goto err_inval;
 789 #ifdef CONFIG_NET_CLS_ROUTE
 790                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
 791                         goto err_inval;
 792 #endif
 793 #else
 794                 goto err_inval;
 795 #endif
 796         } else {
 797                 struct fib_nh *nh = fi->fib_nh;
 798
 799                 nh->nh_oif = cfg->fc_oif;
 800                 nh->nh_gw = cfg->fc_gw;
 801                 nh->nh_flags = cfg->fc_flags;
 802 #ifdef CONFIG_NET_CLS_ROUTE
 803                 nh->nh_tclassid = cfg->fc_flow;
 804 #endif
 805 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 806                 nh->nh_weight = 1;
 807 #endif
 808         }
 809
 810         if (fib_props[cfg->fc_type].error) {
 811                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
 812                         goto err_inval;
 813                 goto link_it;
 814         }
 815
 816         if (cfg->fc_scope > RT_SCOPE_HOST)
 817                 goto err_inval;
 818
 819         if (cfg->fc_scope == RT_SCOPE_HOST) {
 820                 struct fib_nh *nh = fi->fib_nh;
 821
 822                 /* Local address is added. */
 823                 if (nhs != 1 || nh->nh_gw)
 824                         goto err_inval;
 825                 nh->nh_scope = RT_SCOPE_NOWHERE;
 826                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
 827                 err = -ENODEV;
 828                 if (nh->nh_dev == NULL)
 829                         goto failure;
 830         } else {
 831                 change_nexthops(fi) {
 832                         err = fib_check_nh(cfg, fi, nexthop_nh);
 833                         if (err != 0)
 834                                 goto failure;
 835                 } endfor_nexthops(fi)
 836         }
 837
 838         if (fi->fib_prefsrc) {
 839                 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 840                     fi->fib_prefsrc != cfg->fc_dst)
 841                         if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
 842                                 goto err_inval;
 843         }
 844
 845 link_it:
 846         ofi = fib_find_info(fi);
 847         if (ofi) {
 848                 fi->fib_dead = 1;
 849                 free_fib_info(fi);
 850                 ofi->fib_treeref++;
 851                 return ofi;
 852         }
 853
 854         fi->fib_treeref++;
 855         atomic_inc(&fi->fib_clntref);
 856         spin_lock_bh(&fib_info_lock);
 857         hlist_add_head(&fi->fib_hash,
 858                        &fib_info_hash[fib_info_hashfn(fi)]);
 859         if (fi->fib_prefsrc) {
 860                 struct hlist_head *head;
 861
 862                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
 863                 hlist_add_head(&fi->fib_lhash, head);
 864         }
 865         change_nexthops(fi) {
 866                 struct hlist_head *head;
 867                 unsigned int hash;
 868
 869                 if (!nexthop_nh->nh_dev)
 870                         continue;
 871                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
 872                 head = &fib_info_devhash[hash];
 873                 hlist_add_head(&nexthop_nh->nh_hash, head);
 874         } endfor_nexthops(fi)
 875         spin_unlock_bh(&fib_info_lock);
 876         return fi;
 877
 878 err_inval:
 879         err = -EINVAL;
 880
 881 failure:
 882         if (fi) {
 883                 fi->fib_dead = 1;
 884                 free_fib_info(fi);
 885         }
 886
 887         return ERR_PTR(err);
 888 }
 889
 890 /* Note! fib_semantic_match intentionally uses  RCU list functions. */
 891 int fib_semantic_match(struct list_head *head, const struct flowi *flp,
 892                        struct fib_result *res, int prefixlen, int fib_flags)
 893 {
 894         struct fib_alias *fa;
 895         int nh_sel = 0;
 896
 897         list_for_each_entry_rcu(fa, head, fa_list) {
 898                 int err;
 899
 900                 if (fa->fa_tos &&
 901                     fa->fa_tos != flp->fl4_tos)
 902                         continue;
 903
 904                 if (fa->fa_scope < flp->fl4_scope)
 905                         continue;
 906
 907                 fa->fa_state |= FA_S_ACCESSED;
 908
 909                 err = fib_props[fa->fa_type].error;
 910                 if (err == 0) {
 911                         struct fib_info *fi = fa->fa_info;
 912
 913                         if (fi->fib_flags & RTNH_F_DEAD)
 914                                 continue;
 915
 916                         switch (fa->fa_type) {
 917                         case RTN_UNICAST:
 918                         case RTN_LOCAL:
 919                         case RTN_BROADCAST:
 920                         case RTN_ANYCAST:
 921                         case RTN_MULTICAST:
 922                                 for_nexthops(fi) {
 923                                         if (nh->nh_flags & RTNH_F_DEAD)
 924                                                 continue;
 925                                         if (!flp->oif || flp->oif == nh->nh_oif)
 926                                                 break;
 927                                 }
 928 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 929                                 if (nhsel < fi->fib_nhs) {
 930                                         nh_sel = nhsel;
 931                                         goto out_fill_res;
 932                                 }
 933 #else
 934                                 if (nhsel < 1)
 935                                         goto out_fill_res;
 936 #endif
 937                                 endfor_nexthops(fi);
 938                                 continue;
 939
 940                         default:
 941                                 pr_warning("fib_semantic_match bad type %#x\n",
 942                                            fa->fa_type);
 943                                 return -EINVAL;
 944                         }
 945                 }
 946                 return err;
 947         }
 948         return 1;
 949
 950 out_fill_res:
 951         res->prefixlen = prefixlen;
 952         res->nh_sel = nh_sel;
 953         res->type = fa->fa_type;
 954         res->scope = fa->fa_scope;
 955         res->fi = fa->fa_info;
 956         if (!(fib_flags & FIB_LOOKUP_NOREF))
 957                 atomic_inc(&res->fi->fib_clntref);
 958         return 0;
 959 }
 960
 961 /* Find appropriate source address to this destination */
 962
 963 __be32 __fib_res_prefsrc(struct fib_result *res)
 964 {
 965         return inet_select_addr(FIB_RES_DEV(*res), FIB_RES_GW(*res), res->scope);
 966 }
 967
 968 int fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
 969                   u32 tb_id, u8 type, u8 scope, __be32 dst, int dst_len, u8 tos,
 970                   struct fib_info *fi, unsigned int flags)
 971 {
 972         struct nlmsghdr *nlh;
 973         struct rtmsg *rtm;
 974
 975         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*rtm), flags);
 976         if (nlh == NULL)
 977                 return -EMSGSIZE;
 978
 979         rtm = nlmsg_data(nlh);
 980         rtm->rtm_family = AF_INET;
 981         rtm->rtm_dst_len = dst_len;
 982         rtm->rtm_src_len = 0;
 983         rtm->rtm_tos = tos;
 984         if (tb_id < 256)
 985                 rtm->rtm_table = tb_id;
 986         else
 987                 rtm->rtm_table = RT_TABLE_COMPAT;
 988         NLA_PUT_U32(skb, RTA_TABLE, tb_id);
 989         rtm->rtm_type = type;
 990         rtm->rtm_flags = fi->fib_flags;
 991         rtm->rtm_scope = scope;
 992         rtm->rtm_protocol = fi->fib_protocol;
 993
 994         if (rtm->rtm_dst_len)
 995                 NLA_PUT_BE32(skb, RTA_DST, dst);
 996
 997         if (fi->fib_priority)
 998                 NLA_PUT_U32(skb, RTA_PRIORITY, fi->fib_priority);
 999
1000         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1001                 goto nla_put_failure;
1002
1003         if (fi->fib_prefsrc)
1004                 NLA_PUT_BE32(skb, RTA_PREFSRC, fi->fib_prefsrc);
1005
1006         if (fi->fib_nhs == 1) {
1007                 if (fi->fib_nh->nh_gw)
1008                         NLA_PUT_BE32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw);
1009
1010                 if (fi->fib_nh->nh_oif)
1011                         NLA_PUT_U32(skb, RTA_OIF, fi->fib_nh->nh_oif);
1012 #ifdef CONFIG_NET_CLS_ROUTE
1013                 if (fi->fib_nh[0].nh_tclassid)
1014                         NLA_PUT_U32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid);
1015 #endif
1016         }
1017 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1018         if (fi->fib_nhs > 1) {
1019                 struct rtnexthop *rtnh;
1020                 struct nlattr *mp;
1021
1022                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1023                 if (mp == NULL)
1024                         goto nla_put_failure;
1025
1026                 for_nexthops(fi) {
1027                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1028                         if (rtnh == NULL)
1029                                 goto nla_put_failure;
1030
1031                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1032                         rtnh->rtnh_hops = nh->nh_weight - 1;
1033                         rtnh->rtnh_ifindex = nh->nh_oif;
1034
1035                         if (nh->nh_gw)
1036                                 NLA_PUT_BE32(skb, RTA_GATEWAY, nh->nh_gw);
1037 #ifdef CONFIG_NET_CLS_ROUTE
1038                         if (nh->nh_tclassid)
1039                                 NLA_PUT_U32(skb, RTA_FLOW, nh->nh_tclassid);
1040 #endif
1041                         /* length of rtnetlink header + attributes */
1042                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1043                 } endfor_nexthops(fi);
1044
1045                 nla_nest_end(skb, mp);
1046         }
1047 #endif
1048         return nlmsg_end(skb, nlh);
1049
1050 nla_put_failure:
1051         nlmsg_cancel(skb, nlh);
1052         return -EMSGSIZE;
1053 }
1054
1055 /*
1056  * Update FIB if:
1057  * - local address disappeared -> we must delete all the entries
1058  *   referring to it.
1059  * - device went down -> we must shutdown all nexthops going via it.
1060  */
1061 int fib_sync_down_addr(struct net *net, __be32 local)
1062 {
1063         int ret = 0;
1064         unsigned int hash = fib_laddr_hashfn(local);
1065         struct hlist_head *head = &fib_info_laddrhash[hash];
1066         struct hlist_node *node;
1067         struct fib_info *fi;
1068
1069         if (fib_info_laddrhash == NULL || local == 0)
1070                 return 0;
1071
1072         hlist_for_each_entry(fi, node, head, fib_lhash) {
1073                 if (!net_eq(fi->fib_net, net))
1074                         continue;
1075                 if (fi->fib_prefsrc == local) {
1076                         fi->fib_flags |= RTNH_F_DEAD;
1077                         ret++;
1078                 }
1079         }
1080         return ret;
1081 }
1082
1083 int fib_sync_down_dev(struct net_device *dev, int force)
1084 {
1085         int ret = 0;
1086         int scope = RT_SCOPE_NOWHERE;
1087         struct fib_info *prev_fi = NULL;
1088         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1089         struct hlist_head *head = &fib_info_devhash[hash];
1090         struct hlist_node *node;
1091         struct fib_nh *nh;
1092
1093         if (force)
1094                 scope = -1;
1095
1096         hlist_for_each_entry(nh, node, head, nh_hash) {
1097                 struct fib_info *fi = nh->nh_parent;
1098                 int dead;
1099
1100                 BUG_ON(!fi->fib_nhs);
1101                 if (nh->nh_dev != dev || fi == prev_fi)
1102                         continue;
1103                 prev_fi = fi;
1104                 dead = 0;
1105                 change_nexthops(fi) {
1106                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1107                                 dead++;
1108                         else if (nexthop_nh->nh_dev == dev &&
1109                                  nexthop_nh->nh_scope != scope) {
1110                                 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1111 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1112                                 spin_lock_bh(&fib_multipath_lock);
1113                                 fi->fib_power -= nexthop_nh->nh_power;
1114                                 nexthop_nh->nh_power = 0;
1115                                 spin_unlock_bh(&fib_multipath_lock);
1116 #endif
1117                                 dead++;
1118                         }
1119 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1120                         if (force > 1 && nexthop_nh->nh_dev == dev) {
1121                                 dead = fi->fib_nhs;
1122                                 break;
1123                         }
1124 #endif
1125                 } endfor_nexthops(fi)
1126                 if (dead == fi->fib_nhs) {
1127                         fi->fib_flags |= RTNH_F_DEAD;
1128                         ret++;
1129                 }
1130         }
1131
1132         return ret;
1133 }
1134
1135 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1136
1137 /*
1138  * Dead device goes up. We wake up dead nexthops.
1139  * It takes sense only on multipath routes.
1140  */
1141 int fib_sync_up(struct net_device *dev)
1142 {
1143         struct fib_info *prev_fi;
1144         unsigned int hash;
1145         struct hlist_head *head;
1146         struct hlist_node *node;
1147         struct fib_nh *nh;
1148         int ret;
1149
1150         if (!(dev->flags & IFF_UP))
1151                 return 0;
1152
1153         prev_fi = NULL;
1154         hash = fib_devindex_hashfn(dev->ifindex);
1155         head = &fib_info_devhash[hash];
1156         ret = 0;
1157
1158         hlist_for_each_entry(nh, node, head, nh_hash) {
1159                 struct fib_info *fi = nh->nh_parent;
1160                 int alive;
1161
1162                 BUG_ON(!fi->fib_nhs);
1163                 if (nh->nh_dev != dev || fi == prev_fi)
1164                         continue;
1165
1166                 prev_fi = fi;
1167                 alive = 0;
1168                 change_nexthops(fi) {
1169                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1170                                 alive++;
1171                                 continue;
1172                         }
1173                         if (nexthop_nh->nh_dev == NULL ||
1174                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1175                                 continue;
1176                         if (nexthop_nh->nh_dev != dev ||
1177                             !__in_dev_get_rtnl(dev))
1178                                 continue;
1179                         alive++;
1180                         spin_lock_bh(&fib_multipath_lock);
1181                         nexthop_nh->nh_power = 0;
1182                         nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1183                         spin_unlock_bh(&fib_multipath_lock);
1184                 } endfor_nexthops(fi)
1185
1186                 if (alive > 0) {
1187                         fi->fib_flags &= ~RTNH_F_DEAD;
1188                         ret++;
1189                 }
1190         }
1191
1192         return ret;
1193 }
1194
1195 /*
1196  * The algorithm is suboptimal, but it provides really
1197  * fair weighted route distribution.
1198  */
1199 void fib_select_multipath(const struct flowi *flp, struct fib_result *res)
1200 {
1201         struct fib_info *fi = res->fi;
1202         int w;
1203
1204         spin_lock_bh(&fib_multipath_lock);
1205         if (fi->fib_power <= 0) {
1206                 int power = 0;
1207                 change_nexthops(fi) {
1208                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1209                                 power += nexthop_nh->nh_weight;
1210                                 nexthop_nh->nh_power = nexthop_nh->nh_weight;
1211                         }
1212                 } endfor_nexthops(fi);
1213                 fi->fib_power = power;
1214                 if (power <= 0) {
1215                         spin_unlock_bh(&fib_multipath_lock);
1216                         /* Race condition: route has just become dead. */
1217                         res->nh_sel = 0;
1218                         return;
1219                 }
1220         }
1221
1222
1223         /* w should be random number [0..fi->fib_power-1],
1224          * it is pretty bad approximation.
1225          */
1226
1227         w = jiffies % fi->fib_power;
1228
1229         change_nexthops(fi) {
1230                 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1231                     nexthop_nh->nh_power) {
1232                         w -= nexthop_nh->nh_power;
1233                         if (w <= 0) {
1234                                 nexthop_nh->nh_power--;
1235                                 fi->fib_power--;
1236                                 res->nh_sel = nhsel;
1237                                 spin_unlock_bh(&fib_multipath_lock);
1238                                 return;
1239                         }
1240                 }
1241         } endfor_nexthops(fi);
1242
1243         /* Race condition: route has just become dead. */
1244         res->nh_sel = 0;
1245         spin_unlock_bh(&fib_multipath_lock);
1246 }
1247 #endif