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                                 .fl4_dst = nh->nh_gw,
 567                                 .fl4_scope = cfg->fc_scope + 1,
 568                                 .oif = nh->nh_oif,
 569                         };
 570
 571                         /* It is not necessary, but requires a bit of thinking */
 572                         if (fl.fl4_scope < RT_SCOPE_LINK)
 573                                 fl.fl4_scope = RT_SCOPE_LINK;
 574                         err = fib_lookup(net, &fl, &res);
 575                         if (err) {
 576                                 rcu_read_unlock();
 577                                 return err;
 578                         }
 579                 }
 580                 err = -EINVAL;
 581                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 582                         goto out;
 583                 nh->nh_scope = res.scope;
 584                 nh->nh_oif = FIB_RES_OIF(res);
 585                 nh->nh_dev = dev = FIB_RES_DEV(res);
 586                 if (!dev)
 587                         goto out;
 588                 dev_hold(dev);
 589                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 590         } else {
 591                 struct in_device *in_dev;
 592
 593                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 594                         return -EINVAL;
 595
 596                 rcu_read_lock();
 597                 err = -ENODEV;
 598                 in_dev = inetdev_by_index(net, nh->nh_oif);
 599                 if (in_dev == NULL)
 600                         goto out;
 601                 err = -ENETDOWN;
 602                 if (!(in_dev->dev->flags & IFF_UP))
 603                         goto out;
 604                 nh->nh_dev = in_dev->dev;
 605                 dev_hold(nh->nh_dev);
 606                 nh->nh_scope = RT_SCOPE_HOST;
 607                 err = 0;
 608         }
 609 out:
 610         rcu_read_unlock();
 611         return err;
 612 }
 613
 614 static inline unsigned int fib_laddr_hashfn(__be32 val)
 615 {
 616         unsigned int mask = (fib_hash_size - 1);
 617
 618         return ((__force u32)val ^
 619                 ((__force u32)val >> 7) ^
 620                 ((__force u32)val >> 14)) & mask;
 621 }
 622
 623 static struct hlist_head *fib_hash_alloc(int bytes)
 624 {
 625         if (bytes <= PAGE_SIZE)
 626                 return kzalloc(bytes, GFP_KERNEL);
 627         else
 628                 return (struct hlist_head *)
 629                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 630                                          get_order(bytes));
 631 }
 632
 633 static void fib_hash_free(struct hlist_head *hash, int bytes)
 634 {
 635         if (!hash)
 636                 return;
 637
 638         if (bytes <= PAGE_SIZE)
 639                 kfree(hash);
 640         else
 641                 free_pages((unsigned long) hash, get_order(bytes));
 642 }
 643
 644 static void fib_hash_move(struct hlist_head *new_info_hash,
 645                           struct hlist_head *new_laddrhash,
 646                           unsigned int new_size)
 647 {
 648         struct hlist_head *old_info_hash, *old_laddrhash;
 649         unsigned int old_size = fib_hash_size;
 650         unsigned int i, bytes;
 651
 652         spin_lock_bh(&fib_info_lock);
 653         old_info_hash = fib_info_hash;
 654         old_laddrhash = fib_info_laddrhash;
 655         fib_hash_size = new_size;
 656
 657         for (i = 0; i < old_size; i++) {
 658                 struct hlist_head *head = &fib_info_hash[i];
 659                 struct hlist_node *node, *n;
 660                 struct fib_info *fi;
 661
 662                 hlist_for_each_entry_safe(fi, node, n, head, fib_hash) {
 663                         struct hlist_head *dest;
 664                         unsigned int new_hash;
 665
 666                         hlist_del(&fi->fib_hash);
 667
 668                         new_hash = fib_info_hashfn(fi);
 669                         dest = &new_info_hash[new_hash];
 670                         hlist_add_head(&fi->fib_hash, dest);
 671                 }
 672         }
 673         fib_info_hash = new_info_hash;
 674
 675         for (i = 0; i < old_size; i++) {
 676                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 677                 struct hlist_node *node, *n;
 678                 struct fib_info *fi;
 679
 680                 hlist_for_each_entry_safe(fi, node, n, lhead, fib_lhash) {
 681                         struct hlist_head *ldest;
 682                         unsigned int new_hash;
 683
 684                         hlist_del(&fi->fib_lhash);
 685
 686                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 687                         ldest = &new_laddrhash[new_hash];
 688                         hlist_add_head(&fi->fib_lhash, ldest);
 689                 }
 690         }
 691         fib_info_laddrhash = new_laddrhash;
 692
 693         spin_unlock_bh(&fib_info_lock);
 694
 695         bytes = old_size * sizeof(struct hlist_head *);
 696         fib_hash_free(old_info_hash, bytes);
 697         fib_hash_free(old_laddrhash, bytes);
 698 }
 699
 700 struct fib_info *fib_create_info(struct fib_config *cfg)
 701 {
 702         int err;
 703         struct fib_info *fi = NULL;
 704         struct fib_info *ofi;
 705         int nhs = 1;
 706         struct net *net = cfg->fc_nlinfo.nl_net;
 707
 708         /* Fast check to catch the most weird cases */
 709         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
 710                 goto err_inval;
 711
 712 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 713         if (cfg->fc_mp) {
 714                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
 715                 if (nhs == 0)
 716                         goto err_inval;
 717         }
 718 #endif
 719
 720         err = -ENOBUFS;
 721         if (fib_info_cnt >= fib_hash_size) {
 722                 unsigned int new_size = fib_hash_size << 1;
 723                 struct hlist_head *new_info_hash;
 724                 struct hlist_head *new_laddrhash;
 725                 unsigned int bytes;
 726
 727                 if (!new_size)
 728                         new_size = 1;
 729                 bytes = new_size * sizeof(struct hlist_head *);
 730                 new_info_hash = fib_hash_alloc(bytes);
 731                 new_laddrhash = fib_hash_alloc(bytes);
 732                 if (!new_info_hash || !new_laddrhash) {
 733                         fib_hash_free(new_info_hash, bytes);
 734                         fib_hash_free(new_laddrhash, bytes);
 735                 } else
 736                         fib_hash_move(new_info_hash, new_laddrhash, new_size);
 737
 738                 if (!fib_hash_size)
 739                         goto failure;
 740         }
 741
 742         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
 743         if (fi == NULL)
 744                 goto failure;
 745         fib_info_cnt++;
 746
 747         fi->fib_net = hold_net(net);
 748         fi->fib_protocol = cfg->fc_protocol;
 749         fi->fib_flags = cfg->fc_flags;
 750         fi->fib_priority = cfg->fc_priority;
 751         fi->fib_prefsrc = cfg->fc_prefsrc;
 752
 753         fi->fib_nhs = nhs;
 754         change_nexthops(fi) {
 755                 nexthop_nh->nh_parent = fi;
 756         } endfor_nexthops(fi)
 757
 758         if (cfg->fc_mx) {
 759                 struct nlattr *nla;
 760                 int remaining;
 761
 762                 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 763                         int type = nla_type(nla);
 764
 765                         if (type) {
 766                                 if (type > RTAX_MAX)
 767                                         goto err_inval;
 768                                 fi->fib_metrics[type - 1] = nla_get_u32(nla);
 769                         }
 770                 }
 771         }
 772
 773         if (cfg->fc_mp) {
 774 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 775                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
 776                 if (err != 0)
 777                         goto failure;
 778                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
 779                         goto err_inval;
 780                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
 781                         goto err_inval;
 782 #ifdef CONFIG_NET_CLS_ROUTE
 783                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
 784                         goto err_inval;
 785 #endif
 786 #else
 787                 goto err_inval;
 788 #endif
 789         } else {
 790                 struct fib_nh *nh = fi->fib_nh;
 791
 792                 nh->nh_oif = cfg->fc_oif;
 793                 nh->nh_gw = cfg->fc_gw;
 794                 nh->nh_flags = cfg->fc_flags;
 795 #ifdef CONFIG_NET_CLS_ROUTE
 796                 nh->nh_tclassid = cfg->fc_flow;
 797 #endif
 798 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 799                 nh->nh_weight = 1;
 800 #endif
 801         }
 802
 803         if (fib_props[cfg->fc_type].error) {
 804                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
 805                         goto err_inval;
 806                 goto link_it;
 807         }
 808
 809         if (cfg->fc_scope > RT_SCOPE_HOST)
 810                 goto err_inval;
 811
 812         if (cfg->fc_scope == RT_SCOPE_HOST) {
 813                 struct fib_nh *nh = fi->fib_nh;
 814
 815                 /* Local address is added. */
 816                 if (nhs != 1 || nh->nh_gw)
 817                         goto err_inval;
 818                 nh->nh_scope = RT_SCOPE_NOWHERE;
 819                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
 820                 err = -ENODEV;
 821                 if (nh->nh_dev == NULL)
 822                         goto failure;
 823         } else {
 824                 change_nexthops(fi) {
 825                         err = fib_check_nh(cfg, fi, nexthop_nh);
 826                         if (err != 0)
 827                                 goto failure;
 828                 } endfor_nexthops(fi)
 829         }
 830
 831         if (fi->fib_prefsrc) {
 832                 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 833                     fi->fib_prefsrc != cfg->fc_dst)
 834                         if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
 835                                 goto err_inval;
 836         }
 837
 838 link_it:
 839         ofi = fib_find_info(fi);
 840         if (ofi) {
 841                 fi->fib_dead = 1;
 842                 free_fib_info(fi);
 843                 ofi->fib_treeref++;
 844                 return ofi;
 845         }
 846
 847         fi->fib_treeref++;
 848         atomic_inc(&fi->fib_clntref);
 849         spin_lock_bh(&fib_info_lock);
 850         hlist_add_head(&fi->fib_hash,
 851                        &fib_info_hash[fib_info_hashfn(fi)]);
 852         if (fi->fib_prefsrc) {
 853                 struct hlist_head *head;
 854
 855                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
 856                 hlist_add_head(&fi->fib_lhash, head);
 857         }
 858         change_nexthops(fi) {
 859                 struct hlist_head *head;
 860                 unsigned int hash;
 861
 862                 if (!nexthop_nh->nh_dev)
 863                         continue;
 864                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
 865                 head = &fib_info_devhash[hash];
 866                 hlist_add_head(&nexthop_nh->nh_hash, head);
 867         } endfor_nexthops(fi)
 868         spin_unlock_bh(&fib_info_lock);
 869         return fi;
 870
 871 err_inval:
 872         err = -EINVAL;
 873
 874 failure:
 875         if (fi) {
 876                 fi->fib_dead = 1;
 877                 free_fib_info(fi);
 878         }
 879
 880         return ERR_PTR(err);
 881 }
 882
 883 /* Note! fib_semantic_match intentionally uses  RCU list functions. */
 884 int fib_semantic_match(struct list_head *head, const struct flowi *flp,
 885                        struct fib_result *res, int prefixlen, int fib_flags)
 886 {
 887         struct fib_alias *fa;
 888         int nh_sel = 0;
 889
 890         list_for_each_entry_rcu(fa, head, fa_list) {
 891                 int err;
 892
 893                 if (fa->fa_tos &&
 894                     fa->fa_tos != flp->fl4_tos)
 895                         continue;
 896
 897                 if (fa->fa_scope < flp->fl4_scope)
 898                         continue;
 899
 900                 fib_alias_accessed(fa);
 901
 902                 err = fib_props[fa->fa_type].error;
 903                 if (err == 0) {
 904                         struct fib_info *fi = fa->fa_info;
 905
 906                         if (fi->fib_flags & RTNH_F_DEAD)
 907                                 continue;
 908
 909                         switch (fa->fa_type) {
 910                         case RTN_UNICAST:
 911                         case RTN_LOCAL:
 912                         case RTN_BROADCAST:
 913                         case RTN_ANYCAST:
 914                         case RTN_MULTICAST:
 915                                 for_nexthops(fi) {
 916                                         if (nh->nh_flags & RTNH_F_DEAD)
 917                                                 continue;
 918                                         if (!flp->oif || flp->oif == nh->nh_oif)
 919                                                 break;
 920                                 }
 921 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 922                                 if (nhsel < fi->fib_nhs) {
 923                                         nh_sel = nhsel;
 924                                         goto out_fill_res;
 925                                 }
 926 #else
 927                                 if (nhsel < 1)
 928                                         goto out_fill_res;
 929 #endif
 930                                 endfor_nexthops(fi);
 931                                 continue;
 932
 933                         default:
 934                                 pr_warning("fib_semantic_match bad type %#x\n",
 935                                            fa->fa_type);
 936                                 return -EINVAL;
 937                         }
 938                 }
 939                 return err;
 940         }
 941         return 1;
 942
 943 out_fill_res:
 944         res->prefixlen = prefixlen;
 945         res->nh_sel = nh_sel;
 946         res->type = fa->fa_type;
 947         res->scope = fa->fa_scope;
 948         res->fi = fa->fa_info;
 949         if (!(fib_flags & FIB_LOOKUP_NOREF))
 950                 atomic_inc(&res->fi->fib_clntref);
 951         return 0;
 952 }
 953
 954 /* Find appropriate source address to this destination */
 955
 956 __be32 __fib_res_prefsrc(struct fib_result *res)
 957 {
 958         return inet_select_addr(FIB_RES_DEV(*res), FIB_RES_GW(*res), res->scope);
 959 }
 960
 961 int fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
 962                   u32 tb_id, u8 type, u8 scope, __be32 dst, int dst_len, u8 tos,
 963                   struct fib_info *fi, unsigned int flags)
 964 {
 965         struct nlmsghdr *nlh;
 966         struct rtmsg *rtm;
 967
 968         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*rtm), flags);
 969         if (nlh == NULL)
 970                 return -EMSGSIZE;
 971
 972         rtm = nlmsg_data(nlh);
 973         rtm->rtm_family = AF_INET;
 974         rtm->rtm_dst_len = dst_len;
 975         rtm->rtm_src_len = 0;
 976         rtm->rtm_tos = tos;
 977         if (tb_id < 256)
 978                 rtm->rtm_table = tb_id;
 979         else
 980                 rtm->rtm_table = RT_TABLE_COMPAT;
 981         NLA_PUT_U32(skb, RTA_TABLE, tb_id);
 982         rtm->rtm_type = type;
 983         rtm->rtm_flags = fi->fib_flags;
 984         rtm->rtm_scope = scope;
 985         rtm->rtm_protocol = fi->fib_protocol;
 986
 987         if (rtm->rtm_dst_len)
 988                 NLA_PUT_BE32(skb, RTA_DST, dst);
 989
 990         if (fi->fib_priority)
 991                 NLA_PUT_U32(skb, RTA_PRIORITY, fi->fib_priority);
 992
 993         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
 994                 goto nla_put_failure;
 995
 996         if (fi->fib_prefsrc)
 997                 NLA_PUT_BE32(skb, RTA_PREFSRC, fi->fib_prefsrc);
 998
 999         if (fi->fib_nhs == 1) {
1000                 if (fi->fib_nh->nh_gw)
1001                         NLA_PUT_BE32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw);
1002
1003                 if (fi->fib_nh->nh_oif)
1004                         NLA_PUT_U32(skb, RTA_OIF, fi->fib_nh->nh_oif);
1005 #ifdef CONFIG_NET_CLS_ROUTE
1006                 if (fi->fib_nh[0].nh_tclassid)
1007                         NLA_PUT_U32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid);
1008 #endif
1009         }
1010 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1011         if (fi->fib_nhs > 1) {
1012                 struct rtnexthop *rtnh;
1013                 struct nlattr *mp;
1014
1015                 mp = nla_nest_start(skb, RTA_MULTIPATH);
1016                 if (mp == NULL)
1017                         goto nla_put_failure;
1018
1019                 for_nexthops(fi) {
1020                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1021                         if (rtnh == NULL)
1022                                 goto nla_put_failure;
1023
1024                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1025                         rtnh->rtnh_hops = nh->nh_weight - 1;
1026                         rtnh->rtnh_ifindex = nh->nh_oif;
1027
1028                         if (nh->nh_gw)
1029                                 NLA_PUT_BE32(skb, RTA_GATEWAY, nh->nh_gw);
1030 #ifdef CONFIG_NET_CLS_ROUTE
1031                         if (nh->nh_tclassid)
1032                                 NLA_PUT_U32(skb, RTA_FLOW, nh->nh_tclassid);
1033 #endif
1034                         /* length of rtnetlink header + attributes */
1035                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1036                 } endfor_nexthops(fi);
1037
1038                 nla_nest_end(skb, mp);
1039         }
1040 #endif
1041         return nlmsg_end(skb, nlh);
1042
1043 nla_put_failure:
1044         nlmsg_cancel(skb, nlh);
1045         return -EMSGSIZE;
1046 }
1047
1048 /*
1049  * Update FIB if:
1050  * - local address disappeared -> we must delete all the entries
1051  *   referring to it.
1052  * - device went down -> we must shutdown all nexthops going via it.
1053  */
1054 int fib_sync_down_addr(struct net *net, __be32 local)
1055 {
1056         int ret = 0;
1057         unsigned int hash = fib_laddr_hashfn(local);
1058         struct hlist_head *head = &fib_info_laddrhash[hash];
1059         struct hlist_node *node;
1060         struct fib_info *fi;
1061
1062         if (fib_info_laddrhash == NULL || local == 0)
1063                 return 0;
1064
1065         hlist_for_each_entry(fi, node, head, fib_lhash) {
1066                 if (!net_eq(fi->fib_net, net))
1067                         continue;
1068                 if (fi->fib_prefsrc == local) {
1069                         fi->fib_flags |= RTNH_F_DEAD;
1070                         ret++;
1071                 }
1072         }
1073         return ret;
1074 }
1075
1076 int fib_sync_down_dev(struct net_device *dev, int force)
1077 {
1078         int ret = 0;
1079         int scope = RT_SCOPE_NOWHERE;
1080         struct fib_info *prev_fi = NULL;
1081         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1082         struct hlist_head *head = &fib_info_devhash[hash];
1083         struct hlist_node *node;
1084         struct fib_nh *nh;
1085
1086         if (force)
1087                 scope = -1;
1088
1089         hlist_for_each_entry(nh, node, head, nh_hash) {
1090                 struct fib_info *fi = nh->nh_parent;
1091                 int dead;
1092
1093                 BUG_ON(!fi->fib_nhs);
1094                 if (nh->nh_dev != dev || fi == prev_fi)
1095                         continue;
1096                 prev_fi = fi;
1097                 dead = 0;
1098                 change_nexthops(fi) {
1099                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1100                                 dead++;
1101                         else if (nexthop_nh->nh_dev == dev &&
1102                                  nexthop_nh->nh_scope != scope) {
1103                                 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1104 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1105                                 spin_lock_bh(&fib_multipath_lock);
1106                                 fi->fib_power -= nexthop_nh->nh_power;
1107                                 nexthop_nh->nh_power = 0;
1108                                 spin_unlock_bh(&fib_multipath_lock);
1109 #endif
1110                                 dead++;
1111                         }
1112 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1113                         if (force > 1 && nexthop_nh->nh_dev == dev) {
1114                                 dead = fi->fib_nhs;
1115                                 break;
1116                         }
1117 #endif
1118                 } endfor_nexthops(fi)
1119                 if (dead == fi->fib_nhs) {
1120                         fi->fib_flags |= RTNH_F_DEAD;
1121                         ret++;
1122                 }
1123         }
1124
1125         return ret;
1126 }
1127
1128 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1129
1130 /*
1131  * Dead device goes up. We wake up dead nexthops.
1132  * It takes sense only on multipath routes.
1133  */
1134 int fib_sync_up(struct net_device *dev)
1135 {
1136         struct fib_info *prev_fi;
1137         unsigned int hash;
1138         struct hlist_head *head;
1139         struct hlist_node *node;
1140         struct fib_nh *nh;
1141         int ret;
1142
1143         if (!(dev->flags & IFF_UP))
1144                 return 0;
1145
1146         prev_fi = NULL;
1147         hash = fib_devindex_hashfn(dev->ifindex);
1148         head = &fib_info_devhash[hash];
1149         ret = 0;
1150
1151         hlist_for_each_entry(nh, node, head, nh_hash) {
1152                 struct fib_info *fi = nh->nh_parent;
1153                 int alive;
1154
1155                 BUG_ON(!fi->fib_nhs);
1156                 if (nh->nh_dev != dev || fi == prev_fi)
1157                         continue;
1158
1159                 prev_fi = fi;
1160                 alive = 0;
1161                 change_nexthops(fi) {
1162                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1163                                 alive++;
1164                                 continue;
1165                         }
1166                         if (nexthop_nh->nh_dev == NULL ||
1167                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1168                                 continue;
1169                         if (nexthop_nh->nh_dev != dev ||
1170                             !__in_dev_get_rtnl(dev))
1171                                 continue;
1172                         alive++;
1173                         spin_lock_bh(&fib_multipath_lock);
1174                         nexthop_nh->nh_power = 0;
1175                         nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1176                         spin_unlock_bh(&fib_multipath_lock);
1177                 } endfor_nexthops(fi)
1178
1179                 if (alive > 0) {
1180                         fi->fib_flags &= ~RTNH_F_DEAD;
1181                         ret++;
1182                 }
1183         }
1184
1185         return ret;
1186 }
1187
1188 /*
1189  * The algorithm is suboptimal, but it provides really
1190  * fair weighted route distribution.
1191  */
1192 void fib_select_multipath(const struct flowi *flp, struct fib_result *res)
1193 {
1194         struct fib_info *fi = res->fi;
1195         int w;
1196
1197         spin_lock_bh(&fib_multipath_lock);
1198         if (fi->fib_power <= 0) {
1199                 int power = 0;
1200                 change_nexthops(fi) {
1201                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1202                                 power += nexthop_nh->nh_weight;
1203                                 nexthop_nh->nh_power = nexthop_nh->nh_weight;
1204                         }
1205                 } endfor_nexthops(fi);
1206                 fi->fib_power = power;
1207                 if (power <= 0) {
1208                         spin_unlock_bh(&fib_multipath_lock);
1209                         /* Race condition: route has just become dead. */
1210                         res->nh_sel = 0;
1211                         return;
1212                 }
1213         }
1214
1215
1216         /* w should be random number [0..fi->fib_power-1],
1217          * it is pretty bad approximation.
1218          */
1219
1220         w = jiffies % fi->fib_power;
1221
1222         change_nexthops(fi) {
1223                 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1224                     nexthop_nh->nh_power) {
1225                         w -= nexthop_nh->nh_power;
1226                         if (w <= 0) {
1227                                 nexthop_nh->nh_power--;
1228                                 fi->fib_power--;
1229                                 res->nh_sel = nhsel;
1230                                 spin_unlock_bh(&fib_multipath_lock);
1231                                 return;
1232                         }
1233                 }
1234         } endfor_nexthops(fi);
1235
1236         /* Race condition: route has just become dead. */
1237         res->nh_sel = 0;
1238         spin_unlock_bh(&fib_multipath_lock);
1239 }
1240 #endif