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_info_hash_size;
  53 static unsigned int fib_info_cnt;
  54
  55 #define DEVINDEX_HASHBITS 8
  56 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
  57 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
  58
  59 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  60
  61 static DEFINE_SPINLOCK(fib_multipath_lock);
  62
  63 #define for_nexthops(fi) {                                              \
  64         int nhsel; const struct fib_nh *nh;                             \
  65         for (nhsel = 0, nh = (fi)->fib_nh;                              \
  66              nhsel < (fi)->fib_nhs;                                     \
  67              nh++, nhsel++)
  68
  69 #define change_nexthops(fi) {                                           \
  70         int nhsel; struct fib_nh *nexthop_nh;                           \
  71         for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh);   \
  72              nhsel < (fi)->fib_nhs;                                     \
  73              nexthop_nh++, nhsel++)
  74
  75 #else /* CONFIG_IP_ROUTE_MULTIPATH */
  76
  77 /* Hope, that gcc will optimize it to get rid of dummy loop */
  78
  79 #define for_nexthops(fi) {                                              \
  80         int nhsel; const struct fib_nh *nh = (fi)->fib_nh;              \
  81         for (nhsel = 0; nhsel < 1; nhsel++)
  82
  83 #define change_nexthops(fi) {                                           \
  84         int nhsel;                                                      \
  85         struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);    \
  86         for (nhsel = 0; nhsel < 1; nhsel++)
  87
  88 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
  89
  90 #define endfor_nexthops(fi) }
  91
  92
  93 const struct fib_prop fib_props[RTN_MAX + 1] = {
  94         [RTN_UNSPEC] = {
  95                 .error  = 0,
  96                 .scope  = RT_SCOPE_NOWHERE,
  97         },
  98         [RTN_UNICAST] = {
  99                 .error  = 0,
 100                 .scope  = RT_SCOPE_UNIVERSE,
 101         },
 102         [RTN_LOCAL] = {
 103                 .error  = 0,
 104                 .scope  = RT_SCOPE_HOST,
 105         },
 106         [RTN_BROADCAST] = {
 107                 .error  = 0,
 108                 .scope  = RT_SCOPE_LINK,
 109         },
 110         [RTN_ANYCAST] = {
 111                 .error  = 0,
 112                 .scope  = RT_SCOPE_LINK,
 113         },
 114         [RTN_MULTICAST] = {
 115                 .error  = 0,
 116                 .scope  = RT_SCOPE_UNIVERSE,
 117         },
 118         [RTN_BLACKHOLE] = {
 119                 .error  = -EINVAL,
 120                 .scope  = RT_SCOPE_UNIVERSE,
 121         },
 122         [RTN_UNREACHABLE] = {
 123                 .error  = -EHOSTUNREACH,
 124                 .scope  = RT_SCOPE_UNIVERSE,
 125         },
 126         [RTN_PROHIBIT] = {
 127                 .error  = -EACCES,
 128                 .scope  = RT_SCOPE_UNIVERSE,
 129         },
 130         [RTN_THROW] = {
 131                 .error  = -EAGAIN,
 132                 .scope  = RT_SCOPE_UNIVERSE,
 133         },
 134         [RTN_NAT] = {
 135                 .error  = -EINVAL,
 136                 .scope  = RT_SCOPE_NOWHERE,
 137         },
 138         [RTN_XRESOLVE] = {
 139                 .error  = -EINVAL,
 140                 .scope  = RT_SCOPE_NOWHERE,
 141         },
 142 };
 143
 144 /* Release a nexthop info record */
 145 static void free_fib_info_rcu(struct rcu_head *head)
 146 {
 147         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 148
 149         change_nexthops(fi) {
 150                 if (nexthop_nh->nh_dev)
 151                         dev_put(nexthop_nh->nh_dev);
 152         } endfor_nexthops(fi);
 153
 154         release_net(fi->fib_net);
 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         fib_info_cnt--;
 165         call_rcu(&fi->rcu, free_fib_info_rcu);
 166 }
 167
 168 void fib_release_info(struct fib_info *fi)
 169 {
 170         spin_lock_bh(&fib_info_lock);
 171         if (fi && --fi->fib_treeref == 0) {
 172                 hlist_del(&fi->fib_hash);
 173                 if (fi->fib_prefsrc)
 174                         hlist_del(&fi->fib_lhash);
 175                 change_nexthops(fi) {
 176                         if (!nexthop_nh->nh_dev)
 177                                 continue;
 178                         hlist_del(&nexthop_nh->nh_hash);
 179                 } endfor_nexthops(fi)
 180                 fi->fib_dead = 1;
 181                 fib_info_put(fi);
 182         }
 183         spin_unlock_bh(&fib_info_lock);
 184 }
 185
 186 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
 187 {
 188         const struct fib_nh *onh = ofi->fib_nh;
 189
 190         for_nexthops(fi) {
 191                 if (nh->nh_oif != onh->nh_oif ||
 192                     nh->nh_gw  != onh->nh_gw ||
 193                     nh->nh_scope != onh->nh_scope ||
 194 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 195                     nh->nh_weight != onh->nh_weight ||
 196 #endif
 197 #ifdef CONFIG_IP_ROUTE_CLASSID
 198                     nh->nh_tclassid != onh->nh_tclassid ||
 199 #endif
 200                     ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
 201                         return -1;
 202                 onh++;
 203         } endfor_nexthops(fi);
 204         return 0;
 205 }
 206
 207 static inline unsigned int fib_devindex_hashfn(unsigned int val)
 208 {
 209         unsigned int mask = DEVINDEX_HASHSIZE - 1;
 210
 211         return (val ^
 212                 (val >> DEVINDEX_HASHBITS) ^
 213                 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
 214 }
 215
 216 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
 217 {
 218         unsigned int mask = (fib_info_hash_size - 1);
 219         unsigned int val = fi->fib_nhs;
 220
 221         val ^= (fi->fib_protocol << 8) | fi->fib_scope;
 222         val ^= (__force u32)fi->fib_prefsrc;
 223         val ^= fi->fib_priority;
 224         for_nexthops(fi) {
 225                 val ^= fib_devindex_hashfn(nh->nh_oif);
 226         } endfor_nexthops(fi)
 227
 228         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 229 }
 230
 231 static struct fib_info *fib_find_info(const struct fib_info *nfi)
 232 {
 233         struct hlist_head *head;
 234         struct hlist_node *node;
 235         struct fib_info *fi;
 236         unsigned int hash;
 237
 238         hash = fib_info_hashfn(nfi);
 239         head = &fib_info_hash[hash];
 240
 241         hlist_for_each_entry(fi, node, head, fib_hash) {
 242                 if (!net_eq(fi->fib_net, nfi->fib_net))
 243                         continue;
 244                 if (fi->fib_nhs != nfi->fib_nhs)
 245                         continue;
 246                 if (nfi->fib_protocol == fi->fib_protocol &&
 247                     nfi->fib_scope == fi->fib_scope &&
 248                     nfi->fib_prefsrc == fi->fib_prefsrc &&
 249                     nfi->fib_priority == fi->fib_priority &&
 250                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 251                            sizeof(u32) * RTAX_MAX) == 0 &&
 252                     ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
 253                     (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
 254                         return fi;
 255         }
 256
 257         return NULL;
 258 }
 259
 260 /* Check, that the gateway is already configured.
 261  * Used only by redirect accept routine.
 262  */
 263 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 264 {
 265         struct hlist_head *head;
 266         struct hlist_node *node;
 267         struct fib_nh *nh;
 268         unsigned int hash;
 269
 270         spin_lock(&fib_info_lock);
 271
 272         hash = fib_devindex_hashfn(dev->ifindex);
 273         head = &fib_info_devhash[hash];
 274         hlist_for_each_entry(nh, node, head, nh_hash) {
 275                 if (nh->nh_dev == dev &&
 276                     nh->nh_gw == gw &&
 277                     !(nh->nh_flags & RTNH_F_DEAD)) {
 278                         spin_unlock(&fib_info_lock);
 279                         return 0;
 280                 }
 281         }
 282
 283         spin_unlock(&fib_info_lock);
 284
 285         return -1;
 286 }
 287
 288 static inline size_t fib_nlmsg_size(struct fib_info *fi)
 289 {
 290         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 291                          + nla_total_size(4) /* RTA_TABLE */
 292                          + nla_total_size(4) /* RTA_DST */
 293                          + nla_total_size(4) /* RTA_PRIORITY */
 294                          + nla_total_size(4); /* RTA_PREFSRC */
 295
 296         /* space for nested metrics */
 297         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 298
 299         if (fi->fib_nhs) {
 300                 /* Also handles the special case fib_nhs == 1 */
 301
 302                 /* each nexthop is packed in an attribute */
 303                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 304
 305                 /* may contain flow and gateway attribute */
 306                 nhsize += 2 * nla_total_size(4);
 307
 308                 /* all nexthops are packed in a nested attribute */
 309                 payload += nla_total_size(fi->fib_nhs * nhsize);
 310         }
 311
 312         return payload;
 313 }
 314
 315 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 316                int dst_len, u32 tb_id, struct nl_info *info,
 317                unsigned int nlm_flags)
 318 {
 319         struct sk_buff *skb;
 320         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 321         int err = -ENOBUFS;
 322
 323         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 324         if (skb == NULL)
 325                 goto errout;
 326
 327         err = fib_dump_info(skb, info->pid, seq, event, tb_id,
 328                             fa->fa_type, key, dst_len,
 329                             fa->fa_tos, fa->fa_info, nlm_flags);
 330         if (err < 0) {
 331                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 332                 WARN_ON(err == -EMSGSIZE);
 333                 kfree_skb(skb);
 334                 goto errout;
 335         }
 336         rtnl_notify(skb, info->nl_net, info->pid, RTNLGRP_IPV4_ROUTE,
 337                     info->nlh, GFP_KERNEL);
 338         return;
 339 errout:
 340         if (err < 0)
 341                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 342 }
 343
 344 /* Return the first fib alias matching TOS with
 345  * priority less than or equal to PRIO.
 346  */
 347 struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
 348 {
 349         if (fah) {
 350                 struct fib_alias *fa;
 351                 list_for_each_entry(fa, fah, fa_list) {
 352                         if (fa->fa_tos > tos)
 353                                 continue;
 354                         if (fa->fa_info->fib_priority >= prio ||
 355                             fa->fa_tos < tos)
 356                                 return fa;
 357                 }
 358         }
 359         return NULL;
 360 }
 361
 362 int fib_detect_death(struct fib_info *fi, int order,
 363                      struct fib_info **last_resort, int *last_idx, int dflt)
 364 {
 365         struct neighbour *n;
 366         int state = NUD_NONE;
 367
 368         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 369         if (n) {
 370                 state = n->nud_state;
 371                 neigh_release(n);
 372         }
 373         if (state == NUD_REACHABLE)
 374                 return 0;
 375         if ((state & NUD_VALID) && order != dflt)
 376                 return 0;
 377         if ((state & NUD_VALID) ||
 378             (*last_idx < 0 && order > dflt)) {
 379                 *last_resort = fi;
 380                 *last_idx = order;
 381         }
 382         return 1;
 383 }
 384
 385 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 386
 387 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
 388 {
 389         int nhs = 0;
 390
 391         while (rtnh_ok(rtnh, remaining)) {
 392                 nhs++;
 393                 rtnh = rtnh_next(rtnh, &remaining);
 394         }
 395
 396         /* leftover implies invalid nexthop configuration, discard it */
 397         return remaining > 0 ? 0 : nhs;
 398 }
 399
 400 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 401                        int remaining, struct fib_config *cfg)
 402 {
 403         change_nexthops(fi) {
 404                 int attrlen;
 405
 406                 if (!rtnh_ok(rtnh, remaining))
 407                         return -EINVAL;
 408
 409                 nexthop_nh->nh_flags =
 410                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 411                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
 412                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
 413
 414                 attrlen = rtnh_attrlen(rtnh);
 415                 if (attrlen > 0) {
 416                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 417
 418                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 419                         nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
 420 #ifdef CONFIG_IP_ROUTE_CLASSID
 421                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 422                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 423 #endif
 424                 }
 425
 426                 rtnh = rtnh_next(rtnh, &remaining);
 427         } endfor_nexthops(fi);
 428
 429         return 0;
 430 }
 431
 432 #endif
 433
 434 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
 435 {
 436 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 437         struct rtnexthop *rtnh;
 438         int remaining;
 439 #endif
 440
 441         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 442                 return 1;
 443
 444         if (cfg->fc_oif || cfg->fc_gw) {
 445                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 446                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 447                         return 0;
 448                 return 1;
 449         }
 450
 451 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 452         if (cfg->fc_mp == NULL)
 453                 return 0;
 454
 455         rtnh = cfg->fc_mp;
 456         remaining = cfg->fc_mp_len;
 457
 458         for_nexthops(fi) {
 459                 int attrlen;
 460
 461                 if (!rtnh_ok(rtnh, remaining))
 462                         return -EINVAL;
 463
 464                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 465                         return 1;
 466
 467                 attrlen = rtnh_attrlen(rtnh);
 468                 if (attrlen < 0) {
 469                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 470
 471                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 472                         if (nla && nla_get_be32(nla) != nh->nh_gw)
 473                                 return 1;
 474 #ifdef CONFIG_IP_ROUTE_CLASSID
 475                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 476                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 477                                 return 1;
 478 #endif
 479                 }
 480
 481                 rtnh = rtnh_next(rtnh, &remaining);
 482         } endfor_nexthops(fi);
 483 #endif
 484         return 0;
 485 }
 486
 487
 488 /*
 489  * Picture
 490  * -------
 491  *
 492  * Semantics of nexthop is very messy by historical reasons.
 493  * We have to take into account, that:
 494  * a) gateway can be actually local interface address,
 495  *    so that gatewayed route is direct.
 496  * b) gateway must be on-link address, possibly
 497  *    described not by an ifaddr, but also by a direct route.
 498  * c) If both gateway and interface are specified, they should not
 499  *    contradict.
 500  * d) If we use tunnel routes, gateway could be not on-link.
 501  *
 502  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 503  * results in pretty ugly and hairy code with obscure logic.
 504  *
 505  * I chose to generalized it instead, so that the size
 506  * of code does not increase practically, but it becomes
 507  * much more general.
 508  * Every prefix is assigned a "scope" value: "host" is local address,
 509  * "link" is direct route,
 510  * [ ... "site" ... "interior" ... ]
 511  * and "universe" is true gateway route with global meaning.
 512  *
 513  * Every prefix refers to a set of "nexthop"s (gw, oif),
 514  * where gw must have narrower scope. This recursion stops
 515  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 516  * which means that gw is forced to be on link.
 517  *
 518  * Code is still hairy, but now it is apparently logically
 519  * consistent and very flexible. F.e. as by-product it allows
 520  * to co-exists in peace independent exterior and interior
 521  * routing processes.
 522  *
 523  * Normally it looks as following.
 524  *
 525  * {universe prefix}  -> (gw, oif) [scope link]
 526  *                |
 527  *                |-> {link prefix} -> (gw, oif) [scope local]
 528  *                                      |
 529  *                                      |-> {local prefix} (terminal node)
 530  */
 531 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
 532                         struct fib_nh *nh)
 533 {
 534         int err;
 535         struct net *net;
 536         struct net_device *dev;
 537
 538         net = cfg->fc_nlinfo.nl_net;
 539         if (nh->nh_gw) {
 540                 struct fib_result res;
 541
 542                 if (nh->nh_flags & RTNH_F_ONLINK) {
 543
 544                         if (cfg->fc_scope >= RT_SCOPE_LINK)
 545                                 return -EINVAL;
 546                         if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
 547                                 return -EINVAL;
 548                         dev = __dev_get_by_index(net, nh->nh_oif);
 549                         if (!dev)
 550                                 return -ENODEV;
 551                         if (!(dev->flags & IFF_UP))
 552                                 return -ENETDOWN;
 553                         nh->nh_dev = dev;
 554                         dev_hold(dev);
 555                         nh->nh_scope = RT_SCOPE_LINK;
 556                         return 0;
 557                 }
 558                 rcu_read_lock();
 559                 {
 560                         struct flowi4 fl4 = {
 561                                 .daddr = nh->nh_gw,
 562                                 .flowi4_scope = cfg->fc_scope + 1,
 563                                 .flowi4_oif = nh->nh_oif,
 564                         };
 565
 566                         /* It is not necessary, but requires a bit of thinking */
 567                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
 568                                 fl4.flowi4_scope = RT_SCOPE_LINK;
 569                         err = fib_lookup(net, &fl4, &res);
 570                         if (err) {
 571                                 rcu_read_unlock();
 572                                 return err;
 573                         }
 574                 }
 575                 err = -EINVAL;
 576                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 577                         goto out;
 578                 nh->nh_scope = res.scope;
 579                 nh->nh_oif = FIB_RES_OIF(res);
 580                 nh->nh_dev = dev = FIB_RES_DEV(res);
 581                 if (!dev)
 582                         goto out;
 583                 dev_hold(dev);
 584                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 585         } else {
 586                 struct in_device *in_dev;
 587
 588                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 589                         return -EINVAL;
 590
 591                 rcu_read_lock();
 592                 err = -ENODEV;
 593                 in_dev = inetdev_by_index(net, nh->nh_oif);
 594                 if (in_dev == NULL)
 595                         goto out;
 596                 err = -ENETDOWN;
 597                 if (!(in_dev->dev->flags & IFF_UP))
 598                         goto out;
 599                 nh->nh_dev = in_dev->dev;
 600                 dev_hold(nh->nh_dev);
 601                 nh->nh_scope = RT_SCOPE_HOST;
 602                 err = 0;
 603         }
 604 out:
 605         rcu_read_unlock();
 606         return err;
 607 }
 608
 609 static inline unsigned int fib_laddr_hashfn(__be32 val)
 610 {
 611         unsigned int mask = (fib_info_hash_size - 1);
 612
 613         return ((__force u32)val ^
 614                 ((__force u32)val >> 7) ^
 615                 ((__force u32)val >> 14)) & mask;
 616 }
 617
 618 static struct hlist_head *fib_info_hash_alloc(int bytes)
 619 {
 620         if (bytes <= PAGE_SIZE)
 621                 return kzalloc(bytes, GFP_KERNEL);
 622         else
 623                 return (struct hlist_head *)
 624                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 625                                          get_order(bytes));
 626 }
 627
 628 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 629 {
 630         if (!hash)
 631                 return;
 632
 633         if (bytes <= PAGE_SIZE)
 634                 kfree(hash);
 635         else
 636                 free_pages((unsigned long) hash, get_order(bytes));
 637 }
 638
 639 static void fib_info_hash_move(struct hlist_head *new_info_hash,
 640                                struct hlist_head *new_laddrhash,
 641                                unsigned int new_size)
 642 {
 643         struct hlist_head *old_info_hash, *old_laddrhash;
 644         unsigned int old_size = fib_info_hash_size;
 645         unsigned int i, bytes;
 646
 647         spin_lock_bh(&fib_info_lock);
 648         old_info_hash = fib_info_hash;
 649         old_laddrhash = fib_info_laddrhash;
 650         fib_info_hash_size = new_size;
 651
 652         for (i = 0; i < old_size; i++) {
 653                 struct hlist_head *head = &fib_info_hash[i];
 654                 struct hlist_node *node, *n;
 655                 struct fib_info *fi;
 656
 657                 hlist_for_each_entry_safe(fi, node, n, head, fib_hash) {
 658                         struct hlist_head *dest;
 659                         unsigned int new_hash;
 660
 661                         hlist_del(&fi->fib_hash);
 662
 663                         new_hash = fib_info_hashfn(fi);
 664                         dest = &new_info_hash[new_hash];
 665                         hlist_add_head(&fi->fib_hash, dest);
 666                 }
 667         }
 668         fib_info_hash = new_info_hash;
 669
 670         for (i = 0; i < old_size; i++) {
 671                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 672                 struct hlist_node *node, *n;
 673                 struct fib_info *fi;
 674
 675                 hlist_for_each_entry_safe(fi, node, n, lhead, fib_lhash) {
 676                         struct hlist_head *ldest;
 677                         unsigned int new_hash;
 678
 679                         hlist_del(&fi->fib_lhash);
 680
 681                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 682                         ldest = &new_laddrhash[new_hash];
 683                         hlist_add_head(&fi->fib_lhash, ldest);
 684                 }
 685         }
 686         fib_info_laddrhash = new_laddrhash;
 687
 688         spin_unlock_bh(&fib_info_lock);
 689
 690         bytes = old_size * sizeof(struct hlist_head *);
 691         fib_info_hash_free(old_info_hash, bytes);
 692         fib_info_hash_free(old_laddrhash, bytes);
 693 }
 694
 695 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 696 {
 697         nh->nh_saddr = inet_select_addr(nh->nh_dev,
 698                                         nh->nh_gw,
 699                                         nh->nh_parent->fib_scope);
 700         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 701
 702         return nh->nh_saddr;
 703 }
 704
 705 struct fib_info *fib_create_info(struct fib_config *cfg)
 706 {
 707         int err;
 708         struct fib_info *fi = NULL;
 709         struct fib_info *ofi;
 710         int nhs = 1;
 711         struct net *net = cfg->fc_nlinfo.nl_net;
 712
 713         if (cfg->fc_type > RTN_MAX)
 714                 goto err_inval;
 715
 716         /* Fast check to catch the most weird cases */
 717         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
 718                 goto err_inval;
 719
 720 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 721         if (cfg->fc_mp) {
 722                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
 723                 if (nhs == 0)
 724                         goto err_inval;
 725         }
 726 #endif
 727
 728         err = -ENOBUFS;
 729         if (fib_info_cnt >= fib_info_hash_size) {
 730                 unsigned int new_size = fib_info_hash_size << 1;
 731                 struct hlist_head *new_info_hash;
 732                 struct hlist_head *new_laddrhash;
 733                 unsigned int bytes;
 734
 735                 if (!new_size)
 736                         new_size = 1;
 737                 bytes = new_size * sizeof(struct hlist_head *);
 738                 new_info_hash = fib_info_hash_alloc(bytes);
 739                 new_laddrhash = fib_info_hash_alloc(bytes);
 740                 if (!new_info_hash || !new_laddrhash) {
 741                         fib_info_hash_free(new_info_hash, bytes);
 742                         fib_info_hash_free(new_laddrhash, bytes);
 743                 } else
 744                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
 745
 746                 if (!fib_info_hash_size)
 747                         goto failure;
 748         }
 749
 750         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
 751         if (fi == NULL)
 752                 goto failure;
 753         if (cfg->fc_mx) {
 754                 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
 755                 if (!fi->fib_metrics)
 756                         goto failure;
 757         } else
 758                 fi->fib_metrics = (u32 *) dst_default_metrics;
 759         fib_info_cnt++;
 760
 761         fi->fib_net = hold_net(net);
 762         fi->fib_protocol = cfg->fc_protocol;
 763         fi->fib_scope = cfg->fc_scope;
 764         fi->fib_flags = cfg->fc_flags;
 765         fi->fib_priority = cfg->fc_priority;
 766         fi->fib_prefsrc = cfg->fc_prefsrc;
 767
 768         fi->fib_nhs = nhs;
 769         change_nexthops(fi) {
 770                 nexthop_nh->nh_parent = fi;
 771         } endfor_nexthops(fi)
 772
 773         if (cfg->fc_mx) {
 774                 struct nlattr *nla;
 775                 int remaining;
 776
 777                 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 778                         int type = nla_type(nla);
 779
 780                         if (type) {
 781                                 if (type > RTAX_MAX)
 782                                         goto err_inval;
 783                                 fi->fib_metrics[type - 1] = nla_get_u32(nla);
 784                         }
 785                 }
 786         }
 787
 788         if (cfg->fc_mp) {
 789 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 790                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
 791                 if (err != 0)
 792                         goto failure;
 793                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
 794                         goto err_inval;
 795                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
 796                         goto err_inval;
 797 #ifdef CONFIG_IP_ROUTE_CLASSID
 798                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
 799                         goto err_inval;
 800 #endif
 801 #else
 802                 goto err_inval;
 803 #endif
 804         } else {
 805                 struct fib_nh *nh = fi->fib_nh;
 806
 807                 nh->nh_oif = cfg->fc_oif;
 808                 nh->nh_gw = cfg->fc_gw;
 809                 nh->nh_flags = cfg->fc_flags;
 810 #ifdef CONFIG_IP_ROUTE_CLASSID
 811                 nh->nh_tclassid = cfg->fc_flow;
 812 #endif
 813 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 814                 nh->nh_weight = 1;
 815 #endif
 816         }
 817
 818         if (fib_props[cfg->fc_type].error) {
 819                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
 820                         goto err_inval;
 821                 goto link_it;
 822         } else {
 823                 switch (cfg->fc_type) {
 824                 case RTN_UNICAST:
 825                 case RTN_LOCAL:
 826                 case RTN_BROADCAST:
 827                 case RTN_ANYCAST:
 828                 case RTN_MULTICAST:
 829                         break;
 830                 default:
 831                         goto err_inval;
 832                 }
 833         }
 834
 835         if (cfg->fc_scope > RT_SCOPE_HOST)
 836                 goto err_inval;
 837
 838         if (cfg->fc_scope == RT_SCOPE_HOST) {
 839                 struct fib_nh *nh = fi->fib_nh;
 840
 841                 /* Local address is added. */
 842                 if (nhs != 1 || nh->nh_gw)
 843                         goto err_inval;
 844                 nh->nh_scope = RT_SCOPE_NOWHERE;
 845                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
 846                 err = -ENODEV;
 847                 if (nh->nh_dev == NULL)
 848                         goto failure;
 849         } else {
 850                 change_nexthops(fi) {
 851                         err = fib_check_nh(cfg, fi, nexthop_nh);
 852                         if (err != 0)
 853                                 goto failure;
 854                 } endfor_nexthops(fi)
 855         }
 856
 857         if (fi->fib_prefsrc) {
 858                 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 859                     fi->fib_prefsrc != cfg->fc_dst)
 860                         if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
 861                                 goto err_inval;
 862         }
 863
 864         change_nexthops(fi) {
 865                 fib_info_update_nh_saddr(net, nexthop_nh);
 866         } endfor_nexthops(fi)
 867
 868 link_it:
 869         ofi = fib_find_info(fi);
 870         if (ofi) {
 871                 fi->fib_dead = 1;
 872                 free_fib_info(fi);
 873                 ofi->fib_treeref++;
 874                 return ofi;
 875         }
 876
 877         fi->fib_treeref++;
 878         atomic_inc(&fi->fib_clntref);
 879         spin_lock_bh(&fib_info_lock);
 880         hlist_add_head(&fi->fib_hash,
 881                        &fib_info_hash[fib_info_hashfn(fi)]);
 882         if (fi->fib_prefsrc) {
 883                 struct hlist_head *head;
 884
 885                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
 886                 hlist_add_head(&fi->fib_lhash, head);
 887         }
 888         change_nexthops(fi) {
 889                 struct hlist_head *head;
 890                 unsigned int hash;
 891
 892                 if (!nexthop_nh->nh_dev)
 893                         continue;
 894                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
 895                 head = &fib_info_devhash[hash];
 896                 hlist_add_head(&nexthop_nh->nh_hash, head);
 897         } endfor_nexthops(fi)
 898         spin_unlock_bh(&fib_info_lock);
 899         return fi;
 900
 901 err_inval:
 902         err = -EINVAL;
 903
 904 failure:
 905         if (fi) {
 906                 fi->fib_dead = 1;
 907                 free_fib_info(fi);
 908         }
 909
 910         return ERR_PTR(err);
 911 }
 912
 913 int fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
 914                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
 915                   struct fib_info *fi, unsigned int flags)
 916 {
 917         struct nlmsghdr *nlh;
 918         struct rtmsg *rtm;
 919
 920         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*rtm), flags);
 921         if (nlh == NULL)
 922                 return -EMSGSIZE;
 923
 924         rtm = nlmsg_data(nlh);
 925         rtm->rtm_family = AF_INET;
 926         rtm->rtm_dst_len = dst_len;
 927         rtm->rtm_src_len = 0;
 928         rtm->rtm_tos = tos;
 929         if (tb_id < 256)
 930                 rtm->rtm_table = tb_id;
 931         else
 932                 rtm->rtm_table = RT_TABLE_COMPAT;
 933         NLA_PUT_U32(skb, RTA_TABLE, tb_id);
 934         rtm->rtm_type = type;
 935         rtm->rtm_flags = fi->fib_flags;
 936         rtm->rtm_scope = fi->fib_scope;
 937         rtm->rtm_protocol = fi->fib_protocol;
 938
 939         if (rtm->rtm_dst_len)
 940                 NLA_PUT_BE32(skb, RTA_DST, dst);
 941
 942         if (fi->fib_priority)
 943                 NLA_PUT_U32(skb, RTA_PRIORITY, fi->fib_priority);
 944
 945         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
 946                 goto nla_put_failure;
 947
 948         if (fi->fib_prefsrc)
 949                 NLA_PUT_BE32(skb, RTA_PREFSRC, fi->fib_prefsrc);
 950
 951         if (fi->fib_nhs == 1) {
 952                 if (fi->fib_nh->nh_gw)
 953                         NLA_PUT_BE32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw);
 954
 955                 if (fi->fib_nh->nh_oif)
 956                         NLA_PUT_U32(skb, RTA_OIF, fi->fib_nh->nh_oif);
 957 #ifdef CONFIG_IP_ROUTE_CLASSID
 958                 if (fi->fib_nh[0].nh_tclassid)
 959                         NLA_PUT_U32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid);
 960 #endif
 961         }
 962 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 963         if (fi->fib_nhs > 1) {
 964                 struct rtnexthop *rtnh;
 965                 struct nlattr *mp;
 966
 967                 mp = nla_nest_start(skb, RTA_MULTIPATH);
 968                 if (mp == NULL)
 969                         goto nla_put_failure;
 970
 971                 for_nexthops(fi) {
 972                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
 973                         if (rtnh == NULL)
 974                                 goto nla_put_failure;
 975
 976                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
 977                         rtnh->rtnh_hops = nh->nh_weight - 1;
 978                         rtnh->rtnh_ifindex = nh->nh_oif;
 979
 980                         if (nh->nh_gw)
 981                                 NLA_PUT_BE32(skb, RTA_GATEWAY, nh->nh_gw);
 982 #ifdef CONFIG_IP_ROUTE_CLASSID
 983                         if (nh->nh_tclassid)
 984                                 NLA_PUT_U32(skb, RTA_FLOW, nh->nh_tclassid);
 985 #endif
 986                         /* length of rtnetlink header + attributes */
 987                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
 988                 } endfor_nexthops(fi);
 989
 990                 nla_nest_end(skb, mp);
 991         }
 992 #endif
 993         return nlmsg_end(skb, nlh);
 994
 995 nla_put_failure:
 996         nlmsg_cancel(skb, nlh);
 997         return -EMSGSIZE;
 998 }
 999
1000 /*
1001  * Update FIB if:
1002  * - local address disappeared -> we must delete all the entries
1003  *   referring to it.
1004  * - device went down -> we must shutdown all nexthops going via it.
1005  */
1006 int fib_sync_down_addr(struct net *net, __be32 local)
1007 {
1008         int ret = 0;
1009         unsigned int hash = fib_laddr_hashfn(local);
1010         struct hlist_head *head = &fib_info_laddrhash[hash];
1011         struct hlist_node *node;
1012         struct fib_info *fi;
1013
1014         if (fib_info_laddrhash == NULL || local == 0)
1015                 return 0;
1016
1017         hlist_for_each_entry(fi, node, head, fib_lhash) {
1018                 if (!net_eq(fi->fib_net, net))
1019                         continue;
1020                 if (fi->fib_prefsrc == local) {
1021                         fi->fib_flags |= RTNH_F_DEAD;
1022                         ret++;
1023                 }
1024         }
1025         return ret;
1026 }
1027
1028 int fib_sync_down_dev(struct net_device *dev, int force)
1029 {
1030         int ret = 0;
1031         int scope = RT_SCOPE_NOWHERE;
1032         struct fib_info *prev_fi = NULL;
1033         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1034         struct hlist_head *head = &fib_info_devhash[hash];
1035         struct hlist_node *node;
1036         struct fib_nh *nh;
1037
1038         if (force)
1039                 scope = -1;
1040
1041         hlist_for_each_entry(nh, node, head, nh_hash) {
1042                 struct fib_info *fi = nh->nh_parent;
1043                 int dead;
1044
1045                 BUG_ON(!fi->fib_nhs);
1046                 if (nh->nh_dev != dev || fi == prev_fi)
1047                         continue;
1048                 prev_fi = fi;
1049                 dead = 0;
1050                 change_nexthops(fi) {
1051                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1052                                 dead++;
1053                         else if (nexthop_nh->nh_dev == dev &&
1054                                  nexthop_nh->nh_scope != scope) {
1055                                 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1056 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1057                                 spin_lock_bh(&fib_multipath_lock);
1058                                 fi->fib_power -= nexthop_nh->nh_power;
1059                                 nexthop_nh->nh_power = 0;
1060                                 spin_unlock_bh(&fib_multipath_lock);
1061 #endif
1062                                 dead++;
1063                         }
1064 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1065                         if (force > 1 && nexthop_nh->nh_dev == dev) {
1066                                 dead = fi->fib_nhs;
1067                                 break;
1068                         }
1069 #endif
1070                 } endfor_nexthops(fi)
1071                 if (dead == fi->fib_nhs) {
1072                         fi->fib_flags |= RTNH_F_DEAD;
1073                         ret++;
1074                 }
1075         }
1076
1077         return ret;
1078 }
1079
1080 /* Must be invoked inside of an RCU protected region.  */
1081 void fib_select_default(struct fib_result *res)
1082 {
1083         struct fib_info *fi = NULL, *last_resort = NULL;
1084         struct list_head *fa_head = res->fa_head;
1085         struct fib_table *tb = res->table;
1086         int order = -1, last_idx = -1;
1087         struct fib_alias *fa;
1088
1089         list_for_each_entry_rcu(fa, fa_head, fa_list) {
1090                 struct fib_info *next_fi = fa->fa_info;
1091
1092                 if (next_fi->fib_scope != res->scope ||
1093                     fa->fa_type != RTN_UNICAST)
1094                         continue;
1095
1096                 if (next_fi->fib_priority > res->fi->fib_priority)
1097                         break;
1098                 if (!next_fi->fib_nh[0].nh_gw ||
1099                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1100                         continue;
1101
1102                 fib_alias_accessed(fa);
1103
1104                 if (fi == NULL) {
1105                         if (next_fi != res->fi)
1106                                 break;
1107                 } else if (!fib_detect_death(fi, order, &last_resort,
1108                                              &last_idx, tb->tb_default)) {
1109                         fib_result_assign(res, fi);
1110                         tb->tb_default = order;
1111                         goto out;
1112                 }
1113                 fi = next_fi;
1114                 order++;
1115         }
1116
1117         if (order <= 0 || fi == NULL) {
1118                 tb->tb_default = -1;
1119                 goto out;
1120         }
1121
1122         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1123                                 tb->tb_default)) {
1124                 fib_result_assign(res, fi);
1125                 tb->tb_default = order;
1126                 goto out;
1127         }
1128
1129         if (last_idx >= 0)
1130                 fib_result_assign(res, last_resort);
1131         tb->tb_default = last_idx;
1132 out:
1133         return;
1134 }
1135
1136 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1137
1138 /*
1139  * Dead device goes up. We wake up dead nexthops.
1140  * It takes sense only on multipath routes.
1141  */
1142 int fib_sync_up(struct net_device *dev)
1143 {
1144         struct fib_info *prev_fi;
1145         unsigned int hash;
1146         struct hlist_head *head;
1147         struct hlist_node *node;
1148         struct fib_nh *nh;
1149         int ret;
1150
1151         if (!(dev->flags & IFF_UP))
1152                 return 0;
1153
1154         prev_fi = NULL;
1155         hash = fib_devindex_hashfn(dev->ifindex);
1156         head = &fib_info_devhash[hash];
1157         ret = 0;
1158
1159         hlist_for_each_entry(nh, node, head, nh_hash) {
1160                 struct fib_info *fi = nh->nh_parent;
1161                 int alive;
1162
1163                 BUG_ON(!fi->fib_nhs);
1164                 if (nh->nh_dev != dev || fi == prev_fi)
1165                         continue;
1166
1167                 prev_fi = fi;
1168                 alive = 0;
1169                 change_nexthops(fi) {
1170                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1171                                 alive++;
1172                                 continue;
1173                         }
1174                         if (nexthop_nh->nh_dev == NULL ||
1175                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1176                                 continue;
1177                         if (nexthop_nh->nh_dev != dev ||
1178                             !__in_dev_get_rtnl(dev))
1179                                 continue;
1180                         alive++;
1181                         spin_lock_bh(&fib_multipath_lock);
1182                         nexthop_nh->nh_power = 0;
1183                         nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1184                         spin_unlock_bh(&fib_multipath_lock);
1185                 } endfor_nexthops(fi)
1186
1187                 if (alive > 0) {
1188                         fi->fib_flags &= ~RTNH_F_DEAD;
1189                         ret++;
1190                 }
1191         }
1192
1193         return ret;
1194 }
1195
1196 /*
1197  * The algorithm is suboptimal, but it provides really
1198  * fair weighted route distribution.
1199  */
1200 void fib_select_multipath(struct fib_result *res)
1201 {
1202         struct fib_info *fi = res->fi;
1203         int w;
1204
1205         spin_lock_bh(&fib_multipath_lock);
1206         if (fi->fib_power <= 0) {
1207                 int power = 0;
1208                 change_nexthops(fi) {
1209                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1210                                 power += nexthop_nh->nh_weight;
1211                                 nexthop_nh->nh_power = nexthop_nh->nh_weight;
1212                         }
1213                 } endfor_nexthops(fi);
1214                 fi->fib_power = power;
1215                 if (power <= 0) {
1216                         spin_unlock_bh(&fib_multipath_lock);
1217                         /* Race condition: route has just become dead. */
1218                         res->nh_sel = 0;
1219                         return;
1220                 }
1221         }
1222
1223
1224         /* w should be random number [0..fi->fib_power-1],
1225          * it is pretty bad approximation.
1226          */
1227
1228         w = jiffies % fi->fib_power;
1229
1230         change_nexthops(fi) {
1231                 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1232                     nexthop_nh->nh_power) {
1233                         w -= nexthop_nh->nh_power;
1234                         if (w <= 0) {
1235                                 nexthop_nh->nh_power--;
1236                                 fi->fib_power--;
1237                                 res->nh_sel = nhsel;
1238                                 spin_unlock_bh(&fib_multipath_lock);
1239                                 return;
1240                         }
1241                 }
1242         } endfor_nexthops(fi);
1243
1244         /* Race condition: route has just become dead. */
1245         res->nh_sel = 0;
1246         spin_unlock_bh(&fib_multipath_lock);
1247 }
1248 #endif