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