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 <linux/bitops.h>
  18 #include <linux/types.h>
  19 #include <linux/kernel.h>
  20 #include <linux/jiffies.h>
  21 #include <linux/mm.h>
  22 #include <linux/string.h>
  23 #include <linux/socket.h>
  24 #include <linux/sockios.h>
  25 #include <linux/errno.h>
  26 #include <linux/in.h>
  27 #include <linux/inet.h>
  28 #include <linux/inetdevice.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/if_arp.h>
  31 #include <linux/proc_fs.h>
  32 #include <linux/skbuff.h>
  33 #include <linux/init.h>
  34 #include <linux/slab.h>
  35
  36 #include <net/arp.h>
  37 #include <net/ip.h>
  38 #include <net/protocol.h>
  39 #include <net/route.h>
  40 #include <net/tcp.h>
  41 #include <net/sock.h>
  42 #include <net/ip_fib.h>
  43 #include <net/netlink.h>
  44 #include <net/nexthop.h>
  45
  46 #include "fib_lookup.h"
  47
  48 static DEFINE_SPINLOCK(fib_info_lock);
  49 static struct hlist_head *fib_info_hash;
  50 static struct hlist_head *fib_info_laddrhash;
  51 static unsigned int fib_info_hash_size;
  52 static unsigned int fib_info_cnt;
  53
  54 #define DEVINDEX_HASHBITS 8
  55 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
  56 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
  57
  58 #ifdef CONFIG_IP_ROUTE_MULTIPATH
  59
  60 static DEFINE_SPINLOCK(fib_multipath_lock);
  61
  62 #define for_nexthops(fi) {                                              \
  63         int nhsel; const struct fib_nh *nh;                             \
  64         for (nhsel = 0, nh = (fi)->fib_nh;                              \
  65              nhsel < (fi)->fib_nhs;                                     \
  66              nh++, nhsel++)
  67
  68 #define change_nexthops(fi) {                                           \
  69         int nhsel; struct fib_nh *nexthop_nh;                           \
  70         for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh);   \
  71              nhsel < (fi)->fib_nhs;                                     \
  72              nexthop_nh++, nhsel++)
  73
  74 #else /* CONFIG_IP_ROUTE_MULTIPATH */
  75
  76 /* Hope, that gcc will optimize it to get rid of dummy loop */
  77
  78 #define for_nexthops(fi) {                                              \
  79         int nhsel; const struct fib_nh *nh = (fi)->fib_nh;              \
  80         for (nhsel = 0; nhsel < 1; nhsel++)
  81
  82 #define change_nexthops(fi) {                                           \
  83         int nhsel;                                                      \
  84         struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh);    \
  85         for (nhsel = 0; nhsel < 1; nhsel++)
  86
  87 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
  88
  89 #define endfor_nexthops(fi) }
  90
  91
  92 const struct fib_prop fib_props[RTN_MAX + 1] = {
  93         [RTN_UNSPEC] = {
  94                 .error  = 0,
  95                 .scope  = RT_SCOPE_NOWHERE,
  96         },
  97         [RTN_UNICAST] = {
  98                 .error  = 0,
  99                 .scope  = RT_SCOPE_UNIVERSE,
 100         },
 101         [RTN_LOCAL] = {
 102                 .error  = 0,
 103                 .scope  = RT_SCOPE_HOST,
 104         },
 105         [RTN_BROADCAST] = {
 106                 .error  = 0,
 107                 .scope  = RT_SCOPE_LINK,
 108         },
 109         [RTN_ANYCAST] = {
 110                 .error  = 0,
 111                 .scope  = RT_SCOPE_LINK,
 112         },
 113         [RTN_MULTICAST] = {
 114                 .error  = 0,
 115                 .scope  = RT_SCOPE_UNIVERSE,
 116         },
 117         [RTN_BLACKHOLE] = {
 118                 .error  = -EINVAL,
 119                 .scope  = RT_SCOPE_UNIVERSE,
 120         },
 121         [RTN_UNREACHABLE] = {
 122                 .error  = -EHOSTUNREACH,
 123                 .scope  = RT_SCOPE_UNIVERSE,
 124         },
 125         [RTN_PROHIBIT] = {
 126                 .error  = -EACCES,
 127                 .scope  = RT_SCOPE_UNIVERSE,
 128         },
 129         [RTN_THROW] = {
 130                 .error  = -EAGAIN,
 131                 .scope  = RT_SCOPE_UNIVERSE,
 132         },
 133         [RTN_NAT] = {
 134                 .error  = -EINVAL,
 135                 .scope  = RT_SCOPE_NOWHERE,
 136         },
 137         [RTN_XRESOLVE] = {
 138                 .error  = -EINVAL,
 139                 .scope  = RT_SCOPE_NOWHERE,
 140         },
 141 };
 142
 143 /* Release a nexthop info record */
 144 static void free_fib_info_rcu(struct rcu_head *head)
 145 {
 146         struct fib_info *fi = container_of(head, struct fib_info, rcu);
 147
 148         change_nexthops(fi) {
 149                 if (nexthop_nh->nh_dev)
 150                         dev_put(nexthop_nh->nh_dev);
 151         } endfor_nexthops(fi);
 152
 153         release_net(fi->fib_net);
 154         if (fi->fib_metrics != (u32 *) dst_default_metrics)
 155                 kfree(fi->fib_metrics);
 156         kfree(fi);
 157 }
 158
 159 void free_fib_info(struct fib_info *fi)
 160 {
 161         if (fi->fib_dead == 0) {
 162                 pr_warn("Freeing alive fib_info %p\n", fi);
 163                 return;
 164         }
 165         fib_info_cnt--;
 166         call_rcu(&fi->rcu, free_fib_info_rcu);
 167 }
 168
 169 void fib_release_info(struct fib_info *fi)
 170 {
 171         spin_lock_bh(&fib_info_lock);
 172         if (fi && --fi->fib_treeref == 0) {
 173                 hlist_del(&fi->fib_hash);
 174                 if (fi->fib_prefsrc)
 175                         hlist_del(&fi->fib_lhash);
 176                 change_nexthops(fi) {
 177                         if (!nexthop_nh->nh_dev)
 178                                 continue;
 179                         hlist_del(&nexthop_nh->nh_hash);
 180                 } endfor_nexthops(fi)
 181                 fi->fib_dead = 1;
 182                 fib_info_put(fi);
 183         }
 184         spin_unlock_bh(&fib_info_lock);
 185 }
 186
 187 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
 188 {
 189         const struct fib_nh *onh = ofi->fib_nh;
 190
 191         for_nexthops(fi) {
 192                 if (nh->nh_oif != onh->nh_oif ||
 193                     nh->nh_gw  != onh->nh_gw ||
 194                     nh->nh_scope != onh->nh_scope ||
 195 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 196                     nh->nh_weight != onh->nh_weight ||
 197 #endif
 198 #ifdef CONFIG_IP_ROUTE_CLASSID
 199                     nh->nh_tclassid != onh->nh_tclassid ||
 200 #endif
 201                     ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
 202                         return -1;
 203                 onh++;
 204         } endfor_nexthops(fi);
 205         return 0;
 206 }
 207
 208 static inline unsigned int fib_devindex_hashfn(unsigned int val)
 209 {
 210         unsigned int mask = DEVINDEX_HASHSIZE - 1;
 211
 212         return (val ^
 213                 (val >> DEVINDEX_HASHBITS) ^
 214                 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
 215 }
 216
 217 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
 218 {
 219         unsigned int mask = (fib_info_hash_size - 1);
 220         unsigned int val = fi->fib_nhs;
 221
 222         val ^= (fi->fib_protocol << 8) | fi->fib_scope;
 223         val ^= (__force u32)fi->fib_prefsrc;
 224         val ^= fi->fib_priority;
 225         for_nexthops(fi) {
 226                 val ^= fib_devindex_hashfn(nh->nh_oif);
 227         } endfor_nexthops(fi)
 228
 229         return (val ^ (val >> 7) ^ (val >> 12)) & mask;
 230 }
 231
 232 static struct fib_info *fib_find_info(const struct fib_info *nfi)
 233 {
 234         struct hlist_head *head;
 235         struct hlist_node *node;
 236         struct fib_info *fi;
 237         unsigned int hash;
 238
 239         hash = fib_info_hashfn(nfi);
 240         head = &fib_info_hash[hash];
 241
 242         hlist_for_each_entry(fi, node, head, fib_hash) {
 243                 if (!net_eq(fi->fib_net, nfi->fib_net))
 244                         continue;
 245                 if (fi->fib_nhs != nfi->fib_nhs)
 246                         continue;
 247                 if (nfi->fib_protocol == fi->fib_protocol &&
 248                     nfi->fib_scope == fi->fib_scope &&
 249                     nfi->fib_prefsrc == fi->fib_prefsrc &&
 250                     nfi->fib_priority == fi->fib_priority &&
 251                     memcmp(nfi->fib_metrics, fi->fib_metrics,
 252                            sizeof(u32) * RTAX_MAX) == 0 &&
 253                     ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
 254                     (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
 255                         return fi;
 256         }
 257
 258         return NULL;
 259 }
 260
 261 /* Check, that the gateway is already configured.
 262  * Used only by redirect accept routine.
 263  */
 264 int ip_fib_check_default(__be32 gw, struct net_device *dev)
 265 {
 266         struct hlist_head *head;
 267         struct hlist_node *node;
 268         struct fib_nh *nh;
 269         unsigned int hash;
 270
 271         spin_lock(&fib_info_lock);
 272
 273         hash = fib_devindex_hashfn(dev->ifindex);
 274         head = &fib_info_devhash[hash];
 275         hlist_for_each_entry(nh, node, head, nh_hash) {
 276                 if (nh->nh_dev == dev &&
 277                     nh->nh_gw == gw &&
 278                     !(nh->nh_flags & RTNH_F_DEAD)) {
 279                         spin_unlock(&fib_info_lock);
 280                         return 0;
 281                 }
 282         }
 283
 284         spin_unlock(&fib_info_lock);
 285
 286         return -1;
 287 }
 288
 289 static inline size_t fib_nlmsg_size(struct fib_info *fi)
 290 {
 291         size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
 292                          + nla_total_size(4) /* RTA_TABLE */
 293                          + nla_total_size(4) /* RTA_DST */
 294                          + nla_total_size(4) /* RTA_PRIORITY */
 295                          + nla_total_size(4); /* RTA_PREFSRC */
 296
 297         /* space for nested metrics */
 298         payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
 299
 300         if (fi->fib_nhs) {
 301                 /* Also handles the special case fib_nhs == 1 */
 302
 303                 /* each nexthop is packed in an attribute */
 304                 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
 305
 306                 /* may contain flow and gateway attribute */
 307                 nhsize += 2 * nla_total_size(4);
 308
 309                 /* all nexthops are packed in a nested attribute */
 310                 payload += nla_total_size(fi->fib_nhs * nhsize);
 311         }
 312
 313         return payload;
 314 }
 315
 316 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
 317                int dst_len, u32 tb_id, struct nl_info *info,
 318                unsigned int nlm_flags)
 319 {
 320         struct sk_buff *skb;
 321         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 322         int err = -ENOBUFS;
 323
 324         skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
 325         if (skb == NULL)
 326                 goto errout;
 327
 328         err = fib_dump_info(skb, info->pid, seq, event, tb_id,
 329                             fa->fa_type, key, dst_len,
 330                             fa->fa_tos, fa->fa_info, nlm_flags);
 331         if (err < 0) {
 332                 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
 333                 WARN_ON(err == -EMSGSIZE);
 334                 kfree_skb(skb);
 335                 goto errout;
 336         }
 337         rtnl_notify(skb, info->nl_net, info->pid, RTNLGRP_IPV4_ROUTE,
 338                     info->nlh, GFP_KERNEL);
 339         return;
 340 errout:
 341         if (err < 0)
 342                 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
 343 }
 344
 345 /* Return the first fib alias matching TOS with
 346  * priority less than or equal to PRIO.
 347  */
 348 struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
 349 {
 350         if (fah) {
 351                 struct fib_alias *fa;
 352                 list_for_each_entry(fa, fah, fa_list) {
 353                         if (fa->fa_tos > tos)
 354                                 continue;
 355                         if (fa->fa_info->fib_priority >= prio ||
 356                             fa->fa_tos < tos)
 357                                 return fa;
 358                 }
 359         }
 360         return NULL;
 361 }
 362
 363 int fib_detect_death(struct fib_info *fi, int order,
 364                      struct fib_info **last_resort, int *last_idx, int dflt)
 365 {
 366         struct neighbour *n;
 367         int state = NUD_NONE;
 368
 369         n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
 370         if (n) {
 371                 state = n->nud_state;
 372                 neigh_release(n);
 373         }
 374         if (state == NUD_REACHABLE)
 375                 return 0;
 376         if ((state & NUD_VALID) && order != dflt)
 377                 return 0;
 378         if ((state & NUD_VALID) ||
 379             (*last_idx < 0 && order > dflt)) {
 380                 *last_resort = fi;
 381                 *last_idx = order;
 382         }
 383         return 1;
 384 }
 385
 386 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 387
 388 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
 389 {
 390         int nhs = 0;
 391
 392         while (rtnh_ok(rtnh, remaining)) {
 393                 nhs++;
 394                 rtnh = rtnh_next(rtnh, &remaining);
 395         }
 396
 397         /* leftover implies invalid nexthop configuration, discard it */
 398         return remaining > 0 ? 0 : nhs;
 399 }
 400
 401 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
 402                        int remaining, struct fib_config *cfg)
 403 {
 404         change_nexthops(fi) {
 405                 int attrlen;
 406
 407                 if (!rtnh_ok(rtnh, remaining))
 408                         return -EINVAL;
 409
 410                 nexthop_nh->nh_flags =
 411                         (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
 412                 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
 413                 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
 414
 415                 attrlen = rtnh_attrlen(rtnh);
 416                 if (attrlen > 0) {
 417                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 418
 419                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 420                         nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
 421 #ifdef CONFIG_IP_ROUTE_CLASSID
 422                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 423                         nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
 424 #endif
 425                 }
 426
 427                 rtnh = rtnh_next(rtnh, &remaining);
 428         } endfor_nexthops(fi);
 429
 430         return 0;
 431 }
 432
 433 #endif
 434
 435 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
 436 {
 437 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 438         struct rtnexthop *rtnh;
 439         int remaining;
 440 #endif
 441
 442         if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
 443                 return 1;
 444
 445         if (cfg->fc_oif || cfg->fc_gw) {
 446                 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
 447                     (!cfg->fc_gw  || cfg->fc_gw == fi->fib_nh->nh_gw))
 448                         return 0;
 449                 return 1;
 450         }
 451
 452 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 453         if (cfg->fc_mp == NULL)
 454                 return 0;
 455
 456         rtnh = cfg->fc_mp;
 457         remaining = cfg->fc_mp_len;
 458
 459         for_nexthops(fi) {
 460                 int attrlen;
 461
 462                 if (!rtnh_ok(rtnh, remaining))
 463                         return -EINVAL;
 464
 465                 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
 466                         return 1;
 467
 468                 attrlen = rtnh_attrlen(rtnh);
 469                 if (attrlen < 0) {
 470                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
 471
 472                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
 473                         if (nla && nla_get_be32(nla) != nh->nh_gw)
 474                                 return 1;
 475 #ifdef CONFIG_IP_ROUTE_CLASSID
 476                         nla = nla_find(attrs, attrlen, RTA_FLOW);
 477                         if (nla && nla_get_u32(nla) != nh->nh_tclassid)
 478                                 return 1;
 479 #endif
 480                 }
 481
 482                 rtnh = rtnh_next(rtnh, &remaining);
 483         } endfor_nexthops(fi);
 484 #endif
 485         return 0;
 486 }
 487
 488
 489 /*
 490  * Picture
 491  * -------
 492  *
 493  * Semantics of nexthop is very messy by historical reasons.
 494  * We have to take into account, that:
 495  * a) gateway can be actually local interface address,
 496  *    so that gatewayed route is direct.
 497  * b) gateway must be on-link address, possibly
 498  *    described not by an ifaddr, but also by a direct route.
 499  * c) If both gateway and interface are specified, they should not
 500  *    contradict.
 501  * d) If we use tunnel routes, gateway could be not on-link.
 502  *
 503  * Attempt to reconcile all of these (alas, self-contradictory) conditions
 504  * results in pretty ugly and hairy code with obscure logic.
 505  *
 506  * I chose to generalized it instead, so that the size
 507  * of code does not increase practically, but it becomes
 508  * much more general.
 509  * Every prefix is assigned a "scope" value: "host" is local address,
 510  * "link" is direct route,
 511  * [ ... "site" ... "interior" ... ]
 512  * and "universe" is true gateway route with global meaning.
 513  *
 514  * Every prefix refers to a set of "nexthop"s (gw, oif),
 515  * where gw must have narrower scope. This recursion stops
 516  * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
 517  * which means that gw is forced to be on link.
 518  *
 519  * Code is still hairy, but now it is apparently logically
 520  * consistent and very flexible. F.e. as by-product it allows
 521  * to co-exists in peace independent exterior and interior
 522  * routing processes.
 523  *
 524  * Normally it looks as following.
 525  *
 526  * {universe prefix}  -> (gw, oif) [scope link]
 527  *                |
 528  *                |-> {link prefix} -> (gw, oif) [scope local]
 529  *                                      |
 530  *                                      |-> {local prefix} (terminal node)
 531  */
 532 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
 533                         struct fib_nh *nh)
 534 {
 535         int err;
 536         struct net *net;
 537         struct net_device *dev;
 538
 539         net = cfg->fc_nlinfo.nl_net;
 540         if (nh->nh_gw) {
 541                 struct fib_result res;
 542
 543                 if (nh->nh_flags & RTNH_F_ONLINK) {
 544
 545                         if (cfg->fc_scope >= RT_SCOPE_LINK)
 546                                 return -EINVAL;
 547                         if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
 548                                 return -EINVAL;
 549                         dev = __dev_get_by_index(net, nh->nh_oif);
 550                         if (!dev)
 551                                 return -ENODEV;
 552                         if (!(dev->flags & IFF_UP))
 553                                 return -ENETDOWN;
 554                         nh->nh_dev = dev;
 555                         dev_hold(dev);
 556                         nh->nh_scope = RT_SCOPE_LINK;
 557                         return 0;
 558                 }
 559                 rcu_read_lock();
 560                 {
 561                         struct flowi4 fl4 = {
 562                                 .daddr = nh->nh_gw,
 563                                 .flowi4_scope = cfg->fc_scope + 1,
 564                                 .flowi4_oif = nh->nh_oif,
 565                         };
 566
 567                         /* It is not necessary, but requires a bit of thinking */
 568                         if (fl4.flowi4_scope < RT_SCOPE_LINK)
 569                                 fl4.flowi4_scope = RT_SCOPE_LINK;
 570                         err = fib_lookup(net, &fl4, &res);
 571                         if (err) {
 572                                 rcu_read_unlock();
 573                                 return err;
 574                         }
 575                 }
 576                 err = -EINVAL;
 577                 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
 578                         goto out;
 579                 nh->nh_scope = res.scope;
 580                 nh->nh_oif = FIB_RES_OIF(res);
 581                 nh->nh_dev = dev = FIB_RES_DEV(res);
 582                 if (!dev)
 583                         goto out;
 584                 dev_hold(dev);
 585                 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
 586         } else {
 587                 struct in_device *in_dev;
 588
 589                 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
 590                         return -EINVAL;
 591
 592                 rcu_read_lock();
 593                 err = -ENODEV;
 594                 in_dev = inetdev_by_index(net, nh->nh_oif);
 595                 if (in_dev == NULL)
 596                         goto out;
 597                 err = -ENETDOWN;
 598                 if (!(in_dev->dev->flags & IFF_UP))
 599                         goto out;
 600                 nh->nh_dev = in_dev->dev;
 601                 dev_hold(nh->nh_dev);
 602                 nh->nh_scope = RT_SCOPE_HOST;
 603                 err = 0;
 604         }
 605 out:
 606         rcu_read_unlock();
 607         return err;
 608 }
 609
 610 static inline unsigned int fib_laddr_hashfn(__be32 val)
 611 {
 612         unsigned int mask = (fib_info_hash_size - 1);
 613
 614         return ((__force u32)val ^
 615                 ((__force u32)val >> 7) ^
 616                 ((__force u32)val >> 14)) & mask;
 617 }
 618
 619 static struct hlist_head *fib_info_hash_alloc(int bytes)
 620 {
 621         if (bytes <= PAGE_SIZE)
 622                 return kzalloc(bytes, GFP_KERNEL);
 623         else
 624                 return (struct hlist_head *)
 625                         __get_free_pages(GFP_KERNEL | __GFP_ZERO,
 626                                          get_order(bytes));
 627 }
 628
 629 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
 630 {
 631         if (!hash)
 632                 return;
 633
 634         if (bytes <= PAGE_SIZE)
 635                 kfree(hash);
 636         else
 637                 free_pages((unsigned long) hash, get_order(bytes));
 638 }
 639
 640 static void fib_info_hash_move(struct hlist_head *new_info_hash,
 641                                struct hlist_head *new_laddrhash,
 642                                unsigned int new_size)
 643 {
 644         struct hlist_head *old_info_hash, *old_laddrhash;
 645         unsigned int old_size = fib_info_hash_size;
 646         unsigned int i, bytes;
 647
 648         spin_lock_bh(&fib_info_lock);
 649         old_info_hash = fib_info_hash;
 650         old_laddrhash = fib_info_laddrhash;
 651         fib_info_hash_size = new_size;
 652
 653         for (i = 0; i < old_size; i++) {
 654                 struct hlist_head *head = &fib_info_hash[i];
 655                 struct hlist_node *node, *n;
 656                 struct fib_info *fi;
 657
 658                 hlist_for_each_entry_safe(fi, node, n, head, fib_hash) {
 659                         struct hlist_head *dest;
 660                         unsigned int new_hash;
 661
 662                         hlist_del(&fi->fib_hash);
 663
 664                         new_hash = fib_info_hashfn(fi);
 665                         dest = &new_info_hash[new_hash];
 666                         hlist_add_head(&fi->fib_hash, dest);
 667                 }
 668         }
 669         fib_info_hash = new_info_hash;
 670
 671         for (i = 0; i < old_size; i++) {
 672                 struct hlist_head *lhead = &fib_info_laddrhash[i];
 673                 struct hlist_node *node, *n;
 674                 struct fib_info *fi;
 675
 676                 hlist_for_each_entry_safe(fi, node, n, lhead, fib_lhash) {
 677                         struct hlist_head *ldest;
 678                         unsigned int new_hash;
 679
 680                         hlist_del(&fi->fib_lhash);
 681
 682                         new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
 683                         ldest = &new_laddrhash[new_hash];
 684                         hlist_add_head(&fi->fib_lhash, ldest);
 685                 }
 686         }
 687         fib_info_laddrhash = new_laddrhash;
 688
 689         spin_unlock_bh(&fib_info_lock);
 690
 691         bytes = old_size * sizeof(struct hlist_head *);
 692         fib_info_hash_free(old_info_hash, bytes);
 693         fib_info_hash_free(old_laddrhash, bytes);
 694 }
 695
 696 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
 697 {
 698         nh->nh_saddr = inet_select_addr(nh->nh_dev,
 699                                         nh->nh_gw,
 700                                         nh->nh_parent->fib_scope);
 701         nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
 702
 703         return nh->nh_saddr;
 704 }
 705
 706 struct fib_info *fib_create_info(struct fib_config *cfg)
 707 {
 708         int err;
 709         struct fib_info *fi = NULL;
 710         struct fib_info *ofi;
 711         int nhs = 1;
 712         struct net *net = cfg->fc_nlinfo.nl_net;
 713
 714         if (cfg->fc_type > RTN_MAX)
 715                 goto err_inval;
 716
 717         /* Fast check to catch the most weird cases */
 718         if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
 719                 goto err_inval;
 720
 721 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 722         if (cfg->fc_mp) {
 723                 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
 724                 if (nhs == 0)
 725                         goto err_inval;
 726         }
 727 #endif
 728
 729         err = -ENOBUFS;
 730         if (fib_info_cnt >= fib_info_hash_size) {
 731                 unsigned int new_size = fib_info_hash_size << 1;
 732                 struct hlist_head *new_info_hash;
 733                 struct hlist_head *new_laddrhash;
 734                 unsigned int bytes;
 735
 736                 if (!new_size)
 737                         new_size = 1;
 738                 bytes = new_size * sizeof(struct hlist_head *);
 739                 new_info_hash = fib_info_hash_alloc(bytes);
 740                 new_laddrhash = fib_info_hash_alloc(bytes);
 741                 if (!new_info_hash || !new_laddrhash) {
 742                         fib_info_hash_free(new_info_hash, bytes);
 743                         fib_info_hash_free(new_laddrhash, bytes);
 744                 } else
 745                         fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
 746
 747                 if (!fib_info_hash_size)
 748                         goto failure;
 749         }
 750
 751         fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
 752         if (fi == NULL)
 753                 goto failure;
 754         if (cfg->fc_mx) {
 755                 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
 756                 if (!fi->fib_metrics)
 757                         goto failure;
 758         } else
 759                 fi->fib_metrics = (u32 *) dst_default_metrics;
 760         fib_info_cnt++;
 761
 762         fi->fib_net = hold_net(net);
 763         fi->fib_protocol = cfg->fc_protocol;
 764         fi->fib_scope = cfg->fc_scope;
 765         fi->fib_flags = cfg->fc_flags;
 766         fi->fib_priority = cfg->fc_priority;
 767         fi->fib_prefsrc = cfg->fc_prefsrc;
 768
 769         fi->fib_nhs = nhs;
 770         change_nexthops(fi) {
 771                 nexthop_nh->nh_parent = fi;
 772         } endfor_nexthops(fi)
 773
 774         if (cfg->fc_mx) {
 775                 struct nlattr *nla;
 776                 int remaining;
 777
 778                 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
 779                         int type = nla_type(nla);
 780
 781                         if (type) {
 782                                 if (type > RTAX_MAX)
 783                                         goto err_inval;
 784                                 fi->fib_metrics[type - 1] = nla_get_u32(nla);
 785                         }
 786                 }
 787         }
 788
 789         if (cfg->fc_mp) {
 790 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 791                 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
 792                 if (err != 0)
 793                         goto failure;
 794                 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
 795                         goto err_inval;
 796                 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
 797                         goto err_inval;
 798 #ifdef CONFIG_IP_ROUTE_CLASSID
 799                 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
 800                         goto err_inval;
 801 #endif
 802 #else
 803                 goto err_inval;
 804 #endif
 805         } else {
 806                 struct fib_nh *nh = fi->fib_nh;
 807
 808                 nh->nh_oif = cfg->fc_oif;
 809                 nh->nh_gw = cfg->fc_gw;
 810                 nh->nh_flags = cfg->fc_flags;
 811 #ifdef CONFIG_IP_ROUTE_CLASSID
 812                 nh->nh_tclassid = cfg->fc_flow;
 813 #endif
 814 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 815                 nh->nh_weight = 1;
 816 #endif
 817         }
 818
 819         if (fib_props[cfg->fc_type].error) {
 820                 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
 821                         goto err_inval;
 822                 goto link_it;
 823         } else {
 824                 switch (cfg->fc_type) {
 825                 case RTN_UNICAST:
 826                 case RTN_LOCAL:
 827                 case RTN_BROADCAST:
 828                 case RTN_ANYCAST:
 829                 case RTN_MULTICAST:
 830                         break;
 831                 default:
 832                         goto err_inval;
 833                 }
 834         }
 835
 836         if (cfg->fc_scope > RT_SCOPE_HOST)
 837                 goto err_inval;
 838
 839         if (cfg->fc_scope == RT_SCOPE_HOST) {
 840                 struct fib_nh *nh = fi->fib_nh;
 841
 842                 /* Local address is added. */
 843                 if (nhs != 1 || nh->nh_gw)
 844                         goto err_inval;
 845                 nh->nh_scope = RT_SCOPE_NOWHERE;
 846                 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
 847                 err = -ENODEV;
 848                 if (nh->nh_dev == NULL)
 849                         goto failure;
 850         } else {
 851                 change_nexthops(fi) {
 852                         err = fib_check_nh(cfg, fi, nexthop_nh);
 853                         if (err != 0)
 854                                 goto failure;
 855                 } endfor_nexthops(fi)
 856         }
 857
 858         if (fi->fib_prefsrc) {
 859                 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
 860                     fi->fib_prefsrc != cfg->fc_dst)
 861                         if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
 862                                 goto err_inval;
 863         }
 864
 865         change_nexthops(fi) {
 866                 fib_info_update_nh_saddr(net, nexthop_nh);
 867         } endfor_nexthops(fi)
 868
 869 link_it:
 870         ofi = fib_find_info(fi);
 871         if (ofi) {
 872                 fi->fib_dead = 1;
 873                 free_fib_info(fi);
 874                 ofi->fib_treeref++;
 875                 return ofi;
 876         }
 877
 878         fi->fib_treeref++;
 879         atomic_inc(&fi->fib_clntref);
 880         spin_lock_bh(&fib_info_lock);
 881         hlist_add_head(&fi->fib_hash,
 882                        &fib_info_hash[fib_info_hashfn(fi)]);
 883         if (fi->fib_prefsrc) {
 884                 struct hlist_head *head;
 885
 886                 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
 887                 hlist_add_head(&fi->fib_lhash, head);
 888         }
 889         change_nexthops(fi) {
 890                 struct hlist_head *head;
 891                 unsigned int hash;
 892
 893                 if (!nexthop_nh->nh_dev)
 894                         continue;
 895                 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
 896                 head = &fib_info_devhash[hash];
 897                 hlist_add_head(&nexthop_nh->nh_hash, head);
 898         } endfor_nexthops(fi)
 899         spin_unlock_bh(&fib_info_lock);
 900         return fi;
 901
 902 err_inval:
 903         err = -EINVAL;
 904
 905 failure:
 906         if (fi) {
 907                 fi->fib_dead = 1;
 908                 free_fib_info(fi);
 909         }
 910
 911         return ERR_PTR(err);
 912 }
 913
 914 int fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
 915                   u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
 916                   struct fib_info *fi, unsigned int flags)
 917 {
 918         struct nlmsghdr *nlh;
 919         struct rtmsg *rtm;
 920
 921         nlh = nlmsg_put(skb, pid, seq, event, sizeof(*rtm), flags);
 922         if (nlh == NULL)
 923                 return -EMSGSIZE;
 924
 925         rtm = nlmsg_data(nlh);
 926         rtm->rtm_family = AF_INET;
 927         rtm->rtm_dst_len = dst_len;
 928         rtm->rtm_src_len = 0;
 929         rtm->rtm_tos = tos;
 930         if (tb_id < 256)
 931                 rtm->rtm_table = tb_id;
 932         else
 933                 rtm->rtm_table = RT_TABLE_COMPAT;
 934         if (nla_put_u32(skb, RTA_TABLE, tb_id))
 935                 goto nla_put_failure;
 936         rtm->rtm_type = type;
 937         rtm->rtm_flags = fi->fib_flags;
 938         rtm->rtm_scope = fi->fib_scope;
 939         rtm->rtm_protocol = fi->fib_protocol;
 940
 941         if (rtm->rtm_dst_len &&
 942             nla_put_be32(skb, RTA_DST, dst))
 943                 goto nla_put_failure;
 944         if (fi->fib_priority &&
 945             nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
 946                 goto nla_put_failure;
 947         if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
 948                 goto nla_put_failure;
 949
 950         if (fi->fib_prefsrc &&
 951             nla_put_be32(skb, RTA_PREFSRC, fi->fib_prefsrc))
 952                 goto nla_put_failure;
 953         if (fi->fib_nhs == 1) {
 954                 if (fi->fib_nh->nh_gw &&
 955                     nla_put_be32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
 956                         goto nla_put_failure;
 957                 if (fi->fib_nh->nh_oif &&
 958                     nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
 959                         goto nla_put_failure;
 960 #ifdef CONFIG_IP_ROUTE_CLASSID
 961                 if (fi->fib_nh[0].nh_tclassid &&
 962                     nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
 963                         goto nla_put_failure;
 964 #endif
 965         }
 966 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 967         if (fi->fib_nhs > 1) {
 968                 struct rtnexthop *rtnh;
 969                 struct nlattr *mp;
 970
 971                 mp = nla_nest_start(skb, RTA_MULTIPATH);
 972                 if (mp == NULL)
 973                         goto nla_put_failure;
 974
 975                 for_nexthops(fi) {
 976                         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
 977                         if (rtnh == NULL)
 978                                 goto nla_put_failure;
 979
 980                         rtnh->rtnh_flags = nh->nh_flags & 0xFF;
 981                         rtnh->rtnh_hops = nh->nh_weight - 1;
 982                         rtnh->rtnh_ifindex = nh->nh_oif;
 983
 984                         if (nh->nh_gw &&
 985                             nla_put_be32(skb, RTA_GATEWAY, nh->nh_gw))
 986                                 goto nla_put_failure;
 987 #ifdef CONFIG_IP_ROUTE_CLASSID
 988                         if (nh->nh_tclassid &&
 989                             nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
 990                                 goto nla_put_failure;
 991 #endif
 992                         /* length of rtnetlink header + attributes */
 993                         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
 994                 } endfor_nexthops(fi);
 995
 996                 nla_nest_end(skb, mp);
 997         }
 998 #endif
 999         return nlmsg_end(skb, nlh);
1000
1001 nla_put_failure:
1002         nlmsg_cancel(skb, nlh);
1003         return -EMSGSIZE;
1004 }
1005
1006 /*
1007  * Update FIB if:
1008  * - local address disappeared -> we must delete all the entries
1009  *   referring to it.
1010  * - device went down -> we must shutdown all nexthops going via it.
1011  */
1012 int fib_sync_down_addr(struct net *net, __be32 local)
1013 {
1014         int ret = 0;
1015         unsigned int hash = fib_laddr_hashfn(local);
1016         struct hlist_head *head = &fib_info_laddrhash[hash];
1017         struct hlist_node *node;
1018         struct fib_info *fi;
1019
1020         if (fib_info_laddrhash == NULL || local == 0)
1021                 return 0;
1022
1023         hlist_for_each_entry(fi, node, head, fib_lhash) {
1024                 if (!net_eq(fi->fib_net, net))
1025                         continue;
1026                 if (fi->fib_prefsrc == local) {
1027                         fi->fib_flags |= RTNH_F_DEAD;
1028                         ret++;
1029                 }
1030         }
1031         return ret;
1032 }
1033
1034 int fib_sync_down_dev(struct net_device *dev, int force)
1035 {
1036         int ret = 0;
1037         int scope = RT_SCOPE_NOWHERE;
1038         struct fib_info *prev_fi = NULL;
1039         unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1040         struct hlist_head *head = &fib_info_devhash[hash];
1041         struct hlist_node *node;
1042         struct fib_nh *nh;
1043
1044         if (force)
1045                 scope = -1;
1046
1047         hlist_for_each_entry(nh, node, head, nh_hash) {
1048                 struct fib_info *fi = nh->nh_parent;
1049                 int dead;
1050
1051                 BUG_ON(!fi->fib_nhs);
1052                 if (nh->nh_dev != dev || fi == prev_fi)
1053                         continue;
1054                 prev_fi = fi;
1055                 dead = 0;
1056                 change_nexthops(fi) {
1057                         if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1058                                 dead++;
1059                         else if (nexthop_nh->nh_dev == dev &&
1060                                  nexthop_nh->nh_scope != scope) {
1061                                 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1062 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1063                                 spin_lock_bh(&fib_multipath_lock);
1064                                 fi->fib_power -= nexthop_nh->nh_power;
1065                                 nexthop_nh->nh_power = 0;
1066                                 spin_unlock_bh(&fib_multipath_lock);
1067 #endif
1068                                 dead++;
1069                         }
1070 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1071                         if (force > 1 && nexthop_nh->nh_dev == dev) {
1072                                 dead = fi->fib_nhs;
1073                                 break;
1074                         }
1075 #endif
1076                 } endfor_nexthops(fi)
1077                 if (dead == fi->fib_nhs) {
1078                         fi->fib_flags |= RTNH_F_DEAD;
1079                         ret++;
1080                 }
1081         }
1082
1083         return ret;
1084 }
1085
1086 /* Must be invoked inside of an RCU protected region.  */
1087 void fib_select_default(struct fib_result *res)
1088 {
1089         struct fib_info *fi = NULL, *last_resort = NULL;
1090         struct list_head *fa_head = res->fa_head;
1091         struct fib_table *tb = res->table;
1092         int order = -1, last_idx = -1;
1093         struct fib_alias *fa;
1094
1095         list_for_each_entry_rcu(fa, fa_head, fa_list) {
1096                 struct fib_info *next_fi = fa->fa_info;
1097
1098                 if (next_fi->fib_scope != res->scope ||
1099                     fa->fa_type != RTN_UNICAST)
1100                         continue;
1101
1102                 if (next_fi->fib_priority > res->fi->fib_priority)
1103                         break;
1104                 if (!next_fi->fib_nh[0].nh_gw ||
1105                     next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1106                         continue;
1107
1108                 fib_alias_accessed(fa);
1109
1110                 if (fi == NULL) {
1111                         if (next_fi != res->fi)
1112                                 break;
1113                 } else if (!fib_detect_death(fi, order, &last_resort,
1114                                              &last_idx, tb->tb_default)) {
1115                         fib_result_assign(res, fi);
1116                         tb->tb_default = order;
1117                         goto out;
1118                 }
1119                 fi = next_fi;
1120                 order++;
1121         }
1122
1123         if (order <= 0 || fi == NULL) {
1124                 tb->tb_default = -1;
1125                 goto out;
1126         }
1127
1128         if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1129                                 tb->tb_default)) {
1130                 fib_result_assign(res, fi);
1131                 tb->tb_default = order;
1132                 goto out;
1133         }
1134
1135         if (last_idx >= 0)
1136                 fib_result_assign(res, last_resort);
1137         tb->tb_default = last_idx;
1138 out:
1139         return;
1140 }
1141
1142 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1143
1144 /*
1145  * Dead device goes up. We wake up dead nexthops.
1146  * It takes sense only on multipath routes.
1147  */
1148 int fib_sync_up(struct net_device *dev)
1149 {
1150         struct fib_info *prev_fi;
1151         unsigned int hash;
1152         struct hlist_head *head;
1153         struct hlist_node *node;
1154         struct fib_nh *nh;
1155         int ret;
1156
1157         if (!(dev->flags & IFF_UP))
1158                 return 0;
1159
1160         prev_fi = NULL;
1161         hash = fib_devindex_hashfn(dev->ifindex);
1162         head = &fib_info_devhash[hash];
1163         ret = 0;
1164
1165         hlist_for_each_entry(nh, node, head, nh_hash) {
1166                 struct fib_info *fi = nh->nh_parent;
1167                 int alive;
1168
1169                 BUG_ON(!fi->fib_nhs);
1170                 if (nh->nh_dev != dev || fi == prev_fi)
1171                         continue;
1172
1173                 prev_fi = fi;
1174                 alive = 0;
1175                 change_nexthops(fi) {
1176                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1177                                 alive++;
1178                                 continue;
1179                         }
1180                         if (nexthop_nh->nh_dev == NULL ||
1181                             !(nexthop_nh->nh_dev->flags & IFF_UP))
1182                                 continue;
1183                         if (nexthop_nh->nh_dev != dev ||
1184                             !__in_dev_get_rtnl(dev))
1185                                 continue;
1186                         alive++;
1187                         spin_lock_bh(&fib_multipath_lock);
1188                         nexthop_nh->nh_power = 0;
1189                         nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1190                         spin_unlock_bh(&fib_multipath_lock);
1191                 } endfor_nexthops(fi)
1192
1193                 if (alive > 0) {
1194                         fi->fib_flags &= ~RTNH_F_DEAD;
1195                         ret++;
1196                 }
1197         }
1198
1199         return ret;
1200 }
1201
1202 /*
1203  * The algorithm is suboptimal, but it provides really
1204  * fair weighted route distribution.
1205  */
1206 void fib_select_multipath(struct fib_result *res)
1207 {
1208         struct fib_info *fi = res->fi;
1209         int w;
1210
1211         spin_lock_bh(&fib_multipath_lock);
1212         if (fi->fib_power <= 0) {
1213                 int power = 0;
1214                 change_nexthops(fi) {
1215                         if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1216                                 power += nexthop_nh->nh_weight;
1217                                 nexthop_nh->nh_power = nexthop_nh->nh_weight;
1218                         }
1219                 } endfor_nexthops(fi);
1220                 fi->fib_power = power;
1221                 if (power <= 0) {
1222                         spin_unlock_bh(&fib_multipath_lock);
1223                         /* Race condition: route has just become dead. */
1224                         res->nh_sel = 0;
1225                         return;
1226                 }
1227         }
1228
1229
1230         /* w should be random number [0..fi->fib_power-1],
1231          * it is pretty bad approximation.
1232          */
1233
1234         w = jiffies % fi->fib_power;
1235
1236         change_nexthops(fi) {
1237                 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1238                     nexthop_nh->nh_power) {
1239                         w -= nexthop_nh->nh_power;
1240                         if (w <= 0) {
1241                                 nexthop_nh->nh_power--;
1242                                 fi->fib_power--;
1243                                 res->nh_sel = nhsel;
1244                                 spin_unlock_bh(&fib_multipath_lock);
1245                                 return;
1246                         }
1247                 }
1248         } endfor_nexthops(fi);
1249
1250         /* Race condition: route has just become dead. */
1251         res->nh_sel = 0;
1252         spin_unlock_bh(&fib_multipath_lock);
1253 }
1254 #endif