net/ipv6/ip6_fib.c

   1 /*
   2  *      Linux INET6 implementation
   3  *      Forwarding Information Database
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <roque@di.fc.ul.pt>
   7  *
   8  *      This program is free software; you can redistribute it and/or
   9  *      modify it under the terms of the GNU General Public License
  10  *      as published by the Free Software Foundation; either version
  11  *      2 of the License, or (at your option) any later version.
  12  *
  13  *      Changes:
  14  *      Yuji SEKIYA @USAGI:     Support default route on router node;
  15  *                              remove ip6_null_entry from the top of
  16  *                              routing table.
  17  *      Ville Nuorvala:         Fixed routing subtrees.
  18  */
  19
  20 #define pr_fmt(fmt) "IPv6: " fmt
  21
  22 #include <linux/errno.h>
  23 #include <linux/types.h>
  24 #include <linux/net.h>
  25 #include <linux/route.h>
  26 #include <linux/netdevice.h>
  27 #include <linux/in6.h>
  28 #include <linux/init.h>
  29 #include <linux/list.h>
  30 #include <linux/slab.h>
  31
  32 #include <net/ipv6.h>
  33 #include <net/ndisc.h>
  34 #include <net/addrconf.h>
  35
  36 #include <net/ip6_fib.h>
  37 #include <net/ip6_route.h>
  38
  39 #define RT6_DEBUG 2
  40
  41 #if RT6_DEBUG >= 3
  42 #define RT6_TRACE(x...) pr_debug(x)
  43 #else
  44 #define RT6_TRACE(x...) do { ; } while (0)
  45 #endif
  46
  47 static struct kmem_cache *fib6_node_kmem __read_mostly;
  48
  49 enum fib_walk_state_t {
  50 #ifdef CONFIG_IPV6_SUBTREES
  51         FWS_S,
  52 #endif
  53         FWS_L,
  54         FWS_R,
  55         FWS_C,
  56         FWS_U
  57 };
  58
  59 struct fib6_cleaner_t {
  60         struct fib6_walker_t w;
  61         struct net *net;
  62         int (*func)(struct rt6_info *, void *arg);
  63         void *arg;
  64 };
  65
  66 static DEFINE_RWLOCK(fib6_walker_lock);
  67
  68 #ifdef CONFIG_IPV6_SUBTREES
  69 #define FWS_INIT FWS_S
  70 #else
  71 #define FWS_INIT FWS_L
  72 #endif
  73
  74 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
  75                               struct rt6_info *rt);
  76 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
  77 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
  78 static int fib6_walk(struct fib6_walker_t *w);
  79 static int fib6_walk_continue(struct fib6_walker_t *w);
  80
  81 /*
  82  *      A routing update causes an increase of the serial number on the
  83  *      affected subtree. This allows for cached routes to be asynchronously
  84  *      tested when modifications are made to the destination cache as a
  85  *      result of redirects, path MTU changes, etc.
  86  */
  87
  88 static __u32 rt_sernum;
  89
  90 static void fib6_gc_timer_cb(unsigned long arg);
  91
  92 static LIST_HEAD(fib6_walkers);
  93 #define FOR_WALKERS(w) list_for_each_entry(w, &fib6_walkers, lh)
  94
  95 static inline void fib6_walker_link(struct fib6_walker_t *w)
  96 {
  97         write_lock_bh(&fib6_walker_lock);
  98         list_add(&w->lh, &fib6_walkers);
  99         write_unlock_bh(&fib6_walker_lock);
 100 }
 101
 102 static inline void fib6_walker_unlink(struct fib6_walker_t *w)
 103 {
 104         write_lock_bh(&fib6_walker_lock);
 105         list_del(&w->lh);
 106         write_unlock_bh(&fib6_walker_lock);
 107 }
 108 static __inline__ u32 fib6_new_sernum(void)
 109 {
 110         u32 n = ++rt_sernum;
 111         if ((__s32)n <= 0)
 112                 rt_sernum = n = 1;
 113         return n;
 114 }
 115
 116 /*
 117  *      Auxiliary address test functions for the radix tree.
 118  *
 119  *      These assume a 32bit processor (although it will work on
 120  *      64bit processors)
 121  */
 122
 123 /*
 124  *      test bit
 125  */
 126 #if defined(__LITTLE_ENDIAN)
 127 # define BITOP_BE32_SWIZZLE     (0x1F & ~7)
 128 #else
 129 # define BITOP_BE32_SWIZZLE     0
 130 #endif
 131
 132 static __inline__ __be32 addr_bit_set(const void *token, int fn_bit)
 133 {
 134         const __be32 *addr = token;
 135         /*
 136          * Here,
 137          *      1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
 138          * is optimized version of
 139          *      htonl(1 << ((~fn_bit)&0x1F))
 140          * See include/asm-generic/bitops/le.h.
 141          */
 142         return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
 143                addr[fn_bit >> 5];
 144 }
 145
 146 static __inline__ struct fib6_node *node_alloc(void)
 147 {
 148         struct fib6_node *fn;
 149
 150         fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
 151
 152         return fn;
 153 }
 154
 155 static __inline__ void node_free(struct fib6_node *fn)
 156 {
 157         kmem_cache_free(fib6_node_kmem, fn);
 158 }
 159
 160 static __inline__ void rt6_release(struct rt6_info *rt)
 161 {
 162         if (atomic_dec_and_test(&rt->rt6i_ref))
 163                 dst_free(&rt->dst);
 164 }
 165
 166 static void fib6_link_table(struct net *net, struct fib6_table *tb)
 167 {
 168         unsigned int h;
 169
 170         /*
 171          * Initialize table lock at a single place to give lockdep a key,
 172          * tables aren't visible prior to being linked to the list.
 173          */
 174         rwlock_init(&tb->tb6_lock);
 175
 176         h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
 177
 178         /*
 179          * No protection necessary, this is the only list mutatation
 180          * operation, tables never disappear once they exist.
 181          */
 182         hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
 183 }
 184
 185 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 186
 187 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
 188 {
 189         struct fib6_table *table;
 190
 191         table = kzalloc(sizeof(*table), GFP_ATOMIC);
 192         if (table) {
 193                 table->tb6_id = id;
 194                 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
 195                 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
 196                 inet_peer_base_init(&table->tb6_peers);
 197         }
 198
 199         return table;
 200 }
 201
 202 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 203 {
 204         struct fib6_table *tb;
 205
 206         if (id == 0)
 207                 id = RT6_TABLE_MAIN;
 208         tb = fib6_get_table(net, id);
 209         if (tb)
 210                 return tb;
 211
 212         tb = fib6_alloc_table(net, id);
 213         if (tb)
 214                 fib6_link_table(net, tb);
 215
 216         return tb;
 217 }
 218
 219 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 220 {
 221         struct fib6_table *tb;
 222         struct hlist_head *head;
 223         unsigned int h;
 224
 225         if (id == 0)
 226                 id = RT6_TABLE_MAIN;
 227         h = id & (FIB6_TABLE_HASHSZ - 1);
 228         rcu_read_lock();
 229         head = &net->ipv6.fib_table_hash[h];
 230         hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 231                 if (tb->tb6_id == id) {
 232                         rcu_read_unlock();
 233                         return tb;
 234                 }
 235         }
 236         rcu_read_unlock();
 237
 238         return NULL;
 239 }
 240
 241 static void __net_init fib6_tables_init(struct net *net)
 242 {
 243         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 244         fib6_link_table(net, net->ipv6.fib6_local_tbl);
 245 }
 246 #else
 247
 248 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 249 {
 250         return fib6_get_table(net, id);
 251 }
 252
 253 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 254 {
 255           return net->ipv6.fib6_main_tbl;
 256 }
 257
 258 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 259                                    int flags, pol_lookup_t lookup)
 260 {
 261         return (struct dst_entry *) lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
 262 }
 263
 264 static void __net_init fib6_tables_init(struct net *net)
 265 {
 266         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 267 }
 268
 269 #endif
 270
 271 static int fib6_dump_node(struct fib6_walker_t *w)
 272 {
 273         int res;
 274         struct rt6_info *rt;
 275
 276         for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
 277                 res = rt6_dump_route(rt, w->args);
 278                 if (res < 0) {
 279                         /* Frame is full, suspend walking */
 280                         w->leaf = rt;
 281                         return 1;
 282                 }
 283                 WARN_ON(res == 0);
 284         }
 285         w->leaf = NULL;
 286         return 0;
 287 }
 288
 289 static void fib6_dump_end(struct netlink_callback *cb)
 290 {
 291         struct fib6_walker_t *w = (void *)cb->args[2];
 292
 293         if (w) {
 294                 if (cb->args[4]) {
 295                         cb->args[4] = 0;
 296                         fib6_walker_unlink(w);
 297                 }
 298                 cb->args[2] = 0;
 299                 kfree(w);
 300         }
 301         cb->done = (void *)cb->args[3];
 302         cb->args[1] = 3;
 303 }
 304
 305 static int fib6_dump_done(struct netlink_callback *cb)
 306 {
 307         fib6_dump_end(cb);
 308         return cb->done ? cb->done(cb) : 0;
 309 }
 310
 311 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
 312                            struct netlink_callback *cb)
 313 {
 314         struct fib6_walker_t *w;
 315         int res;
 316
 317         w = (void *)cb->args[2];
 318         w->root = &table->tb6_root;
 319
 320         if (cb->args[4] == 0) {
 321                 w->count = 0;
 322                 w->skip = 0;
 323
 324                 read_lock_bh(&table->tb6_lock);
 325                 res = fib6_walk(w);
 326                 read_unlock_bh(&table->tb6_lock);
 327                 if (res > 0) {
 328                         cb->args[4] = 1;
 329                         cb->args[5] = w->root->fn_sernum;
 330                 }
 331         } else {
 332                 if (cb->args[5] != w->root->fn_sernum) {
 333                         /* Begin at the root if the tree changed */
 334                         cb->args[5] = w->root->fn_sernum;
 335                         w->state = FWS_INIT;
 336                         w->node = w->root;
 337                         w->skip = w->count;
 338                 } else
 339                         w->skip = 0;
 340
 341                 read_lock_bh(&table->tb6_lock);
 342                 res = fib6_walk_continue(w);
 343                 read_unlock_bh(&table->tb6_lock);
 344                 if (res <= 0) {
 345                         fib6_walker_unlink(w);
 346                         cb->args[4] = 0;
 347                 }
 348         }
 349
 350         return res;
 351 }
 352
 353 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 354 {
 355         struct net *net = sock_net(skb->sk);
 356         unsigned int h, s_h;
 357         unsigned int e = 0, s_e;
 358         struct rt6_rtnl_dump_arg arg;
 359         struct fib6_walker_t *w;
 360         struct fib6_table *tb;
 361         struct hlist_head *head;
 362         int res = 0;
 363
 364         s_h = cb->args[0];
 365         s_e = cb->args[1];
 366
 367         w = (void *)cb->args[2];
 368         if (!w) {
 369                 /* New dump:
 370                  *
 371                  * 1. hook callback destructor.
 372                  */
 373                 cb->args[3] = (long)cb->done;
 374                 cb->done = fib6_dump_done;
 375
 376                 /*
 377                  * 2. allocate and initialize walker.
 378                  */
 379                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
 380                 if (!w)
 381                         return -ENOMEM;
 382                 w->func = fib6_dump_node;
 383                 cb->args[2] = (long)w;
 384         }
 385
 386         arg.skb = skb;
 387         arg.cb = cb;
 388         arg.net = net;
 389         w->args = &arg;
 390
 391         rcu_read_lock();
 392         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
 393                 e = 0;
 394                 head = &net->ipv6.fib_table_hash[h];
 395                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 396                         if (e < s_e)
 397                                 goto next;
 398                         res = fib6_dump_table(tb, skb, cb);
 399                         if (res != 0)
 400                                 goto out;
 401 next:
 402                         e++;
 403                 }
 404         }
 405 out:
 406         rcu_read_unlock();
 407         cb->args[1] = e;
 408         cb->args[0] = h;
 409
 410         res = res < 0 ? res : skb->len;
 411         if (res <= 0)
 412                 fib6_dump_end(cb);
 413         return res;
 414 }
 415
 416 /*
 417  *      Routing Table
 418  *
 419  *      return the appropriate node for a routing tree "add" operation
 420  *      by either creating and inserting or by returning an existing
 421  *      node.
 422  */
 423
 424 static struct fib6_node *fib6_add_1(struct fib6_node *root,
 425                                      struct in6_addr *addr, int plen,
 426                                      int offset, int allow_create,
 427                                      int replace_required)
 428 {
 429         struct fib6_node *fn, *in, *ln;
 430         struct fib6_node *pn = NULL;
 431         struct rt6key *key;
 432         int     bit;
 433         __be32  dir = 0;
 434         __u32   sernum = fib6_new_sernum();
 435
 436         RT6_TRACE("fib6_add_1\n");
 437
 438         /* insert node in tree */
 439
 440         fn = root;
 441
 442         do {
 443                 key = (struct rt6key *)((u8 *)fn->leaf + offset);
 444
 445                 /*
 446                  *      Prefix match
 447                  */
 448                 if (plen < fn->fn_bit ||
 449                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
 450                         if (!allow_create) {
 451                                 if (replace_required) {
 452                                         pr_warn("Can't replace route, no match found\n");
 453                                         return ERR_PTR(-ENOENT);
 454                                 }
 455                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 456                         }
 457                         goto insert_above;
 458                 }
 459
 460                 /*
 461                  *      Exact match ?
 462                  */
 463
 464                 if (plen == fn->fn_bit) {
 465                         /* clean up an intermediate node */
 466                         if (!(fn->fn_flags & RTN_RTINFO)) {
 467                                 rt6_release(fn->leaf);
 468                                 fn->leaf = NULL;
 469                         }
 470
 471                         fn->fn_sernum = sernum;
 472
 473                         return fn;
 474                 }
 475
 476                 /*
 477                  *      We have more bits to go
 478                  */
 479
 480                 /* Try to walk down on tree. */
 481                 fn->fn_sernum = sernum;
 482                 dir = addr_bit_set(addr, fn->fn_bit);
 483                 pn = fn;
 484                 fn = dir ? fn->right : fn->left;
 485         } while (fn);
 486
 487         if (!allow_create) {
 488                 /* We should not create new node because
 489                  * NLM_F_REPLACE was specified without NLM_F_CREATE
 490                  * I assume it is safe to require NLM_F_CREATE when
 491                  * REPLACE flag is used! Later we may want to remove the
 492                  * check for replace_required, because according
 493                  * to netlink specification, NLM_F_CREATE
 494                  * MUST be specified if new route is created.
 495                  * That would keep IPv6 consistent with IPv4
 496                  */
 497                 if (replace_required) {
 498                         pr_warn("Can't replace route, no match found\n");
 499                         return ERR_PTR(-ENOENT);
 500                 }
 501                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 502         }
 503         /*
 504          *      We walked to the bottom of tree.
 505          *      Create new leaf node without children.
 506          */
 507
 508         ln = node_alloc();
 509
 510         if (!ln)
 511                 return ERR_PTR(-ENOMEM);
 512         ln->fn_bit = plen;
 513
 514         ln->parent = pn;
 515         ln->fn_sernum = sernum;
 516
 517         if (dir)
 518                 pn->right = ln;
 519         else
 520                 pn->left  = ln;
 521
 522         return ln;
 523
 524
 525 insert_above:
 526         /*
 527          * split since we don't have a common prefix anymore or
 528          * we have a less significant route.
 529          * we've to insert an intermediate node on the list
 530          * this new node will point to the one we need to create
 531          * and the current
 532          */
 533
 534         pn = fn->parent;
 535
 536         /* find 1st bit in difference between the 2 addrs.
 537
 538            See comment in __ipv6_addr_diff: bit may be an invalid value,
 539            but if it is >= plen, the value is ignored in any case.
 540          */
 541
 542         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
 543
 544         /*
 545          *              (intermediate)[in]
 546          *                /        \
 547          *      (new leaf node)[ln] (old node)[fn]
 548          */
 549         if (plen > bit) {
 550                 in = node_alloc();
 551                 ln = node_alloc();
 552
 553                 if (!in || !ln) {
 554                         if (in)
 555                                 node_free(in);
 556                         if (ln)
 557                                 node_free(ln);
 558                         return ERR_PTR(-ENOMEM);
 559                 }
 560
 561                 /*
 562                  * new intermediate node.
 563                  * RTN_RTINFO will
 564                  * be off since that an address that chooses one of
 565                  * the branches would not match less specific routes
 566                  * in the other branch
 567                  */
 568
 569                 in->fn_bit = bit;
 570
 571                 in->parent = pn;
 572                 in->leaf = fn->leaf;
 573                 atomic_inc(&in->leaf->rt6i_ref);
 574
 575                 in->fn_sernum = sernum;
 576
 577                 /* update parent pointer */
 578                 if (dir)
 579                         pn->right = in;
 580                 else
 581                         pn->left  = in;
 582
 583                 ln->fn_bit = plen;
 584
 585                 ln->parent = in;
 586                 fn->parent = in;
 587
 588                 ln->fn_sernum = sernum;
 589
 590                 if (addr_bit_set(addr, bit)) {
 591                         in->right = ln;
 592                         in->left  = fn;
 593                 } else {
 594                         in->left  = ln;
 595                         in->right = fn;
 596                 }
 597         } else { /* plen <= bit */
 598
 599                 /*
 600                  *              (new leaf node)[ln]
 601                  *                /        \
 602                  *           (old node)[fn] NULL
 603                  */
 604
 605                 ln = node_alloc();
 606
 607                 if (!ln)
 608                         return ERR_PTR(-ENOMEM);
 609
 610                 ln->fn_bit = plen;
 611
 612                 ln->parent = pn;
 613
 614                 ln->fn_sernum = sernum;
 615
 616                 if (dir)
 617                         pn->right = ln;
 618                 else
 619                         pn->left  = ln;
 620
 621                 if (addr_bit_set(&key->addr, plen))
 622                         ln->right = fn;
 623                 else
 624                         ln->left  = fn;
 625
 626                 fn->parent = ln;
 627         }
 628         return ln;
 629 }
 630
 631 static inline bool rt6_qualify_for_ecmp(struct rt6_info *rt)
 632 {
 633         return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
 634                RTF_GATEWAY;
 635 }
 636
 637 static int fib6_commit_metrics(struct dst_entry *dst,
 638                                struct nlattr *mx, int mx_len)
 639 {
 640         struct nlattr *nla;
 641         int remaining;
 642         u32 *mp;
 643
 644         if (dst->flags & DST_HOST) {
 645                 mp = dst_metrics_write_ptr(dst);
 646         } else {
 647                 mp = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
 648                 if (!mp)
 649                         return -ENOMEM;
 650                 dst_init_metrics(dst, mp, 0);
 651         }
 652
 653         nla_for_each_attr(nla, mx, mx_len, remaining) {
 654                 int type = nla_type(nla);
 655
 656                 if (type) {
 657                         if (type > RTAX_MAX)
 658                                 return -EINVAL;
 659
 660                         mp[type - 1] = nla_get_u32(nla);
 661                 }
 662         }
 663         return 0;
 664 }
 665
 666 /*
 667  *      Insert routing information in a node.
 668  */
 669
 670 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
 671                             struct nl_info *info, struct nlattr *mx, int mx_len)
 672 {
 673         struct rt6_info *iter = NULL;
 674         struct rt6_info **ins;
 675         int replace = (info->nlh &&
 676                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
 677         int add = (!info->nlh ||
 678                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
 679         int found = 0;
 680         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
 681         int err;
 682
 683         ins = &fn->leaf;
 684
 685         for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
 686                 /*
 687                  *      Search for duplicates
 688                  */
 689
 690                 if (iter->rt6i_metric == rt->rt6i_metric) {
 691                         /*
 692                          *      Same priority level
 693                          */
 694                         if (info->nlh &&
 695                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
 696                                 return -EEXIST;
 697                         if (replace) {
 698                                 found++;
 699                                 break;
 700                         }
 701
 702                         if (iter->dst.dev == rt->dst.dev &&
 703                             iter->rt6i_idev == rt->rt6i_idev &&
 704                             ipv6_addr_equal(&iter->rt6i_gateway,
 705                                             &rt->rt6i_gateway)) {
 706                                 if (rt->rt6i_nsiblings)
 707                                         rt->rt6i_nsiblings = 0;
 708                                 if (!(iter->rt6i_flags & RTF_EXPIRES))
 709                                         return -EEXIST;
 710                                 if (!(rt->rt6i_flags & RTF_EXPIRES))
 711                                         rt6_clean_expires(iter);
 712                                 else
 713                                         rt6_set_expires(iter, rt->dst.expires);
 714                                 return -EEXIST;
 715                         }
 716                         /* If we have the same destination and the same metric,
 717                          * but not the same gateway, then the route we try to
 718                          * add is sibling to this route, increment our counter
 719                          * of siblings, and later we will add our route to the
 720                          * list.
 721                          * Only static routes (which don't have flag
 722                          * RTF_EXPIRES) are used for ECMPv6.
 723                          *
 724                          * To avoid long list, we only had siblings if the
 725                          * route have a gateway.
 726                          */
 727                         if (rt_can_ecmp &&
 728                             rt6_qualify_for_ecmp(iter))
 729                                 rt->rt6i_nsiblings++;
 730                 }
 731
 732                 if (iter->rt6i_metric > rt->rt6i_metric)
 733                         break;
 734
 735                 ins = &iter->dst.rt6_next;
 736         }
 737
 738         /* Reset round-robin state, if necessary */
 739         if (ins == &fn->leaf)
 740                 fn->rr_ptr = NULL;
 741
 742         /* Link this route to others same route. */
 743         if (rt->rt6i_nsiblings) {
 744                 unsigned int rt6i_nsiblings;
 745                 struct rt6_info *sibling, *temp_sibling;
 746
 747                 /* Find the first route that have the same metric */
 748                 sibling = fn->leaf;
 749                 while (sibling) {
 750                         if (sibling->rt6i_metric == rt->rt6i_metric &&
 751                             rt6_qualify_for_ecmp(sibling)) {
 752                                 list_add_tail(&rt->rt6i_siblings,
 753                                               &sibling->rt6i_siblings);
 754                                 break;
 755                         }
 756                         sibling = sibling->dst.rt6_next;
 757                 }
 758                 /* For each sibling in the list, increment the counter of
 759                  * siblings. BUG() if counters does not match, list of siblings
 760                  * is broken!
 761                  */
 762                 rt6i_nsiblings = 0;
 763                 list_for_each_entry_safe(sibling, temp_sibling,
 764                                          &rt->rt6i_siblings, rt6i_siblings) {
 765                         sibling->rt6i_nsiblings++;
 766                         BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
 767                         rt6i_nsiblings++;
 768                 }
 769                 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
 770         }
 771
 772         /*
 773          *      insert node
 774          */
 775         if (!replace) {
 776                 if (!add)
 777                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
 778
 779 add:
 780                 if (mx) {
 781                         err = fib6_commit_metrics(&rt->dst, mx, mx_len);
 782                         if (err)
 783                                 return err;
 784                 }
 785                 rt->dst.rt6_next = iter;
 786                 *ins = rt;
 787                 rt->rt6i_node = fn;
 788                 atomic_inc(&rt->rt6i_ref);
 789                 inet6_rt_notify(RTM_NEWROUTE, rt, info);
 790                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
 791
 792                 if (!(fn->fn_flags & RTN_RTINFO)) {
 793                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
 794                         fn->fn_flags |= RTN_RTINFO;
 795                 }
 796
 797         } else {
 798                 if (!found) {
 799                         if (add)
 800                                 goto add;
 801                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
 802                         return -ENOENT;
 803                 }
 804                 if (mx) {
 805                         err = fib6_commit_metrics(&rt->dst, mx, mx_len);
 806                         if (err)
 807                                 return err;
 808                 }
 809                 *ins = rt;
 810                 rt->rt6i_node = fn;
 811                 rt->dst.rt6_next = iter->dst.rt6_next;
 812                 atomic_inc(&rt->rt6i_ref);
 813                 inet6_rt_notify(RTM_NEWROUTE, rt, info);
 814                 rt6_release(iter);
 815                 if (!(fn->fn_flags & RTN_RTINFO)) {
 816                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
 817                         fn->fn_flags |= RTN_RTINFO;
 818                 }
 819         }
 820
 821         return 0;
 822 }
 823
 824 static __inline__ void fib6_start_gc(struct net *net, struct rt6_info *rt)
 825 {
 826         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
 827             (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
 828                 mod_timer(&net->ipv6.ip6_fib_timer,
 829                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
 830 }
 831
 832 void fib6_force_start_gc(struct net *net)
 833 {
 834         if (!timer_pending(&net->ipv6.ip6_fib_timer))
 835                 mod_timer(&net->ipv6.ip6_fib_timer,
 836                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
 837 }
 838
 839 /*
 840  *      Add routing information to the routing tree.
 841  *      <destination addr>/<source addr>
 842  *      with source addr info in sub-trees
 843  */
 844
 845 int fib6_add(struct fib6_node *root, struct rt6_info *rt, struct nl_info *info,
 846              struct nlattr *mx, int mx_len)
 847 {
 848         struct fib6_node *fn, *pn = NULL;
 849         int err = -ENOMEM;
 850         int allow_create = 1;
 851         int replace_required = 0;
 852
 853         if (info->nlh) {
 854                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
 855                         allow_create = 0;
 856                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
 857                         replace_required = 1;
 858         }
 859         if (!allow_create && !replace_required)
 860                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
 861
 862         fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
 863                         offsetof(struct rt6_info, rt6i_dst), allow_create,
 864                         replace_required);
 865         if (IS_ERR(fn)) {
 866                 err = PTR_ERR(fn);
 867                 fn = NULL;
 868                 goto out;
 869         }
 870
 871         pn = fn;
 872
 873 #ifdef CONFIG_IPV6_SUBTREES
 874         if (rt->rt6i_src.plen) {
 875                 struct fib6_node *sn;
 876
 877                 if (!fn->subtree) {
 878                         struct fib6_node *sfn;
 879
 880                         /*
 881                          * Create subtree.
 882                          *
 883                          *              fn[main tree]
 884                          *              |
 885                          *              sfn[subtree root]
 886                          *                 \
 887                          *                  sn[new leaf node]
 888                          */
 889
 890                         /* Create subtree root node */
 891                         sfn = node_alloc();
 892                         if (!sfn)
 893                                 goto st_failure;
 894
 895                         sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
 896                         atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
 897                         sfn->fn_flags = RTN_ROOT;
 898                         sfn->fn_sernum = fib6_new_sernum();
 899
 900                         /* Now add the first leaf node to new subtree */
 901
 902                         sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
 903                                         rt->rt6i_src.plen,
 904                                         offsetof(struct rt6_info, rt6i_src),
 905                                         allow_create, replace_required);
 906
 907                         if (IS_ERR(sn)) {
 908                                 /* If it is failed, discard just allocated
 909                                    root, and then (in st_failure) stale node
 910                                    in main tree.
 911                                  */
 912                                 node_free(sfn);
 913                                 err = PTR_ERR(sn);
 914                                 goto st_failure;
 915                         }
 916
 917                         /* Now link new subtree to main tree */
 918                         sfn->parent = fn;
 919                         fn->subtree = sfn;
 920                 } else {
 921                         sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
 922                                         rt->rt6i_src.plen,
 923                                         offsetof(struct rt6_info, rt6i_src),
 924                                         allow_create, replace_required);
 925
 926                         if (IS_ERR(sn)) {
 927                                 err = PTR_ERR(sn);
 928                                 goto st_failure;
 929                         }
 930                 }
 931
 932                 if (!fn->leaf) {
 933                         fn->leaf = rt;
 934                         atomic_inc(&rt->rt6i_ref);
 935                 }
 936                 fn = sn;
 937         }
 938 #endif
 939
 940         err = fib6_add_rt2node(fn, rt, info, mx, mx_len);
 941         if (!err) {
 942                 fib6_start_gc(info->nl_net, rt);
 943                 if (!(rt->rt6i_flags & RTF_CACHE))
 944                         fib6_prune_clones(info->nl_net, pn, rt);
 945         }
 946
 947 out:
 948         if (err) {
 949 #ifdef CONFIG_IPV6_SUBTREES
 950                 /*
 951                  * If fib6_add_1 has cleared the old leaf pointer in the
 952                  * super-tree leaf node we have to find a new one for it.
 953                  */
 954                 if (pn != fn && pn->leaf == rt) {
 955                         pn->leaf = NULL;
 956                         atomic_dec(&rt->rt6i_ref);
 957                 }
 958                 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
 959                         pn->leaf = fib6_find_prefix(info->nl_net, pn);
 960 #if RT6_DEBUG >= 2
 961                         if (!pn->leaf) {
 962                                 WARN_ON(pn->leaf == NULL);
 963                                 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
 964                         }
 965 #endif
 966                         atomic_inc(&pn->leaf->rt6i_ref);
 967                 }
 968 #endif
 969                 dst_free(&rt->dst);
 970         }
 971         return err;
 972
 973 #ifdef CONFIG_IPV6_SUBTREES
 974         /* Subtree creation failed, probably main tree node
 975            is orphan. If it is, shoot it.
 976          */
 977 st_failure:
 978         if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
 979                 fib6_repair_tree(info->nl_net, fn);
 980         dst_free(&rt->dst);
 981         return err;
 982 #endif
 983 }
 984
 985 /*
 986  *      Routing tree lookup
 987  *
 988  */
 989
 990 struct lookup_args {
 991         int                     offset;         /* key offset on rt6_info       */
 992         const struct in6_addr   *addr;          /* search key                   */
 993 };
 994
 995 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
 996                                        struct lookup_args *args)
 997 {
 998         struct fib6_node *fn;
 999         __be32 dir;
1000
1001         if (unlikely(args->offset == 0))
1002                 return NULL;
1003
1004         /*
1005          *      Descend on a tree
1006          */
1007
1008         fn = root;
1009
1010         for (;;) {
1011                 struct fib6_node *next;
1012
1013                 dir = addr_bit_set(args->addr, fn->fn_bit);
1014
1015                 next = dir ? fn->right : fn->left;
1016
1017                 if (next) {
1018                         fn = next;
1019                         continue;
1020                 }
1021                 break;
1022         }
1023
1024         while (fn) {
1025                 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1026                         struct rt6key *key;
1027
1028                         key = (struct rt6key *) ((u8 *) fn->leaf +
1029                                                  args->offset);
1030
1031                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1032 #ifdef CONFIG_IPV6_SUBTREES
1033                                 if (fn->subtree) {
1034                                         struct fib6_node *sfn;
1035                                         sfn = fib6_lookup_1(fn->subtree,
1036                                                             args + 1);
1037                                         if (!sfn)
1038                                                 goto backtrack;
1039                                         fn = sfn;
1040                                 }
1041 #endif
1042                                 if (fn->fn_flags & RTN_RTINFO)
1043                                         return fn;
1044                         }
1045                 }
1046 #ifdef CONFIG_IPV6_SUBTREES
1047 backtrack:
1048 #endif
1049                 if (fn->fn_flags & RTN_ROOT)
1050                         break;
1051
1052                 fn = fn->parent;
1053         }
1054
1055         return NULL;
1056 }
1057
1058 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1059                               const struct in6_addr *saddr)
1060 {
1061         struct fib6_node *fn;
1062         struct lookup_args args[] = {
1063                 {
1064                         .offset = offsetof(struct rt6_info, rt6i_dst),
1065                         .addr = daddr,
1066                 },
1067 #ifdef CONFIG_IPV6_SUBTREES
1068                 {
1069                         .offset = offsetof(struct rt6_info, rt6i_src),
1070                         .addr = saddr,
1071                 },
1072 #endif
1073                 {
1074                         .offset = 0,    /* sentinel */
1075                 }
1076         };
1077
1078         fn = fib6_lookup_1(root, daddr ? args : args + 1);
1079         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1080                 fn = root;
1081
1082         return fn;
1083 }
1084
1085 /*
1086  *      Get node with specified destination prefix (and source prefix,
1087  *      if subtrees are used)
1088  */
1089
1090
1091 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1092                                        const struct in6_addr *addr,
1093                                        int plen, int offset)
1094 {
1095         struct fib6_node *fn;
1096
1097         for (fn = root; fn ; ) {
1098                 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1099
1100                 /*
1101                  *      Prefix match
1102                  */
1103                 if (plen < fn->fn_bit ||
1104                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1105                         return NULL;
1106
1107                 if (plen == fn->fn_bit)
1108                         return fn;
1109
1110                 /*
1111                  *      We have more bits to go
1112                  */
1113                 if (addr_bit_set(addr, fn->fn_bit))
1114                         fn = fn->right;
1115                 else
1116                         fn = fn->left;
1117         }
1118         return NULL;
1119 }
1120
1121 struct fib6_node *fib6_locate(struct fib6_node *root,
1122                               const struct in6_addr *daddr, int dst_len,
1123                               const struct in6_addr *saddr, int src_len)
1124 {
1125         struct fib6_node *fn;
1126
1127         fn = fib6_locate_1(root, daddr, dst_len,
1128                            offsetof(struct rt6_info, rt6i_dst));
1129
1130 #ifdef CONFIG_IPV6_SUBTREES
1131         if (src_len) {
1132                 WARN_ON(saddr == NULL);
1133                 if (fn && fn->subtree)
1134                         fn = fib6_locate_1(fn->subtree, saddr, src_len,
1135                                            offsetof(struct rt6_info, rt6i_src));
1136         }
1137 #endif
1138
1139         if (fn && fn->fn_flags & RTN_RTINFO)
1140                 return fn;
1141
1142         return NULL;
1143 }
1144
1145
1146 /*
1147  *      Deletion
1148  *
1149  */
1150
1151 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1152 {
1153         if (fn->fn_flags & RTN_ROOT)
1154                 return net->ipv6.ip6_null_entry;
1155
1156         while (fn) {
1157                 if (fn->left)
1158                         return fn->left->leaf;
1159                 if (fn->right)
1160                         return fn->right->leaf;
1161
1162                 fn = FIB6_SUBTREE(fn);
1163         }
1164         return NULL;
1165 }
1166
1167 /*
1168  *      Called to trim the tree of intermediate nodes when possible. "fn"
1169  *      is the node we want to try and remove.
1170  */
1171
1172 static struct fib6_node *fib6_repair_tree(struct net *net,
1173                                            struct fib6_node *fn)
1174 {
1175         int children;
1176         int nstate;
1177         struct fib6_node *child, *pn;
1178         struct fib6_walker_t *w;
1179         int iter = 0;
1180
1181         for (;;) {
1182                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1183                 iter++;
1184
1185                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1186                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1187                 WARN_ON(fn->leaf != NULL);
1188
1189                 children = 0;
1190                 child = NULL;
1191                 if (fn->right)
1192                         child = fn->right, children |= 1;
1193                 if (fn->left)
1194                         child = fn->left, children |= 2;
1195
1196                 if (children == 3 || FIB6_SUBTREE(fn)
1197 #ifdef CONFIG_IPV6_SUBTREES
1198                     /* Subtree root (i.e. fn) may have one child */
1199                     || (children && fn->fn_flags & RTN_ROOT)
1200 #endif
1201                     ) {
1202                         fn->leaf = fib6_find_prefix(net, fn);
1203 #if RT6_DEBUG >= 2
1204                         if (!fn->leaf) {
1205                                 WARN_ON(!fn->leaf);
1206                                 fn->leaf = net->ipv6.ip6_null_entry;
1207                         }
1208 #endif
1209                         atomic_inc(&fn->leaf->rt6i_ref);
1210                         return fn->parent;
1211                 }
1212
1213                 pn = fn->parent;
1214 #ifdef CONFIG_IPV6_SUBTREES
1215                 if (FIB6_SUBTREE(pn) == fn) {
1216                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1217                         FIB6_SUBTREE(pn) = NULL;
1218                         nstate = FWS_L;
1219                 } else {
1220                         WARN_ON(fn->fn_flags & RTN_ROOT);
1221 #endif
1222                         if (pn->right == fn)
1223                                 pn->right = child;
1224                         else if (pn->left == fn)
1225                                 pn->left = child;
1226 #if RT6_DEBUG >= 2
1227                         else
1228                                 WARN_ON(1);
1229 #endif
1230                         if (child)
1231                                 child->parent = pn;
1232                         nstate = FWS_R;
1233 #ifdef CONFIG_IPV6_SUBTREES
1234                 }
1235 #endif
1236
1237                 read_lock(&fib6_walker_lock);
1238                 FOR_WALKERS(w) {
1239                         if (!child) {
1240                                 if (w->root == fn) {
1241                                         w->root = w->node = NULL;
1242                                         RT6_TRACE("W %p adjusted by delroot 1\n", w);
1243                                 } else if (w->node == fn) {
1244                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1245                                         w->node = pn;
1246                                         w->state = nstate;
1247                                 }
1248                         } else {
1249                                 if (w->root == fn) {
1250                                         w->root = child;
1251                                         RT6_TRACE("W %p adjusted by delroot 2\n", w);
1252                                 }
1253                                 if (w->node == fn) {
1254                                         w->node = child;
1255                                         if (children&2) {
1256                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1257                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1258                                         } else {
1259                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1260                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1261                                         }
1262                                 }
1263                         }
1264                 }
1265                 read_unlock(&fib6_walker_lock);
1266
1267                 node_free(fn);
1268                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1269                         return pn;
1270
1271                 rt6_release(pn->leaf);
1272                 pn->leaf = NULL;
1273                 fn = pn;
1274         }
1275 }
1276
1277 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1278                            struct nl_info *info)
1279 {
1280         struct fib6_walker_t *w;
1281         struct rt6_info *rt = *rtp;
1282         struct net *net = info->nl_net;
1283
1284         RT6_TRACE("fib6_del_route\n");
1285
1286         /* Unlink it */
1287         *rtp = rt->dst.rt6_next;
1288         rt->rt6i_node = NULL;
1289         net->ipv6.rt6_stats->fib_rt_entries--;
1290         net->ipv6.rt6_stats->fib_discarded_routes++;
1291
1292         /* Reset round-robin state, if necessary */
1293         if (fn->rr_ptr == rt)
1294                 fn->rr_ptr = NULL;
1295
1296         /* Remove this entry from other siblings */
1297         if (rt->rt6i_nsiblings) {
1298                 struct rt6_info *sibling, *next_sibling;
1299
1300                 list_for_each_entry_safe(sibling, next_sibling,
1301                                          &rt->rt6i_siblings, rt6i_siblings)
1302                         sibling->rt6i_nsiblings--;
1303                 rt->rt6i_nsiblings = 0;
1304                 list_del_init(&rt->rt6i_siblings);
1305         }
1306
1307         /* Adjust walkers */
1308         read_lock(&fib6_walker_lock);
1309         FOR_WALKERS(w) {
1310                 if (w->state == FWS_C && w->leaf == rt) {
1311                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1312                         w->leaf = rt->dst.rt6_next;
1313                         if (!w->leaf)
1314                                 w->state = FWS_U;
1315                 }
1316         }
1317         read_unlock(&fib6_walker_lock);
1318
1319         rt->dst.rt6_next = NULL;
1320
1321         /* If it was last route, expunge its radix tree node */
1322         if (!fn->leaf) {
1323                 fn->fn_flags &= ~RTN_RTINFO;
1324                 net->ipv6.rt6_stats->fib_route_nodes--;
1325                 fn = fib6_repair_tree(net, fn);
1326         }
1327
1328         if (atomic_read(&rt->rt6i_ref) != 1) {
1329                 /* This route is used as dummy address holder in some split
1330                  * nodes. It is not leaked, but it still holds other resources,
1331                  * which must be released in time. So, scan ascendant nodes
1332                  * and replace dummy references to this route with references
1333                  * to still alive ones.
1334                  */
1335                 while (fn) {
1336                         if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
1337                                 fn->leaf = fib6_find_prefix(net, fn);
1338                                 atomic_inc(&fn->leaf->rt6i_ref);
1339                                 rt6_release(rt);
1340                         }
1341                         fn = fn->parent;
1342                 }
1343                 /* No more references are possible at this point. */
1344                 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
1345         }
1346
1347         inet6_rt_notify(RTM_DELROUTE, rt, info);
1348         rt6_release(rt);
1349 }
1350
1351 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1352 {
1353         struct net *net = info->nl_net;
1354         struct fib6_node *fn = rt->rt6i_node;
1355         struct rt6_info **rtp;
1356
1357 #if RT6_DEBUG >= 2
1358         if (rt->dst.obsolete > 0) {
1359                 WARN_ON(fn != NULL);
1360                 return -ENOENT;
1361         }
1362 #endif
1363         if (!fn || rt == net->ipv6.ip6_null_entry)
1364                 return -ENOENT;
1365
1366         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1367
1368         if (!(rt->rt6i_flags & RTF_CACHE)) {
1369                 struct fib6_node *pn = fn;
1370 #ifdef CONFIG_IPV6_SUBTREES
1371                 /* clones of this route might be in another subtree */
1372                 if (rt->rt6i_src.plen) {
1373                         while (!(pn->fn_flags & RTN_ROOT))
1374                                 pn = pn->parent;
1375                         pn = pn->parent;
1376                 }
1377 #endif
1378                 fib6_prune_clones(info->nl_net, pn, rt);
1379         }
1380
1381         /*
1382          *      Walk the leaf entries looking for ourself
1383          */
1384
1385         for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1386                 if (*rtp == rt) {
1387                         fib6_del_route(fn, rtp, info);
1388                         return 0;
1389                 }
1390         }
1391         return -ENOENT;
1392 }
1393
1394 /*
1395  *      Tree traversal function.
1396  *
1397  *      Certainly, it is not interrupt safe.
1398  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1399  *      It means, that we can modify tree during walking
1400  *      and use this function for garbage collection, clone pruning,
1401  *      cleaning tree when a device goes down etc. etc.
1402  *
1403  *      It guarantees that every node will be traversed,
1404  *      and that it will be traversed only once.
1405  *
1406  *      Callback function w->func may return:
1407  *      0 -> continue walking.
1408  *      positive value -> walking is suspended (used by tree dumps,
1409  *      and probably by gc, if it will be split to several slices)
1410  *      negative value -> terminate walking.
1411  *
1412  *      The function itself returns:
1413  *      0   -> walk is complete.
1414  *      >0  -> walk is incomplete (i.e. suspended)
1415  *      <0  -> walk is terminated by an error.
1416  */
1417
1418 static int fib6_walk_continue(struct fib6_walker_t *w)
1419 {
1420         struct fib6_node *fn, *pn;
1421
1422         for (;;) {
1423                 fn = w->node;
1424                 if (!fn)
1425                         return 0;
1426
1427                 if (w->prune && fn != w->root &&
1428                     fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1429                         w->state = FWS_C;
1430                         w->leaf = fn->leaf;
1431                 }
1432                 switch (w->state) {
1433 #ifdef CONFIG_IPV6_SUBTREES
1434                 case FWS_S:
1435                         if (FIB6_SUBTREE(fn)) {
1436                                 w->node = FIB6_SUBTREE(fn);
1437                                 continue;
1438                         }
1439                         w->state = FWS_L;
1440 #endif
1441                 case FWS_L:
1442                         if (fn->left) {
1443                                 w->node = fn->left;
1444                                 w->state = FWS_INIT;
1445                                 continue;
1446                         }
1447                         w->state = FWS_R;
1448                 case FWS_R:
1449                         if (fn->right) {
1450                                 w->node = fn->right;
1451                                 w->state = FWS_INIT;
1452                                 continue;
1453                         }
1454                         w->state = FWS_C;
1455                         w->leaf = fn->leaf;
1456                 case FWS_C:
1457                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1458                                 int err;
1459
1460                                 if (w->skip) {
1461                                         w->skip--;
1462                                         continue;
1463                                 }
1464
1465                                 err = w->func(w);
1466                                 if (err)
1467                                         return err;
1468
1469                                 w->count++;
1470                                 continue;
1471                         }
1472                         w->state = FWS_U;
1473                 case FWS_U:
1474                         if (fn == w->root)
1475                                 return 0;
1476                         pn = fn->parent;
1477                         w->node = pn;
1478 #ifdef CONFIG_IPV6_SUBTREES
1479                         if (FIB6_SUBTREE(pn) == fn) {
1480                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1481                                 w->state = FWS_L;
1482                                 continue;
1483                         }
1484 #endif
1485                         if (pn->left == fn) {
1486                                 w->state = FWS_R;
1487                                 continue;
1488                         }
1489                         if (pn->right == fn) {
1490                                 w->state = FWS_C;
1491                                 w->leaf = w->node->leaf;
1492                                 continue;
1493                         }
1494 #if RT6_DEBUG >= 2
1495                         WARN_ON(1);
1496 #endif
1497                 }
1498         }
1499 }
1500
1501 static int fib6_walk(struct fib6_walker_t *w)
1502 {
1503         int res;
1504
1505         w->state = FWS_INIT;
1506         w->node = w->root;
1507
1508         fib6_walker_link(w);
1509         res = fib6_walk_continue(w);
1510         if (res <= 0)
1511                 fib6_walker_unlink(w);
1512         return res;
1513 }
1514
1515 static int fib6_clean_node(struct fib6_walker_t *w)
1516 {
1517         int res;
1518         struct rt6_info *rt;
1519         struct fib6_cleaner_t *c = container_of(w, struct fib6_cleaner_t, w);
1520         struct nl_info info = {
1521                 .nl_net = c->net,
1522         };
1523
1524         for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1525                 res = c->func(rt, c->arg);
1526                 if (res < 0) {
1527                         w->leaf = rt;
1528                         res = fib6_del(rt, &info);
1529                         if (res) {
1530 #if RT6_DEBUG >= 2
1531                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1532                                          __func__, rt, rt->rt6i_node, res);
1533 #endif
1534                                 continue;
1535                         }
1536                         return 0;
1537                 }
1538                 WARN_ON(res != 0);
1539         }
1540         w->leaf = rt;
1541         return 0;
1542 }
1543
1544 /*
1545  *      Convenient frontend to tree walker.
1546  *
1547  *      func is called on each route.
1548  *              It may return -1 -> delete this route.
1549  *                            0  -> continue walking
1550  *
1551  *      prune==1 -> only immediate children of node (certainly,
1552  *      ignoring pure split nodes) will be scanned.
1553  */
1554
1555 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1556                             int (*func)(struct rt6_info *, void *arg),
1557                             int prune, void *arg)
1558 {
1559         struct fib6_cleaner_t c;
1560
1561         c.w.root = root;
1562         c.w.func = fib6_clean_node;
1563         c.w.prune = prune;
1564         c.w.count = 0;
1565         c.w.skip = 0;
1566         c.func = func;
1567         c.arg = arg;
1568         c.net = net;
1569
1570         fib6_walk(&c.w);
1571 }
1572
1573 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *arg),
1574                     void *arg)
1575 {
1576         struct fib6_table *table;
1577         struct hlist_head *head;
1578         unsigned int h;
1579
1580         rcu_read_lock();
1581         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1582                 head = &net->ipv6.fib_table_hash[h];
1583                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1584                         write_lock_bh(&table->tb6_lock);
1585                         fib6_clean_tree(net, &table->tb6_root,
1586                                         func, 0, arg);
1587                         write_unlock_bh(&table->tb6_lock);
1588                 }
1589         }
1590         rcu_read_unlock();
1591 }
1592
1593 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1594 {
1595         if (rt->rt6i_flags & RTF_CACHE) {
1596                 RT6_TRACE("pruning clone %p\n", rt);
1597                 return -1;
1598         }
1599
1600         return 0;
1601 }
1602
1603 static void fib6_prune_clones(struct net *net, struct fib6_node *fn,
1604                               struct rt6_info *rt)
1605 {
1606         fib6_clean_tree(net, fn, fib6_prune_clone, 1, rt);
1607 }
1608
1609 /*
1610  *      Garbage collection
1611  */
1612
1613 static struct fib6_gc_args
1614 {
1615         int                     timeout;
1616         int                     more;
1617 } gc_args;
1618
1619 static int fib6_age(struct rt6_info *rt, void *arg)
1620 {
1621         unsigned long now = jiffies;
1622
1623         /*
1624          *      check addrconf expiration here.
1625          *      Routes are expired even if they are in use.
1626          *
1627          *      Also age clones. Note, that clones are aged out
1628          *      only if they are not in use now.
1629          */
1630
1631         if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1632                 if (time_after(now, rt->dst.expires)) {
1633                         RT6_TRACE("expiring %p\n", rt);
1634                         return -1;
1635                 }
1636                 gc_args.more++;
1637         } else if (rt->rt6i_flags & RTF_CACHE) {
1638                 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1639                     time_after_eq(now, rt->dst.lastuse + gc_args.timeout)) {
1640                         RT6_TRACE("aging clone %p\n", rt);
1641                         return -1;
1642                 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1643                         struct neighbour *neigh;
1644                         __u8 neigh_flags = 0;
1645
1646                         neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1647                         if (neigh) {
1648                                 neigh_flags = neigh->flags;
1649                                 neigh_release(neigh);
1650                         }
1651                         if (!(neigh_flags & NTF_ROUTER)) {
1652                                 RT6_TRACE("purging route %p via non-router but gateway\n",
1653                                           rt);
1654                                 return -1;
1655                         }
1656                 }
1657                 gc_args.more++;
1658         }
1659
1660         return 0;
1661 }
1662
1663 static DEFINE_SPINLOCK(fib6_gc_lock);
1664
1665 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1666 {
1667         unsigned long now;
1668
1669         if (force) {
1670                 spin_lock_bh(&fib6_gc_lock);
1671         } else if (!spin_trylock_bh(&fib6_gc_lock)) {
1672                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1673                 return;
1674         }
1675         gc_args.timeout = expires ? (int)expires :
1676                           net->ipv6.sysctl.ip6_rt_gc_interval;
1677
1678         gc_args.more = icmp6_dst_gc();
1679
1680         fib6_clean_all(net, fib6_age, NULL);
1681         now = jiffies;
1682         net->ipv6.ip6_rt_last_gc = now;
1683
1684         if (gc_args.more)
1685                 mod_timer(&net->ipv6.ip6_fib_timer,
1686                           round_jiffies(now
1687                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
1688         else
1689                 del_timer(&net->ipv6.ip6_fib_timer);
1690         spin_unlock_bh(&fib6_gc_lock);
1691 }
1692
1693 static void fib6_gc_timer_cb(unsigned long arg)
1694 {
1695         fib6_run_gc(0, (struct net *)arg, true);
1696 }
1697
1698 static int __net_init fib6_net_init(struct net *net)
1699 {
1700         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1701
1702         setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1703
1704         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1705         if (!net->ipv6.rt6_stats)
1706                 goto out_timer;
1707
1708         /* Avoid false sharing : Use at least a full cache line */
1709         size = max_t(size_t, size, L1_CACHE_BYTES);
1710
1711         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1712         if (!net->ipv6.fib_table_hash)
1713                 goto out_rt6_stats;
1714
1715         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1716                                           GFP_KERNEL);
1717         if (!net->ipv6.fib6_main_tbl)
1718                 goto out_fib_table_hash;
1719
1720         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1721         net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1722         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1723                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1724         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1725
1726 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1727         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1728                                            GFP_KERNEL);
1729         if (!net->ipv6.fib6_local_tbl)
1730                 goto out_fib6_main_tbl;
1731         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1732         net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1733         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1734                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1735         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1736 #endif
1737         fib6_tables_init(net);
1738
1739         return 0;
1740
1741 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1742 out_fib6_main_tbl:
1743         kfree(net->ipv6.fib6_main_tbl);
1744 #endif
1745 out_fib_table_hash:
1746         kfree(net->ipv6.fib_table_hash);
1747 out_rt6_stats:
1748         kfree(net->ipv6.rt6_stats);
1749 out_timer:
1750         return -ENOMEM;
1751 }
1752
1753 static void fib6_net_exit(struct net *net)
1754 {
1755         rt6_ifdown(net, NULL);
1756         del_timer_sync(&net->ipv6.ip6_fib_timer);
1757
1758 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1759         inetpeer_invalidate_tree(&net->ipv6.fib6_local_tbl->tb6_peers);
1760         kfree(net->ipv6.fib6_local_tbl);
1761 #endif
1762         inetpeer_invalidate_tree(&net->ipv6.fib6_main_tbl->tb6_peers);
1763         kfree(net->ipv6.fib6_main_tbl);
1764         kfree(net->ipv6.fib_table_hash);
1765         kfree(net->ipv6.rt6_stats);
1766 }
1767
1768 static struct pernet_operations fib6_net_ops = {
1769         .init = fib6_net_init,
1770         .exit = fib6_net_exit,
1771 };
1772
1773 int __init fib6_init(void)
1774 {
1775         int ret = -ENOMEM;
1776
1777         fib6_node_kmem = kmem_cache_create("fib6_nodes",
1778                                            sizeof(struct fib6_node),
1779                                            0, SLAB_HWCACHE_ALIGN,
1780                                            NULL);
1781         if (!fib6_node_kmem)
1782                 goto out;
1783
1784         ret = register_pernet_subsys(&fib6_net_ops);
1785         if (ret)
1786                 goto out_kmem_cache_create;
1787
1788         ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1789                               NULL);
1790         if (ret)
1791                 goto out_unregister_subsys;
1792 out:
1793         return ret;
1794
1795 out_unregister_subsys:
1796         unregister_pernet_subsys(&fib6_net_ops);
1797 out_kmem_cache_create:
1798         kmem_cache_destroy(fib6_node_kmem);
1799         goto out;
1800 }
1801
1802 void fib6_gc_cleanup(void)
1803 {
1804         unregister_pernet_subsys(&fib6_net_ops);
1805         kmem_cache_destroy(fib6_node_kmem);
1806 }
1807
1808 #ifdef CONFIG_PROC_FS
1809
1810 struct ipv6_route_iter {
1811         struct seq_net_private p;
1812         struct fib6_walker_t w;
1813         loff_t skip;
1814         struct fib6_table *tbl;
1815         __u32 sernum;
1816 };
1817
1818 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1819 {
1820         struct rt6_info *rt = v;
1821         struct ipv6_route_iter *iter = seq->private;
1822
1823         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1824
1825 #ifdef CONFIG_IPV6_SUBTREES
1826         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1827 #else
1828         seq_puts(seq, "00000000000000000000000000000000 00 ");
1829 #endif
1830         if (rt->rt6i_flags & RTF_GATEWAY)
1831                 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1832         else
1833                 seq_puts(seq, "00000000000000000000000000000000");
1834
1835         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1836                    rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1837                    rt->dst.__use, rt->rt6i_flags,
1838                    rt->dst.dev ? rt->dst.dev->name : "");
1839         iter->w.leaf = NULL;
1840         return 0;
1841 }
1842
1843 static int ipv6_route_yield(struct fib6_walker_t *w)
1844 {
1845         struct ipv6_route_iter *iter = w->args;
1846
1847         if (!iter->skip)
1848                 return 1;
1849
1850         do {
1851                 iter->w.leaf = iter->w.leaf->dst.rt6_next;
1852                 iter->skip--;
1853                 if (!iter->skip && iter->w.leaf)
1854                         return 1;
1855         } while (iter->w.leaf);
1856
1857         return 0;
1858 }
1859
1860 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter)
1861 {
1862         memset(&iter->w, 0, sizeof(iter->w));
1863         iter->w.func = ipv6_route_yield;
1864         iter->w.root = &iter->tbl->tb6_root;
1865         iter->w.state = FWS_INIT;
1866         iter->w.node = iter->w.root;
1867         iter->w.args = iter;
1868         iter->sernum = iter->w.root->fn_sernum;
1869         INIT_LIST_HEAD(&iter->w.lh);
1870         fib6_walker_link(&iter->w);
1871 }
1872
1873 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
1874                                                     struct net *net)
1875 {
1876         unsigned int h;
1877         struct hlist_node *node;
1878
1879         if (tbl) {
1880                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
1881                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
1882         } else {
1883                 h = 0;
1884                 node = NULL;
1885         }
1886
1887         while (!node && h < FIB6_TABLE_HASHSZ) {
1888                 node = rcu_dereference_bh(
1889                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
1890         }
1891         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
1892 }
1893
1894 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
1895 {
1896         if (iter->sernum != iter->w.root->fn_sernum) {
1897                 iter->sernum = iter->w.root->fn_sernum;
1898                 iter->w.state = FWS_INIT;
1899                 iter->w.node = iter->w.root;
1900                 WARN_ON(iter->w.skip);
1901                 iter->w.skip = iter->w.count;
1902         }
1903 }
1904
1905 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1906 {
1907         int r;
1908         struct rt6_info *n;
1909         struct net *net = seq_file_net(seq);
1910         struct ipv6_route_iter *iter = seq->private;
1911
1912         if (!v)
1913                 goto iter_table;
1914
1915         n = ((struct rt6_info *)v)->dst.rt6_next;
1916         if (n) {
1917                 ++*pos;
1918                 return n;
1919         }
1920
1921 iter_table:
1922         ipv6_route_check_sernum(iter);
1923         read_lock(&iter->tbl->tb6_lock);
1924         r = fib6_walk_continue(&iter->w);
1925         read_unlock(&iter->tbl->tb6_lock);
1926         if (r > 0) {
1927                 if (v)
1928                         ++*pos;
1929                 return iter->w.leaf;
1930         } else if (r < 0) {
1931                 fib6_walker_unlink(&iter->w);
1932                 return NULL;
1933         }
1934         fib6_walker_unlink(&iter->w);
1935
1936         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
1937         if (!iter->tbl)
1938                 return NULL;
1939
1940         ipv6_route_seq_setup_walk(iter);
1941         goto iter_table;
1942 }
1943
1944 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
1945         __acquires(RCU_BH)
1946 {
1947         struct net *net = seq_file_net(seq);
1948         struct ipv6_route_iter *iter = seq->private;
1949
1950         rcu_read_lock_bh();
1951         iter->tbl = ipv6_route_seq_next_table(NULL, net);
1952         iter->skip = *pos;
1953
1954         if (iter->tbl) {
1955                 ipv6_route_seq_setup_walk(iter);
1956                 return ipv6_route_seq_next(seq, NULL, pos);
1957         } else {
1958                 return NULL;
1959         }
1960 }
1961
1962 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
1963 {
1964         struct fib6_walker_t *w = &iter->w;
1965         return w->node && !(w->state == FWS_U && w->node == w->root);
1966 }
1967
1968 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
1969         __releases(RCU_BH)
1970 {
1971         struct ipv6_route_iter *iter = seq->private;
1972
1973         if (ipv6_route_iter_active(iter))
1974                 fib6_walker_unlink(&iter->w);
1975
1976         rcu_read_unlock_bh();
1977 }
1978
1979 static const struct seq_operations ipv6_route_seq_ops = {
1980         .start  = ipv6_route_seq_start,
1981         .next   = ipv6_route_seq_next,
1982         .stop   = ipv6_route_seq_stop,
1983         .show   = ipv6_route_seq_show
1984 };
1985
1986 int ipv6_route_open(struct inode *inode, struct file *file)
1987 {
1988         return seq_open_net(inode, file, &ipv6_route_seq_ops,
1989                             sizeof(struct ipv6_route_iter));
1990 }
1991
1992 #endif /* CONFIG_PROC_FS */