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