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 #include <net/lwtunnel.h>
  36
  37 #include <net/ip6_fib.h>
  38 #include <net/ip6_route.h>
  39
  40 #define RT6_DEBUG 2
  41
  42 #if RT6_DEBUG >= 3
  43 #define RT6_TRACE(x...) pr_debug(x)
  44 #else
  45 #define RT6_TRACE(x...) do { ; } while (0)
  46 #endif
  47
  48 static struct kmem_cache *fib6_node_kmem __read_mostly;
  49
  50 struct fib6_cleaner {
  51         struct fib6_walker w;
  52         struct net *net;
  53         int (*func)(struct rt6_info *, void *arg);
  54         int sernum;
  55         void *arg;
  56 };
  57
  58 #ifdef CONFIG_IPV6_SUBTREES
  59 #define FWS_INIT FWS_S
  60 #else
  61 #define FWS_INIT FWS_L
  62 #endif
  63
  64 static void fib6_prune_clones(struct net *net, struct fib6_node *fn);
  65 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
  66 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
  67 static int fib6_walk(struct net *net, struct fib6_walker *w);
  68 static int fib6_walk_continue(struct fib6_walker *w);
  69
  70 /*
  71  *      A routing update causes an increase of the serial number on the
  72  *      affected subtree. This allows for cached routes to be asynchronously
  73  *      tested when modifications are made to the destination cache as a
  74  *      result of redirects, path MTU changes, etc.
  75  */
  76
  77 static void fib6_gc_timer_cb(unsigned long arg);
  78
  79 #define FOR_WALKERS(net, w) \
  80         list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
  81
  82 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
  83 {
  84         write_lock_bh(&net->ipv6.fib6_walker_lock);
  85         list_add(&w->lh, &net->ipv6.fib6_walkers);
  86         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  87 }
  88
  89 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
  90 {
  91         write_lock_bh(&net->ipv6.fib6_walker_lock);
  92         list_del(&w->lh);
  93         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  94 }
  95
  96 static int fib6_new_sernum(struct net *net)
  97 {
  98         int new, old;
  99
 100         do {
 101                 old = atomic_read(&net->ipv6.fib6_sernum);
 102                 new = old < INT_MAX ? old + 1 : 1;
 103         } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
 104                                 old, new) != old);
 105         return new;
 106 }
 107
 108 enum {
 109         FIB6_NO_SERNUM_CHANGE = 0,
 110 };
 111
 112 /*
 113  *      Auxiliary address test functions for the radix tree.
 114  *
 115  *      These assume a 32bit processor (although it will work on
 116  *      64bit processors)
 117  */
 118
 119 /*
 120  *      test bit
 121  */
 122 #if defined(__LITTLE_ENDIAN)
 123 # define BITOP_BE32_SWIZZLE     (0x1F & ~7)
 124 #else
 125 # define BITOP_BE32_SWIZZLE     0
 126 #endif
 127
 128 static __be32 addr_bit_set(const void *token, int fn_bit)
 129 {
 130         const __be32 *addr = token;
 131         /*
 132          * Here,
 133          *      1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
 134          * is optimized version of
 135          *      htonl(1 << ((~fn_bit)&0x1F))
 136          * See include/asm-generic/bitops/le.h.
 137          */
 138         return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
 139                addr[fn_bit >> 5];
 140 }
 141
 142 static struct fib6_node *node_alloc(void)
 143 {
 144         struct fib6_node *fn;
 145
 146         fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
 147
 148         return fn;
 149 }
 150
 151 static void node_free_immediate(struct fib6_node *fn)
 152 {
 153         kmem_cache_free(fib6_node_kmem, fn);
 154 }
 155
 156 static void node_free_rcu(struct rcu_head *head)
 157 {
 158         struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
 159
 160         kmem_cache_free(fib6_node_kmem, fn);
 161 }
 162
 163 static void node_free(struct fib6_node *fn)
 164 {
 165         call_rcu(&fn->rcu, node_free_rcu);
 166 }
 167
 168 static void rt6_rcu_free(struct rt6_info *rt)
 169 {
 170         call_rcu(&rt->dst.rcu_head, dst_rcu_free);
 171 }
 172
 173 static void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
 174 {
 175         int cpu;
 176
 177         if (!non_pcpu_rt->rt6i_pcpu)
 178                 return;
 179
 180         for_each_possible_cpu(cpu) {
 181                 struct rt6_info **ppcpu_rt;
 182                 struct rt6_info *pcpu_rt;
 183
 184                 ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
 185                 pcpu_rt = *ppcpu_rt;
 186                 if (pcpu_rt) {
 187                         rt6_rcu_free(pcpu_rt);
 188                         *ppcpu_rt = NULL;
 189                 }
 190         }
 191
 192         free_percpu(non_pcpu_rt->rt6i_pcpu);
 193         non_pcpu_rt->rt6i_pcpu = NULL;
 194 }
 195
 196 static void rt6_release(struct rt6_info *rt)
 197 {
 198         if (atomic_dec_and_test(&rt->rt6i_ref)) {
 199                 rt6_free_pcpu(rt);
 200                 rt6_rcu_free(rt);
 201         }
 202 }
 203
 204 static void fib6_free_table(struct fib6_table *table)
 205 {
 206         inetpeer_invalidate_tree(&table->tb6_peers);
 207         kfree(table);
 208 }
 209
 210 static void fib6_link_table(struct net *net, struct fib6_table *tb)
 211 {
 212         unsigned int h;
 213
 214         /*
 215          * Initialize table lock at a single place to give lockdep a key,
 216          * tables aren't visible prior to being linked to the list.
 217          */
 218         rwlock_init(&tb->tb6_lock);
 219
 220         h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
 221
 222         /*
 223          * No protection necessary, this is the only list mutatation
 224          * operation, tables never disappear once they exist.
 225          */
 226         hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
 227 }
 228
 229 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 230
 231 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
 232 {
 233         struct fib6_table *table;
 234
 235         table = kzalloc(sizeof(*table), GFP_ATOMIC);
 236         if (table) {
 237                 table->tb6_id = id;
 238                 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
 239                 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
 240                 inet_peer_base_init(&table->tb6_peers);
 241         }
 242
 243         return table;
 244 }
 245
 246 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 247 {
 248         struct fib6_table *tb;
 249
 250         if (id == 0)
 251                 id = RT6_TABLE_MAIN;
 252         tb = fib6_get_table(net, id);
 253         if (tb)
 254                 return tb;
 255
 256         tb = fib6_alloc_table(net, id);
 257         if (tb)
 258                 fib6_link_table(net, tb);
 259
 260         return tb;
 261 }
 262 EXPORT_SYMBOL_GPL(fib6_new_table);
 263
 264 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 265 {
 266         struct fib6_table *tb;
 267         struct hlist_head *head;
 268         unsigned int h;
 269
 270         if (id == 0)
 271                 id = RT6_TABLE_MAIN;
 272         h = id & (FIB6_TABLE_HASHSZ - 1);
 273         rcu_read_lock();
 274         head = &net->ipv6.fib_table_hash[h];
 275         hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 276                 if (tb->tb6_id == id) {
 277                         rcu_read_unlock();
 278                         return tb;
 279                 }
 280         }
 281         rcu_read_unlock();
 282
 283         return NULL;
 284 }
 285 EXPORT_SYMBOL_GPL(fib6_get_table);
 286
 287 static void __net_init fib6_tables_init(struct net *net)
 288 {
 289         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 290         fib6_link_table(net, net->ipv6.fib6_local_tbl);
 291 }
 292 #else
 293
 294 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 295 {
 296         return fib6_get_table(net, id);
 297 }
 298
 299 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 300 {
 301           return net->ipv6.fib6_main_tbl;
 302 }
 303
 304 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 305                                    int flags, pol_lookup_t lookup)
 306 {
 307         struct rt6_info *rt;
 308
 309         rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
 310         if (rt->dst.error == -EAGAIN) {
 311                 ip6_rt_put(rt);
 312                 rt = net->ipv6.ip6_null_entry;
 313                 dst_hold(&rt->dst);
 314         }
 315
 316         return &rt->dst;
 317 }
 318
 319 static void __net_init fib6_tables_init(struct net *net)
 320 {
 321         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 322 }
 323
 324 #endif
 325
 326 static int fib6_dump_node(struct fib6_walker *w)
 327 {
 328         int res;
 329         struct rt6_info *rt;
 330
 331         for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
 332                 res = rt6_dump_route(rt, w->args);
 333                 if (res < 0) {
 334                         /* Frame is full, suspend walking */
 335                         w->leaf = rt;
 336                         return 1;
 337                 }
 338         }
 339         w->leaf = NULL;
 340         return 0;
 341 }
 342
 343 static void fib6_dump_end(struct netlink_callback *cb)
 344 {
 345         struct net *net = sock_net(cb->skb->sk);
 346         struct fib6_walker *w = (void *)cb->args[2];
 347
 348         if (w) {
 349                 if (cb->args[4]) {
 350                         cb->args[4] = 0;
 351                         fib6_walker_unlink(net, w);
 352                 }
 353                 cb->args[2] = 0;
 354                 kfree(w);
 355         }
 356         cb->done = (void *)cb->args[3];
 357         cb->args[1] = 3;
 358 }
 359
 360 static int fib6_dump_done(struct netlink_callback *cb)
 361 {
 362         fib6_dump_end(cb);
 363         return cb->done ? cb->done(cb) : 0;
 364 }
 365
 366 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
 367                            struct netlink_callback *cb)
 368 {
 369         struct net *net = sock_net(skb->sk);
 370         struct fib6_walker *w;
 371         int res;
 372
 373         w = (void *)cb->args[2];
 374         w->root = &table->tb6_root;
 375
 376         if (cb->args[4] == 0) {
 377                 w->count = 0;
 378                 w->skip = 0;
 379
 380                 read_lock_bh(&table->tb6_lock);
 381                 res = fib6_walk(net, w);
 382                 read_unlock_bh(&table->tb6_lock);
 383                 if (res > 0) {
 384                         cb->args[4] = 1;
 385                         cb->args[5] = w->root->fn_sernum;
 386                 }
 387         } else {
 388                 if (cb->args[5] != w->root->fn_sernum) {
 389                         /* Begin at the root if the tree changed */
 390                         cb->args[5] = w->root->fn_sernum;
 391                         w->state = FWS_INIT;
 392                         w->node = w->root;
 393                         w->skip = w->count;
 394                 } else
 395                         w->skip = 0;
 396
 397                 read_lock_bh(&table->tb6_lock);
 398                 res = fib6_walk_continue(w);
 399                 read_unlock_bh(&table->tb6_lock);
 400                 if (res <= 0) {
 401                         fib6_walker_unlink(net, w);
 402                         cb->args[4] = 0;
 403                 }
 404         }
 405
 406         return res;
 407 }
 408
 409 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 410 {
 411         struct net *net = sock_net(skb->sk);
 412         unsigned int h, s_h;
 413         unsigned int e = 0, s_e;
 414         struct rt6_rtnl_dump_arg arg;
 415         struct fib6_walker *w;
 416         struct fib6_table *tb;
 417         struct hlist_head *head;
 418         int res = 0;
 419
 420         s_h = cb->args[0];
 421         s_e = cb->args[1];
 422
 423         w = (void *)cb->args[2];
 424         if (!w) {
 425                 /* New dump:
 426                  *
 427                  * 1. hook callback destructor.
 428                  */
 429                 cb->args[3] = (long)cb->done;
 430                 cb->done = fib6_dump_done;
 431
 432                 /*
 433                  * 2. allocate and initialize walker.
 434                  */
 435                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
 436                 if (!w)
 437                         return -ENOMEM;
 438                 w->func = fib6_dump_node;
 439                 cb->args[2] = (long)w;
 440         }
 441
 442         arg.skb = skb;
 443         arg.cb = cb;
 444         arg.net = net;
 445         w->args = &arg;
 446
 447         rcu_read_lock();
 448         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
 449                 e = 0;
 450                 head = &net->ipv6.fib_table_hash[h];
 451                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 452                         if (e < s_e)
 453                                 goto next;
 454                         res = fib6_dump_table(tb, skb, cb);
 455                         if (res != 0)
 456                                 goto out;
 457 next:
 458                         e++;
 459                 }
 460         }
 461 out:
 462         rcu_read_unlock();
 463         cb->args[1] = e;
 464         cb->args[0] = h;
 465
 466         res = res < 0 ? res : skb->len;
 467         if (res <= 0)
 468                 fib6_dump_end(cb);
 469         return res;
 470 }
 471
 472 /*
 473  *      Routing Table
 474  *
 475  *      return the appropriate node for a routing tree "add" operation
 476  *      by either creating and inserting or by returning an existing
 477  *      node.
 478  */
 479
 480 static struct fib6_node *fib6_add_1(struct fib6_node *root,
 481                                      struct in6_addr *addr, int plen,
 482                                      int offset, int allow_create,
 483                                      int replace_required, int sernum)
 484 {
 485         struct fib6_node *fn, *in, *ln;
 486         struct fib6_node *pn = NULL;
 487         struct rt6key *key;
 488         int     bit;
 489         __be32  dir = 0;
 490
 491         RT6_TRACE("fib6_add_1\n");
 492
 493         /* insert node in tree */
 494
 495         fn = root;
 496
 497         do {
 498                 key = (struct rt6key *)((u8 *)fn->leaf + offset);
 499
 500                 /*
 501                  *      Prefix match
 502                  */
 503                 if (plen < fn->fn_bit ||
 504                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
 505                         if (!allow_create) {
 506                                 if (replace_required) {
 507                                         pr_warn("Can't replace route, no match found\n");
 508                                         return ERR_PTR(-ENOENT);
 509                                 }
 510                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 511                         }
 512                         goto insert_above;
 513                 }
 514
 515                 /*
 516                  *      Exact match ?
 517                  */
 518
 519                 if (plen == fn->fn_bit) {
 520                         /* clean up an intermediate node */
 521                         if (!(fn->fn_flags & RTN_RTINFO)) {
 522                                 rt6_release(fn->leaf);
 523                                 fn->leaf = NULL;
 524                         }
 525
 526                         fn->fn_sernum = sernum;
 527
 528                         return fn;
 529                 }
 530
 531                 /*
 532                  *      We have more bits to go
 533                  */
 534
 535                 /* Try to walk down on tree. */
 536                 fn->fn_sernum = sernum;
 537                 dir = addr_bit_set(addr, fn->fn_bit);
 538                 pn = fn;
 539                 fn = dir ? fn->right : fn->left;
 540         } while (fn);
 541
 542         if (!allow_create) {
 543                 /* We should not create new node because
 544                  * NLM_F_REPLACE was specified without NLM_F_CREATE
 545                  * I assume it is safe to require NLM_F_CREATE when
 546                  * REPLACE flag is used! Later we may want to remove the
 547                  * check for replace_required, because according
 548                  * to netlink specification, NLM_F_CREATE
 549                  * MUST be specified if new route is created.
 550                  * That would keep IPv6 consistent with IPv4
 551                  */
 552                 if (replace_required) {
 553                         pr_warn("Can't replace route, no match found\n");
 554                         return ERR_PTR(-ENOENT);
 555                 }
 556                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 557         }
 558         /*
 559          *      We walked to the bottom of tree.
 560          *      Create new leaf node without children.
 561          */
 562
 563         ln = node_alloc();
 564
 565         if (!ln)
 566                 return ERR_PTR(-ENOMEM);
 567         ln->fn_bit = plen;
 568
 569         ln->parent = pn;
 570         ln->fn_sernum = sernum;
 571
 572         if (dir)
 573                 pn->right = ln;
 574         else
 575                 pn->left  = ln;
 576
 577         return ln;
 578
 579
 580 insert_above:
 581         /*
 582          * split since we don't have a common prefix anymore or
 583          * we have a less significant route.
 584          * we've to insert an intermediate node on the list
 585          * this new node will point to the one we need to create
 586          * and the current
 587          */
 588
 589         pn = fn->parent;
 590
 591         /* find 1st bit in difference between the 2 addrs.
 592
 593            See comment in __ipv6_addr_diff: bit may be an invalid value,
 594            but if it is >= plen, the value is ignored in any case.
 595          */
 596
 597         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
 598
 599         /*
 600          *              (intermediate)[in]
 601          *                /        \
 602          *      (new leaf node)[ln] (old node)[fn]
 603          */
 604         if (plen > bit) {
 605                 in = node_alloc();
 606                 ln = node_alloc();
 607
 608                 if (!in || !ln) {
 609                         if (in)
 610                                 node_free_immediate(in);
 611                         if (ln)
 612                                 node_free_immediate(ln);
 613                         return ERR_PTR(-ENOMEM);
 614                 }
 615
 616                 /*
 617                  * new intermediate node.
 618                  * RTN_RTINFO will
 619                  * be off since that an address that chooses one of
 620                  * the branches would not match less specific routes
 621                  * in the other branch
 622                  */
 623
 624                 in->fn_bit = bit;
 625
 626                 in->parent = pn;
 627                 in->leaf = fn->leaf;
 628                 atomic_inc(&in->leaf->rt6i_ref);
 629
 630                 in->fn_sernum = sernum;
 631
 632                 /* update parent pointer */
 633                 if (dir)
 634                         pn->right = in;
 635                 else
 636                         pn->left  = in;
 637
 638                 ln->fn_bit = plen;
 639
 640                 ln->parent = in;
 641                 fn->parent = in;
 642
 643                 ln->fn_sernum = sernum;
 644
 645                 if (addr_bit_set(addr, bit)) {
 646                         in->right = ln;
 647                         in->left  = fn;
 648                 } else {
 649                         in->left  = ln;
 650                         in->right = fn;
 651                 }
 652         } else { /* plen <= bit */
 653
 654                 /*
 655                  *              (new leaf node)[ln]
 656                  *                /        \
 657                  *           (old node)[fn] NULL
 658                  */
 659
 660                 ln = node_alloc();
 661
 662                 if (!ln)
 663                         return ERR_PTR(-ENOMEM);
 664
 665                 ln->fn_bit = plen;
 666
 667                 ln->parent = pn;
 668
 669                 ln->fn_sernum = sernum;
 670
 671                 if (dir)
 672                         pn->right = ln;
 673                 else
 674                         pn->left  = ln;
 675
 676                 if (addr_bit_set(&key->addr, plen))
 677                         ln->right = fn;
 678                 else
 679                         ln->left  = fn;
 680
 681                 fn->parent = ln;
 682         }
 683         return ln;
 684 }
 685
 686 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
 687 {
 688         return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
 689                RTF_GATEWAY;
 690 }
 691
 692 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
 693 {
 694         int i;
 695
 696         for (i = 0; i < RTAX_MAX; i++) {
 697                 if (test_bit(i, mxc->mx_valid))
 698                         mp[i] = mxc->mx[i];
 699         }
 700 }
 701
 702 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
 703 {
 704         if (!mxc->mx)
 705                 return 0;
 706
 707         if (dst->flags & DST_HOST) {
 708                 u32 *mp = dst_metrics_write_ptr(dst);
 709
 710                 if (unlikely(!mp))
 711                         return -ENOMEM;
 712
 713                 fib6_copy_metrics(mp, mxc);
 714         } else {
 715                 dst_init_metrics(dst, mxc->mx, false);
 716
 717                 /* We've stolen mx now. */
 718                 mxc->mx = NULL;
 719         }
 720
 721         return 0;
 722 }
 723
 724 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
 725                           struct net *net)
 726 {
 727         if (atomic_read(&rt->rt6i_ref) != 1) {
 728                 /* This route is used as dummy address holder in some split
 729                  * nodes. It is not leaked, but it still holds other resources,
 730                  * which must be released in time. So, scan ascendant nodes
 731                  * and replace dummy references to this route with references
 732                  * to still alive ones.
 733                  */
 734                 while (fn) {
 735                         if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
 736                                 fn->leaf = fib6_find_prefix(net, fn);
 737                                 atomic_inc(&fn->leaf->rt6i_ref);
 738                                 rt6_release(rt);
 739                         }
 740                         fn = fn->parent;
 741                 }
 742                 /* No more references are possible at this point. */
 743                 BUG_ON(atomic_read(&rt->rt6i_ref) != 1);
 744         }
 745 }
 746
 747 /*
 748  *      Insert routing information in a node.
 749  */
 750
 751 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
 752                             struct nl_info *info, struct mx6_config *mxc)
 753 {
 754         struct rt6_info *iter = NULL;
 755         struct rt6_info **ins;
 756         struct rt6_info **fallback_ins = NULL;
 757         int replace = (info->nlh &&
 758                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
 759         int add = (!info->nlh ||
 760                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
 761         int found = 0;
 762         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
 763         u16 nlflags = NLM_F_EXCL;
 764         int err;
 765
 766         ins = &fn->leaf;
 767
 768         for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
 769                 /*
 770                  *      Search for duplicates
 771                  */
 772
 773                 if (iter->rt6i_metric == rt->rt6i_metric) {
 774                         /*
 775                          *      Same priority level
 776                          */
 777                         if (info->nlh &&
 778                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
 779                                 return -EEXIST;
 780
 781                         nlflags &= ~NLM_F_EXCL;
 782                         if (replace) {
 783                                 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
 784                                         found++;
 785                                         break;
 786                                 }
 787                                 fallback_ins = fallback_ins ?: ins;
 788                                 goto next_iter;
 789                         }
 790
 791                         if (rt6_duplicate_nexthop(iter, rt)) {
 792                                 if (rt->rt6i_nsiblings)
 793                                         rt->rt6i_nsiblings = 0;
 794                                 if (!(iter->rt6i_flags & RTF_EXPIRES))
 795                                         return -EEXIST;
 796                                 if (!(rt->rt6i_flags & RTF_EXPIRES))
 797                                         rt6_clean_expires(iter);
 798                                 else
 799                                         rt6_set_expires(iter, rt->dst.expires);
 800                                 iter->rt6i_pmtu = rt->rt6i_pmtu;
 801                                 return -EEXIST;
 802                         }
 803                         /* If we have the same destination and the same metric,
 804                          * but not the same gateway, then the route we try to
 805                          * add is sibling to this route, increment our counter
 806                          * of siblings, and later we will add our route to the
 807                          * list.
 808                          * Only static routes (which don't have flag
 809                          * RTF_EXPIRES) are used for ECMPv6.
 810                          *
 811                          * To avoid long list, we only had siblings if the
 812                          * route have a gateway.
 813                          */
 814                         if (rt_can_ecmp &&
 815                             rt6_qualify_for_ecmp(iter))
 816                                 rt->rt6i_nsiblings++;
 817                 }
 818
 819                 if (iter->rt6i_metric > rt->rt6i_metric)
 820                         break;
 821
 822 next_iter:
 823                 ins = &iter->dst.rt6_next;
 824         }
 825
 826         if (fallback_ins && !found) {
 827                 /* No matching route with same ecmp-able-ness found, replace
 828                  * first matching route
 829                  */
 830                 ins = fallback_ins;
 831                 iter = *ins;
 832                 found++;
 833         }
 834
 835         /* Reset round-robin state, if necessary */
 836         if (ins == &fn->leaf)
 837                 fn->rr_ptr = NULL;
 838
 839         /* Link this route to others same route. */
 840         if (rt->rt6i_nsiblings) {
 841                 unsigned int rt6i_nsiblings;
 842                 struct rt6_info *sibling, *temp_sibling;
 843
 844                 /* Find the first route that have the same metric */
 845                 sibling = fn->leaf;
 846                 while (sibling) {
 847                         if (sibling->rt6i_metric == rt->rt6i_metric &&
 848                             rt6_qualify_for_ecmp(sibling)) {
 849                                 list_add_tail(&rt->rt6i_siblings,
 850                                               &sibling->rt6i_siblings);
 851                                 break;
 852                         }
 853                         sibling = sibling->dst.rt6_next;
 854                 }
 855                 /* For each sibling in the list, increment the counter of
 856                  * siblings. BUG() if counters does not match, list of siblings
 857                  * is broken!
 858                  */
 859                 rt6i_nsiblings = 0;
 860                 list_for_each_entry_safe(sibling, temp_sibling,
 861                                          &rt->rt6i_siblings, rt6i_siblings) {
 862                         sibling->rt6i_nsiblings++;
 863                         BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
 864                         rt6i_nsiblings++;
 865                 }
 866                 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
 867         }
 868
 869         /*
 870          *      insert node
 871          */
 872         if (!replace) {
 873                 if (!add)
 874                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
 875
 876 add:
 877                 nlflags |= NLM_F_CREATE;
 878                 err = fib6_commit_metrics(&rt->dst, mxc);
 879                 if (err)
 880                         return err;
 881
 882                 rt->dst.rt6_next = iter;
 883                 *ins = rt;
 884                 rcu_assign_pointer(rt->rt6i_node, fn);
 885                 atomic_inc(&rt->rt6i_ref);
 886                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
 887                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
 888
 889                 if (!(fn->fn_flags & RTN_RTINFO)) {
 890                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
 891                         fn->fn_flags |= RTN_RTINFO;
 892                 }
 893
 894         } else {
 895                 int nsiblings;
 896
 897                 if (!found) {
 898                         if (add)
 899                                 goto add;
 900                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
 901                         return -ENOENT;
 902                 }
 903
 904                 err = fib6_commit_metrics(&rt->dst, mxc);
 905                 if (err)
 906                         return err;
 907
 908                 *ins = rt;
 909                 rcu_assign_pointer(rt->rt6i_node, fn);
 910                 rt->dst.rt6_next = iter->dst.rt6_next;
 911                 atomic_inc(&rt->rt6i_ref);
 912                 inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
 913                 if (!(fn->fn_flags & RTN_RTINFO)) {
 914                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
 915                         fn->fn_flags |= RTN_RTINFO;
 916                 }
 917                 nsiblings = iter->rt6i_nsiblings;
 918                 fib6_purge_rt(iter, fn, info->nl_net);
 919                 if (fn->rr_ptr == iter)
 920                         fn->rr_ptr = NULL;
 921                 rt6_release(iter);
 922
 923                 if (nsiblings) {
 924                         /* Replacing an ECMP route, remove all siblings */
 925                         ins = &rt->dst.rt6_next;
 926                         iter = *ins;
 927                         while (iter) {
 928                                 if (iter->rt6i_metric > rt->rt6i_metric)
 929                                         break;
 930                                 if (rt6_qualify_for_ecmp(iter)) {
 931                                         *ins = iter->dst.rt6_next;
 932                                         fib6_purge_rt(iter, fn, info->nl_net);
 933                                         if (fn->rr_ptr == iter)
 934                                                 fn->rr_ptr = NULL;
 935                                         rt6_release(iter);
 936                                         nsiblings--;
 937                                 } else {
 938                                         ins = &iter->dst.rt6_next;
 939                                 }
 940                                 iter = *ins;
 941                         }
 942                         WARN_ON(nsiblings != 0);
 943                 }
 944         }
 945
 946         return 0;
 947 }
 948
 949 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
 950 {
 951         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
 952             (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
 953                 mod_timer(&net->ipv6.ip6_fib_timer,
 954                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
 955 }
 956
 957 void fib6_force_start_gc(struct net *net)
 958 {
 959         if (!timer_pending(&net->ipv6.ip6_fib_timer))
 960                 mod_timer(&net->ipv6.ip6_fib_timer,
 961                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
 962 }
 963
 964 /*
 965  *      Add routing information to the routing tree.
 966  *      <destination addr>/<source addr>
 967  *      with source addr info in sub-trees
 968  */
 969
 970 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
 971              struct nl_info *info, struct mx6_config *mxc)
 972 {
 973         struct fib6_node *fn, *pn = NULL;
 974         int err = -ENOMEM;
 975         int allow_create = 1;
 976         int replace_required = 0;
 977         int sernum = fib6_new_sernum(info->nl_net);
 978
 979         if (WARN_ON_ONCE((rt->dst.flags & DST_NOCACHE) &&
 980                          !atomic_read(&rt->dst.__refcnt)))
 981                 return -EINVAL;
 982
 983         if (info->nlh) {
 984                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
 985                         allow_create = 0;
 986                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
 987                         replace_required = 1;
 988         }
 989         if (!allow_create && !replace_required)
 990                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
 991
 992         fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
 993                         offsetof(struct rt6_info, rt6i_dst), allow_create,
 994                         replace_required, sernum);
 995         if (IS_ERR(fn)) {
 996                 err = PTR_ERR(fn);
 997                 fn = NULL;
 998                 goto out;
 999         }
1000
1001         pn = fn;
1002
1003 #ifdef CONFIG_IPV6_SUBTREES
1004         if (rt->rt6i_src.plen) {
1005                 struct fib6_node *sn;
1006
1007                 if (!fn->subtree) {
1008                         struct fib6_node *sfn;
1009
1010                         /*
1011                          * Create subtree.
1012                          *
1013                          *              fn[main tree]
1014                          *              |
1015                          *              sfn[subtree root]
1016                          *                 \
1017                          *                  sn[new leaf node]
1018                          */
1019
1020                         /* Create subtree root node */
1021                         sfn = node_alloc();
1022                         if (!sfn)
1023                                 goto failure;
1024
1025                         sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
1026                         atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1027                         sfn->fn_flags = RTN_ROOT;
1028                         sfn->fn_sernum = sernum;
1029
1030                         /* Now add the first leaf node to new subtree */
1031
1032                         sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
1033                                         rt->rt6i_src.plen,
1034                                         offsetof(struct rt6_info, rt6i_src),
1035                                         allow_create, replace_required, sernum);
1036
1037                         if (IS_ERR(sn)) {
1038                                 /* If it is failed, discard just allocated
1039                                    root, and then (in failure) stale node
1040                                    in main tree.
1041                                  */
1042                                 node_free_immediate(sfn);
1043                                 err = PTR_ERR(sn);
1044                                 goto failure;
1045                         }
1046
1047                         /* Now link new subtree to main tree */
1048                         sfn->parent = fn;
1049                         fn->subtree = sfn;
1050                 } else {
1051                         sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1052                                         rt->rt6i_src.plen,
1053                                         offsetof(struct rt6_info, rt6i_src),
1054                                         allow_create, replace_required, sernum);
1055
1056                         if (IS_ERR(sn)) {
1057                                 err = PTR_ERR(sn);
1058                                 goto failure;
1059                         }
1060                 }
1061
1062                 if (!fn->leaf) {
1063                         fn->leaf = rt;
1064                         atomic_inc(&rt->rt6i_ref);
1065                 }
1066                 fn = sn;
1067         }
1068 #endif
1069
1070         err = fib6_add_rt2node(fn, rt, info, mxc);
1071         if (!err) {
1072                 fib6_start_gc(info->nl_net, rt);
1073                 if (!(rt->rt6i_flags & RTF_CACHE))
1074                         fib6_prune_clones(info->nl_net, pn);
1075                 rt->dst.flags &= ~DST_NOCACHE;
1076         }
1077
1078 out:
1079         if (err) {
1080 #ifdef CONFIG_IPV6_SUBTREES
1081                 /*
1082                  * If fib6_add_1 has cleared the old leaf pointer in the
1083                  * super-tree leaf node we have to find a new one for it.
1084                  */
1085                 if (pn != fn && pn->leaf == rt) {
1086                         pn->leaf = NULL;
1087                         atomic_dec(&rt->rt6i_ref);
1088                 }
1089                 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1090                         pn->leaf = fib6_find_prefix(info->nl_net, pn);
1091 #if RT6_DEBUG >= 2
1092                         if (!pn->leaf) {
1093                                 WARN_ON(pn->leaf == NULL);
1094                                 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1095                         }
1096 #endif
1097                         atomic_inc(&pn->leaf->rt6i_ref);
1098                 }
1099 #endif
1100                 goto failure;
1101         }
1102         return err;
1103
1104 failure:
1105         /* fn->leaf could be NULL if fn is an intermediate node and we
1106          * failed to add the new route to it in both subtree creation
1107          * failure and fib6_add_rt2node() failure case.
1108          * In both cases, fib6_repair_tree() should be called to fix
1109          * fn->leaf.
1110          */
1111         if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1112                 fib6_repair_tree(info->nl_net, fn);
1113         if (!(rt->dst.flags & DST_NOCACHE))
1114                 dst_free(&rt->dst);
1115         return err;
1116 }
1117
1118 /*
1119  *      Routing tree lookup
1120  *
1121  */
1122
1123 struct lookup_args {
1124         int                     offset;         /* key offset on rt6_info       */
1125         const struct in6_addr   *addr;          /* search key                   */
1126 };
1127
1128 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1129                                        struct lookup_args *args)
1130 {
1131         struct fib6_node *fn;
1132         __be32 dir;
1133
1134         if (unlikely(args->offset == 0))
1135                 return NULL;
1136
1137         /*
1138          *      Descend on a tree
1139          */
1140
1141         fn = root;
1142
1143         for (;;) {
1144                 struct fib6_node *next;
1145
1146                 dir = addr_bit_set(args->addr, fn->fn_bit);
1147
1148                 next = dir ? fn->right : fn->left;
1149
1150                 if (next) {
1151                         fn = next;
1152                         continue;
1153                 }
1154                 break;
1155         }
1156
1157         while (fn) {
1158                 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1159                         struct rt6key *key;
1160
1161                         key = (struct rt6key *) ((u8 *) fn->leaf +
1162                                                  args->offset);
1163
1164                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1165 #ifdef CONFIG_IPV6_SUBTREES
1166                                 if (fn->subtree) {
1167                                         struct fib6_node *sfn;
1168                                         sfn = fib6_lookup_1(fn->subtree,
1169                                                             args + 1);
1170                                         if (!sfn)
1171                                                 goto backtrack;
1172                                         fn = sfn;
1173                                 }
1174 #endif
1175                                 if (fn->fn_flags & RTN_RTINFO)
1176                                         return fn;
1177                         }
1178                 }
1179 #ifdef CONFIG_IPV6_SUBTREES
1180 backtrack:
1181 #endif
1182                 if (fn->fn_flags & RTN_ROOT)
1183                         break;
1184
1185                 fn = fn->parent;
1186         }
1187
1188         return NULL;
1189 }
1190
1191 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1192                               const struct in6_addr *saddr)
1193 {
1194         struct fib6_node *fn;
1195         struct lookup_args args[] = {
1196                 {
1197                         .offset = offsetof(struct rt6_info, rt6i_dst),
1198                         .addr = daddr,
1199                 },
1200 #ifdef CONFIG_IPV6_SUBTREES
1201                 {
1202                         .offset = offsetof(struct rt6_info, rt6i_src),
1203                         .addr = saddr,
1204                 },
1205 #endif
1206                 {
1207                         .offset = 0,    /* sentinel */
1208                 }
1209         };
1210
1211         fn = fib6_lookup_1(root, daddr ? args : args + 1);
1212         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1213                 fn = root;
1214
1215         return fn;
1216 }
1217
1218 /*
1219  *      Get node with specified destination prefix (and source prefix,
1220  *      if subtrees are used)
1221  */
1222
1223
1224 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1225                                        const struct in6_addr *addr,
1226                                        int plen, int offset)
1227 {
1228         struct fib6_node *fn;
1229
1230         for (fn = root; fn ; ) {
1231                 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1232
1233                 /*
1234                  *      Prefix match
1235                  */
1236                 if (plen < fn->fn_bit ||
1237                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1238                         return NULL;
1239
1240                 if (plen == fn->fn_bit)
1241                         return fn;
1242
1243                 /*
1244                  *      We have more bits to go
1245                  */
1246                 if (addr_bit_set(addr, fn->fn_bit))
1247                         fn = fn->right;
1248                 else
1249                         fn = fn->left;
1250         }
1251         return NULL;
1252 }
1253
1254 struct fib6_node *fib6_locate(struct fib6_node *root,
1255                               const struct in6_addr *daddr, int dst_len,
1256                               const struct in6_addr *saddr, int src_len)
1257 {
1258         struct fib6_node *fn;
1259
1260         fn = fib6_locate_1(root, daddr, dst_len,
1261                            offsetof(struct rt6_info, rt6i_dst));
1262
1263 #ifdef CONFIG_IPV6_SUBTREES
1264         if (src_len) {
1265                 WARN_ON(saddr == NULL);
1266                 if (fn && fn->subtree)
1267                         fn = fib6_locate_1(fn->subtree, saddr, src_len,
1268                                            offsetof(struct rt6_info, rt6i_src));
1269         }
1270 #endif
1271
1272         if (fn && fn->fn_flags & RTN_RTINFO)
1273                 return fn;
1274
1275         return NULL;
1276 }
1277
1278
1279 /*
1280  *      Deletion
1281  *
1282  */
1283
1284 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1285 {
1286         if (fn->fn_flags & RTN_ROOT)
1287                 return net->ipv6.ip6_null_entry;
1288
1289         while (fn) {
1290                 if (fn->left)
1291                         return fn->left->leaf;
1292                 if (fn->right)
1293                         return fn->right->leaf;
1294
1295                 fn = FIB6_SUBTREE(fn);
1296         }
1297         return NULL;
1298 }
1299
1300 /*
1301  *      Called to trim the tree of intermediate nodes when possible. "fn"
1302  *      is the node we want to try and remove.
1303  */
1304
1305 static struct fib6_node *fib6_repair_tree(struct net *net,
1306                                            struct fib6_node *fn)
1307 {
1308         int children;
1309         int nstate;
1310         struct fib6_node *child, *pn;
1311         struct fib6_walker *w;
1312         int iter = 0;
1313
1314         for (;;) {
1315                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1316                 iter++;
1317
1318                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1319                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1320                 WARN_ON(fn->leaf);
1321
1322                 children = 0;
1323                 child = NULL;
1324                 if (fn->right)
1325                         child = fn->right, children |= 1;
1326                 if (fn->left)
1327                         child = fn->left, children |= 2;
1328
1329                 if (children == 3 || FIB6_SUBTREE(fn)
1330 #ifdef CONFIG_IPV6_SUBTREES
1331                     /* Subtree root (i.e. fn) may have one child */
1332                     || (children && fn->fn_flags & RTN_ROOT)
1333 #endif
1334                     ) {
1335                         fn->leaf = fib6_find_prefix(net, fn);
1336 #if RT6_DEBUG >= 2
1337                         if (!fn->leaf) {
1338                                 WARN_ON(!fn->leaf);
1339                                 fn->leaf = net->ipv6.ip6_null_entry;
1340                         }
1341 #endif
1342                         atomic_inc(&fn->leaf->rt6i_ref);
1343                         return fn->parent;
1344                 }
1345
1346                 pn = fn->parent;
1347 #ifdef CONFIG_IPV6_SUBTREES
1348                 if (FIB6_SUBTREE(pn) == fn) {
1349                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1350                         FIB6_SUBTREE(pn) = NULL;
1351                         nstate = FWS_L;
1352                 } else {
1353                         WARN_ON(fn->fn_flags & RTN_ROOT);
1354 #endif
1355                         if (pn->right == fn)
1356                                 pn->right = child;
1357                         else if (pn->left == fn)
1358                                 pn->left = child;
1359 #if RT6_DEBUG >= 2
1360                         else
1361                                 WARN_ON(1);
1362 #endif
1363                         if (child)
1364                                 child->parent = pn;
1365                         nstate = FWS_R;
1366 #ifdef CONFIG_IPV6_SUBTREES
1367                 }
1368 #endif
1369
1370                 read_lock(&net->ipv6.fib6_walker_lock);
1371                 FOR_WALKERS(net, w) {
1372                         if (!child) {
1373                                 if (w->root == fn) {
1374                                         w->root = w->node = NULL;
1375                                         RT6_TRACE("W %p adjusted by delroot 1\n", w);
1376                                 } else if (w->node == fn) {
1377                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1378                                         w->node = pn;
1379                                         w->state = nstate;
1380                                 }
1381                         } else {
1382                                 if (w->root == fn) {
1383                                         w->root = child;
1384                                         RT6_TRACE("W %p adjusted by delroot 2\n", w);
1385                                 }
1386                                 if (w->node == fn) {
1387                                         w->node = child;
1388                                         if (children&2) {
1389                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1390                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1391                                         } else {
1392                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1393                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1394                                         }
1395                                 }
1396                         }
1397                 }
1398                 read_unlock(&net->ipv6.fib6_walker_lock);
1399
1400                 node_free(fn);
1401                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1402                         return pn;
1403
1404                 rt6_release(pn->leaf);
1405                 pn->leaf = NULL;
1406                 fn = pn;
1407         }
1408 }
1409
1410 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1411                            struct nl_info *info)
1412 {
1413         struct fib6_walker *w;
1414         struct rt6_info *rt = *rtp;
1415         struct net *net = info->nl_net;
1416
1417         RT6_TRACE("fib6_del_route\n");
1418
1419         /* Unlink it */
1420         *rtp = rt->dst.rt6_next;
1421         rt->rt6i_node = NULL;
1422         net->ipv6.rt6_stats->fib_rt_entries--;
1423         net->ipv6.rt6_stats->fib_discarded_routes++;
1424
1425         /* Reset round-robin state, if necessary */
1426         if (fn->rr_ptr == rt)
1427                 fn->rr_ptr = NULL;
1428
1429         /* Remove this entry from other siblings */
1430         if (rt->rt6i_nsiblings) {
1431                 struct rt6_info *sibling, *next_sibling;
1432
1433                 list_for_each_entry_safe(sibling, next_sibling,
1434                                          &rt->rt6i_siblings, rt6i_siblings)
1435                         sibling->rt6i_nsiblings--;
1436                 rt->rt6i_nsiblings = 0;
1437                 list_del_init(&rt->rt6i_siblings);
1438         }
1439
1440         /* Adjust walkers */
1441         read_lock(&net->ipv6.fib6_walker_lock);
1442         FOR_WALKERS(net, w) {
1443                 if (w->state == FWS_C && w->leaf == rt) {
1444                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1445                         w->leaf = rt->dst.rt6_next;
1446                         if (!w->leaf)
1447                                 w->state = FWS_U;
1448                 }
1449         }
1450         read_unlock(&net->ipv6.fib6_walker_lock);
1451
1452         rt->dst.rt6_next = NULL;
1453
1454         /* If it was last route, expunge its radix tree node */
1455         if (!fn->leaf) {
1456                 fn->fn_flags &= ~RTN_RTINFO;
1457                 net->ipv6.rt6_stats->fib_route_nodes--;
1458                 fn = fib6_repair_tree(net, fn);
1459         }
1460
1461         fib6_purge_rt(rt, fn, net);
1462
1463         inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1464         rt6_release(rt);
1465 }
1466
1467 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1468 {
1469         struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1470                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
1471         struct net *net = info->nl_net;
1472         struct rt6_info **rtp;
1473
1474 #if RT6_DEBUG >= 2
1475         if (rt->dst.obsolete > 0) {
1476                 WARN_ON(fn);
1477                 return -ENOENT;
1478         }
1479 #endif
1480         if (!fn || rt == net->ipv6.ip6_null_entry)
1481                 return -ENOENT;
1482
1483         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1484
1485         if (!(rt->rt6i_flags & RTF_CACHE)) {
1486                 struct fib6_node *pn = fn;
1487 #ifdef CONFIG_IPV6_SUBTREES
1488                 /* clones of this route might be in another subtree */
1489                 if (rt->rt6i_src.plen) {
1490                         while (!(pn->fn_flags & RTN_ROOT))
1491                                 pn = pn->parent;
1492                         pn = pn->parent;
1493                 }
1494 #endif
1495                 fib6_prune_clones(info->nl_net, pn);
1496         }
1497
1498         /*
1499          *      Walk the leaf entries looking for ourself
1500          */
1501
1502         for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1503                 if (*rtp == rt) {
1504                         fib6_del_route(fn, rtp, info);
1505                         return 0;
1506                 }
1507         }
1508         return -ENOENT;
1509 }
1510
1511 /*
1512  *      Tree traversal function.
1513  *
1514  *      Certainly, it is not interrupt safe.
1515  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1516  *      It means, that we can modify tree during walking
1517  *      and use this function for garbage collection, clone pruning,
1518  *      cleaning tree when a device goes down etc. etc.
1519  *
1520  *      It guarantees that every node will be traversed,
1521  *      and that it will be traversed only once.
1522  *
1523  *      Callback function w->func may return:
1524  *      0 -> continue walking.
1525  *      positive value -> walking is suspended (used by tree dumps,
1526  *      and probably by gc, if it will be split to several slices)
1527  *      negative value -> terminate walking.
1528  *
1529  *      The function itself returns:
1530  *      0   -> walk is complete.
1531  *      >0  -> walk is incomplete (i.e. suspended)
1532  *      <0  -> walk is terminated by an error.
1533  */
1534
1535 static int fib6_walk_continue(struct fib6_walker *w)
1536 {
1537         struct fib6_node *fn, *pn;
1538
1539         for (;;) {
1540                 fn = w->node;
1541                 if (!fn)
1542                         return 0;
1543
1544                 if (w->prune && fn != w->root &&
1545                     fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1546                         w->state = FWS_C;
1547                         w->leaf = fn->leaf;
1548                 }
1549                 switch (w->state) {
1550 #ifdef CONFIG_IPV6_SUBTREES
1551                 case FWS_S:
1552                         if (FIB6_SUBTREE(fn)) {
1553                                 w->node = FIB6_SUBTREE(fn);
1554                                 continue;
1555                         }
1556                         w->state = FWS_L;
1557 #endif
1558                 case FWS_L:
1559                         if (fn->left) {
1560                                 w->node = fn->left;
1561                                 w->state = FWS_INIT;
1562                                 continue;
1563                         }
1564                         w->state = FWS_R;
1565                 case FWS_R:
1566                         if (fn->right) {
1567                                 w->node = fn->right;
1568                                 w->state = FWS_INIT;
1569                                 continue;
1570                         }
1571                         w->state = FWS_C;
1572                         w->leaf = fn->leaf;
1573                 case FWS_C:
1574                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1575                                 int err;
1576
1577                                 if (w->skip) {
1578                                         w->skip--;
1579                                         goto skip;
1580                                 }
1581
1582                                 err = w->func(w);
1583                                 if (err)
1584                                         return err;
1585
1586                                 w->count++;
1587                                 continue;
1588                         }
1589 skip:
1590                         w->state = FWS_U;
1591                 case FWS_U:
1592                         if (fn == w->root)
1593                                 return 0;
1594                         pn = fn->parent;
1595                         w->node = pn;
1596 #ifdef CONFIG_IPV6_SUBTREES
1597                         if (FIB6_SUBTREE(pn) == fn) {
1598                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1599                                 w->state = FWS_L;
1600                                 continue;
1601                         }
1602 #endif
1603                         if (pn->left == fn) {
1604                                 w->state = FWS_R;
1605                                 continue;
1606                         }
1607                         if (pn->right == fn) {
1608                                 w->state = FWS_C;
1609                                 w->leaf = w->node->leaf;
1610                                 continue;
1611                         }
1612 #if RT6_DEBUG >= 2
1613                         WARN_ON(1);
1614 #endif
1615                 }
1616         }
1617 }
1618
1619 static int fib6_walk(struct net *net, struct fib6_walker *w)
1620 {
1621         int res;
1622
1623         w->state = FWS_INIT;
1624         w->node = w->root;
1625
1626         fib6_walker_link(net, w);
1627         res = fib6_walk_continue(w);
1628         if (res <= 0)
1629                 fib6_walker_unlink(net, w);
1630         return res;
1631 }
1632
1633 static int fib6_clean_node(struct fib6_walker *w)
1634 {
1635         int res;
1636         struct rt6_info *rt;
1637         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1638         struct nl_info info = {
1639                 .nl_net = c->net,
1640         };
1641
1642         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1643             w->node->fn_sernum != c->sernum)
1644                 w->node->fn_sernum = c->sernum;
1645
1646         if (!c->func) {
1647                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1648                 w->leaf = NULL;
1649                 return 0;
1650         }
1651
1652         for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1653                 res = c->func(rt, c->arg);
1654                 if (res < 0) {
1655                         w->leaf = rt;
1656                         res = fib6_del(rt, &info);
1657                         if (res) {
1658 #if RT6_DEBUG >= 2
1659                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1660                                          __func__, rt,
1661                                          rcu_access_pointer(rt->rt6i_node),
1662                                          res);
1663 #endif
1664                                 continue;
1665                         }
1666                         return 0;
1667                 }
1668                 WARN_ON(res != 0);
1669         }
1670         w->leaf = rt;
1671         return 0;
1672 }
1673
1674 /*
1675  *      Convenient frontend to tree walker.
1676  *
1677  *      func is called on each route.
1678  *              It may return -1 -> delete this route.
1679  *                            0  -> continue walking
1680  *
1681  *      prune==1 -> only immediate children of node (certainly,
1682  *      ignoring pure split nodes) will be scanned.
1683  */
1684
1685 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1686                             int (*func)(struct rt6_info *, void *arg),
1687                             bool prune, int sernum, void *arg)
1688 {
1689         struct fib6_cleaner c;
1690
1691         c.w.root = root;
1692         c.w.func = fib6_clean_node;
1693         c.w.prune = prune;
1694         c.w.count = 0;
1695         c.w.skip = 0;
1696         c.func = func;
1697         c.sernum = sernum;
1698         c.arg = arg;
1699         c.net = net;
1700
1701         fib6_walk(net, &c.w);
1702 }
1703
1704 static void __fib6_clean_all(struct net *net,
1705                              int (*func)(struct rt6_info *, void *),
1706                              int sernum, void *arg)
1707 {
1708         struct fib6_table *table;
1709         struct hlist_head *head;
1710         unsigned int h;
1711
1712         rcu_read_lock();
1713         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1714                 head = &net->ipv6.fib_table_hash[h];
1715                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1716                         write_lock_bh(&table->tb6_lock);
1717                         fib6_clean_tree(net, &table->tb6_root,
1718                                         func, false, sernum, arg);
1719                         write_unlock_bh(&table->tb6_lock);
1720                 }
1721         }
1722         rcu_read_unlock();
1723 }
1724
1725 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1726                     void *arg)
1727 {
1728         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1729 }
1730
1731 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1732 {
1733         if (rt->rt6i_flags & RTF_CACHE) {
1734                 RT6_TRACE("pruning clone %p\n", rt);
1735                 return -1;
1736         }
1737
1738         return 0;
1739 }
1740
1741 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1742 {
1743         fib6_clean_tree(net, fn, fib6_prune_clone, true,
1744                         FIB6_NO_SERNUM_CHANGE, NULL);
1745 }
1746
1747 static void fib6_flush_trees(struct net *net)
1748 {
1749         int new_sernum = fib6_new_sernum(net);
1750
1751         __fib6_clean_all(net, NULL, new_sernum, NULL);
1752 }
1753
1754 /*
1755  *      Garbage collection
1756  */
1757
1758 struct fib6_gc_args
1759 {
1760         int                     timeout;
1761         int                     more;
1762 };
1763
1764 static int fib6_age(struct rt6_info *rt, void *arg)
1765 {
1766         struct fib6_gc_args *gc_args = arg;
1767         unsigned long now = jiffies;
1768
1769         /*
1770          *      check addrconf expiration here.
1771          *      Routes are expired even if they are in use.
1772          *
1773          *      Also age clones. Note, that clones are aged out
1774          *      only if they are not in use now.
1775          */
1776
1777         if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1778                 if (time_after(now, rt->dst.expires)) {
1779                         RT6_TRACE("expiring %p\n", rt);
1780                         return -1;
1781                 }
1782                 gc_args->more++;
1783         } else if (rt->rt6i_flags & RTF_CACHE) {
1784                 if (atomic_read(&rt->dst.__refcnt) == 0 &&
1785                     time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
1786                         RT6_TRACE("aging clone %p\n", rt);
1787                         return -1;
1788                 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1789                         struct neighbour *neigh;
1790                         __u8 neigh_flags = 0;
1791
1792                         neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1793                         if (neigh) {
1794                                 neigh_flags = neigh->flags;
1795                                 neigh_release(neigh);
1796                         }
1797                         if (!(neigh_flags & NTF_ROUTER)) {
1798                                 RT6_TRACE("purging route %p via non-router but gateway\n",
1799                                           rt);
1800                                 return -1;
1801                         }
1802                 }
1803                 gc_args->more++;
1804         }
1805
1806         return 0;
1807 }
1808
1809 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1810 {
1811         struct fib6_gc_args gc_args;
1812         unsigned long now;
1813
1814         if (force) {
1815                 spin_lock_bh(&net->ipv6.fib6_gc_lock);
1816         } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
1817                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1818                 return;
1819         }
1820         gc_args.timeout = expires ? (int)expires :
1821                           net->ipv6.sysctl.ip6_rt_gc_interval;
1822
1823         gc_args.more = icmp6_dst_gc();
1824
1825         fib6_clean_all(net, fib6_age, &gc_args);
1826         now = jiffies;
1827         net->ipv6.ip6_rt_last_gc = now;
1828
1829         if (gc_args.more)
1830                 mod_timer(&net->ipv6.ip6_fib_timer,
1831                           round_jiffies(now
1832                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
1833         else
1834                 del_timer(&net->ipv6.ip6_fib_timer);
1835         spin_unlock_bh(&net->ipv6.fib6_gc_lock);
1836 }
1837
1838 static void fib6_gc_timer_cb(unsigned long arg)
1839 {
1840         fib6_run_gc(0, (struct net *)arg, true);
1841 }
1842
1843 static int __net_init fib6_net_init(struct net *net)
1844 {
1845         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1846
1847         spin_lock_init(&net->ipv6.fib6_gc_lock);
1848         rwlock_init(&net->ipv6.fib6_walker_lock);
1849         INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
1850         setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1851
1852         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1853         if (!net->ipv6.rt6_stats)
1854                 goto out_timer;
1855
1856         /* Avoid false sharing : Use at least a full cache line */
1857         size = max_t(size_t, size, L1_CACHE_BYTES);
1858
1859         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1860         if (!net->ipv6.fib_table_hash)
1861                 goto out_rt6_stats;
1862
1863         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1864                                           GFP_KERNEL);
1865         if (!net->ipv6.fib6_main_tbl)
1866                 goto out_fib_table_hash;
1867
1868         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1869         net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1870         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1871                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1872         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
1873
1874 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1875         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
1876                                            GFP_KERNEL);
1877         if (!net->ipv6.fib6_local_tbl)
1878                 goto out_fib6_main_tbl;
1879         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
1880         net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1881         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
1882                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1883         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
1884 #endif
1885         fib6_tables_init(net);
1886
1887         return 0;
1888
1889 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
1890 out_fib6_main_tbl:
1891         kfree(net->ipv6.fib6_main_tbl);
1892 #endif
1893 out_fib_table_hash:
1894         kfree(net->ipv6.fib_table_hash);
1895 out_rt6_stats:
1896         kfree(net->ipv6.rt6_stats);
1897 out_timer:
1898         return -ENOMEM;
1899 }
1900
1901 static void fib6_net_exit(struct net *net)
1902 {
1903         unsigned int i;
1904
1905         rt6_ifdown(net, NULL);
1906         del_timer_sync(&net->ipv6.ip6_fib_timer);
1907
1908         for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
1909                 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
1910                 struct hlist_node *tmp;
1911                 struct fib6_table *tb;
1912
1913                 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
1914                         hlist_del(&tb->tb6_hlist);
1915                         fib6_free_table(tb);
1916                 }
1917         }
1918
1919         kfree(net->ipv6.fib_table_hash);
1920         kfree(net->ipv6.rt6_stats);
1921 }
1922
1923 static struct pernet_operations fib6_net_ops = {
1924         .init = fib6_net_init,
1925         .exit = fib6_net_exit,
1926 };
1927
1928 int __init fib6_init(void)
1929 {
1930         int ret = -ENOMEM;
1931
1932         fib6_node_kmem = kmem_cache_create("fib6_nodes",
1933                                            sizeof(struct fib6_node),
1934                                            0, SLAB_HWCACHE_ALIGN,
1935                                            NULL);
1936         if (!fib6_node_kmem)
1937                 goto out;
1938
1939         ret = register_pernet_subsys(&fib6_net_ops);
1940         if (ret)
1941                 goto out_kmem_cache_create;
1942
1943         ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
1944                               NULL);
1945         if (ret)
1946                 goto out_unregister_subsys;
1947
1948         __fib6_flush_trees = fib6_flush_trees;
1949 out:
1950         return ret;
1951
1952 out_unregister_subsys:
1953         unregister_pernet_subsys(&fib6_net_ops);
1954 out_kmem_cache_create:
1955         kmem_cache_destroy(fib6_node_kmem);
1956         goto out;
1957 }
1958
1959 void fib6_gc_cleanup(void)
1960 {
1961         unregister_pernet_subsys(&fib6_net_ops);
1962         kmem_cache_destroy(fib6_node_kmem);
1963 }
1964
1965 #ifdef CONFIG_PROC_FS
1966
1967 struct ipv6_route_iter {
1968         struct seq_net_private p;
1969         struct fib6_walker w;
1970         loff_t skip;
1971         struct fib6_table *tbl;
1972         int sernum;
1973 };
1974
1975 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
1976 {
1977         struct rt6_info *rt = v;
1978         struct ipv6_route_iter *iter = seq->private;
1979
1980         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
1981
1982 #ifdef CONFIG_IPV6_SUBTREES
1983         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
1984 #else
1985         seq_puts(seq, "00000000000000000000000000000000 00 ");
1986 #endif
1987         if (rt->rt6i_flags & RTF_GATEWAY)
1988                 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
1989         else
1990                 seq_puts(seq, "00000000000000000000000000000000");
1991
1992         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
1993                    rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
1994                    rt->dst.__use, rt->rt6i_flags,
1995                    rt->dst.dev ? rt->dst.dev->name : "");
1996         iter->w.leaf = NULL;
1997         return 0;
1998 }
1999
2000 static int ipv6_route_yield(struct fib6_walker *w)
2001 {
2002         struct ipv6_route_iter *iter = w->args;
2003
2004         if (!iter->skip)
2005                 return 1;
2006
2007         do {
2008                 iter->w.leaf = iter->w.leaf->dst.rt6_next;
2009                 iter->skip--;
2010                 if (!iter->skip && iter->w.leaf)
2011                         return 1;
2012         } while (iter->w.leaf);
2013
2014         return 0;
2015 }
2016
2017 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2018                                       struct net *net)
2019 {
2020         memset(&iter->w, 0, sizeof(iter->w));
2021         iter->w.func = ipv6_route_yield;
2022         iter->w.root = &iter->tbl->tb6_root;
2023         iter->w.state = FWS_INIT;
2024         iter->w.node = iter->w.root;
2025         iter->w.args = iter;
2026         iter->sernum = iter->w.root->fn_sernum;
2027         INIT_LIST_HEAD(&iter->w.lh);
2028         fib6_walker_link(net, &iter->w);
2029 }
2030
2031 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2032                                                     struct net *net)
2033 {
2034         unsigned int h;
2035         struct hlist_node *node;
2036
2037         if (tbl) {
2038                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2039                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2040         } else {
2041                 h = 0;
2042                 node = NULL;
2043         }
2044
2045         while (!node && h < FIB6_TABLE_HASHSZ) {
2046                 node = rcu_dereference_bh(
2047                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2048         }
2049         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2050 }
2051
2052 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2053 {
2054         if (iter->sernum != iter->w.root->fn_sernum) {
2055                 iter->sernum = iter->w.root->fn_sernum;
2056                 iter->w.state = FWS_INIT;
2057                 iter->w.node = iter->w.root;
2058                 WARN_ON(iter->w.skip);
2059                 iter->w.skip = iter->w.count;
2060         }
2061 }
2062
2063 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2064 {
2065         int r;
2066         struct rt6_info *n;
2067         struct net *net = seq_file_net(seq);
2068         struct ipv6_route_iter *iter = seq->private;
2069
2070         if (!v)
2071                 goto iter_table;
2072
2073         n = ((struct rt6_info *)v)->dst.rt6_next;
2074         if (n) {
2075                 ++*pos;
2076                 return n;
2077         }
2078
2079 iter_table:
2080         ipv6_route_check_sernum(iter);
2081         read_lock(&iter->tbl->tb6_lock);
2082         r = fib6_walk_continue(&iter->w);
2083         read_unlock(&iter->tbl->tb6_lock);
2084         if (r > 0) {
2085                 if (v)
2086                         ++*pos;
2087                 return iter->w.leaf;
2088         } else if (r < 0) {
2089                 fib6_walker_unlink(net, &iter->w);
2090                 return NULL;
2091         }
2092         fib6_walker_unlink(net, &iter->w);
2093
2094         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2095         if (!iter->tbl)
2096                 return NULL;
2097
2098         ipv6_route_seq_setup_walk(iter, net);
2099         goto iter_table;
2100 }
2101
2102 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2103         __acquires(RCU_BH)
2104 {
2105         struct net *net = seq_file_net(seq);
2106         struct ipv6_route_iter *iter = seq->private;
2107
2108         rcu_read_lock_bh();
2109         iter->tbl = ipv6_route_seq_next_table(NULL, net);
2110         iter->skip = *pos;
2111
2112         if (iter->tbl) {
2113                 ipv6_route_seq_setup_walk(iter, net);
2114                 return ipv6_route_seq_next(seq, NULL, pos);
2115         } else {
2116                 return NULL;
2117         }
2118 }
2119
2120 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2121 {
2122         struct fib6_walker *w = &iter->w;
2123         return w->node && !(w->state == FWS_U && w->node == w->root);
2124 }
2125
2126 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2127         __releases(RCU_BH)
2128 {
2129         struct net *net = seq_file_net(seq);
2130         struct ipv6_route_iter *iter = seq->private;
2131
2132         if (ipv6_route_iter_active(iter))
2133                 fib6_walker_unlink(net, &iter->w);
2134
2135         rcu_read_unlock_bh();
2136 }
2137
2138 static const struct seq_operations ipv6_route_seq_ops = {
2139         .start  = ipv6_route_seq_start,
2140         .next   = ipv6_route_seq_next,
2141         .stop   = ipv6_route_seq_stop,
2142         .show   = ipv6_route_seq_show
2143 };
2144
2145 int ipv6_route_open(struct inode *inode, struct file *file)
2146 {
2147         return seq_open_net(inode, file, &ipv6_route_seq_ops,
2148                             sizeof(struct ipv6_route_iter));
2149 }
2150
2151 #endif /* CONFIG_PROC_FS */