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 #include <net/fib_notifier.h>
  37
  38 #include <net/ip6_fib.h>
  39 #include <net/ip6_route.h>
  40
  41 #define RT6_DEBUG 2
  42
  43 #if RT6_DEBUG >= 3
  44 #define RT6_TRACE(x...) pr_debug(x)
  45 #else
  46 #define RT6_TRACE(x...) do { ; } while (0)
  47 #endif
  48
  49 static struct kmem_cache *fib6_node_kmem __read_mostly;
  50
  51 struct fib6_cleaner {
  52         struct fib6_walker w;
  53         struct net *net;
  54         int (*func)(struct rt6_info *, void *arg);
  55         int sernum;
  56         void *arg;
  57 };
  58
  59 #ifdef CONFIG_IPV6_SUBTREES
  60 #define FWS_INIT FWS_S
  61 #else
  62 #define FWS_INIT FWS_L
  63 #endif
  64
  65 static void fib6_prune_clones(struct net *net, struct fib6_node *fn);
  66 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
  67 static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
  68 static int fib6_walk(struct net *net, struct fib6_walker *w);
  69 static int fib6_walk_continue(struct fib6_walker *w);
  70
  71 /*
  72  *      A routing update causes an increase of the serial number on the
  73  *      affected subtree. This allows for cached routes to be asynchronously
  74  *      tested when modifications are made to the destination cache as a
  75  *      result of redirects, path MTU changes, etc.
  76  */
  77
  78 static void fib6_gc_timer_cb(unsigned long arg);
  79
  80 #define FOR_WALKERS(net, w) \
  81         list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
  82
  83 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
  84 {
  85         write_lock_bh(&net->ipv6.fib6_walker_lock);
  86         list_add(&w->lh, &net->ipv6.fib6_walkers);
  87         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  88 }
  89
  90 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
  91 {
  92         write_lock_bh(&net->ipv6.fib6_walker_lock);
  93         list_del(&w->lh);
  94         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  95 }
  96
  97 static int fib6_new_sernum(struct net *net)
  98 {
  99         int new, old;
 100
 101         do {
 102                 old = atomic_read(&net->ipv6.fib6_sernum);
 103                 new = old < INT_MAX ? old + 1 : 1;
 104         } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
 105                                 old, new) != old);
 106         return new;
 107 }
 108
 109 enum {
 110         FIB6_NO_SERNUM_CHANGE = 0,
 111 };
 112
 113 /*
 114  *      Auxiliary address test functions for the radix tree.
 115  *
 116  *      These assume a 32bit processor (although it will work on
 117  *      64bit processors)
 118  */
 119
 120 /*
 121  *      test bit
 122  */
 123 #if defined(__LITTLE_ENDIAN)
 124 # define BITOP_BE32_SWIZZLE     (0x1F & ~7)
 125 #else
 126 # define BITOP_BE32_SWIZZLE     0
 127 #endif
 128
 129 static __be32 addr_bit_set(const void *token, int fn_bit)
 130 {
 131         const __be32 *addr = token;
 132         /*
 133          * Here,
 134          *      1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
 135          * is optimized version of
 136          *      htonl(1 << ((~fn_bit)&0x1F))
 137          * See include/asm-generic/bitops/le.h.
 138          */
 139         return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
 140                addr[fn_bit >> 5];
 141 }
 142
 143 static struct fib6_node *node_alloc(void)
 144 {
 145         struct fib6_node *fn;
 146
 147         fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
 148
 149         return fn;
 150 }
 151
 152 static void node_free_immediate(struct fib6_node *fn)
 153 {
 154         kmem_cache_free(fib6_node_kmem, fn);
 155 }
 156
 157 static void node_free_rcu(struct rcu_head *head)
 158 {
 159         struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
 160
 161         kmem_cache_free(fib6_node_kmem, fn);
 162 }
 163
 164 static void node_free(struct fib6_node *fn)
 165 {
 166         call_rcu(&fn->rcu, node_free_rcu);
 167 }
 168
 169 void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
 170 {
 171         int cpu;
 172
 173         if (!non_pcpu_rt->rt6i_pcpu)
 174                 return;
 175
 176         for_each_possible_cpu(cpu) {
 177                 struct rt6_info **ppcpu_rt;
 178                 struct rt6_info *pcpu_rt;
 179
 180                 ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
 181                 pcpu_rt = *ppcpu_rt;
 182                 if (pcpu_rt) {
 183                         dst_dev_put(&pcpu_rt->dst);
 184                         dst_release(&pcpu_rt->dst);
 185                         *ppcpu_rt = NULL;
 186                 }
 187         }
 188
 189         free_percpu(non_pcpu_rt->rt6i_pcpu);
 190         non_pcpu_rt->rt6i_pcpu = NULL;
 191 }
 192 EXPORT_SYMBOL_GPL(rt6_free_pcpu);
 193
 194 static void fib6_free_table(struct fib6_table *table)
 195 {
 196         inetpeer_invalidate_tree(&table->tb6_peers);
 197         kfree(table);
 198 }
 199
 200 static void fib6_link_table(struct net *net, struct fib6_table *tb)
 201 {
 202         unsigned int h;
 203
 204         /*
 205          * Initialize table lock at a single place to give lockdep a key,
 206          * tables aren't visible prior to being linked to the list.
 207          */
 208         rwlock_init(&tb->tb6_lock);
 209
 210         h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
 211
 212         /*
 213          * No protection necessary, this is the only list mutatation
 214          * operation, tables never disappear once they exist.
 215          */
 216         hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
 217 }
 218
 219 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 220
 221 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
 222 {
 223         struct fib6_table *table;
 224
 225         table = kzalloc(sizeof(*table), GFP_ATOMIC);
 226         if (table) {
 227                 table->tb6_id = id;
 228                 table->tb6_root.leaf = net->ipv6.ip6_null_entry;
 229                 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
 230                 inet_peer_base_init(&table->tb6_peers);
 231         }
 232
 233         return table;
 234 }
 235
 236 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 237 {
 238         struct fib6_table *tb;
 239
 240         if (id == 0)
 241                 id = RT6_TABLE_MAIN;
 242         tb = fib6_get_table(net, id);
 243         if (tb)
 244                 return tb;
 245
 246         tb = fib6_alloc_table(net, id);
 247         if (tb)
 248                 fib6_link_table(net, tb);
 249
 250         return tb;
 251 }
 252 EXPORT_SYMBOL_GPL(fib6_new_table);
 253
 254 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 255 {
 256         struct fib6_table *tb;
 257         struct hlist_head *head;
 258         unsigned int h;
 259
 260         if (id == 0)
 261                 id = RT6_TABLE_MAIN;
 262         h = id & (FIB6_TABLE_HASHSZ - 1);
 263         rcu_read_lock();
 264         head = &net->ipv6.fib_table_hash[h];
 265         hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 266                 if (tb->tb6_id == id) {
 267                         rcu_read_unlock();
 268                         return tb;
 269                 }
 270         }
 271         rcu_read_unlock();
 272
 273         return NULL;
 274 }
 275 EXPORT_SYMBOL_GPL(fib6_get_table);
 276
 277 static void __net_init fib6_tables_init(struct net *net)
 278 {
 279         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 280         fib6_link_table(net, net->ipv6.fib6_local_tbl);
 281 }
 282 #else
 283
 284 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 285 {
 286         return fib6_get_table(net, id);
 287 }
 288
 289 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 290 {
 291           return net->ipv6.fib6_main_tbl;
 292 }
 293
 294 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 295                                    int flags, pol_lookup_t lookup)
 296 {
 297         struct rt6_info *rt;
 298
 299         rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
 300         if (rt->dst.error == -EAGAIN) {
 301                 ip6_rt_put(rt);
 302                 rt = net->ipv6.ip6_null_entry;
 303                 dst_hold(&rt->dst);
 304         }
 305
 306         return &rt->dst;
 307 }
 308
 309 static void __net_init fib6_tables_init(struct net *net)
 310 {
 311         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 312 }
 313
 314 #endif
 315
 316 unsigned int fib6_tables_seq_read(struct net *net)
 317 {
 318         unsigned int h, fib_seq = 0;
 319
 320         rcu_read_lock();
 321         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 322                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 323                 struct fib6_table *tb;
 324
 325                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 326                         read_lock_bh(&tb->tb6_lock);
 327                         fib_seq += tb->fib_seq;
 328                         read_unlock_bh(&tb->tb6_lock);
 329                 }
 330         }
 331         rcu_read_unlock();
 332
 333         return fib_seq;
 334 }
 335
 336 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
 337                                     enum fib_event_type event_type,
 338                                     struct rt6_info *rt)
 339 {
 340         struct fib6_entry_notifier_info info = {
 341                 .rt = rt,
 342         };
 343
 344         return call_fib6_notifier(nb, net, event_type, &info.info);
 345 }
 346
 347 static int call_fib6_entry_notifiers(struct net *net,
 348                                      enum fib_event_type event_type,
 349                                      struct rt6_info *rt)
 350 {
 351         struct fib6_entry_notifier_info info = {
 352                 .rt = rt,
 353         };
 354
 355         rt->rt6i_table->fib_seq++;
 356         return call_fib6_notifiers(net, event_type, &info.info);
 357 }
 358
 359 struct fib6_dump_arg {
 360         struct net *net;
 361         struct notifier_block *nb;
 362 };
 363
 364 static void fib6_rt_dump(struct rt6_info *rt, struct fib6_dump_arg *arg)
 365 {
 366         if (rt == arg->net->ipv6.ip6_null_entry)
 367                 return;
 368         call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
 369 }
 370
 371 static int fib6_node_dump(struct fib6_walker *w)
 372 {
 373         struct rt6_info *rt;
 374
 375         for (rt = w->leaf; rt; rt = rt->dst.rt6_next)
 376                 fib6_rt_dump(rt, w->args);
 377         w->leaf = NULL;
 378         return 0;
 379 }
 380
 381 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
 382                             struct fib6_walker *w)
 383 {
 384         w->root = &tb->tb6_root;
 385         read_lock_bh(&tb->tb6_lock);
 386         fib6_walk(net, w);
 387         read_unlock_bh(&tb->tb6_lock);
 388 }
 389
 390 /* Called with rcu_read_lock() */
 391 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
 392 {
 393         struct fib6_dump_arg arg;
 394         struct fib6_walker *w;
 395         unsigned int h;
 396
 397         w = kzalloc(sizeof(*w), GFP_ATOMIC);
 398         if (!w)
 399                 return -ENOMEM;
 400
 401         w->func = fib6_node_dump;
 402         arg.net = net;
 403         arg.nb = nb;
 404         w->args = &arg;
 405
 406         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 407                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 408                 struct fib6_table *tb;
 409
 410                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 411                         fib6_table_dump(net, tb, w);
 412         }
 413
 414         kfree(w);
 415
 416         return 0;
 417 }
 418
 419 static int fib6_dump_node(struct fib6_walker *w)
 420 {
 421         int res;
 422         struct rt6_info *rt;
 423
 424         for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
 425                 res = rt6_dump_route(rt, w->args);
 426                 if (res < 0) {
 427                         /* Frame is full, suspend walking */
 428                         w->leaf = rt;
 429                         return 1;
 430                 }
 431
 432                 /* Multipath routes are dumped in one route with the
 433                  * RTA_MULTIPATH attribute. Jump 'rt' to point to the
 434                  * last sibling of this route (no need to dump the
 435                  * sibling routes again)
 436                  */
 437                 if (rt->rt6i_nsiblings)
 438                         rt = list_last_entry(&rt->rt6i_siblings,
 439                                              struct rt6_info,
 440                                              rt6i_siblings);
 441         }
 442         w->leaf = NULL;
 443         return 0;
 444 }
 445
 446 static void fib6_dump_end(struct netlink_callback *cb)
 447 {
 448         struct net *net = sock_net(cb->skb->sk);
 449         struct fib6_walker *w = (void *)cb->args[2];
 450
 451         if (w) {
 452                 if (cb->args[4]) {
 453                         cb->args[4] = 0;
 454                         fib6_walker_unlink(net, w);
 455                 }
 456                 cb->args[2] = 0;
 457                 kfree(w);
 458         }
 459         cb->done = (void *)cb->args[3];
 460         cb->args[1] = 3;
 461 }
 462
 463 static int fib6_dump_done(struct netlink_callback *cb)
 464 {
 465         fib6_dump_end(cb);
 466         return cb->done ? cb->done(cb) : 0;
 467 }
 468
 469 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
 470                            struct netlink_callback *cb)
 471 {
 472         struct net *net = sock_net(skb->sk);
 473         struct fib6_walker *w;
 474         int res;
 475
 476         w = (void *)cb->args[2];
 477         w->root = &table->tb6_root;
 478
 479         if (cb->args[4] == 0) {
 480                 w->count = 0;
 481                 w->skip = 0;
 482
 483                 read_lock_bh(&table->tb6_lock);
 484                 res = fib6_walk(net, w);
 485                 read_unlock_bh(&table->tb6_lock);
 486                 if (res > 0) {
 487                         cb->args[4] = 1;
 488                         cb->args[5] = w->root->fn_sernum;
 489                 }
 490         } else {
 491                 if (cb->args[5] != w->root->fn_sernum) {
 492                         /* Begin at the root if the tree changed */
 493                         cb->args[5] = w->root->fn_sernum;
 494                         w->state = FWS_INIT;
 495                         w->node = w->root;
 496                         w->skip = w->count;
 497                 } else
 498                         w->skip = 0;
 499
 500                 read_lock_bh(&table->tb6_lock);
 501                 res = fib6_walk_continue(w);
 502                 read_unlock_bh(&table->tb6_lock);
 503                 if (res <= 0) {
 504                         fib6_walker_unlink(net, w);
 505                         cb->args[4] = 0;
 506                 }
 507         }
 508
 509         return res;
 510 }
 511
 512 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 513 {
 514         struct net *net = sock_net(skb->sk);
 515         unsigned int h, s_h;
 516         unsigned int e = 0, s_e;
 517         struct rt6_rtnl_dump_arg arg;
 518         struct fib6_walker *w;
 519         struct fib6_table *tb;
 520         struct hlist_head *head;
 521         int res = 0;
 522
 523         s_h = cb->args[0];
 524         s_e = cb->args[1];
 525
 526         w = (void *)cb->args[2];
 527         if (!w) {
 528                 /* New dump:
 529                  *
 530                  * 1. hook callback destructor.
 531                  */
 532                 cb->args[3] = (long)cb->done;
 533                 cb->done = fib6_dump_done;
 534
 535                 /*
 536                  * 2. allocate and initialize walker.
 537                  */
 538                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
 539                 if (!w)
 540                         return -ENOMEM;
 541                 w->func = fib6_dump_node;
 542                 cb->args[2] = (long)w;
 543         }
 544
 545         arg.skb = skb;
 546         arg.cb = cb;
 547         arg.net = net;
 548         w->args = &arg;
 549
 550         rcu_read_lock();
 551         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
 552                 e = 0;
 553                 head = &net->ipv6.fib_table_hash[h];
 554                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 555                         if (e < s_e)
 556                                 goto next;
 557                         res = fib6_dump_table(tb, skb, cb);
 558                         if (res != 0)
 559                                 goto out;
 560 next:
 561                         e++;
 562                 }
 563         }
 564 out:
 565         rcu_read_unlock();
 566         cb->args[1] = e;
 567         cb->args[0] = h;
 568
 569         res = res < 0 ? res : skb->len;
 570         if (res <= 0)
 571                 fib6_dump_end(cb);
 572         return res;
 573 }
 574
 575 /*
 576  *      Routing Table
 577  *
 578  *      return the appropriate node for a routing tree "add" operation
 579  *      by either creating and inserting or by returning an existing
 580  *      node.
 581  */
 582
 583 static struct fib6_node *fib6_add_1(struct fib6_node *root,
 584                                      struct in6_addr *addr, int plen,
 585                                      int offset, int allow_create,
 586                                      int replace_required, int sernum,
 587                                      struct netlink_ext_ack *extack)
 588 {
 589         struct fib6_node *fn, *in, *ln;
 590         struct fib6_node *pn = NULL;
 591         struct rt6key *key;
 592         int     bit;
 593         __be32  dir = 0;
 594
 595         RT6_TRACE("fib6_add_1\n");
 596
 597         /* insert node in tree */
 598
 599         fn = root;
 600
 601         do {
 602                 key = (struct rt6key *)((u8 *)fn->leaf + offset);
 603
 604                 /*
 605                  *      Prefix match
 606                  */
 607                 if (plen < fn->fn_bit ||
 608                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
 609                         if (!allow_create) {
 610                                 if (replace_required) {
 611                                         NL_SET_ERR_MSG(extack,
 612                                                        "Can not replace route - no match found");
 613                                         pr_warn("Can't replace route, no match found\n");
 614                                         return ERR_PTR(-ENOENT);
 615                                 }
 616                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 617                         }
 618                         goto insert_above;
 619                 }
 620
 621                 /*
 622                  *      Exact match ?
 623                  */
 624
 625                 if (plen == fn->fn_bit) {
 626                         /* clean up an intermediate node */
 627                         if (!(fn->fn_flags & RTN_RTINFO)) {
 628                                 rt6_release(fn->leaf);
 629                                 fn->leaf = NULL;
 630                         }
 631
 632                         fn->fn_sernum = sernum;
 633
 634                         return fn;
 635                 }
 636
 637                 /*
 638                  *      We have more bits to go
 639                  */
 640
 641                 /* Try to walk down on tree. */
 642                 fn->fn_sernum = sernum;
 643                 dir = addr_bit_set(addr, fn->fn_bit);
 644                 pn = fn;
 645                 fn = dir ? fn->right : fn->left;
 646         } while (fn);
 647
 648         if (!allow_create) {
 649                 /* We should not create new node because
 650                  * NLM_F_REPLACE was specified without NLM_F_CREATE
 651                  * I assume it is safe to require NLM_F_CREATE when
 652                  * REPLACE flag is used! Later we may want to remove the
 653                  * check for replace_required, because according
 654                  * to netlink specification, NLM_F_CREATE
 655                  * MUST be specified if new route is created.
 656                  * That would keep IPv6 consistent with IPv4
 657                  */
 658                 if (replace_required) {
 659                         NL_SET_ERR_MSG(extack,
 660                                        "Can not replace route - no match found");
 661                         pr_warn("Can't replace route, no match found\n");
 662                         return ERR_PTR(-ENOENT);
 663                 }
 664                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 665         }
 666         /*
 667          *      We walked to the bottom of tree.
 668          *      Create new leaf node without children.
 669          */
 670
 671         ln = node_alloc();
 672
 673         if (!ln)
 674                 return ERR_PTR(-ENOMEM);
 675         ln->fn_bit = plen;
 676
 677         ln->parent = pn;
 678         ln->fn_sernum = sernum;
 679
 680         if (dir)
 681                 pn->right = ln;
 682         else
 683                 pn->left  = ln;
 684
 685         return ln;
 686
 687
 688 insert_above:
 689         /*
 690          * split since we don't have a common prefix anymore or
 691          * we have a less significant route.
 692          * we've to insert an intermediate node on the list
 693          * this new node will point to the one we need to create
 694          * and the current
 695          */
 696
 697         pn = fn->parent;
 698
 699         /* find 1st bit in difference between the 2 addrs.
 700
 701            See comment in __ipv6_addr_diff: bit may be an invalid value,
 702            but if it is >= plen, the value is ignored in any case.
 703          */
 704
 705         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
 706
 707         /*
 708          *              (intermediate)[in]
 709          *                /        \
 710          *      (new leaf node)[ln] (old node)[fn]
 711          */
 712         if (plen > bit) {
 713                 in = node_alloc();
 714                 ln = node_alloc();
 715
 716                 if (!in || !ln) {
 717                         if (in)
 718                                 node_free_immediate(in);
 719                         if (ln)
 720                                 node_free_immediate(ln);
 721                         return ERR_PTR(-ENOMEM);
 722                 }
 723
 724                 /*
 725                  * new intermediate node.
 726                  * RTN_RTINFO will
 727                  * be off since that an address that chooses one of
 728                  * the branches would not match less specific routes
 729                  * in the other branch
 730                  */
 731
 732                 in->fn_bit = bit;
 733
 734                 in->parent = pn;
 735                 in->leaf = fn->leaf;
 736                 atomic_inc(&in->leaf->rt6i_ref);
 737
 738                 in->fn_sernum = sernum;
 739
 740                 /* update parent pointer */
 741                 if (dir)
 742                         pn->right = in;
 743                 else
 744                         pn->left  = in;
 745
 746                 ln->fn_bit = plen;
 747
 748                 ln->parent = in;
 749                 fn->parent = in;
 750
 751                 ln->fn_sernum = sernum;
 752
 753                 if (addr_bit_set(addr, bit)) {
 754                         in->right = ln;
 755                         in->left  = fn;
 756                 } else {
 757                         in->left  = ln;
 758                         in->right = fn;
 759                 }
 760         } else { /* plen <= bit */
 761
 762                 /*
 763                  *              (new leaf node)[ln]
 764                  *                /        \
 765                  *           (old node)[fn] NULL
 766                  */
 767
 768                 ln = node_alloc();
 769
 770                 if (!ln)
 771                         return ERR_PTR(-ENOMEM);
 772
 773                 ln->fn_bit = plen;
 774
 775                 ln->parent = pn;
 776
 777                 ln->fn_sernum = sernum;
 778
 779                 if (dir)
 780                         pn->right = ln;
 781                 else
 782                         pn->left  = ln;
 783
 784                 if (addr_bit_set(&key->addr, plen))
 785                         ln->right = fn;
 786                 else
 787                         ln->left  = fn;
 788
 789                 fn->parent = ln;
 790         }
 791         return ln;
 792 }
 793
 794 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
 795 {
 796         return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
 797                RTF_GATEWAY;
 798 }
 799
 800 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
 801 {
 802         int i;
 803
 804         for (i = 0; i < RTAX_MAX; i++) {
 805                 if (test_bit(i, mxc->mx_valid))
 806                         mp[i] = mxc->mx[i];
 807         }
 808 }
 809
 810 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
 811 {
 812         if (!mxc->mx)
 813                 return 0;
 814
 815         if (dst->flags & DST_HOST) {
 816                 u32 *mp = dst_metrics_write_ptr(dst);
 817
 818                 if (unlikely(!mp))
 819                         return -ENOMEM;
 820
 821                 fib6_copy_metrics(mp, mxc);
 822         } else {
 823                 dst_init_metrics(dst, mxc->mx, false);
 824
 825                 /* We've stolen mx now. */
 826                 mxc->mx = NULL;
 827         }
 828
 829         return 0;
 830 }
 831
 832 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
 833                           struct net *net)
 834 {
 835         if (atomic_read(&rt->rt6i_ref) != 1) {
 836                 /* This route is used as dummy address holder in some split
 837                  * nodes. It is not leaked, but it still holds other resources,
 838                  * which must be released in time. So, scan ascendant nodes
 839                  * and replace dummy references to this route with references
 840                  * to still alive ones.
 841                  */
 842                 while (fn) {
 843                         if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
 844                                 fn->leaf = fib6_find_prefix(net, fn);
 845                                 atomic_inc(&fn->leaf->rt6i_ref);
 846                                 rt6_release(rt);
 847                         }
 848                         fn = fn->parent;
 849                 }
 850         }
 851 }
 852
 853 /*
 854  *      Insert routing information in a node.
 855  */
 856
 857 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
 858                             struct nl_info *info, struct mx6_config *mxc)
 859 {
 860         struct rt6_info *iter = NULL;
 861         struct rt6_info **ins;
 862         struct rt6_info **fallback_ins = NULL;
 863         int replace = (info->nlh &&
 864                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
 865         int add = (!info->nlh ||
 866                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
 867         int found = 0;
 868         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
 869         u16 nlflags = NLM_F_EXCL;
 870         int err;
 871
 872         if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
 873                 nlflags |= NLM_F_APPEND;
 874
 875         ins = &fn->leaf;
 876
 877         for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
 878                 /*
 879                  *      Search for duplicates
 880                  */
 881
 882                 if (iter->rt6i_metric == rt->rt6i_metric) {
 883                         /*
 884                          *      Same priority level
 885                          */
 886                         if (info->nlh &&
 887                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
 888                                 return -EEXIST;
 889
 890                         nlflags &= ~NLM_F_EXCL;
 891                         if (replace) {
 892                                 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
 893                                         found++;
 894                                         break;
 895                                 }
 896                                 if (rt_can_ecmp)
 897                                         fallback_ins = fallback_ins ?: ins;
 898                                 goto next_iter;
 899                         }
 900
 901                         if (rt6_duplicate_nexthop(iter, rt)) {
 902                                 if (rt->rt6i_nsiblings)
 903                                         rt->rt6i_nsiblings = 0;
 904                                 if (!(iter->rt6i_flags & RTF_EXPIRES))
 905                                         return -EEXIST;
 906                                 if (!(rt->rt6i_flags & RTF_EXPIRES))
 907                                         rt6_clean_expires(iter);
 908                                 else
 909                                         rt6_set_expires(iter, rt->dst.expires);
 910                                 iter->rt6i_pmtu = rt->rt6i_pmtu;
 911                                 return -EEXIST;
 912                         }
 913                         /* If we have the same destination and the same metric,
 914                          * but not the same gateway, then the route we try to
 915                          * add is sibling to this route, increment our counter
 916                          * of siblings, and later we will add our route to the
 917                          * list.
 918                          * Only static routes (which don't have flag
 919                          * RTF_EXPIRES) are used for ECMPv6.
 920                          *
 921                          * To avoid long list, we only had siblings if the
 922                          * route have a gateway.
 923                          */
 924                         if (rt_can_ecmp &&
 925                             rt6_qualify_for_ecmp(iter))
 926                                 rt->rt6i_nsiblings++;
 927                 }
 928
 929                 if (iter->rt6i_metric > rt->rt6i_metric)
 930                         break;
 931
 932 next_iter:
 933                 ins = &iter->dst.rt6_next;
 934         }
 935
 936         if (fallback_ins && !found) {
 937                 /* No ECMP-able route found, replace first non-ECMP one */
 938                 ins = fallback_ins;
 939                 iter = *ins;
 940                 found++;
 941         }
 942
 943         /* Reset round-robin state, if necessary */
 944         if (ins == &fn->leaf)
 945                 fn->rr_ptr = NULL;
 946
 947         /* Link this route to others same route. */
 948         if (rt->rt6i_nsiblings) {
 949                 unsigned int rt6i_nsiblings;
 950                 struct rt6_info *sibling, *temp_sibling;
 951
 952                 /* Find the first route that have the same metric */
 953                 sibling = fn->leaf;
 954                 while (sibling) {
 955                         if (sibling->rt6i_metric == rt->rt6i_metric &&
 956                             rt6_qualify_for_ecmp(sibling)) {
 957                                 list_add_tail(&rt->rt6i_siblings,
 958                                               &sibling->rt6i_siblings);
 959                                 break;
 960                         }
 961                         sibling = sibling->dst.rt6_next;
 962                 }
 963                 /* For each sibling in the list, increment the counter of
 964                  * siblings. BUG() if counters does not match, list of siblings
 965                  * is broken!
 966                  */
 967                 rt6i_nsiblings = 0;
 968                 list_for_each_entry_safe(sibling, temp_sibling,
 969                                          &rt->rt6i_siblings, rt6i_siblings) {
 970                         sibling->rt6i_nsiblings++;
 971                         BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
 972                         rt6i_nsiblings++;
 973                 }
 974                 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
 975         }
 976
 977         /*
 978          *      insert node
 979          */
 980         if (!replace) {
 981                 if (!add)
 982                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
 983
 984 add:
 985                 nlflags |= NLM_F_CREATE;
 986                 err = fib6_commit_metrics(&rt->dst, mxc);
 987                 if (err)
 988                         return err;
 989
 990                 rt->dst.rt6_next = iter;
 991                 *ins = rt;
 992                 rcu_assign_pointer(rt->rt6i_node, fn);
 993                 atomic_inc(&rt->rt6i_ref);
 994                 call_fib6_entry_notifiers(info->nl_net, FIB_EVENT_ENTRY_ADD,
 995                                           rt);
 996                 if (!info->skip_notify)
 997                         inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
 998                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
 999
1000                 if (!(fn->fn_flags & RTN_RTINFO)) {
1001                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1002                         fn->fn_flags |= RTN_RTINFO;
1003                 }
1004
1005         } else {
1006                 int nsiblings;
1007
1008                 if (!found) {
1009                         if (add)
1010                                 goto add;
1011                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1012                         return -ENOENT;
1013                 }
1014
1015                 err = fib6_commit_metrics(&rt->dst, mxc);
1016                 if (err)
1017                         return err;
1018
1019                 *ins = rt;
1020                 rcu_assign_pointer(rt->rt6i_node, fn);
1021                 rt->dst.rt6_next = iter->dst.rt6_next;
1022                 atomic_inc(&rt->rt6i_ref);
1023                 call_fib6_entry_notifiers(info->nl_net, FIB_EVENT_ENTRY_REPLACE,
1024                                           rt);
1025                 if (!info->skip_notify)
1026                         inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1027                 if (!(fn->fn_flags & RTN_RTINFO)) {
1028                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1029                         fn->fn_flags |= RTN_RTINFO;
1030                 }
1031                 nsiblings = iter->rt6i_nsiblings;
1032                 iter->rt6i_node = NULL;
1033                 fib6_purge_rt(iter, fn, info->nl_net);
1034                 if (fn->rr_ptr == iter)
1035                         fn->rr_ptr = NULL;
1036                 rt6_release(iter);
1037
1038                 if (nsiblings) {
1039                         /* Replacing an ECMP route, remove all siblings */
1040                         ins = &rt->dst.rt6_next;
1041                         iter = *ins;
1042                         while (iter) {
1043                                 if (iter->rt6i_metric > rt->rt6i_metric)
1044                                         break;
1045                                 if (rt6_qualify_for_ecmp(iter)) {
1046                                         *ins = iter->dst.rt6_next;
1047                                         iter->rt6i_node = NULL;
1048                                         fib6_purge_rt(iter, fn, info->nl_net);
1049                                         if (fn->rr_ptr == iter)
1050                                                 fn->rr_ptr = NULL;
1051                                         rt6_release(iter);
1052                                         nsiblings--;
1053                                 } else {
1054                                         ins = &iter->dst.rt6_next;
1055                                 }
1056                                 iter = *ins;
1057                         }
1058                         WARN_ON(nsiblings != 0);
1059                 }
1060         }
1061
1062         return 0;
1063 }
1064
1065 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
1066 {
1067         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1068             (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
1069                 mod_timer(&net->ipv6.ip6_fib_timer,
1070                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1071 }
1072
1073 void fib6_force_start_gc(struct net *net)
1074 {
1075         if (!timer_pending(&net->ipv6.ip6_fib_timer))
1076                 mod_timer(&net->ipv6.ip6_fib_timer,
1077                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1078 }
1079
1080 /*
1081  *      Add routing information to the routing tree.
1082  *      <destination addr>/<source addr>
1083  *      with source addr info in sub-trees
1084  */
1085
1086 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
1087              struct nl_info *info, struct mx6_config *mxc,
1088              struct netlink_ext_ack *extack)
1089 {
1090         struct fib6_node *fn, *pn = NULL;
1091         int err = -ENOMEM;
1092         int allow_create = 1;
1093         int replace_required = 0;
1094         int sernum = fib6_new_sernum(info->nl_net);
1095
1096         if (WARN_ON_ONCE(!atomic_read(&rt->dst.__refcnt)))
1097                 return -EINVAL;
1098
1099         if (info->nlh) {
1100                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1101                         allow_create = 0;
1102                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1103                         replace_required = 1;
1104         }
1105         if (!allow_create && !replace_required)
1106                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1107
1108         fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
1109                         offsetof(struct rt6_info, rt6i_dst), allow_create,
1110                         replace_required, sernum, extack);
1111         if (IS_ERR(fn)) {
1112                 err = PTR_ERR(fn);
1113                 fn = NULL;
1114                 goto out;
1115         }
1116
1117         pn = fn;
1118
1119 #ifdef CONFIG_IPV6_SUBTREES
1120         if (rt->rt6i_src.plen) {
1121                 struct fib6_node *sn;
1122
1123                 if (!fn->subtree) {
1124                         struct fib6_node *sfn;
1125
1126                         /*
1127                          * Create subtree.
1128                          *
1129                          *              fn[main tree]
1130                          *              |
1131                          *              sfn[subtree root]
1132                          *                 \
1133                          *                  sn[new leaf node]
1134                          */
1135
1136                         /* Create subtree root node */
1137                         sfn = node_alloc();
1138                         if (!sfn)
1139                                 goto failure;
1140
1141                         sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
1142                         atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1143                         sfn->fn_flags = RTN_ROOT;
1144                         sfn->fn_sernum = sernum;
1145
1146                         /* Now add the first leaf node to new subtree */
1147
1148                         sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
1149                                         rt->rt6i_src.plen,
1150                                         offsetof(struct rt6_info, rt6i_src),
1151                                         allow_create, replace_required, sernum,
1152                                         extack);
1153
1154                         if (IS_ERR(sn)) {
1155                                 /* If it is failed, discard just allocated
1156                                    root, and then (in failure) stale node
1157                                    in main tree.
1158                                  */
1159                                 node_free_immediate(sfn);
1160                                 err = PTR_ERR(sn);
1161                                 goto failure;
1162                         }
1163
1164                         /* Now link new subtree to main tree */
1165                         sfn->parent = fn;
1166                         fn->subtree = sfn;
1167                 } else {
1168                         sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
1169                                         rt->rt6i_src.plen,
1170                                         offsetof(struct rt6_info, rt6i_src),
1171                                         allow_create, replace_required, sernum,
1172                                         extack);
1173
1174                         if (IS_ERR(sn)) {
1175                                 err = PTR_ERR(sn);
1176                                 goto failure;
1177                         }
1178                 }
1179
1180                 if (!fn->leaf) {
1181                         fn->leaf = rt;
1182                         atomic_inc(&rt->rt6i_ref);
1183                 }
1184                 fn = sn;
1185         }
1186 #endif
1187
1188         err = fib6_add_rt2node(fn, rt, info, mxc);
1189         if (!err) {
1190                 fib6_start_gc(info->nl_net, rt);
1191                 if (!(rt->rt6i_flags & RTF_CACHE))
1192                         fib6_prune_clones(info->nl_net, pn);
1193         }
1194
1195 out:
1196         if (err) {
1197 #ifdef CONFIG_IPV6_SUBTREES
1198                 /*
1199                  * If fib6_add_1 has cleared the old leaf pointer in the
1200                  * super-tree leaf node we have to find a new one for it.
1201                  */
1202                 if (pn != fn && pn->leaf == rt) {
1203                         pn->leaf = NULL;
1204                         atomic_dec(&rt->rt6i_ref);
1205                 }
1206                 if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
1207                         pn->leaf = fib6_find_prefix(info->nl_net, pn);
1208 #if RT6_DEBUG >= 2
1209                         if (!pn->leaf) {
1210                                 WARN_ON(pn->leaf == NULL);
1211                                 pn->leaf = info->nl_net->ipv6.ip6_null_entry;
1212                         }
1213 #endif
1214                         atomic_inc(&pn->leaf->rt6i_ref);
1215                 }
1216 #endif
1217                 goto failure;
1218         }
1219         return err;
1220
1221 failure:
1222         /* fn->leaf could be NULL if fn is an intermediate node and we
1223          * failed to add the new route to it in both subtree creation
1224          * failure and fib6_add_rt2node() failure case.
1225          * In both cases, fib6_repair_tree() should be called to fix
1226          * fn->leaf.
1227          */
1228         if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1229                 fib6_repair_tree(info->nl_net, fn);
1230         /* Always release dst as dst->__refcnt is guaranteed
1231          * to be taken before entering this function
1232          */
1233         dst_release_immediate(&rt->dst);
1234         return err;
1235 }
1236
1237 /*
1238  *      Routing tree lookup
1239  *
1240  */
1241
1242 struct lookup_args {
1243         int                     offset;         /* key offset on rt6_info       */
1244         const struct in6_addr   *addr;          /* search key                   */
1245 };
1246
1247 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1248                                        struct lookup_args *args)
1249 {
1250         struct fib6_node *fn;
1251         __be32 dir;
1252
1253         if (unlikely(args->offset == 0))
1254                 return NULL;
1255
1256         /*
1257          *      Descend on a tree
1258          */
1259
1260         fn = root;
1261
1262         for (;;) {
1263                 struct fib6_node *next;
1264
1265                 dir = addr_bit_set(args->addr, fn->fn_bit);
1266
1267                 next = dir ? fn->right : fn->left;
1268
1269                 if (next) {
1270                         fn = next;
1271                         continue;
1272                 }
1273                 break;
1274         }
1275
1276         while (fn) {
1277                 if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
1278                         struct rt6key *key;
1279
1280                         key = (struct rt6key *) ((u8 *) fn->leaf +
1281                                                  args->offset);
1282
1283                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1284 #ifdef CONFIG_IPV6_SUBTREES
1285                                 if (fn->subtree) {
1286                                         struct fib6_node *sfn;
1287                                         sfn = fib6_lookup_1(fn->subtree,
1288                                                             args + 1);
1289                                         if (!sfn)
1290                                                 goto backtrack;
1291                                         fn = sfn;
1292                                 }
1293 #endif
1294                                 if (fn->fn_flags & RTN_RTINFO)
1295                                         return fn;
1296                         }
1297                 }
1298 #ifdef CONFIG_IPV6_SUBTREES
1299 backtrack:
1300 #endif
1301                 if (fn->fn_flags & RTN_ROOT)
1302                         break;
1303
1304                 fn = fn->parent;
1305         }
1306
1307         return NULL;
1308 }
1309
1310 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1311                               const struct in6_addr *saddr)
1312 {
1313         struct fib6_node *fn;
1314         struct lookup_args args[] = {
1315                 {
1316                         .offset = offsetof(struct rt6_info, rt6i_dst),
1317                         .addr = daddr,
1318                 },
1319 #ifdef CONFIG_IPV6_SUBTREES
1320                 {
1321                         .offset = offsetof(struct rt6_info, rt6i_src),
1322                         .addr = saddr,
1323                 },
1324 #endif
1325                 {
1326                         .offset = 0,    /* sentinel */
1327                 }
1328         };
1329
1330         fn = fib6_lookup_1(root, daddr ? args : args + 1);
1331         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1332                 fn = root;
1333
1334         return fn;
1335 }
1336
1337 /*
1338  *      Get node with specified destination prefix (and source prefix,
1339  *      if subtrees are used)
1340  */
1341
1342
1343 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1344                                        const struct in6_addr *addr,
1345                                        int plen, int offset)
1346 {
1347         struct fib6_node *fn;
1348
1349         for (fn = root; fn ; ) {
1350                 struct rt6key *key = (struct rt6key *)((u8 *)fn->leaf + offset);
1351
1352                 /*
1353                  *      Prefix match
1354                  */
1355                 if (plen < fn->fn_bit ||
1356                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1357                         return NULL;
1358
1359                 if (plen == fn->fn_bit)
1360                         return fn;
1361
1362                 /*
1363                  *      We have more bits to go
1364                  */
1365                 if (addr_bit_set(addr, fn->fn_bit))
1366                         fn = fn->right;
1367                 else
1368                         fn = fn->left;
1369         }
1370         return NULL;
1371 }
1372
1373 struct fib6_node *fib6_locate(struct fib6_node *root,
1374                               const struct in6_addr *daddr, int dst_len,
1375                               const struct in6_addr *saddr, int src_len)
1376 {
1377         struct fib6_node *fn;
1378
1379         fn = fib6_locate_1(root, daddr, dst_len,
1380                            offsetof(struct rt6_info, rt6i_dst));
1381
1382 #ifdef CONFIG_IPV6_SUBTREES
1383         if (src_len) {
1384                 WARN_ON(saddr == NULL);
1385                 if (fn && fn->subtree)
1386                         fn = fib6_locate_1(fn->subtree, saddr, src_len,
1387                                            offsetof(struct rt6_info, rt6i_src));
1388         }
1389 #endif
1390
1391         if (fn && fn->fn_flags & RTN_RTINFO)
1392                 return fn;
1393
1394         return NULL;
1395 }
1396
1397
1398 /*
1399  *      Deletion
1400  *
1401  */
1402
1403 static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
1404 {
1405         if (fn->fn_flags & RTN_ROOT)
1406                 return net->ipv6.ip6_null_entry;
1407
1408         while (fn) {
1409                 if (fn->left)
1410                         return fn->left->leaf;
1411                 if (fn->right)
1412                         return fn->right->leaf;
1413
1414                 fn = FIB6_SUBTREE(fn);
1415         }
1416         return NULL;
1417 }
1418
1419 /*
1420  *      Called to trim the tree of intermediate nodes when possible. "fn"
1421  *      is the node we want to try and remove.
1422  */
1423
1424 static struct fib6_node *fib6_repair_tree(struct net *net,
1425                                            struct fib6_node *fn)
1426 {
1427         int children;
1428         int nstate;
1429         struct fib6_node *child, *pn;
1430         struct fib6_walker *w;
1431         int iter = 0;
1432
1433         for (;;) {
1434                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1435                 iter++;
1436
1437                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1438                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1439                 WARN_ON(fn->leaf);
1440
1441                 children = 0;
1442                 child = NULL;
1443                 if (fn->right)
1444                         child = fn->right, children |= 1;
1445                 if (fn->left)
1446                         child = fn->left, children |= 2;
1447
1448                 if (children == 3 || FIB6_SUBTREE(fn)
1449 #ifdef CONFIG_IPV6_SUBTREES
1450                     /* Subtree root (i.e. fn) may have one child */
1451                     || (children && fn->fn_flags & RTN_ROOT)
1452 #endif
1453                     ) {
1454                         fn->leaf = fib6_find_prefix(net, fn);
1455 #if RT6_DEBUG >= 2
1456                         if (!fn->leaf) {
1457                                 WARN_ON(!fn->leaf);
1458                                 fn->leaf = net->ipv6.ip6_null_entry;
1459                         }
1460 #endif
1461                         atomic_inc(&fn->leaf->rt6i_ref);
1462                         return fn->parent;
1463                 }
1464
1465                 pn = fn->parent;
1466 #ifdef CONFIG_IPV6_SUBTREES
1467                 if (FIB6_SUBTREE(pn) == fn) {
1468                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1469                         FIB6_SUBTREE(pn) = NULL;
1470                         nstate = FWS_L;
1471                 } else {
1472                         WARN_ON(fn->fn_flags & RTN_ROOT);
1473 #endif
1474                         if (pn->right == fn)
1475                                 pn->right = child;
1476                         else if (pn->left == fn)
1477                                 pn->left = child;
1478 #if RT6_DEBUG >= 2
1479                         else
1480                                 WARN_ON(1);
1481 #endif
1482                         if (child)
1483                                 child->parent = pn;
1484                         nstate = FWS_R;
1485 #ifdef CONFIG_IPV6_SUBTREES
1486                 }
1487 #endif
1488
1489                 read_lock(&net->ipv6.fib6_walker_lock);
1490                 FOR_WALKERS(net, w) {
1491                         if (!child) {
1492                                 if (w->root == fn) {
1493                                         w->root = w->node = NULL;
1494                                         RT6_TRACE("W %p adjusted by delroot 1\n", w);
1495                                 } else if (w->node == fn) {
1496                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1497                                         w->node = pn;
1498                                         w->state = nstate;
1499                                 }
1500                         } else {
1501                                 if (w->root == fn) {
1502                                         w->root = child;
1503                                         RT6_TRACE("W %p adjusted by delroot 2\n", w);
1504                                 }
1505                                 if (w->node == fn) {
1506                                         w->node = child;
1507                                         if (children&2) {
1508                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1509                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1510                                         } else {
1511                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1512                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1513                                         }
1514                                 }
1515                         }
1516                 }
1517                 read_unlock(&net->ipv6.fib6_walker_lock);
1518
1519                 node_free(fn);
1520                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1521                         return pn;
1522
1523                 rt6_release(pn->leaf);
1524                 pn->leaf = NULL;
1525                 fn = pn;
1526         }
1527 }
1528
1529 static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
1530                            struct nl_info *info)
1531 {
1532         struct fib6_walker *w;
1533         struct rt6_info *rt = *rtp;
1534         struct net *net = info->nl_net;
1535
1536         RT6_TRACE("fib6_del_route\n");
1537
1538         /* Unlink it */
1539         *rtp = rt->dst.rt6_next;
1540         rt->rt6i_node = NULL;
1541         net->ipv6.rt6_stats->fib_rt_entries--;
1542         net->ipv6.rt6_stats->fib_discarded_routes++;
1543
1544         /* Reset round-robin state, if necessary */
1545         if (fn->rr_ptr == rt)
1546                 fn->rr_ptr = NULL;
1547
1548         /* Remove this entry from other siblings */
1549         if (rt->rt6i_nsiblings) {
1550                 struct rt6_info *sibling, *next_sibling;
1551
1552                 list_for_each_entry_safe(sibling, next_sibling,
1553                                          &rt->rt6i_siblings, rt6i_siblings)
1554                         sibling->rt6i_nsiblings--;
1555                 rt->rt6i_nsiblings = 0;
1556                 list_del_init(&rt->rt6i_siblings);
1557         }
1558
1559         /* Adjust walkers */
1560         read_lock(&net->ipv6.fib6_walker_lock);
1561         FOR_WALKERS(net, w) {
1562                 if (w->state == FWS_C && w->leaf == rt) {
1563                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1564                         w->leaf = rt->dst.rt6_next;
1565                         if (!w->leaf)
1566                                 w->state = FWS_U;
1567                 }
1568         }
1569         read_unlock(&net->ipv6.fib6_walker_lock);
1570
1571         rt->dst.rt6_next = NULL;
1572
1573         /* If it was last route, expunge its radix tree node */
1574         if (!fn->leaf) {
1575                 fn->fn_flags &= ~RTN_RTINFO;
1576                 net->ipv6.rt6_stats->fib_route_nodes--;
1577                 fn = fib6_repair_tree(net, fn);
1578         }
1579
1580         fib6_purge_rt(rt, fn, net);
1581
1582         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt);
1583         if (!info->skip_notify)
1584                 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1585         rt6_release(rt);
1586 }
1587
1588 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1589 {
1590         struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1591                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
1592         struct net *net = info->nl_net;
1593         struct rt6_info **rtp;
1594
1595 #if RT6_DEBUG >= 2
1596         if (rt->dst.obsolete > 0) {
1597                 WARN_ON(fn);
1598                 return -ENOENT;
1599         }
1600 #endif
1601         if (!fn || rt == net->ipv6.ip6_null_entry)
1602                 return -ENOENT;
1603
1604         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1605
1606         if (!(rt->rt6i_flags & RTF_CACHE)) {
1607                 struct fib6_node *pn = fn;
1608 #ifdef CONFIG_IPV6_SUBTREES
1609                 /* clones of this route might be in another subtree */
1610                 if (rt->rt6i_src.plen) {
1611                         while (!(pn->fn_flags & RTN_ROOT))
1612                                 pn = pn->parent;
1613                         pn = pn->parent;
1614                 }
1615 #endif
1616                 fib6_prune_clones(info->nl_net, pn);
1617         }
1618
1619         /*
1620          *      Walk the leaf entries looking for ourself
1621          */
1622
1623         for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
1624                 if (*rtp == rt) {
1625                         fib6_del_route(fn, rtp, info);
1626                         return 0;
1627                 }
1628         }
1629         return -ENOENT;
1630 }
1631
1632 /*
1633  *      Tree traversal function.
1634  *
1635  *      Certainly, it is not interrupt safe.
1636  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1637  *      It means, that we can modify tree during walking
1638  *      and use this function for garbage collection, clone pruning,
1639  *      cleaning tree when a device goes down etc. etc.
1640  *
1641  *      It guarantees that every node will be traversed,
1642  *      and that it will be traversed only once.
1643  *
1644  *      Callback function w->func may return:
1645  *      0 -> continue walking.
1646  *      positive value -> walking is suspended (used by tree dumps,
1647  *      and probably by gc, if it will be split to several slices)
1648  *      negative value -> terminate walking.
1649  *
1650  *      The function itself returns:
1651  *      0   -> walk is complete.
1652  *      >0  -> walk is incomplete (i.e. suspended)
1653  *      <0  -> walk is terminated by an error.
1654  */
1655
1656 static int fib6_walk_continue(struct fib6_walker *w)
1657 {
1658         struct fib6_node *fn, *pn;
1659
1660         for (;;) {
1661                 fn = w->node;
1662                 if (!fn)
1663                         return 0;
1664
1665                 if (w->prune && fn != w->root &&
1666                     fn->fn_flags & RTN_RTINFO && w->state < FWS_C) {
1667                         w->state = FWS_C;
1668                         w->leaf = fn->leaf;
1669                 }
1670                 switch (w->state) {
1671 #ifdef CONFIG_IPV6_SUBTREES
1672                 case FWS_S:
1673                         if (FIB6_SUBTREE(fn)) {
1674                                 w->node = FIB6_SUBTREE(fn);
1675                                 continue;
1676                         }
1677                         w->state = FWS_L;
1678 #endif
1679                 case FWS_L:
1680                         if (fn->left) {
1681                                 w->node = fn->left;
1682                                 w->state = FWS_INIT;
1683                                 continue;
1684                         }
1685                         w->state = FWS_R;
1686                 case FWS_R:
1687                         if (fn->right) {
1688                                 w->node = fn->right;
1689                                 w->state = FWS_INIT;
1690                                 continue;
1691                         }
1692                         w->state = FWS_C;
1693                         w->leaf = fn->leaf;
1694                 case FWS_C:
1695                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1696                                 int err;
1697
1698                                 if (w->skip) {
1699                                         w->skip--;
1700                                         goto skip;
1701                                 }
1702
1703                                 err = w->func(w);
1704                                 if (err)
1705                                         return err;
1706
1707                                 w->count++;
1708                                 continue;
1709                         }
1710 skip:
1711                         w->state = FWS_U;
1712                 case FWS_U:
1713                         if (fn == w->root)
1714                                 return 0;
1715                         pn = fn->parent;
1716                         w->node = pn;
1717 #ifdef CONFIG_IPV6_SUBTREES
1718                         if (FIB6_SUBTREE(pn) == fn) {
1719                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1720                                 w->state = FWS_L;
1721                                 continue;
1722                         }
1723 #endif
1724                         if (pn->left == fn) {
1725                                 w->state = FWS_R;
1726                                 continue;
1727                         }
1728                         if (pn->right == fn) {
1729                                 w->state = FWS_C;
1730                                 w->leaf = w->node->leaf;
1731                                 continue;
1732                         }
1733 #if RT6_DEBUG >= 2
1734                         WARN_ON(1);
1735 #endif
1736                 }
1737         }
1738 }
1739
1740 static int fib6_walk(struct net *net, struct fib6_walker *w)
1741 {
1742         int res;
1743
1744         w->state = FWS_INIT;
1745         w->node = w->root;
1746
1747         fib6_walker_link(net, w);
1748         res = fib6_walk_continue(w);
1749         if (res <= 0)
1750                 fib6_walker_unlink(net, w);
1751         return res;
1752 }
1753
1754 static int fib6_clean_node(struct fib6_walker *w)
1755 {
1756         int res;
1757         struct rt6_info *rt;
1758         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1759         struct nl_info info = {
1760                 .nl_net = c->net,
1761         };
1762
1763         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1764             w->node->fn_sernum != c->sernum)
1765                 w->node->fn_sernum = c->sernum;
1766
1767         if (!c->func) {
1768                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1769                 w->leaf = NULL;
1770                 return 0;
1771         }
1772
1773         for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
1774                 res = c->func(rt, c->arg);
1775                 if (res < 0) {
1776                         w->leaf = rt;
1777                         res = fib6_del(rt, &info);
1778                         if (res) {
1779 #if RT6_DEBUG >= 2
1780                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1781                                          __func__, rt,
1782                                          rcu_access_pointer(rt->rt6i_node),
1783                                          res);
1784 #endif
1785                                 continue;
1786                         }
1787                         return 0;
1788                 }
1789                 WARN_ON(res != 0);
1790         }
1791         w->leaf = rt;
1792         return 0;
1793 }
1794
1795 /*
1796  *      Convenient frontend to tree walker.
1797  *
1798  *      func is called on each route.
1799  *              It may return -1 -> delete this route.
1800  *                            0  -> continue walking
1801  *
1802  *      prune==1 -> only immediate children of node (certainly,
1803  *      ignoring pure split nodes) will be scanned.
1804  */
1805
1806 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1807                             int (*func)(struct rt6_info *, void *arg),
1808                             bool prune, int sernum, void *arg)
1809 {
1810         struct fib6_cleaner c;
1811
1812         c.w.root = root;
1813         c.w.func = fib6_clean_node;
1814         c.w.prune = prune;
1815         c.w.count = 0;
1816         c.w.skip = 0;
1817         c.func = func;
1818         c.sernum = sernum;
1819         c.arg = arg;
1820         c.net = net;
1821
1822         fib6_walk(net, &c.w);
1823 }
1824
1825 static void __fib6_clean_all(struct net *net,
1826                              int (*func)(struct rt6_info *, void *),
1827                              int sernum, void *arg)
1828 {
1829         struct fib6_table *table;
1830         struct hlist_head *head;
1831         unsigned int h;
1832
1833         rcu_read_lock();
1834         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1835                 head = &net->ipv6.fib_table_hash[h];
1836                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1837                         write_lock_bh(&table->tb6_lock);
1838                         fib6_clean_tree(net, &table->tb6_root,
1839                                         func, false, sernum, arg);
1840                         write_unlock_bh(&table->tb6_lock);
1841                 }
1842         }
1843         rcu_read_unlock();
1844 }
1845
1846 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1847                     void *arg)
1848 {
1849         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1850 }
1851
1852 static int fib6_prune_clone(struct rt6_info *rt, void *arg)
1853 {
1854         if (rt->rt6i_flags & RTF_CACHE) {
1855                 RT6_TRACE("pruning clone %p\n", rt);
1856                 return -1;
1857         }
1858
1859         return 0;
1860 }
1861
1862 static void fib6_prune_clones(struct net *net, struct fib6_node *fn)
1863 {
1864         fib6_clean_tree(net, fn, fib6_prune_clone, true,
1865                         FIB6_NO_SERNUM_CHANGE, NULL);
1866 }
1867
1868 static void fib6_flush_trees(struct net *net)
1869 {
1870         int new_sernum = fib6_new_sernum(net);
1871
1872         __fib6_clean_all(net, NULL, new_sernum, NULL);
1873 }
1874
1875 /*
1876  *      Garbage collection
1877  */
1878
1879 struct fib6_gc_args
1880 {
1881         int                     timeout;
1882         int                     more;
1883 };
1884
1885 static int fib6_age(struct rt6_info *rt, void *arg)
1886 {
1887         struct fib6_gc_args *gc_args = arg;
1888         unsigned long now = jiffies;
1889
1890         /*
1891          *      check addrconf expiration here.
1892          *      Routes are expired even if they are in use.
1893          *
1894          *      Also age clones. Note, that clones are aged out
1895          *      only if they are not in use now.
1896          */
1897
1898         if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1899                 if (time_after(now, rt->dst.expires)) {
1900                         RT6_TRACE("expiring %p\n", rt);
1901                         return -1;
1902                 }
1903                 gc_args->more++;
1904         } else if (rt->rt6i_flags & RTF_CACHE) {
1905                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout))
1906                         rt->dst.obsolete = DST_OBSOLETE_KILL;
1907                 if (atomic_read(&rt->dst.__refcnt) == 1 &&
1908                     rt->dst.obsolete == DST_OBSOLETE_KILL) {
1909                         RT6_TRACE("aging clone %p\n", rt);
1910                         return -1;
1911                 } else if (rt->rt6i_flags & RTF_GATEWAY) {
1912                         struct neighbour *neigh;
1913                         __u8 neigh_flags = 0;
1914
1915                         neigh = dst_neigh_lookup(&rt->dst, &rt->rt6i_gateway);
1916                         if (neigh) {
1917                                 neigh_flags = neigh->flags;
1918                                 neigh_release(neigh);
1919                         }
1920                         if (!(neigh_flags & NTF_ROUTER)) {
1921                                 RT6_TRACE("purging route %p via non-router but gateway\n",
1922                                           rt);
1923                                 return -1;
1924                         }
1925                 }
1926                 gc_args->more++;
1927         }
1928
1929         return 0;
1930 }
1931
1932 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
1933 {
1934         struct fib6_gc_args gc_args;
1935         unsigned long now;
1936
1937         if (force) {
1938                 spin_lock_bh(&net->ipv6.fib6_gc_lock);
1939         } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
1940                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
1941                 return;
1942         }
1943         gc_args.timeout = expires ? (int)expires :
1944                           net->ipv6.sysctl.ip6_rt_gc_interval;
1945         gc_args.more = 0;
1946
1947         fib6_clean_all(net, fib6_age, &gc_args);
1948         now = jiffies;
1949         net->ipv6.ip6_rt_last_gc = now;
1950
1951         if (gc_args.more)
1952                 mod_timer(&net->ipv6.ip6_fib_timer,
1953                           round_jiffies(now
1954                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
1955         else
1956                 del_timer(&net->ipv6.ip6_fib_timer);
1957         spin_unlock_bh(&net->ipv6.fib6_gc_lock);
1958 }
1959
1960 static void fib6_gc_timer_cb(unsigned long arg)
1961 {
1962         fib6_run_gc(0, (struct net *)arg, true);
1963 }
1964
1965 static int __net_init fib6_net_init(struct net *net)
1966 {
1967         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
1968         int err;
1969
1970         err = fib6_notifier_init(net);
1971         if (err)
1972                 return err;
1973
1974         spin_lock_init(&net->ipv6.fib6_gc_lock);
1975         rwlock_init(&net->ipv6.fib6_walker_lock);
1976         INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
1977         setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
1978
1979         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
1980         if (!net->ipv6.rt6_stats)
1981                 goto out_timer;
1982
1983         /* Avoid false sharing : Use at least a full cache line */
1984         size = max_t(size_t, size, L1_CACHE_BYTES);
1985
1986         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
1987         if (!net->ipv6.fib_table_hash)
1988                 goto out_rt6_stats;
1989
1990         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
1991                                           GFP_KERNEL);
1992         if (!net->ipv6.fib6_main_tbl)
1993                 goto out_fib_table_hash;
1994
1995         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
1996         net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
1997         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
1998                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
1999         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2000
2001 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2002         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2003                                            GFP_KERNEL);
2004         if (!net->ipv6.fib6_local_tbl)
2005                 goto out_fib6_main_tbl;
2006         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2007         net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
2008         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2009                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2010         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2011 #endif
2012         fib6_tables_init(net);
2013
2014         return 0;
2015
2016 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2017 out_fib6_main_tbl:
2018         kfree(net->ipv6.fib6_main_tbl);
2019 #endif
2020 out_fib_table_hash:
2021         kfree(net->ipv6.fib_table_hash);
2022 out_rt6_stats:
2023         kfree(net->ipv6.rt6_stats);
2024 out_timer:
2025         fib6_notifier_exit(net);
2026         return -ENOMEM;
2027 }
2028
2029 static void fib6_net_exit(struct net *net)
2030 {
2031         unsigned int i;
2032
2033         rt6_ifdown(net, NULL);
2034         del_timer_sync(&net->ipv6.ip6_fib_timer);
2035
2036         for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2037                 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2038                 struct hlist_node *tmp;
2039                 struct fib6_table *tb;
2040
2041                 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2042                         hlist_del(&tb->tb6_hlist);
2043                         fib6_free_table(tb);
2044                 }
2045         }
2046
2047         kfree(net->ipv6.fib_table_hash);
2048         kfree(net->ipv6.rt6_stats);
2049         fib6_notifier_exit(net);
2050 }
2051
2052 static struct pernet_operations fib6_net_ops = {
2053         .init = fib6_net_init,
2054         .exit = fib6_net_exit,
2055 };
2056
2057 int __init fib6_init(void)
2058 {
2059         int ret = -ENOMEM;
2060
2061         fib6_node_kmem = kmem_cache_create("fib6_nodes",
2062                                            sizeof(struct fib6_node),
2063                                            0, SLAB_HWCACHE_ALIGN,
2064                                            NULL);
2065         if (!fib6_node_kmem)
2066                 goto out;
2067
2068         ret = register_pernet_subsys(&fib6_net_ops);
2069         if (ret)
2070                 goto out_kmem_cache_create;
2071
2072         ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
2073                               0);
2074         if (ret)
2075                 goto out_unregister_subsys;
2076
2077         __fib6_flush_trees = fib6_flush_trees;
2078 out:
2079         return ret;
2080
2081 out_unregister_subsys:
2082         unregister_pernet_subsys(&fib6_net_ops);
2083 out_kmem_cache_create:
2084         kmem_cache_destroy(fib6_node_kmem);
2085         goto out;
2086 }
2087
2088 void fib6_gc_cleanup(void)
2089 {
2090         unregister_pernet_subsys(&fib6_net_ops);
2091         kmem_cache_destroy(fib6_node_kmem);
2092 }
2093
2094 #ifdef CONFIG_PROC_FS
2095
2096 struct ipv6_route_iter {
2097         struct seq_net_private p;
2098         struct fib6_walker w;
2099         loff_t skip;
2100         struct fib6_table *tbl;
2101         int sernum;
2102 };
2103
2104 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2105 {
2106         struct rt6_info *rt = v;
2107         struct ipv6_route_iter *iter = seq->private;
2108
2109         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
2110
2111 #ifdef CONFIG_IPV6_SUBTREES
2112         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
2113 #else
2114         seq_puts(seq, "00000000000000000000000000000000 00 ");
2115 #endif
2116         if (rt->rt6i_flags & RTF_GATEWAY)
2117                 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
2118         else
2119                 seq_puts(seq, "00000000000000000000000000000000");
2120
2121         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2122                    rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
2123                    rt->dst.__use, rt->rt6i_flags,
2124                    rt->dst.dev ? rt->dst.dev->name : "");
2125         iter->w.leaf = NULL;
2126         return 0;
2127 }
2128
2129 static int ipv6_route_yield(struct fib6_walker *w)
2130 {
2131         struct ipv6_route_iter *iter = w->args;
2132
2133         if (!iter->skip)
2134                 return 1;
2135
2136         do {
2137                 iter->w.leaf = iter->w.leaf->dst.rt6_next;
2138                 iter->skip--;
2139                 if (!iter->skip && iter->w.leaf)
2140                         return 1;
2141         } while (iter->w.leaf);
2142
2143         return 0;
2144 }
2145
2146 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2147                                       struct net *net)
2148 {
2149         memset(&iter->w, 0, sizeof(iter->w));
2150         iter->w.func = ipv6_route_yield;
2151         iter->w.root = &iter->tbl->tb6_root;
2152         iter->w.state = FWS_INIT;
2153         iter->w.node = iter->w.root;
2154         iter->w.args = iter;
2155         iter->sernum = iter->w.root->fn_sernum;
2156         INIT_LIST_HEAD(&iter->w.lh);
2157         fib6_walker_link(net, &iter->w);
2158 }
2159
2160 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2161                                                     struct net *net)
2162 {
2163         unsigned int h;
2164         struct hlist_node *node;
2165
2166         if (tbl) {
2167                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2168                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2169         } else {
2170                 h = 0;
2171                 node = NULL;
2172         }
2173
2174         while (!node && h < FIB6_TABLE_HASHSZ) {
2175                 node = rcu_dereference_bh(
2176                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2177         }
2178         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2179 }
2180
2181 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2182 {
2183         if (iter->sernum != iter->w.root->fn_sernum) {
2184                 iter->sernum = iter->w.root->fn_sernum;
2185                 iter->w.state = FWS_INIT;
2186                 iter->w.node = iter->w.root;
2187                 WARN_ON(iter->w.skip);
2188                 iter->w.skip = iter->w.count;
2189         }
2190 }
2191
2192 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2193 {
2194         int r;
2195         struct rt6_info *n;
2196         struct net *net = seq_file_net(seq);
2197         struct ipv6_route_iter *iter = seq->private;
2198
2199         if (!v)
2200                 goto iter_table;
2201
2202         n = ((struct rt6_info *)v)->dst.rt6_next;
2203         if (n) {
2204                 ++*pos;
2205                 return n;
2206         }
2207
2208 iter_table:
2209         ipv6_route_check_sernum(iter);
2210         read_lock(&iter->tbl->tb6_lock);
2211         r = fib6_walk_continue(&iter->w);
2212         read_unlock(&iter->tbl->tb6_lock);
2213         if (r > 0) {
2214                 if (v)
2215                         ++*pos;
2216                 return iter->w.leaf;
2217         } else if (r < 0) {
2218                 fib6_walker_unlink(net, &iter->w);
2219                 return NULL;
2220         }
2221         fib6_walker_unlink(net, &iter->w);
2222
2223         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2224         if (!iter->tbl)
2225                 return NULL;
2226
2227         ipv6_route_seq_setup_walk(iter, net);
2228         goto iter_table;
2229 }
2230
2231 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2232         __acquires(RCU_BH)
2233 {
2234         struct net *net = seq_file_net(seq);
2235         struct ipv6_route_iter *iter = seq->private;
2236
2237         rcu_read_lock_bh();
2238         iter->tbl = ipv6_route_seq_next_table(NULL, net);
2239         iter->skip = *pos;
2240
2241         if (iter->tbl) {
2242                 ipv6_route_seq_setup_walk(iter, net);
2243                 return ipv6_route_seq_next(seq, NULL, pos);
2244         } else {
2245                 return NULL;
2246         }
2247 }
2248
2249 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2250 {
2251         struct fib6_walker *w = &iter->w;
2252         return w->node && !(w->state == FWS_U && w->node == w->root);
2253 }
2254
2255 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2256         __releases(RCU_BH)
2257 {
2258         struct net *net = seq_file_net(seq);
2259         struct ipv6_route_iter *iter = seq->private;
2260
2261         if (ipv6_route_iter_active(iter))
2262                 fib6_walker_unlink(net, &iter->w);
2263
2264         rcu_read_unlock_bh();
2265 }
2266
2267 static const struct seq_operations ipv6_route_seq_ops = {
2268         .start  = ipv6_route_seq_start,
2269         .next   = ipv6_route_seq_next,
2270         .stop   = ipv6_route_seq_stop,
2271         .show   = ipv6_route_seq_show
2272 };
2273
2274 int ipv6_route_open(struct inode *inode, struct file *file)
2275 {
2276         return seq_open_net(inode, file, &ipv6_route_seq_ops,
2277                             sizeof(struct ipv6_route_iter));
2278 }
2279
2280 #endif /* CONFIG_PROC_FS */