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 static struct kmem_cache *fib6_node_kmem __read_mostly;
  42
  43 struct fib6_cleaner {
  44         struct fib6_walker w;
  45         struct net *net;
  46         int (*func)(struct rt6_info *, void *arg);
  47         int sernum;
  48         void *arg;
  49 };
  50
  51 #ifdef CONFIG_IPV6_SUBTREES
  52 #define FWS_INIT FWS_S
  53 #else
  54 #define FWS_INIT FWS_L
  55 #endif
  56
  57 static struct rt6_info *fib6_find_prefix(struct net *net,
  58                                          struct fib6_table *table,
  59                                          struct fib6_node *fn);
  60 static struct fib6_node *fib6_repair_tree(struct net *net,
  61                                           struct fib6_table *table,
  62                                           struct fib6_node *fn);
  63 static int fib6_walk(struct net *net, struct fib6_walker *w);
  64 static int fib6_walk_continue(struct fib6_walker *w);
  65
  66 /*
  67  *      A routing update causes an increase of the serial number on the
  68  *      affected subtree. This allows for cached routes to be asynchronously
  69  *      tested when modifications are made to the destination cache as a
  70  *      result of redirects, path MTU changes, etc.
  71  */
  72
  73 static void fib6_gc_timer_cb(unsigned long arg);
  74
  75 #define FOR_WALKERS(net, w) \
  76         list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
  77
  78 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
  79 {
  80         write_lock_bh(&net->ipv6.fib6_walker_lock);
  81         list_add(&w->lh, &net->ipv6.fib6_walkers);
  82         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  83 }
  84
  85 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
  86 {
  87         write_lock_bh(&net->ipv6.fib6_walker_lock);
  88         list_del(&w->lh);
  89         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  90 }
  91
  92 static int fib6_new_sernum(struct net *net)
  93 {
  94         int new, old;
  95
  96         do {
  97                 old = atomic_read(&net->ipv6.fib6_sernum);
  98                 new = old < INT_MAX ? old + 1 : 1;
  99         } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
 100                                 old, new) != old);
 101         return new;
 102 }
 103
 104 enum {
 105         FIB6_NO_SERNUM_CHANGE = 0,
 106 };
 107
 108 void fib6_update_sernum(struct rt6_info *rt)
 109 {
 110         struct fib6_table *table = rt->rt6i_table;
 111         struct net *net = dev_net(rt->dst.dev);
 112         struct fib6_node *fn;
 113
 114         spin_lock_bh(&table->tb6_lock);
 115         fn = rcu_dereference_protected(rt->rt6i_node,
 116                         lockdep_is_held(&table->tb6_lock));
 117         if (fn)
 118                 fn->fn_sernum = fib6_new_sernum(net);
 119         spin_unlock_bh(&table->tb6_lock);
 120 }
 121
 122 /*
 123  *      Auxiliary address test functions for the radix tree.
 124  *
 125  *      These assume a 32bit processor (although it will work on
 126  *      64bit processors)
 127  */
 128
 129 /*
 130  *      test bit
 131  */
 132 #if defined(__LITTLE_ENDIAN)
 133 # define BITOP_BE32_SWIZZLE     (0x1F & ~7)
 134 #else
 135 # define BITOP_BE32_SWIZZLE     0
 136 #endif
 137
 138 static __be32 addr_bit_set(const void *token, int fn_bit)
 139 {
 140         const __be32 *addr = token;
 141         /*
 142          * Here,
 143          *      1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
 144          * is optimized version of
 145          *      htonl(1 << ((~fn_bit)&0x1F))
 146          * See include/asm-generic/bitops/le.h.
 147          */
 148         return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
 149                addr[fn_bit >> 5];
 150 }
 151
 152 static struct fib6_node *node_alloc(struct net *net)
 153 {
 154         struct fib6_node *fn;
 155
 156         fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
 157         if (fn)
 158                 net->ipv6.rt6_stats->fib_nodes++;
 159
 160         return fn;
 161 }
 162
 163 static void node_free_immediate(struct net *net, struct fib6_node *fn)
 164 {
 165         kmem_cache_free(fib6_node_kmem, fn);
 166         net->ipv6.rt6_stats->fib_nodes--;
 167 }
 168
 169 static void node_free_rcu(struct rcu_head *head)
 170 {
 171         struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
 172
 173         kmem_cache_free(fib6_node_kmem, fn);
 174 }
 175
 176 static void node_free(struct net *net, struct fib6_node *fn)
 177 {
 178         call_rcu(&fn->rcu, node_free_rcu);
 179         net->ipv6.rt6_stats->fib_nodes--;
 180 }
 181
 182 void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
 183 {
 184         int cpu;
 185
 186         if (!non_pcpu_rt->rt6i_pcpu)
 187                 return;
 188
 189         for_each_possible_cpu(cpu) {
 190                 struct rt6_info **ppcpu_rt;
 191                 struct rt6_info *pcpu_rt;
 192
 193                 ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
 194                 pcpu_rt = *ppcpu_rt;
 195                 if (pcpu_rt) {
 196                         dst_dev_put(&pcpu_rt->dst);
 197                         dst_release(&pcpu_rt->dst);
 198                         *ppcpu_rt = NULL;
 199                 }
 200         }
 201 }
 202 EXPORT_SYMBOL_GPL(rt6_free_pcpu);
 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         spin_lock_init(&tb->tb6_lock);
 219         h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
 220
 221         /*
 222          * No protection necessary, this is the only list mutatation
 223          * operation, tables never disappear once they exist.
 224          */
 225         hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
 226 }
 227
 228 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 229
 230 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
 231 {
 232         struct fib6_table *table;
 233
 234         table = kzalloc(sizeof(*table), GFP_ATOMIC);
 235         if (table) {
 236                 table->tb6_id = id;
 237                 rcu_assign_pointer(table->tb6_root.leaf,
 238                                    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 unsigned int fib6_tables_seq_read(struct net *net)
 327 {
 328         unsigned int h, fib_seq = 0;
 329
 330         rcu_read_lock();
 331         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 332                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 333                 struct fib6_table *tb;
 334
 335                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 336                         fib_seq += tb->fib_seq;
 337         }
 338         rcu_read_unlock();
 339
 340         return fib_seq;
 341 }
 342
 343 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
 344                                     enum fib_event_type event_type,
 345                                     struct rt6_info *rt)
 346 {
 347         struct fib6_entry_notifier_info info = {
 348                 .rt = rt,
 349         };
 350
 351         return call_fib6_notifier(nb, net, event_type, &info.info);
 352 }
 353
 354 static int call_fib6_entry_notifiers(struct net *net,
 355                                      enum fib_event_type event_type,
 356                                      struct rt6_info *rt,
 357                                      struct netlink_ext_ack *extack)
 358 {
 359         struct fib6_entry_notifier_info info = {
 360                 .info.extack = extack,
 361                 .rt = rt,
 362         };
 363
 364         rt->rt6i_table->fib_seq++;
 365         return call_fib6_notifiers(net, event_type, &info.info);
 366 }
 367
 368 struct fib6_dump_arg {
 369         struct net *net;
 370         struct notifier_block *nb;
 371 };
 372
 373 static void fib6_rt_dump(struct rt6_info *rt, struct fib6_dump_arg *arg)
 374 {
 375         if (rt == arg->net->ipv6.ip6_null_entry)
 376                 return;
 377         call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
 378 }
 379
 380 static int fib6_node_dump(struct fib6_walker *w)
 381 {
 382         struct rt6_info *rt;
 383
 384         for_each_fib6_walker_rt(w)
 385                 fib6_rt_dump(rt, w->args);
 386         w->leaf = NULL;
 387         return 0;
 388 }
 389
 390 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
 391                             struct fib6_walker *w)
 392 {
 393         w->root = &tb->tb6_root;
 394         spin_lock_bh(&tb->tb6_lock);
 395         fib6_walk(net, w);
 396         spin_unlock_bh(&tb->tb6_lock);
 397 }
 398
 399 /* Called with rcu_read_lock() */
 400 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
 401 {
 402         struct fib6_dump_arg arg;
 403         struct fib6_walker *w;
 404         unsigned int h;
 405
 406         w = kzalloc(sizeof(*w), GFP_ATOMIC);
 407         if (!w)
 408                 return -ENOMEM;
 409
 410         w->func = fib6_node_dump;
 411         arg.net = net;
 412         arg.nb = nb;
 413         w->args = &arg;
 414
 415         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 416                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 417                 struct fib6_table *tb;
 418
 419                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 420                         fib6_table_dump(net, tb, w);
 421         }
 422
 423         kfree(w);
 424
 425         return 0;
 426 }
 427
 428 static int fib6_dump_node(struct fib6_walker *w)
 429 {
 430         int res;
 431         struct rt6_info *rt;
 432
 433         for_each_fib6_walker_rt(w) {
 434                 res = rt6_dump_route(rt, w->args);
 435                 if (res < 0) {
 436                         /* Frame is full, suspend walking */
 437                         w->leaf = rt;
 438                         return 1;
 439                 }
 440
 441                 /* Multipath routes are dumped in one route with the
 442                  * RTA_MULTIPATH attribute. Jump 'rt' to point to the
 443                  * last sibling of this route (no need to dump the
 444                  * sibling routes again)
 445                  */
 446                 if (rt->rt6i_nsiblings)
 447                         rt = list_last_entry(&rt->rt6i_siblings,
 448                                              struct rt6_info,
 449                                              rt6i_siblings);
 450         }
 451         w->leaf = NULL;
 452         return 0;
 453 }
 454
 455 static void fib6_dump_end(struct netlink_callback *cb)
 456 {
 457         struct net *net = sock_net(cb->skb->sk);
 458         struct fib6_walker *w = (void *)cb->args[2];
 459
 460         if (w) {
 461                 if (cb->args[4]) {
 462                         cb->args[4] = 0;
 463                         fib6_walker_unlink(net, w);
 464                 }
 465                 cb->args[2] = 0;
 466                 kfree(w);
 467         }
 468         cb->done = (void *)cb->args[3];
 469         cb->args[1] = 3;
 470 }
 471
 472 static int fib6_dump_done(struct netlink_callback *cb)
 473 {
 474         fib6_dump_end(cb);
 475         return cb->done ? cb->done(cb) : 0;
 476 }
 477
 478 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
 479                            struct netlink_callback *cb)
 480 {
 481         struct net *net = sock_net(skb->sk);
 482         struct fib6_walker *w;
 483         int res;
 484
 485         w = (void *)cb->args[2];
 486         w->root = &table->tb6_root;
 487
 488         if (cb->args[4] == 0) {
 489                 w->count = 0;
 490                 w->skip = 0;
 491
 492                 spin_lock_bh(&table->tb6_lock);
 493                 res = fib6_walk(net, w);
 494                 spin_unlock_bh(&table->tb6_lock);
 495                 if (res > 0) {
 496                         cb->args[4] = 1;
 497                         cb->args[5] = w->root->fn_sernum;
 498                 }
 499         } else {
 500                 if (cb->args[5] != w->root->fn_sernum) {
 501                         /* Begin at the root if the tree changed */
 502                         cb->args[5] = w->root->fn_sernum;
 503                         w->state = FWS_INIT;
 504                         w->node = w->root;
 505                         w->skip = w->count;
 506                 } else
 507                         w->skip = 0;
 508
 509                 spin_lock_bh(&table->tb6_lock);
 510                 res = fib6_walk_continue(w);
 511                 spin_unlock_bh(&table->tb6_lock);
 512                 if (res <= 0) {
 513                         fib6_walker_unlink(net, w);
 514                         cb->args[4] = 0;
 515                 }
 516         }
 517
 518         return res;
 519 }
 520
 521 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 522 {
 523         struct net *net = sock_net(skb->sk);
 524         unsigned int h, s_h;
 525         unsigned int e = 0, s_e;
 526         struct rt6_rtnl_dump_arg arg;
 527         struct fib6_walker *w;
 528         struct fib6_table *tb;
 529         struct hlist_head *head;
 530         int res = 0;
 531
 532         s_h = cb->args[0];
 533         s_e = cb->args[1];
 534
 535         w = (void *)cb->args[2];
 536         if (!w) {
 537                 /* New dump:
 538                  *
 539                  * 1. hook callback destructor.
 540                  */
 541                 cb->args[3] = (long)cb->done;
 542                 cb->done = fib6_dump_done;
 543
 544                 /*
 545                  * 2. allocate and initialize walker.
 546                  */
 547                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
 548                 if (!w)
 549                         return -ENOMEM;
 550                 w->func = fib6_dump_node;
 551                 cb->args[2] = (long)w;
 552         }
 553
 554         arg.skb = skb;
 555         arg.cb = cb;
 556         arg.net = net;
 557         w->args = &arg;
 558
 559         rcu_read_lock();
 560         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
 561                 e = 0;
 562                 head = &net->ipv6.fib_table_hash[h];
 563                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 564                         if (e < s_e)
 565                                 goto next;
 566                         res = fib6_dump_table(tb, skb, cb);
 567                         if (res != 0)
 568                                 goto out;
 569 next:
 570                         e++;
 571                 }
 572         }
 573 out:
 574         rcu_read_unlock();
 575         cb->args[1] = e;
 576         cb->args[0] = h;
 577
 578         res = res < 0 ? res : skb->len;
 579         if (res <= 0)
 580                 fib6_dump_end(cb);
 581         return res;
 582 }
 583
 584 /*
 585  *      Routing Table
 586  *
 587  *      return the appropriate node for a routing tree "add" operation
 588  *      by either creating and inserting or by returning an existing
 589  *      node.
 590  */
 591
 592 static struct fib6_node *fib6_add_1(struct net *net,
 593                                     struct fib6_table *table,
 594                                     struct fib6_node *root,
 595                                     struct in6_addr *addr, int plen,
 596                                     int offset, int allow_create,
 597                                     int replace_required,
 598                                     struct netlink_ext_ack *extack)
 599 {
 600         struct fib6_node *fn, *in, *ln;
 601         struct fib6_node *pn = NULL;
 602         struct rt6key *key;
 603         int     bit;
 604         __be32  dir = 0;
 605
 606         RT6_TRACE("fib6_add_1\n");
 607
 608         /* insert node in tree */
 609
 610         fn = root;
 611
 612         do {
 613                 struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
 614                                             lockdep_is_held(&table->tb6_lock));
 615                 key = (struct rt6key *)((u8 *)leaf + offset);
 616
 617                 /*
 618                  *      Prefix match
 619                  */
 620                 if (plen < fn->fn_bit ||
 621                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
 622                         if (!allow_create) {
 623                                 if (replace_required) {
 624                                         NL_SET_ERR_MSG(extack,
 625                                                        "Can not replace route - no match found");
 626                                         pr_warn("Can't replace route, no match found\n");
 627                                         return ERR_PTR(-ENOENT);
 628                                 }
 629                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 630                         }
 631                         goto insert_above;
 632                 }
 633
 634                 /*
 635                  *      Exact match ?
 636                  */
 637
 638                 if (plen == fn->fn_bit) {
 639                         /* clean up an intermediate node */
 640                         if (!(fn->fn_flags & RTN_RTINFO)) {
 641                                 RCU_INIT_POINTER(fn->leaf, NULL);
 642                                 rt6_release(leaf);
 643                         }
 644
 645                         return fn;
 646                 }
 647
 648                 /*
 649                  *      We have more bits to go
 650                  */
 651
 652                 /* Try to walk down on tree. */
 653                 dir = addr_bit_set(addr, fn->fn_bit);
 654                 pn = fn;
 655                 fn = dir ?
 656                      rcu_dereference_protected(fn->right,
 657                                         lockdep_is_held(&table->tb6_lock)) :
 658                      rcu_dereference_protected(fn->left,
 659                                         lockdep_is_held(&table->tb6_lock));
 660         } while (fn);
 661
 662         if (!allow_create) {
 663                 /* We should not create new node because
 664                  * NLM_F_REPLACE was specified without NLM_F_CREATE
 665                  * I assume it is safe to require NLM_F_CREATE when
 666                  * REPLACE flag is used! Later we may want to remove the
 667                  * check for replace_required, because according
 668                  * to netlink specification, NLM_F_CREATE
 669                  * MUST be specified if new route is created.
 670                  * That would keep IPv6 consistent with IPv4
 671                  */
 672                 if (replace_required) {
 673                         NL_SET_ERR_MSG(extack,
 674                                        "Can not replace route - no match found");
 675                         pr_warn("Can't replace route, no match found\n");
 676                         return ERR_PTR(-ENOENT);
 677                 }
 678                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 679         }
 680         /*
 681          *      We walked to the bottom of tree.
 682          *      Create new leaf node without children.
 683          */
 684
 685         ln = node_alloc(net);
 686
 687         if (!ln)
 688                 return ERR_PTR(-ENOMEM);
 689         ln->fn_bit = plen;
 690         RCU_INIT_POINTER(ln->parent, pn);
 691
 692         if (dir)
 693                 rcu_assign_pointer(pn->right, ln);
 694         else
 695                 rcu_assign_pointer(pn->left, ln);
 696
 697         return ln;
 698
 699
 700 insert_above:
 701         /*
 702          * split since we don't have a common prefix anymore or
 703          * we have a less significant route.
 704          * we've to insert an intermediate node on the list
 705          * this new node will point to the one we need to create
 706          * and the current
 707          */
 708
 709         pn = rcu_dereference_protected(fn->parent,
 710                                        lockdep_is_held(&table->tb6_lock));
 711
 712         /* find 1st bit in difference between the 2 addrs.
 713
 714            See comment in __ipv6_addr_diff: bit may be an invalid value,
 715            but if it is >= plen, the value is ignored in any case.
 716          */
 717
 718         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
 719
 720         /*
 721          *              (intermediate)[in]
 722          *                /        \
 723          *      (new leaf node)[ln] (old node)[fn]
 724          */
 725         if (plen > bit) {
 726                 in = node_alloc(net);
 727                 ln = node_alloc(net);
 728
 729                 if (!in || !ln) {
 730                         if (in)
 731                                 node_free_immediate(net, in);
 732                         if (ln)
 733                                 node_free_immediate(net, ln);
 734                         return ERR_PTR(-ENOMEM);
 735                 }
 736
 737                 /*
 738                  * new intermediate node.
 739                  * RTN_RTINFO will
 740                  * be off since that an address that chooses one of
 741                  * the branches would not match less specific routes
 742                  * in the other branch
 743                  */
 744
 745                 in->fn_bit = bit;
 746
 747                 RCU_INIT_POINTER(in->parent, pn);
 748                 in->leaf = fn->leaf;
 749                 atomic_inc(&rcu_dereference_protected(in->leaf,
 750                                 lockdep_is_held(&table->tb6_lock))->rt6i_ref);
 751
 752                 /* update parent pointer */
 753                 if (dir)
 754                         rcu_assign_pointer(pn->right, in);
 755                 else
 756                         rcu_assign_pointer(pn->left, in);
 757
 758                 ln->fn_bit = plen;
 759
 760                 RCU_INIT_POINTER(ln->parent, in);
 761                 rcu_assign_pointer(fn->parent, in);
 762
 763                 if (addr_bit_set(addr, bit)) {
 764                         rcu_assign_pointer(in->right, ln);
 765                         rcu_assign_pointer(in->left, fn);
 766                 } else {
 767                         rcu_assign_pointer(in->left, ln);
 768                         rcu_assign_pointer(in->right, fn);
 769                 }
 770         } else { /* plen <= bit */
 771
 772                 /*
 773                  *              (new leaf node)[ln]
 774                  *                /        \
 775                  *           (old node)[fn] NULL
 776                  */
 777
 778                 ln = node_alloc(net);
 779
 780                 if (!ln)
 781                         return ERR_PTR(-ENOMEM);
 782
 783                 ln->fn_bit = plen;
 784
 785                 RCU_INIT_POINTER(ln->parent, pn);
 786
 787                 if (addr_bit_set(&key->addr, plen))
 788                         RCU_INIT_POINTER(ln->right, fn);
 789                 else
 790                         RCU_INIT_POINTER(ln->left, fn);
 791
 792                 rcu_assign_pointer(fn->parent, ln);
 793
 794                 if (dir)
 795                         rcu_assign_pointer(pn->right, ln);
 796                 else
 797                         rcu_assign_pointer(pn->left, ln);
 798         }
 799         return ln;
 800 }
 801
 802 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
 803 {
 804         return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
 805                RTF_GATEWAY;
 806 }
 807
 808 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
 809 {
 810         int i;
 811
 812         for (i = 0; i < RTAX_MAX; i++) {
 813                 if (test_bit(i, mxc->mx_valid))
 814                         mp[i] = mxc->mx[i];
 815         }
 816 }
 817
 818 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
 819 {
 820         if (!mxc->mx)
 821                 return 0;
 822
 823         if (dst->flags & DST_HOST) {
 824                 u32 *mp = dst_metrics_write_ptr(dst);
 825
 826                 if (unlikely(!mp))
 827                         return -ENOMEM;
 828
 829                 fib6_copy_metrics(mp, mxc);
 830         } else {
 831                 dst_init_metrics(dst, mxc->mx, false);
 832
 833                 /* We've stolen mx now. */
 834                 mxc->mx = NULL;
 835         }
 836
 837         return 0;
 838 }
 839
 840 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
 841                           struct net *net)
 842 {
 843         struct fib6_table *table = rt->rt6i_table;
 844
 845         if (atomic_read(&rt->rt6i_ref) != 1) {
 846                 /* This route is used as dummy address holder in some split
 847                  * nodes. It is not leaked, but it still holds other resources,
 848                  * which must be released in time. So, scan ascendant nodes
 849                  * and replace dummy references to this route with references
 850                  * to still alive ones.
 851                  */
 852                 while (fn) {
 853                         struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
 854                                             lockdep_is_held(&table->tb6_lock));
 855                         struct rt6_info *new_leaf;
 856                         if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
 857                                 new_leaf = fib6_find_prefix(net, table, fn);
 858                                 atomic_inc(&new_leaf->rt6i_ref);
 859                                 rcu_assign_pointer(fn->leaf, new_leaf);
 860                                 rt6_release(rt);
 861                         }
 862                         fn = rcu_dereference_protected(fn->parent,
 863                                     lockdep_is_held(&table->tb6_lock));
 864                 }
 865         }
 866 }
 867
 868 /*
 869  *      Insert routing information in a node.
 870  */
 871
 872 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
 873                             struct nl_info *info, struct mx6_config *mxc,
 874                             struct netlink_ext_ack *extack)
 875 {
 876         struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
 877                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
 878         struct rt6_info *iter = NULL;
 879         struct rt6_info __rcu **ins;
 880         struct rt6_info __rcu **fallback_ins = NULL;
 881         int replace = (info->nlh &&
 882                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
 883         int add = (!info->nlh ||
 884                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
 885         int found = 0;
 886         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
 887         u16 nlflags = NLM_F_EXCL;
 888         int err;
 889
 890         if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
 891                 nlflags |= NLM_F_APPEND;
 892
 893         ins = &fn->leaf;
 894
 895         for (iter = leaf; iter;
 896              iter = rcu_dereference_protected(iter->dst.rt6_next,
 897                                 lockdep_is_held(&rt->rt6i_table->tb6_lock))) {
 898                 /*
 899                  *      Search for duplicates
 900                  */
 901
 902                 if (iter->rt6i_metric == rt->rt6i_metric) {
 903                         /*
 904                          *      Same priority level
 905                          */
 906                         if (info->nlh &&
 907                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
 908                                 return -EEXIST;
 909
 910                         nlflags &= ~NLM_F_EXCL;
 911                         if (replace) {
 912                                 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
 913                                         found++;
 914                                         break;
 915                                 }
 916                                 if (rt_can_ecmp)
 917                                         fallback_ins = fallback_ins ?: ins;
 918                                 goto next_iter;
 919                         }
 920
 921                         if (rt6_duplicate_nexthop(iter, rt)) {
 922                                 if (rt->rt6i_nsiblings)
 923                                         rt->rt6i_nsiblings = 0;
 924                                 if (!(iter->rt6i_flags & RTF_EXPIRES))
 925                                         return -EEXIST;
 926                                 if (!(rt->rt6i_flags & RTF_EXPIRES))
 927                                         rt6_clean_expires(iter);
 928                                 else
 929                                         rt6_set_expires(iter, rt->dst.expires);
 930                                 iter->rt6i_pmtu = rt->rt6i_pmtu;
 931                                 return -EEXIST;
 932                         }
 933                         /* If we have the same destination and the same metric,
 934                          * but not the same gateway, then the route we try to
 935                          * add is sibling to this route, increment our counter
 936                          * of siblings, and later we will add our route to the
 937                          * list.
 938                          * Only static routes (which don't have flag
 939                          * RTF_EXPIRES) are used for ECMPv6.
 940                          *
 941                          * To avoid long list, we only had siblings if the
 942                          * route have a gateway.
 943                          */
 944                         if (rt_can_ecmp &&
 945                             rt6_qualify_for_ecmp(iter))
 946                                 rt->rt6i_nsiblings++;
 947                 }
 948
 949                 if (iter->rt6i_metric > rt->rt6i_metric)
 950                         break;
 951
 952 next_iter:
 953                 ins = &iter->dst.rt6_next;
 954         }
 955
 956         if (fallback_ins && !found) {
 957                 /* No ECMP-able route found, replace first non-ECMP one */
 958                 ins = fallback_ins;
 959                 iter = rcu_dereference_protected(*ins,
 960                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
 961                 found++;
 962         }
 963
 964         /* Reset round-robin state, if necessary */
 965         if (ins == &fn->leaf)
 966                 fn->rr_ptr = NULL;
 967
 968         /* Link this route to others same route. */
 969         if (rt->rt6i_nsiblings) {
 970                 unsigned int rt6i_nsiblings;
 971                 struct rt6_info *sibling, *temp_sibling;
 972
 973                 /* Find the first route that have the same metric */
 974                 sibling = leaf;
 975                 while (sibling) {
 976                         if (sibling->rt6i_metric == rt->rt6i_metric &&
 977                             rt6_qualify_for_ecmp(sibling)) {
 978                                 list_add_tail(&rt->rt6i_siblings,
 979                                               &sibling->rt6i_siblings);
 980                                 break;
 981                         }
 982                         sibling = rcu_dereference_protected(sibling->dst.rt6_next,
 983                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
 984                 }
 985                 /* For each sibling in the list, increment the counter of
 986                  * siblings. BUG() if counters does not match, list of siblings
 987                  * is broken!
 988                  */
 989                 rt6i_nsiblings = 0;
 990                 list_for_each_entry_safe(sibling, temp_sibling,
 991                                          &rt->rt6i_siblings, rt6i_siblings) {
 992                         sibling->rt6i_nsiblings++;
 993                         BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
 994                         rt6i_nsiblings++;
 995                 }
 996                 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
 997         }
 998
 999         /*
1000          *      insert node
1001          */
1002         if (!replace) {
1003                 if (!add)
1004                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
1005
1006 add:
1007                 nlflags |= NLM_F_CREATE;
1008                 err = fib6_commit_metrics(&rt->dst, mxc);
1009                 if (err)
1010                         return err;
1011
1012                 rcu_assign_pointer(rt->dst.rt6_next, iter);
1013                 atomic_inc(&rt->rt6i_ref);
1014                 rcu_assign_pointer(rt->rt6i_node, fn);
1015                 rcu_assign_pointer(*ins, rt);
1016                 call_fib6_entry_notifiers(info->nl_net, FIB_EVENT_ENTRY_ADD,
1017                                           rt, extack);
1018                 if (!info->skip_notify)
1019                         inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1020                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1021
1022                 if (!(fn->fn_flags & RTN_RTINFO)) {
1023                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1024                         fn->fn_flags |= RTN_RTINFO;
1025                 }
1026
1027         } else {
1028                 int nsiblings;
1029
1030                 if (!found) {
1031                         if (add)
1032                                 goto add;
1033                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1034                         return -ENOENT;
1035                 }
1036
1037                 err = fib6_commit_metrics(&rt->dst, mxc);
1038                 if (err)
1039                         return err;
1040
1041                 atomic_inc(&rt->rt6i_ref);
1042                 rcu_assign_pointer(rt->rt6i_node, fn);
1043                 rt->dst.rt6_next = iter->dst.rt6_next;
1044                 rcu_assign_pointer(*ins, rt);
1045                 call_fib6_entry_notifiers(info->nl_net, FIB_EVENT_ENTRY_REPLACE,
1046                                           rt, extack);
1047                 if (!info->skip_notify)
1048                         inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1049                 if (!(fn->fn_flags & RTN_RTINFO)) {
1050                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1051                         fn->fn_flags |= RTN_RTINFO;
1052                 }
1053                 nsiblings = iter->rt6i_nsiblings;
1054                 iter->rt6i_node = NULL;
1055                 fib6_purge_rt(iter, fn, info->nl_net);
1056                 if (rcu_access_pointer(fn->rr_ptr) == iter)
1057                         fn->rr_ptr = NULL;
1058                 rt6_release(iter);
1059
1060                 if (nsiblings) {
1061                         /* Replacing an ECMP route, remove all siblings */
1062                         ins = &rt->dst.rt6_next;
1063                         iter = rcu_dereference_protected(*ins,
1064                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
1065                         while (iter) {
1066                                 if (iter->rt6i_metric > rt->rt6i_metric)
1067                                         break;
1068                                 if (rt6_qualify_for_ecmp(iter)) {
1069                                         *ins = iter->dst.rt6_next;
1070                                         iter->rt6i_node = NULL;
1071                                         fib6_purge_rt(iter, fn, info->nl_net);
1072                                         if (rcu_access_pointer(fn->rr_ptr) == iter)
1073                                                 fn->rr_ptr = NULL;
1074                                         rt6_release(iter);
1075                                         nsiblings--;
1076                                         info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1077                                 } else {
1078                                         ins = &iter->dst.rt6_next;
1079                                 }
1080                                 iter = rcu_dereference_protected(*ins,
1081                                         lockdep_is_held(&rt->rt6i_table->tb6_lock));
1082                         }
1083                         WARN_ON(nsiblings != 0);
1084                 }
1085         }
1086
1087         return 0;
1088 }
1089
1090 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
1091 {
1092         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1093             (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
1094                 mod_timer(&net->ipv6.ip6_fib_timer,
1095                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1096 }
1097
1098 void fib6_force_start_gc(struct net *net)
1099 {
1100         if (!timer_pending(&net->ipv6.ip6_fib_timer))
1101                 mod_timer(&net->ipv6.ip6_fib_timer,
1102                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1103 }
1104
1105 static void fib6_update_sernum_upto_root(struct rt6_info *rt,
1106                                          int sernum)
1107 {
1108         struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1109                                 lockdep_is_held(&rt->rt6i_table->tb6_lock));
1110
1111         /* paired with smp_rmb() in rt6_get_cookie_safe() */
1112         smp_wmb();
1113         while (fn) {
1114                 fn->fn_sernum = sernum;
1115                 fn = rcu_dereference_protected(fn->parent,
1116                                 lockdep_is_held(&rt->rt6i_table->tb6_lock));
1117         }
1118 }
1119
1120 /*
1121  *      Add routing information to the routing tree.
1122  *      <destination addr>/<source addr>
1123  *      with source addr info in sub-trees
1124  *      Need to own table->tb6_lock
1125  */
1126
1127 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
1128              struct nl_info *info, struct mx6_config *mxc,
1129              struct netlink_ext_ack *extack)
1130 {
1131         struct fib6_table *table = rt->rt6i_table;
1132         struct fib6_node *fn, *pn = NULL;
1133         int err = -ENOMEM;
1134         int allow_create = 1;
1135         int replace_required = 0;
1136         int sernum = fib6_new_sernum(info->nl_net);
1137
1138         if (WARN_ON_ONCE(!atomic_read(&rt->dst.__refcnt)))
1139                 return -EINVAL;
1140         if (WARN_ON_ONCE(rt->rt6i_flags & RTF_CACHE))
1141                 return -EINVAL;
1142
1143         if (info->nlh) {
1144                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1145                         allow_create = 0;
1146                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1147                         replace_required = 1;
1148         }
1149         if (!allow_create && !replace_required)
1150                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1151
1152         fn = fib6_add_1(info->nl_net, table, root,
1153                         &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
1154                         offsetof(struct rt6_info, rt6i_dst), allow_create,
1155                         replace_required, extack);
1156         if (IS_ERR(fn)) {
1157                 err = PTR_ERR(fn);
1158                 fn = NULL;
1159                 goto out;
1160         }
1161
1162         pn = fn;
1163
1164 #ifdef CONFIG_IPV6_SUBTREES
1165         if (rt->rt6i_src.plen) {
1166                 struct fib6_node *sn;
1167
1168                 if (!rcu_access_pointer(fn->subtree)) {
1169                         struct fib6_node *sfn;
1170
1171                         /*
1172                          * Create subtree.
1173                          *
1174                          *              fn[main tree]
1175                          *              |
1176                          *              sfn[subtree root]
1177                          *                 \
1178                          *                  sn[new leaf node]
1179                          */
1180
1181                         /* Create subtree root node */
1182                         sfn = node_alloc(info->nl_net);
1183                         if (!sfn)
1184                                 goto failure;
1185
1186                         atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1187                         rcu_assign_pointer(sfn->leaf,
1188                                            info->nl_net->ipv6.ip6_null_entry);
1189                         sfn->fn_flags = RTN_ROOT;
1190
1191                         /* Now add the first leaf node to new subtree */
1192
1193                         sn = fib6_add_1(info->nl_net, table, sfn,
1194                                         &rt->rt6i_src.addr, rt->rt6i_src.plen,
1195                                         offsetof(struct rt6_info, rt6i_src),
1196                                         allow_create, replace_required, extack);
1197
1198                         if (IS_ERR(sn)) {
1199                                 /* If it is failed, discard just allocated
1200                                    root, and then (in failure) stale node
1201                                    in main tree.
1202                                  */
1203                                 node_free_immediate(info->nl_net, sfn);
1204                                 err = PTR_ERR(sn);
1205                                 goto failure;
1206                         }
1207
1208                         /* Now link new subtree to main tree */
1209                         rcu_assign_pointer(sfn->parent, fn);
1210                         rcu_assign_pointer(fn->subtree, sfn);
1211                 } else {
1212                         sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1213                                         &rt->rt6i_src.addr, rt->rt6i_src.plen,
1214                                         offsetof(struct rt6_info, rt6i_src),
1215                                         allow_create, replace_required, extack);
1216
1217                         if (IS_ERR(sn)) {
1218                                 err = PTR_ERR(sn);
1219                                 goto failure;
1220                         }
1221                 }
1222
1223                 if (!rcu_access_pointer(fn->leaf)) {
1224                         atomic_inc(&rt->rt6i_ref);
1225                         rcu_assign_pointer(fn->leaf, rt);
1226                 }
1227                 fn = sn;
1228         }
1229 #endif
1230
1231         err = fib6_add_rt2node(fn, rt, info, mxc, extack);
1232         if (!err) {
1233                 fib6_update_sernum_upto_root(rt, sernum);
1234                 fib6_start_gc(info->nl_net, rt);
1235         }
1236
1237 out:
1238         if (err) {
1239 #ifdef CONFIG_IPV6_SUBTREES
1240                 /*
1241                  * If fib6_add_1 has cleared the old leaf pointer in the
1242                  * super-tree leaf node we have to find a new one for it.
1243                  */
1244                 struct rt6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1245                                             lockdep_is_held(&table->tb6_lock));
1246                 if (pn != fn && pn_leaf == rt) {
1247                         pn_leaf = NULL;
1248                         RCU_INIT_POINTER(pn->leaf, NULL);
1249                         atomic_dec(&rt->rt6i_ref);
1250                 }
1251                 if (pn != fn && !pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1252                         pn_leaf = fib6_find_prefix(info->nl_net, table, pn);
1253 #if RT6_DEBUG >= 2
1254                         if (!pn_leaf) {
1255                                 WARN_ON(!pn_leaf);
1256                                 pn_leaf = info->nl_net->ipv6.ip6_null_entry;
1257                         }
1258 #endif
1259                         atomic_inc(&pn_leaf->rt6i_ref);
1260                         rcu_assign_pointer(pn->leaf, pn_leaf);
1261                 }
1262 #endif
1263                 goto failure;
1264         }
1265         return err;
1266
1267 failure:
1268         /* fn->leaf could be NULL if fn is an intermediate node and we
1269          * failed to add the new route to it in both subtree creation
1270          * failure and fib6_add_rt2node() failure case.
1271          * In both cases, fib6_repair_tree() should be called to fix
1272          * fn->leaf.
1273          */
1274         if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
1275                 fib6_repair_tree(info->nl_net, table, fn);
1276         /* Always release dst as dst->__refcnt is guaranteed
1277          * to be taken before entering this function
1278          */
1279         dst_release_immediate(&rt->dst);
1280         return err;
1281 }
1282
1283 /*
1284  *      Routing tree lookup
1285  *
1286  */
1287
1288 struct lookup_args {
1289         int                     offset;         /* key offset on rt6_info       */
1290         const struct in6_addr   *addr;          /* search key                   */
1291 };
1292
1293 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1294                                        struct lookup_args *args)
1295 {
1296         struct fib6_node *fn;
1297         __be32 dir;
1298
1299         if (unlikely(args->offset == 0))
1300                 return NULL;
1301
1302         /*
1303          *      Descend on a tree
1304          */
1305
1306         fn = root;
1307
1308         for (;;) {
1309                 struct fib6_node *next;
1310
1311                 dir = addr_bit_set(args->addr, fn->fn_bit);
1312
1313                 next = dir ? rcu_dereference(fn->right) :
1314                              rcu_dereference(fn->left);
1315
1316                 if (next) {
1317                         fn = next;
1318                         continue;
1319                 }
1320                 break;
1321         }
1322
1323         while (fn) {
1324                 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1325
1326                 if (subtree || fn->fn_flags & RTN_RTINFO) {
1327                         struct rt6_info *leaf = rcu_dereference(fn->leaf);
1328                         struct rt6key *key;
1329
1330                         if (!leaf)
1331                                 goto backtrack;
1332
1333                         key = (struct rt6key *) ((u8 *)leaf + args->offset);
1334
1335                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1336 #ifdef CONFIG_IPV6_SUBTREES
1337                                 if (subtree) {
1338                                         struct fib6_node *sfn;
1339                                         sfn = fib6_lookup_1(subtree, args + 1);
1340                                         if (!sfn)
1341                                                 goto backtrack;
1342                                         fn = sfn;
1343                                 }
1344 #endif
1345                                 if (fn->fn_flags & RTN_RTINFO)
1346                                         return fn;
1347                         }
1348                 }
1349 backtrack:
1350                 if (fn->fn_flags & RTN_ROOT)
1351                         break;
1352
1353                 fn = rcu_dereference(fn->parent);
1354         }
1355
1356         return NULL;
1357 }
1358
1359 /* called with rcu_read_lock() held
1360  */
1361 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1362                               const struct in6_addr *saddr)
1363 {
1364         struct fib6_node *fn;
1365         struct lookup_args args[] = {
1366                 {
1367                         .offset = offsetof(struct rt6_info, rt6i_dst),
1368                         .addr = daddr,
1369                 },
1370 #ifdef CONFIG_IPV6_SUBTREES
1371                 {
1372                         .offset = offsetof(struct rt6_info, rt6i_src),
1373                         .addr = saddr,
1374                 },
1375 #endif
1376                 {
1377                         .offset = 0,    /* sentinel */
1378                 }
1379         };
1380
1381         fn = fib6_lookup_1(root, daddr ? args : args + 1);
1382         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1383                 fn = root;
1384
1385         return fn;
1386 }
1387
1388 /*
1389  *      Get node with specified destination prefix (and source prefix,
1390  *      if subtrees are used)
1391  *      exact_match == true means we try to find fn with exact match of
1392  *      the passed in prefix addr
1393  *      exact_match == false means we try to find fn with longest prefix
1394  *      match of the passed in prefix addr. This is useful for finding fn
1395  *      for cached route as it will be stored in the exception table under
1396  *      the node with longest prefix length.
1397  */
1398
1399
1400 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1401                                        const struct in6_addr *addr,
1402                                        int plen, int offset,
1403                                        bool exact_match)
1404 {
1405         struct fib6_node *fn, *prev = NULL;
1406
1407         for (fn = root; fn ; ) {
1408                 struct rt6_info *leaf = rcu_dereference(fn->leaf);
1409                 struct rt6key *key;
1410
1411                 /* This node is being deleted */
1412                 if (!leaf) {
1413                         if (plen <= fn->fn_bit)
1414                                 goto out;
1415                         else
1416                                 goto next;
1417                 }
1418
1419                 key = (struct rt6key *)((u8 *)leaf + offset);
1420
1421                 /*
1422                  *      Prefix match
1423                  */
1424                 if (plen < fn->fn_bit ||
1425                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1426                         goto out;
1427
1428                 if (plen == fn->fn_bit)
1429                         return fn;
1430
1431                 prev = fn;
1432
1433 next:
1434                 /*
1435                  *      We have more bits to go
1436                  */
1437                 if (addr_bit_set(addr, fn->fn_bit))
1438                         fn = rcu_dereference(fn->right);
1439                 else
1440                         fn = rcu_dereference(fn->left);
1441         }
1442 out:
1443         if (exact_match)
1444                 return NULL;
1445         else
1446                 return prev;
1447 }
1448
1449 struct fib6_node *fib6_locate(struct fib6_node *root,
1450                               const struct in6_addr *daddr, int dst_len,
1451                               const struct in6_addr *saddr, int src_len,
1452                               bool exact_match)
1453 {
1454         struct fib6_node *fn;
1455
1456         fn = fib6_locate_1(root, daddr, dst_len,
1457                            offsetof(struct rt6_info, rt6i_dst),
1458                            exact_match);
1459
1460 #ifdef CONFIG_IPV6_SUBTREES
1461         if (src_len) {
1462                 WARN_ON(saddr == NULL);
1463                 if (fn) {
1464                         struct fib6_node *subtree = FIB6_SUBTREE(fn);
1465
1466                         if (subtree) {
1467                                 fn = fib6_locate_1(subtree, saddr, src_len,
1468                                            offsetof(struct rt6_info, rt6i_src),
1469                                            exact_match);
1470                         }
1471                 }
1472         }
1473 #endif
1474
1475         if (fn && fn->fn_flags & RTN_RTINFO)
1476                 return fn;
1477
1478         return NULL;
1479 }
1480
1481
1482 /*
1483  *      Deletion
1484  *
1485  */
1486
1487 static struct rt6_info *fib6_find_prefix(struct net *net,
1488                                          struct fib6_table *table,
1489                                          struct fib6_node *fn)
1490 {
1491         struct fib6_node *child_left, *child_right;
1492
1493         if (fn->fn_flags & RTN_ROOT)
1494                 return net->ipv6.ip6_null_entry;
1495
1496         while (fn) {
1497                 child_left = rcu_dereference_protected(fn->left,
1498                                     lockdep_is_held(&table->tb6_lock));
1499                 child_right = rcu_dereference_protected(fn->right,
1500                                     lockdep_is_held(&table->tb6_lock));
1501                 if (child_left)
1502                         return rcu_dereference_protected(child_left->leaf,
1503                                         lockdep_is_held(&table->tb6_lock));
1504                 if (child_right)
1505                         return rcu_dereference_protected(child_right->leaf,
1506                                         lockdep_is_held(&table->tb6_lock));
1507
1508                 fn = FIB6_SUBTREE(fn);
1509         }
1510         return NULL;
1511 }
1512
1513 /*
1514  *      Called to trim the tree of intermediate nodes when possible. "fn"
1515  *      is the node we want to try and remove.
1516  *      Need to own table->tb6_lock
1517  */
1518
1519 static struct fib6_node *fib6_repair_tree(struct net *net,
1520                                           struct fib6_table *table,
1521                                           struct fib6_node *fn)
1522 {
1523         int children;
1524         int nstate;
1525         struct fib6_node *child;
1526         struct fib6_walker *w;
1527         int iter = 0;
1528
1529         for (;;) {
1530                 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1531                                             lockdep_is_held(&table->tb6_lock));
1532                 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1533                                             lockdep_is_held(&table->tb6_lock));
1534                 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1535                                             lockdep_is_held(&table->tb6_lock));
1536                 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1537                                             lockdep_is_held(&table->tb6_lock));
1538                 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1539                                             lockdep_is_held(&table->tb6_lock));
1540                 struct rt6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1541                                             lockdep_is_held(&table->tb6_lock));
1542                 struct rt6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1543                                             lockdep_is_held(&table->tb6_lock));
1544                 struct rt6_info *new_fn_leaf;
1545
1546                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1547                 iter++;
1548
1549                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1550                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1551                 WARN_ON(fn_leaf);
1552
1553                 children = 0;
1554                 child = NULL;
1555                 if (fn_r)
1556                         child = fn_r, children |= 1;
1557                 if (fn_l)
1558                         child = fn_l, children |= 2;
1559
1560                 if (children == 3 || FIB6_SUBTREE(fn)
1561 #ifdef CONFIG_IPV6_SUBTREES
1562                     /* Subtree root (i.e. fn) may have one child */
1563                     || (children && fn->fn_flags & RTN_ROOT)
1564 #endif
1565                     ) {
1566                         new_fn_leaf = fib6_find_prefix(net, table, fn);
1567 #if RT6_DEBUG >= 2
1568                         if (!new_fn_leaf) {
1569                                 WARN_ON(!new_fn_leaf);
1570                                 new_fn_leaf = net->ipv6.ip6_null_entry;
1571                         }
1572 #endif
1573                         atomic_inc(&new_fn_leaf->rt6i_ref);
1574                         rcu_assign_pointer(fn->leaf, new_fn_leaf);
1575                         return pn;
1576                 }
1577
1578 #ifdef CONFIG_IPV6_SUBTREES
1579                 if (FIB6_SUBTREE(pn) == fn) {
1580                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1581                         RCU_INIT_POINTER(pn->subtree, NULL);
1582                         nstate = FWS_L;
1583                 } else {
1584                         WARN_ON(fn->fn_flags & RTN_ROOT);
1585 #endif
1586                         if (pn_r == fn)
1587                                 rcu_assign_pointer(pn->right, child);
1588                         else if (pn_l == fn)
1589                                 rcu_assign_pointer(pn->left, child);
1590 #if RT6_DEBUG >= 2
1591                         else
1592                                 WARN_ON(1);
1593 #endif
1594                         if (child)
1595                                 rcu_assign_pointer(child->parent, pn);
1596                         nstate = FWS_R;
1597 #ifdef CONFIG_IPV6_SUBTREES
1598                 }
1599 #endif
1600
1601                 read_lock(&net->ipv6.fib6_walker_lock);
1602                 FOR_WALKERS(net, w) {
1603                         if (!child) {
1604                                 if (w->node == fn) {
1605                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1606                                         w->node = pn;
1607                                         w->state = nstate;
1608                                 }
1609                         } else {
1610                                 if (w->node == fn) {
1611                                         w->node = child;
1612                                         if (children&2) {
1613                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1614                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1615                                         } else {
1616                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1617                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1618                                         }
1619                                 }
1620                         }
1621                 }
1622                 read_unlock(&net->ipv6.fib6_walker_lock);
1623
1624                 node_free(net, fn);
1625                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1626                         return pn;
1627
1628                 RCU_INIT_POINTER(pn->leaf, NULL);
1629                 rt6_release(pn_leaf);
1630                 fn = pn;
1631         }
1632 }
1633
1634 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1635                            struct rt6_info __rcu **rtp, struct nl_info *info)
1636 {
1637         struct fib6_walker *w;
1638         struct rt6_info *rt = rcu_dereference_protected(*rtp,
1639                                     lockdep_is_held(&table->tb6_lock));
1640         struct net *net = info->nl_net;
1641
1642         RT6_TRACE("fib6_del_route\n");
1643
1644         WARN_ON_ONCE(rt->rt6i_flags & RTF_CACHE);
1645
1646         /* Unlink it */
1647         *rtp = rt->dst.rt6_next;
1648         rt->rt6i_node = NULL;
1649         net->ipv6.rt6_stats->fib_rt_entries--;
1650         net->ipv6.rt6_stats->fib_discarded_routes++;
1651
1652         /* Flush all cached dst in exception table */
1653         rt6_flush_exceptions(rt);
1654
1655         /* Reset round-robin state, if necessary */
1656         if (rcu_access_pointer(fn->rr_ptr) == rt)
1657                 fn->rr_ptr = NULL;
1658
1659         /* Remove this entry from other siblings */
1660         if (rt->rt6i_nsiblings) {
1661                 struct rt6_info *sibling, *next_sibling;
1662
1663                 list_for_each_entry_safe(sibling, next_sibling,
1664                                          &rt->rt6i_siblings, rt6i_siblings)
1665                         sibling->rt6i_nsiblings--;
1666                 rt->rt6i_nsiblings = 0;
1667                 list_del_init(&rt->rt6i_siblings);
1668         }
1669
1670         /* Adjust walkers */
1671         read_lock(&net->ipv6.fib6_walker_lock);
1672         FOR_WALKERS(net, w) {
1673                 if (w->state == FWS_C && w->leaf == rt) {
1674                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1675                         w->leaf = rcu_dereference_protected(rt->dst.rt6_next,
1676                                             lockdep_is_held(&table->tb6_lock));
1677                         if (!w->leaf)
1678                                 w->state = FWS_U;
1679                 }
1680         }
1681         read_unlock(&net->ipv6.fib6_walker_lock);
1682
1683         /* If it was last route, expunge its radix tree node */
1684         if (!rcu_access_pointer(fn->leaf)) {
1685                 fn->fn_flags &= ~RTN_RTINFO;
1686                 net->ipv6.rt6_stats->fib_route_nodes--;
1687                 fn = fib6_repair_tree(net, table, fn);
1688         }
1689
1690         fib6_purge_rt(rt, fn, net);
1691
1692         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1693         if (!info->skip_notify)
1694                 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1695         rt6_release(rt);
1696 }
1697
1698 /* Need to own table->tb6_lock */
1699 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1700 {
1701         struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1702                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
1703         struct fib6_table *table = rt->rt6i_table;
1704         struct net *net = info->nl_net;
1705         struct rt6_info __rcu **rtp;
1706         struct rt6_info __rcu **rtp_next;
1707
1708 #if RT6_DEBUG >= 2
1709         if (rt->dst.obsolete > 0) {
1710                 WARN_ON(fn);
1711                 return -ENOENT;
1712         }
1713 #endif
1714         if (!fn || rt == net->ipv6.ip6_null_entry)
1715                 return -ENOENT;
1716
1717         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1718
1719         /* remove cached dst from exception table */
1720         if (rt->rt6i_flags & RTF_CACHE)
1721                 return rt6_remove_exception_rt(rt);
1722
1723         /*
1724          *      Walk the leaf entries looking for ourself
1725          */
1726
1727         for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1728                 struct rt6_info *cur = rcu_dereference_protected(*rtp,
1729                                         lockdep_is_held(&table->tb6_lock));
1730                 if (rt == cur) {
1731                         fib6_del_route(table, fn, rtp, info);
1732                         return 0;
1733                 }
1734                 rtp_next = &cur->dst.rt6_next;
1735         }
1736         return -ENOENT;
1737 }
1738
1739 /*
1740  *      Tree traversal function.
1741  *
1742  *      Certainly, it is not interrupt safe.
1743  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1744  *      It means, that we can modify tree during walking
1745  *      and use this function for garbage collection, clone pruning,
1746  *      cleaning tree when a device goes down etc. etc.
1747  *
1748  *      It guarantees that every node will be traversed,
1749  *      and that it will be traversed only once.
1750  *
1751  *      Callback function w->func may return:
1752  *      0 -> continue walking.
1753  *      positive value -> walking is suspended (used by tree dumps,
1754  *      and probably by gc, if it will be split to several slices)
1755  *      negative value -> terminate walking.
1756  *
1757  *      The function itself returns:
1758  *      0   -> walk is complete.
1759  *      >0  -> walk is incomplete (i.e. suspended)
1760  *      <0  -> walk is terminated by an error.
1761  *
1762  *      This function is called with tb6_lock held.
1763  */
1764
1765 static int fib6_walk_continue(struct fib6_walker *w)
1766 {
1767         struct fib6_node *fn, *pn, *left, *right;
1768
1769         /* w->root should always be table->tb6_root */
1770         WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1771
1772         for (;;) {
1773                 fn = w->node;
1774                 if (!fn)
1775                         return 0;
1776
1777                 switch (w->state) {
1778 #ifdef CONFIG_IPV6_SUBTREES
1779                 case FWS_S:
1780                         if (FIB6_SUBTREE(fn)) {
1781                                 w->node = FIB6_SUBTREE(fn);
1782                                 continue;
1783                         }
1784                         w->state = FWS_L;
1785 #endif
1786                         /* fall through */
1787                 case FWS_L:
1788                         left = rcu_dereference_protected(fn->left, 1);
1789                         if (left) {
1790                                 w->node = left;
1791                                 w->state = FWS_INIT;
1792                                 continue;
1793                         }
1794                         w->state = FWS_R;
1795                         /* fall through */
1796                 case FWS_R:
1797                         right = rcu_dereference_protected(fn->right, 1);
1798                         if (right) {
1799                                 w->node = right;
1800                                 w->state = FWS_INIT;
1801                                 continue;
1802                         }
1803                         w->state = FWS_C;
1804                         w->leaf = rcu_dereference_protected(fn->leaf, 1);
1805                         /* fall through */
1806                 case FWS_C:
1807                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1808                                 int err;
1809
1810                                 if (w->skip) {
1811                                         w->skip--;
1812                                         goto skip;
1813                                 }
1814
1815                                 err = w->func(w);
1816                                 if (err)
1817                                         return err;
1818
1819                                 w->count++;
1820                                 continue;
1821                         }
1822 skip:
1823                         w->state = FWS_U;
1824                         /* fall through */
1825                 case FWS_U:
1826                         if (fn == w->root)
1827                                 return 0;
1828                         pn = rcu_dereference_protected(fn->parent, 1);
1829                         left = rcu_dereference_protected(pn->left, 1);
1830                         right = rcu_dereference_protected(pn->right, 1);
1831                         w->node = pn;
1832 #ifdef CONFIG_IPV6_SUBTREES
1833                         if (FIB6_SUBTREE(pn) == fn) {
1834                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1835                                 w->state = FWS_L;
1836                                 continue;
1837                         }
1838 #endif
1839                         if (left == fn) {
1840                                 w->state = FWS_R;
1841                                 continue;
1842                         }
1843                         if (right == fn) {
1844                                 w->state = FWS_C;
1845                                 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1846                                 continue;
1847                         }
1848 #if RT6_DEBUG >= 2
1849                         WARN_ON(1);
1850 #endif
1851                 }
1852         }
1853 }
1854
1855 static int fib6_walk(struct net *net, struct fib6_walker *w)
1856 {
1857         int res;
1858
1859         w->state = FWS_INIT;
1860         w->node = w->root;
1861
1862         fib6_walker_link(net, w);
1863         res = fib6_walk_continue(w);
1864         if (res <= 0)
1865                 fib6_walker_unlink(net, w);
1866         return res;
1867 }
1868
1869 static int fib6_clean_node(struct fib6_walker *w)
1870 {
1871         int res;
1872         struct rt6_info *rt;
1873         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1874         struct nl_info info = {
1875                 .nl_net = c->net,
1876         };
1877
1878         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1879             w->node->fn_sernum != c->sernum)
1880                 w->node->fn_sernum = c->sernum;
1881
1882         if (!c->func) {
1883                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1884                 w->leaf = NULL;
1885                 return 0;
1886         }
1887
1888         for_each_fib6_walker_rt(w) {
1889                 res = c->func(rt, c->arg);
1890                 if (res < 0) {
1891                         w->leaf = rt;
1892                         res = fib6_del(rt, &info);
1893                         if (res) {
1894 #if RT6_DEBUG >= 2
1895                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1896                                          __func__, rt,
1897                                          rcu_access_pointer(rt->rt6i_node),
1898                                          res);
1899 #endif
1900                                 continue;
1901                         }
1902                         return 0;
1903                 }
1904                 WARN_ON(res != 0);
1905         }
1906         w->leaf = rt;
1907         return 0;
1908 }
1909
1910 /*
1911  *      Convenient frontend to tree walker.
1912  *
1913  *      func is called on each route.
1914  *              It may return -1 -> delete this route.
1915  *                            0  -> continue walking
1916  */
1917
1918 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1919                             int (*func)(struct rt6_info *, void *arg),
1920                             int sernum, void *arg)
1921 {
1922         struct fib6_cleaner c;
1923
1924         c.w.root = root;
1925         c.w.func = fib6_clean_node;
1926         c.w.count = 0;
1927         c.w.skip = 0;
1928         c.func = func;
1929         c.sernum = sernum;
1930         c.arg = arg;
1931         c.net = net;
1932
1933         fib6_walk(net, &c.w);
1934 }
1935
1936 static void __fib6_clean_all(struct net *net,
1937                              int (*func)(struct rt6_info *, void *),
1938                              int sernum, void *arg)
1939 {
1940         struct fib6_table *table;
1941         struct hlist_head *head;
1942         unsigned int h;
1943
1944         rcu_read_lock();
1945         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1946                 head = &net->ipv6.fib_table_hash[h];
1947                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1948                         spin_lock_bh(&table->tb6_lock);
1949                         fib6_clean_tree(net, &table->tb6_root,
1950                                         func, sernum, arg);
1951                         spin_unlock_bh(&table->tb6_lock);
1952                 }
1953         }
1954         rcu_read_unlock();
1955 }
1956
1957 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1958                     void *arg)
1959 {
1960         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1961 }
1962
1963 static void fib6_flush_trees(struct net *net)
1964 {
1965         int new_sernum = fib6_new_sernum(net);
1966
1967         __fib6_clean_all(net, NULL, new_sernum, NULL);
1968 }
1969
1970 /*
1971  *      Garbage collection
1972  */
1973
1974 static int fib6_age(struct rt6_info *rt, void *arg)
1975 {
1976         struct fib6_gc_args *gc_args = arg;
1977         unsigned long now = jiffies;
1978
1979         /*
1980          *      check addrconf expiration here.
1981          *      Routes are expired even if they are in use.
1982          */
1983
1984         if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
1985                 if (time_after(now, rt->dst.expires)) {
1986                         RT6_TRACE("expiring %p\n", rt);
1987                         return -1;
1988                 }
1989                 gc_args->more++;
1990         }
1991
1992         /*      Also age clones in the exception table.
1993          *      Note, that clones are aged out
1994          *      only if they are not in use now.
1995          */
1996         rt6_age_exceptions(rt, gc_args, now);
1997
1998         return 0;
1999 }
2000
2001 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2002 {
2003         struct fib6_gc_args gc_args;
2004         unsigned long now;
2005
2006         if (force) {
2007                 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2008         } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2009                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2010                 return;
2011         }
2012         gc_args.timeout = expires ? (int)expires :
2013                           net->ipv6.sysctl.ip6_rt_gc_interval;
2014         gc_args.more = 0;
2015
2016         fib6_clean_all(net, fib6_age, &gc_args);
2017         now = jiffies;
2018         net->ipv6.ip6_rt_last_gc = now;
2019
2020         if (gc_args.more)
2021                 mod_timer(&net->ipv6.ip6_fib_timer,
2022                           round_jiffies(now
2023                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
2024         else
2025                 del_timer(&net->ipv6.ip6_fib_timer);
2026         spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2027 }
2028
2029 static void fib6_gc_timer_cb(unsigned long arg)
2030 {
2031         fib6_run_gc(0, (struct net *)arg, true);
2032 }
2033
2034 static int __net_init fib6_net_init(struct net *net)
2035 {
2036         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2037         int err;
2038
2039         err = fib6_notifier_init(net);
2040         if (err)
2041                 return err;
2042
2043         spin_lock_init(&net->ipv6.fib6_gc_lock);
2044         rwlock_init(&net->ipv6.fib6_walker_lock);
2045         INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2046         setup_timer(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, (unsigned long)net);
2047
2048         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2049         if (!net->ipv6.rt6_stats)
2050                 goto out_timer;
2051
2052         /* Avoid false sharing : Use at least a full cache line */
2053         size = max_t(size_t, size, L1_CACHE_BYTES);
2054
2055         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2056         if (!net->ipv6.fib_table_hash)
2057                 goto out_rt6_stats;
2058
2059         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2060                                           GFP_KERNEL);
2061         if (!net->ipv6.fib6_main_tbl)
2062                 goto out_fib_table_hash;
2063
2064         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2065         rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2066                            net->ipv6.ip6_null_entry);
2067         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2068                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2069         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2070
2071 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2072         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2073                                            GFP_KERNEL);
2074         if (!net->ipv6.fib6_local_tbl)
2075                 goto out_fib6_main_tbl;
2076         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2077         rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2078                            net->ipv6.ip6_null_entry);
2079         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2080                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2081         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2082 #endif
2083         fib6_tables_init(net);
2084
2085         return 0;
2086
2087 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2088 out_fib6_main_tbl:
2089         kfree(net->ipv6.fib6_main_tbl);
2090 #endif
2091 out_fib_table_hash:
2092         kfree(net->ipv6.fib_table_hash);
2093 out_rt6_stats:
2094         kfree(net->ipv6.rt6_stats);
2095 out_timer:
2096         fib6_notifier_exit(net);
2097         return -ENOMEM;
2098 }
2099
2100 static void fib6_net_exit(struct net *net)
2101 {
2102         unsigned int i;
2103
2104         rt6_ifdown(net, NULL);
2105         del_timer_sync(&net->ipv6.ip6_fib_timer);
2106
2107         for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2108                 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2109                 struct hlist_node *tmp;
2110                 struct fib6_table *tb;
2111
2112                 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2113                         hlist_del(&tb->tb6_hlist);
2114                         fib6_free_table(tb);
2115                 }
2116         }
2117
2118         kfree(net->ipv6.fib_table_hash);
2119         kfree(net->ipv6.rt6_stats);
2120         fib6_notifier_exit(net);
2121 }
2122
2123 static struct pernet_operations fib6_net_ops = {
2124         .init = fib6_net_init,
2125         .exit = fib6_net_exit,
2126 };
2127
2128 int __init fib6_init(void)
2129 {
2130         int ret = -ENOMEM;
2131
2132         fib6_node_kmem = kmem_cache_create("fib6_nodes",
2133                                            sizeof(struct fib6_node),
2134                                            0, SLAB_HWCACHE_ALIGN,
2135                                            NULL);
2136         if (!fib6_node_kmem)
2137                 goto out;
2138
2139         ret = register_pernet_subsys(&fib6_net_ops);
2140         if (ret)
2141                 goto out_kmem_cache_create;
2142
2143         ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
2144                               0);
2145         if (ret)
2146                 goto out_unregister_subsys;
2147
2148         __fib6_flush_trees = fib6_flush_trees;
2149 out:
2150         return ret;
2151
2152 out_unregister_subsys:
2153         unregister_pernet_subsys(&fib6_net_ops);
2154 out_kmem_cache_create:
2155         kmem_cache_destroy(fib6_node_kmem);
2156         goto out;
2157 }
2158
2159 void fib6_gc_cleanup(void)
2160 {
2161         unregister_pernet_subsys(&fib6_net_ops);
2162         kmem_cache_destroy(fib6_node_kmem);
2163 }
2164
2165 #ifdef CONFIG_PROC_FS
2166
2167 struct ipv6_route_iter {
2168         struct seq_net_private p;
2169         struct fib6_walker w;
2170         loff_t skip;
2171         struct fib6_table *tbl;
2172         int sernum;
2173 };
2174
2175 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2176 {
2177         struct rt6_info *rt = v;
2178         struct ipv6_route_iter *iter = seq->private;
2179
2180         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
2181
2182 #ifdef CONFIG_IPV6_SUBTREES
2183         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
2184 #else
2185         seq_puts(seq, "00000000000000000000000000000000 00 ");
2186 #endif
2187         if (rt->rt6i_flags & RTF_GATEWAY)
2188                 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
2189         else
2190                 seq_puts(seq, "00000000000000000000000000000000");
2191
2192         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2193                    rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
2194                    rt->dst.__use, rt->rt6i_flags,
2195                    rt->dst.dev ? rt->dst.dev->name : "");
2196         iter->w.leaf = NULL;
2197         return 0;
2198 }
2199
2200 static int ipv6_route_yield(struct fib6_walker *w)
2201 {
2202         struct ipv6_route_iter *iter = w->args;
2203
2204         if (!iter->skip)
2205                 return 1;
2206
2207         do {
2208                 iter->w.leaf = rcu_dereference_protected(
2209                                 iter->w.leaf->dst.rt6_next,
2210                                 lockdep_is_held(&iter->tbl->tb6_lock));
2211                 iter->skip--;
2212                 if (!iter->skip && iter->w.leaf)
2213                         return 1;
2214         } while (iter->w.leaf);
2215
2216         return 0;
2217 }
2218
2219 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2220                                       struct net *net)
2221 {
2222         memset(&iter->w, 0, sizeof(iter->w));
2223         iter->w.func = ipv6_route_yield;
2224         iter->w.root = &iter->tbl->tb6_root;
2225         iter->w.state = FWS_INIT;
2226         iter->w.node = iter->w.root;
2227         iter->w.args = iter;
2228         iter->sernum = iter->w.root->fn_sernum;
2229         INIT_LIST_HEAD(&iter->w.lh);
2230         fib6_walker_link(net, &iter->w);
2231 }
2232
2233 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2234                                                     struct net *net)
2235 {
2236         unsigned int h;
2237         struct hlist_node *node;
2238
2239         if (tbl) {
2240                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2241                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2242         } else {
2243                 h = 0;
2244                 node = NULL;
2245         }
2246
2247         while (!node && h < FIB6_TABLE_HASHSZ) {
2248                 node = rcu_dereference_bh(
2249                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2250         }
2251         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2252 }
2253
2254 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2255 {
2256         if (iter->sernum != iter->w.root->fn_sernum) {
2257                 iter->sernum = iter->w.root->fn_sernum;
2258                 iter->w.state = FWS_INIT;
2259                 iter->w.node = iter->w.root;
2260                 WARN_ON(iter->w.skip);
2261                 iter->w.skip = iter->w.count;
2262         }
2263 }
2264
2265 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2266 {
2267         int r;
2268         struct rt6_info *n;
2269         struct net *net = seq_file_net(seq);
2270         struct ipv6_route_iter *iter = seq->private;
2271
2272         if (!v)
2273                 goto iter_table;
2274
2275         n = rcu_dereference_bh(((struct rt6_info *)v)->dst.rt6_next);
2276         if (n) {
2277                 ++*pos;
2278                 return n;
2279         }
2280
2281 iter_table:
2282         ipv6_route_check_sernum(iter);
2283         spin_lock_bh(&iter->tbl->tb6_lock);
2284         r = fib6_walk_continue(&iter->w);
2285         spin_unlock_bh(&iter->tbl->tb6_lock);
2286         if (r > 0) {
2287                 if (v)
2288                         ++*pos;
2289                 return iter->w.leaf;
2290         } else if (r < 0) {
2291                 fib6_walker_unlink(net, &iter->w);
2292                 return NULL;
2293         }
2294         fib6_walker_unlink(net, &iter->w);
2295
2296         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2297         if (!iter->tbl)
2298                 return NULL;
2299
2300         ipv6_route_seq_setup_walk(iter, net);
2301         goto iter_table;
2302 }
2303
2304 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2305         __acquires(RCU_BH)
2306 {
2307         struct net *net = seq_file_net(seq);
2308         struct ipv6_route_iter *iter = seq->private;
2309
2310         rcu_read_lock_bh();
2311         iter->tbl = ipv6_route_seq_next_table(NULL, net);
2312         iter->skip = *pos;
2313
2314         if (iter->tbl) {
2315                 ipv6_route_seq_setup_walk(iter, net);
2316                 return ipv6_route_seq_next(seq, NULL, pos);
2317         } else {
2318                 return NULL;
2319         }
2320 }
2321
2322 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2323 {
2324         struct fib6_walker *w = &iter->w;
2325         return w->node && !(w->state == FWS_U && w->node == w->root);
2326 }
2327
2328 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2329         __releases(RCU_BH)
2330 {
2331         struct net *net = seq_file_net(seq);
2332         struct ipv6_route_iter *iter = seq->private;
2333
2334         if (ipv6_route_iter_active(iter))
2335                 fib6_walker_unlink(net, &iter->w);
2336
2337         rcu_read_unlock_bh();
2338 }
2339
2340 static const struct seq_operations ipv6_route_seq_ops = {
2341         .start  = ipv6_route_seq_start,
2342         .next   = ipv6_route_seq_next,
2343         .stop   = ipv6_route_seq_stop,
2344         .show   = ipv6_route_seq_show
2345 };
2346
2347 int ipv6_route_open(struct inode *inode, struct file *file)
2348 {
2349         return seq_open_net(inode, file, &ipv6_route_seq_ops,
2350                             sizeof(struct ipv6_route_iter));
2351 }
2352
2353 #endif /* CONFIG_PROC_FS */