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(struct timer_list *t);
  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                         /* remove null_entry in the root node */
 644                         } else if (fn->fn_flags & RTN_TL_ROOT &&
 645                                    rcu_access_pointer(fn->leaf) ==
 646                                    net->ipv6.ip6_null_entry) {
 647                                 RCU_INIT_POINTER(fn->leaf, NULL);
 648                         }
 649
 650                         return fn;
 651                 }
 652
 653                 /*
 654                  *      We have more bits to go
 655                  */
 656
 657                 /* Try to walk down on tree. */
 658                 dir = addr_bit_set(addr, fn->fn_bit);
 659                 pn = fn;
 660                 fn = dir ?
 661                      rcu_dereference_protected(fn->right,
 662                                         lockdep_is_held(&table->tb6_lock)) :
 663                      rcu_dereference_protected(fn->left,
 664                                         lockdep_is_held(&table->tb6_lock));
 665         } while (fn);
 666
 667         if (!allow_create) {
 668                 /* We should not create new node because
 669                  * NLM_F_REPLACE was specified without NLM_F_CREATE
 670                  * I assume it is safe to require NLM_F_CREATE when
 671                  * REPLACE flag is used! Later we may want to remove the
 672                  * check for replace_required, because according
 673                  * to netlink specification, NLM_F_CREATE
 674                  * MUST be specified if new route is created.
 675                  * That would keep IPv6 consistent with IPv4
 676                  */
 677                 if (replace_required) {
 678                         NL_SET_ERR_MSG(extack,
 679                                        "Can not replace route - no match found");
 680                         pr_warn("Can't replace route, no match found\n");
 681                         return ERR_PTR(-ENOENT);
 682                 }
 683                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 684         }
 685         /*
 686          *      We walked to the bottom of tree.
 687          *      Create new leaf node without children.
 688          */
 689
 690         ln = node_alloc(net);
 691
 692         if (!ln)
 693                 return ERR_PTR(-ENOMEM);
 694         ln->fn_bit = plen;
 695         RCU_INIT_POINTER(ln->parent, pn);
 696
 697         if (dir)
 698                 rcu_assign_pointer(pn->right, ln);
 699         else
 700                 rcu_assign_pointer(pn->left, ln);
 701
 702         return ln;
 703
 704
 705 insert_above:
 706         /*
 707          * split since we don't have a common prefix anymore or
 708          * we have a less significant route.
 709          * we've to insert an intermediate node on the list
 710          * this new node will point to the one we need to create
 711          * and the current
 712          */
 713
 714         pn = rcu_dereference_protected(fn->parent,
 715                                        lockdep_is_held(&table->tb6_lock));
 716
 717         /* find 1st bit in difference between the 2 addrs.
 718
 719            See comment in __ipv6_addr_diff: bit may be an invalid value,
 720            but if it is >= plen, the value is ignored in any case.
 721          */
 722
 723         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
 724
 725         /*
 726          *              (intermediate)[in]
 727          *                /        \
 728          *      (new leaf node)[ln] (old node)[fn]
 729          */
 730         if (plen > bit) {
 731                 in = node_alloc(net);
 732                 ln = node_alloc(net);
 733
 734                 if (!in || !ln) {
 735                         if (in)
 736                                 node_free_immediate(net, in);
 737                         if (ln)
 738                                 node_free_immediate(net, ln);
 739                         return ERR_PTR(-ENOMEM);
 740                 }
 741
 742                 /*
 743                  * new intermediate node.
 744                  * RTN_RTINFO will
 745                  * be off since that an address that chooses one of
 746                  * the branches would not match less specific routes
 747                  * in the other branch
 748                  */
 749
 750                 in->fn_bit = bit;
 751
 752                 RCU_INIT_POINTER(in->parent, pn);
 753                 in->leaf = fn->leaf;
 754                 atomic_inc(&rcu_dereference_protected(in->leaf,
 755                                 lockdep_is_held(&table->tb6_lock))->rt6i_ref);
 756
 757                 /* update parent pointer */
 758                 if (dir)
 759                         rcu_assign_pointer(pn->right, in);
 760                 else
 761                         rcu_assign_pointer(pn->left, in);
 762
 763                 ln->fn_bit = plen;
 764
 765                 RCU_INIT_POINTER(ln->parent, in);
 766                 rcu_assign_pointer(fn->parent, in);
 767
 768                 if (addr_bit_set(addr, bit)) {
 769                         rcu_assign_pointer(in->right, ln);
 770                         rcu_assign_pointer(in->left, fn);
 771                 } else {
 772                         rcu_assign_pointer(in->left, ln);
 773                         rcu_assign_pointer(in->right, fn);
 774                 }
 775         } else { /* plen <= bit */
 776
 777                 /*
 778                  *              (new leaf node)[ln]
 779                  *                /        \
 780                  *           (old node)[fn] NULL
 781                  */
 782
 783                 ln = node_alloc(net);
 784
 785                 if (!ln)
 786                         return ERR_PTR(-ENOMEM);
 787
 788                 ln->fn_bit = plen;
 789
 790                 RCU_INIT_POINTER(ln->parent, pn);
 791
 792                 if (addr_bit_set(&key->addr, plen))
 793                         RCU_INIT_POINTER(ln->right, fn);
 794                 else
 795                         RCU_INIT_POINTER(ln->left, fn);
 796
 797                 rcu_assign_pointer(fn->parent, ln);
 798
 799                 if (dir)
 800                         rcu_assign_pointer(pn->right, ln);
 801                 else
 802                         rcu_assign_pointer(pn->left, ln);
 803         }
 804         return ln;
 805 }
 806
 807 static bool rt6_qualify_for_ecmp(struct rt6_info *rt)
 808 {
 809         return (rt->rt6i_flags & (RTF_GATEWAY|RTF_ADDRCONF|RTF_DYNAMIC)) ==
 810                RTF_GATEWAY;
 811 }
 812
 813 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
 814 {
 815         int i;
 816
 817         for (i = 0; i < RTAX_MAX; i++) {
 818                 if (test_bit(i, mxc->mx_valid))
 819                         mp[i] = mxc->mx[i];
 820         }
 821 }
 822
 823 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
 824 {
 825         if (!mxc->mx)
 826                 return 0;
 827
 828         if (dst->flags & DST_HOST) {
 829                 u32 *mp = dst_metrics_write_ptr(dst);
 830
 831                 if (unlikely(!mp))
 832                         return -ENOMEM;
 833
 834                 fib6_copy_metrics(mp, mxc);
 835         } else {
 836                 dst_init_metrics(dst, mxc->mx, false);
 837
 838                 /* We've stolen mx now. */
 839                 mxc->mx = NULL;
 840         }
 841
 842         return 0;
 843 }
 844
 845 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
 846                           struct net *net)
 847 {
 848         struct fib6_table *table = rt->rt6i_table;
 849
 850         if (atomic_read(&rt->rt6i_ref) != 1) {
 851                 /* This route is used as dummy address holder in some split
 852                  * nodes. It is not leaked, but it still holds other resources,
 853                  * which must be released in time. So, scan ascendant nodes
 854                  * and replace dummy references to this route with references
 855                  * to still alive ones.
 856                  */
 857                 while (fn) {
 858                         struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
 859                                             lockdep_is_held(&table->tb6_lock));
 860                         struct rt6_info *new_leaf;
 861                         if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
 862                                 new_leaf = fib6_find_prefix(net, table, fn);
 863                                 atomic_inc(&new_leaf->rt6i_ref);
 864                                 rcu_assign_pointer(fn->leaf, new_leaf);
 865                                 rt6_release(rt);
 866                         }
 867                         fn = rcu_dereference_protected(fn->parent,
 868                                     lockdep_is_held(&table->tb6_lock));
 869                 }
 870         }
 871 }
 872
 873 /*
 874  *      Insert routing information in a node.
 875  */
 876
 877 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
 878                             struct nl_info *info, struct mx6_config *mxc,
 879                             struct netlink_ext_ack *extack)
 880 {
 881         struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
 882                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
 883         struct rt6_info *iter = NULL;
 884         struct rt6_info __rcu **ins;
 885         struct rt6_info __rcu **fallback_ins = NULL;
 886         int replace = (info->nlh &&
 887                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
 888         int add = (!info->nlh ||
 889                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
 890         int found = 0;
 891         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
 892         u16 nlflags = NLM_F_EXCL;
 893         int err;
 894
 895         if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
 896                 nlflags |= NLM_F_APPEND;
 897
 898         ins = &fn->leaf;
 899
 900         for (iter = leaf; iter;
 901              iter = rcu_dereference_protected(iter->dst.rt6_next,
 902                                 lockdep_is_held(&rt->rt6i_table->tb6_lock))) {
 903                 /*
 904                  *      Search for duplicates
 905                  */
 906
 907                 if (iter->rt6i_metric == rt->rt6i_metric) {
 908                         /*
 909                          *      Same priority level
 910                          */
 911                         if (info->nlh &&
 912                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
 913                                 return -EEXIST;
 914
 915                         nlflags &= ~NLM_F_EXCL;
 916                         if (replace) {
 917                                 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
 918                                         found++;
 919                                         break;
 920                                 }
 921                                 if (rt_can_ecmp)
 922                                         fallback_ins = fallback_ins ?: ins;
 923                                 goto next_iter;
 924                         }
 925
 926                         if (rt6_duplicate_nexthop(iter, rt)) {
 927                                 if (rt->rt6i_nsiblings)
 928                                         rt->rt6i_nsiblings = 0;
 929                                 if (!(iter->rt6i_flags & RTF_EXPIRES))
 930                                         return -EEXIST;
 931                                 if (!(rt->rt6i_flags & RTF_EXPIRES))
 932                                         rt6_clean_expires(iter);
 933                                 else
 934                                         rt6_set_expires(iter, rt->dst.expires);
 935                                 iter->rt6i_pmtu = rt->rt6i_pmtu;
 936                                 return -EEXIST;
 937                         }
 938                         /* If we have the same destination and the same metric,
 939                          * but not the same gateway, then the route we try to
 940                          * add is sibling to this route, increment our counter
 941                          * of siblings, and later we will add our route to the
 942                          * list.
 943                          * Only static routes (which don't have flag
 944                          * RTF_EXPIRES) are used for ECMPv6.
 945                          *
 946                          * To avoid long list, we only had siblings if the
 947                          * route have a gateway.
 948                          */
 949                         if (rt_can_ecmp &&
 950                             rt6_qualify_for_ecmp(iter))
 951                                 rt->rt6i_nsiblings++;
 952                 }
 953
 954                 if (iter->rt6i_metric > rt->rt6i_metric)
 955                         break;
 956
 957 next_iter:
 958                 ins = &iter->dst.rt6_next;
 959         }
 960
 961         if (fallback_ins && !found) {
 962                 /* No ECMP-able route found, replace first non-ECMP one */
 963                 ins = fallback_ins;
 964                 iter = rcu_dereference_protected(*ins,
 965                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
 966                 found++;
 967         }
 968
 969         /* Reset round-robin state, if necessary */
 970         if (ins == &fn->leaf)
 971                 fn->rr_ptr = NULL;
 972
 973         /* Link this route to others same route. */
 974         if (rt->rt6i_nsiblings) {
 975                 unsigned int rt6i_nsiblings;
 976                 struct rt6_info *sibling, *temp_sibling;
 977
 978                 /* Find the first route that have the same metric */
 979                 sibling = leaf;
 980                 while (sibling) {
 981                         if (sibling->rt6i_metric == rt->rt6i_metric &&
 982                             rt6_qualify_for_ecmp(sibling)) {
 983                                 list_add_tail(&rt->rt6i_siblings,
 984                                               &sibling->rt6i_siblings);
 985                                 break;
 986                         }
 987                         sibling = rcu_dereference_protected(sibling->dst.rt6_next,
 988                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
 989                 }
 990                 /* For each sibling in the list, increment the counter of
 991                  * siblings. BUG() if counters does not match, list of siblings
 992                  * is broken!
 993                  */
 994                 rt6i_nsiblings = 0;
 995                 list_for_each_entry_safe(sibling, temp_sibling,
 996                                          &rt->rt6i_siblings, rt6i_siblings) {
 997                         sibling->rt6i_nsiblings++;
 998                         BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
 999                         rt6i_nsiblings++;
1000                 }
1001                 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
1002         }
1003
1004         /*
1005          *      insert node
1006          */
1007         if (!replace) {
1008                 if (!add)
1009                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
1010
1011 add:
1012                 nlflags |= NLM_F_CREATE;
1013                 err = fib6_commit_metrics(&rt->dst, mxc);
1014                 if (err)
1015                         return err;
1016
1017                 rcu_assign_pointer(rt->dst.rt6_next, iter);
1018                 atomic_inc(&rt->rt6i_ref);
1019                 rcu_assign_pointer(rt->rt6i_node, fn);
1020                 rcu_assign_pointer(*ins, rt);
1021                 call_fib6_entry_notifiers(info->nl_net, FIB_EVENT_ENTRY_ADD,
1022                                           rt, extack);
1023                 if (!info->skip_notify)
1024                         inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1025                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1026
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
1032         } else {
1033                 int nsiblings;
1034
1035                 if (!found) {
1036                         if (add)
1037                                 goto add;
1038                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1039                         return -ENOENT;
1040                 }
1041
1042                 err = fib6_commit_metrics(&rt->dst, mxc);
1043                 if (err)
1044                         return err;
1045
1046                 atomic_inc(&rt->rt6i_ref);
1047                 rcu_assign_pointer(rt->rt6i_node, fn);
1048                 rt->dst.rt6_next = iter->dst.rt6_next;
1049                 rcu_assign_pointer(*ins, rt);
1050                 call_fib6_entry_notifiers(info->nl_net, FIB_EVENT_ENTRY_REPLACE,
1051                                           rt, extack);
1052                 if (!info->skip_notify)
1053                         inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1054                 if (!(fn->fn_flags & RTN_RTINFO)) {
1055                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1056                         fn->fn_flags |= RTN_RTINFO;
1057                 }
1058                 nsiblings = iter->rt6i_nsiblings;
1059                 iter->rt6i_node = NULL;
1060                 fib6_purge_rt(iter, fn, info->nl_net);
1061                 if (rcu_access_pointer(fn->rr_ptr) == iter)
1062                         fn->rr_ptr = NULL;
1063                 rt6_release(iter);
1064
1065                 if (nsiblings) {
1066                         /* Replacing an ECMP route, remove all siblings */
1067                         ins = &rt->dst.rt6_next;
1068                         iter = rcu_dereference_protected(*ins,
1069                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
1070                         while (iter) {
1071                                 if (iter->rt6i_metric > rt->rt6i_metric)
1072                                         break;
1073                                 if (rt6_qualify_for_ecmp(iter)) {
1074                                         *ins = iter->dst.rt6_next;
1075                                         iter->rt6i_node = NULL;
1076                                         fib6_purge_rt(iter, fn, info->nl_net);
1077                                         if (rcu_access_pointer(fn->rr_ptr) == iter)
1078                                                 fn->rr_ptr = NULL;
1079                                         rt6_release(iter);
1080                                         nsiblings--;
1081                                         info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1082                                 } else {
1083                                         ins = &iter->dst.rt6_next;
1084                                 }
1085                                 iter = rcu_dereference_protected(*ins,
1086                                         lockdep_is_held(&rt->rt6i_table->tb6_lock));
1087                         }
1088                         WARN_ON(nsiblings != 0);
1089                 }
1090         }
1091
1092         return 0;
1093 }
1094
1095 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
1096 {
1097         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1098             (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
1099                 mod_timer(&net->ipv6.ip6_fib_timer,
1100                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1101 }
1102
1103 void fib6_force_start_gc(struct net *net)
1104 {
1105         if (!timer_pending(&net->ipv6.ip6_fib_timer))
1106                 mod_timer(&net->ipv6.ip6_fib_timer,
1107                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1108 }
1109
1110 static void fib6_update_sernum_upto_root(struct rt6_info *rt,
1111                                          int sernum)
1112 {
1113         struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1114                                 lockdep_is_held(&rt->rt6i_table->tb6_lock));
1115
1116         /* paired with smp_rmb() in rt6_get_cookie_safe() */
1117         smp_wmb();
1118         while (fn) {
1119                 fn->fn_sernum = sernum;
1120                 fn = rcu_dereference_protected(fn->parent,
1121                                 lockdep_is_held(&rt->rt6i_table->tb6_lock));
1122         }
1123 }
1124
1125 /*
1126  *      Add routing information to the routing tree.
1127  *      <destination addr>/<source addr>
1128  *      with source addr info in sub-trees
1129  *      Need to own table->tb6_lock
1130  */
1131
1132 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
1133              struct nl_info *info, struct mx6_config *mxc,
1134              struct netlink_ext_ack *extack)
1135 {
1136         struct fib6_table *table = rt->rt6i_table;
1137         struct fib6_node *fn, *pn = NULL;
1138         int err = -ENOMEM;
1139         int allow_create = 1;
1140         int replace_required = 0;
1141         int sernum = fib6_new_sernum(info->nl_net);
1142
1143         if (WARN_ON_ONCE(!atomic_read(&rt->dst.__refcnt)))
1144                 return -EINVAL;
1145         if (WARN_ON_ONCE(rt->rt6i_flags & RTF_CACHE))
1146                 return -EINVAL;
1147
1148         if (info->nlh) {
1149                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1150                         allow_create = 0;
1151                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1152                         replace_required = 1;
1153         }
1154         if (!allow_create && !replace_required)
1155                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1156
1157         fn = fib6_add_1(info->nl_net, table, root,
1158                         &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
1159                         offsetof(struct rt6_info, rt6i_dst), allow_create,
1160                         replace_required, extack);
1161         if (IS_ERR(fn)) {
1162                 err = PTR_ERR(fn);
1163                 fn = NULL;
1164                 goto out;
1165         }
1166
1167         pn = fn;
1168
1169 #ifdef CONFIG_IPV6_SUBTREES
1170         if (rt->rt6i_src.plen) {
1171                 struct fib6_node *sn;
1172
1173                 if (!rcu_access_pointer(fn->subtree)) {
1174                         struct fib6_node *sfn;
1175
1176                         /*
1177                          * Create subtree.
1178                          *
1179                          *              fn[main tree]
1180                          *              |
1181                          *              sfn[subtree root]
1182                          *                 \
1183                          *                  sn[new leaf node]
1184                          */
1185
1186                         /* Create subtree root node */
1187                         sfn = node_alloc(info->nl_net);
1188                         if (!sfn)
1189                                 goto failure;
1190
1191                         atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1192                         rcu_assign_pointer(sfn->leaf,
1193                                            info->nl_net->ipv6.ip6_null_entry);
1194                         sfn->fn_flags = RTN_ROOT;
1195
1196                         /* Now add the first leaf node to new subtree */
1197
1198                         sn = fib6_add_1(info->nl_net, table, sfn,
1199                                         &rt->rt6i_src.addr, rt->rt6i_src.plen,
1200                                         offsetof(struct rt6_info, rt6i_src),
1201                                         allow_create, replace_required, extack);
1202
1203                         if (IS_ERR(sn)) {
1204                                 /* If it is failed, discard just allocated
1205                                    root, and then (in failure) stale node
1206                                    in main tree.
1207                                  */
1208                                 node_free_immediate(info->nl_net, sfn);
1209                                 err = PTR_ERR(sn);
1210                                 goto failure;
1211                         }
1212
1213                         /* Now link new subtree to main tree */
1214                         rcu_assign_pointer(sfn->parent, fn);
1215                         rcu_assign_pointer(fn->subtree, sfn);
1216                 } else {
1217                         sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1218                                         &rt->rt6i_src.addr, rt->rt6i_src.plen,
1219                                         offsetof(struct rt6_info, rt6i_src),
1220                                         allow_create, replace_required, extack);
1221
1222                         if (IS_ERR(sn)) {
1223                                 err = PTR_ERR(sn);
1224                                 goto failure;
1225                         }
1226                 }
1227
1228                 if (!rcu_access_pointer(fn->leaf)) {
1229                         if (fn->fn_flags & RTN_TL_ROOT) {
1230                                 /* put back null_entry for root node */
1231                                 rcu_assign_pointer(fn->leaf,
1232                                             info->nl_net->ipv6.ip6_null_entry);
1233                         } else {
1234                                 atomic_inc(&rt->rt6i_ref);
1235                                 rcu_assign_pointer(fn->leaf, rt);
1236                         }
1237                 }
1238                 fn = sn;
1239         }
1240 #endif
1241
1242         err = fib6_add_rt2node(fn, rt, info, mxc, extack);
1243         if (!err) {
1244                 fib6_update_sernum_upto_root(rt, sernum);
1245                 fib6_start_gc(info->nl_net, rt);
1246         }
1247
1248 out:
1249         if (err) {
1250 #ifdef CONFIG_IPV6_SUBTREES
1251                 /*
1252                  * If fib6_add_1 has cleared the old leaf pointer in the
1253                  * super-tree leaf node we have to find a new one for it.
1254                  */
1255                 if (pn != fn) {
1256                         struct rt6_info *pn_leaf =
1257                                 rcu_dereference_protected(pn->leaf,
1258                                     lockdep_is_held(&table->tb6_lock));
1259                         if (pn_leaf == rt) {
1260                                 pn_leaf = NULL;
1261                                 RCU_INIT_POINTER(pn->leaf, NULL);
1262                                 atomic_dec(&rt->rt6i_ref);
1263                         }
1264                         if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1265                                 pn_leaf = fib6_find_prefix(info->nl_net, table,
1266                                                            pn);
1267 #if RT6_DEBUG >= 2
1268                                 if (!pn_leaf) {
1269                                         WARN_ON(!pn_leaf);
1270                                         pn_leaf =
1271                                             info->nl_net->ipv6.ip6_null_entry;
1272                                 }
1273 #endif
1274                                 atomic_inc(&pn_leaf->rt6i_ref);
1275                                 rcu_assign_pointer(pn->leaf, pn_leaf);
1276                         }
1277                 }
1278 #endif
1279                 goto failure;
1280         }
1281         return err;
1282
1283 failure:
1284         /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1285          * 1. fn is an intermediate node and we failed to add the new
1286          * route to it in both subtree creation failure and fib6_add_rt2node()
1287          * failure case.
1288          * 2. fn is the root node in the table and we fail to add the first
1289          * default route to it.
1290          */
1291         if (fn &&
1292             (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1293              (fn->fn_flags & RTN_TL_ROOT &&
1294               !rcu_access_pointer(fn->leaf))))
1295                 fib6_repair_tree(info->nl_net, table, fn);
1296         /* Always release dst as dst->__refcnt is guaranteed
1297          * to be taken before entering this function
1298          */
1299         dst_release_immediate(&rt->dst);
1300         return err;
1301 }
1302
1303 /*
1304  *      Routing tree lookup
1305  *
1306  */
1307
1308 struct lookup_args {
1309         int                     offset;         /* key offset on rt6_info       */
1310         const struct in6_addr   *addr;          /* search key                   */
1311 };
1312
1313 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1314                                        struct lookup_args *args)
1315 {
1316         struct fib6_node *fn;
1317         __be32 dir;
1318
1319         if (unlikely(args->offset == 0))
1320                 return NULL;
1321
1322         /*
1323          *      Descend on a tree
1324          */
1325
1326         fn = root;
1327
1328         for (;;) {
1329                 struct fib6_node *next;
1330
1331                 dir = addr_bit_set(args->addr, fn->fn_bit);
1332
1333                 next = dir ? rcu_dereference(fn->right) :
1334                              rcu_dereference(fn->left);
1335
1336                 if (next) {
1337                         fn = next;
1338                         continue;
1339                 }
1340                 break;
1341         }
1342
1343         while (fn) {
1344                 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1345
1346                 if (subtree || fn->fn_flags & RTN_RTINFO) {
1347                         struct rt6_info *leaf = rcu_dereference(fn->leaf);
1348                         struct rt6key *key;
1349
1350                         if (!leaf)
1351                                 goto backtrack;
1352
1353                         key = (struct rt6key *) ((u8 *)leaf + args->offset);
1354
1355                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1356 #ifdef CONFIG_IPV6_SUBTREES
1357                                 if (subtree) {
1358                                         struct fib6_node *sfn;
1359                                         sfn = fib6_lookup_1(subtree, args + 1);
1360                                         if (!sfn)
1361                                                 goto backtrack;
1362                                         fn = sfn;
1363                                 }
1364 #endif
1365                                 if (fn->fn_flags & RTN_RTINFO)
1366                                         return fn;
1367                         }
1368                 }
1369 backtrack:
1370                 if (fn->fn_flags & RTN_ROOT)
1371                         break;
1372
1373                 fn = rcu_dereference(fn->parent);
1374         }
1375
1376         return NULL;
1377 }
1378
1379 /* called with rcu_read_lock() held
1380  */
1381 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1382                               const struct in6_addr *saddr)
1383 {
1384         struct fib6_node *fn;
1385         struct lookup_args args[] = {
1386                 {
1387                         .offset = offsetof(struct rt6_info, rt6i_dst),
1388                         .addr = daddr,
1389                 },
1390 #ifdef CONFIG_IPV6_SUBTREES
1391                 {
1392                         .offset = offsetof(struct rt6_info, rt6i_src),
1393                         .addr = saddr,
1394                 },
1395 #endif
1396                 {
1397                         .offset = 0,    /* sentinel */
1398                 }
1399         };
1400
1401         fn = fib6_lookup_1(root, daddr ? args : args + 1);
1402         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1403                 fn = root;
1404
1405         return fn;
1406 }
1407
1408 /*
1409  *      Get node with specified destination prefix (and source prefix,
1410  *      if subtrees are used)
1411  *      exact_match == true means we try to find fn with exact match of
1412  *      the passed in prefix addr
1413  *      exact_match == false means we try to find fn with longest prefix
1414  *      match of the passed in prefix addr. This is useful for finding fn
1415  *      for cached route as it will be stored in the exception table under
1416  *      the node with longest prefix length.
1417  */
1418
1419
1420 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1421                                        const struct in6_addr *addr,
1422                                        int plen, int offset,
1423                                        bool exact_match)
1424 {
1425         struct fib6_node *fn, *prev = NULL;
1426
1427         for (fn = root; fn ; ) {
1428                 struct rt6_info *leaf = rcu_dereference(fn->leaf);
1429                 struct rt6key *key;
1430
1431                 /* This node is being deleted */
1432                 if (!leaf) {
1433                         if (plen <= fn->fn_bit)
1434                                 goto out;
1435                         else
1436                                 goto next;
1437                 }
1438
1439                 key = (struct rt6key *)((u8 *)leaf + offset);
1440
1441                 /*
1442                  *      Prefix match
1443                  */
1444                 if (plen < fn->fn_bit ||
1445                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1446                         goto out;
1447
1448                 if (plen == fn->fn_bit)
1449                         return fn;
1450
1451                 prev = fn;
1452
1453 next:
1454                 /*
1455                  *      We have more bits to go
1456                  */
1457                 if (addr_bit_set(addr, fn->fn_bit))
1458                         fn = rcu_dereference(fn->right);
1459                 else
1460                         fn = rcu_dereference(fn->left);
1461         }
1462 out:
1463         if (exact_match)
1464                 return NULL;
1465         else
1466                 return prev;
1467 }
1468
1469 struct fib6_node *fib6_locate(struct fib6_node *root,
1470                               const struct in6_addr *daddr, int dst_len,
1471                               const struct in6_addr *saddr, int src_len,
1472                               bool exact_match)
1473 {
1474         struct fib6_node *fn;
1475
1476         fn = fib6_locate_1(root, daddr, dst_len,
1477                            offsetof(struct rt6_info, rt6i_dst),
1478                            exact_match);
1479
1480 #ifdef CONFIG_IPV6_SUBTREES
1481         if (src_len) {
1482                 WARN_ON(saddr == NULL);
1483                 if (fn) {
1484                         struct fib6_node *subtree = FIB6_SUBTREE(fn);
1485
1486                         if (subtree) {
1487                                 fn = fib6_locate_1(subtree, saddr, src_len,
1488                                            offsetof(struct rt6_info, rt6i_src),
1489                                            exact_match);
1490                         }
1491                 }
1492         }
1493 #endif
1494
1495         if (fn && fn->fn_flags & RTN_RTINFO)
1496                 return fn;
1497
1498         return NULL;
1499 }
1500
1501
1502 /*
1503  *      Deletion
1504  *
1505  */
1506
1507 static struct rt6_info *fib6_find_prefix(struct net *net,
1508                                          struct fib6_table *table,
1509                                          struct fib6_node *fn)
1510 {
1511         struct fib6_node *child_left, *child_right;
1512
1513         if (fn->fn_flags & RTN_ROOT)
1514                 return net->ipv6.ip6_null_entry;
1515
1516         while (fn) {
1517                 child_left = rcu_dereference_protected(fn->left,
1518                                     lockdep_is_held(&table->tb6_lock));
1519                 child_right = rcu_dereference_protected(fn->right,
1520                                     lockdep_is_held(&table->tb6_lock));
1521                 if (child_left)
1522                         return rcu_dereference_protected(child_left->leaf,
1523                                         lockdep_is_held(&table->tb6_lock));
1524                 if (child_right)
1525                         return rcu_dereference_protected(child_right->leaf,
1526                                         lockdep_is_held(&table->tb6_lock));
1527
1528                 fn = FIB6_SUBTREE(fn);
1529         }
1530         return NULL;
1531 }
1532
1533 /*
1534  *      Called to trim the tree of intermediate nodes when possible. "fn"
1535  *      is the node we want to try and remove.
1536  *      Need to own table->tb6_lock
1537  */
1538
1539 static struct fib6_node *fib6_repair_tree(struct net *net,
1540                                           struct fib6_table *table,
1541                                           struct fib6_node *fn)
1542 {
1543         int children;
1544         int nstate;
1545         struct fib6_node *child;
1546         struct fib6_walker *w;
1547         int iter = 0;
1548
1549         /* Set fn->leaf to null_entry for root node. */
1550         if (fn->fn_flags & RTN_TL_ROOT) {
1551                 rcu_assign_pointer(fn->leaf, net->ipv6.ip6_null_entry);
1552                 return fn;
1553         }
1554
1555         for (;;) {
1556                 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1557                                             lockdep_is_held(&table->tb6_lock));
1558                 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1559                                             lockdep_is_held(&table->tb6_lock));
1560                 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1561                                             lockdep_is_held(&table->tb6_lock));
1562                 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1563                                             lockdep_is_held(&table->tb6_lock));
1564                 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1565                                             lockdep_is_held(&table->tb6_lock));
1566                 struct rt6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1567                                             lockdep_is_held(&table->tb6_lock));
1568                 struct rt6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1569                                             lockdep_is_held(&table->tb6_lock));
1570                 struct rt6_info *new_fn_leaf;
1571
1572                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1573                 iter++;
1574
1575                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1576                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1577                 WARN_ON(fn_leaf);
1578
1579                 children = 0;
1580                 child = NULL;
1581                 if (fn_r)
1582                         child = fn_r, children |= 1;
1583                 if (fn_l)
1584                         child = fn_l, children |= 2;
1585
1586                 if (children == 3 || FIB6_SUBTREE(fn)
1587 #ifdef CONFIG_IPV6_SUBTREES
1588                     /* Subtree root (i.e. fn) may have one child */
1589                     || (children && fn->fn_flags & RTN_ROOT)
1590 #endif
1591                     ) {
1592                         new_fn_leaf = fib6_find_prefix(net, table, fn);
1593 #if RT6_DEBUG >= 2
1594                         if (!new_fn_leaf) {
1595                                 WARN_ON(!new_fn_leaf);
1596                                 new_fn_leaf = net->ipv6.ip6_null_entry;
1597                         }
1598 #endif
1599                         atomic_inc(&new_fn_leaf->rt6i_ref);
1600                         rcu_assign_pointer(fn->leaf, new_fn_leaf);
1601                         return pn;
1602                 }
1603
1604 #ifdef CONFIG_IPV6_SUBTREES
1605                 if (FIB6_SUBTREE(pn) == fn) {
1606                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1607                         RCU_INIT_POINTER(pn->subtree, NULL);
1608                         nstate = FWS_L;
1609                 } else {
1610                         WARN_ON(fn->fn_flags & RTN_ROOT);
1611 #endif
1612                         if (pn_r == fn)
1613                                 rcu_assign_pointer(pn->right, child);
1614                         else if (pn_l == fn)
1615                                 rcu_assign_pointer(pn->left, child);
1616 #if RT6_DEBUG >= 2
1617                         else
1618                                 WARN_ON(1);
1619 #endif
1620                         if (child)
1621                                 rcu_assign_pointer(child->parent, pn);
1622                         nstate = FWS_R;
1623 #ifdef CONFIG_IPV6_SUBTREES
1624                 }
1625 #endif
1626
1627                 read_lock(&net->ipv6.fib6_walker_lock);
1628                 FOR_WALKERS(net, w) {
1629                         if (!child) {
1630                                 if (w->node == fn) {
1631                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1632                                         w->node = pn;
1633                                         w->state = nstate;
1634                                 }
1635                         } else {
1636                                 if (w->node == fn) {
1637                                         w->node = child;
1638                                         if (children&2) {
1639                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1640                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1641                                         } else {
1642                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1643                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1644                                         }
1645                                 }
1646                         }
1647                 }
1648                 read_unlock(&net->ipv6.fib6_walker_lock);
1649
1650                 node_free(net, fn);
1651                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1652                         return pn;
1653
1654                 RCU_INIT_POINTER(pn->leaf, NULL);
1655                 rt6_release(pn_leaf);
1656                 fn = pn;
1657         }
1658 }
1659
1660 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1661                            struct rt6_info __rcu **rtp, struct nl_info *info)
1662 {
1663         struct fib6_walker *w;
1664         struct rt6_info *rt = rcu_dereference_protected(*rtp,
1665                                     lockdep_is_held(&table->tb6_lock));
1666         struct net *net = info->nl_net;
1667
1668         RT6_TRACE("fib6_del_route\n");
1669
1670         WARN_ON_ONCE(rt->rt6i_flags & RTF_CACHE);
1671
1672         /* Unlink it */
1673         *rtp = rt->dst.rt6_next;
1674         rt->rt6i_node = NULL;
1675         net->ipv6.rt6_stats->fib_rt_entries--;
1676         net->ipv6.rt6_stats->fib_discarded_routes++;
1677
1678         /* Flush all cached dst in exception table */
1679         rt6_flush_exceptions(rt);
1680
1681         /* Reset round-robin state, if necessary */
1682         if (rcu_access_pointer(fn->rr_ptr) == rt)
1683                 fn->rr_ptr = NULL;
1684
1685         /* Remove this entry from other siblings */
1686         if (rt->rt6i_nsiblings) {
1687                 struct rt6_info *sibling, *next_sibling;
1688
1689                 list_for_each_entry_safe(sibling, next_sibling,
1690                                          &rt->rt6i_siblings, rt6i_siblings)
1691                         sibling->rt6i_nsiblings--;
1692                 rt->rt6i_nsiblings = 0;
1693                 list_del_init(&rt->rt6i_siblings);
1694         }
1695
1696         /* Adjust walkers */
1697         read_lock(&net->ipv6.fib6_walker_lock);
1698         FOR_WALKERS(net, w) {
1699                 if (w->state == FWS_C && w->leaf == rt) {
1700                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1701                         w->leaf = rcu_dereference_protected(rt->dst.rt6_next,
1702                                             lockdep_is_held(&table->tb6_lock));
1703                         if (!w->leaf)
1704                                 w->state = FWS_U;
1705                 }
1706         }
1707         read_unlock(&net->ipv6.fib6_walker_lock);
1708
1709         /* If it was last route, call fib6_repair_tree() to:
1710          * 1. For root node, put back null_entry as how the table was created.
1711          * 2. For other nodes, expunge its radix tree node.
1712          */
1713         if (!rcu_access_pointer(fn->leaf)) {
1714                 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1715                         fn->fn_flags &= ~RTN_RTINFO;
1716                         net->ipv6.rt6_stats->fib_route_nodes--;
1717                 }
1718                 fn = fib6_repair_tree(net, table, fn);
1719         }
1720
1721         fib6_purge_rt(rt, fn, net);
1722
1723         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1724         if (!info->skip_notify)
1725                 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1726         rt6_release(rt);
1727 }
1728
1729 /* Need to own table->tb6_lock */
1730 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1731 {
1732         struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1733                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
1734         struct fib6_table *table = rt->rt6i_table;
1735         struct net *net = info->nl_net;
1736         struct rt6_info __rcu **rtp;
1737         struct rt6_info __rcu **rtp_next;
1738
1739 #if RT6_DEBUG >= 2
1740         if (rt->dst.obsolete > 0) {
1741                 WARN_ON(fn);
1742                 return -ENOENT;
1743         }
1744 #endif
1745         if (!fn || rt == net->ipv6.ip6_null_entry)
1746                 return -ENOENT;
1747
1748         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1749
1750         /* remove cached dst from exception table */
1751         if (rt->rt6i_flags & RTF_CACHE)
1752                 return rt6_remove_exception_rt(rt);
1753
1754         /*
1755          *      Walk the leaf entries looking for ourself
1756          */
1757
1758         for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1759                 struct rt6_info *cur = rcu_dereference_protected(*rtp,
1760                                         lockdep_is_held(&table->tb6_lock));
1761                 if (rt == cur) {
1762                         fib6_del_route(table, fn, rtp, info);
1763                         return 0;
1764                 }
1765                 rtp_next = &cur->dst.rt6_next;
1766         }
1767         return -ENOENT;
1768 }
1769
1770 /*
1771  *      Tree traversal function.
1772  *
1773  *      Certainly, it is not interrupt safe.
1774  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1775  *      It means, that we can modify tree during walking
1776  *      and use this function for garbage collection, clone pruning,
1777  *      cleaning tree when a device goes down etc. etc.
1778  *
1779  *      It guarantees that every node will be traversed,
1780  *      and that it will be traversed only once.
1781  *
1782  *      Callback function w->func may return:
1783  *      0 -> continue walking.
1784  *      positive value -> walking is suspended (used by tree dumps,
1785  *      and probably by gc, if it will be split to several slices)
1786  *      negative value -> terminate walking.
1787  *
1788  *      The function itself returns:
1789  *      0   -> walk is complete.
1790  *      >0  -> walk is incomplete (i.e. suspended)
1791  *      <0  -> walk is terminated by an error.
1792  *
1793  *      This function is called with tb6_lock held.
1794  */
1795
1796 static int fib6_walk_continue(struct fib6_walker *w)
1797 {
1798         struct fib6_node *fn, *pn, *left, *right;
1799
1800         /* w->root should always be table->tb6_root */
1801         WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1802
1803         for (;;) {
1804                 fn = w->node;
1805                 if (!fn)
1806                         return 0;
1807
1808                 switch (w->state) {
1809 #ifdef CONFIG_IPV6_SUBTREES
1810                 case FWS_S:
1811                         if (FIB6_SUBTREE(fn)) {
1812                                 w->node = FIB6_SUBTREE(fn);
1813                                 continue;
1814                         }
1815                         w->state = FWS_L;
1816 #endif
1817                         /* fall through */
1818                 case FWS_L:
1819                         left = rcu_dereference_protected(fn->left, 1);
1820                         if (left) {
1821                                 w->node = left;
1822                                 w->state = FWS_INIT;
1823                                 continue;
1824                         }
1825                         w->state = FWS_R;
1826                         /* fall through */
1827                 case FWS_R:
1828                         right = rcu_dereference_protected(fn->right, 1);
1829                         if (right) {
1830                                 w->node = right;
1831                                 w->state = FWS_INIT;
1832                                 continue;
1833                         }
1834                         w->state = FWS_C;
1835                         w->leaf = rcu_dereference_protected(fn->leaf, 1);
1836                         /* fall through */
1837                 case FWS_C:
1838                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1839                                 int err;
1840
1841                                 if (w->skip) {
1842                                         w->skip--;
1843                                         goto skip;
1844                                 }
1845
1846                                 err = w->func(w);
1847                                 if (err)
1848                                         return err;
1849
1850                                 w->count++;
1851                                 continue;
1852                         }
1853 skip:
1854                         w->state = FWS_U;
1855                         /* fall through */
1856                 case FWS_U:
1857                         if (fn == w->root)
1858                                 return 0;
1859                         pn = rcu_dereference_protected(fn->parent, 1);
1860                         left = rcu_dereference_protected(pn->left, 1);
1861                         right = rcu_dereference_protected(pn->right, 1);
1862                         w->node = pn;
1863 #ifdef CONFIG_IPV6_SUBTREES
1864                         if (FIB6_SUBTREE(pn) == fn) {
1865                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1866                                 w->state = FWS_L;
1867                                 continue;
1868                         }
1869 #endif
1870                         if (left == fn) {
1871                                 w->state = FWS_R;
1872                                 continue;
1873                         }
1874                         if (right == fn) {
1875                                 w->state = FWS_C;
1876                                 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1877                                 continue;
1878                         }
1879 #if RT6_DEBUG >= 2
1880                         WARN_ON(1);
1881 #endif
1882                 }
1883         }
1884 }
1885
1886 static int fib6_walk(struct net *net, struct fib6_walker *w)
1887 {
1888         int res;
1889
1890         w->state = FWS_INIT;
1891         w->node = w->root;
1892
1893         fib6_walker_link(net, w);
1894         res = fib6_walk_continue(w);
1895         if (res <= 0)
1896                 fib6_walker_unlink(net, w);
1897         return res;
1898 }
1899
1900 static int fib6_clean_node(struct fib6_walker *w)
1901 {
1902         int res;
1903         struct rt6_info *rt;
1904         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1905         struct nl_info info = {
1906                 .nl_net = c->net,
1907         };
1908
1909         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1910             w->node->fn_sernum != c->sernum)
1911                 w->node->fn_sernum = c->sernum;
1912
1913         if (!c->func) {
1914                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1915                 w->leaf = NULL;
1916                 return 0;
1917         }
1918
1919         for_each_fib6_walker_rt(w) {
1920                 res = c->func(rt, c->arg);
1921                 if (res < 0) {
1922                         w->leaf = rt;
1923                         res = fib6_del(rt, &info);
1924                         if (res) {
1925 #if RT6_DEBUG >= 2
1926                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1927                                          __func__, rt,
1928                                          rcu_access_pointer(rt->rt6i_node),
1929                                          res);
1930 #endif
1931                                 continue;
1932                         }
1933                         return 0;
1934                 }
1935                 WARN_ON(res != 0);
1936         }
1937         w->leaf = rt;
1938         return 0;
1939 }
1940
1941 /*
1942  *      Convenient frontend to tree walker.
1943  *
1944  *      func is called on each route.
1945  *              It may return -1 -> delete this route.
1946  *                            0  -> continue walking
1947  */
1948
1949 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1950                             int (*func)(struct rt6_info *, void *arg),
1951                             int sernum, void *arg)
1952 {
1953         struct fib6_cleaner c;
1954
1955         c.w.root = root;
1956         c.w.func = fib6_clean_node;
1957         c.w.count = 0;
1958         c.w.skip = 0;
1959         c.func = func;
1960         c.sernum = sernum;
1961         c.arg = arg;
1962         c.net = net;
1963
1964         fib6_walk(net, &c.w);
1965 }
1966
1967 static void __fib6_clean_all(struct net *net,
1968                              int (*func)(struct rt6_info *, void *),
1969                              int sernum, void *arg)
1970 {
1971         struct fib6_table *table;
1972         struct hlist_head *head;
1973         unsigned int h;
1974
1975         rcu_read_lock();
1976         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1977                 head = &net->ipv6.fib_table_hash[h];
1978                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1979                         spin_lock_bh(&table->tb6_lock);
1980                         fib6_clean_tree(net, &table->tb6_root,
1981                                         func, sernum, arg);
1982                         spin_unlock_bh(&table->tb6_lock);
1983                 }
1984         }
1985         rcu_read_unlock();
1986 }
1987
1988 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
1989                     void *arg)
1990 {
1991         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
1992 }
1993
1994 static void fib6_flush_trees(struct net *net)
1995 {
1996         int new_sernum = fib6_new_sernum(net);
1997
1998         __fib6_clean_all(net, NULL, new_sernum, NULL);
1999 }
2000
2001 /*
2002  *      Garbage collection
2003  */
2004
2005 static int fib6_age(struct rt6_info *rt, void *arg)
2006 {
2007         struct fib6_gc_args *gc_args = arg;
2008         unsigned long now = jiffies;
2009
2010         /*
2011          *      check addrconf expiration here.
2012          *      Routes are expired even if they are in use.
2013          */
2014
2015         if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
2016                 if (time_after(now, rt->dst.expires)) {
2017                         RT6_TRACE("expiring %p\n", rt);
2018                         return -1;
2019                 }
2020                 gc_args->more++;
2021         }
2022
2023         /*      Also age clones in the exception table.
2024          *      Note, that clones are aged out
2025          *      only if they are not in use now.
2026          */
2027         rt6_age_exceptions(rt, gc_args, now);
2028
2029         return 0;
2030 }
2031
2032 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2033 {
2034         struct fib6_gc_args gc_args;
2035         unsigned long now;
2036
2037         if (force) {
2038                 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2039         } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2040                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2041                 return;
2042         }
2043         gc_args.timeout = expires ? (int)expires :
2044                           net->ipv6.sysctl.ip6_rt_gc_interval;
2045         gc_args.more = 0;
2046
2047         fib6_clean_all(net, fib6_age, &gc_args);
2048         now = jiffies;
2049         net->ipv6.ip6_rt_last_gc = now;
2050
2051         if (gc_args.more)
2052                 mod_timer(&net->ipv6.ip6_fib_timer,
2053                           round_jiffies(now
2054                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
2055         else
2056                 del_timer(&net->ipv6.ip6_fib_timer);
2057         spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2058 }
2059
2060 static void fib6_gc_timer_cb(struct timer_list *t)
2061 {
2062         struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2063
2064         fib6_run_gc(0, arg, true);
2065 }
2066
2067 static int __net_init fib6_net_init(struct net *net)
2068 {
2069         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2070         int err;
2071
2072         err = fib6_notifier_init(net);
2073         if (err)
2074                 return err;
2075
2076         spin_lock_init(&net->ipv6.fib6_gc_lock);
2077         rwlock_init(&net->ipv6.fib6_walker_lock);
2078         INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2079         timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2080
2081         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2082         if (!net->ipv6.rt6_stats)
2083                 goto out_timer;
2084
2085         /* Avoid false sharing : Use at least a full cache line */
2086         size = max_t(size_t, size, L1_CACHE_BYTES);
2087
2088         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2089         if (!net->ipv6.fib_table_hash)
2090                 goto out_rt6_stats;
2091
2092         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2093                                           GFP_KERNEL);
2094         if (!net->ipv6.fib6_main_tbl)
2095                 goto out_fib_table_hash;
2096
2097         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2098         rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2099                            net->ipv6.ip6_null_entry);
2100         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2101                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2102         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2103
2104 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2105         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2106                                            GFP_KERNEL);
2107         if (!net->ipv6.fib6_local_tbl)
2108                 goto out_fib6_main_tbl;
2109         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2110         rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2111                            net->ipv6.ip6_null_entry);
2112         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2113                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2114         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2115 #endif
2116         fib6_tables_init(net);
2117
2118         return 0;
2119
2120 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2121 out_fib6_main_tbl:
2122         kfree(net->ipv6.fib6_main_tbl);
2123 #endif
2124 out_fib_table_hash:
2125         kfree(net->ipv6.fib_table_hash);
2126 out_rt6_stats:
2127         kfree(net->ipv6.rt6_stats);
2128 out_timer:
2129         fib6_notifier_exit(net);
2130         return -ENOMEM;
2131 }
2132
2133 static void fib6_net_exit(struct net *net)
2134 {
2135         unsigned int i;
2136
2137         rt6_ifdown(net, NULL);
2138         del_timer_sync(&net->ipv6.ip6_fib_timer);
2139
2140         for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2141                 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2142                 struct hlist_node *tmp;
2143                 struct fib6_table *tb;
2144
2145                 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2146                         hlist_del(&tb->tb6_hlist);
2147                         fib6_free_table(tb);
2148                 }
2149         }
2150
2151         kfree(net->ipv6.fib_table_hash);
2152         kfree(net->ipv6.rt6_stats);
2153         fib6_notifier_exit(net);
2154 }
2155
2156 static struct pernet_operations fib6_net_ops = {
2157         .init = fib6_net_init,
2158         .exit = fib6_net_exit,
2159 };
2160
2161 int __init fib6_init(void)
2162 {
2163         int ret = -ENOMEM;
2164
2165         fib6_node_kmem = kmem_cache_create("fib6_nodes",
2166                                            sizeof(struct fib6_node),
2167                                            0, SLAB_HWCACHE_ALIGN,
2168                                            NULL);
2169         if (!fib6_node_kmem)
2170                 goto out;
2171
2172         ret = register_pernet_subsys(&fib6_net_ops);
2173         if (ret)
2174                 goto out_kmem_cache_create;
2175
2176         ret = __rtnl_register(PF_INET6, RTM_GETROUTE, NULL, inet6_dump_fib,
2177                               0);
2178         if (ret)
2179                 goto out_unregister_subsys;
2180
2181         __fib6_flush_trees = fib6_flush_trees;
2182 out:
2183         return ret;
2184
2185 out_unregister_subsys:
2186         unregister_pernet_subsys(&fib6_net_ops);
2187 out_kmem_cache_create:
2188         kmem_cache_destroy(fib6_node_kmem);
2189         goto out;
2190 }
2191
2192 void fib6_gc_cleanup(void)
2193 {
2194         unregister_pernet_subsys(&fib6_net_ops);
2195         kmem_cache_destroy(fib6_node_kmem);
2196 }
2197
2198 #ifdef CONFIG_PROC_FS
2199
2200 struct ipv6_route_iter {
2201         struct seq_net_private p;
2202         struct fib6_walker w;
2203         loff_t skip;
2204         struct fib6_table *tbl;
2205         int sernum;
2206 };
2207
2208 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2209 {
2210         struct rt6_info *rt = v;
2211         struct ipv6_route_iter *iter = seq->private;
2212
2213         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
2214
2215 #ifdef CONFIG_IPV6_SUBTREES
2216         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
2217 #else
2218         seq_puts(seq, "00000000000000000000000000000000 00 ");
2219 #endif
2220         if (rt->rt6i_flags & RTF_GATEWAY)
2221                 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
2222         else
2223                 seq_puts(seq, "00000000000000000000000000000000");
2224
2225         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2226                    rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
2227                    rt->dst.__use, rt->rt6i_flags,
2228                    rt->dst.dev ? rt->dst.dev->name : "");
2229         iter->w.leaf = NULL;
2230         return 0;
2231 }
2232
2233 static int ipv6_route_yield(struct fib6_walker *w)
2234 {
2235         struct ipv6_route_iter *iter = w->args;
2236
2237         if (!iter->skip)
2238                 return 1;
2239
2240         do {
2241                 iter->w.leaf = rcu_dereference_protected(
2242                                 iter->w.leaf->dst.rt6_next,
2243                                 lockdep_is_held(&iter->tbl->tb6_lock));
2244                 iter->skip--;
2245                 if (!iter->skip && iter->w.leaf)
2246                         return 1;
2247         } while (iter->w.leaf);
2248
2249         return 0;
2250 }
2251
2252 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2253                                       struct net *net)
2254 {
2255         memset(&iter->w, 0, sizeof(iter->w));
2256         iter->w.func = ipv6_route_yield;
2257         iter->w.root = &iter->tbl->tb6_root;
2258         iter->w.state = FWS_INIT;
2259         iter->w.node = iter->w.root;
2260         iter->w.args = iter;
2261         iter->sernum = iter->w.root->fn_sernum;
2262         INIT_LIST_HEAD(&iter->w.lh);
2263         fib6_walker_link(net, &iter->w);
2264 }
2265
2266 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2267                                                     struct net *net)
2268 {
2269         unsigned int h;
2270         struct hlist_node *node;
2271
2272         if (tbl) {
2273                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2274                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2275         } else {
2276                 h = 0;
2277                 node = NULL;
2278         }
2279
2280         while (!node && h < FIB6_TABLE_HASHSZ) {
2281                 node = rcu_dereference_bh(
2282                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2283         }
2284         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2285 }
2286
2287 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2288 {
2289         if (iter->sernum != iter->w.root->fn_sernum) {
2290                 iter->sernum = iter->w.root->fn_sernum;
2291                 iter->w.state = FWS_INIT;
2292                 iter->w.node = iter->w.root;
2293                 WARN_ON(iter->w.skip);
2294                 iter->w.skip = iter->w.count;
2295         }
2296 }
2297
2298 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2299 {
2300         int r;
2301         struct rt6_info *n;
2302         struct net *net = seq_file_net(seq);
2303         struct ipv6_route_iter *iter = seq->private;
2304
2305         if (!v)
2306                 goto iter_table;
2307
2308         n = rcu_dereference_bh(((struct rt6_info *)v)->dst.rt6_next);
2309         if (n) {
2310                 ++*pos;
2311                 return n;
2312         }
2313
2314 iter_table:
2315         ipv6_route_check_sernum(iter);
2316         spin_lock_bh(&iter->tbl->tb6_lock);
2317         r = fib6_walk_continue(&iter->w);
2318         spin_unlock_bh(&iter->tbl->tb6_lock);
2319         if (r > 0) {
2320                 if (v)
2321                         ++*pos;
2322                 return iter->w.leaf;
2323         } else if (r < 0) {
2324                 fib6_walker_unlink(net, &iter->w);
2325                 return NULL;
2326         }
2327         fib6_walker_unlink(net, &iter->w);
2328
2329         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2330         if (!iter->tbl)
2331                 return NULL;
2332
2333         ipv6_route_seq_setup_walk(iter, net);
2334         goto iter_table;
2335 }
2336
2337 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2338         __acquires(RCU_BH)
2339 {
2340         struct net *net = seq_file_net(seq);
2341         struct ipv6_route_iter *iter = seq->private;
2342
2343         rcu_read_lock_bh();
2344         iter->tbl = ipv6_route_seq_next_table(NULL, net);
2345         iter->skip = *pos;
2346
2347         if (iter->tbl) {
2348                 ipv6_route_seq_setup_walk(iter, net);
2349                 return ipv6_route_seq_next(seq, NULL, pos);
2350         } else {
2351                 return NULL;
2352         }
2353 }
2354
2355 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2356 {
2357         struct fib6_walker *w = &iter->w;
2358         return w->node && !(w->state == FWS_U && w->node == w->root);
2359 }
2360
2361 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2362         __releases(RCU_BH)
2363 {
2364         struct net *net = seq_file_net(seq);
2365         struct ipv6_route_iter *iter = seq->private;
2366
2367         if (ipv6_route_iter_active(iter))
2368                 fib6_walker_unlink(net, &iter->w);
2369
2370         rcu_read_unlock_bh();
2371 }
2372
2373 static const struct seq_operations ipv6_route_seq_ops = {
2374         .start  = ipv6_route_seq_start,
2375         .next   = ipv6_route_seq_next,
2376         .stop   = ipv6_route_seq_stop,
2377         .show   = ipv6_route_seq_show
2378 };
2379
2380 int ipv6_route_open(struct inode *inode, struct file *file)
2381 {
2382         return seq_open_net(inode, file, &ipv6_route_seq_ops,
2383                             sizeof(struct ipv6_route_iter));
2384 }
2385
2386 #endif /* CONFIG_PROC_FS */