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 net *net = dev_net(rt->dst.dev);
 111         struct fib6_node *fn;
 112
 113         fn = rcu_dereference_protected(rt->rt6i_node,
 114                         lockdep_is_held(&rt->rt6i_table->tb6_lock));
 115         if (fn)
 116                 fn->fn_sernum = fib6_new_sernum(net);
 117 }
 118
 119 /*
 120  *      Auxiliary address test functions for the radix tree.
 121  *
 122  *      These assume a 32bit processor (although it will work on
 123  *      64bit processors)
 124  */
 125
 126 /*
 127  *      test bit
 128  */
 129 #if defined(__LITTLE_ENDIAN)
 130 # define BITOP_BE32_SWIZZLE     (0x1F & ~7)
 131 #else
 132 # define BITOP_BE32_SWIZZLE     0
 133 #endif
 134
 135 static __be32 addr_bit_set(const void *token, int fn_bit)
 136 {
 137         const __be32 *addr = token;
 138         /*
 139          * Here,
 140          *      1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
 141          * is optimized version of
 142          *      htonl(1 << ((~fn_bit)&0x1F))
 143          * See include/asm-generic/bitops/le.h.
 144          */
 145         return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
 146                addr[fn_bit >> 5];
 147 }
 148
 149 static struct fib6_node *node_alloc(struct net *net)
 150 {
 151         struct fib6_node *fn;
 152
 153         fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
 154         if (fn)
 155                 net->ipv6.rt6_stats->fib_nodes++;
 156
 157         return fn;
 158 }
 159
 160 static void node_free_immediate(struct net *net, struct fib6_node *fn)
 161 {
 162         kmem_cache_free(fib6_node_kmem, fn);
 163         net->ipv6.rt6_stats->fib_nodes--;
 164 }
 165
 166 static void node_free_rcu(struct rcu_head *head)
 167 {
 168         struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
 169
 170         kmem_cache_free(fib6_node_kmem, fn);
 171 }
 172
 173 static void node_free(struct net *net, struct fib6_node *fn)
 174 {
 175         call_rcu(&fn->rcu, node_free_rcu);
 176         net->ipv6.rt6_stats->fib_nodes--;
 177 }
 178
 179 void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
 180 {
 181         int cpu;
 182
 183         if (!non_pcpu_rt->rt6i_pcpu)
 184                 return;
 185
 186         for_each_possible_cpu(cpu) {
 187                 struct rt6_info **ppcpu_rt;
 188                 struct rt6_info *pcpu_rt;
 189
 190                 ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
 191                 pcpu_rt = *ppcpu_rt;
 192                 if (pcpu_rt) {
 193                         dst_dev_put(&pcpu_rt->dst);
 194                         dst_release(&pcpu_rt->dst);
 195                         *ppcpu_rt = NULL;
 196                 }
 197         }
 198 }
 199 EXPORT_SYMBOL_GPL(rt6_free_pcpu);
 200
 201 static void fib6_free_table(struct fib6_table *table)
 202 {
 203         inetpeer_invalidate_tree(&table->tb6_peers);
 204         kfree(table);
 205 }
 206
 207 static void fib6_link_table(struct net *net, struct fib6_table *tb)
 208 {
 209         unsigned int h;
 210
 211         /*
 212          * Initialize table lock at a single place to give lockdep a key,
 213          * tables aren't visible prior to being linked to the list.
 214          */
 215         spin_lock_init(&tb->tb6_lock);
 216         h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
 217
 218         /*
 219          * No protection necessary, this is the only list mutatation
 220          * operation, tables never disappear once they exist.
 221          */
 222         hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
 223 }
 224
 225 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 226
 227 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
 228 {
 229         struct fib6_table *table;
 230
 231         table = kzalloc(sizeof(*table), GFP_ATOMIC);
 232         if (table) {
 233                 table->tb6_id = id;
 234                 rcu_assign_pointer(table->tb6_root.leaf,
 235                                    net->ipv6.ip6_null_entry);
 236                 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
 237                 inet_peer_base_init(&table->tb6_peers);
 238         }
 239
 240         return table;
 241 }
 242
 243 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 244 {
 245         struct fib6_table *tb;
 246
 247         if (id == 0)
 248                 id = RT6_TABLE_MAIN;
 249         tb = fib6_get_table(net, id);
 250         if (tb)
 251                 return tb;
 252
 253         tb = fib6_alloc_table(net, id);
 254         if (tb)
 255                 fib6_link_table(net, tb);
 256
 257         return tb;
 258 }
 259 EXPORT_SYMBOL_GPL(fib6_new_table);
 260
 261 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 262 {
 263         struct fib6_table *tb;
 264         struct hlist_head *head;
 265         unsigned int h;
 266
 267         if (id == 0)
 268                 id = RT6_TABLE_MAIN;
 269         h = id & (FIB6_TABLE_HASHSZ - 1);
 270         rcu_read_lock();
 271         head = &net->ipv6.fib_table_hash[h];
 272         hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 273                 if (tb->tb6_id == id) {
 274                         rcu_read_unlock();
 275                         return tb;
 276                 }
 277         }
 278         rcu_read_unlock();
 279
 280         return NULL;
 281 }
 282 EXPORT_SYMBOL_GPL(fib6_get_table);
 283
 284 static void __net_init fib6_tables_init(struct net *net)
 285 {
 286         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 287         fib6_link_table(net, net->ipv6.fib6_local_tbl);
 288 }
 289 #else
 290
 291 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 292 {
 293         return fib6_get_table(net, id);
 294 }
 295
 296 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 297 {
 298           return net->ipv6.fib6_main_tbl;
 299 }
 300
 301 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 302                                    const struct sk_buff *skb,
 303                                    int flags, pol_lookup_t lookup)
 304 {
 305         struct rt6_info *rt;
 306
 307         rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
 308         if (rt->dst.error == -EAGAIN) {
 309                 ip6_rt_put(rt);
 310                 rt = net->ipv6.ip6_null_entry;
 311                 dst_hold(&rt->dst);
 312         }
 313
 314         return &rt->dst;
 315 }
 316
 317 static void __net_init fib6_tables_init(struct net *net)
 318 {
 319         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 320 }
 321
 322 #endif
 323
 324 unsigned int fib6_tables_seq_read(struct net *net)
 325 {
 326         unsigned int h, fib_seq = 0;
 327
 328         rcu_read_lock();
 329         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 330                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 331                 struct fib6_table *tb;
 332
 333                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 334                         fib_seq += tb->fib_seq;
 335         }
 336         rcu_read_unlock();
 337
 338         return fib_seq;
 339 }
 340
 341 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
 342                                     enum fib_event_type event_type,
 343                                     struct rt6_info *rt)
 344 {
 345         struct fib6_entry_notifier_info info = {
 346                 .rt = rt,
 347         };
 348
 349         return call_fib6_notifier(nb, net, event_type, &info.info);
 350 }
 351
 352 static int call_fib6_entry_notifiers(struct net *net,
 353                                      enum fib_event_type event_type,
 354                                      struct rt6_info *rt,
 355                                      struct netlink_ext_ack *extack)
 356 {
 357         struct fib6_entry_notifier_info info = {
 358                 .info.extack = extack,
 359                 .rt = rt,
 360         };
 361
 362         rt->rt6i_table->fib_seq++;
 363         return call_fib6_notifiers(net, event_type, &info.info);
 364 }
 365
 366 struct fib6_dump_arg {
 367         struct net *net;
 368         struct notifier_block *nb;
 369 };
 370
 371 static void fib6_rt_dump(struct rt6_info *rt, struct fib6_dump_arg *arg)
 372 {
 373         if (rt == arg->net->ipv6.ip6_null_entry)
 374                 return;
 375         call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
 376 }
 377
 378 static int fib6_node_dump(struct fib6_walker *w)
 379 {
 380         struct rt6_info *rt;
 381
 382         for_each_fib6_walker_rt(w)
 383                 fib6_rt_dump(rt, w->args);
 384         w->leaf = NULL;
 385         return 0;
 386 }
 387
 388 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
 389                             struct fib6_walker *w)
 390 {
 391         w->root = &tb->tb6_root;
 392         spin_lock_bh(&tb->tb6_lock);
 393         fib6_walk(net, w);
 394         spin_unlock_bh(&tb->tb6_lock);
 395 }
 396
 397 /* Called with rcu_read_lock() */
 398 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
 399 {
 400         struct fib6_dump_arg arg;
 401         struct fib6_walker *w;
 402         unsigned int h;
 403
 404         w = kzalloc(sizeof(*w), GFP_ATOMIC);
 405         if (!w)
 406                 return -ENOMEM;
 407
 408         w->func = fib6_node_dump;
 409         arg.net = net;
 410         arg.nb = nb;
 411         w->args = &arg;
 412
 413         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 414                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 415                 struct fib6_table *tb;
 416
 417                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 418                         fib6_table_dump(net, tb, w);
 419         }
 420
 421         kfree(w);
 422
 423         return 0;
 424 }
 425
 426 static int fib6_dump_node(struct fib6_walker *w)
 427 {
 428         int res;
 429         struct rt6_info *rt;
 430
 431         for_each_fib6_walker_rt(w) {
 432                 res = rt6_dump_route(rt, w->args);
 433                 if (res < 0) {
 434                         /* Frame is full, suspend walking */
 435                         w->leaf = rt;
 436                         return 1;
 437                 }
 438
 439                 /* Multipath routes are dumped in one route with the
 440                  * RTA_MULTIPATH attribute. Jump 'rt' to point to the
 441                  * last sibling of this route (no need to dump the
 442                  * sibling routes again)
 443                  */
 444                 if (rt->rt6i_nsiblings)
 445                         rt = list_last_entry(&rt->rt6i_siblings,
 446                                              struct rt6_info,
 447                                              rt6i_siblings);
 448         }
 449         w->leaf = NULL;
 450         return 0;
 451 }
 452
 453 static void fib6_dump_end(struct netlink_callback *cb)
 454 {
 455         struct net *net = sock_net(cb->skb->sk);
 456         struct fib6_walker *w = (void *)cb->args[2];
 457
 458         if (w) {
 459                 if (cb->args[4]) {
 460                         cb->args[4] = 0;
 461                         fib6_walker_unlink(net, w);
 462                 }
 463                 cb->args[2] = 0;
 464                 kfree(w);
 465         }
 466         cb->done = (void *)cb->args[3];
 467         cb->args[1] = 3;
 468 }
 469
 470 static int fib6_dump_done(struct netlink_callback *cb)
 471 {
 472         fib6_dump_end(cb);
 473         return cb->done ? cb->done(cb) : 0;
 474 }
 475
 476 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
 477                            struct netlink_callback *cb)
 478 {
 479         struct net *net = sock_net(skb->sk);
 480         struct fib6_walker *w;
 481         int res;
 482
 483         w = (void *)cb->args[2];
 484         w->root = &table->tb6_root;
 485
 486         if (cb->args[4] == 0) {
 487                 w->count = 0;
 488                 w->skip = 0;
 489
 490                 spin_lock_bh(&table->tb6_lock);
 491                 res = fib6_walk(net, w);
 492                 spin_unlock_bh(&table->tb6_lock);
 493                 if (res > 0) {
 494                         cb->args[4] = 1;
 495                         cb->args[5] = w->root->fn_sernum;
 496                 }
 497         } else {
 498                 if (cb->args[5] != w->root->fn_sernum) {
 499                         /* Begin at the root if the tree changed */
 500                         cb->args[5] = w->root->fn_sernum;
 501                         w->state = FWS_INIT;
 502                         w->node = w->root;
 503                         w->skip = w->count;
 504                 } else
 505                         w->skip = 0;
 506
 507                 spin_lock_bh(&table->tb6_lock);
 508                 res = fib6_walk_continue(w);
 509                 spin_unlock_bh(&table->tb6_lock);
 510                 if (res <= 0) {
 511                         fib6_walker_unlink(net, w);
 512                         cb->args[4] = 0;
 513                 }
 514         }
 515
 516         return res;
 517 }
 518
 519 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 520 {
 521         struct net *net = sock_net(skb->sk);
 522         unsigned int h, s_h;
 523         unsigned int e = 0, s_e;
 524         struct rt6_rtnl_dump_arg arg;
 525         struct fib6_walker *w;
 526         struct fib6_table *tb;
 527         struct hlist_head *head;
 528         int res = 0;
 529
 530         s_h = cb->args[0];
 531         s_e = cb->args[1];
 532
 533         w = (void *)cb->args[2];
 534         if (!w) {
 535                 /* New dump:
 536                  *
 537                  * 1. hook callback destructor.
 538                  */
 539                 cb->args[3] = (long)cb->done;
 540                 cb->done = fib6_dump_done;
 541
 542                 /*
 543                  * 2. allocate and initialize walker.
 544                  */
 545                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
 546                 if (!w)
 547                         return -ENOMEM;
 548                 w->func = fib6_dump_node;
 549                 cb->args[2] = (long)w;
 550         }
 551
 552         arg.skb = skb;
 553         arg.cb = cb;
 554         arg.net = net;
 555         w->args = &arg;
 556
 557         rcu_read_lock();
 558         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
 559                 e = 0;
 560                 head = &net->ipv6.fib_table_hash[h];
 561                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 562                         if (e < s_e)
 563                                 goto next;
 564                         res = fib6_dump_table(tb, skb, cb);
 565                         if (res != 0)
 566                                 goto out;
 567 next:
 568                         e++;
 569                 }
 570         }
 571 out:
 572         rcu_read_unlock();
 573         cb->args[1] = e;
 574         cb->args[0] = h;
 575
 576         res = res < 0 ? res : skb->len;
 577         if (res <= 0)
 578                 fib6_dump_end(cb);
 579         return res;
 580 }
 581
 582 /*
 583  *      Routing Table
 584  *
 585  *      return the appropriate node for a routing tree "add" operation
 586  *      by either creating and inserting or by returning an existing
 587  *      node.
 588  */
 589
 590 static struct fib6_node *fib6_add_1(struct net *net,
 591                                     struct fib6_table *table,
 592                                     struct fib6_node *root,
 593                                     struct in6_addr *addr, int plen,
 594                                     int offset, int allow_create,
 595                                     int replace_required,
 596                                     struct netlink_ext_ack *extack)
 597 {
 598         struct fib6_node *fn, *in, *ln;
 599         struct fib6_node *pn = NULL;
 600         struct rt6key *key;
 601         int     bit;
 602         __be32  dir = 0;
 603
 604         RT6_TRACE("fib6_add_1\n");
 605
 606         /* insert node in tree */
 607
 608         fn = root;
 609
 610         do {
 611                 struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
 612                                             lockdep_is_held(&table->tb6_lock));
 613                 key = (struct rt6key *)((u8 *)leaf + offset);
 614
 615                 /*
 616                  *      Prefix match
 617                  */
 618                 if (plen < fn->fn_bit ||
 619                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
 620                         if (!allow_create) {
 621                                 if (replace_required) {
 622                                         NL_SET_ERR_MSG(extack,
 623                                                        "Can not replace route - no match found");
 624                                         pr_warn("Can't replace route, no match found\n");
 625                                         return ERR_PTR(-ENOENT);
 626                                 }
 627                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 628                         }
 629                         goto insert_above;
 630                 }
 631
 632                 /*
 633                  *      Exact match ?
 634                  */
 635
 636                 if (plen == fn->fn_bit) {
 637                         /* clean up an intermediate node */
 638                         if (!(fn->fn_flags & RTN_RTINFO)) {
 639                                 RCU_INIT_POINTER(fn->leaf, NULL);
 640                                 rt6_release(leaf);
 641                         /* remove null_entry in the root node */
 642                         } else if (fn->fn_flags & RTN_TL_ROOT &&
 643                                    rcu_access_pointer(fn->leaf) ==
 644                                    net->ipv6.ip6_null_entry) {
 645                                 RCU_INIT_POINTER(fn->leaf, NULL);
 646                         }
 647
 648                         return fn;
 649                 }
 650
 651                 /*
 652                  *      We have more bits to go
 653                  */
 654
 655                 /* Try to walk down on tree. */
 656                 dir = addr_bit_set(addr, fn->fn_bit);
 657                 pn = fn;
 658                 fn = dir ?
 659                      rcu_dereference_protected(fn->right,
 660                                         lockdep_is_held(&table->tb6_lock)) :
 661                      rcu_dereference_protected(fn->left,
 662                                         lockdep_is_held(&table->tb6_lock));
 663         } while (fn);
 664
 665         if (!allow_create) {
 666                 /* We should not create new node because
 667                  * NLM_F_REPLACE was specified without NLM_F_CREATE
 668                  * I assume it is safe to require NLM_F_CREATE when
 669                  * REPLACE flag is used! Later we may want to remove the
 670                  * check for replace_required, because according
 671                  * to netlink specification, NLM_F_CREATE
 672                  * MUST be specified if new route is created.
 673                  * That would keep IPv6 consistent with IPv4
 674                  */
 675                 if (replace_required) {
 676                         NL_SET_ERR_MSG(extack,
 677                                        "Can not replace route - no match found");
 678                         pr_warn("Can't replace route, no match found\n");
 679                         return ERR_PTR(-ENOENT);
 680                 }
 681                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 682         }
 683         /*
 684          *      We walked to the bottom of tree.
 685          *      Create new leaf node without children.
 686          */
 687
 688         ln = node_alloc(net);
 689
 690         if (!ln)
 691                 return ERR_PTR(-ENOMEM);
 692         ln->fn_bit = plen;
 693         RCU_INIT_POINTER(ln->parent, pn);
 694
 695         if (dir)
 696                 rcu_assign_pointer(pn->right, ln);
 697         else
 698                 rcu_assign_pointer(pn->left, ln);
 699
 700         return ln;
 701
 702
 703 insert_above:
 704         /*
 705          * split since we don't have a common prefix anymore or
 706          * we have a less significant route.
 707          * we've to insert an intermediate node on the list
 708          * this new node will point to the one we need to create
 709          * and the current
 710          */
 711
 712         pn = rcu_dereference_protected(fn->parent,
 713                                        lockdep_is_held(&table->tb6_lock));
 714
 715         /* find 1st bit in difference between the 2 addrs.
 716
 717            See comment in __ipv6_addr_diff: bit may be an invalid value,
 718            but if it is >= plen, the value is ignored in any case.
 719          */
 720
 721         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
 722
 723         /*
 724          *              (intermediate)[in]
 725          *                /        \
 726          *      (new leaf node)[ln] (old node)[fn]
 727          */
 728         if (plen > bit) {
 729                 in = node_alloc(net);
 730                 ln = node_alloc(net);
 731
 732                 if (!in || !ln) {
 733                         if (in)
 734                                 node_free_immediate(net, in);
 735                         if (ln)
 736                                 node_free_immediate(net, ln);
 737                         return ERR_PTR(-ENOMEM);
 738                 }
 739
 740                 /*
 741                  * new intermediate node.
 742                  * RTN_RTINFO will
 743                  * be off since that an address that chooses one of
 744                  * the branches would not match less specific routes
 745                  * in the other branch
 746                  */
 747
 748                 in->fn_bit = bit;
 749
 750                 RCU_INIT_POINTER(in->parent, pn);
 751                 in->leaf = fn->leaf;
 752                 atomic_inc(&rcu_dereference_protected(in->leaf,
 753                                 lockdep_is_held(&table->tb6_lock))->rt6i_ref);
 754
 755                 /* update parent pointer */
 756                 if (dir)
 757                         rcu_assign_pointer(pn->right, in);
 758                 else
 759                         rcu_assign_pointer(pn->left, in);
 760
 761                 ln->fn_bit = plen;
 762
 763                 RCU_INIT_POINTER(ln->parent, in);
 764                 rcu_assign_pointer(fn->parent, in);
 765
 766                 if (addr_bit_set(addr, bit)) {
 767                         rcu_assign_pointer(in->right, ln);
 768                         rcu_assign_pointer(in->left, fn);
 769                 } else {
 770                         rcu_assign_pointer(in->left, ln);
 771                         rcu_assign_pointer(in->right, fn);
 772                 }
 773         } else { /* plen <= bit */
 774
 775                 /*
 776                  *              (new leaf node)[ln]
 777                  *                /        \
 778                  *           (old node)[fn] NULL
 779                  */
 780
 781                 ln = node_alloc(net);
 782
 783                 if (!ln)
 784                         return ERR_PTR(-ENOMEM);
 785
 786                 ln->fn_bit = plen;
 787
 788                 RCU_INIT_POINTER(ln->parent, pn);
 789
 790                 if (addr_bit_set(&key->addr, plen))
 791                         RCU_INIT_POINTER(ln->right, fn);
 792                 else
 793                         RCU_INIT_POINTER(ln->left, fn);
 794
 795                 rcu_assign_pointer(fn->parent, ln);
 796
 797                 if (dir)
 798                         rcu_assign_pointer(pn->right, ln);
 799                 else
 800                         rcu_assign_pointer(pn->left, ln);
 801         }
 802         return ln;
 803 }
 804
 805 static void fib6_copy_metrics(u32 *mp, const struct mx6_config *mxc)
 806 {
 807         int i;
 808
 809         for (i = 0; i < RTAX_MAX; i++) {
 810                 if (test_bit(i, mxc->mx_valid))
 811                         mp[i] = mxc->mx[i];
 812         }
 813 }
 814
 815 static int fib6_commit_metrics(struct dst_entry *dst, struct mx6_config *mxc)
 816 {
 817         if (!mxc->mx)
 818                 return 0;
 819
 820         if (dst->flags & DST_HOST) {
 821                 u32 *mp = dst_metrics_write_ptr(dst);
 822
 823                 if (unlikely(!mp))
 824                         return -ENOMEM;
 825
 826                 fib6_copy_metrics(mp, mxc);
 827         } else {
 828                 dst_init_metrics(dst, mxc->mx, false);
 829
 830                 /* We've stolen mx now. */
 831                 mxc->mx = NULL;
 832         }
 833
 834         return 0;
 835 }
 836
 837 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
 838                           struct net *net)
 839 {
 840         struct fib6_table *table = rt->rt6i_table;
 841
 842         if (atomic_read(&rt->rt6i_ref) != 1) {
 843                 /* This route is used as dummy address holder in some split
 844                  * nodes. It is not leaked, but it still holds other resources,
 845                  * which must be released in time. So, scan ascendant nodes
 846                  * and replace dummy references to this route with references
 847                  * to still alive ones.
 848                  */
 849                 while (fn) {
 850                         struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
 851                                             lockdep_is_held(&table->tb6_lock));
 852                         struct rt6_info *new_leaf;
 853                         if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
 854                                 new_leaf = fib6_find_prefix(net, table, fn);
 855                                 atomic_inc(&new_leaf->rt6i_ref);
 856                                 rcu_assign_pointer(fn->leaf, new_leaf);
 857                                 rt6_release(rt);
 858                         }
 859                         fn = rcu_dereference_protected(fn->parent,
 860                                     lockdep_is_held(&table->tb6_lock));
 861                 }
 862         }
 863 }
 864
 865 /*
 866  *      Insert routing information in a node.
 867  */
 868
 869 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
 870                             struct nl_info *info, struct mx6_config *mxc,
 871                             struct netlink_ext_ack *extack)
 872 {
 873         struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
 874                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
 875         struct rt6_info *iter = NULL;
 876         struct rt6_info __rcu **ins;
 877         struct rt6_info __rcu **fallback_ins = NULL;
 878         int replace = (info->nlh &&
 879                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
 880         int add = (!info->nlh ||
 881                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
 882         int found = 0;
 883         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
 884         u16 nlflags = NLM_F_EXCL;
 885         int err;
 886
 887         if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
 888                 nlflags |= NLM_F_APPEND;
 889
 890         ins = &fn->leaf;
 891
 892         for (iter = leaf; iter;
 893              iter = rcu_dereference_protected(iter->rt6_next,
 894                                 lockdep_is_held(&rt->rt6i_table->tb6_lock))) {
 895                 /*
 896                  *      Search for duplicates
 897                  */
 898
 899                 if (iter->rt6i_metric == rt->rt6i_metric) {
 900                         /*
 901                          *      Same priority level
 902                          */
 903                         if (info->nlh &&
 904                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
 905                                 return -EEXIST;
 906
 907                         nlflags &= ~NLM_F_EXCL;
 908                         if (replace) {
 909                                 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
 910                                         found++;
 911                                         break;
 912                                 }
 913                                 if (rt_can_ecmp)
 914                                         fallback_ins = fallback_ins ?: ins;
 915                                 goto next_iter;
 916                         }
 917
 918                         if (rt6_duplicate_nexthop(iter, rt)) {
 919                                 if (rt->rt6i_nsiblings)
 920                                         rt->rt6i_nsiblings = 0;
 921                                 if (!(iter->rt6i_flags & RTF_EXPIRES))
 922                                         return -EEXIST;
 923                                 if (!(rt->rt6i_flags & RTF_EXPIRES))
 924                                         rt6_clean_expires(iter);
 925                                 else
 926                                         rt6_set_expires(iter, rt->dst.expires);
 927                                 iter->rt6i_pmtu = rt->rt6i_pmtu;
 928                                 return -EEXIST;
 929                         }
 930                         /* If we have the same destination and the same metric,
 931                          * but not the same gateway, then the route we try to
 932                          * add is sibling to this route, increment our counter
 933                          * of siblings, and later we will add our route to the
 934                          * list.
 935                          * Only static routes (which don't have flag
 936                          * RTF_EXPIRES) are used for ECMPv6.
 937                          */
 938                         if (rt_can_ecmp &&
 939                             rt6_qualify_for_ecmp(iter))
 940                                 rt->rt6i_nsiblings++;
 941                 }
 942
 943                 if (iter->rt6i_metric > rt->rt6i_metric)
 944                         break;
 945
 946 next_iter:
 947                 ins = &iter->rt6_next;
 948         }
 949
 950         if (fallback_ins && !found) {
 951                 /* No ECMP-able route found, replace first non-ECMP one */
 952                 ins = fallback_ins;
 953                 iter = rcu_dereference_protected(*ins,
 954                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
 955                 found++;
 956         }
 957
 958         /* Reset round-robin state, if necessary */
 959         if (ins == &fn->leaf)
 960                 fn->rr_ptr = NULL;
 961
 962         /* Link this route to others same route. */
 963         if (rt->rt6i_nsiblings) {
 964                 unsigned int rt6i_nsiblings;
 965                 struct rt6_info *sibling, *temp_sibling;
 966
 967                 /* Find the first route that have the same metric */
 968                 sibling = leaf;
 969                 while (sibling) {
 970                         if (sibling->rt6i_metric == rt->rt6i_metric &&
 971                             rt6_qualify_for_ecmp(sibling)) {
 972                                 list_add_tail(&rt->rt6i_siblings,
 973                                               &sibling->rt6i_siblings);
 974                                 break;
 975                         }
 976                         sibling = rcu_dereference_protected(sibling->rt6_next,
 977                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
 978                 }
 979                 /* For each sibling in the list, increment the counter of
 980                  * siblings. BUG() if counters does not match, list of siblings
 981                  * is broken!
 982                  */
 983                 rt6i_nsiblings = 0;
 984                 list_for_each_entry_safe(sibling, temp_sibling,
 985                                          &rt->rt6i_siblings, rt6i_siblings) {
 986                         sibling->rt6i_nsiblings++;
 987                         BUG_ON(sibling->rt6i_nsiblings != rt->rt6i_nsiblings);
 988                         rt6i_nsiblings++;
 989                 }
 990                 BUG_ON(rt6i_nsiblings != rt->rt6i_nsiblings);
 991                 rt6_multipath_rebalance(temp_sibling);
 992         }
 993
 994         /*
 995          *      insert node
 996          */
 997         if (!replace) {
 998                 if (!add)
 999                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
1000
1001 add:
1002                 nlflags |= NLM_F_CREATE;
1003                 err = fib6_commit_metrics(&rt->dst, mxc);
1004                 if (err)
1005                         return err;
1006
1007                 err = call_fib6_entry_notifiers(info->nl_net,
1008                                                 FIB_EVENT_ENTRY_ADD,
1009                                                 rt, extack);
1010                 if (err)
1011                         return err;
1012
1013                 rcu_assign_pointer(rt->rt6_next, iter);
1014                 atomic_inc(&rt->rt6i_ref);
1015                 rcu_assign_pointer(rt->rt6i_node, fn);
1016                 rcu_assign_pointer(*ins, rt);
1017                 if (!info->skip_notify)
1018                         inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1019                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1020
1021                 if (!(fn->fn_flags & RTN_RTINFO)) {
1022                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1023                         fn->fn_flags |= RTN_RTINFO;
1024                 }
1025
1026         } else {
1027                 int nsiblings;
1028
1029                 if (!found) {
1030                         if (add)
1031                                 goto add;
1032                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1033                         return -ENOENT;
1034                 }
1035
1036                 err = fib6_commit_metrics(&rt->dst, mxc);
1037                 if (err)
1038                         return err;
1039
1040                 err = call_fib6_entry_notifiers(info->nl_net,
1041                                                 FIB_EVENT_ENTRY_REPLACE,
1042                                                 rt, extack);
1043                 if (err)
1044                         return err;
1045
1046                 atomic_inc(&rt->rt6i_ref);
1047                 rcu_assign_pointer(rt->rt6i_node, fn);
1048                 rt->rt6_next = iter->rt6_next;
1049                 rcu_assign_pointer(*ins, rt);
1050                 if (!info->skip_notify)
1051                         inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1052                 if (!(fn->fn_flags & RTN_RTINFO)) {
1053                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1054                         fn->fn_flags |= RTN_RTINFO;
1055                 }
1056                 nsiblings = iter->rt6i_nsiblings;
1057                 iter->rt6i_node = NULL;
1058                 fib6_purge_rt(iter, fn, info->nl_net);
1059                 if (rcu_access_pointer(fn->rr_ptr) == iter)
1060                         fn->rr_ptr = NULL;
1061                 rt6_release(iter);
1062
1063                 if (nsiblings) {
1064                         /* Replacing an ECMP route, remove all siblings */
1065                         ins = &rt->rt6_next;
1066                         iter = rcu_dereference_protected(*ins,
1067                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
1068                         while (iter) {
1069                                 if (iter->rt6i_metric > rt->rt6i_metric)
1070                                         break;
1071                                 if (rt6_qualify_for_ecmp(iter)) {
1072                                         *ins = iter->rt6_next;
1073                                         iter->rt6i_node = NULL;
1074                                         fib6_purge_rt(iter, fn, info->nl_net);
1075                                         if (rcu_access_pointer(fn->rr_ptr) == iter)
1076                                                 fn->rr_ptr = NULL;
1077                                         rt6_release(iter);
1078                                         nsiblings--;
1079                                         info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1080                                 } else {
1081                                         ins = &iter->rt6_next;
1082                                 }
1083                                 iter = rcu_dereference_protected(*ins,
1084                                         lockdep_is_held(&rt->rt6i_table->tb6_lock));
1085                         }
1086                         WARN_ON(nsiblings != 0);
1087                 }
1088         }
1089
1090         return 0;
1091 }
1092
1093 static void fib6_start_gc(struct net *net, struct rt6_info *rt)
1094 {
1095         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1096             (rt->rt6i_flags & (RTF_EXPIRES | RTF_CACHE)))
1097                 mod_timer(&net->ipv6.ip6_fib_timer,
1098                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1099 }
1100
1101 void fib6_force_start_gc(struct net *net)
1102 {
1103         if (!timer_pending(&net->ipv6.ip6_fib_timer))
1104                 mod_timer(&net->ipv6.ip6_fib_timer,
1105                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1106 }
1107
1108 static void __fib6_update_sernum_upto_root(struct rt6_info *rt,
1109                                            int sernum)
1110 {
1111         struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1112                                 lockdep_is_held(&rt->rt6i_table->tb6_lock));
1113
1114         /* paired with smp_rmb() in rt6_get_cookie_safe() */
1115         smp_wmb();
1116         while (fn) {
1117                 fn->fn_sernum = sernum;
1118                 fn = rcu_dereference_protected(fn->parent,
1119                                 lockdep_is_held(&rt->rt6i_table->tb6_lock));
1120         }
1121 }
1122
1123 void fib6_update_sernum_upto_root(struct net *net, struct rt6_info *rt)
1124 {
1125         __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1126 }
1127
1128 /*
1129  *      Add routing information to the routing tree.
1130  *      <destination addr>/<source addr>
1131  *      with source addr info in sub-trees
1132  *      Need to own table->tb6_lock
1133  */
1134
1135 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
1136              struct nl_info *info, struct mx6_config *mxc,
1137              struct netlink_ext_ack *extack)
1138 {
1139         struct fib6_table *table = rt->rt6i_table;
1140         struct fib6_node *fn, *pn = NULL;
1141         int err = -ENOMEM;
1142         int allow_create = 1;
1143         int replace_required = 0;
1144         int sernum = fib6_new_sernum(info->nl_net);
1145
1146         if (WARN_ON_ONCE(!atomic_read(&rt->dst.__refcnt)))
1147                 return -EINVAL;
1148         if (WARN_ON_ONCE(rt->rt6i_flags & RTF_CACHE))
1149                 return -EINVAL;
1150
1151         if (info->nlh) {
1152                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1153                         allow_create = 0;
1154                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1155                         replace_required = 1;
1156         }
1157         if (!allow_create && !replace_required)
1158                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1159
1160         fn = fib6_add_1(info->nl_net, table, root,
1161                         &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
1162                         offsetof(struct rt6_info, rt6i_dst), allow_create,
1163                         replace_required, extack);
1164         if (IS_ERR(fn)) {
1165                 err = PTR_ERR(fn);
1166                 fn = NULL;
1167                 goto out;
1168         }
1169
1170         pn = fn;
1171
1172 #ifdef CONFIG_IPV6_SUBTREES
1173         if (rt->rt6i_src.plen) {
1174                 struct fib6_node *sn;
1175
1176                 if (!rcu_access_pointer(fn->subtree)) {
1177                         struct fib6_node *sfn;
1178
1179                         /*
1180                          * Create subtree.
1181                          *
1182                          *              fn[main tree]
1183                          *              |
1184                          *              sfn[subtree root]
1185                          *                 \
1186                          *                  sn[new leaf node]
1187                          */
1188
1189                         /* Create subtree root node */
1190                         sfn = node_alloc(info->nl_net);
1191                         if (!sfn)
1192                                 goto failure;
1193
1194                         atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
1195                         rcu_assign_pointer(sfn->leaf,
1196                                            info->nl_net->ipv6.ip6_null_entry);
1197                         sfn->fn_flags = RTN_ROOT;
1198
1199                         /* Now add the first leaf node to new subtree */
1200
1201                         sn = fib6_add_1(info->nl_net, table, sfn,
1202                                         &rt->rt6i_src.addr, rt->rt6i_src.plen,
1203                                         offsetof(struct rt6_info, rt6i_src),
1204                                         allow_create, replace_required, extack);
1205
1206                         if (IS_ERR(sn)) {
1207                                 /* If it is failed, discard just allocated
1208                                    root, and then (in failure) stale node
1209                                    in main tree.
1210                                  */
1211                                 node_free_immediate(info->nl_net, sfn);
1212                                 err = PTR_ERR(sn);
1213                                 goto failure;
1214                         }
1215
1216                         /* Now link new subtree to main tree */
1217                         rcu_assign_pointer(sfn->parent, fn);
1218                         rcu_assign_pointer(fn->subtree, sfn);
1219                 } else {
1220                         sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1221                                         &rt->rt6i_src.addr, rt->rt6i_src.plen,
1222                                         offsetof(struct rt6_info, rt6i_src),
1223                                         allow_create, replace_required, extack);
1224
1225                         if (IS_ERR(sn)) {
1226                                 err = PTR_ERR(sn);
1227                                 goto failure;
1228                         }
1229                 }
1230
1231                 if (!rcu_access_pointer(fn->leaf)) {
1232                         if (fn->fn_flags & RTN_TL_ROOT) {
1233                                 /* put back null_entry for root node */
1234                                 rcu_assign_pointer(fn->leaf,
1235                                             info->nl_net->ipv6.ip6_null_entry);
1236                         } else {
1237                                 atomic_inc(&rt->rt6i_ref);
1238                                 rcu_assign_pointer(fn->leaf, rt);
1239                         }
1240                 }
1241                 fn = sn;
1242         }
1243 #endif
1244
1245         err = fib6_add_rt2node(fn, rt, info, mxc, extack);
1246         if (!err) {
1247                 __fib6_update_sernum_upto_root(rt, sernum);
1248                 fib6_start_gc(info->nl_net, rt);
1249         }
1250
1251 out:
1252         if (err) {
1253 #ifdef CONFIG_IPV6_SUBTREES
1254                 /*
1255                  * If fib6_add_1 has cleared the old leaf pointer in the
1256                  * super-tree leaf node we have to find a new one for it.
1257                  */
1258                 if (pn != fn) {
1259                         struct rt6_info *pn_leaf =
1260                                 rcu_dereference_protected(pn->leaf,
1261                                     lockdep_is_held(&table->tb6_lock));
1262                         if (pn_leaf == rt) {
1263                                 pn_leaf = NULL;
1264                                 RCU_INIT_POINTER(pn->leaf, NULL);
1265                                 atomic_dec(&rt->rt6i_ref);
1266                         }
1267                         if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1268                                 pn_leaf = fib6_find_prefix(info->nl_net, table,
1269                                                            pn);
1270 #if RT6_DEBUG >= 2
1271                                 if (!pn_leaf) {
1272                                         WARN_ON(!pn_leaf);
1273                                         pn_leaf =
1274                                             info->nl_net->ipv6.ip6_null_entry;
1275                                 }
1276 #endif
1277                                 atomic_inc(&pn_leaf->rt6i_ref);
1278                                 rcu_assign_pointer(pn->leaf, pn_leaf);
1279                         }
1280                 }
1281 #endif
1282                 goto failure;
1283         }
1284         return err;
1285
1286 failure:
1287         /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1288          * 1. fn is an intermediate node and we failed to add the new
1289          * route to it in both subtree creation failure and fib6_add_rt2node()
1290          * failure case.
1291          * 2. fn is the root node in the table and we fail to add the first
1292          * default route to it.
1293          */
1294         if (fn &&
1295             (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1296              (fn->fn_flags & RTN_TL_ROOT &&
1297               !rcu_access_pointer(fn->leaf))))
1298                 fib6_repair_tree(info->nl_net, table, fn);
1299         /* Always release dst as dst->__refcnt is guaranteed
1300          * to be taken before entering this function
1301          */
1302         dst_release_immediate(&rt->dst);
1303         return err;
1304 }
1305
1306 /*
1307  *      Routing tree lookup
1308  *
1309  */
1310
1311 struct lookup_args {
1312         int                     offset;         /* key offset on rt6_info       */
1313         const struct in6_addr   *addr;          /* search key                   */
1314 };
1315
1316 static struct fib6_node *fib6_lookup_1(struct fib6_node *root,
1317                                        struct lookup_args *args)
1318 {
1319         struct fib6_node *fn;
1320         __be32 dir;
1321
1322         if (unlikely(args->offset == 0))
1323                 return NULL;
1324
1325         /*
1326          *      Descend on a tree
1327          */
1328
1329         fn = root;
1330
1331         for (;;) {
1332                 struct fib6_node *next;
1333
1334                 dir = addr_bit_set(args->addr, fn->fn_bit);
1335
1336                 next = dir ? rcu_dereference(fn->right) :
1337                              rcu_dereference(fn->left);
1338
1339                 if (next) {
1340                         fn = next;
1341                         continue;
1342                 }
1343                 break;
1344         }
1345
1346         while (fn) {
1347                 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1348
1349                 if (subtree || fn->fn_flags & RTN_RTINFO) {
1350                         struct rt6_info *leaf = rcu_dereference(fn->leaf);
1351                         struct rt6key *key;
1352
1353                         if (!leaf)
1354                                 goto backtrack;
1355
1356                         key = (struct rt6key *) ((u8 *)leaf + args->offset);
1357
1358                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1359 #ifdef CONFIG_IPV6_SUBTREES
1360                                 if (subtree) {
1361                                         struct fib6_node *sfn;
1362                                         sfn = fib6_lookup_1(subtree, args + 1);
1363                                         if (!sfn)
1364                                                 goto backtrack;
1365                                         fn = sfn;
1366                                 }
1367 #endif
1368                                 if (fn->fn_flags & RTN_RTINFO)
1369                                         return fn;
1370                         }
1371                 }
1372 backtrack:
1373                 if (fn->fn_flags & RTN_ROOT)
1374                         break;
1375
1376                 fn = rcu_dereference(fn->parent);
1377         }
1378
1379         return NULL;
1380 }
1381
1382 /* called with rcu_read_lock() held
1383  */
1384 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
1385                               const struct in6_addr *saddr)
1386 {
1387         struct fib6_node *fn;
1388         struct lookup_args args[] = {
1389                 {
1390                         .offset = offsetof(struct rt6_info, rt6i_dst),
1391                         .addr = daddr,
1392                 },
1393 #ifdef CONFIG_IPV6_SUBTREES
1394                 {
1395                         .offset = offsetof(struct rt6_info, rt6i_src),
1396                         .addr = saddr,
1397                 },
1398 #endif
1399                 {
1400                         .offset = 0,    /* sentinel */
1401                 }
1402         };
1403
1404         fn = fib6_lookup_1(root, daddr ? args : args + 1);
1405         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1406                 fn = root;
1407
1408         return fn;
1409 }
1410
1411 /*
1412  *      Get node with specified destination prefix (and source prefix,
1413  *      if subtrees are used)
1414  *      exact_match == true means we try to find fn with exact match of
1415  *      the passed in prefix addr
1416  *      exact_match == false means we try to find fn with longest prefix
1417  *      match of the passed in prefix addr. This is useful for finding fn
1418  *      for cached route as it will be stored in the exception table under
1419  *      the node with longest prefix length.
1420  */
1421
1422
1423 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1424                                        const struct in6_addr *addr,
1425                                        int plen, int offset,
1426                                        bool exact_match)
1427 {
1428         struct fib6_node *fn, *prev = NULL;
1429
1430         for (fn = root; fn ; ) {
1431                 struct rt6_info *leaf = rcu_dereference(fn->leaf);
1432                 struct rt6key *key;
1433
1434                 /* This node is being deleted */
1435                 if (!leaf) {
1436                         if (plen <= fn->fn_bit)
1437                                 goto out;
1438                         else
1439                                 goto next;
1440                 }
1441
1442                 key = (struct rt6key *)((u8 *)leaf + offset);
1443
1444                 /*
1445                  *      Prefix match
1446                  */
1447                 if (plen < fn->fn_bit ||
1448                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1449                         goto out;
1450
1451                 if (plen == fn->fn_bit)
1452                         return fn;
1453
1454                 prev = fn;
1455
1456 next:
1457                 /*
1458                  *      We have more bits to go
1459                  */
1460                 if (addr_bit_set(addr, fn->fn_bit))
1461                         fn = rcu_dereference(fn->right);
1462                 else
1463                         fn = rcu_dereference(fn->left);
1464         }
1465 out:
1466         if (exact_match)
1467                 return NULL;
1468         else
1469                 return prev;
1470 }
1471
1472 struct fib6_node *fib6_locate(struct fib6_node *root,
1473                               const struct in6_addr *daddr, int dst_len,
1474                               const struct in6_addr *saddr, int src_len,
1475                               bool exact_match)
1476 {
1477         struct fib6_node *fn;
1478
1479         fn = fib6_locate_1(root, daddr, dst_len,
1480                            offsetof(struct rt6_info, rt6i_dst),
1481                            exact_match);
1482
1483 #ifdef CONFIG_IPV6_SUBTREES
1484         if (src_len) {
1485                 WARN_ON(saddr == NULL);
1486                 if (fn) {
1487                         struct fib6_node *subtree = FIB6_SUBTREE(fn);
1488
1489                         if (subtree) {
1490                                 fn = fib6_locate_1(subtree, saddr, src_len,
1491                                            offsetof(struct rt6_info, rt6i_src),
1492                                            exact_match);
1493                         }
1494                 }
1495         }
1496 #endif
1497
1498         if (fn && fn->fn_flags & RTN_RTINFO)
1499                 return fn;
1500
1501         return NULL;
1502 }
1503
1504
1505 /*
1506  *      Deletion
1507  *
1508  */
1509
1510 static struct rt6_info *fib6_find_prefix(struct net *net,
1511                                          struct fib6_table *table,
1512                                          struct fib6_node *fn)
1513 {
1514         struct fib6_node *child_left, *child_right;
1515
1516         if (fn->fn_flags & RTN_ROOT)
1517                 return net->ipv6.ip6_null_entry;
1518
1519         while (fn) {
1520                 child_left = rcu_dereference_protected(fn->left,
1521                                     lockdep_is_held(&table->tb6_lock));
1522                 child_right = rcu_dereference_protected(fn->right,
1523                                     lockdep_is_held(&table->tb6_lock));
1524                 if (child_left)
1525                         return rcu_dereference_protected(child_left->leaf,
1526                                         lockdep_is_held(&table->tb6_lock));
1527                 if (child_right)
1528                         return rcu_dereference_protected(child_right->leaf,
1529                                         lockdep_is_held(&table->tb6_lock));
1530
1531                 fn = FIB6_SUBTREE(fn);
1532         }
1533         return NULL;
1534 }
1535
1536 /*
1537  *      Called to trim the tree of intermediate nodes when possible. "fn"
1538  *      is the node we want to try and remove.
1539  *      Need to own table->tb6_lock
1540  */
1541
1542 static struct fib6_node *fib6_repair_tree(struct net *net,
1543                                           struct fib6_table *table,
1544                                           struct fib6_node *fn)
1545 {
1546         int children;
1547         int nstate;
1548         struct fib6_node *child;
1549         struct fib6_walker *w;
1550         int iter = 0;
1551
1552         /* Set fn->leaf to null_entry for root node. */
1553         if (fn->fn_flags & RTN_TL_ROOT) {
1554                 rcu_assign_pointer(fn->leaf, net->ipv6.ip6_null_entry);
1555                 return fn;
1556         }
1557
1558         for (;;) {
1559                 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1560                                             lockdep_is_held(&table->tb6_lock));
1561                 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1562                                             lockdep_is_held(&table->tb6_lock));
1563                 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1564                                             lockdep_is_held(&table->tb6_lock));
1565                 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1566                                             lockdep_is_held(&table->tb6_lock));
1567                 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1568                                             lockdep_is_held(&table->tb6_lock));
1569                 struct rt6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1570                                             lockdep_is_held(&table->tb6_lock));
1571                 struct rt6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1572                                             lockdep_is_held(&table->tb6_lock));
1573                 struct rt6_info *new_fn_leaf;
1574
1575                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1576                 iter++;
1577
1578                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1579                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1580                 WARN_ON(fn_leaf);
1581
1582                 children = 0;
1583                 child = NULL;
1584                 if (fn_r)
1585                         child = fn_r, children |= 1;
1586                 if (fn_l)
1587                         child = fn_l, children |= 2;
1588
1589                 if (children == 3 || FIB6_SUBTREE(fn)
1590 #ifdef CONFIG_IPV6_SUBTREES
1591                     /* Subtree root (i.e. fn) may have one child */
1592                     || (children && fn->fn_flags & RTN_ROOT)
1593 #endif
1594                     ) {
1595                         new_fn_leaf = fib6_find_prefix(net, table, fn);
1596 #if RT6_DEBUG >= 2
1597                         if (!new_fn_leaf) {
1598                                 WARN_ON(!new_fn_leaf);
1599                                 new_fn_leaf = net->ipv6.ip6_null_entry;
1600                         }
1601 #endif
1602                         atomic_inc(&new_fn_leaf->rt6i_ref);
1603                         rcu_assign_pointer(fn->leaf, new_fn_leaf);
1604                         return pn;
1605                 }
1606
1607 #ifdef CONFIG_IPV6_SUBTREES
1608                 if (FIB6_SUBTREE(pn) == fn) {
1609                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1610                         RCU_INIT_POINTER(pn->subtree, NULL);
1611                         nstate = FWS_L;
1612                 } else {
1613                         WARN_ON(fn->fn_flags & RTN_ROOT);
1614 #endif
1615                         if (pn_r == fn)
1616                                 rcu_assign_pointer(pn->right, child);
1617                         else if (pn_l == fn)
1618                                 rcu_assign_pointer(pn->left, child);
1619 #if RT6_DEBUG >= 2
1620                         else
1621                                 WARN_ON(1);
1622 #endif
1623                         if (child)
1624                                 rcu_assign_pointer(child->parent, pn);
1625                         nstate = FWS_R;
1626 #ifdef CONFIG_IPV6_SUBTREES
1627                 }
1628 #endif
1629
1630                 read_lock(&net->ipv6.fib6_walker_lock);
1631                 FOR_WALKERS(net, w) {
1632                         if (!child) {
1633                                 if (w->node == fn) {
1634                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1635                                         w->node = pn;
1636                                         w->state = nstate;
1637                                 }
1638                         } else {
1639                                 if (w->node == fn) {
1640                                         w->node = child;
1641                                         if (children&2) {
1642                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1643                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1644                                         } else {
1645                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1646                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1647                                         }
1648                                 }
1649                         }
1650                 }
1651                 read_unlock(&net->ipv6.fib6_walker_lock);
1652
1653                 node_free(net, fn);
1654                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1655                         return pn;
1656
1657                 RCU_INIT_POINTER(pn->leaf, NULL);
1658                 rt6_release(pn_leaf);
1659                 fn = pn;
1660         }
1661 }
1662
1663 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1664                            struct rt6_info __rcu **rtp, struct nl_info *info)
1665 {
1666         struct fib6_walker *w;
1667         struct rt6_info *rt = rcu_dereference_protected(*rtp,
1668                                     lockdep_is_held(&table->tb6_lock));
1669         struct net *net = info->nl_net;
1670
1671         RT6_TRACE("fib6_del_route\n");
1672
1673         WARN_ON_ONCE(rt->rt6i_flags & RTF_CACHE);
1674
1675         /* Unlink it */
1676         *rtp = rt->rt6_next;
1677         rt->rt6i_node = NULL;
1678         net->ipv6.rt6_stats->fib_rt_entries--;
1679         net->ipv6.rt6_stats->fib_discarded_routes++;
1680
1681         /* Flush all cached dst in exception table */
1682         rt6_flush_exceptions(rt);
1683
1684         /* Reset round-robin state, if necessary */
1685         if (rcu_access_pointer(fn->rr_ptr) == rt)
1686                 fn->rr_ptr = NULL;
1687
1688         /* Remove this entry from other siblings */
1689         if (rt->rt6i_nsiblings) {
1690                 struct rt6_info *sibling, *next_sibling;
1691
1692                 list_for_each_entry_safe(sibling, next_sibling,
1693                                          &rt->rt6i_siblings, rt6i_siblings)
1694                         sibling->rt6i_nsiblings--;
1695                 rt->rt6i_nsiblings = 0;
1696                 list_del_init(&rt->rt6i_siblings);
1697                 rt6_multipath_rebalance(next_sibling);
1698         }
1699
1700         /* Adjust walkers */
1701         read_lock(&net->ipv6.fib6_walker_lock);
1702         FOR_WALKERS(net, w) {
1703                 if (w->state == FWS_C && w->leaf == rt) {
1704                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1705                         w->leaf = rcu_dereference_protected(rt->rt6_next,
1706                                             lockdep_is_held(&table->tb6_lock));
1707                         if (!w->leaf)
1708                                 w->state = FWS_U;
1709                 }
1710         }
1711         read_unlock(&net->ipv6.fib6_walker_lock);
1712
1713         /* If it was last route, call fib6_repair_tree() to:
1714          * 1. For root node, put back null_entry as how the table was created.
1715          * 2. For other nodes, expunge its radix tree node.
1716          */
1717         if (!rcu_access_pointer(fn->leaf)) {
1718                 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1719                         fn->fn_flags &= ~RTN_RTINFO;
1720                         net->ipv6.rt6_stats->fib_route_nodes--;
1721                 }
1722                 fn = fib6_repair_tree(net, table, fn);
1723         }
1724
1725         fib6_purge_rt(rt, fn, net);
1726
1727         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1728         if (!info->skip_notify)
1729                 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1730         rt6_release(rt);
1731 }
1732
1733 /* Need to own table->tb6_lock */
1734 int fib6_del(struct rt6_info *rt, struct nl_info *info)
1735 {
1736         struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
1737                                     lockdep_is_held(&rt->rt6i_table->tb6_lock));
1738         struct fib6_table *table = rt->rt6i_table;
1739         struct net *net = info->nl_net;
1740         struct rt6_info __rcu **rtp;
1741         struct rt6_info __rcu **rtp_next;
1742
1743 #if RT6_DEBUG >= 2
1744         if (rt->dst.obsolete > 0) {
1745                 WARN_ON(fn);
1746                 return -ENOENT;
1747         }
1748 #endif
1749         if (!fn || rt == net->ipv6.ip6_null_entry)
1750                 return -ENOENT;
1751
1752         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1753
1754         /* remove cached dst from exception table */
1755         if (rt->rt6i_flags & RTF_CACHE)
1756                 return rt6_remove_exception_rt(rt);
1757
1758         /*
1759          *      Walk the leaf entries looking for ourself
1760          */
1761
1762         for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1763                 struct rt6_info *cur = rcu_dereference_protected(*rtp,
1764                                         lockdep_is_held(&table->tb6_lock));
1765                 if (rt == cur) {
1766                         fib6_del_route(table, fn, rtp, info);
1767                         return 0;
1768                 }
1769                 rtp_next = &cur->rt6_next;
1770         }
1771         return -ENOENT;
1772 }
1773
1774 /*
1775  *      Tree traversal function.
1776  *
1777  *      Certainly, it is not interrupt safe.
1778  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1779  *      It means, that we can modify tree during walking
1780  *      and use this function for garbage collection, clone pruning,
1781  *      cleaning tree when a device goes down etc. etc.
1782  *
1783  *      It guarantees that every node will be traversed,
1784  *      and that it will be traversed only once.
1785  *
1786  *      Callback function w->func may return:
1787  *      0 -> continue walking.
1788  *      positive value -> walking is suspended (used by tree dumps,
1789  *      and probably by gc, if it will be split to several slices)
1790  *      negative value -> terminate walking.
1791  *
1792  *      The function itself returns:
1793  *      0   -> walk is complete.
1794  *      >0  -> walk is incomplete (i.e. suspended)
1795  *      <0  -> walk is terminated by an error.
1796  *
1797  *      This function is called with tb6_lock held.
1798  */
1799
1800 static int fib6_walk_continue(struct fib6_walker *w)
1801 {
1802         struct fib6_node *fn, *pn, *left, *right;
1803
1804         /* w->root should always be table->tb6_root */
1805         WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1806
1807         for (;;) {
1808                 fn = w->node;
1809                 if (!fn)
1810                         return 0;
1811
1812                 switch (w->state) {
1813 #ifdef CONFIG_IPV6_SUBTREES
1814                 case FWS_S:
1815                         if (FIB6_SUBTREE(fn)) {
1816                                 w->node = FIB6_SUBTREE(fn);
1817                                 continue;
1818                         }
1819                         w->state = FWS_L;
1820 #endif
1821                         /* fall through */
1822                 case FWS_L:
1823                         left = rcu_dereference_protected(fn->left, 1);
1824                         if (left) {
1825                                 w->node = left;
1826                                 w->state = FWS_INIT;
1827                                 continue;
1828                         }
1829                         w->state = FWS_R;
1830                         /* fall through */
1831                 case FWS_R:
1832                         right = rcu_dereference_protected(fn->right, 1);
1833                         if (right) {
1834                                 w->node = right;
1835                                 w->state = FWS_INIT;
1836                                 continue;
1837                         }
1838                         w->state = FWS_C;
1839                         w->leaf = rcu_dereference_protected(fn->leaf, 1);
1840                         /* fall through */
1841                 case FWS_C:
1842                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1843                                 int err;
1844
1845                                 if (w->skip) {
1846                                         w->skip--;
1847                                         goto skip;
1848                                 }
1849
1850                                 err = w->func(w);
1851                                 if (err)
1852                                         return err;
1853
1854                                 w->count++;
1855                                 continue;
1856                         }
1857 skip:
1858                         w->state = FWS_U;
1859                         /* fall through */
1860                 case FWS_U:
1861                         if (fn == w->root)
1862                                 return 0;
1863                         pn = rcu_dereference_protected(fn->parent, 1);
1864                         left = rcu_dereference_protected(pn->left, 1);
1865                         right = rcu_dereference_protected(pn->right, 1);
1866                         w->node = pn;
1867 #ifdef CONFIG_IPV6_SUBTREES
1868                         if (FIB6_SUBTREE(pn) == fn) {
1869                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1870                                 w->state = FWS_L;
1871                                 continue;
1872                         }
1873 #endif
1874                         if (left == fn) {
1875                                 w->state = FWS_R;
1876                                 continue;
1877                         }
1878                         if (right == fn) {
1879                                 w->state = FWS_C;
1880                                 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1881                                 continue;
1882                         }
1883 #if RT6_DEBUG >= 2
1884                         WARN_ON(1);
1885 #endif
1886                 }
1887         }
1888 }
1889
1890 static int fib6_walk(struct net *net, struct fib6_walker *w)
1891 {
1892         int res;
1893
1894         w->state = FWS_INIT;
1895         w->node = w->root;
1896
1897         fib6_walker_link(net, w);
1898         res = fib6_walk_continue(w);
1899         if (res <= 0)
1900                 fib6_walker_unlink(net, w);
1901         return res;
1902 }
1903
1904 static int fib6_clean_node(struct fib6_walker *w)
1905 {
1906         int res;
1907         struct rt6_info *rt;
1908         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1909         struct nl_info info = {
1910                 .nl_net = c->net,
1911         };
1912
1913         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1914             w->node->fn_sernum != c->sernum)
1915                 w->node->fn_sernum = c->sernum;
1916
1917         if (!c->func) {
1918                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1919                 w->leaf = NULL;
1920                 return 0;
1921         }
1922
1923         for_each_fib6_walker_rt(w) {
1924                 res = c->func(rt, c->arg);
1925                 if (res == -1) {
1926                         w->leaf = rt;
1927                         res = fib6_del(rt, &info);
1928                         if (res) {
1929 #if RT6_DEBUG >= 2
1930                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
1931                                          __func__, rt,
1932                                          rcu_access_pointer(rt->rt6i_node),
1933                                          res);
1934 #endif
1935                                 continue;
1936                         }
1937                         return 0;
1938                 } else if (res == -2) {
1939                         if (WARN_ON(!rt->rt6i_nsiblings))
1940                                 continue;
1941                         rt = list_last_entry(&rt->rt6i_siblings,
1942                                              struct rt6_info, rt6i_siblings);
1943                         continue;
1944                 }
1945                 WARN_ON(res != 0);
1946         }
1947         w->leaf = rt;
1948         return 0;
1949 }
1950
1951 /*
1952  *      Convenient frontend to tree walker.
1953  *
1954  *      func is called on each route.
1955  *              It may return -2 -> skip multipath route.
1956  *                            -1 -> delete this route.
1957  *                            0  -> continue walking
1958  */
1959
1960 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
1961                             int (*func)(struct rt6_info *, void *arg),
1962                             int sernum, void *arg)
1963 {
1964         struct fib6_cleaner c;
1965
1966         c.w.root = root;
1967         c.w.func = fib6_clean_node;
1968         c.w.count = 0;
1969         c.w.skip = 0;
1970         c.func = func;
1971         c.sernum = sernum;
1972         c.arg = arg;
1973         c.net = net;
1974
1975         fib6_walk(net, &c.w);
1976 }
1977
1978 static void __fib6_clean_all(struct net *net,
1979                              int (*func)(struct rt6_info *, void *),
1980                              int sernum, void *arg)
1981 {
1982         struct fib6_table *table;
1983         struct hlist_head *head;
1984         unsigned int h;
1985
1986         rcu_read_lock();
1987         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
1988                 head = &net->ipv6.fib_table_hash[h];
1989                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
1990                         spin_lock_bh(&table->tb6_lock);
1991                         fib6_clean_tree(net, &table->tb6_root,
1992                                         func, sernum, arg);
1993                         spin_unlock_bh(&table->tb6_lock);
1994                 }
1995         }
1996         rcu_read_unlock();
1997 }
1998
1999 void fib6_clean_all(struct net *net, int (*func)(struct rt6_info *, void *),
2000                     void *arg)
2001 {
2002         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg);
2003 }
2004
2005 static void fib6_flush_trees(struct net *net)
2006 {
2007         int new_sernum = fib6_new_sernum(net);
2008
2009         __fib6_clean_all(net, NULL, new_sernum, NULL);
2010 }
2011
2012 /*
2013  *      Garbage collection
2014  */
2015
2016 static int fib6_age(struct rt6_info *rt, void *arg)
2017 {
2018         struct fib6_gc_args *gc_args = arg;
2019         unsigned long now = jiffies;
2020
2021         /*
2022          *      check addrconf expiration here.
2023          *      Routes are expired even if they are in use.
2024          */
2025
2026         if (rt->rt6i_flags & RTF_EXPIRES && rt->dst.expires) {
2027                 if (time_after(now, rt->dst.expires)) {
2028                         RT6_TRACE("expiring %p\n", rt);
2029                         return -1;
2030                 }
2031                 gc_args->more++;
2032         }
2033
2034         /*      Also age clones in the exception table.
2035          *      Note, that clones are aged out
2036          *      only if they are not in use now.
2037          */
2038         rt6_age_exceptions(rt, gc_args, now);
2039
2040         return 0;
2041 }
2042
2043 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2044 {
2045         struct fib6_gc_args gc_args;
2046         unsigned long now;
2047
2048         if (force) {
2049                 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2050         } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2051                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2052                 return;
2053         }
2054         gc_args.timeout = expires ? (int)expires :
2055                           net->ipv6.sysctl.ip6_rt_gc_interval;
2056         gc_args.more = 0;
2057
2058         fib6_clean_all(net, fib6_age, &gc_args);
2059         now = jiffies;
2060         net->ipv6.ip6_rt_last_gc = now;
2061
2062         if (gc_args.more)
2063                 mod_timer(&net->ipv6.ip6_fib_timer,
2064                           round_jiffies(now
2065                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
2066         else
2067                 del_timer(&net->ipv6.ip6_fib_timer);
2068         spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2069 }
2070
2071 static void fib6_gc_timer_cb(struct timer_list *t)
2072 {
2073         struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2074
2075         fib6_run_gc(0, arg, true);
2076 }
2077
2078 static int __net_init fib6_net_init(struct net *net)
2079 {
2080         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2081         int err;
2082
2083         err = fib6_notifier_init(net);
2084         if (err)
2085                 return err;
2086
2087         spin_lock_init(&net->ipv6.fib6_gc_lock);
2088         rwlock_init(&net->ipv6.fib6_walker_lock);
2089         INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2090         timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2091
2092         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2093         if (!net->ipv6.rt6_stats)
2094                 goto out_timer;
2095
2096         /* Avoid false sharing : Use at least a full cache line */
2097         size = max_t(size_t, size, L1_CACHE_BYTES);
2098
2099         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2100         if (!net->ipv6.fib_table_hash)
2101                 goto out_rt6_stats;
2102
2103         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2104                                           GFP_KERNEL);
2105         if (!net->ipv6.fib6_main_tbl)
2106                 goto out_fib_table_hash;
2107
2108         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2109         rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2110                            net->ipv6.ip6_null_entry);
2111         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2112                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2113         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2114
2115 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2116         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2117                                            GFP_KERNEL);
2118         if (!net->ipv6.fib6_local_tbl)
2119                 goto out_fib6_main_tbl;
2120         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2121         rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2122                            net->ipv6.ip6_null_entry);
2123         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2124                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2125         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2126 #endif
2127         fib6_tables_init(net);
2128
2129         return 0;
2130
2131 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2132 out_fib6_main_tbl:
2133         kfree(net->ipv6.fib6_main_tbl);
2134 #endif
2135 out_fib_table_hash:
2136         kfree(net->ipv6.fib_table_hash);
2137 out_rt6_stats:
2138         kfree(net->ipv6.rt6_stats);
2139 out_timer:
2140         fib6_notifier_exit(net);
2141         return -ENOMEM;
2142 }
2143
2144 static void fib6_net_exit(struct net *net)
2145 {
2146         unsigned int i;
2147
2148         del_timer_sync(&net->ipv6.ip6_fib_timer);
2149
2150         for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2151                 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2152                 struct hlist_node *tmp;
2153                 struct fib6_table *tb;
2154
2155                 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2156                         hlist_del(&tb->tb6_hlist);
2157                         fib6_free_table(tb);
2158                 }
2159         }
2160
2161         kfree(net->ipv6.fib_table_hash);
2162         kfree(net->ipv6.rt6_stats);
2163         fib6_notifier_exit(net);
2164 }
2165
2166 static struct pernet_operations fib6_net_ops = {
2167         .init = fib6_net_init,
2168         .exit = fib6_net_exit,
2169 };
2170
2171 int __init fib6_init(void)
2172 {
2173         int ret = -ENOMEM;
2174
2175         fib6_node_kmem = kmem_cache_create("fib6_nodes",
2176                                            sizeof(struct fib6_node),
2177                                            0, SLAB_HWCACHE_ALIGN,
2178                                            NULL);
2179         if (!fib6_node_kmem)
2180                 goto out;
2181
2182         ret = register_pernet_subsys(&fib6_net_ops);
2183         if (ret)
2184                 goto out_kmem_cache_create;
2185
2186         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2187                                    inet6_dump_fib, 0);
2188         if (ret)
2189                 goto out_unregister_subsys;
2190
2191         __fib6_flush_trees = fib6_flush_trees;
2192 out:
2193         return ret;
2194
2195 out_unregister_subsys:
2196         unregister_pernet_subsys(&fib6_net_ops);
2197 out_kmem_cache_create:
2198         kmem_cache_destroy(fib6_node_kmem);
2199         goto out;
2200 }
2201
2202 void fib6_gc_cleanup(void)
2203 {
2204         unregister_pernet_subsys(&fib6_net_ops);
2205         kmem_cache_destroy(fib6_node_kmem);
2206 }
2207
2208 #ifdef CONFIG_PROC_FS
2209
2210 struct ipv6_route_iter {
2211         struct seq_net_private p;
2212         struct fib6_walker w;
2213         loff_t skip;
2214         struct fib6_table *tbl;
2215         int sernum;
2216 };
2217
2218 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2219 {
2220         struct rt6_info *rt = v;
2221         struct ipv6_route_iter *iter = seq->private;
2222
2223         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen);
2224
2225 #ifdef CONFIG_IPV6_SUBTREES
2226         seq_printf(seq, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen);
2227 #else
2228         seq_puts(seq, "00000000000000000000000000000000 00 ");
2229 #endif
2230         if (rt->rt6i_flags & RTF_GATEWAY)
2231                 seq_printf(seq, "%pi6", &rt->rt6i_gateway);
2232         else
2233                 seq_puts(seq, "00000000000000000000000000000000");
2234
2235         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2236                    rt->rt6i_metric, atomic_read(&rt->dst.__refcnt),
2237                    rt->dst.__use, rt->rt6i_flags,
2238                    rt->dst.dev ? rt->dst.dev->name : "");
2239         iter->w.leaf = NULL;
2240         return 0;
2241 }
2242
2243 static int ipv6_route_yield(struct fib6_walker *w)
2244 {
2245         struct ipv6_route_iter *iter = w->args;
2246
2247         if (!iter->skip)
2248                 return 1;
2249
2250         do {
2251                 iter->w.leaf = rcu_dereference_protected(
2252                                 iter->w.leaf->rt6_next,
2253                                 lockdep_is_held(&iter->tbl->tb6_lock));
2254                 iter->skip--;
2255                 if (!iter->skip && iter->w.leaf)
2256                         return 1;
2257         } while (iter->w.leaf);
2258
2259         return 0;
2260 }
2261
2262 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2263                                       struct net *net)
2264 {
2265         memset(&iter->w, 0, sizeof(iter->w));
2266         iter->w.func = ipv6_route_yield;
2267         iter->w.root = &iter->tbl->tb6_root;
2268         iter->w.state = FWS_INIT;
2269         iter->w.node = iter->w.root;
2270         iter->w.args = iter;
2271         iter->sernum = iter->w.root->fn_sernum;
2272         INIT_LIST_HEAD(&iter->w.lh);
2273         fib6_walker_link(net, &iter->w);
2274 }
2275
2276 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2277                                                     struct net *net)
2278 {
2279         unsigned int h;
2280         struct hlist_node *node;
2281
2282         if (tbl) {
2283                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2284                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2285         } else {
2286                 h = 0;
2287                 node = NULL;
2288         }
2289
2290         while (!node && h < FIB6_TABLE_HASHSZ) {
2291                 node = rcu_dereference_bh(
2292                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2293         }
2294         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2295 }
2296
2297 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2298 {
2299         if (iter->sernum != iter->w.root->fn_sernum) {
2300                 iter->sernum = iter->w.root->fn_sernum;
2301                 iter->w.state = FWS_INIT;
2302                 iter->w.node = iter->w.root;
2303                 WARN_ON(iter->w.skip);
2304                 iter->w.skip = iter->w.count;
2305         }
2306 }
2307
2308 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2309 {
2310         int r;
2311         struct rt6_info *n;
2312         struct net *net = seq_file_net(seq);
2313         struct ipv6_route_iter *iter = seq->private;
2314
2315         if (!v)
2316                 goto iter_table;
2317
2318         n = rcu_dereference_bh(((struct rt6_info *)v)->rt6_next);
2319         if (n) {
2320                 ++*pos;
2321                 return n;
2322         }
2323
2324 iter_table:
2325         ipv6_route_check_sernum(iter);
2326         spin_lock_bh(&iter->tbl->tb6_lock);
2327         r = fib6_walk_continue(&iter->w);
2328         spin_unlock_bh(&iter->tbl->tb6_lock);
2329         if (r > 0) {
2330                 if (v)
2331                         ++*pos;
2332                 return iter->w.leaf;
2333         } else if (r < 0) {
2334                 fib6_walker_unlink(net, &iter->w);
2335                 return NULL;
2336         }
2337         fib6_walker_unlink(net, &iter->w);
2338
2339         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2340         if (!iter->tbl)
2341                 return NULL;
2342
2343         ipv6_route_seq_setup_walk(iter, net);
2344         goto iter_table;
2345 }
2346
2347 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2348         __acquires(RCU_BH)
2349 {
2350         struct net *net = seq_file_net(seq);
2351         struct ipv6_route_iter *iter = seq->private;
2352
2353         rcu_read_lock_bh();
2354         iter->tbl = ipv6_route_seq_next_table(NULL, net);
2355         iter->skip = *pos;
2356
2357         if (iter->tbl) {
2358                 ipv6_route_seq_setup_walk(iter, net);
2359                 return ipv6_route_seq_next(seq, NULL, pos);
2360         } else {
2361                 return NULL;
2362         }
2363 }
2364
2365 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2366 {
2367         struct fib6_walker *w = &iter->w;
2368         return w->node && !(w->state == FWS_U && w->node == w->root);
2369 }
2370
2371 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2372         __releases(RCU_BH)
2373 {
2374         struct net *net = seq_file_net(seq);
2375         struct ipv6_route_iter *iter = seq->private;
2376
2377         if (ipv6_route_iter_active(iter))
2378                 fib6_walker_unlink(net, &iter->w);
2379
2380         rcu_read_unlock_bh();
2381 }
2382
2383 static const struct seq_operations ipv6_route_seq_ops = {
2384         .start  = ipv6_route_seq_start,
2385         .next   = ipv6_route_seq_next,
2386         .stop   = ipv6_route_seq_stop,
2387         .show   = ipv6_route_seq_show
2388 };
2389
2390 int ipv6_route_open(struct inode *inode, struct file *file)
2391 {
2392         return seq_open_net(inode, file, &ipv6_route_seq_ops,
2393                             sizeof(struct ipv6_route_iter));
2394 }
2395
2396 #endif /* CONFIG_PROC_FS */