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/ip.h>
  33 #include <net/ipv6.h>
  34 #include <net/ndisc.h>
  35 #include <net/addrconf.h>
  36 #include <net/lwtunnel.h>
  37 #include <net/fib_notifier.h>
  38
  39 #include <net/ip6_fib.h>
  40 #include <net/ip6_route.h>
  41
  42 static struct kmem_cache *fib6_node_kmem __read_mostly;
  43
  44 struct fib6_cleaner {
  45         struct fib6_walker w;
  46         struct net *net;
  47         int (*func)(struct fib6_info *, void *arg);
  48         int sernum;
  49         void *arg;
  50         bool skip_notify;
  51 };
  52
  53 #ifdef CONFIG_IPV6_SUBTREES
  54 #define FWS_INIT FWS_S
  55 #else
  56 #define FWS_INIT FWS_L
  57 #endif
  58
  59 static struct fib6_info *fib6_find_prefix(struct net *net,
  60                                          struct fib6_table *table,
  61                                          struct fib6_node *fn);
  62 static struct fib6_node *fib6_repair_tree(struct net *net,
  63                                           struct fib6_table *table,
  64                                           struct fib6_node *fn);
  65 static int fib6_walk(struct net *net, struct fib6_walker *w);
  66 static int fib6_walk_continue(struct fib6_walker *w);
  67
  68 /*
  69  *      A routing update causes an increase of the serial number on the
  70  *      affected subtree. This allows for cached routes to be asynchronously
  71  *      tested when modifications are made to the destination cache as a
  72  *      result of redirects, path MTU changes, etc.
  73  */
  74
  75 static void fib6_gc_timer_cb(struct timer_list *t);
  76
  77 #define FOR_WALKERS(net, w) \
  78         list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
  79
  80 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
  81 {
  82         write_lock_bh(&net->ipv6.fib6_walker_lock);
  83         list_add(&w->lh, &net->ipv6.fib6_walkers);
  84         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  85 }
  86
  87 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
  88 {
  89         write_lock_bh(&net->ipv6.fib6_walker_lock);
  90         list_del(&w->lh);
  91         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  92 }
  93
  94 static int fib6_new_sernum(struct net *net)
  95 {
  96         int new, old;
  97
  98         do {
  99                 old = atomic_read(&net->ipv6.fib6_sernum);
 100                 new = old < INT_MAX ? old + 1 : 1;
 101         } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
 102                                 old, new) != old);
 103         return new;
 104 }
 105
 106 enum {
 107         FIB6_NO_SERNUM_CHANGE = 0,
 108 };
 109
 110 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
 111 {
 112         struct fib6_node *fn;
 113
 114         fn = rcu_dereference_protected(f6i->fib6_node,
 115                         lockdep_is_held(&f6i->fib6_table->tb6_lock));
 116         if (fn)
 117                 fn->fn_sernum = fib6_new_sernum(net);
 118 }
 119
 120 /*
 121  *      Auxiliary address test functions for the radix tree.
 122  *
 123  *      These assume a 32bit processor (although it will work on
 124  *      64bit processors)
 125  */
 126
 127 /*
 128  *      test bit
 129  */
 130 #if defined(__LITTLE_ENDIAN)
 131 # define BITOP_BE32_SWIZZLE     (0x1F & ~7)
 132 #else
 133 # define BITOP_BE32_SWIZZLE     0
 134 #endif
 135
 136 static __be32 addr_bit_set(const void *token, int fn_bit)
 137 {
 138         const __be32 *addr = token;
 139         /*
 140          * Here,
 141          *      1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
 142          * is optimized version of
 143          *      htonl(1 << ((~fn_bit)&0x1F))
 144          * See include/asm-generic/bitops/le.h.
 145          */
 146         return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
 147                addr[fn_bit >> 5];
 148 }
 149
 150 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags)
 151 {
 152         struct fib6_info *f6i;
 153
 154         f6i = kzalloc(sizeof(*f6i), gfp_flags);
 155         if (!f6i)
 156                 return NULL;
 157
 158         f6i->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
 159         if (!f6i->rt6i_pcpu) {
 160                 kfree(f6i);
 161                 return NULL;
 162         }
 163
 164         INIT_LIST_HEAD(&f6i->fib6_siblings);
 165         atomic_inc(&f6i->fib6_ref);
 166
 167         return f6i;
 168 }
 169
 170 void fib6_info_destroy_rcu(struct rcu_head *head)
 171 {
 172         struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
 173         struct rt6_exception_bucket *bucket;
 174
 175         WARN_ON(f6i->fib6_node);
 176
 177         bucket = rcu_dereference_protected(f6i->rt6i_exception_bucket, 1);
 178         if (bucket) {
 179                 f6i->rt6i_exception_bucket = NULL;
 180                 kfree(bucket);
 181         }
 182
 183         if (f6i->rt6i_pcpu) {
 184                 int cpu;
 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(f6i->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                 free_percpu(f6i->rt6i_pcpu);
 200         }
 201
 202         lwtstate_put(f6i->fib6_nh.nh_lwtstate);
 203
 204         if (f6i->fib6_nh.nh_dev)
 205                 dev_put(f6i->fib6_nh.nh_dev);
 206
 207         ip_fib_metrics_put(f6i->fib6_metrics);
 208
 209         kfree(f6i);
 210 }
 211 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
 212
 213 static struct fib6_node *node_alloc(struct net *net)
 214 {
 215         struct fib6_node *fn;
 216
 217         fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
 218         if (fn)
 219                 net->ipv6.rt6_stats->fib_nodes++;
 220
 221         return fn;
 222 }
 223
 224 static void node_free_immediate(struct net *net, struct fib6_node *fn)
 225 {
 226         kmem_cache_free(fib6_node_kmem, fn);
 227         net->ipv6.rt6_stats->fib_nodes--;
 228 }
 229
 230 static void node_free_rcu(struct rcu_head *head)
 231 {
 232         struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
 233
 234         kmem_cache_free(fib6_node_kmem, fn);
 235 }
 236
 237 static void node_free(struct net *net, struct fib6_node *fn)
 238 {
 239         call_rcu(&fn->rcu, node_free_rcu);
 240         net->ipv6.rt6_stats->fib_nodes--;
 241 }
 242
 243 static void fib6_free_table(struct fib6_table *table)
 244 {
 245         inetpeer_invalidate_tree(&table->tb6_peers);
 246         kfree(table);
 247 }
 248
 249 static void fib6_link_table(struct net *net, struct fib6_table *tb)
 250 {
 251         unsigned int h;
 252
 253         /*
 254          * Initialize table lock at a single place to give lockdep a key,
 255          * tables aren't visible prior to being linked to the list.
 256          */
 257         spin_lock_init(&tb->tb6_lock);
 258         h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
 259
 260         /*
 261          * No protection necessary, this is the only list mutatation
 262          * operation, tables never disappear once they exist.
 263          */
 264         hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
 265 }
 266
 267 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 268
 269 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
 270 {
 271         struct fib6_table *table;
 272
 273         table = kzalloc(sizeof(*table), GFP_ATOMIC);
 274         if (table) {
 275                 table->tb6_id = id;
 276                 rcu_assign_pointer(table->tb6_root.leaf,
 277                                    net->ipv6.fib6_null_entry);
 278                 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
 279                 inet_peer_base_init(&table->tb6_peers);
 280         }
 281
 282         return table;
 283 }
 284
 285 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 286 {
 287         struct fib6_table *tb;
 288
 289         if (id == 0)
 290                 id = RT6_TABLE_MAIN;
 291         tb = fib6_get_table(net, id);
 292         if (tb)
 293                 return tb;
 294
 295         tb = fib6_alloc_table(net, id);
 296         if (tb)
 297                 fib6_link_table(net, tb);
 298
 299         return tb;
 300 }
 301 EXPORT_SYMBOL_GPL(fib6_new_table);
 302
 303 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 304 {
 305         struct fib6_table *tb;
 306         struct hlist_head *head;
 307         unsigned int h;
 308
 309         if (id == 0)
 310                 id = RT6_TABLE_MAIN;
 311         h = id & (FIB6_TABLE_HASHSZ - 1);
 312         rcu_read_lock();
 313         head = &net->ipv6.fib_table_hash[h];
 314         hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 315                 if (tb->tb6_id == id) {
 316                         rcu_read_unlock();
 317                         return tb;
 318                 }
 319         }
 320         rcu_read_unlock();
 321
 322         return NULL;
 323 }
 324 EXPORT_SYMBOL_GPL(fib6_get_table);
 325
 326 static void __net_init fib6_tables_init(struct net *net)
 327 {
 328         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 329         fib6_link_table(net, net->ipv6.fib6_local_tbl);
 330 }
 331 #else
 332
 333 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 334 {
 335         return fib6_get_table(net, id);
 336 }
 337
 338 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 339 {
 340           return net->ipv6.fib6_main_tbl;
 341 }
 342
 343 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 344                                    const struct sk_buff *skb,
 345                                    int flags, pol_lookup_t lookup)
 346 {
 347         struct rt6_info *rt;
 348
 349         rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
 350         if (rt->dst.error == -EAGAIN) {
 351                 ip6_rt_put(rt);
 352                 rt = net->ipv6.ip6_null_entry;
 353                 dst_hold(&rt->dst);
 354         }
 355
 356         return &rt->dst;
 357 }
 358
 359 /* called with rcu lock held; no reference taken on fib6_info */
 360 struct fib6_info *fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
 361                               int flags)
 362 {
 363         return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6, flags);
 364 }
 365
 366 static void __net_init fib6_tables_init(struct net *net)
 367 {
 368         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 369 }
 370
 371 #endif
 372
 373 unsigned int fib6_tables_seq_read(struct net *net)
 374 {
 375         unsigned int h, fib_seq = 0;
 376
 377         rcu_read_lock();
 378         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 379                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 380                 struct fib6_table *tb;
 381
 382                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 383                         fib_seq += tb->fib_seq;
 384         }
 385         rcu_read_unlock();
 386
 387         return fib_seq;
 388 }
 389
 390 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
 391                                     enum fib_event_type event_type,
 392                                     struct fib6_info *rt)
 393 {
 394         struct fib6_entry_notifier_info info = {
 395                 .rt = rt,
 396         };
 397
 398         return call_fib6_notifier(nb, net, event_type, &info.info);
 399 }
 400
 401 static int call_fib6_entry_notifiers(struct net *net,
 402                                      enum fib_event_type event_type,
 403                                      struct fib6_info *rt,
 404                                      struct netlink_ext_ack *extack)
 405 {
 406         struct fib6_entry_notifier_info info = {
 407                 .info.extack = extack,
 408                 .rt = rt,
 409         };
 410
 411         rt->fib6_table->fib_seq++;
 412         return call_fib6_notifiers(net, event_type, &info.info);
 413 }
 414
 415 struct fib6_dump_arg {
 416         struct net *net;
 417         struct notifier_block *nb;
 418 };
 419
 420 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
 421 {
 422         if (rt == arg->net->ipv6.fib6_null_entry)
 423                 return;
 424         call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
 425 }
 426
 427 static int fib6_node_dump(struct fib6_walker *w)
 428 {
 429         struct fib6_info *rt;
 430
 431         for_each_fib6_walker_rt(w)
 432                 fib6_rt_dump(rt, w->args);
 433         w->leaf = NULL;
 434         return 0;
 435 }
 436
 437 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
 438                             struct fib6_walker *w)
 439 {
 440         w->root = &tb->tb6_root;
 441         spin_lock_bh(&tb->tb6_lock);
 442         fib6_walk(net, w);
 443         spin_unlock_bh(&tb->tb6_lock);
 444 }
 445
 446 /* Called with rcu_read_lock() */
 447 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
 448 {
 449         struct fib6_dump_arg arg;
 450         struct fib6_walker *w;
 451         unsigned int h;
 452
 453         w = kzalloc(sizeof(*w), GFP_ATOMIC);
 454         if (!w)
 455                 return -ENOMEM;
 456
 457         w->func = fib6_node_dump;
 458         arg.net = net;
 459         arg.nb = nb;
 460         w->args = &arg;
 461
 462         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 463                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 464                 struct fib6_table *tb;
 465
 466                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 467                         fib6_table_dump(net, tb, w);
 468         }
 469
 470         kfree(w);
 471
 472         return 0;
 473 }
 474
 475 static int fib6_dump_node(struct fib6_walker *w)
 476 {
 477         int res;
 478         struct fib6_info *rt;
 479
 480         for_each_fib6_walker_rt(w) {
 481                 res = rt6_dump_route(rt, w->args);
 482                 if (res < 0) {
 483                         /* Frame is full, suspend walking */
 484                         w->leaf = rt;
 485                         return 1;
 486                 }
 487
 488                 /* Multipath routes are dumped in one route with the
 489                  * RTA_MULTIPATH attribute. Jump 'rt' to point to the
 490                  * last sibling of this route (no need to dump the
 491                  * sibling routes again)
 492                  */
 493                 if (rt->fib6_nsiblings)
 494                         rt = list_last_entry(&rt->fib6_siblings,
 495                                              struct fib6_info,
 496                                              fib6_siblings);
 497         }
 498         w->leaf = NULL;
 499         return 0;
 500 }
 501
 502 static void fib6_dump_end(struct netlink_callback *cb)
 503 {
 504         struct net *net = sock_net(cb->skb->sk);
 505         struct fib6_walker *w = (void *)cb->args[2];
 506
 507         if (w) {
 508                 if (cb->args[4]) {
 509                         cb->args[4] = 0;
 510                         fib6_walker_unlink(net, w);
 511                 }
 512                 cb->args[2] = 0;
 513                 kfree(w);
 514         }
 515         cb->done = (void *)cb->args[3];
 516         cb->args[1] = 3;
 517 }
 518
 519 static int fib6_dump_done(struct netlink_callback *cb)
 520 {
 521         fib6_dump_end(cb);
 522         return cb->done ? cb->done(cb) : 0;
 523 }
 524
 525 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
 526                            struct netlink_callback *cb)
 527 {
 528         struct net *net = sock_net(skb->sk);
 529         struct fib6_walker *w;
 530         int res;
 531
 532         w = (void *)cb->args[2];
 533         w->root = &table->tb6_root;
 534
 535         if (cb->args[4] == 0) {
 536                 w->count = 0;
 537                 w->skip = 0;
 538
 539                 spin_lock_bh(&table->tb6_lock);
 540                 res = fib6_walk(net, w);
 541                 spin_unlock_bh(&table->tb6_lock);
 542                 if (res > 0) {
 543                         cb->args[4] = 1;
 544                         cb->args[5] = w->root->fn_sernum;
 545                 }
 546         } else {
 547                 if (cb->args[5] != w->root->fn_sernum) {
 548                         /* Begin at the root if the tree changed */
 549                         cb->args[5] = w->root->fn_sernum;
 550                         w->state = FWS_INIT;
 551                         w->node = w->root;
 552                         w->skip = w->count;
 553                 } else
 554                         w->skip = 0;
 555
 556                 spin_lock_bh(&table->tb6_lock);
 557                 res = fib6_walk_continue(w);
 558                 spin_unlock_bh(&table->tb6_lock);
 559                 if (res <= 0) {
 560                         fib6_walker_unlink(net, w);
 561                         cb->args[4] = 0;
 562                 }
 563         }
 564
 565         return res;
 566 }
 567
 568 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 569 {
 570         const struct nlmsghdr *nlh = cb->nlh;
 571         struct net *net = sock_net(skb->sk);
 572         struct rt6_rtnl_dump_arg arg = {};
 573         unsigned int h, s_h;
 574         unsigned int e = 0, s_e;
 575         struct fib6_walker *w;
 576         struct fib6_table *tb;
 577         struct hlist_head *head;
 578         int res = 0;
 579
 580         if (cb->strict_check) {
 581                 int err;
 582
 583                 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
 584                 if (err < 0)
 585                         return err;
 586         } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
 587                 struct rtmsg *rtm = nlmsg_data(nlh);
 588
 589                 arg.filter.flags = rtm->rtm_flags & (RTM_F_PREFIX|RTM_F_CLONED);
 590         }
 591
 592         /* fib entries are never clones */
 593         if (arg.filter.flags & RTM_F_CLONED)
 594                 goto out;
 595
 596         w = (void *)cb->args[2];
 597         if (!w) {
 598                 /* New dump:
 599                  *
 600                  * 1. hook callback destructor.
 601                  */
 602                 cb->args[3] = (long)cb->done;
 603                 cb->done = fib6_dump_done;
 604
 605                 /*
 606                  * 2. allocate and initialize walker.
 607                  */
 608                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
 609                 if (!w)
 610                         return -ENOMEM;
 611                 w->func = fib6_dump_node;
 612                 cb->args[2] = (long)w;
 613         }
 614
 615         arg.skb = skb;
 616         arg.cb = cb;
 617         arg.net = net;
 618         w->args = &arg;
 619
 620         if (arg.filter.table_id) {
 621                 tb = fib6_get_table(net, arg.filter.table_id);
 622                 if (!tb) {
 623                         if (arg.filter.dump_all_families)
 624                                 goto out;
 625
 626                         NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
 627                         return -ENOENT;
 628                 }
 629
 630                 if (!cb->args[0]) {
 631                         res = fib6_dump_table(tb, skb, cb);
 632                         if (!res)
 633                                 cb->args[0] = 1;
 634                 }
 635                 goto out;
 636         }
 637
 638         s_h = cb->args[0];
 639         s_e = cb->args[1];
 640
 641         rcu_read_lock();
 642         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
 643                 e = 0;
 644                 head = &net->ipv6.fib_table_hash[h];
 645                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 646                         if (e < s_e)
 647                                 goto next;
 648                         res = fib6_dump_table(tb, skb, cb);
 649                         if (res != 0)
 650                                 goto out_unlock;
 651 next:
 652                         e++;
 653                 }
 654         }
 655 out_unlock:
 656         rcu_read_unlock();
 657         cb->args[1] = e;
 658         cb->args[0] = h;
 659 out:
 660         res = res < 0 ? res : skb->len;
 661         if (res <= 0)
 662                 fib6_dump_end(cb);
 663         return res;
 664 }
 665
 666 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
 667 {
 668         if (!f6i)
 669                 return;
 670
 671         if (f6i->fib6_metrics == &dst_default_metrics) {
 672                 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
 673
 674                 if (!p)
 675                         return;
 676
 677                 refcount_set(&p->refcnt, 1);
 678                 f6i->fib6_metrics = p;
 679         }
 680
 681         f6i->fib6_metrics->metrics[metric - 1] = val;
 682 }
 683
 684 /*
 685  *      Routing Table
 686  *
 687  *      return the appropriate node for a routing tree "add" operation
 688  *      by either creating and inserting or by returning an existing
 689  *      node.
 690  */
 691
 692 static struct fib6_node *fib6_add_1(struct net *net,
 693                                     struct fib6_table *table,
 694                                     struct fib6_node *root,
 695                                     struct in6_addr *addr, int plen,
 696                                     int offset, int allow_create,
 697                                     int replace_required,
 698                                     struct netlink_ext_ack *extack)
 699 {
 700         struct fib6_node *fn, *in, *ln;
 701         struct fib6_node *pn = NULL;
 702         struct rt6key *key;
 703         int     bit;
 704         __be32  dir = 0;
 705
 706         RT6_TRACE("fib6_add_1\n");
 707
 708         /* insert node in tree */
 709
 710         fn = root;
 711
 712         do {
 713                 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
 714                                             lockdep_is_held(&table->tb6_lock));
 715                 key = (struct rt6key *)((u8 *)leaf + offset);
 716
 717                 /*
 718                  *      Prefix match
 719                  */
 720                 if (plen < fn->fn_bit ||
 721                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
 722                         if (!allow_create) {
 723                                 if (replace_required) {
 724                                         NL_SET_ERR_MSG(extack,
 725                                                        "Can not replace route - no match found");
 726                                         pr_warn("Can't replace route, no match found\n");
 727                                         return ERR_PTR(-ENOENT);
 728                                 }
 729                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 730                         }
 731                         goto insert_above;
 732                 }
 733
 734                 /*
 735                  *      Exact match ?
 736                  */
 737
 738                 if (plen == fn->fn_bit) {
 739                         /* clean up an intermediate node */
 740                         if (!(fn->fn_flags & RTN_RTINFO)) {
 741                                 RCU_INIT_POINTER(fn->leaf, NULL);
 742                                 fib6_info_release(leaf);
 743                         /* remove null_entry in the root node */
 744                         } else if (fn->fn_flags & RTN_TL_ROOT &&
 745                                    rcu_access_pointer(fn->leaf) ==
 746                                    net->ipv6.fib6_null_entry) {
 747                                 RCU_INIT_POINTER(fn->leaf, NULL);
 748                         }
 749
 750                         return fn;
 751                 }
 752
 753                 /*
 754                  *      We have more bits to go
 755                  */
 756
 757                 /* Try to walk down on tree. */
 758                 dir = addr_bit_set(addr, fn->fn_bit);
 759                 pn = fn;
 760                 fn = dir ?
 761                      rcu_dereference_protected(fn->right,
 762                                         lockdep_is_held(&table->tb6_lock)) :
 763                      rcu_dereference_protected(fn->left,
 764                                         lockdep_is_held(&table->tb6_lock));
 765         } while (fn);
 766
 767         if (!allow_create) {
 768                 /* We should not create new node because
 769                  * NLM_F_REPLACE was specified without NLM_F_CREATE
 770                  * I assume it is safe to require NLM_F_CREATE when
 771                  * REPLACE flag is used! Later we may want to remove the
 772                  * check for replace_required, because according
 773                  * to netlink specification, NLM_F_CREATE
 774                  * MUST be specified if new route is created.
 775                  * That would keep IPv6 consistent with IPv4
 776                  */
 777                 if (replace_required) {
 778                         NL_SET_ERR_MSG(extack,
 779                                        "Can not replace route - no match found");
 780                         pr_warn("Can't replace route, no match found\n");
 781                         return ERR_PTR(-ENOENT);
 782                 }
 783                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 784         }
 785         /*
 786          *      We walked to the bottom of tree.
 787          *      Create new leaf node without children.
 788          */
 789
 790         ln = node_alloc(net);
 791
 792         if (!ln)
 793                 return ERR_PTR(-ENOMEM);
 794         ln->fn_bit = plen;
 795         RCU_INIT_POINTER(ln->parent, pn);
 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 insert_above:
 806         /*
 807          * split since we don't have a common prefix anymore or
 808          * we have a less significant route.
 809          * we've to insert an intermediate node on the list
 810          * this new node will point to the one we need to create
 811          * and the current
 812          */
 813
 814         pn = rcu_dereference_protected(fn->parent,
 815                                        lockdep_is_held(&table->tb6_lock));
 816
 817         /* find 1st bit in difference between the 2 addrs.
 818
 819            See comment in __ipv6_addr_diff: bit may be an invalid value,
 820            but if it is >= plen, the value is ignored in any case.
 821          */
 822
 823         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
 824
 825         /*
 826          *              (intermediate)[in]
 827          *                /        \
 828          *      (new leaf node)[ln] (old node)[fn]
 829          */
 830         if (plen > bit) {
 831                 in = node_alloc(net);
 832                 ln = node_alloc(net);
 833
 834                 if (!in || !ln) {
 835                         if (in)
 836                                 node_free_immediate(net, in);
 837                         if (ln)
 838                                 node_free_immediate(net, ln);
 839                         return ERR_PTR(-ENOMEM);
 840                 }
 841
 842                 /*
 843                  * new intermediate node.
 844                  * RTN_RTINFO will
 845                  * be off since that an address that chooses one of
 846                  * the branches would not match less specific routes
 847                  * in the other branch
 848                  */
 849
 850                 in->fn_bit = bit;
 851
 852                 RCU_INIT_POINTER(in->parent, pn);
 853                 in->leaf = fn->leaf;
 854                 atomic_inc(&rcu_dereference_protected(in->leaf,
 855                                 lockdep_is_held(&table->tb6_lock))->fib6_ref);
 856
 857                 /* update parent pointer */
 858                 if (dir)
 859                         rcu_assign_pointer(pn->right, in);
 860                 else
 861                         rcu_assign_pointer(pn->left, in);
 862
 863                 ln->fn_bit = plen;
 864
 865                 RCU_INIT_POINTER(ln->parent, in);
 866                 rcu_assign_pointer(fn->parent, in);
 867
 868                 if (addr_bit_set(addr, bit)) {
 869                         rcu_assign_pointer(in->right, ln);
 870                         rcu_assign_pointer(in->left, fn);
 871                 } else {
 872                         rcu_assign_pointer(in->left, ln);
 873                         rcu_assign_pointer(in->right, fn);
 874                 }
 875         } else { /* plen <= bit */
 876
 877                 /*
 878                  *              (new leaf node)[ln]
 879                  *                /        \
 880                  *           (old node)[fn] NULL
 881                  */
 882
 883                 ln = node_alloc(net);
 884
 885                 if (!ln)
 886                         return ERR_PTR(-ENOMEM);
 887
 888                 ln->fn_bit = plen;
 889
 890                 RCU_INIT_POINTER(ln->parent, pn);
 891
 892                 if (addr_bit_set(&key->addr, plen))
 893                         RCU_INIT_POINTER(ln->right, fn);
 894                 else
 895                         RCU_INIT_POINTER(ln->left, fn);
 896
 897                 rcu_assign_pointer(fn->parent, ln);
 898
 899                 if (dir)
 900                         rcu_assign_pointer(pn->right, ln);
 901                 else
 902                         rcu_assign_pointer(pn->left, ln);
 903         }
 904         return ln;
 905 }
 906
 907 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
 908                                 const struct fib6_table *table)
 909 {
 910         int cpu;
 911
 912         /* release the reference to this fib entry from
 913          * all of its cached pcpu routes
 914          */
 915         for_each_possible_cpu(cpu) {
 916                 struct rt6_info **ppcpu_rt;
 917                 struct rt6_info *pcpu_rt;
 918
 919                 ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
 920                 pcpu_rt = *ppcpu_rt;
 921                 if (pcpu_rt) {
 922                         struct fib6_info *from;
 923
 924                         from = rcu_dereference_protected(pcpu_rt->from,
 925                                              lockdep_is_held(&table->tb6_lock));
 926                         rcu_assign_pointer(pcpu_rt->from, NULL);
 927                         fib6_info_release(from);
 928                 }
 929         }
 930 }
 931
 932 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
 933                           struct net *net)
 934 {
 935         struct fib6_table *table = rt->fib6_table;
 936
 937         if (atomic_read(&rt->fib6_ref) != 1) {
 938                 /* This route is used as dummy address holder in some split
 939                  * nodes. It is not leaked, but it still holds other resources,
 940                  * which must be released in time. So, scan ascendant nodes
 941                  * and replace dummy references to this route with references
 942                  * to still alive ones.
 943                  */
 944                 while (fn) {
 945                         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
 946                                             lockdep_is_held(&table->tb6_lock));
 947                         struct fib6_info *new_leaf;
 948                         if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
 949                                 new_leaf = fib6_find_prefix(net, table, fn);
 950                                 atomic_inc(&new_leaf->fib6_ref);
 951
 952                                 rcu_assign_pointer(fn->leaf, new_leaf);
 953                                 fib6_info_release(rt);
 954                         }
 955                         fn = rcu_dereference_protected(fn->parent,
 956                                     lockdep_is_held(&table->tb6_lock));
 957                 }
 958
 959                 if (rt->rt6i_pcpu)
 960                         fib6_drop_pcpu_from(rt, table);
 961         }
 962 }
 963
 964 /*
 965  *      Insert routing information in a node.
 966  */
 967
 968 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
 969                             struct nl_info *info,
 970                             struct netlink_ext_ack *extack)
 971 {
 972         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
 973                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
 974         struct fib6_info *iter = NULL;
 975         struct fib6_info __rcu **ins;
 976         struct fib6_info __rcu **fallback_ins = NULL;
 977         int replace = (info->nlh &&
 978                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
 979         int add = (!info->nlh ||
 980                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
 981         int found = 0;
 982         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
 983         u16 nlflags = NLM_F_EXCL;
 984         int err;
 985
 986         if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
 987                 nlflags |= NLM_F_APPEND;
 988
 989         ins = &fn->leaf;
 990
 991         for (iter = leaf; iter;
 992              iter = rcu_dereference_protected(iter->fib6_next,
 993                                 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
 994                 /*
 995                  *      Search for duplicates
 996                  */
 997
 998                 if (iter->fib6_metric == rt->fib6_metric) {
 999                         /*
1000                          *      Same priority level
1001                          */
1002                         if (info->nlh &&
1003                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
1004                                 return -EEXIST;
1005
1006                         nlflags &= ~NLM_F_EXCL;
1007                         if (replace) {
1008                                 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1009                                         found++;
1010                                         break;
1011                                 }
1012                                 if (rt_can_ecmp)
1013                                         fallback_ins = fallback_ins ?: ins;
1014                                 goto next_iter;
1015                         }
1016
1017                         if (rt6_duplicate_nexthop(iter, rt)) {
1018                                 if (rt->fib6_nsiblings)
1019                                         rt->fib6_nsiblings = 0;
1020                                 if (!(iter->fib6_flags & RTF_EXPIRES))
1021                                         return -EEXIST;
1022                                 if (!(rt->fib6_flags & RTF_EXPIRES))
1023                                         fib6_clean_expires(iter);
1024                                 else
1025                                         fib6_set_expires(iter, rt->expires);
1026
1027                                 if (rt->fib6_pmtu)
1028                                         fib6_metric_set(iter, RTAX_MTU,
1029                                                         rt->fib6_pmtu);
1030                                 return -EEXIST;
1031                         }
1032                         /* If we have the same destination and the same metric,
1033                          * but not the same gateway, then the route we try to
1034                          * add is sibling to this route, increment our counter
1035                          * of siblings, and later we will add our route to the
1036                          * list.
1037                          * Only static routes (which don't have flag
1038                          * RTF_EXPIRES) are used for ECMPv6.
1039                          *
1040                          * To avoid long list, we only had siblings if the
1041                          * route have a gateway.
1042                          */
1043                         if (rt_can_ecmp &&
1044                             rt6_qualify_for_ecmp(iter))
1045                                 rt->fib6_nsiblings++;
1046                 }
1047
1048                 if (iter->fib6_metric > rt->fib6_metric)
1049                         break;
1050
1051 next_iter:
1052                 ins = &iter->fib6_next;
1053         }
1054
1055         if (fallback_ins && !found) {
1056                 /* No ECMP-able route found, replace first non-ECMP one */
1057                 ins = fallback_ins;
1058                 iter = rcu_dereference_protected(*ins,
1059                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1060                 found++;
1061         }
1062
1063         /* Reset round-robin state, if necessary */
1064         if (ins == &fn->leaf)
1065                 fn->rr_ptr = NULL;
1066
1067         /* Link this route to others same route. */
1068         if (rt->fib6_nsiblings) {
1069                 unsigned int fib6_nsiblings;
1070                 struct fib6_info *sibling, *temp_sibling;
1071
1072                 /* Find the first route that have the same metric */
1073                 sibling = leaf;
1074                 while (sibling) {
1075                         if (sibling->fib6_metric == rt->fib6_metric &&
1076                             rt6_qualify_for_ecmp(sibling)) {
1077                                 list_add_tail(&rt->fib6_siblings,
1078                                               &sibling->fib6_siblings);
1079                                 break;
1080                         }
1081                         sibling = rcu_dereference_protected(sibling->fib6_next,
1082                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1083                 }
1084                 /* For each sibling in the list, increment the counter of
1085                  * siblings. BUG() if counters does not match, list of siblings
1086                  * is broken!
1087                  */
1088                 fib6_nsiblings = 0;
1089                 list_for_each_entry_safe(sibling, temp_sibling,
1090                                          &rt->fib6_siblings, fib6_siblings) {
1091                         sibling->fib6_nsiblings++;
1092                         BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1093                         fib6_nsiblings++;
1094                 }
1095                 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1096                 rt6_multipath_rebalance(temp_sibling);
1097         }
1098
1099         /*
1100          *      insert node
1101          */
1102         if (!replace) {
1103                 if (!add)
1104                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
1105
1106 add:
1107                 nlflags |= NLM_F_CREATE;
1108
1109                 err = call_fib6_entry_notifiers(info->nl_net,
1110                                                 FIB_EVENT_ENTRY_ADD,
1111                                                 rt, extack);
1112                 if (err)
1113                         return err;
1114
1115                 rcu_assign_pointer(rt->fib6_next, iter);
1116                 atomic_inc(&rt->fib6_ref);
1117                 rcu_assign_pointer(rt->fib6_node, fn);
1118                 rcu_assign_pointer(*ins, rt);
1119                 if (!info->skip_notify)
1120                         inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1121                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1122
1123                 if (!(fn->fn_flags & RTN_RTINFO)) {
1124                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1125                         fn->fn_flags |= RTN_RTINFO;
1126                 }
1127
1128         } else {
1129                 int nsiblings;
1130
1131                 if (!found) {
1132                         if (add)
1133                                 goto add;
1134                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1135                         return -ENOENT;
1136                 }
1137
1138                 err = call_fib6_entry_notifiers(info->nl_net,
1139                                                 FIB_EVENT_ENTRY_REPLACE,
1140                                                 rt, extack);
1141                 if (err)
1142                         return err;
1143
1144                 atomic_inc(&rt->fib6_ref);
1145                 rcu_assign_pointer(rt->fib6_node, fn);
1146                 rt->fib6_next = iter->fib6_next;
1147                 rcu_assign_pointer(*ins, rt);
1148                 if (!info->skip_notify)
1149                         inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1150                 if (!(fn->fn_flags & RTN_RTINFO)) {
1151                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1152                         fn->fn_flags |= RTN_RTINFO;
1153                 }
1154                 nsiblings = iter->fib6_nsiblings;
1155                 iter->fib6_node = NULL;
1156                 fib6_purge_rt(iter, fn, info->nl_net);
1157                 if (rcu_access_pointer(fn->rr_ptr) == iter)
1158                         fn->rr_ptr = NULL;
1159                 fib6_info_release(iter);
1160
1161                 if (nsiblings) {
1162                         /* Replacing an ECMP route, remove all siblings */
1163                         ins = &rt->fib6_next;
1164                         iter = rcu_dereference_protected(*ins,
1165                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1166                         while (iter) {
1167                                 if (iter->fib6_metric > rt->fib6_metric)
1168                                         break;
1169                                 if (rt6_qualify_for_ecmp(iter)) {
1170                                         *ins = iter->fib6_next;
1171                                         iter->fib6_node = NULL;
1172                                         fib6_purge_rt(iter, fn, info->nl_net);
1173                                         if (rcu_access_pointer(fn->rr_ptr) == iter)
1174                                                 fn->rr_ptr = NULL;
1175                                         fib6_info_release(iter);
1176                                         nsiblings--;
1177                                         info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1178                                 } else {
1179                                         ins = &iter->fib6_next;
1180                                 }
1181                                 iter = rcu_dereference_protected(*ins,
1182                                         lockdep_is_held(&rt->fib6_table->tb6_lock));
1183                         }
1184                         WARN_ON(nsiblings != 0);
1185                 }
1186         }
1187
1188         return 0;
1189 }
1190
1191 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1192 {
1193         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1194             (rt->fib6_flags & RTF_EXPIRES))
1195                 mod_timer(&net->ipv6.ip6_fib_timer,
1196                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1197 }
1198
1199 void fib6_force_start_gc(struct net *net)
1200 {
1201         if (!timer_pending(&net->ipv6.ip6_fib_timer))
1202                 mod_timer(&net->ipv6.ip6_fib_timer,
1203                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1204 }
1205
1206 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1207                                            int sernum)
1208 {
1209         struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1210                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1211
1212         /* paired with smp_rmb() in rt6_get_cookie_safe() */
1213         smp_wmb();
1214         while (fn) {
1215                 fn->fn_sernum = sernum;
1216                 fn = rcu_dereference_protected(fn->parent,
1217                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1218         }
1219 }
1220
1221 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1222 {
1223         __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1224 }
1225
1226 /*
1227  *      Add routing information to the routing tree.
1228  *      <destination addr>/<source addr>
1229  *      with source addr info in sub-trees
1230  *      Need to own table->tb6_lock
1231  */
1232
1233 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1234              struct nl_info *info, struct netlink_ext_ack *extack)
1235 {
1236         struct fib6_table *table = rt->fib6_table;
1237         struct fib6_node *fn, *pn = NULL;
1238         int err = -ENOMEM;
1239         int allow_create = 1;
1240         int replace_required = 0;
1241         int sernum = fib6_new_sernum(info->nl_net);
1242
1243         if (info->nlh) {
1244                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1245                         allow_create = 0;
1246                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1247                         replace_required = 1;
1248         }
1249         if (!allow_create && !replace_required)
1250                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1251
1252         fn = fib6_add_1(info->nl_net, table, root,
1253                         &rt->fib6_dst.addr, rt->fib6_dst.plen,
1254                         offsetof(struct fib6_info, fib6_dst), allow_create,
1255                         replace_required, extack);
1256         if (IS_ERR(fn)) {
1257                 err = PTR_ERR(fn);
1258                 fn = NULL;
1259                 goto out;
1260         }
1261
1262         pn = fn;
1263
1264 #ifdef CONFIG_IPV6_SUBTREES
1265         if (rt->fib6_src.plen) {
1266                 struct fib6_node *sn;
1267
1268                 if (!rcu_access_pointer(fn->subtree)) {
1269                         struct fib6_node *sfn;
1270
1271                         /*
1272                          * Create subtree.
1273                          *
1274                          *              fn[main tree]
1275                          *              |
1276                          *              sfn[subtree root]
1277                          *                 \
1278                          *                  sn[new leaf node]
1279                          */
1280
1281                         /* Create subtree root node */
1282                         sfn = node_alloc(info->nl_net);
1283                         if (!sfn)
1284                                 goto failure;
1285
1286                         atomic_inc(&info->nl_net->ipv6.fib6_null_entry->fib6_ref);
1287                         rcu_assign_pointer(sfn->leaf,
1288                                            info->nl_net->ipv6.fib6_null_entry);
1289                         sfn->fn_flags = RTN_ROOT;
1290
1291                         /* Now add the first leaf node to new subtree */
1292
1293                         sn = fib6_add_1(info->nl_net, table, sfn,
1294                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1295                                         offsetof(struct fib6_info, fib6_src),
1296                                         allow_create, replace_required, extack);
1297
1298                         if (IS_ERR(sn)) {
1299                                 /* If it is failed, discard just allocated
1300                                    root, and then (in failure) stale node
1301                                    in main tree.
1302                                  */
1303                                 node_free_immediate(info->nl_net, sfn);
1304                                 err = PTR_ERR(sn);
1305                                 goto failure;
1306                         }
1307
1308                         /* Now link new subtree to main tree */
1309                         rcu_assign_pointer(sfn->parent, fn);
1310                         rcu_assign_pointer(fn->subtree, sfn);
1311                 } else {
1312                         sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1313                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1314                                         offsetof(struct fib6_info, fib6_src),
1315                                         allow_create, replace_required, extack);
1316
1317                         if (IS_ERR(sn)) {
1318                                 err = PTR_ERR(sn);
1319                                 goto failure;
1320                         }
1321                 }
1322
1323                 if (!rcu_access_pointer(fn->leaf)) {
1324                         if (fn->fn_flags & RTN_TL_ROOT) {
1325                                 /* put back null_entry for root node */
1326                                 rcu_assign_pointer(fn->leaf,
1327                                             info->nl_net->ipv6.fib6_null_entry);
1328                         } else {
1329                                 atomic_inc(&rt->fib6_ref);
1330                                 rcu_assign_pointer(fn->leaf, rt);
1331                         }
1332                 }
1333                 fn = sn;
1334         }
1335 #endif
1336
1337         err = fib6_add_rt2node(fn, rt, info, extack);
1338         if (!err) {
1339                 __fib6_update_sernum_upto_root(rt, sernum);
1340                 fib6_start_gc(info->nl_net, rt);
1341         }
1342
1343 out:
1344         if (err) {
1345 #ifdef CONFIG_IPV6_SUBTREES
1346                 /*
1347                  * If fib6_add_1 has cleared the old leaf pointer in the
1348                  * super-tree leaf node we have to find a new one for it.
1349                  */
1350                 if (pn != fn) {
1351                         struct fib6_info *pn_leaf =
1352                                 rcu_dereference_protected(pn->leaf,
1353                                     lockdep_is_held(&table->tb6_lock));
1354                         if (pn_leaf == rt) {
1355                                 pn_leaf = NULL;
1356                                 RCU_INIT_POINTER(pn->leaf, NULL);
1357                                 fib6_info_release(rt);
1358                         }
1359                         if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1360                                 pn_leaf = fib6_find_prefix(info->nl_net, table,
1361                                                            pn);
1362 #if RT6_DEBUG >= 2
1363                                 if (!pn_leaf) {
1364                                         WARN_ON(!pn_leaf);
1365                                         pn_leaf =
1366                                             info->nl_net->ipv6.fib6_null_entry;
1367                                 }
1368 #endif
1369                                 fib6_info_hold(pn_leaf);
1370                                 rcu_assign_pointer(pn->leaf, pn_leaf);
1371                         }
1372                 }
1373 #endif
1374                 goto failure;
1375         }
1376         return err;
1377
1378 failure:
1379         /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1380          * 1. fn is an intermediate node and we failed to add the new
1381          * route to it in both subtree creation failure and fib6_add_rt2node()
1382          * failure case.
1383          * 2. fn is the root node in the table and we fail to add the first
1384          * default route to it.
1385          */
1386         if (fn &&
1387             (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1388              (fn->fn_flags & RTN_TL_ROOT &&
1389               !rcu_access_pointer(fn->leaf))))
1390                 fib6_repair_tree(info->nl_net, table, fn);
1391         return err;
1392 }
1393
1394 /*
1395  *      Routing tree lookup
1396  *
1397  */
1398
1399 struct lookup_args {
1400         int                     offset;         /* key offset on fib6_info */
1401         const struct in6_addr   *addr;          /* search key                   */
1402 };
1403
1404 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1405                                             struct lookup_args *args)
1406 {
1407         struct fib6_node *fn;
1408         __be32 dir;
1409
1410         if (unlikely(args->offset == 0))
1411                 return NULL;
1412
1413         /*
1414          *      Descend on a tree
1415          */
1416
1417         fn = root;
1418
1419         for (;;) {
1420                 struct fib6_node *next;
1421
1422                 dir = addr_bit_set(args->addr, fn->fn_bit);
1423
1424                 next = dir ? rcu_dereference(fn->right) :
1425                              rcu_dereference(fn->left);
1426
1427                 if (next) {
1428                         fn = next;
1429                         continue;
1430                 }
1431                 break;
1432         }
1433
1434         while (fn) {
1435                 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1436
1437                 if (subtree || fn->fn_flags & RTN_RTINFO) {
1438                         struct fib6_info *leaf = rcu_dereference(fn->leaf);
1439                         struct rt6key *key;
1440
1441                         if (!leaf)
1442                                 goto backtrack;
1443
1444                         key = (struct rt6key *) ((u8 *)leaf + args->offset);
1445
1446                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1447 #ifdef CONFIG_IPV6_SUBTREES
1448                                 if (subtree) {
1449                                         struct fib6_node *sfn;
1450                                         sfn = fib6_node_lookup_1(subtree,
1451                                                                  args + 1);
1452                                         if (!sfn)
1453                                                 goto backtrack;
1454                                         fn = sfn;
1455                                 }
1456 #endif
1457                                 if (fn->fn_flags & RTN_RTINFO)
1458                                         return fn;
1459                         }
1460                 }
1461 backtrack:
1462                 if (fn->fn_flags & RTN_ROOT)
1463                         break;
1464
1465                 fn = rcu_dereference(fn->parent);
1466         }
1467
1468         return NULL;
1469 }
1470
1471 /* called with rcu_read_lock() held
1472  */
1473 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1474                                    const struct in6_addr *daddr,
1475                                    const struct in6_addr *saddr)
1476 {
1477         struct fib6_node *fn;
1478         struct lookup_args args[] = {
1479                 {
1480                         .offset = offsetof(struct fib6_info, fib6_dst),
1481                         .addr = daddr,
1482                 },
1483 #ifdef CONFIG_IPV6_SUBTREES
1484                 {
1485                         .offset = offsetof(struct fib6_info, fib6_src),
1486                         .addr = saddr,
1487                 },
1488 #endif
1489                 {
1490                         .offset = 0,    /* sentinel */
1491                 }
1492         };
1493
1494         fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1495         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1496                 fn = root;
1497
1498         return fn;
1499 }
1500
1501 /*
1502  *      Get node with specified destination prefix (and source prefix,
1503  *      if subtrees are used)
1504  *      exact_match == true means we try to find fn with exact match of
1505  *      the passed in prefix addr
1506  *      exact_match == false means we try to find fn with longest prefix
1507  *      match of the passed in prefix addr. This is useful for finding fn
1508  *      for cached route as it will be stored in the exception table under
1509  *      the node with longest prefix length.
1510  */
1511
1512
1513 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1514                                        const struct in6_addr *addr,
1515                                        int plen, int offset,
1516                                        bool exact_match)
1517 {
1518         struct fib6_node *fn, *prev = NULL;
1519
1520         for (fn = root; fn ; ) {
1521                 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1522                 struct rt6key *key;
1523
1524                 /* This node is being deleted */
1525                 if (!leaf) {
1526                         if (plen <= fn->fn_bit)
1527                                 goto out;
1528                         else
1529                                 goto next;
1530                 }
1531
1532                 key = (struct rt6key *)((u8 *)leaf + offset);
1533
1534                 /*
1535                  *      Prefix match
1536                  */
1537                 if (plen < fn->fn_bit ||
1538                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1539                         goto out;
1540
1541                 if (plen == fn->fn_bit)
1542                         return fn;
1543
1544                 prev = fn;
1545
1546 next:
1547                 /*
1548                  *      We have more bits to go
1549                  */
1550                 if (addr_bit_set(addr, fn->fn_bit))
1551                         fn = rcu_dereference(fn->right);
1552                 else
1553                         fn = rcu_dereference(fn->left);
1554         }
1555 out:
1556         if (exact_match)
1557                 return NULL;
1558         else
1559                 return prev;
1560 }
1561
1562 struct fib6_node *fib6_locate(struct fib6_node *root,
1563                               const struct in6_addr *daddr, int dst_len,
1564                               const struct in6_addr *saddr, int src_len,
1565                               bool exact_match)
1566 {
1567         struct fib6_node *fn;
1568
1569         fn = fib6_locate_1(root, daddr, dst_len,
1570                            offsetof(struct fib6_info, fib6_dst),
1571                            exact_match);
1572
1573 #ifdef CONFIG_IPV6_SUBTREES
1574         if (src_len) {
1575                 WARN_ON(saddr == NULL);
1576                 if (fn) {
1577                         struct fib6_node *subtree = FIB6_SUBTREE(fn);
1578
1579                         if (subtree) {
1580                                 fn = fib6_locate_1(subtree, saddr, src_len,
1581                                            offsetof(struct fib6_info, fib6_src),
1582                                            exact_match);
1583                         }
1584                 }
1585         }
1586 #endif
1587
1588         if (fn && fn->fn_flags & RTN_RTINFO)
1589                 return fn;
1590
1591         return NULL;
1592 }
1593
1594
1595 /*
1596  *      Deletion
1597  *
1598  */
1599
1600 static struct fib6_info *fib6_find_prefix(struct net *net,
1601                                          struct fib6_table *table,
1602                                          struct fib6_node *fn)
1603 {
1604         struct fib6_node *child_left, *child_right;
1605
1606         if (fn->fn_flags & RTN_ROOT)
1607                 return net->ipv6.fib6_null_entry;
1608
1609         while (fn) {
1610                 child_left = rcu_dereference_protected(fn->left,
1611                                     lockdep_is_held(&table->tb6_lock));
1612                 child_right = rcu_dereference_protected(fn->right,
1613                                     lockdep_is_held(&table->tb6_lock));
1614                 if (child_left)
1615                         return rcu_dereference_protected(child_left->leaf,
1616                                         lockdep_is_held(&table->tb6_lock));
1617                 if (child_right)
1618                         return rcu_dereference_protected(child_right->leaf,
1619                                         lockdep_is_held(&table->tb6_lock));
1620
1621                 fn = FIB6_SUBTREE(fn);
1622         }
1623         return NULL;
1624 }
1625
1626 /*
1627  *      Called to trim the tree of intermediate nodes when possible. "fn"
1628  *      is the node we want to try and remove.
1629  *      Need to own table->tb6_lock
1630  */
1631
1632 static struct fib6_node *fib6_repair_tree(struct net *net,
1633                                           struct fib6_table *table,
1634                                           struct fib6_node *fn)
1635 {
1636         int children;
1637         int nstate;
1638         struct fib6_node *child;
1639         struct fib6_walker *w;
1640         int iter = 0;
1641
1642         /* Set fn->leaf to null_entry for root node. */
1643         if (fn->fn_flags & RTN_TL_ROOT) {
1644                 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1645                 return fn;
1646         }
1647
1648         for (;;) {
1649                 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1650                                             lockdep_is_held(&table->tb6_lock));
1651                 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1652                                             lockdep_is_held(&table->tb6_lock));
1653                 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1654                                             lockdep_is_held(&table->tb6_lock));
1655                 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1656                                             lockdep_is_held(&table->tb6_lock));
1657                 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1658                                             lockdep_is_held(&table->tb6_lock));
1659                 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1660                                             lockdep_is_held(&table->tb6_lock));
1661                 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1662                                             lockdep_is_held(&table->tb6_lock));
1663                 struct fib6_info *new_fn_leaf;
1664
1665                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1666                 iter++;
1667
1668                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1669                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1670                 WARN_ON(fn_leaf);
1671
1672                 children = 0;
1673                 child = NULL;
1674                 if (fn_r)
1675                         child = fn_r, children |= 1;
1676                 if (fn_l)
1677                         child = fn_l, children |= 2;
1678
1679                 if (children == 3 || FIB6_SUBTREE(fn)
1680 #ifdef CONFIG_IPV6_SUBTREES
1681                     /* Subtree root (i.e. fn) may have one child */
1682                     || (children && fn->fn_flags & RTN_ROOT)
1683 #endif
1684                     ) {
1685                         new_fn_leaf = fib6_find_prefix(net, table, fn);
1686 #if RT6_DEBUG >= 2
1687                         if (!new_fn_leaf) {
1688                                 WARN_ON(!new_fn_leaf);
1689                                 new_fn_leaf = net->ipv6.fib6_null_entry;
1690                         }
1691 #endif
1692                         fib6_info_hold(new_fn_leaf);
1693                         rcu_assign_pointer(fn->leaf, new_fn_leaf);
1694                         return pn;
1695                 }
1696
1697 #ifdef CONFIG_IPV6_SUBTREES
1698                 if (FIB6_SUBTREE(pn) == fn) {
1699                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1700                         RCU_INIT_POINTER(pn->subtree, NULL);
1701                         nstate = FWS_L;
1702                 } else {
1703                         WARN_ON(fn->fn_flags & RTN_ROOT);
1704 #endif
1705                         if (pn_r == fn)
1706                                 rcu_assign_pointer(pn->right, child);
1707                         else if (pn_l == fn)
1708                                 rcu_assign_pointer(pn->left, child);
1709 #if RT6_DEBUG >= 2
1710                         else
1711                                 WARN_ON(1);
1712 #endif
1713                         if (child)
1714                                 rcu_assign_pointer(child->parent, pn);
1715                         nstate = FWS_R;
1716 #ifdef CONFIG_IPV6_SUBTREES
1717                 }
1718 #endif
1719
1720                 read_lock(&net->ipv6.fib6_walker_lock);
1721                 FOR_WALKERS(net, w) {
1722                         if (!child) {
1723                                 if (w->node == fn) {
1724                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1725                                         w->node = pn;
1726                                         w->state = nstate;
1727                                 }
1728                         } else {
1729                                 if (w->node == fn) {
1730                                         w->node = child;
1731                                         if (children&2) {
1732                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1733                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1734                                         } else {
1735                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1736                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1737                                         }
1738                                 }
1739                         }
1740                 }
1741                 read_unlock(&net->ipv6.fib6_walker_lock);
1742
1743                 node_free(net, fn);
1744                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1745                         return pn;
1746
1747                 RCU_INIT_POINTER(pn->leaf, NULL);
1748                 fib6_info_release(pn_leaf);
1749                 fn = pn;
1750         }
1751 }
1752
1753 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1754                            struct fib6_info __rcu **rtp, struct nl_info *info)
1755 {
1756         struct fib6_walker *w;
1757         struct fib6_info *rt = rcu_dereference_protected(*rtp,
1758                                     lockdep_is_held(&table->tb6_lock));
1759         struct net *net = info->nl_net;
1760
1761         RT6_TRACE("fib6_del_route\n");
1762
1763         /* Unlink it */
1764         *rtp = rt->fib6_next;
1765         rt->fib6_node = NULL;
1766         net->ipv6.rt6_stats->fib_rt_entries--;
1767         net->ipv6.rt6_stats->fib_discarded_routes++;
1768
1769         /* Flush all cached dst in exception table */
1770         rt6_flush_exceptions(rt);
1771
1772         /* Reset round-robin state, if necessary */
1773         if (rcu_access_pointer(fn->rr_ptr) == rt)
1774                 fn->rr_ptr = NULL;
1775
1776         /* Remove this entry from other siblings */
1777         if (rt->fib6_nsiblings) {
1778                 struct fib6_info *sibling, *next_sibling;
1779
1780                 list_for_each_entry_safe(sibling, next_sibling,
1781                                          &rt->fib6_siblings, fib6_siblings)
1782                         sibling->fib6_nsiblings--;
1783                 rt->fib6_nsiblings = 0;
1784                 list_del_init(&rt->fib6_siblings);
1785                 rt6_multipath_rebalance(next_sibling);
1786         }
1787
1788         /* Adjust walkers */
1789         read_lock(&net->ipv6.fib6_walker_lock);
1790         FOR_WALKERS(net, w) {
1791                 if (w->state == FWS_C && w->leaf == rt) {
1792                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1793                         w->leaf = rcu_dereference_protected(rt->fib6_next,
1794                                             lockdep_is_held(&table->tb6_lock));
1795                         if (!w->leaf)
1796                                 w->state = FWS_U;
1797                 }
1798         }
1799         read_unlock(&net->ipv6.fib6_walker_lock);
1800
1801         /* If it was last route, call fib6_repair_tree() to:
1802          * 1. For root node, put back null_entry as how the table was created.
1803          * 2. For other nodes, expunge its radix tree node.
1804          */
1805         if (!rcu_access_pointer(fn->leaf)) {
1806                 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1807                         fn->fn_flags &= ~RTN_RTINFO;
1808                         net->ipv6.rt6_stats->fib_route_nodes--;
1809                 }
1810                 fn = fib6_repair_tree(net, table, fn);
1811         }
1812
1813         fib6_purge_rt(rt, fn, net);
1814
1815         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1816         if (!info->skip_notify)
1817                 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1818         fib6_info_release(rt);
1819 }
1820
1821 /* Need to own table->tb6_lock */
1822 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1823 {
1824         struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1825                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1826         struct fib6_table *table = rt->fib6_table;
1827         struct net *net = info->nl_net;
1828         struct fib6_info __rcu **rtp;
1829         struct fib6_info __rcu **rtp_next;
1830
1831         if (!fn || rt == net->ipv6.fib6_null_entry)
1832                 return -ENOENT;
1833
1834         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1835
1836         /*
1837          *      Walk the leaf entries looking for ourself
1838          */
1839
1840         for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1841                 struct fib6_info *cur = rcu_dereference_protected(*rtp,
1842                                         lockdep_is_held(&table->tb6_lock));
1843                 if (rt == cur) {
1844                         fib6_del_route(table, fn, rtp, info);
1845                         return 0;
1846                 }
1847                 rtp_next = &cur->fib6_next;
1848         }
1849         return -ENOENT;
1850 }
1851
1852 /*
1853  *      Tree traversal function.
1854  *
1855  *      Certainly, it is not interrupt safe.
1856  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1857  *      It means, that we can modify tree during walking
1858  *      and use this function for garbage collection, clone pruning,
1859  *      cleaning tree when a device goes down etc. etc.
1860  *
1861  *      It guarantees that every node will be traversed,
1862  *      and that it will be traversed only once.
1863  *
1864  *      Callback function w->func may return:
1865  *      0 -> continue walking.
1866  *      positive value -> walking is suspended (used by tree dumps,
1867  *      and probably by gc, if it will be split to several slices)
1868  *      negative value -> terminate walking.
1869  *
1870  *      The function itself returns:
1871  *      0   -> walk is complete.
1872  *      >0  -> walk is incomplete (i.e. suspended)
1873  *      <0  -> walk is terminated by an error.
1874  *
1875  *      This function is called with tb6_lock held.
1876  */
1877
1878 static int fib6_walk_continue(struct fib6_walker *w)
1879 {
1880         struct fib6_node *fn, *pn, *left, *right;
1881
1882         /* w->root should always be table->tb6_root */
1883         WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1884
1885         for (;;) {
1886                 fn = w->node;
1887                 if (!fn)
1888                         return 0;
1889
1890                 switch (w->state) {
1891 #ifdef CONFIG_IPV6_SUBTREES
1892                 case FWS_S:
1893                         if (FIB6_SUBTREE(fn)) {
1894                                 w->node = FIB6_SUBTREE(fn);
1895                                 continue;
1896                         }
1897                         w->state = FWS_L;
1898 #endif
1899                         /* fall through */
1900                 case FWS_L:
1901                         left = rcu_dereference_protected(fn->left, 1);
1902                         if (left) {
1903                                 w->node = left;
1904                                 w->state = FWS_INIT;
1905                                 continue;
1906                         }
1907                         w->state = FWS_R;
1908                         /* fall through */
1909                 case FWS_R:
1910                         right = rcu_dereference_protected(fn->right, 1);
1911                         if (right) {
1912                                 w->node = right;
1913                                 w->state = FWS_INIT;
1914                                 continue;
1915                         }
1916                         w->state = FWS_C;
1917                         w->leaf = rcu_dereference_protected(fn->leaf, 1);
1918                         /* fall through */
1919                 case FWS_C:
1920                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1921                                 int err;
1922
1923                                 if (w->skip) {
1924                                         w->skip--;
1925                                         goto skip;
1926                                 }
1927
1928                                 err = w->func(w);
1929                                 if (err)
1930                                         return err;
1931
1932                                 w->count++;
1933                                 continue;
1934                         }
1935 skip:
1936                         w->state = FWS_U;
1937                         /* fall through */
1938                 case FWS_U:
1939                         if (fn == w->root)
1940                                 return 0;
1941                         pn = rcu_dereference_protected(fn->parent, 1);
1942                         left = rcu_dereference_protected(pn->left, 1);
1943                         right = rcu_dereference_protected(pn->right, 1);
1944                         w->node = pn;
1945 #ifdef CONFIG_IPV6_SUBTREES
1946                         if (FIB6_SUBTREE(pn) == fn) {
1947                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1948                                 w->state = FWS_L;
1949                                 continue;
1950                         }
1951 #endif
1952                         if (left == fn) {
1953                                 w->state = FWS_R;
1954                                 continue;
1955                         }
1956                         if (right == fn) {
1957                                 w->state = FWS_C;
1958                                 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1959                                 continue;
1960                         }
1961 #if RT6_DEBUG >= 2
1962                         WARN_ON(1);
1963 #endif
1964                 }
1965         }
1966 }
1967
1968 static int fib6_walk(struct net *net, struct fib6_walker *w)
1969 {
1970         int res;
1971
1972         w->state = FWS_INIT;
1973         w->node = w->root;
1974
1975         fib6_walker_link(net, w);
1976         res = fib6_walk_continue(w);
1977         if (res <= 0)
1978                 fib6_walker_unlink(net, w);
1979         return res;
1980 }
1981
1982 static int fib6_clean_node(struct fib6_walker *w)
1983 {
1984         int res;
1985         struct fib6_info *rt;
1986         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1987         struct nl_info info = {
1988                 .nl_net = c->net,
1989                 .skip_notify = c->skip_notify,
1990         };
1991
1992         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1993             w->node->fn_sernum != c->sernum)
1994                 w->node->fn_sernum = c->sernum;
1995
1996         if (!c->func) {
1997                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1998                 w->leaf = NULL;
1999                 return 0;
2000         }
2001
2002         for_each_fib6_walker_rt(w) {
2003                 res = c->func(rt, c->arg);
2004                 if (res == -1) {
2005                         w->leaf = rt;
2006                         res = fib6_del(rt, &info);
2007                         if (res) {
2008 #if RT6_DEBUG >= 2
2009                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2010                                          __func__, rt,
2011                                          rcu_access_pointer(rt->fib6_node),
2012                                          res);
2013 #endif
2014                                 continue;
2015                         }
2016                         return 0;
2017                 } else if (res == -2) {
2018                         if (WARN_ON(!rt->fib6_nsiblings))
2019                                 continue;
2020                         rt = list_last_entry(&rt->fib6_siblings,
2021                                              struct fib6_info, fib6_siblings);
2022                         continue;
2023                 }
2024                 WARN_ON(res != 0);
2025         }
2026         w->leaf = rt;
2027         return 0;
2028 }
2029
2030 /*
2031  *      Convenient frontend to tree walker.
2032  *
2033  *      func is called on each route.
2034  *              It may return -2 -> skip multipath route.
2035  *                            -1 -> delete this route.
2036  *                            0  -> continue walking
2037  */
2038
2039 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2040                             int (*func)(struct fib6_info *, void *arg),
2041                             int sernum, void *arg, bool skip_notify)
2042 {
2043         struct fib6_cleaner c;
2044
2045         c.w.root = root;
2046         c.w.func = fib6_clean_node;
2047         c.w.count = 0;
2048         c.w.skip = 0;
2049         c.func = func;
2050         c.sernum = sernum;
2051         c.arg = arg;
2052         c.net = net;
2053         c.skip_notify = skip_notify;
2054
2055         fib6_walk(net, &c.w);
2056 }
2057
2058 static void __fib6_clean_all(struct net *net,
2059                              int (*func)(struct fib6_info *, void *),
2060                              int sernum, void *arg, bool skip_notify)
2061 {
2062         struct fib6_table *table;
2063         struct hlist_head *head;
2064         unsigned int h;
2065
2066         rcu_read_lock();
2067         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2068                 head = &net->ipv6.fib_table_hash[h];
2069                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2070                         spin_lock_bh(&table->tb6_lock);
2071                         fib6_clean_tree(net, &table->tb6_root,
2072                                         func, sernum, arg, skip_notify);
2073                         spin_unlock_bh(&table->tb6_lock);
2074                 }
2075         }
2076         rcu_read_unlock();
2077 }
2078
2079 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2080                     void *arg)
2081 {
2082         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2083 }
2084
2085 void fib6_clean_all_skip_notify(struct net *net,
2086                                 int (*func)(struct fib6_info *, void *),
2087                                 void *arg)
2088 {
2089         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2090 }
2091
2092 static void fib6_flush_trees(struct net *net)
2093 {
2094         int new_sernum = fib6_new_sernum(net);
2095
2096         __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2097 }
2098
2099 /*
2100  *      Garbage collection
2101  */
2102
2103 static int fib6_age(struct fib6_info *rt, void *arg)
2104 {
2105         struct fib6_gc_args *gc_args = arg;
2106         unsigned long now = jiffies;
2107
2108         /*
2109          *      check addrconf expiration here.
2110          *      Routes are expired even if they are in use.
2111          */
2112
2113         if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2114                 if (time_after(now, rt->expires)) {
2115                         RT6_TRACE("expiring %p\n", rt);
2116                         return -1;
2117                 }
2118                 gc_args->more++;
2119         }
2120
2121         /*      Also age clones in the exception table.
2122          *      Note, that clones are aged out
2123          *      only if they are not in use now.
2124          */
2125         rt6_age_exceptions(rt, gc_args, now);
2126
2127         return 0;
2128 }
2129
2130 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2131 {
2132         struct fib6_gc_args gc_args;
2133         unsigned long now;
2134
2135         if (force) {
2136                 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2137         } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2138                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2139                 return;
2140         }
2141         gc_args.timeout = expires ? (int)expires :
2142                           net->ipv6.sysctl.ip6_rt_gc_interval;
2143         gc_args.more = 0;
2144
2145         fib6_clean_all(net, fib6_age, &gc_args);
2146         now = jiffies;
2147         net->ipv6.ip6_rt_last_gc = now;
2148
2149         if (gc_args.more)
2150                 mod_timer(&net->ipv6.ip6_fib_timer,
2151                           round_jiffies(now
2152                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
2153         else
2154                 del_timer(&net->ipv6.ip6_fib_timer);
2155         spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2156 }
2157
2158 static void fib6_gc_timer_cb(struct timer_list *t)
2159 {
2160         struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2161
2162         fib6_run_gc(0, arg, true);
2163 }
2164
2165 static int __net_init fib6_net_init(struct net *net)
2166 {
2167         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2168         int err;
2169
2170         err = fib6_notifier_init(net);
2171         if (err)
2172                 return err;
2173
2174         spin_lock_init(&net->ipv6.fib6_gc_lock);
2175         rwlock_init(&net->ipv6.fib6_walker_lock);
2176         INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2177         timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2178
2179         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2180         if (!net->ipv6.rt6_stats)
2181                 goto out_timer;
2182
2183         /* Avoid false sharing : Use at least a full cache line */
2184         size = max_t(size_t, size, L1_CACHE_BYTES);
2185
2186         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2187         if (!net->ipv6.fib_table_hash)
2188                 goto out_rt6_stats;
2189
2190         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2191                                           GFP_KERNEL);
2192         if (!net->ipv6.fib6_main_tbl)
2193                 goto out_fib_table_hash;
2194
2195         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2196         rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2197                            net->ipv6.fib6_null_entry);
2198         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2199                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2200         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2201
2202 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2203         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2204                                            GFP_KERNEL);
2205         if (!net->ipv6.fib6_local_tbl)
2206                 goto out_fib6_main_tbl;
2207         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2208         rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2209                            net->ipv6.fib6_null_entry);
2210         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2211                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2212         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2213 #endif
2214         fib6_tables_init(net);
2215
2216         return 0;
2217
2218 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2219 out_fib6_main_tbl:
2220         kfree(net->ipv6.fib6_main_tbl);
2221 #endif
2222 out_fib_table_hash:
2223         kfree(net->ipv6.fib_table_hash);
2224 out_rt6_stats:
2225         kfree(net->ipv6.rt6_stats);
2226 out_timer:
2227         fib6_notifier_exit(net);
2228         return -ENOMEM;
2229 }
2230
2231 static void fib6_net_exit(struct net *net)
2232 {
2233         unsigned int i;
2234
2235         del_timer_sync(&net->ipv6.ip6_fib_timer);
2236
2237         for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2238                 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2239                 struct hlist_node *tmp;
2240                 struct fib6_table *tb;
2241
2242                 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2243                         hlist_del(&tb->tb6_hlist);
2244                         fib6_free_table(tb);
2245                 }
2246         }
2247
2248         kfree(net->ipv6.fib_table_hash);
2249         kfree(net->ipv6.rt6_stats);
2250         fib6_notifier_exit(net);
2251 }
2252
2253 static struct pernet_operations fib6_net_ops = {
2254         .init = fib6_net_init,
2255         .exit = fib6_net_exit,
2256 };
2257
2258 int __init fib6_init(void)
2259 {
2260         int ret = -ENOMEM;
2261
2262         fib6_node_kmem = kmem_cache_create("fib6_nodes",
2263                                            sizeof(struct fib6_node),
2264                                            0, SLAB_HWCACHE_ALIGN,
2265                                            NULL);
2266         if (!fib6_node_kmem)
2267                 goto out;
2268
2269         ret = register_pernet_subsys(&fib6_net_ops);
2270         if (ret)
2271                 goto out_kmem_cache_create;
2272
2273         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2274                                    inet6_dump_fib, 0);
2275         if (ret)
2276                 goto out_unregister_subsys;
2277
2278         __fib6_flush_trees = fib6_flush_trees;
2279 out:
2280         return ret;
2281
2282 out_unregister_subsys:
2283         unregister_pernet_subsys(&fib6_net_ops);
2284 out_kmem_cache_create:
2285         kmem_cache_destroy(fib6_node_kmem);
2286         goto out;
2287 }
2288
2289 void fib6_gc_cleanup(void)
2290 {
2291         unregister_pernet_subsys(&fib6_net_ops);
2292         kmem_cache_destroy(fib6_node_kmem);
2293 }
2294
2295 #ifdef CONFIG_PROC_FS
2296 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2297 {
2298         struct fib6_info *rt = v;
2299         struct ipv6_route_iter *iter = seq->private;
2300         const struct net_device *dev;
2301
2302         seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2303
2304 #ifdef CONFIG_IPV6_SUBTREES
2305         seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2306 #else
2307         seq_puts(seq, "00000000000000000000000000000000 00 ");
2308 #endif
2309         if (rt->fib6_flags & RTF_GATEWAY)
2310                 seq_printf(seq, "%pi6", &rt->fib6_nh.nh_gw);
2311         else
2312                 seq_puts(seq, "00000000000000000000000000000000");
2313
2314         dev = rt->fib6_nh.nh_dev;
2315         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2316                    rt->fib6_metric, atomic_read(&rt->fib6_ref), 0,
2317                    rt->fib6_flags, dev ? dev->name : "");
2318         iter->w.leaf = NULL;
2319         return 0;
2320 }
2321
2322 static int ipv6_route_yield(struct fib6_walker *w)
2323 {
2324         struct ipv6_route_iter *iter = w->args;
2325
2326         if (!iter->skip)
2327                 return 1;
2328
2329         do {
2330                 iter->w.leaf = rcu_dereference_protected(
2331                                 iter->w.leaf->fib6_next,
2332                                 lockdep_is_held(&iter->tbl->tb6_lock));
2333                 iter->skip--;
2334                 if (!iter->skip && iter->w.leaf)
2335                         return 1;
2336         } while (iter->w.leaf);
2337
2338         return 0;
2339 }
2340
2341 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2342                                       struct net *net)
2343 {
2344         memset(&iter->w, 0, sizeof(iter->w));
2345         iter->w.func = ipv6_route_yield;
2346         iter->w.root = &iter->tbl->tb6_root;
2347         iter->w.state = FWS_INIT;
2348         iter->w.node = iter->w.root;
2349         iter->w.args = iter;
2350         iter->sernum = iter->w.root->fn_sernum;
2351         INIT_LIST_HEAD(&iter->w.lh);
2352         fib6_walker_link(net, &iter->w);
2353 }
2354
2355 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2356                                                     struct net *net)
2357 {
2358         unsigned int h;
2359         struct hlist_node *node;
2360
2361         if (tbl) {
2362                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2363                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2364         } else {
2365                 h = 0;
2366                 node = NULL;
2367         }
2368
2369         while (!node && h < FIB6_TABLE_HASHSZ) {
2370                 node = rcu_dereference_bh(
2371                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2372         }
2373         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2374 }
2375
2376 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2377 {
2378         if (iter->sernum != iter->w.root->fn_sernum) {
2379                 iter->sernum = iter->w.root->fn_sernum;
2380                 iter->w.state = FWS_INIT;
2381                 iter->w.node = iter->w.root;
2382                 WARN_ON(iter->w.skip);
2383                 iter->w.skip = iter->w.count;
2384         }
2385 }
2386
2387 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2388 {
2389         int r;
2390         struct fib6_info *n;
2391         struct net *net = seq_file_net(seq);
2392         struct ipv6_route_iter *iter = seq->private;
2393
2394         if (!v)
2395                 goto iter_table;
2396
2397         n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2398         if (n) {
2399                 ++*pos;
2400                 return n;
2401         }
2402
2403 iter_table:
2404         ipv6_route_check_sernum(iter);
2405         spin_lock_bh(&iter->tbl->tb6_lock);
2406         r = fib6_walk_continue(&iter->w);
2407         spin_unlock_bh(&iter->tbl->tb6_lock);
2408         if (r > 0) {
2409                 if (v)
2410                         ++*pos;
2411                 return iter->w.leaf;
2412         } else if (r < 0) {
2413                 fib6_walker_unlink(net, &iter->w);
2414                 return NULL;
2415         }
2416         fib6_walker_unlink(net, &iter->w);
2417
2418         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2419         if (!iter->tbl)
2420                 return NULL;
2421
2422         ipv6_route_seq_setup_walk(iter, net);
2423         goto iter_table;
2424 }
2425
2426 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2427         __acquires(RCU_BH)
2428 {
2429         struct net *net = seq_file_net(seq);
2430         struct ipv6_route_iter *iter = seq->private;
2431
2432         rcu_read_lock_bh();
2433         iter->tbl = ipv6_route_seq_next_table(NULL, net);
2434         iter->skip = *pos;
2435
2436         if (iter->tbl) {
2437                 ipv6_route_seq_setup_walk(iter, net);
2438                 return ipv6_route_seq_next(seq, NULL, pos);
2439         } else {
2440                 return NULL;
2441         }
2442 }
2443
2444 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2445 {
2446         struct fib6_walker *w = &iter->w;
2447         return w->node && !(w->state == FWS_U && w->node == w->root);
2448 }
2449
2450 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2451         __releases(RCU_BH)
2452 {
2453         struct net *net = seq_file_net(seq);
2454         struct ipv6_route_iter *iter = seq->private;
2455
2456         if (ipv6_route_iter_active(iter))
2457                 fib6_walker_unlink(net, &iter->w);
2458
2459         rcu_read_unlock_bh();
2460 }
2461
2462 const struct seq_operations ipv6_route_seq_ops = {
2463         .start  = ipv6_route_seq_start,
2464         .next   = ipv6_route_seq_next,
2465         .stop   = ipv6_route_seq_stop,
2466         .show   = ipv6_route_seq_show
2467 };
2468 #endif /* CONFIG_PROC_FS */