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         INIT_LIST_HEAD(&f6i->fib6_siblings);
 159         refcount_set(&f6i->fib6_ref, 1);
 160
 161         return f6i;
 162 }
 163
 164 void fib6_info_destroy_rcu(struct rcu_head *head)
 165 {
 166         struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
 167
 168         WARN_ON(f6i->fib6_node);
 169
 170         fib6_nh_release(&f6i->fib6_nh);
 171         ip_fib_metrics_put(f6i->fib6_metrics);
 172         kfree(f6i);
 173 }
 174 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
 175
 176 static struct fib6_node *node_alloc(struct net *net)
 177 {
 178         struct fib6_node *fn;
 179
 180         fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
 181         if (fn)
 182                 net->ipv6.rt6_stats->fib_nodes++;
 183
 184         return fn;
 185 }
 186
 187 static void node_free_immediate(struct net *net, struct fib6_node *fn)
 188 {
 189         kmem_cache_free(fib6_node_kmem, fn);
 190         net->ipv6.rt6_stats->fib_nodes--;
 191 }
 192
 193 static void node_free_rcu(struct rcu_head *head)
 194 {
 195         struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
 196
 197         kmem_cache_free(fib6_node_kmem, fn);
 198 }
 199
 200 static void node_free(struct net *net, struct fib6_node *fn)
 201 {
 202         call_rcu(&fn->rcu, node_free_rcu);
 203         net->ipv6.rt6_stats->fib_nodes--;
 204 }
 205
 206 static void fib6_free_table(struct fib6_table *table)
 207 {
 208         inetpeer_invalidate_tree(&table->tb6_peers);
 209         kfree(table);
 210 }
 211
 212 static void fib6_link_table(struct net *net, struct fib6_table *tb)
 213 {
 214         unsigned int h;
 215
 216         /*
 217          * Initialize table lock at a single place to give lockdep a key,
 218          * tables aren't visible prior to being linked to the list.
 219          */
 220         spin_lock_init(&tb->tb6_lock);
 221         h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
 222
 223         /*
 224          * No protection necessary, this is the only list mutatation
 225          * operation, tables never disappear once they exist.
 226          */
 227         hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
 228 }
 229
 230 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 231
 232 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
 233 {
 234         struct fib6_table *table;
 235
 236         table = kzalloc(sizeof(*table), GFP_ATOMIC);
 237         if (table) {
 238                 table->tb6_id = id;
 239                 rcu_assign_pointer(table->tb6_root.leaf,
 240                                    net->ipv6.fib6_null_entry);
 241                 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
 242                 inet_peer_base_init(&table->tb6_peers);
 243         }
 244
 245         return table;
 246 }
 247
 248 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 249 {
 250         struct fib6_table *tb;
 251
 252         if (id == 0)
 253                 id = RT6_TABLE_MAIN;
 254         tb = fib6_get_table(net, id);
 255         if (tb)
 256                 return tb;
 257
 258         tb = fib6_alloc_table(net, id);
 259         if (tb)
 260                 fib6_link_table(net, tb);
 261
 262         return tb;
 263 }
 264 EXPORT_SYMBOL_GPL(fib6_new_table);
 265
 266 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 267 {
 268         struct fib6_table *tb;
 269         struct hlist_head *head;
 270         unsigned int h;
 271
 272         if (id == 0)
 273                 id = RT6_TABLE_MAIN;
 274         h = id & (FIB6_TABLE_HASHSZ - 1);
 275         rcu_read_lock();
 276         head = &net->ipv6.fib_table_hash[h];
 277         hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 278                 if (tb->tb6_id == id) {
 279                         rcu_read_unlock();
 280                         return tb;
 281                 }
 282         }
 283         rcu_read_unlock();
 284
 285         return NULL;
 286 }
 287 EXPORT_SYMBOL_GPL(fib6_get_table);
 288
 289 static void __net_init fib6_tables_init(struct net *net)
 290 {
 291         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 292         fib6_link_table(net, net->ipv6.fib6_local_tbl);
 293 }
 294 #else
 295
 296 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 297 {
 298         return fib6_get_table(net, id);
 299 }
 300
 301 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 302 {
 303           return net->ipv6.fib6_main_tbl;
 304 }
 305
 306 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 307                                    const struct sk_buff *skb,
 308                                    int flags, pol_lookup_t lookup)
 309 {
 310         struct rt6_info *rt;
 311
 312         rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
 313         if (rt->dst.error == -EAGAIN) {
 314                 ip6_rt_put(rt);
 315                 rt = net->ipv6.ip6_null_entry;
 316                 dst_hold(&rt->dst);
 317         }
 318
 319         return &rt->dst;
 320 }
 321
 322 /* called with rcu lock held; no reference taken on fib6_info */
 323 int fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
 324                 struct fib6_result *res, int flags)
 325 {
 326         return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6,
 327                                  res, flags);
 328 }
 329
 330 static void __net_init fib6_tables_init(struct net *net)
 331 {
 332         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 333 }
 334
 335 #endif
 336
 337 unsigned int fib6_tables_seq_read(struct net *net)
 338 {
 339         unsigned int h, fib_seq = 0;
 340
 341         rcu_read_lock();
 342         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 343                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 344                 struct fib6_table *tb;
 345
 346                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 347                         fib_seq += tb->fib_seq;
 348         }
 349         rcu_read_unlock();
 350
 351         return fib_seq;
 352 }
 353
 354 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
 355                                     enum fib_event_type event_type,
 356                                     struct fib6_info *rt)
 357 {
 358         struct fib6_entry_notifier_info info = {
 359                 .rt = rt,
 360         };
 361
 362         return call_fib6_notifier(nb, net, event_type, &info.info);
 363 }
 364
 365 int call_fib6_entry_notifiers(struct net *net,
 366                               enum fib_event_type event_type,
 367                               struct fib6_info *rt,
 368                               struct netlink_ext_ack *extack)
 369 {
 370         struct fib6_entry_notifier_info info = {
 371                 .info.extack = extack,
 372                 .rt = rt,
 373         };
 374
 375         rt->fib6_table->fib_seq++;
 376         return call_fib6_notifiers(net, event_type, &info.info);
 377 }
 378
 379 struct fib6_dump_arg {
 380         struct net *net;
 381         struct notifier_block *nb;
 382 };
 383
 384 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
 385 {
 386         if (rt == arg->net->ipv6.fib6_null_entry)
 387                 return;
 388         call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
 389 }
 390
 391 static int fib6_node_dump(struct fib6_walker *w)
 392 {
 393         struct fib6_info *rt;
 394
 395         for_each_fib6_walker_rt(w)
 396                 fib6_rt_dump(rt, w->args);
 397         w->leaf = NULL;
 398         return 0;
 399 }
 400
 401 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
 402                             struct fib6_walker *w)
 403 {
 404         w->root = &tb->tb6_root;
 405         spin_lock_bh(&tb->tb6_lock);
 406         fib6_walk(net, w);
 407         spin_unlock_bh(&tb->tb6_lock);
 408 }
 409
 410 /* Called with rcu_read_lock() */
 411 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
 412 {
 413         struct fib6_dump_arg arg;
 414         struct fib6_walker *w;
 415         unsigned int h;
 416
 417         w = kzalloc(sizeof(*w), GFP_ATOMIC);
 418         if (!w)
 419                 return -ENOMEM;
 420
 421         w->func = fib6_node_dump;
 422         arg.net = net;
 423         arg.nb = nb;
 424         w->args = &arg;
 425
 426         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 427                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 428                 struct fib6_table *tb;
 429
 430                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 431                         fib6_table_dump(net, tb, w);
 432         }
 433
 434         kfree(w);
 435
 436         return 0;
 437 }
 438
 439 static int fib6_dump_node(struct fib6_walker *w)
 440 {
 441         int res;
 442         struct fib6_info *rt;
 443
 444         for_each_fib6_walker_rt(w) {
 445                 res = rt6_dump_route(rt, w->args);
 446                 if (res < 0) {
 447                         /* Frame is full, suspend walking */
 448                         w->leaf = rt;
 449                         return 1;
 450                 }
 451
 452                 /* Multipath routes are dumped in one route with the
 453                  * RTA_MULTIPATH attribute. Jump 'rt' to point to the
 454                  * last sibling of this route (no need to dump the
 455                  * sibling routes again)
 456                  */
 457                 if (rt->fib6_nsiblings)
 458                         rt = list_last_entry(&rt->fib6_siblings,
 459                                              struct fib6_info,
 460                                              fib6_siblings);
 461         }
 462         w->leaf = NULL;
 463         return 0;
 464 }
 465
 466 static void fib6_dump_end(struct netlink_callback *cb)
 467 {
 468         struct net *net = sock_net(cb->skb->sk);
 469         struct fib6_walker *w = (void *)cb->args[2];
 470
 471         if (w) {
 472                 if (cb->args[4]) {
 473                         cb->args[4] = 0;
 474                         fib6_walker_unlink(net, w);
 475                 }
 476                 cb->args[2] = 0;
 477                 kfree(w);
 478         }
 479         cb->done = (void *)cb->args[3];
 480         cb->args[1] = 3;
 481 }
 482
 483 static int fib6_dump_done(struct netlink_callback *cb)
 484 {
 485         fib6_dump_end(cb);
 486         return cb->done ? cb->done(cb) : 0;
 487 }
 488
 489 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
 490                            struct netlink_callback *cb)
 491 {
 492         struct net *net = sock_net(skb->sk);
 493         struct fib6_walker *w;
 494         int res;
 495
 496         w = (void *)cb->args[2];
 497         w->root = &table->tb6_root;
 498
 499         if (cb->args[4] == 0) {
 500                 w->count = 0;
 501                 w->skip = 0;
 502
 503                 spin_lock_bh(&table->tb6_lock);
 504                 res = fib6_walk(net, w);
 505                 spin_unlock_bh(&table->tb6_lock);
 506                 if (res > 0) {
 507                         cb->args[4] = 1;
 508                         cb->args[5] = w->root->fn_sernum;
 509                 }
 510         } else {
 511                 if (cb->args[5] != w->root->fn_sernum) {
 512                         /* Begin at the root if the tree changed */
 513                         cb->args[5] = w->root->fn_sernum;
 514                         w->state = FWS_INIT;
 515                         w->node = w->root;
 516                         w->skip = w->count;
 517                 } else
 518                         w->skip = 0;
 519
 520                 spin_lock_bh(&table->tb6_lock);
 521                 res = fib6_walk_continue(w);
 522                 spin_unlock_bh(&table->tb6_lock);
 523                 if (res <= 0) {
 524                         fib6_walker_unlink(net, w);
 525                         cb->args[4] = 0;
 526                 }
 527         }
 528
 529         return res;
 530 }
 531
 532 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 533 {
 534         const struct nlmsghdr *nlh = cb->nlh;
 535         struct net *net = sock_net(skb->sk);
 536         struct rt6_rtnl_dump_arg arg = {};
 537         unsigned int h, s_h;
 538         unsigned int e = 0, s_e;
 539         struct fib6_walker *w;
 540         struct fib6_table *tb;
 541         struct hlist_head *head;
 542         int res = 0;
 543
 544         if (cb->strict_check) {
 545                 int err;
 546
 547                 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
 548                 if (err < 0)
 549                         return err;
 550         } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
 551                 struct rtmsg *rtm = nlmsg_data(nlh);
 552
 553                 arg.filter.flags = rtm->rtm_flags & (RTM_F_PREFIX|RTM_F_CLONED);
 554         }
 555
 556         /* fib entries are never clones */
 557         if (arg.filter.flags & RTM_F_CLONED)
 558                 goto out;
 559
 560         w = (void *)cb->args[2];
 561         if (!w) {
 562                 /* New dump:
 563                  *
 564                  * 1. hook callback destructor.
 565                  */
 566                 cb->args[3] = (long)cb->done;
 567                 cb->done = fib6_dump_done;
 568
 569                 /*
 570                  * 2. allocate and initialize walker.
 571                  */
 572                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
 573                 if (!w)
 574                         return -ENOMEM;
 575                 w->func = fib6_dump_node;
 576                 cb->args[2] = (long)w;
 577         }
 578
 579         arg.skb = skb;
 580         arg.cb = cb;
 581         arg.net = net;
 582         w->args = &arg;
 583
 584         if (arg.filter.table_id) {
 585                 tb = fib6_get_table(net, arg.filter.table_id);
 586                 if (!tb) {
 587                         if (arg.filter.dump_all_families)
 588                                 goto out;
 589
 590                         NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
 591                         return -ENOENT;
 592                 }
 593
 594                 if (!cb->args[0]) {
 595                         res = fib6_dump_table(tb, skb, cb);
 596                         if (!res)
 597                                 cb->args[0] = 1;
 598                 }
 599                 goto out;
 600         }
 601
 602         s_h = cb->args[0];
 603         s_e = cb->args[1];
 604
 605         rcu_read_lock();
 606         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
 607                 e = 0;
 608                 head = &net->ipv6.fib_table_hash[h];
 609                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 610                         if (e < s_e)
 611                                 goto next;
 612                         res = fib6_dump_table(tb, skb, cb);
 613                         if (res != 0)
 614                                 goto out_unlock;
 615 next:
 616                         e++;
 617                 }
 618         }
 619 out_unlock:
 620         rcu_read_unlock();
 621         cb->args[1] = e;
 622         cb->args[0] = h;
 623 out:
 624         res = res < 0 ? res : skb->len;
 625         if (res <= 0)
 626                 fib6_dump_end(cb);
 627         return res;
 628 }
 629
 630 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
 631 {
 632         if (!f6i)
 633                 return;
 634
 635         if (f6i->fib6_metrics == &dst_default_metrics) {
 636                 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
 637
 638                 if (!p)
 639                         return;
 640
 641                 refcount_set(&p->refcnt, 1);
 642                 f6i->fib6_metrics = p;
 643         }
 644
 645         f6i->fib6_metrics->metrics[metric - 1] = val;
 646 }
 647
 648 /*
 649  *      Routing Table
 650  *
 651  *      return the appropriate node for a routing tree "add" operation
 652  *      by either creating and inserting or by returning an existing
 653  *      node.
 654  */
 655
 656 static struct fib6_node *fib6_add_1(struct net *net,
 657                                     struct fib6_table *table,
 658                                     struct fib6_node *root,
 659                                     struct in6_addr *addr, int plen,
 660                                     int offset, int allow_create,
 661                                     int replace_required,
 662                                     struct netlink_ext_ack *extack)
 663 {
 664         struct fib6_node *fn, *in, *ln;
 665         struct fib6_node *pn = NULL;
 666         struct rt6key *key;
 667         int     bit;
 668         __be32  dir = 0;
 669
 670         RT6_TRACE("fib6_add_1\n");
 671
 672         /* insert node in tree */
 673
 674         fn = root;
 675
 676         do {
 677                 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
 678                                             lockdep_is_held(&table->tb6_lock));
 679                 key = (struct rt6key *)((u8 *)leaf + offset);
 680
 681                 /*
 682                  *      Prefix match
 683                  */
 684                 if (plen < fn->fn_bit ||
 685                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
 686                         if (!allow_create) {
 687                                 if (replace_required) {
 688                                         NL_SET_ERR_MSG(extack,
 689                                                        "Can not replace route - no match found");
 690                                         pr_warn("Can't replace route, no match found\n");
 691                                         return ERR_PTR(-ENOENT);
 692                                 }
 693                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 694                         }
 695                         goto insert_above;
 696                 }
 697
 698                 /*
 699                  *      Exact match ?
 700                  */
 701
 702                 if (plen == fn->fn_bit) {
 703                         /* clean up an intermediate node */
 704                         if (!(fn->fn_flags & RTN_RTINFO)) {
 705                                 RCU_INIT_POINTER(fn->leaf, NULL);
 706                                 fib6_info_release(leaf);
 707                         /* remove null_entry in the root node */
 708                         } else if (fn->fn_flags & RTN_TL_ROOT &&
 709                                    rcu_access_pointer(fn->leaf) ==
 710                                    net->ipv6.fib6_null_entry) {
 711                                 RCU_INIT_POINTER(fn->leaf, NULL);
 712                         }
 713
 714                         return fn;
 715                 }
 716
 717                 /*
 718                  *      We have more bits to go
 719                  */
 720
 721                 /* Try to walk down on tree. */
 722                 dir = addr_bit_set(addr, fn->fn_bit);
 723                 pn = fn;
 724                 fn = dir ?
 725                      rcu_dereference_protected(fn->right,
 726                                         lockdep_is_held(&table->tb6_lock)) :
 727                      rcu_dereference_protected(fn->left,
 728                                         lockdep_is_held(&table->tb6_lock));
 729         } while (fn);
 730
 731         if (!allow_create) {
 732                 /* We should not create new node because
 733                  * NLM_F_REPLACE was specified without NLM_F_CREATE
 734                  * I assume it is safe to require NLM_F_CREATE when
 735                  * REPLACE flag is used! Later we may want to remove the
 736                  * check for replace_required, because according
 737                  * to netlink specification, NLM_F_CREATE
 738                  * MUST be specified if new route is created.
 739                  * That would keep IPv6 consistent with IPv4
 740                  */
 741                 if (replace_required) {
 742                         NL_SET_ERR_MSG(extack,
 743                                        "Can not replace route - no match found");
 744                         pr_warn("Can't replace route, no match found\n");
 745                         return ERR_PTR(-ENOENT);
 746                 }
 747                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 748         }
 749         /*
 750          *      We walked to the bottom of tree.
 751          *      Create new leaf node without children.
 752          */
 753
 754         ln = node_alloc(net);
 755
 756         if (!ln)
 757                 return ERR_PTR(-ENOMEM);
 758         ln->fn_bit = plen;
 759         RCU_INIT_POINTER(ln->parent, pn);
 760
 761         if (dir)
 762                 rcu_assign_pointer(pn->right, ln);
 763         else
 764                 rcu_assign_pointer(pn->left, ln);
 765
 766         return ln;
 767
 768
 769 insert_above:
 770         /*
 771          * split since we don't have a common prefix anymore or
 772          * we have a less significant route.
 773          * we've to insert an intermediate node on the list
 774          * this new node will point to the one we need to create
 775          * and the current
 776          */
 777
 778         pn = rcu_dereference_protected(fn->parent,
 779                                        lockdep_is_held(&table->tb6_lock));
 780
 781         /* find 1st bit in difference between the 2 addrs.
 782
 783            See comment in __ipv6_addr_diff: bit may be an invalid value,
 784            but if it is >= plen, the value is ignored in any case.
 785          */
 786
 787         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
 788
 789         /*
 790          *              (intermediate)[in]
 791          *                /        \
 792          *      (new leaf node)[ln] (old node)[fn]
 793          */
 794         if (plen > bit) {
 795                 in = node_alloc(net);
 796                 ln = node_alloc(net);
 797
 798                 if (!in || !ln) {
 799                         if (in)
 800                                 node_free_immediate(net, in);
 801                         if (ln)
 802                                 node_free_immediate(net, ln);
 803                         return ERR_PTR(-ENOMEM);
 804                 }
 805
 806                 /*
 807                  * new intermediate node.
 808                  * RTN_RTINFO will
 809                  * be off since that an address that chooses one of
 810                  * the branches would not match less specific routes
 811                  * in the other branch
 812                  */
 813
 814                 in->fn_bit = bit;
 815
 816                 RCU_INIT_POINTER(in->parent, pn);
 817                 in->leaf = fn->leaf;
 818                 fib6_info_hold(rcu_dereference_protected(in->leaf,
 819                                 lockdep_is_held(&table->tb6_lock)));
 820
 821                 /* update parent pointer */
 822                 if (dir)
 823                         rcu_assign_pointer(pn->right, in);
 824                 else
 825                         rcu_assign_pointer(pn->left, in);
 826
 827                 ln->fn_bit = plen;
 828
 829                 RCU_INIT_POINTER(ln->parent, in);
 830                 rcu_assign_pointer(fn->parent, in);
 831
 832                 if (addr_bit_set(addr, bit)) {
 833                         rcu_assign_pointer(in->right, ln);
 834                         rcu_assign_pointer(in->left, fn);
 835                 } else {
 836                         rcu_assign_pointer(in->left, ln);
 837                         rcu_assign_pointer(in->right, fn);
 838                 }
 839         } else { /* plen <= bit */
 840
 841                 /*
 842                  *              (new leaf node)[ln]
 843                  *                /        \
 844                  *           (old node)[fn] NULL
 845                  */
 846
 847                 ln = node_alloc(net);
 848
 849                 if (!ln)
 850                         return ERR_PTR(-ENOMEM);
 851
 852                 ln->fn_bit = plen;
 853
 854                 RCU_INIT_POINTER(ln->parent, pn);
 855
 856                 if (addr_bit_set(&key->addr, plen))
 857                         RCU_INIT_POINTER(ln->right, fn);
 858                 else
 859                         RCU_INIT_POINTER(ln->left, fn);
 860
 861                 rcu_assign_pointer(fn->parent, ln);
 862
 863                 if (dir)
 864                         rcu_assign_pointer(pn->right, ln);
 865                 else
 866                         rcu_assign_pointer(pn->left, ln);
 867         }
 868         return ln;
 869 }
 870
 871 static void __fib6_drop_pcpu_from(struct fib6_nh *fib6_nh,
 872                                   const struct fib6_info *match,
 873                                   const struct fib6_table *table)
 874 {
 875         int cpu;
 876
 877         if (!fib6_nh->rt6i_pcpu)
 878                 return;
 879
 880         /* release the reference to this fib entry from
 881          * all of its cached pcpu routes
 882          */
 883         for_each_possible_cpu(cpu) {
 884                 struct rt6_info **ppcpu_rt;
 885                 struct rt6_info *pcpu_rt;
 886
 887                 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
 888                 pcpu_rt = *ppcpu_rt;
 889
 890                 /* only dropping the 'from' reference if the cached route
 891                  * is using 'match'. The cached pcpu_rt->from only changes
 892                  * from a fib6_info to NULL (ip6_dst_destroy); it can never
 893                  * change from one fib6_info reference to another
 894                  */
 895                 if (pcpu_rt && rcu_access_pointer(pcpu_rt->from) == match) {
 896                         struct fib6_info *from;
 897
 898                         from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
 899                         fib6_info_release(from);
 900                 }
 901         }
 902 }
 903
 904 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
 905                                 const struct fib6_table *table)
 906 {
 907         struct fib6_nh *fib6_nh;
 908
 909         /* Make sure rt6_make_pcpu_route() wont add other percpu routes
 910          * while we are cleaning them here.
 911          */
 912         f6i->fib6_destroying = 1;
 913         mb(); /* paired with the cmpxchg() in rt6_make_pcpu_route() */
 914
 915         fib6_nh = &f6i->fib6_nh;
 916         __fib6_drop_pcpu_from(fib6_nh, f6i, table);
 917 }
 918
 919 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
 920                           struct net *net)
 921 {
 922         struct fib6_table *table = rt->fib6_table;
 923
 924         fib6_drop_pcpu_from(rt, table);
 925
 926         if (refcount_read(&rt->fib6_ref) != 1) {
 927                 /* This route is used as dummy address holder in some split
 928                  * nodes. It is not leaked, but it still holds other resources,
 929                  * which must be released in time. So, scan ascendant nodes
 930                  * and replace dummy references to this route with references
 931                  * to still alive ones.
 932                  */
 933                 while (fn) {
 934                         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
 935                                             lockdep_is_held(&table->tb6_lock));
 936                         struct fib6_info *new_leaf;
 937                         if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
 938                                 new_leaf = fib6_find_prefix(net, table, fn);
 939                                 fib6_info_hold(new_leaf);
 940
 941                                 rcu_assign_pointer(fn->leaf, new_leaf);
 942                                 fib6_info_release(rt);
 943                         }
 944                         fn = rcu_dereference_protected(fn->parent,
 945                                     lockdep_is_held(&table->tb6_lock));
 946                 }
 947         }
 948 }
 949
 950 /*
 951  *      Insert routing information in a node.
 952  */
 953
 954 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
 955                             struct nl_info *info,
 956                             struct netlink_ext_ack *extack)
 957 {
 958         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
 959                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
 960         struct fib6_info *iter = NULL;
 961         struct fib6_info __rcu **ins;
 962         struct fib6_info __rcu **fallback_ins = NULL;
 963         int replace = (info->nlh &&
 964                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
 965         int add = (!info->nlh ||
 966                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
 967         int found = 0;
 968         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
 969         u16 nlflags = NLM_F_EXCL;
 970         int err;
 971
 972         if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
 973                 nlflags |= NLM_F_APPEND;
 974
 975         ins = &fn->leaf;
 976
 977         for (iter = leaf; iter;
 978              iter = rcu_dereference_protected(iter->fib6_next,
 979                                 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
 980                 /*
 981                  *      Search for duplicates
 982                  */
 983
 984                 if (iter->fib6_metric == rt->fib6_metric) {
 985                         /*
 986                          *      Same priority level
 987                          */
 988                         if (info->nlh &&
 989                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
 990                                 return -EEXIST;
 991
 992                         nlflags &= ~NLM_F_EXCL;
 993                         if (replace) {
 994                                 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
 995                                         found++;
 996                                         break;
 997                                 }
 998                                 if (rt_can_ecmp)
 999                                         fallback_ins = fallback_ins ?: ins;
1000                                 goto next_iter;
1001                         }
1002
1003                         if (rt6_duplicate_nexthop(iter, rt)) {
1004                                 if (rt->fib6_nsiblings)
1005                                         rt->fib6_nsiblings = 0;
1006                                 if (!(iter->fib6_flags & RTF_EXPIRES))
1007                                         return -EEXIST;
1008                                 if (!(rt->fib6_flags & RTF_EXPIRES))
1009                                         fib6_clean_expires(iter);
1010                                 else
1011                                         fib6_set_expires(iter, rt->expires);
1012
1013                                 if (rt->fib6_pmtu)
1014                                         fib6_metric_set(iter, RTAX_MTU,
1015                                                         rt->fib6_pmtu);
1016                                 return -EEXIST;
1017                         }
1018                         /* If we have the same destination and the same metric,
1019                          * but not the same gateway, then the route we try to
1020                          * add is sibling to this route, increment our counter
1021                          * of siblings, and later we will add our route to the
1022                          * list.
1023                          * Only static routes (which don't have flag
1024                          * RTF_EXPIRES) are used for ECMPv6.
1025                          *
1026                          * To avoid long list, we only had siblings if the
1027                          * route have a gateway.
1028                          */
1029                         if (rt_can_ecmp &&
1030                             rt6_qualify_for_ecmp(iter))
1031                                 rt->fib6_nsiblings++;
1032                 }
1033
1034                 if (iter->fib6_metric > rt->fib6_metric)
1035                         break;
1036
1037 next_iter:
1038                 ins = &iter->fib6_next;
1039         }
1040
1041         if (fallback_ins && !found) {
1042                 /* No ECMP-able route found, replace first non-ECMP one */
1043                 ins = fallback_ins;
1044                 iter = rcu_dereference_protected(*ins,
1045                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1046                 found++;
1047         }
1048
1049         /* Reset round-robin state, if necessary */
1050         if (ins == &fn->leaf)
1051                 fn->rr_ptr = NULL;
1052
1053         /* Link this route to others same route. */
1054         if (rt->fib6_nsiblings) {
1055                 unsigned int fib6_nsiblings;
1056                 struct fib6_info *sibling, *temp_sibling;
1057
1058                 /* Find the first route that have the same metric */
1059                 sibling = leaf;
1060                 while (sibling) {
1061                         if (sibling->fib6_metric == rt->fib6_metric &&
1062                             rt6_qualify_for_ecmp(sibling)) {
1063                                 list_add_tail(&rt->fib6_siblings,
1064                                               &sibling->fib6_siblings);
1065                                 break;
1066                         }
1067                         sibling = rcu_dereference_protected(sibling->fib6_next,
1068                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1069                 }
1070                 /* For each sibling in the list, increment the counter of
1071                  * siblings. BUG() if counters does not match, list of siblings
1072                  * is broken!
1073                  */
1074                 fib6_nsiblings = 0;
1075                 list_for_each_entry_safe(sibling, temp_sibling,
1076                                          &rt->fib6_siblings, fib6_siblings) {
1077                         sibling->fib6_nsiblings++;
1078                         BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1079                         fib6_nsiblings++;
1080                 }
1081                 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1082                 rt6_multipath_rebalance(temp_sibling);
1083         }
1084
1085         /*
1086          *      insert node
1087          */
1088         if (!replace) {
1089                 if (!add)
1090                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
1091
1092 add:
1093                 nlflags |= NLM_F_CREATE;
1094
1095                 err = call_fib6_entry_notifiers(info->nl_net,
1096                                                 FIB_EVENT_ENTRY_ADD,
1097                                                 rt, extack);
1098                 if (err)
1099                         return err;
1100
1101                 rcu_assign_pointer(rt->fib6_next, iter);
1102                 fib6_info_hold(rt);
1103                 rcu_assign_pointer(rt->fib6_node, fn);
1104                 rcu_assign_pointer(*ins, rt);
1105                 if (!info->skip_notify)
1106                         inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1107                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1108
1109                 if (!(fn->fn_flags & RTN_RTINFO)) {
1110                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1111                         fn->fn_flags |= RTN_RTINFO;
1112                 }
1113
1114         } else {
1115                 int nsiblings;
1116
1117                 if (!found) {
1118                         if (add)
1119                                 goto add;
1120                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1121                         return -ENOENT;
1122                 }
1123
1124                 err = call_fib6_entry_notifiers(info->nl_net,
1125                                                 FIB_EVENT_ENTRY_REPLACE,
1126                                                 rt, extack);
1127                 if (err)
1128                         return err;
1129
1130                 fib6_info_hold(rt);
1131                 rcu_assign_pointer(rt->fib6_node, fn);
1132                 rt->fib6_next = iter->fib6_next;
1133                 rcu_assign_pointer(*ins, rt);
1134                 if (!info->skip_notify)
1135                         inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1136                 if (!(fn->fn_flags & RTN_RTINFO)) {
1137                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1138                         fn->fn_flags |= RTN_RTINFO;
1139                 }
1140                 nsiblings = iter->fib6_nsiblings;
1141                 iter->fib6_node = NULL;
1142                 fib6_purge_rt(iter, fn, info->nl_net);
1143                 if (rcu_access_pointer(fn->rr_ptr) == iter)
1144                         fn->rr_ptr = NULL;
1145                 fib6_info_release(iter);
1146
1147                 if (nsiblings) {
1148                         /* Replacing an ECMP route, remove all siblings */
1149                         ins = &rt->fib6_next;
1150                         iter = rcu_dereference_protected(*ins,
1151                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1152                         while (iter) {
1153                                 if (iter->fib6_metric > rt->fib6_metric)
1154                                         break;
1155                                 if (rt6_qualify_for_ecmp(iter)) {
1156                                         *ins = iter->fib6_next;
1157                                         iter->fib6_node = NULL;
1158                                         fib6_purge_rt(iter, fn, info->nl_net);
1159                                         if (rcu_access_pointer(fn->rr_ptr) == iter)
1160                                                 fn->rr_ptr = NULL;
1161                                         fib6_info_release(iter);
1162                                         nsiblings--;
1163                                         info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1164                                 } else {
1165                                         ins = &iter->fib6_next;
1166                                 }
1167                                 iter = rcu_dereference_protected(*ins,
1168                                         lockdep_is_held(&rt->fib6_table->tb6_lock));
1169                         }
1170                         WARN_ON(nsiblings != 0);
1171                 }
1172         }
1173
1174         return 0;
1175 }
1176
1177 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1178 {
1179         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1180             (rt->fib6_flags & RTF_EXPIRES))
1181                 mod_timer(&net->ipv6.ip6_fib_timer,
1182                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1183 }
1184
1185 void fib6_force_start_gc(struct net *net)
1186 {
1187         if (!timer_pending(&net->ipv6.ip6_fib_timer))
1188                 mod_timer(&net->ipv6.ip6_fib_timer,
1189                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1190 }
1191
1192 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1193                                            int sernum)
1194 {
1195         struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1196                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1197
1198         /* paired with smp_rmb() in rt6_get_cookie_safe() */
1199         smp_wmb();
1200         while (fn) {
1201                 fn->fn_sernum = sernum;
1202                 fn = rcu_dereference_protected(fn->parent,
1203                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1204         }
1205 }
1206
1207 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1208 {
1209         __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1210 }
1211
1212 /* allow ipv4 to update sernum via ipv6_stub */
1213 void fib6_update_sernum_stub(struct net *net, struct fib6_info *f6i)
1214 {
1215         spin_lock_bh(&f6i->fib6_table->tb6_lock);
1216         fib6_update_sernum_upto_root(net, f6i);
1217         spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1218 }
1219
1220 /*
1221  *      Add routing information to the routing tree.
1222  *      <destination addr>/<source addr>
1223  *      with source addr info in sub-trees
1224  *      Need to own table->tb6_lock
1225  */
1226
1227 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1228              struct nl_info *info, struct netlink_ext_ack *extack)
1229 {
1230         struct fib6_table *table = rt->fib6_table;
1231         struct fib6_node *fn, *pn = NULL;
1232         int err = -ENOMEM;
1233         int allow_create = 1;
1234         int replace_required = 0;
1235         int sernum = fib6_new_sernum(info->nl_net);
1236
1237         if (info->nlh) {
1238                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1239                         allow_create = 0;
1240                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1241                         replace_required = 1;
1242         }
1243         if (!allow_create && !replace_required)
1244                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1245
1246         fn = fib6_add_1(info->nl_net, table, root,
1247                         &rt->fib6_dst.addr, rt->fib6_dst.plen,
1248                         offsetof(struct fib6_info, fib6_dst), allow_create,
1249                         replace_required, extack);
1250         if (IS_ERR(fn)) {
1251                 err = PTR_ERR(fn);
1252                 fn = NULL;
1253                 goto out;
1254         }
1255
1256         pn = fn;
1257
1258 #ifdef CONFIG_IPV6_SUBTREES
1259         if (rt->fib6_src.plen) {
1260                 struct fib6_node *sn;
1261
1262                 if (!rcu_access_pointer(fn->subtree)) {
1263                         struct fib6_node *sfn;
1264
1265                         /*
1266                          * Create subtree.
1267                          *
1268                          *              fn[main tree]
1269                          *              |
1270                          *              sfn[subtree root]
1271                          *                 \
1272                          *                  sn[new leaf node]
1273                          */
1274
1275                         /* Create subtree root node */
1276                         sfn = node_alloc(info->nl_net);
1277                         if (!sfn)
1278                                 goto failure;
1279
1280                         fib6_info_hold(info->nl_net->ipv6.fib6_null_entry);
1281                         rcu_assign_pointer(sfn->leaf,
1282                                            info->nl_net->ipv6.fib6_null_entry);
1283                         sfn->fn_flags = RTN_ROOT;
1284
1285                         /* Now add the first leaf node to new subtree */
1286
1287                         sn = fib6_add_1(info->nl_net, table, sfn,
1288                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1289                                         offsetof(struct fib6_info, fib6_src),
1290                                         allow_create, replace_required, extack);
1291
1292                         if (IS_ERR(sn)) {
1293                                 /* If it is failed, discard just allocated
1294                                    root, and then (in failure) stale node
1295                                    in main tree.
1296                                  */
1297                                 node_free_immediate(info->nl_net, sfn);
1298                                 err = PTR_ERR(sn);
1299                                 goto failure;
1300                         }
1301
1302                         /* Now link new subtree to main tree */
1303                         rcu_assign_pointer(sfn->parent, fn);
1304                         rcu_assign_pointer(fn->subtree, sfn);
1305                 } else {
1306                         sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1307                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1308                                         offsetof(struct fib6_info, fib6_src),
1309                                         allow_create, replace_required, extack);
1310
1311                         if (IS_ERR(sn)) {
1312                                 err = PTR_ERR(sn);
1313                                 goto failure;
1314                         }
1315                 }
1316
1317                 if (!rcu_access_pointer(fn->leaf)) {
1318                         if (fn->fn_flags & RTN_TL_ROOT) {
1319                                 /* put back null_entry for root node */
1320                                 rcu_assign_pointer(fn->leaf,
1321                                             info->nl_net->ipv6.fib6_null_entry);
1322                         } else {
1323                                 fib6_info_hold(rt);
1324                                 rcu_assign_pointer(fn->leaf, rt);
1325                         }
1326                 }
1327                 fn = sn;
1328         }
1329 #endif
1330
1331         err = fib6_add_rt2node(fn, rt, info, extack);
1332         if (!err) {
1333                 __fib6_update_sernum_upto_root(rt, sernum);
1334                 fib6_start_gc(info->nl_net, rt);
1335         }
1336
1337 out:
1338         if (err) {
1339 #ifdef CONFIG_IPV6_SUBTREES
1340                 /*
1341                  * If fib6_add_1 has cleared the old leaf pointer in the
1342                  * super-tree leaf node we have to find a new one for it.
1343                  */
1344                 if (pn != fn) {
1345                         struct fib6_info *pn_leaf =
1346                                 rcu_dereference_protected(pn->leaf,
1347                                     lockdep_is_held(&table->tb6_lock));
1348                         if (pn_leaf == rt) {
1349                                 pn_leaf = NULL;
1350                                 RCU_INIT_POINTER(pn->leaf, NULL);
1351                                 fib6_info_release(rt);
1352                         }
1353                         if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1354                                 pn_leaf = fib6_find_prefix(info->nl_net, table,
1355                                                            pn);
1356 #if RT6_DEBUG >= 2
1357                                 if (!pn_leaf) {
1358                                         WARN_ON(!pn_leaf);
1359                                         pn_leaf =
1360                                             info->nl_net->ipv6.fib6_null_entry;
1361                                 }
1362 #endif
1363                                 fib6_info_hold(pn_leaf);
1364                                 rcu_assign_pointer(pn->leaf, pn_leaf);
1365                         }
1366                 }
1367 #endif
1368                 goto failure;
1369         }
1370         return err;
1371
1372 failure:
1373         /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1374          * 1. fn is an intermediate node and we failed to add the new
1375          * route to it in both subtree creation failure and fib6_add_rt2node()
1376          * failure case.
1377          * 2. fn is the root node in the table and we fail to add the first
1378          * default route to it.
1379          */
1380         if (fn &&
1381             (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1382              (fn->fn_flags & RTN_TL_ROOT &&
1383               !rcu_access_pointer(fn->leaf))))
1384                 fib6_repair_tree(info->nl_net, table, fn);
1385         return err;
1386 }
1387
1388 /*
1389  *      Routing tree lookup
1390  *
1391  */
1392
1393 struct lookup_args {
1394         int                     offset;         /* key offset on fib6_info */
1395         const struct in6_addr   *addr;          /* search key                   */
1396 };
1397
1398 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1399                                             struct lookup_args *args)
1400 {
1401         struct fib6_node *fn;
1402         __be32 dir;
1403
1404         if (unlikely(args->offset == 0))
1405                 return NULL;
1406
1407         /*
1408          *      Descend on a tree
1409          */
1410
1411         fn = root;
1412
1413         for (;;) {
1414                 struct fib6_node *next;
1415
1416                 dir = addr_bit_set(args->addr, fn->fn_bit);
1417
1418                 next = dir ? rcu_dereference(fn->right) :
1419                              rcu_dereference(fn->left);
1420
1421                 if (next) {
1422                         fn = next;
1423                         continue;
1424                 }
1425                 break;
1426         }
1427
1428         while (fn) {
1429                 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1430
1431                 if (subtree || fn->fn_flags & RTN_RTINFO) {
1432                         struct fib6_info *leaf = rcu_dereference(fn->leaf);
1433                         struct rt6key *key;
1434
1435                         if (!leaf)
1436                                 goto backtrack;
1437
1438                         key = (struct rt6key *) ((u8 *)leaf + args->offset);
1439
1440                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1441 #ifdef CONFIG_IPV6_SUBTREES
1442                                 if (subtree) {
1443                                         struct fib6_node *sfn;
1444                                         sfn = fib6_node_lookup_1(subtree,
1445                                                                  args + 1);
1446                                         if (!sfn)
1447                                                 goto backtrack;
1448                                         fn = sfn;
1449                                 }
1450 #endif
1451                                 if (fn->fn_flags & RTN_RTINFO)
1452                                         return fn;
1453                         }
1454                 }
1455 backtrack:
1456                 if (fn->fn_flags & RTN_ROOT)
1457                         break;
1458
1459                 fn = rcu_dereference(fn->parent);
1460         }
1461
1462         return NULL;
1463 }
1464
1465 /* called with rcu_read_lock() held
1466  */
1467 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1468                                    const struct in6_addr *daddr,
1469                                    const struct in6_addr *saddr)
1470 {
1471         struct fib6_node *fn;
1472         struct lookup_args args[] = {
1473                 {
1474                         .offset = offsetof(struct fib6_info, fib6_dst),
1475                         .addr = daddr,
1476                 },
1477 #ifdef CONFIG_IPV6_SUBTREES
1478                 {
1479                         .offset = offsetof(struct fib6_info, fib6_src),
1480                         .addr = saddr,
1481                 },
1482 #endif
1483                 {
1484                         .offset = 0,    /* sentinel */
1485                 }
1486         };
1487
1488         fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1489         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1490                 fn = root;
1491
1492         return fn;
1493 }
1494
1495 /*
1496  *      Get node with specified destination prefix (and source prefix,
1497  *      if subtrees are used)
1498  *      exact_match == true means we try to find fn with exact match of
1499  *      the passed in prefix addr
1500  *      exact_match == false means we try to find fn with longest prefix
1501  *      match of the passed in prefix addr. This is useful for finding fn
1502  *      for cached route as it will be stored in the exception table under
1503  *      the node with longest prefix length.
1504  */
1505
1506
1507 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1508                                        const struct in6_addr *addr,
1509                                        int plen, int offset,
1510                                        bool exact_match)
1511 {
1512         struct fib6_node *fn, *prev = NULL;
1513
1514         for (fn = root; fn ; ) {
1515                 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1516                 struct rt6key *key;
1517
1518                 /* This node is being deleted */
1519                 if (!leaf) {
1520                         if (plen <= fn->fn_bit)
1521                                 goto out;
1522                         else
1523                                 goto next;
1524                 }
1525
1526                 key = (struct rt6key *)((u8 *)leaf + offset);
1527
1528                 /*
1529                  *      Prefix match
1530                  */
1531                 if (plen < fn->fn_bit ||
1532                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1533                         goto out;
1534
1535                 if (plen == fn->fn_bit)
1536                         return fn;
1537
1538                 prev = fn;
1539
1540 next:
1541                 /*
1542                  *      We have more bits to go
1543                  */
1544                 if (addr_bit_set(addr, fn->fn_bit))
1545                         fn = rcu_dereference(fn->right);
1546                 else
1547                         fn = rcu_dereference(fn->left);
1548         }
1549 out:
1550         if (exact_match)
1551                 return NULL;
1552         else
1553                 return prev;
1554 }
1555
1556 struct fib6_node *fib6_locate(struct fib6_node *root,
1557                               const struct in6_addr *daddr, int dst_len,
1558                               const struct in6_addr *saddr, int src_len,
1559                               bool exact_match)
1560 {
1561         struct fib6_node *fn;
1562
1563         fn = fib6_locate_1(root, daddr, dst_len,
1564                            offsetof(struct fib6_info, fib6_dst),
1565                            exact_match);
1566
1567 #ifdef CONFIG_IPV6_SUBTREES
1568         if (src_len) {
1569                 WARN_ON(saddr == NULL);
1570                 if (fn) {
1571                         struct fib6_node *subtree = FIB6_SUBTREE(fn);
1572
1573                         if (subtree) {
1574                                 fn = fib6_locate_1(subtree, saddr, src_len,
1575                                            offsetof(struct fib6_info, fib6_src),
1576                                            exact_match);
1577                         }
1578                 }
1579         }
1580 #endif
1581
1582         if (fn && fn->fn_flags & RTN_RTINFO)
1583                 return fn;
1584
1585         return NULL;
1586 }
1587
1588
1589 /*
1590  *      Deletion
1591  *
1592  */
1593
1594 static struct fib6_info *fib6_find_prefix(struct net *net,
1595                                          struct fib6_table *table,
1596                                          struct fib6_node *fn)
1597 {
1598         struct fib6_node *child_left, *child_right;
1599
1600         if (fn->fn_flags & RTN_ROOT)
1601                 return net->ipv6.fib6_null_entry;
1602
1603         while (fn) {
1604                 child_left = rcu_dereference_protected(fn->left,
1605                                     lockdep_is_held(&table->tb6_lock));
1606                 child_right = rcu_dereference_protected(fn->right,
1607                                     lockdep_is_held(&table->tb6_lock));
1608                 if (child_left)
1609                         return rcu_dereference_protected(child_left->leaf,
1610                                         lockdep_is_held(&table->tb6_lock));
1611                 if (child_right)
1612                         return rcu_dereference_protected(child_right->leaf,
1613                                         lockdep_is_held(&table->tb6_lock));
1614
1615                 fn = FIB6_SUBTREE(fn);
1616         }
1617         return NULL;
1618 }
1619
1620 /*
1621  *      Called to trim the tree of intermediate nodes when possible. "fn"
1622  *      is the node we want to try and remove.
1623  *      Need to own table->tb6_lock
1624  */
1625
1626 static struct fib6_node *fib6_repair_tree(struct net *net,
1627                                           struct fib6_table *table,
1628                                           struct fib6_node *fn)
1629 {
1630         int children;
1631         int nstate;
1632         struct fib6_node *child;
1633         struct fib6_walker *w;
1634         int iter = 0;
1635
1636         /* Set fn->leaf to null_entry for root node. */
1637         if (fn->fn_flags & RTN_TL_ROOT) {
1638                 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1639                 return fn;
1640         }
1641
1642         for (;;) {
1643                 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1644                                             lockdep_is_held(&table->tb6_lock));
1645                 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1646                                             lockdep_is_held(&table->tb6_lock));
1647                 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1648                                             lockdep_is_held(&table->tb6_lock));
1649                 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1650                                             lockdep_is_held(&table->tb6_lock));
1651                 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1652                                             lockdep_is_held(&table->tb6_lock));
1653                 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1654                                             lockdep_is_held(&table->tb6_lock));
1655                 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1656                                             lockdep_is_held(&table->tb6_lock));
1657                 struct fib6_info *new_fn_leaf;
1658
1659                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1660                 iter++;
1661
1662                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1663                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1664                 WARN_ON(fn_leaf);
1665
1666                 children = 0;
1667                 child = NULL;
1668                 if (fn_r)
1669                         child = fn_r, children |= 1;
1670                 if (fn_l)
1671                         child = fn_l, children |= 2;
1672
1673                 if (children == 3 || FIB6_SUBTREE(fn)
1674 #ifdef CONFIG_IPV6_SUBTREES
1675                     /* Subtree root (i.e. fn) may have one child */
1676                     || (children && fn->fn_flags & RTN_ROOT)
1677 #endif
1678                     ) {
1679                         new_fn_leaf = fib6_find_prefix(net, table, fn);
1680 #if RT6_DEBUG >= 2
1681                         if (!new_fn_leaf) {
1682                                 WARN_ON(!new_fn_leaf);
1683                                 new_fn_leaf = net->ipv6.fib6_null_entry;
1684                         }
1685 #endif
1686                         fib6_info_hold(new_fn_leaf);
1687                         rcu_assign_pointer(fn->leaf, new_fn_leaf);
1688                         return pn;
1689                 }
1690
1691 #ifdef CONFIG_IPV6_SUBTREES
1692                 if (FIB6_SUBTREE(pn) == fn) {
1693                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1694                         RCU_INIT_POINTER(pn->subtree, NULL);
1695                         nstate = FWS_L;
1696                 } else {
1697                         WARN_ON(fn->fn_flags & RTN_ROOT);
1698 #endif
1699                         if (pn_r == fn)
1700                                 rcu_assign_pointer(pn->right, child);
1701                         else if (pn_l == fn)
1702                                 rcu_assign_pointer(pn->left, child);
1703 #if RT6_DEBUG >= 2
1704                         else
1705                                 WARN_ON(1);
1706 #endif
1707                         if (child)
1708                                 rcu_assign_pointer(child->parent, pn);
1709                         nstate = FWS_R;
1710 #ifdef CONFIG_IPV6_SUBTREES
1711                 }
1712 #endif
1713
1714                 read_lock(&net->ipv6.fib6_walker_lock);
1715                 FOR_WALKERS(net, w) {
1716                         if (!child) {
1717                                 if (w->node == fn) {
1718                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1719                                         w->node = pn;
1720                                         w->state = nstate;
1721                                 }
1722                         } else {
1723                                 if (w->node == fn) {
1724                                         w->node = child;
1725                                         if (children&2) {
1726                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1727                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1728                                         } else {
1729                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1730                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1731                                         }
1732                                 }
1733                         }
1734                 }
1735                 read_unlock(&net->ipv6.fib6_walker_lock);
1736
1737                 node_free(net, fn);
1738                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1739                         return pn;
1740
1741                 RCU_INIT_POINTER(pn->leaf, NULL);
1742                 fib6_info_release(pn_leaf);
1743                 fn = pn;
1744         }
1745 }
1746
1747 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1748                            struct fib6_info __rcu **rtp, struct nl_info *info)
1749 {
1750         struct fib6_walker *w;
1751         struct fib6_info *rt = rcu_dereference_protected(*rtp,
1752                                     lockdep_is_held(&table->tb6_lock));
1753         struct net *net = info->nl_net;
1754
1755         RT6_TRACE("fib6_del_route\n");
1756
1757         /* Unlink it */
1758         *rtp = rt->fib6_next;
1759         rt->fib6_node = NULL;
1760         net->ipv6.rt6_stats->fib_rt_entries--;
1761         net->ipv6.rt6_stats->fib_discarded_routes++;
1762
1763         /* Flush all cached dst in exception table */
1764         rt6_flush_exceptions(rt);
1765
1766         /* Reset round-robin state, if necessary */
1767         if (rcu_access_pointer(fn->rr_ptr) == rt)
1768                 fn->rr_ptr = NULL;
1769
1770         /* Remove this entry from other siblings */
1771         if (rt->fib6_nsiblings) {
1772                 struct fib6_info *sibling, *next_sibling;
1773
1774                 list_for_each_entry_safe(sibling, next_sibling,
1775                                          &rt->fib6_siblings, fib6_siblings)
1776                         sibling->fib6_nsiblings--;
1777                 rt->fib6_nsiblings = 0;
1778                 list_del_init(&rt->fib6_siblings);
1779                 rt6_multipath_rebalance(next_sibling);
1780         }
1781
1782         /* Adjust walkers */
1783         read_lock(&net->ipv6.fib6_walker_lock);
1784         FOR_WALKERS(net, w) {
1785                 if (w->state == FWS_C && w->leaf == rt) {
1786                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1787                         w->leaf = rcu_dereference_protected(rt->fib6_next,
1788                                             lockdep_is_held(&table->tb6_lock));
1789                         if (!w->leaf)
1790                                 w->state = FWS_U;
1791                 }
1792         }
1793         read_unlock(&net->ipv6.fib6_walker_lock);
1794
1795         /* If it was last route, call fib6_repair_tree() to:
1796          * 1. For root node, put back null_entry as how the table was created.
1797          * 2. For other nodes, expunge its radix tree node.
1798          */
1799         if (!rcu_access_pointer(fn->leaf)) {
1800                 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1801                         fn->fn_flags &= ~RTN_RTINFO;
1802                         net->ipv6.rt6_stats->fib_route_nodes--;
1803                 }
1804                 fn = fib6_repair_tree(net, table, fn);
1805         }
1806
1807         fib6_purge_rt(rt, fn, net);
1808
1809         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1810         if (!info->skip_notify)
1811                 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1812         fib6_info_release(rt);
1813 }
1814
1815 /* Need to own table->tb6_lock */
1816 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1817 {
1818         struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1819                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1820         struct fib6_table *table = rt->fib6_table;
1821         struct net *net = info->nl_net;
1822         struct fib6_info __rcu **rtp;
1823         struct fib6_info __rcu **rtp_next;
1824
1825         if (!fn || rt == net->ipv6.fib6_null_entry)
1826                 return -ENOENT;
1827
1828         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1829
1830         /*
1831          *      Walk the leaf entries looking for ourself
1832          */
1833
1834         for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1835                 struct fib6_info *cur = rcu_dereference_protected(*rtp,
1836                                         lockdep_is_held(&table->tb6_lock));
1837                 if (rt == cur) {
1838                         fib6_del_route(table, fn, rtp, info);
1839                         return 0;
1840                 }
1841                 rtp_next = &cur->fib6_next;
1842         }
1843         return -ENOENT;
1844 }
1845
1846 /*
1847  *      Tree traversal function.
1848  *
1849  *      Certainly, it is not interrupt safe.
1850  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1851  *      It means, that we can modify tree during walking
1852  *      and use this function for garbage collection, clone pruning,
1853  *      cleaning tree when a device goes down etc. etc.
1854  *
1855  *      It guarantees that every node will be traversed,
1856  *      and that it will be traversed only once.
1857  *
1858  *      Callback function w->func may return:
1859  *      0 -> continue walking.
1860  *      positive value -> walking is suspended (used by tree dumps,
1861  *      and probably by gc, if it will be split to several slices)
1862  *      negative value -> terminate walking.
1863  *
1864  *      The function itself returns:
1865  *      0   -> walk is complete.
1866  *      >0  -> walk is incomplete (i.e. suspended)
1867  *      <0  -> walk is terminated by an error.
1868  *
1869  *      This function is called with tb6_lock held.
1870  */
1871
1872 static int fib6_walk_continue(struct fib6_walker *w)
1873 {
1874         struct fib6_node *fn, *pn, *left, *right;
1875
1876         /* w->root should always be table->tb6_root */
1877         WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1878
1879         for (;;) {
1880                 fn = w->node;
1881                 if (!fn)
1882                         return 0;
1883
1884                 switch (w->state) {
1885 #ifdef CONFIG_IPV6_SUBTREES
1886                 case FWS_S:
1887                         if (FIB6_SUBTREE(fn)) {
1888                                 w->node = FIB6_SUBTREE(fn);
1889                                 continue;
1890                         }
1891                         w->state = FWS_L;
1892 #endif
1893                         /* fall through */
1894                 case FWS_L:
1895                         left = rcu_dereference_protected(fn->left, 1);
1896                         if (left) {
1897                                 w->node = left;
1898                                 w->state = FWS_INIT;
1899                                 continue;
1900                         }
1901                         w->state = FWS_R;
1902                         /* fall through */
1903                 case FWS_R:
1904                         right = rcu_dereference_protected(fn->right, 1);
1905                         if (right) {
1906                                 w->node = right;
1907                                 w->state = FWS_INIT;
1908                                 continue;
1909                         }
1910                         w->state = FWS_C;
1911                         w->leaf = rcu_dereference_protected(fn->leaf, 1);
1912                         /* fall through */
1913                 case FWS_C:
1914                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1915                                 int err;
1916
1917                                 if (w->skip) {
1918                                         w->skip--;
1919                                         goto skip;
1920                                 }
1921
1922                                 err = w->func(w);
1923                                 if (err)
1924                                         return err;
1925
1926                                 w->count++;
1927                                 continue;
1928                         }
1929 skip:
1930                         w->state = FWS_U;
1931                         /* fall through */
1932                 case FWS_U:
1933                         if (fn == w->root)
1934                                 return 0;
1935                         pn = rcu_dereference_protected(fn->parent, 1);
1936                         left = rcu_dereference_protected(pn->left, 1);
1937                         right = rcu_dereference_protected(pn->right, 1);
1938                         w->node = pn;
1939 #ifdef CONFIG_IPV6_SUBTREES
1940                         if (FIB6_SUBTREE(pn) == fn) {
1941                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1942                                 w->state = FWS_L;
1943                                 continue;
1944                         }
1945 #endif
1946                         if (left == fn) {
1947                                 w->state = FWS_R;
1948                                 continue;
1949                         }
1950                         if (right == fn) {
1951                                 w->state = FWS_C;
1952                                 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1953                                 continue;
1954                         }
1955 #if RT6_DEBUG >= 2
1956                         WARN_ON(1);
1957 #endif
1958                 }
1959         }
1960 }
1961
1962 static int fib6_walk(struct net *net, struct fib6_walker *w)
1963 {
1964         int res;
1965
1966         w->state = FWS_INIT;
1967         w->node = w->root;
1968
1969         fib6_walker_link(net, w);
1970         res = fib6_walk_continue(w);
1971         if (res <= 0)
1972                 fib6_walker_unlink(net, w);
1973         return res;
1974 }
1975
1976 static int fib6_clean_node(struct fib6_walker *w)
1977 {
1978         int res;
1979         struct fib6_info *rt;
1980         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1981         struct nl_info info = {
1982                 .nl_net = c->net,
1983                 .skip_notify = c->skip_notify,
1984         };
1985
1986         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1987             w->node->fn_sernum != c->sernum)
1988                 w->node->fn_sernum = c->sernum;
1989
1990         if (!c->func) {
1991                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1992                 w->leaf = NULL;
1993                 return 0;
1994         }
1995
1996         for_each_fib6_walker_rt(w) {
1997                 res = c->func(rt, c->arg);
1998                 if (res == -1) {
1999                         w->leaf = rt;
2000                         res = fib6_del(rt, &info);
2001                         if (res) {
2002 #if RT6_DEBUG >= 2
2003                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2004                                          __func__, rt,
2005                                          rcu_access_pointer(rt->fib6_node),
2006                                          res);
2007 #endif
2008                                 continue;
2009                         }
2010                         return 0;
2011                 } else if (res == -2) {
2012                         if (WARN_ON(!rt->fib6_nsiblings))
2013                                 continue;
2014                         rt = list_last_entry(&rt->fib6_siblings,
2015                                              struct fib6_info, fib6_siblings);
2016                         continue;
2017                 }
2018                 WARN_ON(res != 0);
2019         }
2020         w->leaf = rt;
2021         return 0;
2022 }
2023
2024 /*
2025  *      Convenient frontend to tree walker.
2026  *
2027  *      func is called on each route.
2028  *              It may return -2 -> skip multipath route.
2029  *                            -1 -> delete this route.
2030  *                            0  -> continue walking
2031  */
2032
2033 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2034                             int (*func)(struct fib6_info *, void *arg),
2035                             int sernum, void *arg, bool skip_notify)
2036 {
2037         struct fib6_cleaner c;
2038
2039         c.w.root = root;
2040         c.w.func = fib6_clean_node;
2041         c.w.count = 0;
2042         c.w.skip = 0;
2043         c.func = func;
2044         c.sernum = sernum;
2045         c.arg = arg;
2046         c.net = net;
2047         c.skip_notify = skip_notify;
2048
2049         fib6_walk(net, &c.w);
2050 }
2051
2052 static void __fib6_clean_all(struct net *net,
2053                              int (*func)(struct fib6_info *, void *),
2054                              int sernum, void *arg, bool skip_notify)
2055 {
2056         struct fib6_table *table;
2057         struct hlist_head *head;
2058         unsigned int h;
2059
2060         rcu_read_lock();
2061         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2062                 head = &net->ipv6.fib_table_hash[h];
2063                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2064                         spin_lock_bh(&table->tb6_lock);
2065                         fib6_clean_tree(net, &table->tb6_root,
2066                                         func, sernum, arg, skip_notify);
2067                         spin_unlock_bh(&table->tb6_lock);
2068                 }
2069         }
2070         rcu_read_unlock();
2071 }
2072
2073 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2074                     void *arg)
2075 {
2076         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2077 }
2078
2079 void fib6_clean_all_skip_notify(struct net *net,
2080                                 int (*func)(struct fib6_info *, void *),
2081                                 void *arg)
2082 {
2083         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2084 }
2085
2086 static void fib6_flush_trees(struct net *net)
2087 {
2088         int new_sernum = fib6_new_sernum(net);
2089
2090         __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2091 }
2092
2093 /*
2094  *      Garbage collection
2095  */
2096
2097 static int fib6_age(struct fib6_info *rt, void *arg)
2098 {
2099         struct fib6_gc_args *gc_args = arg;
2100         unsigned long now = jiffies;
2101
2102         /*
2103          *      check addrconf expiration here.
2104          *      Routes are expired even if they are in use.
2105          */
2106
2107         if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2108                 if (time_after(now, rt->expires)) {
2109                         RT6_TRACE("expiring %p\n", rt);
2110                         return -1;
2111                 }
2112                 gc_args->more++;
2113         }
2114
2115         /*      Also age clones in the exception table.
2116          *      Note, that clones are aged out
2117          *      only if they are not in use now.
2118          */
2119         rt6_age_exceptions(rt, gc_args, now);
2120
2121         return 0;
2122 }
2123
2124 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2125 {
2126         struct fib6_gc_args gc_args;
2127         unsigned long now;
2128
2129         if (force) {
2130                 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2131         } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2132                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2133                 return;
2134         }
2135         gc_args.timeout = expires ? (int)expires :
2136                           net->ipv6.sysctl.ip6_rt_gc_interval;
2137         gc_args.more = 0;
2138
2139         fib6_clean_all(net, fib6_age, &gc_args);
2140         now = jiffies;
2141         net->ipv6.ip6_rt_last_gc = now;
2142
2143         if (gc_args.more)
2144                 mod_timer(&net->ipv6.ip6_fib_timer,
2145                           round_jiffies(now
2146                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
2147         else
2148                 del_timer(&net->ipv6.ip6_fib_timer);
2149         spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2150 }
2151
2152 static void fib6_gc_timer_cb(struct timer_list *t)
2153 {
2154         struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2155
2156         fib6_run_gc(0, arg, true);
2157 }
2158
2159 static int __net_init fib6_net_init(struct net *net)
2160 {
2161         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2162         int err;
2163
2164         err = fib6_notifier_init(net);
2165         if (err)
2166                 return err;
2167
2168         spin_lock_init(&net->ipv6.fib6_gc_lock);
2169         rwlock_init(&net->ipv6.fib6_walker_lock);
2170         INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2171         timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2172
2173         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2174         if (!net->ipv6.rt6_stats)
2175                 goto out_timer;
2176
2177         /* Avoid false sharing : Use at least a full cache line */
2178         size = max_t(size_t, size, L1_CACHE_BYTES);
2179
2180         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2181         if (!net->ipv6.fib_table_hash)
2182                 goto out_rt6_stats;
2183
2184         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2185                                           GFP_KERNEL);
2186         if (!net->ipv6.fib6_main_tbl)
2187                 goto out_fib_table_hash;
2188
2189         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2190         rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2191                            net->ipv6.fib6_null_entry);
2192         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2193                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2194         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2195
2196 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2197         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2198                                            GFP_KERNEL);
2199         if (!net->ipv6.fib6_local_tbl)
2200                 goto out_fib6_main_tbl;
2201         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2202         rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2203                            net->ipv6.fib6_null_entry);
2204         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2205                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2206         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2207 #endif
2208         fib6_tables_init(net);
2209
2210         return 0;
2211
2212 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2213 out_fib6_main_tbl:
2214         kfree(net->ipv6.fib6_main_tbl);
2215 #endif
2216 out_fib_table_hash:
2217         kfree(net->ipv6.fib_table_hash);
2218 out_rt6_stats:
2219         kfree(net->ipv6.rt6_stats);
2220 out_timer:
2221         fib6_notifier_exit(net);
2222         return -ENOMEM;
2223 }
2224
2225 static void fib6_net_exit(struct net *net)
2226 {
2227         unsigned int i;
2228
2229         del_timer_sync(&net->ipv6.ip6_fib_timer);
2230
2231         for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2232                 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2233                 struct hlist_node *tmp;
2234                 struct fib6_table *tb;
2235
2236                 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2237                         hlist_del(&tb->tb6_hlist);
2238                         fib6_free_table(tb);
2239                 }
2240         }
2241
2242         kfree(net->ipv6.fib_table_hash);
2243         kfree(net->ipv6.rt6_stats);
2244         fib6_notifier_exit(net);
2245 }
2246
2247 static struct pernet_operations fib6_net_ops = {
2248         .init = fib6_net_init,
2249         .exit = fib6_net_exit,
2250 };
2251
2252 int __init fib6_init(void)
2253 {
2254         int ret = -ENOMEM;
2255
2256         fib6_node_kmem = kmem_cache_create("fib6_nodes",
2257                                            sizeof(struct fib6_node),
2258                                            0, SLAB_HWCACHE_ALIGN,
2259                                            NULL);
2260         if (!fib6_node_kmem)
2261                 goto out;
2262
2263         ret = register_pernet_subsys(&fib6_net_ops);
2264         if (ret)
2265                 goto out_kmem_cache_create;
2266
2267         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2268                                    inet6_dump_fib, 0);
2269         if (ret)
2270                 goto out_unregister_subsys;
2271
2272         __fib6_flush_trees = fib6_flush_trees;
2273 out:
2274         return ret;
2275
2276 out_unregister_subsys:
2277         unregister_pernet_subsys(&fib6_net_ops);
2278 out_kmem_cache_create:
2279         kmem_cache_destroy(fib6_node_kmem);
2280         goto out;
2281 }
2282
2283 void fib6_gc_cleanup(void)
2284 {
2285         unregister_pernet_subsys(&fib6_net_ops);
2286         kmem_cache_destroy(fib6_node_kmem);
2287 }
2288
2289 #ifdef CONFIG_PROC_FS
2290 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2291 {
2292         struct fib6_info *rt = v;
2293         struct ipv6_route_iter *iter = seq->private;
2294         unsigned int flags = rt->fib6_flags;
2295         const struct net_device *dev;
2296
2297         seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2298
2299 #ifdef CONFIG_IPV6_SUBTREES
2300         seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2301 #else
2302         seq_puts(seq, "00000000000000000000000000000000 00 ");
2303 #endif
2304         if (rt->fib6_nh.fib_nh_gw_family) {
2305                 flags |= RTF_GATEWAY;
2306                 seq_printf(seq, "%pi6", &rt->fib6_nh.fib_nh_gw6);
2307         } else {
2308                 seq_puts(seq, "00000000000000000000000000000000");
2309         }
2310
2311         dev = rt->fib6_nh.fib_nh_dev;
2312         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2313                    rt->fib6_metric, refcount_read(&rt->fib6_ref), 0,
2314                    flags, dev ? dev->name : "");
2315         iter->w.leaf = NULL;
2316         return 0;
2317 }
2318
2319 static int ipv6_route_yield(struct fib6_walker *w)
2320 {
2321         struct ipv6_route_iter *iter = w->args;
2322
2323         if (!iter->skip)
2324                 return 1;
2325
2326         do {
2327                 iter->w.leaf = rcu_dereference_protected(
2328                                 iter->w.leaf->fib6_next,
2329                                 lockdep_is_held(&iter->tbl->tb6_lock));
2330                 iter->skip--;
2331                 if (!iter->skip && iter->w.leaf)
2332                         return 1;
2333         } while (iter->w.leaf);
2334
2335         return 0;
2336 }
2337
2338 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2339                                       struct net *net)
2340 {
2341         memset(&iter->w, 0, sizeof(iter->w));
2342         iter->w.func = ipv6_route_yield;
2343         iter->w.root = &iter->tbl->tb6_root;
2344         iter->w.state = FWS_INIT;
2345         iter->w.node = iter->w.root;
2346         iter->w.args = iter;
2347         iter->sernum = iter->w.root->fn_sernum;
2348         INIT_LIST_HEAD(&iter->w.lh);
2349         fib6_walker_link(net, &iter->w);
2350 }
2351
2352 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2353                                                     struct net *net)
2354 {
2355         unsigned int h;
2356         struct hlist_node *node;
2357
2358         if (tbl) {
2359                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2360                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2361         } else {
2362                 h = 0;
2363                 node = NULL;
2364         }
2365
2366         while (!node && h < FIB6_TABLE_HASHSZ) {
2367                 node = rcu_dereference_bh(
2368                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2369         }
2370         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2371 }
2372
2373 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2374 {
2375         if (iter->sernum != iter->w.root->fn_sernum) {
2376                 iter->sernum = iter->w.root->fn_sernum;
2377                 iter->w.state = FWS_INIT;
2378                 iter->w.node = iter->w.root;
2379                 WARN_ON(iter->w.skip);
2380                 iter->w.skip = iter->w.count;
2381         }
2382 }
2383
2384 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2385 {
2386         int r;
2387         struct fib6_info *n;
2388         struct net *net = seq_file_net(seq);
2389         struct ipv6_route_iter *iter = seq->private;
2390
2391         if (!v)
2392                 goto iter_table;
2393
2394         n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2395         if (n) {
2396                 ++*pos;
2397                 return n;
2398         }
2399
2400 iter_table:
2401         ipv6_route_check_sernum(iter);
2402         spin_lock_bh(&iter->tbl->tb6_lock);
2403         r = fib6_walk_continue(&iter->w);
2404         spin_unlock_bh(&iter->tbl->tb6_lock);
2405         if (r > 0) {
2406                 if (v)
2407                         ++*pos;
2408                 return iter->w.leaf;
2409         } else if (r < 0) {
2410                 fib6_walker_unlink(net, &iter->w);
2411                 return NULL;
2412         }
2413         fib6_walker_unlink(net, &iter->w);
2414
2415         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2416         if (!iter->tbl)
2417                 return NULL;
2418
2419         ipv6_route_seq_setup_walk(iter, net);
2420         goto iter_table;
2421 }
2422
2423 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2424         __acquires(RCU_BH)
2425 {
2426         struct net *net = seq_file_net(seq);
2427         struct ipv6_route_iter *iter = seq->private;
2428
2429         rcu_read_lock_bh();
2430         iter->tbl = ipv6_route_seq_next_table(NULL, net);
2431         iter->skip = *pos;
2432
2433         if (iter->tbl) {
2434                 ipv6_route_seq_setup_walk(iter, net);
2435                 return ipv6_route_seq_next(seq, NULL, pos);
2436         } else {
2437                 return NULL;
2438         }
2439 }
2440
2441 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2442 {
2443         struct fib6_walker *w = &iter->w;
2444         return w->node && !(w->state == FWS_U && w->node == w->root);
2445 }
2446
2447 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2448         __releases(RCU_BH)
2449 {
2450         struct net *net = seq_file_net(seq);
2451         struct ipv6_route_iter *iter = seq->private;
2452
2453         if (ipv6_route_iter_active(iter))
2454                 fib6_walker_unlink(net, &iter->w);
2455
2456         rcu_read_unlock_bh();
2457 }
2458
2459 const struct seq_operations ipv6_route_seq_ops = {
2460         .start  = ipv6_route_seq_start,
2461         .next   = ipv6_route_seq_next,
2462         .stop   = ipv6_route_seq_stop,
2463         .show   = ipv6_route_seq_show
2464 };
2465 #endif /* CONFIG_PROC_FS */