net/ipv6/ip6_fib.c

   1 /*
   2  *      Linux INET6 implementation
   3  *      Forwarding Information Database
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <roque@di.fc.ul.pt>
   7  *
   8  *      This program is free software; you can redistribute it and/or
   9  *      modify it under the terms of the GNU General Public License
  10  *      as published by the Free Software Foundation; either version
  11  *      2 of the License, or (at your option) any later version.
  12  *
  13  *      Changes:
  14  *      Yuji SEKIYA @USAGI:     Support default route on router node;
  15  *                              remove ip6_null_entry from the top of
  16  *                              routing table.
  17  *      Ville Nuorvala:         Fixed routing subtrees.
  18  */
  19
  20 #define pr_fmt(fmt) "IPv6: " fmt
  21
  22 #include <linux/errno.h>
  23 #include <linux/types.h>
  24 #include <linux/net.h>
  25 #include <linux/route.h>
  26 #include <linux/netdevice.h>
  27 #include <linux/in6.h>
  28 #include <linux/init.h>
  29 #include <linux/list.h>
  30 #include <linux/slab.h>
  31
  32 #include <net/ip.h>
  33 #include <net/ipv6.h>
  34 #include <net/ndisc.h>
  35 #include <net/addrconf.h>
  36 #include <net/lwtunnel.h>
  37 #include <net/fib_notifier.h>
  38
  39 #include <net/ip6_fib.h>
  40 #include <net/ip6_route.h>
  41
  42 static struct kmem_cache *fib6_node_kmem __read_mostly;
  43
  44 struct fib6_cleaner {
  45         struct fib6_walker w;
  46         struct net *net;
  47         int (*func)(struct fib6_info *, void *arg);
  48         int sernum;
  49         void *arg;
  50         bool skip_notify;
  51 };
  52
  53 #ifdef CONFIG_IPV6_SUBTREES
  54 #define FWS_INIT FWS_S
  55 #else
  56 #define FWS_INIT FWS_L
  57 #endif
  58
  59 static struct fib6_info *fib6_find_prefix(struct net *net,
  60                                          struct fib6_table *table,
  61                                          struct fib6_node *fn);
  62 static struct fib6_node *fib6_repair_tree(struct net *net,
  63                                           struct fib6_table *table,
  64                                           struct fib6_node *fn);
  65 static int fib6_walk(struct net *net, struct fib6_walker *w);
  66 static int fib6_walk_continue(struct fib6_walker *w);
  67
  68 /*
  69  *      A routing update causes an increase of the serial number on the
  70  *      affected subtree. This allows for cached routes to be asynchronously
  71  *      tested when modifications are made to the destination cache as a
  72  *      result of redirects, path MTU changes, etc.
  73  */
  74
  75 static void fib6_gc_timer_cb(struct timer_list *t);
  76
  77 #define FOR_WALKERS(net, w) \
  78         list_for_each_entry(w, &(net)->ipv6.fib6_walkers, lh)
  79
  80 static void fib6_walker_link(struct net *net, struct fib6_walker *w)
  81 {
  82         write_lock_bh(&net->ipv6.fib6_walker_lock);
  83         list_add(&w->lh, &net->ipv6.fib6_walkers);
  84         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  85 }
  86
  87 static void fib6_walker_unlink(struct net *net, struct fib6_walker *w)
  88 {
  89         write_lock_bh(&net->ipv6.fib6_walker_lock);
  90         list_del(&w->lh);
  91         write_unlock_bh(&net->ipv6.fib6_walker_lock);
  92 }
  93
  94 static int fib6_new_sernum(struct net *net)
  95 {
  96         int new, old;
  97
  98         do {
  99                 old = atomic_read(&net->ipv6.fib6_sernum);
 100                 new = old < INT_MAX ? old + 1 : 1;
 101         } while (atomic_cmpxchg(&net->ipv6.fib6_sernum,
 102                                 old, new) != old);
 103         return new;
 104 }
 105
 106 enum {
 107         FIB6_NO_SERNUM_CHANGE = 0,
 108 };
 109
 110 void fib6_update_sernum(struct net *net, struct fib6_info *f6i)
 111 {
 112         struct fib6_node *fn;
 113
 114         fn = rcu_dereference_protected(f6i->fib6_node,
 115                         lockdep_is_held(&f6i->fib6_table->tb6_lock));
 116         if (fn)
 117                 fn->fn_sernum = fib6_new_sernum(net);
 118 }
 119
 120 /*
 121  *      Auxiliary address test functions for the radix tree.
 122  *
 123  *      These assume a 32bit processor (although it will work on
 124  *      64bit processors)
 125  */
 126
 127 /*
 128  *      test bit
 129  */
 130 #if defined(__LITTLE_ENDIAN)
 131 # define BITOP_BE32_SWIZZLE     (0x1F & ~7)
 132 #else
 133 # define BITOP_BE32_SWIZZLE     0
 134 #endif
 135
 136 static __be32 addr_bit_set(const void *token, int fn_bit)
 137 {
 138         const __be32 *addr = token;
 139         /*
 140          * Here,
 141          *      1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)
 142          * is optimized version of
 143          *      htonl(1 << ((~fn_bit)&0x1F))
 144          * See include/asm-generic/bitops/le.h.
 145          */
 146         return (__force __be32)(1 << ((~fn_bit ^ BITOP_BE32_SWIZZLE) & 0x1f)) &
 147                addr[fn_bit >> 5];
 148 }
 149
 150 struct fib6_info *fib6_info_alloc(gfp_t gfp_flags)
 151 {
 152         struct fib6_info *f6i;
 153
 154         f6i = kzalloc(sizeof(*f6i), gfp_flags);
 155         if (!f6i)
 156                 return NULL;
 157
 158         f6i->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
 159         if (!f6i->rt6i_pcpu) {
 160                 kfree(f6i);
 161                 return NULL;
 162         }
 163
 164         INIT_LIST_HEAD(&f6i->fib6_siblings);
 165         atomic_inc(&f6i->fib6_ref);
 166
 167         return f6i;
 168 }
 169
 170 void fib6_info_destroy_rcu(struct rcu_head *head)
 171 {
 172         struct fib6_info *f6i = container_of(head, struct fib6_info, rcu);
 173         struct rt6_exception_bucket *bucket;
 174
 175         WARN_ON(f6i->fib6_node);
 176
 177         bucket = rcu_dereference_protected(f6i->rt6i_exception_bucket, 1);
 178         if (bucket) {
 179                 f6i->rt6i_exception_bucket = NULL;
 180                 kfree(bucket);
 181         }
 182
 183         if (f6i->rt6i_pcpu) {
 184                 int cpu;
 185
 186                 for_each_possible_cpu(cpu) {
 187                         struct rt6_info **ppcpu_rt;
 188                         struct rt6_info *pcpu_rt;
 189
 190                         ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
 191                         pcpu_rt = *ppcpu_rt;
 192                         if (pcpu_rt) {
 193                                 dst_dev_put(&pcpu_rt->dst);
 194                                 dst_release(&pcpu_rt->dst);
 195                                 *ppcpu_rt = NULL;
 196                         }
 197                 }
 198
 199                 free_percpu(f6i->rt6i_pcpu);
 200         }
 201
 202         lwtstate_put(f6i->fib6_nh.nh_lwtstate);
 203
 204         if (f6i->fib6_nh.nh_dev)
 205                 dev_put(f6i->fib6_nh.nh_dev);
 206
 207         ip_fib_metrics_put(f6i->fib6_metrics);
 208
 209         kfree(f6i);
 210 }
 211 EXPORT_SYMBOL_GPL(fib6_info_destroy_rcu);
 212
 213 static struct fib6_node *node_alloc(struct net *net)
 214 {
 215         struct fib6_node *fn;
 216
 217         fn = kmem_cache_zalloc(fib6_node_kmem, GFP_ATOMIC);
 218         if (fn)
 219                 net->ipv6.rt6_stats->fib_nodes++;
 220
 221         return fn;
 222 }
 223
 224 static void node_free_immediate(struct net *net, struct fib6_node *fn)
 225 {
 226         kmem_cache_free(fib6_node_kmem, fn);
 227         net->ipv6.rt6_stats->fib_nodes--;
 228 }
 229
 230 static void node_free_rcu(struct rcu_head *head)
 231 {
 232         struct fib6_node *fn = container_of(head, struct fib6_node, rcu);
 233
 234         kmem_cache_free(fib6_node_kmem, fn);
 235 }
 236
 237 static void node_free(struct net *net, struct fib6_node *fn)
 238 {
 239         call_rcu(&fn->rcu, node_free_rcu);
 240         net->ipv6.rt6_stats->fib_nodes--;
 241 }
 242
 243 static void fib6_free_table(struct fib6_table *table)
 244 {
 245         inetpeer_invalidate_tree(&table->tb6_peers);
 246         kfree(table);
 247 }
 248
 249 static void fib6_link_table(struct net *net, struct fib6_table *tb)
 250 {
 251         unsigned int h;
 252
 253         /*
 254          * Initialize table lock at a single place to give lockdep a key,
 255          * tables aren't visible prior to being linked to the list.
 256          */
 257         spin_lock_init(&tb->tb6_lock);
 258         h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
 259
 260         /*
 261          * No protection necessary, this is the only list mutatation
 262          * operation, tables never disappear once they exist.
 263          */
 264         hlist_add_head_rcu(&tb->tb6_hlist, &net->ipv6.fib_table_hash[h]);
 265 }
 266
 267 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 268
 269 static struct fib6_table *fib6_alloc_table(struct net *net, u32 id)
 270 {
 271         struct fib6_table *table;
 272
 273         table = kzalloc(sizeof(*table), GFP_ATOMIC);
 274         if (table) {
 275                 table->tb6_id = id;
 276                 rcu_assign_pointer(table->tb6_root.leaf,
 277                                    net->ipv6.fib6_null_entry);
 278                 table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
 279                 inet_peer_base_init(&table->tb6_peers);
 280         }
 281
 282         return table;
 283 }
 284
 285 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 286 {
 287         struct fib6_table *tb;
 288
 289         if (id == 0)
 290                 id = RT6_TABLE_MAIN;
 291         tb = fib6_get_table(net, id);
 292         if (tb)
 293                 return tb;
 294
 295         tb = fib6_alloc_table(net, id);
 296         if (tb)
 297                 fib6_link_table(net, tb);
 298
 299         return tb;
 300 }
 301 EXPORT_SYMBOL_GPL(fib6_new_table);
 302
 303 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 304 {
 305         struct fib6_table *tb;
 306         struct hlist_head *head;
 307         unsigned int h;
 308
 309         if (id == 0)
 310                 id = RT6_TABLE_MAIN;
 311         h = id & (FIB6_TABLE_HASHSZ - 1);
 312         rcu_read_lock();
 313         head = &net->ipv6.fib_table_hash[h];
 314         hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 315                 if (tb->tb6_id == id) {
 316                         rcu_read_unlock();
 317                         return tb;
 318                 }
 319         }
 320         rcu_read_unlock();
 321
 322         return NULL;
 323 }
 324 EXPORT_SYMBOL_GPL(fib6_get_table);
 325
 326 static void __net_init fib6_tables_init(struct net *net)
 327 {
 328         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 329         fib6_link_table(net, net->ipv6.fib6_local_tbl);
 330 }
 331 #else
 332
 333 struct fib6_table *fib6_new_table(struct net *net, u32 id)
 334 {
 335         return fib6_get_table(net, id);
 336 }
 337
 338 struct fib6_table *fib6_get_table(struct net *net, u32 id)
 339 {
 340           return net->ipv6.fib6_main_tbl;
 341 }
 342
 343 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 344                                    const struct sk_buff *skb,
 345                                    int flags, pol_lookup_t lookup)
 346 {
 347         struct rt6_info *rt;
 348
 349         rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, skb, flags);
 350         if (rt->dst.error == -EAGAIN) {
 351                 ip6_rt_put(rt);
 352                 rt = net->ipv6.ip6_null_entry;
 353                 dst_hold(&rt->dst);
 354         }
 355
 356         return &rt->dst;
 357 }
 358
 359 /* called with rcu lock held; no reference taken on fib6_info */
 360 struct fib6_info *fib6_lookup(struct net *net, int oif, struct flowi6 *fl6,
 361                               int flags)
 362 {
 363         return fib6_table_lookup(net, net->ipv6.fib6_main_tbl, oif, fl6, flags);
 364 }
 365
 366 static void __net_init fib6_tables_init(struct net *net)
 367 {
 368         fib6_link_table(net, net->ipv6.fib6_main_tbl);
 369 }
 370
 371 #endif
 372
 373 unsigned int fib6_tables_seq_read(struct net *net)
 374 {
 375         unsigned int h, fib_seq = 0;
 376
 377         rcu_read_lock();
 378         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 379                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 380                 struct fib6_table *tb;
 381
 382                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 383                         fib_seq += tb->fib_seq;
 384         }
 385         rcu_read_unlock();
 386
 387         return fib_seq;
 388 }
 389
 390 static int call_fib6_entry_notifier(struct notifier_block *nb, struct net *net,
 391                                     enum fib_event_type event_type,
 392                                     struct fib6_info *rt)
 393 {
 394         struct fib6_entry_notifier_info info = {
 395                 .rt = rt,
 396         };
 397
 398         return call_fib6_notifier(nb, net, event_type, &info.info);
 399 }
 400
 401 static int call_fib6_entry_notifiers(struct net *net,
 402                                      enum fib_event_type event_type,
 403                                      struct fib6_info *rt,
 404                                      struct netlink_ext_ack *extack)
 405 {
 406         struct fib6_entry_notifier_info info = {
 407                 .info.extack = extack,
 408                 .rt = rt,
 409         };
 410
 411         rt->fib6_table->fib_seq++;
 412         return call_fib6_notifiers(net, event_type, &info.info);
 413 }
 414
 415 struct fib6_dump_arg {
 416         struct net *net;
 417         struct notifier_block *nb;
 418 };
 419
 420 static void fib6_rt_dump(struct fib6_info *rt, struct fib6_dump_arg *arg)
 421 {
 422         if (rt == arg->net->ipv6.fib6_null_entry)
 423                 return;
 424         call_fib6_entry_notifier(arg->nb, arg->net, FIB_EVENT_ENTRY_ADD, rt);
 425 }
 426
 427 static int fib6_node_dump(struct fib6_walker *w)
 428 {
 429         struct fib6_info *rt;
 430
 431         for_each_fib6_walker_rt(w)
 432                 fib6_rt_dump(rt, w->args);
 433         w->leaf = NULL;
 434         return 0;
 435 }
 436
 437 static void fib6_table_dump(struct net *net, struct fib6_table *tb,
 438                             struct fib6_walker *w)
 439 {
 440         w->root = &tb->tb6_root;
 441         spin_lock_bh(&tb->tb6_lock);
 442         fib6_walk(net, w);
 443         spin_unlock_bh(&tb->tb6_lock);
 444 }
 445
 446 /* Called with rcu_read_lock() */
 447 int fib6_tables_dump(struct net *net, struct notifier_block *nb)
 448 {
 449         struct fib6_dump_arg arg;
 450         struct fib6_walker *w;
 451         unsigned int h;
 452
 453         w = kzalloc(sizeof(*w), GFP_ATOMIC);
 454         if (!w)
 455                 return -ENOMEM;
 456
 457         w->func = fib6_node_dump;
 458         arg.net = net;
 459         arg.nb = nb;
 460         w->args = &arg;
 461
 462         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
 463                 struct hlist_head *head = &net->ipv6.fib_table_hash[h];
 464                 struct fib6_table *tb;
 465
 466                 hlist_for_each_entry_rcu(tb, head, tb6_hlist)
 467                         fib6_table_dump(net, tb, w);
 468         }
 469
 470         kfree(w);
 471
 472         return 0;
 473 }
 474
 475 static int fib6_dump_node(struct fib6_walker *w)
 476 {
 477         int res;
 478         struct fib6_info *rt;
 479
 480         for_each_fib6_walker_rt(w) {
 481                 res = rt6_dump_route(rt, w->args);
 482                 if (res < 0) {
 483                         /* Frame is full, suspend walking */
 484                         w->leaf = rt;
 485                         return 1;
 486                 }
 487
 488                 /* Multipath routes are dumped in one route with the
 489                  * RTA_MULTIPATH attribute. Jump 'rt' to point to the
 490                  * last sibling of this route (no need to dump the
 491                  * sibling routes again)
 492                  */
 493                 if (rt->fib6_nsiblings)
 494                         rt = list_last_entry(&rt->fib6_siblings,
 495                                              struct fib6_info,
 496                                              fib6_siblings);
 497         }
 498         w->leaf = NULL;
 499         return 0;
 500 }
 501
 502 static void fib6_dump_end(struct netlink_callback *cb)
 503 {
 504         struct net *net = sock_net(cb->skb->sk);
 505         struct fib6_walker *w = (void *)cb->args[2];
 506
 507         if (w) {
 508                 if (cb->args[4]) {
 509                         cb->args[4] = 0;
 510                         fib6_walker_unlink(net, w);
 511                 }
 512                 cb->args[2] = 0;
 513                 kfree(w);
 514         }
 515         cb->done = (void *)cb->args[3];
 516         cb->args[1] = 3;
 517 }
 518
 519 static int fib6_dump_done(struct netlink_callback *cb)
 520 {
 521         fib6_dump_end(cb);
 522         return cb->done ? cb->done(cb) : 0;
 523 }
 524
 525 static int fib6_dump_table(struct fib6_table *table, struct sk_buff *skb,
 526                            struct netlink_callback *cb)
 527 {
 528         struct net *net = sock_net(skb->sk);
 529         struct fib6_walker *w;
 530         int res;
 531
 532         w = (void *)cb->args[2];
 533         w->root = &table->tb6_root;
 534
 535         if (cb->args[4] == 0) {
 536                 w->count = 0;
 537                 w->skip = 0;
 538
 539                 spin_lock_bh(&table->tb6_lock);
 540                 res = fib6_walk(net, w);
 541                 spin_unlock_bh(&table->tb6_lock);
 542                 if (res > 0) {
 543                         cb->args[4] = 1;
 544                         cb->args[5] = w->root->fn_sernum;
 545                 }
 546         } else {
 547                 if (cb->args[5] != w->root->fn_sernum) {
 548                         /* Begin at the root if the tree changed */
 549                         cb->args[5] = w->root->fn_sernum;
 550                         w->state = FWS_INIT;
 551                         w->node = w->root;
 552                         w->skip = w->count;
 553                 } else
 554                         w->skip = 0;
 555
 556                 spin_lock_bh(&table->tb6_lock);
 557                 res = fib6_walk_continue(w);
 558                 spin_unlock_bh(&table->tb6_lock);
 559                 if (res <= 0) {
 560                         fib6_walker_unlink(net, w);
 561                         cb->args[4] = 0;
 562                 }
 563         }
 564
 565         return res;
 566 }
 567
 568 static int inet6_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 569 {
 570         const struct nlmsghdr *nlh = cb->nlh;
 571         struct net *net = sock_net(skb->sk);
 572         struct rt6_rtnl_dump_arg arg = {};
 573         unsigned int h, s_h;
 574         unsigned int e = 0, s_e;
 575         struct fib6_walker *w;
 576         struct fib6_table *tb;
 577         struct hlist_head *head;
 578         int res = 0;
 579
 580         if (cb->strict_check) {
 581                 int err;
 582
 583                 err = ip_valid_fib_dump_req(net, nlh, &arg.filter, cb);
 584                 if (err < 0)
 585                         return err;
 586         } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
 587                 struct rtmsg *rtm = nlmsg_data(nlh);
 588
 589                 arg.filter.flags = rtm->rtm_flags & (RTM_F_PREFIX|RTM_F_CLONED);
 590         }
 591
 592         /* fib entries are never clones */
 593         if (arg.filter.flags & RTM_F_CLONED)
 594                 goto out;
 595
 596         w = (void *)cb->args[2];
 597         if (!w) {
 598                 /* New dump:
 599                  *
 600                  * 1. hook callback destructor.
 601                  */
 602                 cb->args[3] = (long)cb->done;
 603                 cb->done = fib6_dump_done;
 604
 605                 /*
 606                  * 2. allocate and initialize walker.
 607                  */
 608                 w = kzalloc(sizeof(*w), GFP_ATOMIC);
 609                 if (!w)
 610                         return -ENOMEM;
 611                 w->func = fib6_dump_node;
 612                 cb->args[2] = (long)w;
 613         }
 614
 615         arg.skb = skb;
 616         arg.cb = cb;
 617         arg.net = net;
 618         w->args = &arg;
 619
 620         if (arg.filter.table_id) {
 621                 tb = fib6_get_table(net, arg.filter.table_id);
 622                 if (!tb) {
 623                         if (arg.filter.dump_all_families)
 624                                 goto out;
 625
 626                         NL_SET_ERR_MSG_MOD(cb->extack, "FIB table does not exist");
 627                         return -ENOENT;
 628                 }
 629
 630                 if (!cb->args[0]) {
 631                         res = fib6_dump_table(tb, skb, cb);
 632                         if (!res)
 633                                 cb->args[0] = 1;
 634                 }
 635                 goto out;
 636         }
 637
 638         s_h = cb->args[0];
 639         s_e = cb->args[1];
 640
 641         rcu_read_lock();
 642         for (h = s_h; h < FIB6_TABLE_HASHSZ; h++, s_e = 0) {
 643                 e = 0;
 644                 head = &net->ipv6.fib_table_hash[h];
 645                 hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
 646                         if (e < s_e)
 647                                 goto next;
 648                         res = fib6_dump_table(tb, skb, cb);
 649                         if (res != 0)
 650                                 goto out_unlock;
 651 next:
 652                         e++;
 653                 }
 654         }
 655 out_unlock:
 656         rcu_read_unlock();
 657         cb->args[1] = e;
 658         cb->args[0] = h;
 659 out:
 660         res = res < 0 ? res : skb->len;
 661         if (res <= 0)
 662                 fib6_dump_end(cb);
 663         return res;
 664 }
 665
 666 void fib6_metric_set(struct fib6_info *f6i, int metric, u32 val)
 667 {
 668         if (!f6i)
 669                 return;
 670
 671         if (f6i->fib6_metrics == &dst_default_metrics) {
 672                 struct dst_metrics *p = kzalloc(sizeof(*p), GFP_ATOMIC);
 673
 674                 if (!p)
 675                         return;
 676
 677                 refcount_set(&p->refcnt, 1);
 678                 f6i->fib6_metrics = p;
 679         }
 680
 681         f6i->fib6_metrics->metrics[metric - 1] = val;
 682 }
 683
 684 /*
 685  *      Routing Table
 686  *
 687  *      return the appropriate node for a routing tree "add" operation
 688  *      by either creating and inserting or by returning an existing
 689  *      node.
 690  */
 691
 692 static struct fib6_node *fib6_add_1(struct net *net,
 693                                     struct fib6_table *table,
 694                                     struct fib6_node *root,
 695                                     struct in6_addr *addr, int plen,
 696                                     int offset, int allow_create,
 697                                     int replace_required,
 698                                     struct netlink_ext_ack *extack)
 699 {
 700         struct fib6_node *fn, *in, *ln;
 701         struct fib6_node *pn = NULL;
 702         struct rt6key *key;
 703         int     bit;
 704         __be32  dir = 0;
 705
 706         RT6_TRACE("fib6_add_1\n");
 707
 708         /* insert node in tree */
 709
 710         fn = root;
 711
 712         do {
 713                 struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
 714                                             lockdep_is_held(&table->tb6_lock));
 715                 key = (struct rt6key *)((u8 *)leaf + offset);
 716
 717                 /*
 718                  *      Prefix match
 719                  */
 720                 if (plen < fn->fn_bit ||
 721                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit)) {
 722                         if (!allow_create) {
 723                                 if (replace_required) {
 724                                         NL_SET_ERR_MSG(extack,
 725                                                        "Can not replace route - no match found");
 726                                         pr_warn("Can't replace route, no match found\n");
 727                                         return ERR_PTR(-ENOENT);
 728                                 }
 729                                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 730                         }
 731                         goto insert_above;
 732                 }
 733
 734                 /*
 735                  *      Exact match ?
 736                  */
 737
 738                 if (plen == fn->fn_bit) {
 739                         /* clean up an intermediate node */
 740                         if (!(fn->fn_flags & RTN_RTINFO)) {
 741                                 RCU_INIT_POINTER(fn->leaf, NULL);
 742                                 fib6_info_release(leaf);
 743                         /* remove null_entry in the root node */
 744                         } else if (fn->fn_flags & RTN_TL_ROOT &&
 745                                    rcu_access_pointer(fn->leaf) ==
 746                                    net->ipv6.fib6_null_entry) {
 747                                 RCU_INIT_POINTER(fn->leaf, NULL);
 748                         }
 749
 750                         return fn;
 751                 }
 752
 753                 /*
 754                  *      We have more bits to go
 755                  */
 756
 757                 /* Try to walk down on tree. */
 758                 dir = addr_bit_set(addr, fn->fn_bit);
 759                 pn = fn;
 760                 fn = dir ?
 761                      rcu_dereference_protected(fn->right,
 762                                         lockdep_is_held(&table->tb6_lock)) :
 763                      rcu_dereference_protected(fn->left,
 764                                         lockdep_is_held(&table->tb6_lock));
 765         } while (fn);
 766
 767         if (!allow_create) {
 768                 /* We should not create new node because
 769                  * NLM_F_REPLACE was specified without NLM_F_CREATE
 770                  * I assume it is safe to require NLM_F_CREATE when
 771                  * REPLACE flag is used! Later we may want to remove the
 772                  * check for replace_required, because according
 773                  * to netlink specification, NLM_F_CREATE
 774                  * MUST be specified if new route is created.
 775                  * That would keep IPv6 consistent with IPv4
 776                  */
 777                 if (replace_required) {
 778                         NL_SET_ERR_MSG(extack,
 779                                        "Can not replace route - no match found");
 780                         pr_warn("Can't replace route, no match found\n");
 781                         return ERR_PTR(-ENOENT);
 782                 }
 783                 pr_warn("NLM_F_CREATE should be set when creating new route\n");
 784         }
 785         /*
 786          *      We walked to the bottom of tree.
 787          *      Create new leaf node without children.
 788          */
 789
 790         ln = node_alloc(net);
 791
 792         if (!ln)
 793                 return ERR_PTR(-ENOMEM);
 794         ln->fn_bit = plen;
 795         RCU_INIT_POINTER(ln->parent, pn);
 796
 797         if (dir)
 798                 rcu_assign_pointer(pn->right, ln);
 799         else
 800                 rcu_assign_pointer(pn->left, ln);
 801
 802         return ln;
 803
 804
 805 insert_above:
 806         /*
 807          * split since we don't have a common prefix anymore or
 808          * we have a less significant route.
 809          * we've to insert an intermediate node on the list
 810          * this new node will point to the one we need to create
 811          * and the current
 812          */
 813
 814         pn = rcu_dereference_protected(fn->parent,
 815                                        lockdep_is_held(&table->tb6_lock));
 816
 817         /* find 1st bit in difference between the 2 addrs.
 818
 819            See comment in __ipv6_addr_diff: bit may be an invalid value,
 820            but if it is >= plen, the value is ignored in any case.
 821          */
 822
 823         bit = __ipv6_addr_diff(addr, &key->addr, sizeof(*addr));
 824
 825         /*
 826          *              (intermediate)[in]
 827          *                /        \
 828          *      (new leaf node)[ln] (old node)[fn]
 829          */
 830         if (plen > bit) {
 831                 in = node_alloc(net);
 832                 ln = node_alloc(net);
 833
 834                 if (!in || !ln) {
 835                         if (in)
 836                                 node_free_immediate(net, in);
 837                         if (ln)
 838                                 node_free_immediate(net, ln);
 839                         return ERR_PTR(-ENOMEM);
 840                 }
 841
 842                 /*
 843                  * new intermediate node.
 844                  * RTN_RTINFO will
 845                  * be off since that an address that chooses one of
 846                  * the branches would not match less specific routes
 847                  * in the other branch
 848                  */
 849
 850                 in->fn_bit = bit;
 851
 852                 RCU_INIT_POINTER(in->parent, pn);
 853                 in->leaf = fn->leaf;
 854                 atomic_inc(&rcu_dereference_protected(in->leaf,
 855                                 lockdep_is_held(&table->tb6_lock))->fib6_ref);
 856
 857                 /* update parent pointer */
 858                 if (dir)
 859                         rcu_assign_pointer(pn->right, in);
 860                 else
 861                         rcu_assign_pointer(pn->left, in);
 862
 863                 ln->fn_bit = plen;
 864
 865                 RCU_INIT_POINTER(ln->parent, in);
 866                 rcu_assign_pointer(fn->parent, in);
 867
 868                 if (addr_bit_set(addr, bit)) {
 869                         rcu_assign_pointer(in->right, ln);
 870                         rcu_assign_pointer(in->left, fn);
 871                 } else {
 872                         rcu_assign_pointer(in->left, ln);
 873                         rcu_assign_pointer(in->right, fn);
 874                 }
 875         } else { /* plen <= bit */
 876
 877                 /*
 878                  *              (new leaf node)[ln]
 879                  *                /        \
 880                  *           (old node)[fn] NULL
 881                  */
 882
 883                 ln = node_alloc(net);
 884
 885                 if (!ln)
 886                         return ERR_PTR(-ENOMEM);
 887
 888                 ln->fn_bit = plen;
 889
 890                 RCU_INIT_POINTER(ln->parent, pn);
 891
 892                 if (addr_bit_set(&key->addr, plen))
 893                         RCU_INIT_POINTER(ln->right, fn);
 894                 else
 895                         RCU_INIT_POINTER(ln->left, fn);
 896
 897                 rcu_assign_pointer(fn->parent, ln);
 898
 899                 if (dir)
 900                         rcu_assign_pointer(pn->right, ln);
 901                 else
 902                         rcu_assign_pointer(pn->left, ln);
 903         }
 904         return ln;
 905 }
 906
 907 static void fib6_drop_pcpu_from(struct fib6_info *f6i,
 908                                 const struct fib6_table *table)
 909 {
 910         int cpu;
 911
 912         /* release the reference to this fib entry from
 913          * all of its cached pcpu routes
 914          */
 915         for_each_possible_cpu(cpu) {
 916                 struct rt6_info **ppcpu_rt;
 917                 struct rt6_info *pcpu_rt;
 918
 919                 ppcpu_rt = per_cpu_ptr(f6i->rt6i_pcpu, cpu);
 920                 pcpu_rt = *ppcpu_rt;
 921                 if (pcpu_rt) {
 922                         struct fib6_info *from;
 923
 924                         from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
 925                         fib6_info_release(from);
 926                 }
 927         }
 928 }
 929
 930 static void fib6_purge_rt(struct fib6_info *rt, struct fib6_node *fn,
 931                           struct net *net)
 932 {
 933         struct fib6_table *table = rt->fib6_table;
 934
 935         if (atomic_read(&rt->fib6_ref) != 1) {
 936                 /* This route is used as dummy address holder in some split
 937                  * nodes. It is not leaked, but it still holds other resources,
 938                  * which must be released in time. So, scan ascendant nodes
 939                  * and replace dummy references to this route with references
 940                  * to still alive ones.
 941                  */
 942                 while (fn) {
 943                         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
 944                                             lockdep_is_held(&table->tb6_lock));
 945                         struct fib6_info *new_leaf;
 946                         if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
 947                                 new_leaf = fib6_find_prefix(net, table, fn);
 948                                 atomic_inc(&new_leaf->fib6_ref);
 949
 950                                 rcu_assign_pointer(fn->leaf, new_leaf);
 951                                 fib6_info_release(rt);
 952                         }
 953                         fn = rcu_dereference_protected(fn->parent,
 954                                     lockdep_is_held(&table->tb6_lock));
 955                 }
 956
 957                 if (rt->rt6i_pcpu)
 958                         fib6_drop_pcpu_from(rt, table);
 959         }
 960 }
 961
 962 /*
 963  *      Insert routing information in a node.
 964  */
 965
 966 static int fib6_add_rt2node(struct fib6_node *fn, struct fib6_info *rt,
 967                             struct nl_info *info,
 968                             struct netlink_ext_ack *extack)
 969 {
 970         struct fib6_info *leaf = rcu_dereference_protected(fn->leaf,
 971                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
 972         struct fib6_info *iter = NULL;
 973         struct fib6_info __rcu **ins;
 974         struct fib6_info __rcu **fallback_ins = NULL;
 975         int replace = (info->nlh &&
 976                        (info->nlh->nlmsg_flags & NLM_F_REPLACE));
 977         int add = (!info->nlh ||
 978                    (info->nlh->nlmsg_flags & NLM_F_CREATE));
 979         int found = 0;
 980         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
 981         u16 nlflags = NLM_F_EXCL;
 982         int err;
 983
 984         if (info->nlh && (info->nlh->nlmsg_flags & NLM_F_APPEND))
 985                 nlflags |= NLM_F_APPEND;
 986
 987         ins = &fn->leaf;
 988
 989         for (iter = leaf; iter;
 990              iter = rcu_dereference_protected(iter->fib6_next,
 991                                 lockdep_is_held(&rt->fib6_table->tb6_lock))) {
 992                 /*
 993                  *      Search for duplicates
 994                  */
 995
 996                 if (iter->fib6_metric == rt->fib6_metric) {
 997                         /*
 998                          *      Same priority level
 999                          */
1000                         if (info->nlh &&
1001                             (info->nlh->nlmsg_flags & NLM_F_EXCL))
1002                                 return -EEXIST;
1003
1004                         nlflags &= ~NLM_F_EXCL;
1005                         if (replace) {
1006                                 if (rt_can_ecmp == rt6_qualify_for_ecmp(iter)) {
1007                                         found++;
1008                                         break;
1009                                 }
1010                                 if (rt_can_ecmp)
1011                                         fallback_ins = fallback_ins ?: ins;
1012                                 goto next_iter;
1013                         }
1014
1015                         if (rt6_duplicate_nexthop(iter, rt)) {
1016                                 if (rt->fib6_nsiblings)
1017                                         rt->fib6_nsiblings = 0;
1018                                 if (!(iter->fib6_flags & RTF_EXPIRES))
1019                                         return -EEXIST;
1020                                 if (!(rt->fib6_flags & RTF_EXPIRES))
1021                                         fib6_clean_expires(iter);
1022                                 else
1023                                         fib6_set_expires(iter, rt->expires);
1024
1025                                 if (rt->fib6_pmtu)
1026                                         fib6_metric_set(iter, RTAX_MTU,
1027                                                         rt->fib6_pmtu);
1028                                 return -EEXIST;
1029                         }
1030                         /* If we have the same destination and the same metric,
1031                          * but not the same gateway, then the route we try to
1032                          * add is sibling to this route, increment our counter
1033                          * of siblings, and later we will add our route to the
1034                          * list.
1035                          * Only static routes (which don't have flag
1036                          * RTF_EXPIRES) are used for ECMPv6.
1037                          *
1038                          * To avoid long list, we only had siblings if the
1039                          * route have a gateway.
1040                          */
1041                         if (rt_can_ecmp &&
1042                             rt6_qualify_for_ecmp(iter))
1043                                 rt->fib6_nsiblings++;
1044                 }
1045
1046                 if (iter->fib6_metric > rt->fib6_metric)
1047                         break;
1048
1049 next_iter:
1050                 ins = &iter->fib6_next;
1051         }
1052
1053         if (fallback_ins && !found) {
1054                 /* No ECMP-able route found, replace first non-ECMP one */
1055                 ins = fallback_ins;
1056                 iter = rcu_dereference_protected(*ins,
1057                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1058                 found++;
1059         }
1060
1061         /* Reset round-robin state, if necessary */
1062         if (ins == &fn->leaf)
1063                 fn->rr_ptr = NULL;
1064
1065         /* Link this route to others same route. */
1066         if (rt->fib6_nsiblings) {
1067                 unsigned int fib6_nsiblings;
1068                 struct fib6_info *sibling, *temp_sibling;
1069
1070                 /* Find the first route that have the same metric */
1071                 sibling = leaf;
1072                 while (sibling) {
1073                         if (sibling->fib6_metric == rt->fib6_metric &&
1074                             rt6_qualify_for_ecmp(sibling)) {
1075                                 list_add_tail(&rt->fib6_siblings,
1076                                               &sibling->fib6_siblings);
1077                                 break;
1078                         }
1079                         sibling = rcu_dereference_protected(sibling->fib6_next,
1080                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1081                 }
1082                 /* For each sibling in the list, increment the counter of
1083                  * siblings. BUG() if counters does not match, list of siblings
1084                  * is broken!
1085                  */
1086                 fib6_nsiblings = 0;
1087                 list_for_each_entry_safe(sibling, temp_sibling,
1088                                          &rt->fib6_siblings, fib6_siblings) {
1089                         sibling->fib6_nsiblings++;
1090                         BUG_ON(sibling->fib6_nsiblings != rt->fib6_nsiblings);
1091                         fib6_nsiblings++;
1092                 }
1093                 BUG_ON(fib6_nsiblings != rt->fib6_nsiblings);
1094                 rt6_multipath_rebalance(temp_sibling);
1095         }
1096
1097         /*
1098          *      insert node
1099          */
1100         if (!replace) {
1101                 if (!add)
1102                         pr_warn("NLM_F_CREATE should be set when creating new route\n");
1103
1104 add:
1105                 nlflags |= NLM_F_CREATE;
1106
1107                 err = call_fib6_entry_notifiers(info->nl_net,
1108                                                 FIB_EVENT_ENTRY_ADD,
1109                                                 rt, extack);
1110                 if (err)
1111                         return err;
1112
1113                 rcu_assign_pointer(rt->fib6_next, iter);
1114                 atomic_inc(&rt->fib6_ref);
1115                 rcu_assign_pointer(rt->fib6_node, fn);
1116                 rcu_assign_pointer(*ins, rt);
1117                 if (!info->skip_notify)
1118                         inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
1119                 info->nl_net->ipv6.rt6_stats->fib_rt_entries++;
1120
1121                 if (!(fn->fn_flags & RTN_RTINFO)) {
1122                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1123                         fn->fn_flags |= RTN_RTINFO;
1124                 }
1125
1126         } else {
1127                 int nsiblings;
1128
1129                 if (!found) {
1130                         if (add)
1131                                 goto add;
1132                         pr_warn("NLM_F_REPLACE set, but no existing node found!\n");
1133                         return -ENOENT;
1134                 }
1135
1136                 err = call_fib6_entry_notifiers(info->nl_net,
1137                                                 FIB_EVENT_ENTRY_REPLACE,
1138                                                 rt, extack);
1139                 if (err)
1140                         return err;
1141
1142                 atomic_inc(&rt->fib6_ref);
1143                 rcu_assign_pointer(rt->fib6_node, fn);
1144                 rt->fib6_next = iter->fib6_next;
1145                 rcu_assign_pointer(*ins, rt);
1146                 if (!info->skip_notify)
1147                         inet6_rt_notify(RTM_NEWROUTE, rt, info, NLM_F_REPLACE);
1148                 if (!(fn->fn_flags & RTN_RTINFO)) {
1149                         info->nl_net->ipv6.rt6_stats->fib_route_nodes++;
1150                         fn->fn_flags |= RTN_RTINFO;
1151                 }
1152                 nsiblings = iter->fib6_nsiblings;
1153                 iter->fib6_node = NULL;
1154                 fib6_purge_rt(iter, fn, info->nl_net);
1155                 if (rcu_access_pointer(fn->rr_ptr) == iter)
1156                         fn->rr_ptr = NULL;
1157                 fib6_info_release(iter);
1158
1159                 if (nsiblings) {
1160                         /* Replacing an ECMP route, remove all siblings */
1161                         ins = &rt->fib6_next;
1162                         iter = rcu_dereference_protected(*ins,
1163                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1164                         while (iter) {
1165                                 if (iter->fib6_metric > rt->fib6_metric)
1166                                         break;
1167                                 if (rt6_qualify_for_ecmp(iter)) {
1168                                         *ins = iter->fib6_next;
1169                                         iter->fib6_node = NULL;
1170                                         fib6_purge_rt(iter, fn, info->nl_net);
1171                                         if (rcu_access_pointer(fn->rr_ptr) == iter)
1172                                                 fn->rr_ptr = NULL;
1173                                         fib6_info_release(iter);
1174                                         nsiblings--;
1175                                         info->nl_net->ipv6.rt6_stats->fib_rt_entries--;
1176                                 } else {
1177                                         ins = &iter->fib6_next;
1178                                 }
1179                                 iter = rcu_dereference_protected(*ins,
1180                                         lockdep_is_held(&rt->fib6_table->tb6_lock));
1181                         }
1182                         WARN_ON(nsiblings != 0);
1183                 }
1184         }
1185
1186         return 0;
1187 }
1188
1189 static void fib6_start_gc(struct net *net, struct fib6_info *rt)
1190 {
1191         if (!timer_pending(&net->ipv6.ip6_fib_timer) &&
1192             (rt->fib6_flags & RTF_EXPIRES))
1193                 mod_timer(&net->ipv6.ip6_fib_timer,
1194                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1195 }
1196
1197 void fib6_force_start_gc(struct net *net)
1198 {
1199         if (!timer_pending(&net->ipv6.ip6_fib_timer))
1200                 mod_timer(&net->ipv6.ip6_fib_timer,
1201                           jiffies + net->ipv6.sysctl.ip6_rt_gc_interval);
1202 }
1203
1204 static void __fib6_update_sernum_upto_root(struct fib6_info *rt,
1205                                            int sernum)
1206 {
1207         struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1208                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1209
1210         /* paired with smp_rmb() in rt6_get_cookie_safe() */
1211         smp_wmb();
1212         while (fn) {
1213                 fn->fn_sernum = sernum;
1214                 fn = rcu_dereference_protected(fn->parent,
1215                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
1216         }
1217 }
1218
1219 void fib6_update_sernum_upto_root(struct net *net, struct fib6_info *rt)
1220 {
1221         __fib6_update_sernum_upto_root(rt, fib6_new_sernum(net));
1222 }
1223
1224 /*
1225  *      Add routing information to the routing tree.
1226  *      <destination addr>/<source addr>
1227  *      with source addr info in sub-trees
1228  *      Need to own table->tb6_lock
1229  */
1230
1231 int fib6_add(struct fib6_node *root, struct fib6_info *rt,
1232              struct nl_info *info, struct netlink_ext_ack *extack)
1233 {
1234         struct fib6_table *table = rt->fib6_table;
1235         struct fib6_node *fn, *pn = NULL;
1236         int err = -ENOMEM;
1237         int allow_create = 1;
1238         int replace_required = 0;
1239         int sernum = fib6_new_sernum(info->nl_net);
1240
1241         if (info->nlh) {
1242                 if (!(info->nlh->nlmsg_flags & NLM_F_CREATE))
1243                         allow_create = 0;
1244                 if (info->nlh->nlmsg_flags & NLM_F_REPLACE)
1245                         replace_required = 1;
1246         }
1247         if (!allow_create && !replace_required)
1248                 pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
1249
1250         fn = fib6_add_1(info->nl_net, table, root,
1251                         &rt->fib6_dst.addr, rt->fib6_dst.plen,
1252                         offsetof(struct fib6_info, fib6_dst), allow_create,
1253                         replace_required, extack);
1254         if (IS_ERR(fn)) {
1255                 err = PTR_ERR(fn);
1256                 fn = NULL;
1257                 goto out;
1258         }
1259
1260         pn = fn;
1261
1262 #ifdef CONFIG_IPV6_SUBTREES
1263         if (rt->fib6_src.plen) {
1264                 struct fib6_node *sn;
1265
1266                 if (!rcu_access_pointer(fn->subtree)) {
1267                         struct fib6_node *sfn;
1268
1269                         /*
1270                          * Create subtree.
1271                          *
1272                          *              fn[main tree]
1273                          *              |
1274                          *              sfn[subtree root]
1275                          *                 \
1276                          *                  sn[new leaf node]
1277                          */
1278
1279                         /* Create subtree root node */
1280                         sfn = node_alloc(info->nl_net);
1281                         if (!sfn)
1282                                 goto failure;
1283
1284                         atomic_inc(&info->nl_net->ipv6.fib6_null_entry->fib6_ref);
1285                         rcu_assign_pointer(sfn->leaf,
1286                                            info->nl_net->ipv6.fib6_null_entry);
1287                         sfn->fn_flags = RTN_ROOT;
1288
1289                         /* Now add the first leaf node to new subtree */
1290
1291                         sn = fib6_add_1(info->nl_net, table, sfn,
1292                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1293                                         offsetof(struct fib6_info, fib6_src),
1294                                         allow_create, replace_required, extack);
1295
1296                         if (IS_ERR(sn)) {
1297                                 /* If it is failed, discard just allocated
1298                                    root, and then (in failure) stale node
1299                                    in main tree.
1300                                  */
1301                                 node_free_immediate(info->nl_net, sfn);
1302                                 err = PTR_ERR(sn);
1303                                 goto failure;
1304                         }
1305
1306                         /* Now link new subtree to main tree */
1307                         rcu_assign_pointer(sfn->parent, fn);
1308                         rcu_assign_pointer(fn->subtree, sfn);
1309                 } else {
1310                         sn = fib6_add_1(info->nl_net, table, FIB6_SUBTREE(fn),
1311                                         &rt->fib6_src.addr, rt->fib6_src.plen,
1312                                         offsetof(struct fib6_info, fib6_src),
1313                                         allow_create, replace_required, extack);
1314
1315                         if (IS_ERR(sn)) {
1316                                 err = PTR_ERR(sn);
1317                                 goto failure;
1318                         }
1319                 }
1320
1321                 if (!rcu_access_pointer(fn->leaf)) {
1322                         if (fn->fn_flags & RTN_TL_ROOT) {
1323                                 /* put back null_entry for root node */
1324                                 rcu_assign_pointer(fn->leaf,
1325                                             info->nl_net->ipv6.fib6_null_entry);
1326                         } else {
1327                                 atomic_inc(&rt->fib6_ref);
1328                                 rcu_assign_pointer(fn->leaf, rt);
1329                         }
1330                 }
1331                 fn = sn;
1332         }
1333 #endif
1334
1335         err = fib6_add_rt2node(fn, rt, info, extack);
1336         if (!err) {
1337                 __fib6_update_sernum_upto_root(rt, sernum);
1338                 fib6_start_gc(info->nl_net, rt);
1339         }
1340
1341 out:
1342         if (err) {
1343 #ifdef CONFIG_IPV6_SUBTREES
1344                 /*
1345                  * If fib6_add_1 has cleared the old leaf pointer in the
1346                  * super-tree leaf node we have to find a new one for it.
1347                  */
1348                 if (pn != fn) {
1349                         struct fib6_info *pn_leaf =
1350                                 rcu_dereference_protected(pn->leaf,
1351                                     lockdep_is_held(&table->tb6_lock));
1352                         if (pn_leaf == rt) {
1353                                 pn_leaf = NULL;
1354                                 RCU_INIT_POINTER(pn->leaf, NULL);
1355                                 fib6_info_release(rt);
1356                         }
1357                         if (!pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
1358                                 pn_leaf = fib6_find_prefix(info->nl_net, table,
1359                                                            pn);
1360 #if RT6_DEBUG >= 2
1361                                 if (!pn_leaf) {
1362                                         WARN_ON(!pn_leaf);
1363                                         pn_leaf =
1364                                             info->nl_net->ipv6.fib6_null_entry;
1365                                 }
1366 #endif
1367                                 fib6_info_hold(pn_leaf);
1368                                 rcu_assign_pointer(pn->leaf, pn_leaf);
1369                         }
1370                 }
1371 #endif
1372                 goto failure;
1373         }
1374         return err;
1375
1376 failure:
1377         /* fn->leaf could be NULL and fib6_repair_tree() needs to be called if:
1378          * 1. fn is an intermediate node and we failed to add the new
1379          * route to it in both subtree creation failure and fib6_add_rt2node()
1380          * failure case.
1381          * 2. fn is the root node in the table and we fail to add the first
1382          * default route to it.
1383          */
1384         if (fn &&
1385             (!(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)) ||
1386              (fn->fn_flags & RTN_TL_ROOT &&
1387               !rcu_access_pointer(fn->leaf))))
1388                 fib6_repair_tree(info->nl_net, table, fn);
1389         return err;
1390 }
1391
1392 /*
1393  *      Routing tree lookup
1394  *
1395  */
1396
1397 struct lookup_args {
1398         int                     offset;         /* key offset on fib6_info */
1399         const struct in6_addr   *addr;          /* search key                   */
1400 };
1401
1402 static struct fib6_node *fib6_node_lookup_1(struct fib6_node *root,
1403                                             struct lookup_args *args)
1404 {
1405         struct fib6_node *fn;
1406         __be32 dir;
1407
1408         if (unlikely(args->offset == 0))
1409                 return NULL;
1410
1411         /*
1412          *      Descend on a tree
1413          */
1414
1415         fn = root;
1416
1417         for (;;) {
1418                 struct fib6_node *next;
1419
1420                 dir = addr_bit_set(args->addr, fn->fn_bit);
1421
1422                 next = dir ? rcu_dereference(fn->right) :
1423                              rcu_dereference(fn->left);
1424
1425                 if (next) {
1426                         fn = next;
1427                         continue;
1428                 }
1429                 break;
1430         }
1431
1432         while (fn) {
1433                 struct fib6_node *subtree = FIB6_SUBTREE(fn);
1434
1435                 if (subtree || fn->fn_flags & RTN_RTINFO) {
1436                         struct fib6_info *leaf = rcu_dereference(fn->leaf);
1437                         struct rt6key *key;
1438
1439                         if (!leaf)
1440                                 goto backtrack;
1441
1442                         key = (struct rt6key *) ((u8 *)leaf + args->offset);
1443
1444                         if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
1445 #ifdef CONFIG_IPV6_SUBTREES
1446                                 if (subtree) {
1447                                         struct fib6_node *sfn;
1448                                         sfn = fib6_node_lookup_1(subtree,
1449                                                                  args + 1);
1450                                         if (!sfn)
1451                                                 goto backtrack;
1452                                         fn = sfn;
1453                                 }
1454 #endif
1455                                 if (fn->fn_flags & RTN_RTINFO)
1456                                         return fn;
1457                         }
1458                 }
1459 backtrack:
1460                 if (fn->fn_flags & RTN_ROOT)
1461                         break;
1462
1463                 fn = rcu_dereference(fn->parent);
1464         }
1465
1466         return NULL;
1467 }
1468
1469 /* called with rcu_read_lock() held
1470  */
1471 struct fib6_node *fib6_node_lookup(struct fib6_node *root,
1472                                    const struct in6_addr *daddr,
1473                                    const struct in6_addr *saddr)
1474 {
1475         struct fib6_node *fn;
1476         struct lookup_args args[] = {
1477                 {
1478                         .offset = offsetof(struct fib6_info, fib6_dst),
1479                         .addr = daddr,
1480                 },
1481 #ifdef CONFIG_IPV6_SUBTREES
1482                 {
1483                         .offset = offsetof(struct fib6_info, fib6_src),
1484                         .addr = saddr,
1485                 },
1486 #endif
1487                 {
1488                         .offset = 0,    /* sentinel */
1489                 }
1490         };
1491
1492         fn = fib6_node_lookup_1(root, daddr ? args : args + 1);
1493         if (!fn || fn->fn_flags & RTN_TL_ROOT)
1494                 fn = root;
1495
1496         return fn;
1497 }
1498
1499 /*
1500  *      Get node with specified destination prefix (and source prefix,
1501  *      if subtrees are used)
1502  *      exact_match == true means we try to find fn with exact match of
1503  *      the passed in prefix addr
1504  *      exact_match == false means we try to find fn with longest prefix
1505  *      match of the passed in prefix addr. This is useful for finding fn
1506  *      for cached route as it will be stored in the exception table under
1507  *      the node with longest prefix length.
1508  */
1509
1510
1511 static struct fib6_node *fib6_locate_1(struct fib6_node *root,
1512                                        const struct in6_addr *addr,
1513                                        int plen, int offset,
1514                                        bool exact_match)
1515 {
1516         struct fib6_node *fn, *prev = NULL;
1517
1518         for (fn = root; fn ; ) {
1519                 struct fib6_info *leaf = rcu_dereference(fn->leaf);
1520                 struct rt6key *key;
1521
1522                 /* This node is being deleted */
1523                 if (!leaf) {
1524                         if (plen <= fn->fn_bit)
1525                                 goto out;
1526                         else
1527                                 goto next;
1528                 }
1529
1530                 key = (struct rt6key *)((u8 *)leaf + offset);
1531
1532                 /*
1533                  *      Prefix match
1534                  */
1535                 if (plen < fn->fn_bit ||
1536                     !ipv6_prefix_equal(&key->addr, addr, fn->fn_bit))
1537                         goto out;
1538
1539                 if (plen == fn->fn_bit)
1540                         return fn;
1541
1542                 prev = fn;
1543
1544 next:
1545                 /*
1546                  *      We have more bits to go
1547                  */
1548                 if (addr_bit_set(addr, fn->fn_bit))
1549                         fn = rcu_dereference(fn->right);
1550                 else
1551                         fn = rcu_dereference(fn->left);
1552         }
1553 out:
1554         if (exact_match)
1555                 return NULL;
1556         else
1557                 return prev;
1558 }
1559
1560 struct fib6_node *fib6_locate(struct fib6_node *root,
1561                               const struct in6_addr *daddr, int dst_len,
1562                               const struct in6_addr *saddr, int src_len,
1563                               bool exact_match)
1564 {
1565         struct fib6_node *fn;
1566
1567         fn = fib6_locate_1(root, daddr, dst_len,
1568                            offsetof(struct fib6_info, fib6_dst),
1569                            exact_match);
1570
1571 #ifdef CONFIG_IPV6_SUBTREES
1572         if (src_len) {
1573                 WARN_ON(saddr == NULL);
1574                 if (fn) {
1575                         struct fib6_node *subtree = FIB6_SUBTREE(fn);
1576
1577                         if (subtree) {
1578                                 fn = fib6_locate_1(subtree, saddr, src_len,
1579                                            offsetof(struct fib6_info, fib6_src),
1580                                            exact_match);
1581                         }
1582                 }
1583         }
1584 #endif
1585
1586         if (fn && fn->fn_flags & RTN_RTINFO)
1587                 return fn;
1588
1589         return NULL;
1590 }
1591
1592
1593 /*
1594  *      Deletion
1595  *
1596  */
1597
1598 static struct fib6_info *fib6_find_prefix(struct net *net,
1599                                          struct fib6_table *table,
1600                                          struct fib6_node *fn)
1601 {
1602         struct fib6_node *child_left, *child_right;
1603
1604         if (fn->fn_flags & RTN_ROOT)
1605                 return net->ipv6.fib6_null_entry;
1606
1607         while (fn) {
1608                 child_left = rcu_dereference_protected(fn->left,
1609                                     lockdep_is_held(&table->tb6_lock));
1610                 child_right = rcu_dereference_protected(fn->right,
1611                                     lockdep_is_held(&table->tb6_lock));
1612                 if (child_left)
1613                         return rcu_dereference_protected(child_left->leaf,
1614                                         lockdep_is_held(&table->tb6_lock));
1615                 if (child_right)
1616                         return rcu_dereference_protected(child_right->leaf,
1617                                         lockdep_is_held(&table->tb6_lock));
1618
1619                 fn = FIB6_SUBTREE(fn);
1620         }
1621         return NULL;
1622 }
1623
1624 /*
1625  *      Called to trim the tree of intermediate nodes when possible. "fn"
1626  *      is the node we want to try and remove.
1627  *      Need to own table->tb6_lock
1628  */
1629
1630 static struct fib6_node *fib6_repair_tree(struct net *net,
1631                                           struct fib6_table *table,
1632                                           struct fib6_node *fn)
1633 {
1634         int children;
1635         int nstate;
1636         struct fib6_node *child;
1637         struct fib6_walker *w;
1638         int iter = 0;
1639
1640         /* Set fn->leaf to null_entry for root node. */
1641         if (fn->fn_flags & RTN_TL_ROOT) {
1642                 rcu_assign_pointer(fn->leaf, net->ipv6.fib6_null_entry);
1643                 return fn;
1644         }
1645
1646         for (;;) {
1647                 struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
1648                                             lockdep_is_held(&table->tb6_lock));
1649                 struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
1650                                             lockdep_is_held(&table->tb6_lock));
1651                 struct fib6_node *pn = rcu_dereference_protected(fn->parent,
1652                                             lockdep_is_held(&table->tb6_lock));
1653                 struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
1654                                             lockdep_is_held(&table->tb6_lock));
1655                 struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
1656                                             lockdep_is_held(&table->tb6_lock));
1657                 struct fib6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
1658                                             lockdep_is_held(&table->tb6_lock));
1659                 struct fib6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
1660                                             lockdep_is_held(&table->tb6_lock));
1661                 struct fib6_info *new_fn_leaf;
1662
1663                 RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
1664                 iter++;
1665
1666                 WARN_ON(fn->fn_flags & RTN_RTINFO);
1667                 WARN_ON(fn->fn_flags & RTN_TL_ROOT);
1668                 WARN_ON(fn_leaf);
1669
1670                 children = 0;
1671                 child = NULL;
1672                 if (fn_r)
1673                         child = fn_r, children |= 1;
1674                 if (fn_l)
1675                         child = fn_l, children |= 2;
1676
1677                 if (children == 3 || FIB6_SUBTREE(fn)
1678 #ifdef CONFIG_IPV6_SUBTREES
1679                     /* Subtree root (i.e. fn) may have one child */
1680                     || (children && fn->fn_flags & RTN_ROOT)
1681 #endif
1682                     ) {
1683                         new_fn_leaf = fib6_find_prefix(net, table, fn);
1684 #if RT6_DEBUG >= 2
1685                         if (!new_fn_leaf) {
1686                                 WARN_ON(!new_fn_leaf);
1687                                 new_fn_leaf = net->ipv6.fib6_null_entry;
1688                         }
1689 #endif
1690                         fib6_info_hold(new_fn_leaf);
1691                         rcu_assign_pointer(fn->leaf, new_fn_leaf);
1692                         return pn;
1693                 }
1694
1695 #ifdef CONFIG_IPV6_SUBTREES
1696                 if (FIB6_SUBTREE(pn) == fn) {
1697                         WARN_ON(!(fn->fn_flags & RTN_ROOT));
1698                         RCU_INIT_POINTER(pn->subtree, NULL);
1699                         nstate = FWS_L;
1700                 } else {
1701                         WARN_ON(fn->fn_flags & RTN_ROOT);
1702 #endif
1703                         if (pn_r == fn)
1704                                 rcu_assign_pointer(pn->right, child);
1705                         else if (pn_l == fn)
1706                                 rcu_assign_pointer(pn->left, child);
1707 #if RT6_DEBUG >= 2
1708                         else
1709                                 WARN_ON(1);
1710 #endif
1711                         if (child)
1712                                 rcu_assign_pointer(child->parent, pn);
1713                         nstate = FWS_R;
1714 #ifdef CONFIG_IPV6_SUBTREES
1715                 }
1716 #endif
1717
1718                 read_lock(&net->ipv6.fib6_walker_lock);
1719                 FOR_WALKERS(net, w) {
1720                         if (!child) {
1721                                 if (w->node == fn) {
1722                                         RT6_TRACE("W %p adjusted by delnode 1, s=%d/%d\n", w, w->state, nstate);
1723                                         w->node = pn;
1724                                         w->state = nstate;
1725                                 }
1726                         } else {
1727                                 if (w->node == fn) {
1728                                         w->node = child;
1729                                         if (children&2) {
1730                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1731                                                 w->state = w->state >= FWS_R ? FWS_U : FWS_INIT;
1732                                         } else {
1733                                                 RT6_TRACE("W %p adjusted by delnode 2, s=%d\n", w, w->state);
1734                                                 w->state = w->state >= FWS_C ? FWS_U : FWS_INIT;
1735                                         }
1736                                 }
1737                         }
1738                 }
1739                 read_unlock(&net->ipv6.fib6_walker_lock);
1740
1741                 node_free(net, fn);
1742                 if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
1743                         return pn;
1744
1745                 RCU_INIT_POINTER(pn->leaf, NULL);
1746                 fib6_info_release(pn_leaf);
1747                 fn = pn;
1748         }
1749 }
1750
1751 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
1752                            struct fib6_info __rcu **rtp, struct nl_info *info)
1753 {
1754         struct fib6_walker *w;
1755         struct fib6_info *rt = rcu_dereference_protected(*rtp,
1756                                     lockdep_is_held(&table->tb6_lock));
1757         struct net *net = info->nl_net;
1758
1759         RT6_TRACE("fib6_del_route\n");
1760
1761         /* Unlink it */
1762         *rtp = rt->fib6_next;
1763         rt->fib6_node = NULL;
1764         net->ipv6.rt6_stats->fib_rt_entries--;
1765         net->ipv6.rt6_stats->fib_discarded_routes++;
1766
1767         /* Flush all cached dst in exception table */
1768         rt6_flush_exceptions(rt);
1769
1770         /* Reset round-robin state, if necessary */
1771         if (rcu_access_pointer(fn->rr_ptr) == rt)
1772                 fn->rr_ptr = NULL;
1773
1774         /* Remove this entry from other siblings */
1775         if (rt->fib6_nsiblings) {
1776                 struct fib6_info *sibling, *next_sibling;
1777
1778                 list_for_each_entry_safe(sibling, next_sibling,
1779                                          &rt->fib6_siblings, fib6_siblings)
1780                         sibling->fib6_nsiblings--;
1781                 rt->fib6_nsiblings = 0;
1782                 list_del_init(&rt->fib6_siblings);
1783                 rt6_multipath_rebalance(next_sibling);
1784         }
1785
1786         /* Adjust walkers */
1787         read_lock(&net->ipv6.fib6_walker_lock);
1788         FOR_WALKERS(net, w) {
1789                 if (w->state == FWS_C && w->leaf == rt) {
1790                         RT6_TRACE("walker %p adjusted by delroute\n", w);
1791                         w->leaf = rcu_dereference_protected(rt->fib6_next,
1792                                             lockdep_is_held(&table->tb6_lock));
1793                         if (!w->leaf)
1794                                 w->state = FWS_U;
1795                 }
1796         }
1797         read_unlock(&net->ipv6.fib6_walker_lock);
1798
1799         /* If it was last route, call fib6_repair_tree() to:
1800          * 1. For root node, put back null_entry as how the table was created.
1801          * 2. For other nodes, expunge its radix tree node.
1802          */
1803         if (!rcu_access_pointer(fn->leaf)) {
1804                 if (!(fn->fn_flags & RTN_TL_ROOT)) {
1805                         fn->fn_flags &= ~RTN_RTINFO;
1806                         net->ipv6.rt6_stats->fib_route_nodes--;
1807                 }
1808                 fn = fib6_repair_tree(net, table, fn);
1809         }
1810
1811         fib6_purge_rt(rt, fn, net);
1812
1813         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
1814         if (!info->skip_notify)
1815                 inet6_rt_notify(RTM_DELROUTE, rt, info, 0);
1816         fib6_info_release(rt);
1817 }
1818
1819 /* Need to own table->tb6_lock */
1820 int fib6_del(struct fib6_info *rt, struct nl_info *info)
1821 {
1822         struct fib6_node *fn = rcu_dereference_protected(rt->fib6_node,
1823                                     lockdep_is_held(&rt->fib6_table->tb6_lock));
1824         struct fib6_table *table = rt->fib6_table;
1825         struct net *net = info->nl_net;
1826         struct fib6_info __rcu **rtp;
1827         struct fib6_info __rcu **rtp_next;
1828
1829         if (!fn || rt == net->ipv6.fib6_null_entry)
1830                 return -ENOENT;
1831
1832         WARN_ON(!(fn->fn_flags & RTN_RTINFO));
1833
1834         /*
1835          *      Walk the leaf entries looking for ourself
1836          */
1837
1838         for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
1839                 struct fib6_info *cur = rcu_dereference_protected(*rtp,
1840                                         lockdep_is_held(&table->tb6_lock));
1841                 if (rt == cur) {
1842                         fib6_del_route(table, fn, rtp, info);
1843                         return 0;
1844                 }
1845                 rtp_next = &cur->fib6_next;
1846         }
1847         return -ENOENT;
1848 }
1849
1850 /*
1851  *      Tree traversal function.
1852  *
1853  *      Certainly, it is not interrupt safe.
1854  *      However, it is internally reenterable wrt itself and fib6_add/fib6_del.
1855  *      It means, that we can modify tree during walking
1856  *      and use this function for garbage collection, clone pruning,
1857  *      cleaning tree when a device goes down etc. etc.
1858  *
1859  *      It guarantees that every node will be traversed,
1860  *      and that it will be traversed only once.
1861  *
1862  *      Callback function w->func may return:
1863  *      0 -> continue walking.
1864  *      positive value -> walking is suspended (used by tree dumps,
1865  *      and probably by gc, if it will be split to several slices)
1866  *      negative value -> terminate walking.
1867  *
1868  *      The function itself returns:
1869  *      0   -> walk is complete.
1870  *      >0  -> walk is incomplete (i.e. suspended)
1871  *      <0  -> walk is terminated by an error.
1872  *
1873  *      This function is called with tb6_lock held.
1874  */
1875
1876 static int fib6_walk_continue(struct fib6_walker *w)
1877 {
1878         struct fib6_node *fn, *pn, *left, *right;
1879
1880         /* w->root should always be table->tb6_root */
1881         WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
1882
1883         for (;;) {
1884                 fn = w->node;
1885                 if (!fn)
1886                         return 0;
1887
1888                 switch (w->state) {
1889 #ifdef CONFIG_IPV6_SUBTREES
1890                 case FWS_S:
1891                         if (FIB6_SUBTREE(fn)) {
1892                                 w->node = FIB6_SUBTREE(fn);
1893                                 continue;
1894                         }
1895                         w->state = FWS_L;
1896 #endif
1897                         /* fall through */
1898                 case FWS_L:
1899                         left = rcu_dereference_protected(fn->left, 1);
1900                         if (left) {
1901                                 w->node = left;
1902                                 w->state = FWS_INIT;
1903                                 continue;
1904                         }
1905                         w->state = FWS_R;
1906                         /* fall through */
1907                 case FWS_R:
1908                         right = rcu_dereference_protected(fn->right, 1);
1909                         if (right) {
1910                                 w->node = right;
1911                                 w->state = FWS_INIT;
1912                                 continue;
1913                         }
1914                         w->state = FWS_C;
1915                         w->leaf = rcu_dereference_protected(fn->leaf, 1);
1916                         /* fall through */
1917                 case FWS_C:
1918                         if (w->leaf && fn->fn_flags & RTN_RTINFO) {
1919                                 int err;
1920
1921                                 if (w->skip) {
1922                                         w->skip--;
1923                                         goto skip;
1924                                 }
1925
1926                                 err = w->func(w);
1927                                 if (err)
1928                                         return err;
1929
1930                                 w->count++;
1931                                 continue;
1932                         }
1933 skip:
1934                         w->state = FWS_U;
1935                         /* fall through */
1936                 case FWS_U:
1937                         if (fn == w->root)
1938                                 return 0;
1939                         pn = rcu_dereference_protected(fn->parent, 1);
1940                         left = rcu_dereference_protected(pn->left, 1);
1941                         right = rcu_dereference_protected(pn->right, 1);
1942                         w->node = pn;
1943 #ifdef CONFIG_IPV6_SUBTREES
1944                         if (FIB6_SUBTREE(pn) == fn) {
1945                                 WARN_ON(!(fn->fn_flags & RTN_ROOT));
1946                                 w->state = FWS_L;
1947                                 continue;
1948                         }
1949 #endif
1950                         if (left == fn) {
1951                                 w->state = FWS_R;
1952                                 continue;
1953                         }
1954                         if (right == fn) {
1955                                 w->state = FWS_C;
1956                                 w->leaf = rcu_dereference_protected(w->node->leaf, 1);
1957                                 continue;
1958                         }
1959 #if RT6_DEBUG >= 2
1960                         WARN_ON(1);
1961 #endif
1962                 }
1963         }
1964 }
1965
1966 static int fib6_walk(struct net *net, struct fib6_walker *w)
1967 {
1968         int res;
1969
1970         w->state = FWS_INIT;
1971         w->node = w->root;
1972
1973         fib6_walker_link(net, w);
1974         res = fib6_walk_continue(w);
1975         if (res <= 0)
1976                 fib6_walker_unlink(net, w);
1977         return res;
1978 }
1979
1980 static int fib6_clean_node(struct fib6_walker *w)
1981 {
1982         int res;
1983         struct fib6_info *rt;
1984         struct fib6_cleaner *c = container_of(w, struct fib6_cleaner, w);
1985         struct nl_info info = {
1986                 .nl_net = c->net,
1987                 .skip_notify = c->skip_notify,
1988         };
1989
1990         if (c->sernum != FIB6_NO_SERNUM_CHANGE &&
1991             w->node->fn_sernum != c->sernum)
1992                 w->node->fn_sernum = c->sernum;
1993
1994         if (!c->func) {
1995                 WARN_ON_ONCE(c->sernum == FIB6_NO_SERNUM_CHANGE);
1996                 w->leaf = NULL;
1997                 return 0;
1998         }
1999
2000         for_each_fib6_walker_rt(w) {
2001                 res = c->func(rt, c->arg);
2002                 if (res == -1) {
2003                         w->leaf = rt;
2004                         res = fib6_del(rt, &info);
2005                         if (res) {
2006 #if RT6_DEBUG >= 2
2007                                 pr_debug("%s: del failed: rt=%p@%p err=%d\n",
2008                                          __func__, rt,
2009                                          rcu_access_pointer(rt->fib6_node),
2010                                          res);
2011 #endif
2012                                 continue;
2013                         }
2014                         return 0;
2015                 } else if (res == -2) {
2016                         if (WARN_ON(!rt->fib6_nsiblings))
2017                                 continue;
2018                         rt = list_last_entry(&rt->fib6_siblings,
2019                                              struct fib6_info, fib6_siblings);
2020                         continue;
2021                 }
2022                 WARN_ON(res != 0);
2023         }
2024         w->leaf = rt;
2025         return 0;
2026 }
2027
2028 /*
2029  *      Convenient frontend to tree walker.
2030  *
2031  *      func is called on each route.
2032  *              It may return -2 -> skip multipath route.
2033  *                            -1 -> delete this route.
2034  *                            0  -> continue walking
2035  */
2036
2037 static void fib6_clean_tree(struct net *net, struct fib6_node *root,
2038                             int (*func)(struct fib6_info *, void *arg),
2039                             int sernum, void *arg, bool skip_notify)
2040 {
2041         struct fib6_cleaner c;
2042
2043         c.w.root = root;
2044         c.w.func = fib6_clean_node;
2045         c.w.count = 0;
2046         c.w.skip = 0;
2047         c.func = func;
2048         c.sernum = sernum;
2049         c.arg = arg;
2050         c.net = net;
2051         c.skip_notify = skip_notify;
2052
2053         fib6_walk(net, &c.w);
2054 }
2055
2056 static void __fib6_clean_all(struct net *net,
2057                              int (*func)(struct fib6_info *, void *),
2058                              int sernum, void *arg, bool skip_notify)
2059 {
2060         struct fib6_table *table;
2061         struct hlist_head *head;
2062         unsigned int h;
2063
2064         rcu_read_lock();
2065         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
2066                 head = &net->ipv6.fib_table_hash[h];
2067                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
2068                         spin_lock_bh(&table->tb6_lock);
2069                         fib6_clean_tree(net, &table->tb6_root,
2070                                         func, sernum, arg, skip_notify);
2071                         spin_unlock_bh(&table->tb6_lock);
2072                 }
2073         }
2074         rcu_read_unlock();
2075 }
2076
2077 void fib6_clean_all(struct net *net, int (*func)(struct fib6_info *, void *),
2078                     void *arg)
2079 {
2080         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, false);
2081 }
2082
2083 void fib6_clean_all_skip_notify(struct net *net,
2084                                 int (*func)(struct fib6_info *, void *),
2085                                 void *arg)
2086 {
2087         __fib6_clean_all(net, func, FIB6_NO_SERNUM_CHANGE, arg, true);
2088 }
2089
2090 static void fib6_flush_trees(struct net *net)
2091 {
2092         int new_sernum = fib6_new_sernum(net);
2093
2094         __fib6_clean_all(net, NULL, new_sernum, NULL, false);
2095 }
2096
2097 /*
2098  *      Garbage collection
2099  */
2100
2101 static int fib6_age(struct fib6_info *rt, void *arg)
2102 {
2103         struct fib6_gc_args *gc_args = arg;
2104         unsigned long now = jiffies;
2105
2106         /*
2107          *      check addrconf expiration here.
2108          *      Routes are expired even if they are in use.
2109          */
2110
2111         if (rt->fib6_flags & RTF_EXPIRES && rt->expires) {
2112                 if (time_after(now, rt->expires)) {
2113                         RT6_TRACE("expiring %p\n", rt);
2114                         return -1;
2115                 }
2116                 gc_args->more++;
2117         }
2118
2119         /*      Also age clones in the exception table.
2120          *      Note, that clones are aged out
2121          *      only if they are not in use now.
2122          */
2123         rt6_age_exceptions(rt, gc_args, now);
2124
2125         return 0;
2126 }
2127
2128 void fib6_run_gc(unsigned long expires, struct net *net, bool force)
2129 {
2130         struct fib6_gc_args gc_args;
2131         unsigned long now;
2132
2133         if (force) {
2134                 spin_lock_bh(&net->ipv6.fib6_gc_lock);
2135         } else if (!spin_trylock_bh(&net->ipv6.fib6_gc_lock)) {
2136                 mod_timer(&net->ipv6.ip6_fib_timer, jiffies + HZ);
2137                 return;
2138         }
2139         gc_args.timeout = expires ? (int)expires :
2140                           net->ipv6.sysctl.ip6_rt_gc_interval;
2141         gc_args.more = 0;
2142
2143         fib6_clean_all(net, fib6_age, &gc_args);
2144         now = jiffies;
2145         net->ipv6.ip6_rt_last_gc = now;
2146
2147         if (gc_args.more)
2148                 mod_timer(&net->ipv6.ip6_fib_timer,
2149                           round_jiffies(now
2150                                         + net->ipv6.sysctl.ip6_rt_gc_interval));
2151         else
2152                 del_timer(&net->ipv6.ip6_fib_timer);
2153         spin_unlock_bh(&net->ipv6.fib6_gc_lock);
2154 }
2155
2156 static void fib6_gc_timer_cb(struct timer_list *t)
2157 {
2158         struct net *arg = from_timer(arg, t, ipv6.ip6_fib_timer);
2159
2160         fib6_run_gc(0, arg, true);
2161 }
2162
2163 static int __net_init fib6_net_init(struct net *net)
2164 {
2165         size_t size = sizeof(struct hlist_head) * FIB6_TABLE_HASHSZ;
2166         int err;
2167
2168         err = fib6_notifier_init(net);
2169         if (err)
2170                 return err;
2171
2172         spin_lock_init(&net->ipv6.fib6_gc_lock);
2173         rwlock_init(&net->ipv6.fib6_walker_lock);
2174         INIT_LIST_HEAD(&net->ipv6.fib6_walkers);
2175         timer_setup(&net->ipv6.ip6_fib_timer, fib6_gc_timer_cb, 0);
2176
2177         net->ipv6.rt6_stats = kzalloc(sizeof(*net->ipv6.rt6_stats), GFP_KERNEL);
2178         if (!net->ipv6.rt6_stats)
2179                 goto out_timer;
2180
2181         /* Avoid false sharing : Use at least a full cache line */
2182         size = max_t(size_t, size, L1_CACHE_BYTES);
2183
2184         net->ipv6.fib_table_hash = kzalloc(size, GFP_KERNEL);
2185         if (!net->ipv6.fib_table_hash)
2186                 goto out_rt6_stats;
2187
2188         net->ipv6.fib6_main_tbl = kzalloc(sizeof(*net->ipv6.fib6_main_tbl),
2189                                           GFP_KERNEL);
2190         if (!net->ipv6.fib6_main_tbl)
2191                 goto out_fib_table_hash;
2192
2193         net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
2194         rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
2195                            net->ipv6.fib6_null_entry);
2196         net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
2197                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2198         inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
2199
2200 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2201         net->ipv6.fib6_local_tbl = kzalloc(sizeof(*net->ipv6.fib6_local_tbl),
2202                                            GFP_KERNEL);
2203         if (!net->ipv6.fib6_local_tbl)
2204                 goto out_fib6_main_tbl;
2205         net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
2206         rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
2207                            net->ipv6.fib6_null_entry);
2208         net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
2209                 RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
2210         inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
2211 #endif
2212         fib6_tables_init(net);
2213
2214         return 0;
2215
2216 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
2217 out_fib6_main_tbl:
2218         kfree(net->ipv6.fib6_main_tbl);
2219 #endif
2220 out_fib_table_hash:
2221         kfree(net->ipv6.fib_table_hash);
2222 out_rt6_stats:
2223         kfree(net->ipv6.rt6_stats);
2224 out_timer:
2225         fib6_notifier_exit(net);
2226         return -ENOMEM;
2227 }
2228
2229 static void fib6_net_exit(struct net *net)
2230 {
2231         unsigned int i;
2232
2233         del_timer_sync(&net->ipv6.ip6_fib_timer);
2234
2235         for (i = 0; i < FIB6_TABLE_HASHSZ; i++) {
2236                 struct hlist_head *head = &net->ipv6.fib_table_hash[i];
2237                 struct hlist_node *tmp;
2238                 struct fib6_table *tb;
2239
2240                 hlist_for_each_entry_safe(tb, tmp, head, tb6_hlist) {
2241                         hlist_del(&tb->tb6_hlist);
2242                         fib6_free_table(tb);
2243                 }
2244         }
2245
2246         kfree(net->ipv6.fib_table_hash);
2247         kfree(net->ipv6.rt6_stats);
2248         fib6_notifier_exit(net);
2249 }
2250
2251 static struct pernet_operations fib6_net_ops = {
2252         .init = fib6_net_init,
2253         .exit = fib6_net_exit,
2254 };
2255
2256 int __init fib6_init(void)
2257 {
2258         int ret = -ENOMEM;
2259
2260         fib6_node_kmem = kmem_cache_create("fib6_nodes",
2261                                            sizeof(struct fib6_node),
2262                                            0, SLAB_HWCACHE_ALIGN,
2263                                            NULL);
2264         if (!fib6_node_kmem)
2265                 goto out;
2266
2267         ret = register_pernet_subsys(&fib6_net_ops);
2268         if (ret)
2269                 goto out_kmem_cache_create;
2270
2271         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE, NULL,
2272                                    inet6_dump_fib, 0);
2273         if (ret)
2274                 goto out_unregister_subsys;
2275
2276         __fib6_flush_trees = fib6_flush_trees;
2277 out:
2278         return ret;
2279
2280 out_unregister_subsys:
2281         unregister_pernet_subsys(&fib6_net_ops);
2282 out_kmem_cache_create:
2283         kmem_cache_destroy(fib6_node_kmem);
2284         goto out;
2285 }
2286
2287 void fib6_gc_cleanup(void)
2288 {
2289         unregister_pernet_subsys(&fib6_net_ops);
2290         kmem_cache_destroy(fib6_node_kmem);
2291 }
2292
2293 #ifdef CONFIG_PROC_FS
2294 static int ipv6_route_seq_show(struct seq_file *seq, void *v)
2295 {
2296         struct fib6_info *rt = v;
2297         struct ipv6_route_iter *iter = seq->private;
2298         const struct net_device *dev;
2299
2300         seq_printf(seq, "%pi6 %02x ", &rt->fib6_dst.addr, rt->fib6_dst.plen);
2301
2302 #ifdef CONFIG_IPV6_SUBTREES
2303         seq_printf(seq, "%pi6 %02x ", &rt->fib6_src.addr, rt->fib6_src.plen);
2304 #else
2305         seq_puts(seq, "00000000000000000000000000000000 00 ");
2306 #endif
2307         if (rt->fib6_flags & RTF_GATEWAY)
2308                 seq_printf(seq, "%pi6", &rt->fib6_nh.nh_gw);
2309         else
2310                 seq_puts(seq, "00000000000000000000000000000000");
2311
2312         dev = rt->fib6_nh.nh_dev;
2313         seq_printf(seq, " %08x %08x %08x %08x %8s\n",
2314                    rt->fib6_metric, atomic_read(&rt->fib6_ref), 0,
2315                    rt->fib6_flags, dev ? dev->name : "");
2316         iter->w.leaf = NULL;
2317         return 0;
2318 }
2319
2320 static int ipv6_route_yield(struct fib6_walker *w)
2321 {
2322         struct ipv6_route_iter *iter = w->args;
2323
2324         if (!iter->skip)
2325                 return 1;
2326
2327         do {
2328                 iter->w.leaf = rcu_dereference_protected(
2329                                 iter->w.leaf->fib6_next,
2330                                 lockdep_is_held(&iter->tbl->tb6_lock));
2331                 iter->skip--;
2332                 if (!iter->skip && iter->w.leaf)
2333                         return 1;
2334         } while (iter->w.leaf);
2335
2336         return 0;
2337 }
2338
2339 static void ipv6_route_seq_setup_walk(struct ipv6_route_iter *iter,
2340                                       struct net *net)
2341 {
2342         memset(&iter->w, 0, sizeof(iter->w));
2343         iter->w.func = ipv6_route_yield;
2344         iter->w.root = &iter->tbl->tb6_root;
2345         iter->w.state = FWS_INIT;
2346         iter->w.node = iter->w.root;
2347         iter->w.args = iter;
2348         iter->sernum = iter->w.root->fn_sernum;
2349         INIT_LIST_HEAD(&iter->w.lh);
2350         fib6_walker_link(net, &iter->w);
2351 }
2352
2353 static struct fib6_table *ipv6_route_seq_next_table(struct fib6_table *tbl,
2354                                                     struct net *net)
2355 {
2356         unsigned int h;
2357         struct hlist_node *node;
2358
2359         if (tbl) {
2360                 h = (tbl->tb6_id & (FIB6_TABLE_HASHSZ - 1)) + 1;
2361                 node = rcu_dereference_bh(hlist_next_rcu(&tbl->tb6_hlist));
2362         } else {
2363                 h = 0;
2364                 node = NULL;
2365         }
2366
2367         while (!node && h < FIB6_TABLE_HASHSZ) {
2368                 node = rcu_dereference_bh(
2369                         hlist_first_rcu(&net->ipv6.fib_table_hash[h++]));
2370         }
2371         return hlist_entry_safe(node, struct fib6_table, tb6_hlist);
2372 }
2373
2374 static void ipv6_route_check_sernum(struct ipv6_route_iter *iter)
2375 {
2376         if (iter->sernum != iter->w.root->fn_sernum) {
2377                 iter->sernum = iter->w.root->fn_sernum;
2378                 iter->w.state = FWS_INIT;
2379                 iter->w.node = iter->w.root;
2380                 WARN_ON(iter->w.skip);
2381                 iter->w.skip = iter->w.count;
2382         }
2383 }
2384
2385 static void *ipv6_route_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2386 {
2387         int r;
2388         struct fib6_info *n;
2389         struct net *net = seq_file_net(seq);
2390         struct ipv6_route_iter *iter = seq->private;
2391
2392         if (!v)
2393                 goto iter_table;
2394
2395         n = rcu_dereference_bh(((struct fib6_info *)v)->fib6_next);
2396         if (n) {
2397                 ++*pos;
2398                 return n;
2399         }
2400
2401 iter_table:
2402         ipv6_route_check_sernum(iter);
2403         spin_lock_bh(&iter->tbl->tb6_lock);
2404         r = fib6_walk_continue(&iter->w);
2405         spin_unlock_bh(&iter->tbl->tb6_lock);
2406         if (r > 0) {
2407                 if (v)
2408                         ++*pos;
2409                 return iter->w.leaf;
2410         } else if (r < 0) {
2411                 fib6_walker_unlink(net, &iter->w);
2412                 return NULL;
2413         }
2414         fib6_walker_unlink(net, &iter->w);
2415
2416         iter->tbl = ipv6_route_seq_next_table(iter->tbl, net);
2417         if (!iter->tbl)
2418                 return NULL;
2419
2420         ipv6_route_seq_setup_walk(iter, net);
2421         goto iter_table;
2422 }
2423
2424 static void *ipv6_route_seq_start(struct seq_file *seq, loff_t *pos)
2425         __acquires(RCU_BH)
2426 {
2427         struct net *net = seq_file_net(seq);
2428         struct ipv6_route_iter *iter = seq->private;
2429
2430         rcu_read_lock_bh();
2431         iter->tbl = ipv6_route_seq_next_table(NULL, net);
2432         iter->skip = *pos;
2433
2434         if (iter->tbl) {
2435                 ipv6_route_seq_setup_walk(iter, net);
2436                 return ipv6_route_seq_next(seq, NULL, pos);
2437         } else {
2438                 return NULL;
2439         }
2440 }
2441
2442 static bool ipv6_route_iter_active(struct ipv6_route_iter *iter)
2443 {
2444         struct fib6_walker *w = &iter->w;
2445         return w->node && !(w->state == FWS_U && w->node == w->root);
2446 }
2447
2448 static void ipv6_route_seq_stop(struct seq_file *seq, void *v)
2449         __releases(RCU_BH)
2450 {
2451         struct net *net = seq_file_net(seq);
2452         struct ipv6_route_iter *iter = seq->private;
2453
2454         if (ipv6_route_iter_active(iter))
2455                 fib6_walker_unlink(net, &iter->w);
2456
2457         rcu_read_unlock_bh();
2458 }
2459
2460 const struct seq_operations ipv6_route_seq_ops = {
2461         .start  = ipv6_route_seq_start,
2462         .next   = ipv6_route_seq_next,
2463         .stop   = ipv6_route_seq_stop,
2464         .show   = ipv6_route_seq_show
2465 };
2466 #endif /* CONFIG_PROC_FS */